diff options
Diffstat (limited to 'src/gallium')
116 files changed, 14018 insertions, 2838 deletions
diff --git a/src/gallium/README.portability b/src/gallium/README.portability index 18a97f4..d5d5987 100644 --- a/src/gallium/README.portability +++ b/src/gallium/README.portability @@ -6,7 +6,7 @@ The state tracker and winsys driver support a rather limited number of platforms. However, the pipe drivers are meant to run in a wide number of platforms. Hence the pipe drivers, the auxiliary modules, and all public -headers in general, should stricly follow these guidelines to ensure +headers in general, should strictly follow these guidelines to ensure = Compiler Support = diff --git a/src/gallium/auxiliary/cso_cache/cso_context.c b/src/gallium/auxiliary/cso_cache/cso_context.c index 7236bff..af4af8a 100644 --- a/src/gallium/auxiliary/cso_cache/cso_context.c +++ b/src/gallium/auxiliary/cso_cache/cso_context.c @@ -769,8 +769,7 @@ void cso_restore_vertex_shader(struct cso_context *ctx) enum pipe_error cso_set_framebuffer(struct cso_context *ctx, const struct pipe_framebuffer_state *fb) { - /* XXX this memcmp() fails to detect buffer size changes */ - if (1/*memcmp(&ctx->fb, fb, sizeof(*fb))*/) { + if (memcmp(&ctx->fb, fb, sizeof(*fb)) != 0) { ctx->fb = *fb; ctx->pipe->set_framebuffer_state(ctx->pipe, fb); } diff --git a/src/gallium/auxiliary/draw/draw_pipe_aaline.c b/src/gallium/auxiliary/draw/draw_pipe_aaline.c index d93708a..fd48b22 100644 --- a/src/gallium/auxiliary/draw/draw_pipe_aaline.c +++ b/src/gallium/auxiliary/draw/draw_pipe_aaline.c @@ -415,8 +415,11 @@ aaline_create_texture(struct aaline_stage *aaline) assert(aaline->texture->width[level] == aaline->texture->height[level]); - surface = screen->get_tex_surface(screen, aaline->texture, 0, level, 0); - data = pipe_surface_map(surface); + /* This texture is new, no need to flush. + */ + surface = screen->get_tex_surface(screen, aaline->texture, 0, level, 0, + PIPE_BUFFER_USAGE_CPU_WRITE); + data = screen->surface_map(screen, surface, PIPE_BUFFER_USAGE_CPU_WRITE); if (data == NULL) return FALSE; @@ -440,9 +443,8 @@ aaline_create_texture(struct aaline_stage *aaline) } /* unmap */ - pipe_surface_unmap(surface); - pipe_surface_reference(&surface, NULL); - pipe->texture_update(pipe, aaline->texture, 0, (1 << level)); + screen->surface_unmap(screen, surface); + screen->tex_surface_release(screen, &surface); } return TRUE; } @@ -713,6 +715,11 @@ static void aaline_destroy(struct draw_stage *stage) { struct aaline_stage *aaline = aaline_stage(stage); + uint i; + + for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { + pipe_texture_reference(&aaline->state.texture[i], NULL); + } if (aaline->sampler_cso) aaline->pipe->delete_sampler_state(aaline->pipe, aaline->sampler_cso); @@ -836,6 +843,9 @@ aaline_set_sampler_textures(struct pipe_context *pipe, for (i = 0; i < num; i++) { pipe_texture_reference(&aaline->state.texture[i], texture[i]); } + for ( ; i < PIPE_MAX_SAMPLERS; i++) { + pipe_texture_reference(&aaline->state.texture[i], NULL); + } aaline->num_textures = num; /* pass-through */ diff --git a/src/gallium/auxiliary/draw/draw_pipe_pstipple.c b/src/gallium/auxiliary/draw/draw_pipe_pstipple.c index 73ee419..4c92416 100644 --- a/src/gallium/auxiliary/draw/draw_pipe_pstipple.c +++ b/src/gallium/auxiliary/draw/draw_pipe_pstipple.c @@ -376,8 +376,14 @@ pstip_update_texture(struct pstip_stage *pstip) uint i, j; ubyte *data; - surface = screen->get_tex_surface(screen, pstip->texture, 0, 0, 0); - data = pipe_surface_map(surface); + /* XXX: want to avoid flushing just because we use stipple: + */ + pipe->flush( pipe, PIPE_FLUSH_TEXTURE_CACHE, NULL ); + + surface = screen->get_tex_surface(screen, pstip->texture, 0, 0, 0, + PIPE_BUFFER_USAGE_CPU_WRITE); + data = screen->surface_map(screen, surface, + PIPE_BUFFER_USAGE_CPU_WRITE); /* * Load alpha texture. @@ -399,9 +405,8 @@ pstip_update_texture(struct pstip_stage *pstip) } /* unmap */ - pipe_surface_unmap(surface); - pipe_surface_reference(&surface, NULL); - pipe->texture_update(pipe, pstip->texture, 0, 0x1); + screen->surface_unmap(screen, surface); + screen->tex_surface_release(screen, &surface); } @@ -559,6 +564,11 @@ static void pstip_destroy(struct draw_stage *stage) { struct pstip_stage *pstip = pstip_stage(stage); + uint i; + + for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { + pipe_texture_reference(&pstip->state.textures[i], NULL); + } pstip->pipe->delete_sampler_state(pstip->pipe, pstip->sampler_cso); diff --git a/src/gallium/auxiliary/draw/draw_pt_elts.c b/src/gallium/auxiliary/draw/draw_pt_elts.c index 2094c08..b7780fb 100644 --- a/src/gallium/auxiliary/draw/draw_pt_elts.c +++ b/src/gallium/auxiliary/draw/draw_pt_elts.c @@ -60,10 +60,10 @@ static unsigned elt_vert( const void *elts, unsigned idx ) pt_elt_func draw_pt_elt_func( struct draw_context *draw ) { switch (draw->pt.user.eltSize) { - case 0: return elt_vert; - case 1: return elt_ubyte; - case 2: return elt_ushort; - case 4: return elt_uint; + case 0: return &elt_vert; + case 1: return &elt_ubyte; + case 2: return &elt_ushort; + case 4: return &elt_uint; default: return NULL; } } diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c index 7781782..2dc9352 100644 --- a/src/gallium/auxiliary/draw/draw_vs_sse.c +++ b/src/gallium/auxiliary/draw/draw_vs_sse.c @@ -31,9 +31,11 @@ * Brian Paul */ +#include "pipe/p_config.h" + #include "draw_vs.h" -#if defined(__i386__) || defined(__386__) +#if defined(PIPE_ARCH_X86) #include "pipe/p_util.h" #include "pipe/p_shader_tokens.h" diff --git a/src/gallium/auxiliary/pipebuffer/Makefile b/src/gallium/auxiliary/pipebuffer/Makefile index d654dbc..ff09011 100644 --- a/src/gallium/auxiliary/pipebuffer/Makefile +++ b/src/gallium/auxiliary/pipebuffer/Makefile @@ -11,6 +11,7 @@ C_SOURCES = \ pb_bufmgr_mm.c \ pb_bufmgr_pool.c \ pb_bufmgr_slab.c \ + pb_validate.c \ pb_winsys.c include ../../Makefile.template diff --git a/src/gallium/auxiliary/pipebuffer/SConscript b/src/gallium/auxiliary/pipebuffer/SConscript index 604a217..9db0c0e 100644 --- a/src/gallium/auxiliary/pipebuffer/SConscript +++ b/src/gallium/auxiliary/pipebuffer/SConscript @@ -10,6 +10,7 @@ pipebuffer = env.ConvenienceLibrary( 'pb_bufmgr_mm.c', 'pb_bufmgr_pool.c', 'pb_bufmgr_slab.c', + 'pb_validate.c', 'pb_winsys.c', ]) diff --git a/src/gallium/auxiliary/pipebuffer/pb_buffer.h b/src/gallium/auxiliary/pipebuffer/pb_buffer.h index 49705cb..857ea53 100644 --- a/src/gallium/auxiliary/pipebuffer/pb_buffer.h +++ b/src/gallium/auxiliary/pipebuffer/pb_buffer.h @@ -146,6 +146,8 @@ pb_map(struct pb_buffer *buf, unsigned flags) { assert(buf); + if(!buf) + return NULL; return buf->vtbl->map(buf, flags); } @@ -154,6 +156,8 @@ static INLINE void pb_unmap(struct pb_buffer *buf) { assert(buf); + if(!buf) + return; buf->vtbl->unmap(buf); } @@ -163,6 +167,12 @@ pb_get_base_buffer( struct pb_buffer *buf, struct pb_buffer **base_buf, unsigned *offset ) { + assert(buf); + if(!buf) { + base_buf = NULL; + offset = 0; + return; + } buf->vtbl->get_base_buffer(buf, base_buf, offset); } @@ -171,7 +181,8 @@ static INLINE void pb_destroy(struct pb_buffer *buf) { assert(buf); - assert(buf->vtbl); + if(!buf) + return; buf->vtbl->destroy(buf); } diff --git a/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c b/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c index 2fa0842..7f23688 100644 --- a/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c +++ b/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c @@ -168,6 +168,28 @@ _fenced_buffer_remove(struct fenced_buffer *fenced_buf) } +static INLINE enum pipe_error +_fenced_buffer_finish(struct fenced_buffer *fenced_buf) +{ + struct fenced_buffer_list *fenced_list = fenced_buf->list; + struct pipe_winsys *winsys = fenced_list->winsys; + + debug_warning("waiting for GPU"); + + assert(fenced_buf->fence); + if(fenced_buf->fence) { + if(winsys->fence_finish(winsys, fenced_buf->fence, 0) != 0) { + return PIPE_ERROR; + } + /* Remove from the fenced list */ + _fenced_buffer_remove(fenced_buf); /* TODO: remove consequents */ + } + + fenced_buf->flags &= ~PIPE_BUFFER_USAGE_GPU_READ_WRITE; + return PIPE_OK; +} + + /** * Free as many fenced buffers from the list head as possible. */ @@ -207,40 +229,6 @@ _fenced_buffer_list_check_free(struct fenced_buffer_list *fenced_list, } -/** - * Serialize writes, but allow concurrent reads. - */ -static INLINE enum pipe_error -fenced_buffer_serialize(struct fenced_buffer *fenced_buf, unsigned flags) -{ - struct fenced_buffer_list *fenced_list = fenced_buf->list; - struct pipe_winsys *winsys = fenced_list->winsys; - - /* Allow concurrent reads */ - if(((fenced_buf->flags | flags) & PIPE_BUFFER_USAGE_WRITE) == 0) - return PIPE_OK; - - /* Wait for the CPU to finish */ - if(fenced_buf->mapcount) { - /* FIXME: Use thread conditions variables to signal when mapcount - * reaches zero */ - debug_warning("attemp to write concurrently to buffer"); - /* XXX: we must not fail here in order to support texture mipmap generation - return PIPE_ERROR_RETRY; - */ - } - - /* Wait for the GPU to finish */ - if(fenced_buf->fence) { - if(winsys->fence_finish(winsys, fenced_buf->fence, 0) != 0) - return PIPE_ERROR_RETRY; - _fenced_buffer_remove(fenced_buf); - } - - return PIPE_OK; -} - - static void fenced_buffer_destroy(struct pb_buffer *buf) { @@ -280,15 +268,28 @@ fenced_buffer_map(struct pb_buffer *buf, { struct fenced_buffer *fenced_buf = fenced_buffer(buf); void *map; - assert((flags & ~PIPE_BUFFER_USAGE_CPU_READ_WRITE) == 0); + + assert(!(flags & ~PIPE_BUFFER_USAGE_CPU_READ_WRITE)); + flags &= PIPE_BUFFER_USAGE_CPU_READ_WRITE; - if(fenced_buffer_serialize(fenced_buf, flags) != PIPE_OK) - return NULL; + /* Check for GPU read/write access */ + if(fenced_buf->flags & PIPE_BUFFER_USAGE_GPU_WRITE) { + /* Wait for the GPU to finish writing */ + _fenced_buffer_finish(fenced_buf); + } + + /* Check for CPU write access (read is OK) */ + if(fenced_buf->flags & PIPE_BUFFER_USAGE_CPU_READ_WRITE) { + /* this is legal -- just for debugging */ + debug_warning("concurrent CPU writes"); + } map = pb_map(fenced_buf->buffer, flags); - if(map) + if(map) { ++fenced_buf->mapcount; - fenced_buf->flags |= flags & PIPE_BUFFER_USAGE_CPU_READ_WRITE; + fenced_buf->flags |= flags; + } + return map; } @@ -298,10 +299,12 @@ fenced_buffer_unmap(struct pb_buffer *buf) { struct fenced_buffer *fenced_buf = fenced_buffer(buf); assert(fenced_buf->mapcount); - pb_unmap(fenced_buf->buffer); - --fenced_buf->mapcount; - if(!fenced_buf->mapcount) - fenced_buf->flags &= ~PIPE_BUFFER_USAGE_CPU_READ_WRITE; + if(fenced_buf->mapcount) { + pb_unmap(fenced_buf->buffer); + --fenced_buf->mapcount; + if(!fenced_buf->mapcount) + fenced_buf->flags &= ~PIPE_BUFFER_USAGE_CPU_READ_WRITE; + } } @@ -334,8 +337,10 @@ fenced_buffer_create(struct fenced_buffer_list *fenced_list, return NULL; buf = CALLOC_STRUCT(fenced_buffer); - if(!buf) + if(!buf) { + pb_reference(&buffer, NULL); return NULL; + } buf->base.base.refcount = 1; buf->base.base.alignment = buffer->base.alignment; @@ -374,7 +379,7 @@ buffer_fence(struct pb_buffer *buf, fenced_list = fenced_buf->list; winsys = fenced_list->winsys; - if(fence == fenced_buf->fence) { + if(!fence || fence == fenced_buf->fence) { /* Handle the same fence case specially, not only because it is a fast * path, but mostly to avoid serializing two writes with the same fence, * as that would bring the hardware down to synchronous operation without @@ -384,11 +389,6 @@ buffer_fence(struct pb_buffer *buf, return; } - if(fenced_buffer_serialize(fenced_buf, flags) != PIPE_OK) { - /* FIXME: propagate error */ - (void)0; - } - _glthread_LOCK_MUTEX(fenced_list->mutex); if (fenced_buf->fence) _fenced_buffer_remove(fenced_buf); diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr.h b/src/gallium/auxiliary/pipebuffer/pb_bufmgr.h index 8de286e..f6cc7a5 100644 --- a/src/gallium/auxiliary/pipebuffer/pb_bufmgr.h +++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr.h @@ -50,7 +50,8 @@ #define PB_BUFMGR_H_ -#include <stddef.h> +#include "pipe/p_compiler.h" +#include "pipe/p_error.h" #ifdef __cplusplus @@ -68,7 +69,6 @@ struct pipe_winsys; */ struct pb_manager { - /* XXX: we will likely need more allocation flags */ struct pb_buffer * (*create_buffer)( struct pb_manager *mgr, size_t size, diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_slab.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_slab.c index b931455..45ba158 100644 --- a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_slab.c +++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_slab.c @@ -47,9 +47,6 @@ #include "pb_bufmgr.h" -#define DRI_SLABPOOL_ALLOC_RETRIES 100 - - struct pb_slab; struct pb_slab_buffer @@ -313,7 +310,6 @@ pb_slab_manager_create_buffer(struct pb_manager *_mgr, static struct pb_slab_buffer *buf; struct pb_slab *slab; struct list_head *list; - int count = DRI_SLABPOOL_ALLOC_RETRIES; /* check size */ assert(size == mgr->bufSize); @@ -331,23 +327,14 @@ pb_slab_manager_create_buffer(struct pb_manager *_mgr, /* XXX: check for compatible buffer usage too? */ _glthread_LOCK_MUTEX(mgr->mutex); - while (mgr->slabs.next == &mgr->slabs && count > 0) { - if (mgr->slabs.next != &mgr->slabs) - break; - - _glthread_UNLOCK_MUTEX(mgr->mutex); - if (count != DRI_SLABPOOL_ALLOC_RETRIES) - util_time_sleep(1); - _glthread_LOCK_MUTEX(mgr->mutex); + if (mgr->slabs.next == &mgr->slabs) { (void) pb_slab_create(mgr); - count--; + if (mgr->slabs.next == &mgr->slabs) { + _glthread_UNLOCK_MUTEX(mgr->mutex); + return NULL; + } } - list = mgr->slabs.next; - if (list == &mgr->slabs) { - _glthread_UNLOCK_MUTEX(mgr->mutex); - return NULL; - } slab = LIST_ENTRY(struct pb_slab, list, head); if (--slab->numFree == 0) LIST_DELINIT(list); diff --git a/src/gallium/auxiliary/pipebuffer/pb_validate.c b/src/gallium/auxiliary/pipebuffer/pb_validate.c new file mode 100644 index 0000000..362fd89 --- /dev/null +++ b/src/gallium/auxiliary/pipebuffer/pb_validate.c @@ -0,0 +1,153 @@ +/************************************************************************** + * + * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @file + * Buffer validation. + * + * @author Jose Fonseca <jrfonseca@tungstengraphics.com> + */ + + +#include "pipe/p_compiler.h" +#include "pipe/p_error.h" +#include "pipe/p_util.h" +#include "pipe/p_debug.h" + +#include "pb_buffer.h" +#include "pb_buffer_fenced.h" +#include "pb_validate.h" + + +#define PB_VALIDATE_INITIAL_SIZE 1 /* 512 */ + + +struct pb_validate +{ + struct pb_buffer **buffers; + unsigned used; + unsigned size; +}; + + +enum pipe_error +pb_validate_add_buffer(struct pb_validate *vl, + struct pb_buffer *buf) +{ + assert(buf); + if(!buf) + return PIPE_ERROR; + + /* We only need to store one reference for each buffer, so avoid storing + * consecutive references for the same buffer. It might not be the more + * common pasttern, but it is easy to implement. + */ + if(vl->used && vl->buffers[vl->used - 1] == buf) { + return PIPE_OK; + } + + /* Grow the table */ + if(vl->used == vl->size) { + unsigned new_size; + struct pb_buffer **new_buffers; + + new_size = vl->size * 2; + if(!new_size) + return PIPE_ERROR_OUT_OF_MEMORY; + + new_buffers = (struct pb_buffer **)REALLOC(vl->buffers, + vl->size*sizeof(struct pb_buffer *), + new_size*sizeof(struct pb_buffer *)); + if(!new_buffers) + return PIPE_ERROR_OUT_OF_MEMORY; + + memset(new_buffers + vl->size, 0, (new_size - vl->size)*sizeof(struct pb_buffer *)); + + vl->size = new_size; + vl->buffers = new_buffers; + } + + assert(!vl->buffers[vl->used]); + pb_reference(&vl->buffers[vl->used], buf); + ++vl->used; + + return PIPE_OK; +} + + +enum pipe_error +pb_validate_validate(struct pb_validate *vl) +{ + /* FIXME: go through each buffer, ensure its not mapped, its address is + * available -- requires a new pb_buffer interface */ + return PIPE_OK; +} + + +void +pb_validate_fence(struct pb_validate *vl, + struct pipe_fence_handle *fence) +{ + unsigned i; + for(i = 0; i < vl->used; ++i) { + buffer_fence(vl->buffers[i], fence); + pb_reference(&vl->buffers[i], NULL); + } + vl->used = 0; +} + + +void +pb_validate_destroy(struct pb_validate *vl) +{ + unsigned i; + for(i = 0; i < vl->used; ++i) + pb_reference(&vl->buffers[i], NULL); + FREE(vl->buffers); + FREE(vl); +} + + +struct pb_validate * +pb_validate_create() +{ + struct pb_validate *vl; + + vl = CALLOC_STRUCT(pb_validate); + if(!vl) + return NULL; + + vl->size = PB_VALIDATE_INITIAL_SIZE; + vl->buffers = (struct pb_buffer **)CALLOC(vl->size, sizeof(struct pb_buffer *)); + if(!vl->buffers) { + FREE(vl); + return NULL; + } + + return vl; +} + diff --git a/src/gallium/auxiliary/pipebuffer/pb_validate.h b/src/gallium/auxiliary/pipebuffer/pb_validate.h new file mode 100644 index 0000000..3db1d53 --- /dev/null +++ b/src/gallium/auxiliary/pipebuffer/pb_validate.h @@ -0,0 +1,91 @@ +/************************************************************************** + * + * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @file + * Buffer validation. + * + * @author Jose Fonseca <jrfonseca@tungstengraphics.com> + */ + +#ifndef PB_VALIDATE_H_ +#define PB_VALIDATE_H_ + + +#include "pipe/p_compiler.h" +#include "pipe/p_error.h" + +#ifdef __cplusplus +extern "C" { +#endif + + +struct pb_buffer; +struct pipe_fence_handle; + + +/** + * Buffer validation list. + * + * It holds a list of buffers to be validated and fenced when flushing. + */ +struct pb_validate; + + +enum pipe_error +pb_validate_add_buffer(struct pb_validate *vl, + struct pb_buffer *buf); + +/** + * Validate all buffers for hardware access. + * + * Should be called right before issuing commands to the hardware. + */ +enum pipe_error +pb_validate_validate(struct pb_validate *vl); + +/** + * Fence all buffers and clear the list. + * + * Should be called right before issuing commands to the hardware. + */ +void +pb_validate_fence(struct pb_validate *vl, + struct pipe_fence_handle *fence); + +struct pb_validate * +pb_validate_create(void); + +void +pb_validate_destroy(struct pb_validate *vl); + + +#ifdef __cplusplus +} +#endif + +#endif /*PB_VALIDATE_H_*/ diff --git a/src/gallium/auxiliary/rtasm/rtasm_cpu.c b/src/gallium/auxiliary/rtasm/rtasm_cpu.c index f01e12f..5499018 100644 --- a/src/gallium/auxiliary/rtasm/rtasm_cpu.c +++ b/src/gallium/auxiliary/rtasm/rtasm_cpu.c @@ -47,7 +47,7 @@ static boolean rtasm_sse_enabled(void) int rtasm_cpu_has_sse(void) { /* FIXME: actually detect this at run-time */ -#if defined(__i386__) || defined(__386__) || defined(i386) +#if defined(PIPE_ARCH_X86) return rtasm_sse_enabled(); #else return 0; @@ -57,7 +57,7 @@ int rtasm_cpu_has_sse(void) int rtasm_cpu_has_sse2(void) { /* FIXME: actually detect this at run-time */ -#if defined(__i386__) || defined(__386__) || defined(i386) +#if defined(PIPE_ARCH_X86) return rtasm_sse_enabled(); #else return 0; diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c index 2415b01..698389f 100644 --- a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c +++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c @@ -21,7 +21,9 @@ * **************************************************************************/ -#if defined(__i386__) || defined(__386__) || defined(i386) +#include "pipe/p_config.h" + +#if defined(PIPE_ARCH_X86) #include "pipe/p_compiler.h" #include "pipe/p_debug.h" diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.h b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h index 63e812f..bd76e17 100644 --- a/src/gallium/auxiliary/rtasm/rtasm_x86sse.h +++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h @@ -24,7 +24,9 @@ #ifndef _RTASM_X86SSE_H_ #define _RTASM_X86SSE_H_ -#if defined(__i386__) || defined(__386__) || defined(i386) +#include "pipe/p_config.h" + +#if defined(PIPE_ARCH_X86) /* It is up to the caller to ensure that instructions issued are * suitable for the host cpu. There are no checks made in this module diff --git a/src/gallium/auxiliary/tgsi/util/tgsi_util.c b/src/gallium/auxiliary/tgsi/util/tgsi_util.c index 4cdd891..56a50d3 100644 --- a/src/gallium/auxiliary/tgsi/util/tgsi_util.c +++ b/src/gallium/auxiliary/tgsi/util/tgsi_util.c @@ -8,7 +8,7 @@ union pointer_hack { void *pointer; - unsigned long long uint64; + uint64_t uint64; }; void * diff --git a/src/gallium/auxiliary/translate/translate.c b/src/gallium/auxiliary/translate/translate.c index b04bc6e..b93fbf9 100644 --- a/src/gallium/auxiliary/translate/translate.c +++ b/src/gallium/auxiliary/translate/translate.c @@ -30,6 +30,7 @@ * Keith Whitwell <keith@tungstengraphics.com> */ +#include "pipe/p_config.h" #include "pipe/p_util.h" #include "pipe/p_state.h" #include "translate.h" @@ -38,7 +39,7 @@ struct translate *translate_create( const struct translate_key *key ) { struct translate *translate = NULL; -#if defined(__i386__) || defined(__386__) || defined(i386) +#if defined(PIPE_ARCH_X86) translate = translate_sse2_create( key ); if (translate) return translate; diff --git a/src/gallium/auxiliary/translate/translate_generic.c b/src/gallium/auxiliary/translate/translate_generic.c index a25d94f..3fec89b 100644 --- a/src/gallium/auxiliary/translate/translate_generic.c +++ b/src/gallium/auxiliary/translate/translate_generic.c @@ -255,140 +255,140 @@ static fetch_func get_fetch_func( enum pipe_format format ) { switch (format) { case PIPE_FORMAT_R64_FLOAT: - return fetch_R64_FLOAT; + return &fetch_R64_FLOAT; case PIPE_FORMAT_R64G64_FLOAT: - return fetch_R64G64_FLOAT; + return &fetch_R64G64_FLOAT; case PIPE_FORMAT_R64G64B64_FLOAT: - return fetch_R64G64B64_FLOAT; + return &fetch_R64G64B64_FLOAT; case PIPE_FORMAT_R64G64B64A64_FLOAT: - return fetch_R64G64B64A64_FLOAT; + return &fetch_R64G64B64A64_FLOAT; case PIPE_FORMAT_R32_FLOAT: - return fetch_R32_FLOAT; + return &fetch_R32_FLOAT; case PIPE_FORMAT_R32G32_FLOAT: - return fetch_R32G32_FLOAT; + return &fetch_R32G32_FLOAT; case PIPE_FORMAT_R32G32B32_FLOAT: - return fetch_R32G32B32_FLOAT; + return &fetch_R32G32B32_FLOAT; case PIPE_FORMAT_R32G32B32A32_FLOAT: - return fetch_R32G32B32A32_FLOAT; + return &fetch_R32G32B32A32_FLOAT; case PIPE_FORMAT_R32_UNORM: - return fetch_R32_UNORM; + return &fetch_R32_UNORM; case PIPE_FORMAT_R32G32_UNORM: - return fetch_R32G32_UNORM; + return &fetch_R32G32_UNORM; case PIPE_FORMAT_R32G32B32_UNORM: - return fetch_R32G32B32_UNORM; + return &fetch_R32G32B32_UNORM; case PIPE_FORMAT_R32G32B32A32_UNORM: - return fetch_R32G32B32A32_UNORM; + return &fetch_R32G32B32A32_UNORM; case PIPE_FORMAT_R32_USCALED: - return fetch_R32_USCALED; + return &fetch_R32_USCALED; case PIPE_FORMAT_R32G32_USCALED: - return fetch_R32G32_USCALED; + return &fetch_R32G32_USCALED; case PIPE_FORMAT_R32G32B32_USCALED: - return fetch_R32G32B32_USCALED; + return &fetch_R32G32B32_USCALED; case PIPE_FORMAT_R32G32B32A32_USCALED: - return fetch_R32G32B32A32_USCALED; + return &fetch_R32G32B32A32_USCALED; case PIPE_FORMAT_R32_SNORM: - return fetch_R32_SNORM; + return &fetch_R32_SNORM; case PIPE_FORMAT_R32G32_SNORM: - return fetch_R32G32_SNORM; + return &fetch_R32G32_SNORM; case PIPE_FORMAT_R32G32B32_SNORM: - return fetch_R32G32B32_SNORM; + return &fetch_R32G32B32_SNORM; case PIPE_FORMAT_R32G32B32A32_SNORM: - return fetch_R32G32B32A32_SNORM; + return &fetch_R32G32B32A32_SNORM; case PIPE_FORMAT_R32_SSCALED: - return fetch_R32_SSCALED; + return &fetch_R32_SSCALED; case PIPE_FORMAT_R32G32_SSCALED: - return fetch_R32G32_SSCALED; + return &fetch_R32G32_SSCALED; case PIPE_FORMAT_R32G32B32_SSCALED: - return fetch_R32G32B32_SSCALED; + return &fetch_R32G32B32_SSCALED; case PIPE_FORMAT_R32G32B32A32_SSCALED: - return fetch_R32G32B32A32_SSCALED; + return &fetch_R32G32B32A32_SSCALED; case PIPE_FORMAT_R16_UNORM: - return fetch_R16_UNORM; + return &fetch_R16_UNORM; case PIPE_FORMAT_R16G16_UNORM: - return fetch_R16G16_UNORM; + return &fetch_R16G16_UNORM; case PIPE_FORMAT_R16G16B16_UNORM: - return fetch_R16G16B16_UNORM; + return &fetch_R16G16B16_UNORM; case PIPE_FORMAT_R16G16B16A16_UNORM: - return fetch_R16G16B16A16_UNORM; + return &fetch_R16G16B16A16_UNORM; case PIPE_FORMAT_R16_USCALED: - return fetch_R16_USCALED; + return &fetch_R16_USCALED; case PIPE_FORMAT_R16G16_USCALED: - return fetch_R16G16_USCALED; + return &fetch_R16G16_USCALED; case PIPE_FORMAT_R16G16B16_USCALED: - return fetch_R16G16B16_USCALED; + return &fetch_R16G16B16_USCALED; case PIPE_FORMAT_R16G16B16A16_USCALED: - return fetch_R16G16B16A16_USCALED; + return &fetch_R16G16B16A16_USCALED; case PIPE_FORMAT_R16_SNORM: - return fetch_R16_SNORM; + return &fetch_R16_SNORM; case PIPE_FORMAT_R16G16_SNORM: - return fetch_R16G16_SNORM; + return &fetch_R16G16_SNORM; case PIPE_FORMAT_R16G16B16_SNORM: - return fetch_R16G16B16_SNORM; + return &fetch_R16G16B16_SNORM; case PIPE_FORMAT_R16G16B16A16_SNORM: - return fetch_R16G16B16A16_SNORM; + return &fetch_R16G16B16A16_SNORM; case PIPE_FORMAT_R16_SSCALED: - return fetch_R16_SSCALED; + return &fetch_R16_SSCALED; case PIPE_FORMAT_R16G16_SSCALED: - return fetch_R16G16_SSCALED; + return &fetch_R16G16_SSCALED; case PIPE_FORMAT_R16G16B16_SSCALED: - return fetch_R16G16B16_SSCALED; + return &fetch_R16G16B16_SSCALED; case PIPE_FORMAT_R16G16B16A16_SSCALED: - return fetch_R16G16B16A16_SSCALED; + return &fetch_R16G16B16A16_SSCALED; case PIPE_FORMAT_R8_UNORM: - return fetch_R8_UNORM; + return &fetch_R8_UNORM; case PIPE_FORMAT_R8G8_UNORM: - return fetch_R8G8_UNORM; + return &fetch_R8G8_UNORM; case PIPE_FORMAT_R8G8B8_UNORM: - return fetch_R8G8B8_UNORM; + return &fetch_R8G8B8_UNORM; case PIPE_FORMAT_R8G8B8A8_UNORM: - return fetch_R8G8B8A8_UNORM; + return &fetch_R8G8B8A8_UNORM; case PIPE_FORMAT_R8_USCALED: - return fetch_R8_USCALED; + return &fetch_R8_USCALED; case PIPE_FORMAT_R8G8_USCALED: - return fetch_R8G8_USCALED; + return &fetch_R8G8_USCALED; case PIPE_FORMAT_R8G8B8_USCALED: - return fetch_R8G8B8_USCALED; + return &fetch_R8G8B8_USCALED; case PIPE_FORMAT_R8G8B8A8_USCALED: - return fetch_R8G8B8A8_USCALED; + return &fetch_R8G8B8A8_USCALED; case PIPE_FORMAT_R8_SNORM: - return fetch_R8_SNORM; + return &fetch_R8_SNORM; case PIPE_FORMAT_R8G8_SNORM: - return fetch_R8G8_SNORM; + return &fetch_R8G8_SNORM; case PIPE_FORMAT_R8G8B8_SNORM: - return fetch_R8G8B8_SNORM; + return &fetch_R8G8B8_SNORM; case PIPE_FORMAT_R8G8B8A8_SNORM: - return fetch_R8G8B8A8_SNORM; + return &fetch_R8G8B8A8_SNORM; case PIPE_FORMAT_R8_SSCALED: - return fetch_R8_SSCALED; + return &fetch_R8_SSCALED; case PIPE_FORMAT_R8G8_SSCALED: - return fetch_R8G8_SSCALED; + return &fetch_R8G8_SSCALED; case PIPE_FORMAT_R8G8B8_SSCALED: - return fetch_R8G8B8_SSCALED; + return &fetch_R8G8B8_SSCALED; case PIPE_FORMAT_R8G8B8A8_SSCALED: - return fetch_R8G8B8A8_SSCALED; + return &fetch_R8G8B8A8_SSCALED; case PIPE_FORMAT_A8R8G8B8_UNORM: - return fetch_A8R8G8B8_UNORM; + return &fetch_A8R8G8B8_UNORM; case PIPE_FORMAT_B8G8R8A8_UNORM: - return fetch_B8G8R8A8_UNORM; + return &fetch_B8G8R8A8_UNORM; default: assert(0); - return fetch_NULL; + return &fetch_NULL; } } @@ -399,140 +399,140 @@ static emit_func get_emit_func( enum pipe_format format ) { switch (format) { case PIPE_FORMAT_R64_FLOAT: - return emit_R64_FLOAT; + return &emit_R64_FLOAT; case PIPE_FORMAT_R64G64_FLOAT: - return emit_R64G64_FLOAT; + return &emit_R64G64_FLOAT; case PIPE_FORMAT_R64G64B64_FLOAT: - return emit_R64G64B64_FLOAT; + return &emit_R64G64B64_FLOAT; case PIPE_FORMAT_R64G64B64A64_FLOAT: - return emit_R64G64B64A64_FLOAT; + return &emit_R64G64B64A64_FLOAT; case PIPE_FORMAT_R32_FLOAT: - return emit_R32_FLOAT; + return &emit_R32_FLOAT; case PIPE_FORMAT_R32G32_FLOAT: - return emit_R32G32_FLOAT; + return &emit_R32G32_FLOAT; case PIPE_FORMAT_R32G32B32_FLOAT: - return emit_R32G32B32_FLOAT; + return &emit_R32G32B32_FLOAT; case PIPE_FORMAT_R32G32B32A32_FLOAT: - return emit_R32G32B32A32_FLOAT; + return &emit_R32G32B32A32_FLOAT; case PIPE_FORMAT_R32_UNORM: - return emit_R32_UNORM; + return &emit_R32_UNORM; case PIPE_FORMAT_R32G32_UNORM: - return emit_R32G32_UNORM; + return &emit_R32G32_UNORM; case PIPE_FORMAT_R32G32B32_UNORM: - return emit_R32G32B32_UNORM; + return &emit_R32G32B32_UNORM; case PIPE_FORMAT_R32G32B32A32_UNORM: - return emit_R32G32B32A32_UNORM; + return &emit_R32G32B32A32_UNORM; case PIPE_FORMAT_R32_USCALED: - return emit_R32_USCALED; + return &emit_R32_USCALED; case PIPE_FORMAT_R32G32_USCALED: - return emit_R32G32_USCALED; + return &emit_R32G32_USCALED; case PIPE_FORMAT_R32G32B32_USCALED: - return emit_R32G32B32_USCALED; + return &emit_R32G32B32_USCALED; case PIPE_FORMAT_R32G32B32A32_USCALED: - return emit_R32G32B32A32_USCALED; + return &emit_R32G32B32A32_USCALED; case PIPE_FORMAT_R32_SNORM: - return emit_R32_SNORM; + return &emit_R32_SNORM; case PIPE_FORMAT_R32G32_SNORM: - return emit_R32G32_SNORM; + return &emit_R32G32_SNORM; case PIPE_FORMAT_R32G32B32_SNORM: - return emit_R32G32B32_SNORM; + return &emit_R32G32B32_SNORM; case PIPE_FORMAT_R32G32B32A32_SNORM: - return emit_R32G32B32A32_SNORM; + return &emit_R32G32B32A32_SNORM; case PIPE_FORMAT_R32_SSCALED: - return emit_R32_SSCALED; + return &emit_R32_SSCALED; case PIPE_FORMAT_R32G32_SSCALED: - return emit_R32G32_SSCALED; + return &emit_R32G32_SSCALED; case PIPE_FORMAT_R32G32B32_SSCALED: - return emit_R32G32B32_SSCALED; + return &emit_R32G32B32_SSCALED; case PIPE_FORMAT_R32G32B32A32_SSCALED: - return emit_R32G32B32A32_SSCALED; + return &emit_R32G32B32A32_SSCALED; case PIPE_FORMAT_R16_UNORM: - return emit_R16_UNORM; + return &emit_R16_UNORM; case PIPE_FORMAT_R16G16_UNORM: - return emit_R16G16_UNORM; + return &emit_R16G16_UNORM; case PIPE_FORMAT_R16G16B16_UNORM: - return emit_R16G16B16_UNORM; + return &emit_R16G16B16_UNORM; case PIPE_FORMAT_R16G16B16A16_UNORM: - return emit_R16G16B16A16_UNORM; + return &emit_R16G16B16A16_UNORM; case PIPE_FORMAT_R16_USCALED: - return emit_R16_USCALED; + return &emit_R16_USCALED; case PIPE_FORMAT_R16G16_USCALED: - return emit_R16G16_USCALED; + return &emit_R16G16_USCALED; case PIPE_FORMAT_R16G16B16_USCALED: - return emit_R16G16B16_USCALED; + return &emit_R16G16B16_USCALED; case PIPE_FORMAT_R16G16B16A16_USCALED: - return emit_R16G16B16A16_USCALED; + return &emit_R16G16B16A16_USCALED; case PIPE_FORMAT_R16_SNORM: - return emit_R16_SNORM; + return &emit_R16_SNORM; case PIPE_FORMAT_R16G16_SNORM: - return emit_R16G16_SNORM; + return &emit_R16G16_SNORM; case PIPE_FORMAT_R16G16B16_SNORM: - return emit_R16G16B16_SNORM; + return &emit_R16G16B16_SNORM; case PIPE_FORMAT_R16G16B16A16_SNORM: - return emit_R16G16B16A16_SNORM; + return &emit_R16G16B16A16_SNORM; case PIPE_FORMAT_R16_SSCALED: - return emit_R16_SSCALED; + return &emit_R16_SSCALED; case PIPE_FORMAT_R16G16_SSCALED: - return emit_R16G16_SSCALED; + return &emit_R16G16_SSCALED; case PIPE_FORMAT_R16G16B16_SSCALED: - return emit_R16G16B16_SSCALED; + return &emit_R16G16B16_SSCALED; case PIPE_FORMAT_R16G16B16A16_SSCALED: - return emit_R16G16B16A16_SSCALED; + return &emit_R16G16B16A16_SSCALED; case PIPE_FORMAT_R8_UNORM: - return emit_R8_UNORM; + return &emit_R8_UNORM; case PIPE_FORMAT_R8G8_UNORM: - return emit_R8G8_UNORM; + return &emit_R8G8_UNORM; case PIPE_FORMAT_R8G8B8_UNORM: - return emit_R8G8B8_UNORM; + return &emit_R8G8B8_UNORM; case PIPE_FORMAT_R8G8B8A8_UNORM: - return emit_R8G8B8A8_UNORM; + return &emit_R8G8B8A8_UNORM; case PIPE_FORMAT_R8_USCALED: - return emit_R8_USCALED; + return &emit_R8_USCALED; case PIPE_FORMAT_R8G8_USCALED: - return emit_R8G8_USCALED; + return &emit_R8G8_USCALED; case PIPE_FORMAT_R8G8B8_USCALED: - return emit_R8G8B8_USCALED; + return &emit_R8G8B8_USCALED; case PIPE_FORMAT_R8G8B8A8_USCALED: - return emit_R8G8B8A8_USCALED; + return &emit_R8G8B8A8_USCALED; case PIPE_FORMAT_R8_SNORM: - return emit_R8_SNORM; + return &emit_R8_SNORM; case PIPE_FORMAT_R8G8_SNORM: - return emit_R8G8_SNORM; + return &emit_R8G8_SNORM; case PIPE_FORMAT_R8G8B8_SNORM: - return emit_R8G8B8_SNORM; + return &emit_R8G8B8_SNORM; case PIPE_FORMAT_R8G8B8A8_SNORM: - return emit_R8G8B8A8_SNORM; + return &emit_R8G8B8A8_SNORM; case PIPE_FORMAT_R8_SSCALED: - return emit_R8_SSCALED; + return &emit_R8_SSCALED; case PIPE_FORMAT_R8G8_SSCALED: - return emit_R8G8_SSCALED; + return &emit_R8G8_SSCALED; case PIPE_FORMAT_R8G8B8_SSCALED: - return emit_R8G8B8_SSCALED; + return &emit_R8G8B8_SSCALED; case PIPE_FORMAT_R8G8B8A8_SSCALED: - return emit_R8G8B8A8_SSCALED; + return &emit_R8G8B8A8_SSCALED; case PIPE_FORMAT_A8R8G8B8_UNORM: - return emit_A8R8G8B8_UNORM; + return &emit_A8R8G8B8_UNORM; case PIPE_FORMAT_B8G8R8A8_UNORM: - return emit_B8G8R8A8_UNORM; + return &emit_B8G8R8A8_UNORM; default: assert(0); - return emit_NULL; + return &emit_NULL; } } diff --git a/src/gallium/auxiliary/translate/translate_sse.c b/src/gallium/auxiliary/translate/translate_sse.c index 2fc8b9d..af86260 100644 --- a/src/gallium/auxiliary/translate/translate_sse.c +++ b/src/gallium/auxiliary/translate/translate_sse.c @@ -26,6 +26,7 @@ */ +#include "pipe/p_config.h" #include "pipe/p_compiler.h" #include "pipe/p_util.h" #include "util/u_simple_list.h" @@ -33,7 +34,7 @@ #include "translate.h" -#if defined(__i386__) || defined(__386__) || defined(i386) +#if defined(PIPE_ARCH_X86) #include "rtasm/rtasm_cpu.h" #include "rtasm/rtasm_x86sse.h" @@ -605,7 +606,7 @@ struct translate *translate_sse2_create( const struct translate_key *key ) #else -void translate_create_sse( const struct translate_key *key ) +struct translate *translate_sse2_create( const struct translate_key *key ) { return NULL; } diff --git a/src/gallium/auxiliary/util/SConscript b/src/gallium/auxiliary/util/SConscript index d55d2c7..0309de1 100644 --- a/src/gallium/auxiliary/util/SConscript +++ b/src/gallium/auxiliary/util/SConscript @@ -5,6 +5,7 @@ util = env.ConvenienceLibrary( source = [ 'p_debug.c', 'p_debug_mem.c', + 'p_debug_prof.c', 'p_tile.c', 'p_util.c', 'u_blit.c', diff --git a/src/gallium/auxiliary/util/p_debug.c b/src/gallium/auxiliary/util/p_debug.c index 4ec1746..ce7fb58 100644 --- a/src/gallium/auxiliary/util/p_debug.c +++ b/src/gallium/auxiliary/util/p_debug.c @@ -67,7 +67,7 @@ void _debug_vprintf(const char *format, va_list ap) static char buf[512 + 1] = {'\0'}; size_t len = strlen(buf); int ret = util_vsnprintf(buf + len, sizeof(buf) - len, format, ap); - if(ret > (int)(sizeof(buf) - len - 1) || strchr(buf + len, '\n')) { + if(ret > (int)(sizeof(buf) - len - 1) || util_strchr(buf + len, '\n')) { _EngDebugPrint("%s", buf); buf[0] = '\0'; } @@ -154,6 +154,7 @@ debug_get_option(const char *name, const char *dfault) { const char *result; #ifdef PIPE_SUBSYSTEM_WINDOWS_DISPLAY +#ifdef DEBUG ULONG_PTR iFile = 0; const void *pMap = NULL; const char *sol, *eol, *sep; @@ -184,6 +185,9 @@ debug_get_option(const char *name, const char *dfault) EngUnmapFile(iFile); } #else + result = dfault; +#endif +#else result = getenv(name); if(!result) @@ -203,15 +207,15 @@ debug_get_bool_option(const char *name, boolean dfault) if(str == NULL) result = dfault; - else if(!strcmp(str, "n")) + else if(!util_strcmp(str, "n")) result = FALSE; - else if(!strcmp(str, "no")) + else if(!util_strcmp(str, "no")) result = FALSE; - else if(!strcmp(str, "0")) + else if(!util_strcmp(str, "0")) result = FALSE; - else if(!strcmp(str, "f")) + else if(!util_strcmp(str, "f")) result = FALSE; - else if(!strcmp(str, "false")) + else if(!util_strcmp(str, "false")) result = FALSE; else result = TRUE; @@ -244,7 +248,7 @@ debug_get_flags_option(const char *name, else { result = 0; while( flags->name ) { - if (!strcmp(str, "all") || strstr(str, flags->name )) + if (!util_strcmp(str, "all") || util_strstr(str, flags->name )) result |= flags->value; ++flags; } @@ -299,10 +303,10 @@ debug_dump_flags(const struct debug_named_value *names, while(names->name) { if((names->value & value) == names->value) { if (!first) - strncat(output, "|", sizeof(output)); + util_strncat(output, "|", sizeof(output)); else first = 0; - strncat(output, names->name, sizeof(output)); + util_strncat(output, names->name, sizeof(output)); value &= ~names->value; } ++names; @@ -310,12 +314,12 @@ debug_dump_flags(const struct debug_named_value *names, if (value) { if (!first) - strncat(output, "|", sizeof(output)); + util_strncat(output, "|", sizeof(output)); else first = 0; util_snprintf(rest, sizeof(rest), "0x%08lx", value); - strncat(output, rest, sizeof(output)); + util_strncat(output, rest, sizeof(output)); } if(first) @@ -325,101 +329,159 @@ debug_dump_flags(const struct debug_named_value *names, } +static const struct debug_named_value pipe_format_names[] = { +#ifdef DEBUG + DEBUG_NAMED_VALUE(PIPE_FORMAT_NONE), + DEBUG_NAMED_VALUE(PIPE_FORMAT_A8R8G8B8_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_X8R8G8B8_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_B8G8R8A8_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_B8G8R8X8_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_A1R5G5B5_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_A4R4G4B4_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R5G6B5_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_L8_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_A8_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_I8_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_A8L8_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_YCBCR), + DEBUG_NAMED_VALUE(PIPE_FORMAT_YCBCR_REV), + DEBUG_NAMED_VALUE(PIPE_FORMAT_Z16_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_Z32_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_Z32_FLOAT), + DEBUG_NAMED_VALUE(PIPE_FORMAT_S8Z24_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_Z24S8_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_X8Z24_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_Z24X8_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_S8_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R64_FLOAT), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R64G64_FLOAT), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R64G64B64_FLOAT), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R64G64B64A64_FLOAT), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R32_FLOAT), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32_FLOAT), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32B32_FLOAT), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32B32A32_FLOAT), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R32_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32B32_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32B32A32_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R32_USCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32_USCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32B32_USCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32B32A32_USCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R32_SNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32_SNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32B32_SNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32B32A32_SNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R32_SSCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32_SSCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32B32_SSCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32B32A32_SSCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R16_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R16G16_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R16G16B16_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R16G16B16A16_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R16_USCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R16G16_USCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R16G16B16_USCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R16G16B16A16_USCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R16_SNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R16G16_SNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R16G16B16_SNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R16G16B16A16_SNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R16_SSCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R16G16_SSCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R16G16B16_SSCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R16G16B16A16_SSCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R8_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8A8_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8X8_UNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R8_USCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8_USCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8_USCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8A8_USCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8X8_USCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R8_SNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8_SNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8_SNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8A8_SNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8X8_SNORM), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R8_SSCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8_SSCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8_SSCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8A8_SSCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8X8_SSCALED), + DEBUG_NAMED_VALUE(PIPE_FORMAT_L8_SRGB), + DEBUG_NAMED_VALUE(PIPE_FORMAT_A8_L8_SRGB), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8_SRGB), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8A8_SRGB), + DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8X8_SRGB), + DEBUG_NAMED_VALUE(PIPE_FORMAT_DXT1_RGB), + DEBUG_NAMED_VALUE(PIPE_FORMAT_DXT1_RGBA), + DEBUG_NAMED_VALUE(PIPE_FORMAT_DXT3_RGBA), + DEBUG_NAMED_VALUE(PIPE_FORMAT_DXT5_RGBA), +#endif + DEBUG_NAMED_VALUE_END +}; - - +#ifdef DEBUG +void debug_print_format(const char *msg, unsigned fmt ) +{ + debug_printf("%s: %s\n", msg, debug_dump_enum(pipe_format_names, fmt)); +} +#endif char *pf_sprint_name( char *str, enum pipe_format format ) { - strcpy( str, "PIPE_FORMAT_" ); - switch (pf_layout( format )) { - case PIPE_FORMAT_LAYOUT_RGBAZS: - { - pipe_format_rgbazs_t rgbazs = (pipe_format_rgbazs_t) format; - uint i; - uint scale = 1 << (pf_exp8( rgbazs ) * 3); - - for (i = 0; i < 4; i++) { - uint size = pf_size_xyzw( rgbazs, i ); - - if (size == 0) { - break; - } - switch (pf_swizzle_xyzw( rgbazs, i )) { - case PIPE_FORMAT_COMP_R: - strcat( str, "R" ); - break; - case PIPE_FORMAT_COMP_G: - strcat( str, "G" ); - break; - case PIPE_FORMAT_COMP_B: - strcat( str, "B" ); - break; - case PIPE_FORMAT_COMP_A: - strcat( str, "A" ); - break; - case PIPE_FORMAT_COMP_0: - strcat( str, "0" ); - break; - case PIPE_FORMAT_COMP_1: - strcat( str, "1" ); - break; - case PIPE_FORMAT_COMP_Z: - strcat( str, "Z" ); - break; - case PIPE_FORMAT_COMP_S: - strcat( str, "S" ); - break; - } - util_snprintf( &str[strlen( str )], 32, "%u", size * scale ); - } - if (i != 0) { - strcat( str, "_" ); - } - switch (pf_type( rgbazs )) { - case PIPE_FORMAT_TYPE_UNKNOWN: - strcat( str, "NONE" ); - break; - case PIPE_FORMAT_TYPE_FLOAT: - strcat( str, "FLOAT" ); - break; - case PIPE_FORMAT_TYPE_UNORM: - strcat( str, "UNORM" ); - break; - case PIPE_FORMAT_TYPE_SNORM: - strcat( str, "SNORM" ); - break; - case PIPE_FORMAT_TYPE_USCALED: - strcat( str, "USCALED" ); - break; - case PIPE_FORMAT_TYPE_SSCALED: - strcat( str, "SSCALED" ); - break; - } - } - break; - case PIPE_FORMAT_LAYOUT_YCBCR: - { - pipe_format_ycbcr_t ycbcr = (pipe_format_ycbcr_t) format; - - strcat( str, "YCBCR" ); - if (pf_rev( ycbcr )) { - strcat( str, "_REV" ); - } - } - break; - } + strcpy( str, debug_dump_enum(pipe_format_names, format) ); return str; } #ifdef DEBUG -void debug_print_format(const char *msg, unsigned fmt ) +void debug_dump_image(const char *prefix, + unsigned format, unsigned cpp, + unsigned width, unsigned height, + unsigned pitch, + const void *data) { - char fmtstr[80]; - - pf_sprint_name(fmtstr, (enum pipe_format)fmt); +#ifdef PIPE_SUBSYSTEM_WINDOWS_DISPLAY + static unsigned no = 0; + char filename[256]; + WCHAR wfilename[sizeof(filename)]; + ULONG_PTR iFile = 0; + struct { + unsigned format; + unsigned cpp; + unsigned width; + unsigned height; + } header; + unsigned char *pMap = NULL; + unsigned i; - debug_printf("%s: %s\n", msg, fmtstr); + util_snprintf(filename, sizeof(filename), "\\??\\c:\\%03u%s.raw", ++no, prefix); + for(i = 0; i < sizeof(filename); ++i) + wfilename[i] = (WCHAR)filename[i]; + + pMap = (unsigned char *)EngMapFile(wfilename, sizeof(header) + cpp*width*height, &iFile); + if(!pMap) + return; + + header.format = format; + header.cpp = cpp; + header.width = width; + header.height = height; + memcpy(pMap, &header, sizeof(header)); + pMap += sizeof(header); + + for(i = 0; i < height; ++i) { + memcpy(pMap, (unsigned char *)data + cpp*pitch*i, cpp*width); + pMap += cpp*width; + } + + EngUnmapFile(iFile); +#endif } #endif diff --git a/src/gallium/auxiliary/util/p_debug_mem.c b/src/gallium/auxiliary/util/p_debug_mem.c index 3b5e4fb..78497c5 100644 --- a/src/gallium/auxiliary/util/p_debug_mem.c +++ b/src/gallium/auxiliary/util/p_debug_mem.c @@ -211,6 +211,7 @@ debug_memory_begin(void) void debug_memory_end(unsigned long start_no) { + size_t total_size = 0; struct list_head *entry; entry = list.prev; @@ -220,9 +221,15 @@ debug_memory_end(unsigned long start_no) hdr = LIST_ENTRY(struct debug_memory_header, entry, head); ptr = data_from_header(hdr); if(start_no <= hdr->no && hdr->no < last_no || - last_no < start_no && (hdr->no < last_no || start_no <= hdr->no)) + last_no < start_no && (hdr->no < last_no || start_no <= hdr->no)) { debug_printf("%s:%u:%s: %u bytes at %p not freed\n", hdr->file, hdr->line, hdr->function, hdr->size, ptr); + total_size += hdr->size; + } + } + if(total_size) { + debug_printf("Total of %u KB of system memory apparently leaked\n", + (total_size + 1023)/1024); } } diff --git a/src/gallium/auxiliary/util/p_debug_prof.c b/src/gallium/auxiliary/util/p_debug_prof.c new file mode 100644 index 0000000..958f99c --- /dev/null +++ b/src/gallium/auxiliary/util/p_debug_prof.c @@ -0,0 +1,175 @@ +/************************************************************************** + * + * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * @file + * Poor-man profiling. + * + * @author José Fonseca <jrfonseca@tungstengraphics.com> + * + * @sa http://blogs.msdn.com/joshpoley/archive/2008/03/12/poor-man-s-profiler.aspx + * @sa http://www.johnpanzer.com/aci_cuj/index.html + */ + +#include "pipe/p_config.h" + +#if defined(PROFILE) && defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY) + +#include <windows.h> +#include <winddi.h> + +#include "pipe/p_debug.h" +#include "util/u_string.h" + + +#define PROFILE_FILE_SIZE 4*1024*1024 +#define FILE_NAME_SIZE 256 + +static WCHAR wFileName[FILE_NAME_SIZE]; +static ULONG_PTR iFile = 0; +static void *pMap = NULL; +static void *pMapEnd = NULL; + + +/** + * Called at the start of every method or function. + * + * @sa http://msdn.microsoft.com/en-us/library/c63a9b7h.aspx + */ +void __declspec(naked) __cdecl +_penter(void) { + _asm { + push ebx + mov ebx, [pMap] + test ebx, ebx + jz done + cmp ebx, [pMapEnd] + je done + push eax + push edx + mov eax, [esp+12] + and eax, 0xfffffffe + mov [ebx], eax + add ebx, 4 + rdtsc + mov [ebx], eax + add ebx, 4 + mov [pMap], ebx + pop edx + pop eax +done: + pop ebx + ret + } +} + + +/** + * Called at the end of Calls the end of every method or function. + * + * @sa http://msdn.microsoft.com/en-us/library/xc11y76y.aspx + */ +void __declspec(naked) __cdecl +_pexit(void) { + _asm { + push ebx + mov ebx, [pMap] + test ebx, ebx + jz done + cmp ebx, [pMapEnd] + je done + push eax + push edx + mov eax, [esp+12] + or eax, 0x00000001 + mov [ebx], eax + add ebx, 4 + rdtsc + mov [ebx], eax + add ebx, 4 + mov [pMap], ebx + pop edx + pop eax +done: + pop ebx + ret + } +} + + +void __declspec(naked) +__debug_profile_reference1(void) { + _asm { + call _penter + call _pexit + ret + } +} + + +void __declspec(naked) +__debug_profile_reference2(void) { + _asm { + call _penter + call __debug_profile_reference1 + call _pexit + ret + } +} + + +void +debug_profile_start(void) +{ + static unsigned no = 0; + char filename[FILE_NAME_SIZE]; + unsigned i; + + util_snprintf(filename, sizeof(filename), "\\??\\c:\\%03u.prof", ++no); + for(i = 0; i < FILE_NAME_SIZE; ++i) + wFileName[i] = (WCHAR)filename[i]; + + pMap = EngMapFile(wFileName, PROFILE_FILE_SIZE, &iFile); + if(pMap) { + pMapEnd = (unsigned char*)pMap + PROFILE_FILE_SIZE; + /* reference functions for calibration purposes */ + __debug_profile_reference2(); + } +} + +void +debug_profile_stop(void) +{ + if(iFile) { + EngUnmapFile(iFile); + /* TODO: truncate file */ + } + iFile = 0; + pMapEnd = pMap = NULL; +} + +#endif /* PROFILE */ diff --git a/src/gallium/auxiliary/util/p_tile.c b/src/gallium/auxiliary/util/p_tile.c index 63e1cc6..5728757 100644 --- a/src/gallium/auxiliary/util/p_tile.c +++ b/src/gallium/auxiliary/util/p_tile.c @@ -50,6 +50,7 @@ pipe_get_tile_raw(struct pipe_context *pipe, uint x, uint y, uint w, uint h, void *p, int dst_stride) { + struct pipe_screen *screen = pipe->screen; const uint cpp = ps->cpp; const ubyte *pSrc; const uint src_stride = ps->pitch * cpp; @@ -63,7 +64,11 @@ pipe_get_tile_raw(struct pipe_context *pipe, if (pipe_clip_tile(x, y, &w, &h, ps)) return; - pSrc = (const ubyte *) pipe_surface_map(ps) + (y * ps->pitch + x) * cpp; + pSrc = (const ubyte *) screen->surface_map(screen, ps, + PIPE_BUFFER_USAGE_CPU_READ); + assert(pSrc); /* XXX: proper error handling! */ + + pSrc += (y * ps->pitch + x) * cpp; pDest = (ubyte *) p; for (i = 0; i < h; i++) { @@ -72,7 +77,7 @@ pipe_get_tile_raw(struct pipe_context *pipe, pSrc += src_stride; } - pipe_surface_unmap(ps); + screen->surface_unmap(screen, ps); } @@ -86,6 +91,7 @@ pipe_put_tile_raw(struct pipe_context *pipe, uint x, uint y, uint w, uint h, const void *p, int src_stride) { + struct pipe_screen *screen = pipe->screen; const uint cpp = ps->cpp; const ubyte *pSrc; const uint dst_stride = ps->pitch * cpp; @@ -100,7 +106,11 @@ pipe_put_tile_raw(struct pipe_context *pipe, return; pSrc = (const ubyte *) p; - pDest = (ubyte *) pipe_surface_map(ps) + (y * ps->pitch + x) * cpp; + + pDest = screen->surface_map(screen, ps, PIPE_BUFFER_USAGE_CPU_WRITE); + assert(pDest); /* XXX: proper error handling */ + + pDest += (y * ps->pitch + x) * cpp; for (i = 0; i < h; i++) { memcpy(pDest, pSrc, w * cpp); @@ -108,7 +118,7 @@ pipe_put_tile_raw(struct pipe_context *pipe, pSrc += src_stride; } - pipe_surface_unmap(ps); + screen->surface_unmap(screen, ps); } @@ -834,18 +844,26 @@ pipe_get_tile_z(struct pipe_context *pipe, uint x, uint y, uint w, uint h, uint *z) { + struct pipe_screen *screen = pipe->screen; const uint dstStride = w; + void *map; uint *pDest = z; uint i, j; if (pipe_clip_tile(x, y, &w, &h, ps)) return; + map = screen->surface_map(screen, ps, PIPE_BUFFER_USAGE_CPU_READ); + if (!map) { + assert(0); + return; + } + switch (ps->format) { case PIPE_FORMAT_Z32_UNORM: { const uint *pSrc - = (const uint *) pipe_surface_map(ps) + (y * ps->pitch + x); + = (const uint *)map + (y * ps->pitch + x); for (i = 0; i < h; i++) { memcpy(pDest, pSrc, 4 * w); pDest += dstStride; @@ -857,7 +875,7 @@ pipe_get_tile_z(struct pipe_context *pipe, case PIPE_FORMAT_X8Z24_UNORM: { const uint *pSrc - = (const uint *) pipe_surface_map(ps) + (y * ps->pitch + x); + = (const uint *)map + (y * ps->pitch + x); for (i = 0; i < h; i++) { for (j = 0; j < w; j++) { /* convert 24-bit Z to 32-bit Z */ @@ -871,7 +889,7 @@ pipe_get_tile_z(struct pipe_context *pipe, case PIPE_FORMAT_Z16_UNORM: { const ushort *pSrc - = (const ushort *) pipe_surface_map(ps) + (y * ps->pitch + x); + = (const ushort *)map + (y * ps->pitch + x); for (i = 0; i < h; i++) { for (j = 0; j < w; j++) { /* convert 16-bit Z to 32-bit Z */ @@ -886,7 +904,7 @@ pipe_get_tile_z(struct pipe_context *pipe, assert(0); } - pipe_surface_unmap(ps); + screen->surface_unmap(screen, ps); } @@ -896,17 +914,25 @@ pipe_put_tile_z(struct pipe_context *pipe, uint x, uint y, uint w, uint h, const uint *zSrc) { + struct pipe_screen *screen = pipe->screen; const uint srcStride = w; const uint *pSrc = zSrc; + void *map; uint i, j; if (pipe_clip_tile(x, y, &w, &h, ps)) return; + map = screen->surface_map(screen, ps, PIPE_BUFFER_USAGE_CPU_WRITE); + if (!map) { + assert(0); + return; + } + switch (ps->format) { case PIPE_FORMAT_Z32_UNORM: { - uint *pDest = (uint *) pipe_surface_map(ps) + (y * ps->pitch + x); + uint *pDest = (uint *) map + (y * ps->pitch + x); for (i = 0; i < h; i++) { memcpy(pDest, pSrc, 4 * w); pDest += ps->pitch; @@ -917,7 +943,7 @@ pipe_put_tile_z(struct pipe_context *pipe, case PIPE_FORMAT_S8Z24_UNORM: case PIPE_FORMAT_X8Z24_UNORM: { - uint *pDest = (uint *) pipe_surface_map(ps) + (y * ps->pitch + x); + uint *pDest = (uint *) map + (y * ps->pitch + x); for (i = 0; i < h; i++) { for (j = 0; j < w; j++) { /* convert 32-bit Z to 24-bit Z (0 stencil) */ @@ -930,7 +956,7 @@ pipe_put_tile_z(struct pipe_context *pipe, break; case PIPE_FORMAT_Z16_UNORM: { - ushort *pDest = (ushort *) pipe_surface_map(ps) + (y * ps->pitch + x); + ushort *pDest = (ushort *) map + (y * ps->pitch + x); for (i = 0; i < h; i++) { for (j = 0; j < w; j++) { /* convert 32-bit Z to 16-bit Z */ @@ -945,7 +971,7 @@ pipe_put_tile_z(struct pipe_context *pipe, assert(0); } - pipe_surface_unmap(ps); + screen->surface_unmap(screen, ps); } diff --git a/src/gallium/auxiliary/util/u_blit.c b/src/gallium/auxiliary/util/u_blit.c index 568d62c..999a3e5 100644 --- a/src/gallium/auxiliary/util/u_blit.c +++ b/src/gallium/auxiliary/util/u_blit.c @@ -287,7 +287,8 @@ util_blit_pixels(struct blit_state *ctx, if (!tex) return; - texSurf = screen->get_tex_surface(screen, tex, 0, 0, 0); + texSurf = screen->get_tex_surface(screen, tex, 0, 0, 0, + PIPE_BUFFER_USAGE_GPU_WRITE); /* load temp texture */ pipe->surface_copy(pipe, FALSE, @@ -295,7 +296,9 @@ util_blit_pixels(struct blit_state *ctx, src, srcLeft, srcTop, /* src */ srcW, srcH); /* size */ - pipe->texture_update(pipe, tex, 0, 1 << 0); + /* free the surface, update the texture if necessary. + */ + screen->tex_surface_release(screen, &texSurf); /* save state (restored below) */ cso_save_blend(ctx->cso); @@ -356,8 +359,6 @@ util_blit_pixels(struct blit_state *ctx, cso_restore_vertex_shader(ctx->cso); cso_restore_viewport(ctx->cso); - /* free the texture */ - pipe_surface_reference(&texSurf, NULL); screen->texture_release(screen, &tex); } diff --git a/src/gallium/auxiliary/util/u_draw_quad.c b/src/gallium/auxiliary/util/u_draw_quad.c index e659edb..bf143815 100644 --- a/src/gallium/auxiliary/util/u_draw_quad.c +++ b/src/gallium/auxiliary/util/u_draw_quad.c @@ -82,52 +82,51 @@ util_draw_texquad(struct pipe_context *pipe, { struct pipe_buffer *vbuf; uint numAttribs = 2, vertexBytes, i, j; - float *v; vertexBytes = 4 * (4 * numAttribs * sizeof(float)); /* XXX create one-time */ vbuf = pipe->winsys->buffer_create(pipe->winsys, 32, PIPE_BUFFER_USAGE_VERTEX, vertexBytes); - assert(vbuf); - - v = (float *) pipe->winsys->buffer_map(pipe->winsys, vbuf, - PIPE_BUFFER_USAGE_CPU_WRITE); - - /* - * Load vertex buffer - */ - for (i = j = 0; i < 4; i++) { - v[j + 2] = z; /* z */ - v[j + 3] = 1.0; /* w */ - v[j + 6] = 0.0; /* r */ - v[j + 7] = 1.0; /* q */ - j += 8; + if (vbuf) { + float *v = (float *) pipe->winsys->buffer_map(pipe->winsys, vbuf, + PIPE_BUFFER_USAGE_CPU_WRITE); + if (v) { + /* + * Load vertex buffer + */ + for (i = j = 0; i < 4; i++) { + v[j + 2] = z; /* z */ + v[j + 3] = 1.0; /* w */ + v[j + 6] = 0.0; /* r */ + v[j + 7] = 1.0; /* q */ + j += 8; + } + + v[0] = x0; + v[1] = y0; + v[4] = 0.0; /*s*/ + v[5] = 0.0; /*t*/ + + v[8] = x1; + v[9] = y0; + v[12] = 1.0; + v[13] = 0.0; + + v[16] = x1; + v[17] = y1; + v[20] = 1.0; + v[21] = 1.0; + + v[24] = x0; + v[25] = y1; + v[28] = 0.0; + v[29] = 1.0; + + pipe->winsys->buffer_unmap(pipe->winsys, vbuf); + util_draw_vertex_buffer(pipe, vbuf, PIPE_PRIM_TRIANGLE_FAN, 4, 2); + } + + pipe_buffer_reference(pipe->winsys, &vbuf, NULL); } - - v[0] = x0; - v[1] = y0; - v[4] = 0.0; /*s*/ - v[5] = 0.0; /*t*/ - - v[8] = x1; - v[9] = y0; - v[12] = 1.0; - v[13] = 0.0; - - v[16] = x1; - v[17] = y1; - v[20] = 1.0; - v[21] = 1.0; - - v[24] = x0; - v[25] = y1; - v[28] = 0.0; - v[29] = 1.0; - - pipe->winsys->buffer_unmap(pipe->winsys, vbuf); - - util_draw_vertex_buffer(pipe, vbuf, PIPE_PRIM_TRIANGLE_FAN, 4, 2); - - pipe_buffer_reference(pipe->winsys, &vbuf, NULL); } diff --git a/src/gallium/auxiliary/util/u_gen_mipmap.c b/src/gallium/auxiliary/util/u_gen_mipmap.c index c53c512..7d71aef 100644 --- a/src/gallium/auxiliary/util/u_gen_mipmap.c +++ b/src/gallium/auxiliary/util/u_gen_mipmap.c @@ -589,8 +589,11 @@ make_1d_mipmap(struct gen_mipmap_state *ctx, struct pipe_surface *srcSurf, *dstSurf; void *srcMap, *dstMap; - srcSurf = screen->get_tex_surface(screen, pt, face, srcLevel, zslice); - dstSurf = screen->get_tex_surface(screen, pt, face, dstLevel, zslice); + srcSurf = screen->get_tex_surface(screen, pt, face, srcLevel, zslice, + PIPE_BUFFER_USAGE_CPU_READ); + + dstSurf = screen->get_tex_surface(screen, pt, face, dstLevel, zslice, + PIPE_BUFFER_USAGE_CPU_WRITE); srcMap = ((ubyte *) winsys->buffer_map(winsys, srcSurf->buffer, PIPE_BUFFER_USAGE_CPU_READ) @@ -629,8 +632,10 @@ make_2d_mipmap(struct gen_mipmap_state *ctx, struct pipe_surface *srcSurf, *dstSurf; ubyte *srcMap, *dstMap; - srcSurf = screen->get_tex_surface(screen, pt, face, srcLevel, zslice); - dstSurf = screen->get_tex_surface(screen, pt, face, dstLevel, zslice); + srcSurf = screen->get_tex_surface(screen, pt, face, srcLevel, zslice, + PIPE_BUFFER_USAGE_CPU_READ); + dstSurf = screen->get_tex_surface(screen, pt, face, dstLevel, zslice, + PIPE_BUFFER_USAGE_CPU_WRITE); srcMap = ((ubyte *) winsys->buffer_map(winsys, srcSurf->buffer, PIPE_BUFFER_USAGE_CPU_READ) @@ -891,10 +896,14 @@ util_gen_mipmap(struct gen_mipmap_state *ctx, for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) { const uint srcLevel = dstLevel - 1; + struct pipe_surface *surf = + screen->get_tex_surface(screen, pt, face, dstLevel, zslice, + PIPE_BUFFER_USAGE_GPU_WRITE); + /* * Setup framebuffer / dest surface */ - fb.cbufs[0] = screen->get_tex_surface(screen, pt, face, dstLevel, zslice); + fb.cbufs[0] = surf; fb.width = pt->width[dstLevel]; fb.height = pt->height[dstLevel]; cso_set_framebuffer(ctx->cso, &fb); @@ -925,7 +934,7 @@ util_gen_mipmap(struct gen_mipmap_state *ctx, pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, NULL); /* need to signal that the texture has changed _after_ rendering to it */ - pipe->texture_update(pipe, pt, face, (1 << dstLevel)); + pipe_surface_reference( &surf, NULL ); } /* restore state we changed */ diff --git a/src/gallium/auxiliary/util/u_hash_table.c b/src/gallium/auxiliary/util/u_hash_table.c index f3f16a8..dd5eca7 100644 --- a/src/gallium/auxiliary/util/u_hash_table.c +++ b/src/gallium/auxiliary/util/u_hash_table.c @@ -67,6 +67,13 @@ struct hash_table_item }; +static INLINE struct hash_table_item * +hash_table_item(struct cso_hash_iter iter) +{ + return (struct hash_table_item *)cso_hash_iter_data(iter); +} + + struct hash_table * hash_table_create(unsigned (*hash)(void *key), int (*compare)(void *key1, void *key2)) @@ -90,7 +97,27 @@ hash_table_create(unsigned (*hash)(void *key), } -static struct hash_table_item * +static INLINE struct cso_hash_iter +hash_table_find_iter(struct hash_table *ht, + void *key, + unsigned key_hash) +{ + struct cso_hash_iter iter; + struct hash_table_item *item; + + iter = cso_hash_find(ht->cso, key_hash); + while (!cso_hash_iter_is_null(iter)) { + item = (struct hash_table_item *)cso_hash_iter_data(iter); + if (!ht->compare(item->key, key)) + break; + iter = cso_hash_iter_next(iter); + } + + return iter; +} + + +static INLINE struct hash_table_item * hash_table_find_item(struct hash_table *ht, void *key, unsigned key_hash) @@ -117,6 +144,7 @@ hash_table_set(struct hash_table *ht, { unsigned key_hash; struct hash_table_item *item; + struct cso_hash_iter iter; assert(ht); @@ -136,9 +164,8 @@ hash_table_set(struct hash_table *ht, item->key = key; item->value = value; - cso_hash_insert(ht->cso, key_hash, item); - /* FIXME: there is no OOM propagation in cso_hash */ - if(0) { + iter = cso_hash_insert(ht->cso, key_hash, item); + if(cso_hash_iter_is_null(iter)) { FREE(item); return PIPE_ERROR_OUT_OF_MEMORY; } @@ -171,19 +198,39 @@ hash_table_remove(struct hash_table *ht, void *key) { unsigned key_hash; + struct cso_hash_iter iter; struct hash_table_item *item; assert(ht); key_hash = ht->hash(key); - item = hash_table_find_item(ht, key, key_hash); - if(!item) + iter = hash_table_find_iter(ht, key, key_hash); + if(cso_hash_iter_is_null(iter)) return; - /* FIXME: cso_hash_take takes the first element of the collision list - * indiscriminately, so we can not take the item down. */ - item->value = NULL; + item = hash_table_item(iter); + assert(item); + FREE(item); + + cso_hash_erase(ht->cso, iter); +} + + +void +hash_table_clear(struct hash_table *ht) +{ + struct cso_hash_iter iter; + struct hash_table_item *item; + + assert(ht); + + iter = cso_hash_first_node(ht->cso); + while (!cso_hash_iter_is_null(iter)) { + item = (struct hash_table_item *)cso_hash_take(ht->cso, cso_hash_iter_key(iter)); + FREE(item); + iter = cso_hash_first_node(ht->cso); + } } @@ -196,6 +243,8 @@ hash_table_foreach(struct hash_table *ht, struct hash_table_item *item; enum pipe_error result; + assert(ht); + iter = cso_hash_first_node(ht->cso); while (!cso_hash_iter_is_null(iter)) { item = (struct hash_table_item *)cso_hash_iter_data(iter); @@ -212,7 +261,17 @@ hash_table_foreach(struct hash_table *ht, void hash_table_destroy(struct hash_table *ht) { + struct cso_hash_iter iter; + struct hash_table_item *item; + assert(ht); + + iter = cso_hash_first_node(ht->cso); + while (!cso_hash_iter_is_null(iter)) { + item = (struct hash_table_item *)cso_hash_iter_data(iter); + FREE(item); + iter = cso_hash_iter_next(iter); + } cso_hash_delete(ht->cso); diff --git a/src/gallium/auxiliary/util/u_hash_table.h b/src/gallium/auxiliary/util/u_hash_table.h index 1583bd7..feee881 100644 --- a/src/gallium/auxiliary/util/u_hash_table.h +++ b/src/gallium/auxiliary/util/u_hash_table.h @@ -75,6 +75,10 @@ hash_table_remove(struct hash_table *ht, void *key); +void +hash_table_clear(struct hash_table *ht); + + enum pipe_error hash_table_foreach(struct hash_table *ht, enum pipe_error (*callback)(void *key, void *value, void *data), diff --git a/src/gallium/auxiliary/util/u_pack_color.h b/src/gallium/auxiliary/util/u_pack_color.h index 0b917c0..655e2c8 100644 --- a/src/gallium/auxiliary/util/u_pack_color.h +++ b/src/gallium/auxiliary/util/u_pack_color.h @@ -101,6 +101,19 @@ util_pack_color_ub(ubyte r, ubyte g, ubyte b, ubyte a, *d = ((a & 0xf0) << 8) | ((r & 0xf0) << 4) | ((g & 0xf0) << 0) | (b >> 4); } return; + case PIPE_FORMAT_A8_UNORM: + { + ubyte *d = (ubyte *) dest; + *d = a; + } + return; + case PIPE_FORMAT_L8_UNORM: + case PIPE_FORMAT_I8_UNORM: + { + ubyte *d = (ubyte *) dest; + *d = r; + } + return; case PIPE_FORMAT_R32G32B32A32_FLOAT: { float *d = (float *) dest; @@ -198,6 +211,19 @@ util_pack_color(const float rgba[4], enum pipe_format format, void *dest) *d = ((a & 0xf0) << 8) | ((r & 0xf0) << 4) | ((g & 0xf0) << 0) | (b >> 4); } return; + case PIPE_FORMAT_A8_UNORM: + { + ubyte *d = (ubyte *) dest; + *d = a; + } + return; + case PIPE_FORMAT_L8_UNORM: + case PIPE_FORMAT_I8_UNORM: + { + ubyte *d = (ubyte *) dest; + *d = r; + } + return; case PIPE_FORMAT_R32G32B32A32_FLOAT: { float *d = (float *) dest; diff --git a/src/gallium/auxiliary/util/u_string.h b/src/gallium/auxiliary/util/u_string.h index b99d4e8..73c88d8 100644 --- a/src/gallium/auxiliary/util/u_string.h +++ b/src/gallium/auxiliary/util/u_string.h @@ -41,6 +41,8 @@ #include <stddef.h> #include <stdarg.h> +#include "pipe/p_compiler.h" + #ifdef __cplusplus extern "C" { @@ -48,11 +50,129 @@ extern "C" { #ifdef WIN32 + int util_vsnprintf(char *, size_t, const char *, va_list); int util_snprintf(char *str, size_t size, const char *format, ...); + +static INLINE void +util_vsprintf(char *str, const char *format, va_list ap) +{ + util_vsnprintf(str, (size_t)-1, format, ap); +} + +static INLINE void +util_sprintf(char *str, const char *format, ...) +{ + va_list ap; + va_start(ap, format); + util_vsnprintf(str, (size_t)-1, format, ap); + va_end(ap); +} + +static INLINE char * +util_strchr(const char *s, char c) +{ + while(*s) { + if(*s == c) + return (char *)s; + ++s; + } + return NULL; +} + +static INLINE char* +util_strncat(char *dst, const char *src, size_t n) +{ + char *p = dst + strlen(dst); + const char *q = src; + size_t i; + + for (i = 0; i < n && *q != '\0'; ++i) + *p++ = *q++; + *p = '\0'; + + return dst; +} + +static INLINE int +util_strcmp(const char *s1, const char *s2) +{ + unsigned char u1, u2; + + while (1) { + u1 = (unsigned char) *s1++; + u2 = (unsigned char) *s2++; + if (u1 != u2) + return u1 - u2; + if (u1 == '\0') + return 0; + } + return 0; +} + +static INLINE int +util_strncmp(const char *s1, const char *s2, size_t n) +{ + unsigned char u1, u2; + + while (n-- > 0) { + u1 = (unsigned char) *s1++; + u2 = (unsigned char) *s2++; + if (u1 != u2) + return u1 - u2; + if (u1 == '\0') + return 0; + } + return 0; +} + +static INLINE char * +util_strstr(const char *haystack, const char *needle) +{ + const char *p = haystack; + int len = strlen(needle); + + for (; (p = util_strchr(p, *needle)) != 0; p++) { + if (util_strncmp(p, needle, len) == 0) { + return (char *)p; + } + } + return NULL; +} + +static INLINE void * +util_memmove(void *dest, const void *src, size_t n) +{ + char *p = (char *)dest; + const char *q = (const char *)src; + if (dest < src) { + while (n--) + *p++ = *q++; + } + else + { + p += n; + q += n; + while (n--) + *--p = *--q; + } + return dest; +} + + #else + #define util_vsnprintf vsnprintf #define util_snprintf snprintf +#define util_vsprintf vsprintf +#define util_sprintf sprintf +#define util_strchr strchr +#define util_strcmp strcmp +#define util_strncmp strncmp +#define util_strncat strncat +#define util_strstr strstr +#define util_memmove memmove + #endif diff --git a/src/gallium/auxiliary/util/u_time.h b/src/gallium/auxiliary/util/u_time.h index 48ec7a4..f9963ce 100644 --- a/src/gallium/auxiliary/util/u_time.h +++ b/src/gallium/auxiliary/util/u_time.h @@ -61,7 +61,7 @@ struct util_time #if defined(PIPE_OS_LINUX) struct timeval tv; #else - long long counter; + int64_t counter; #endif }; diff --git a/src/gallium/drivers/failover/fo_context.c b/src/gallium/drivers/failover/fo_context.c index cb95ba5..014a3e3 100644 --- a/src/gallium/drivers/failover/fo_context.c +++ b/src/gallium/drivers/failover/fo_context.c @@ -147,8 +147,8 @@ struct pipe_context *failover_create( struct pipe_context *hw, failover->pipe.texture_create = hw->texture_create; failover->pipe.texture_release = hw->texture_release; failover->pipe.get_tex_surface = hw->get_tex_surface; -#endif failover->pipe.texture_update = hw->texture_update; +#endif failover->pipe.flush = hw->flush; diff --git a/src/gallium/drivers/i915simple/i915_context.h b/src/gallium/drivers/i915simple/i915_context.h index 38e6a34..53fc5ed 100644 --- a/src/gallium/drivers/i915simple/i915_context.h +++ b/src/gallium/drivers/i915simple/i915_context.h @@ -1,341 +1,341 @@ - /**************************************************************************
- *
- * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
- * All Rights Reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- *
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- *
- **************************************************************************/
-
-#ifndef I915_CONTEXT_H
-#define I915_CONTEXT_H
-
-
-#include "pipe/p_context.h"
-#include "pipe/p_defines.h"
-#include "pipe/p_state.h"
-
-#include "draw/draw_vertex.h"
-
-#include "tgsi/util/tgsi_scan.h"
-
-
-#define I915_TEX_UNITS 8
-
-#define I915_DYNAMIC_MODES4 0
-#define I915_DYNAMIC_DEPTHSCALE_0 1 /* just the header */
-#define I915_DYNAMIC_DEPTHSCALE_1 2
-#define I915_DYNAMIC_IAB 3
-#define I915_DYNAMIC_BC_0 4 /* just the header */
-#define I915_DYNAMIC_BC_1 5
-#define I915_DYNAMIC_BFO_0 6
-#define I915_DYNAMIC_BFO_1 7
-#define I915_DYNAMIC_STP_0 8
-#define I915_DYNAMIC_STP_1 9
-#define I915_DYNAMIC_SC_ENA_0 10
-#define I915_DYNAMIC_SC_RECT_0 11
-#define I915_DYNAMIC_SC_RECT_1 12
-#define I915_DYNAMIC_SC_RECT_2 13
-#define I915_MAX_DYNAMIC 14
-
-
-#define I915_IMMEDIATE_S0 0
-#define I915_IMMEDIATE_S1 1
-#define I915_IMMEDIATE_S2 2
-#define I915_IMMEDIATE_S3 3
-#define I915_IMMEDIATE_S4 4
-#define I915_IMMEDIATE_S5 5
-#define I915_IMMEDIATE_S6 6
-#define I915_IMMEDIATE_S7 7
-#define I915_MAX_IMMEDIATE 8
-
-/* These must mach the order of LI0_STATE_* bits, as they will be used
- * to generate hardware packets:
- */
-#define I915_CACHE_STATIC 0
-#define I915_CACHE_DYNAMIC 1 /* handled specially */
-#define I915_CACHE_SAMPLER 2
-#define I915_CACHE_MAP 3
-#define I915_CACHE_PROGRAM 4
-#define I915_CACHE_CONSTANTS 5
-#define I915_MAX_CACHE 6
-
-#define I915_MAX_CONSTANT 32
-
-
-/** See constant_flags[] below */
-#define I915_CONSTFLAG_USER 0x1f
-
-
-/**
- * Subclass of pipe_shader_state
- */
-struct i915_fragment_shader
-{
- struct pipe_shader_state state;
-
- struct tgsi_shader_info info;
-
- uint *program;
- uint program_len;
-
- /**
- * constants introduced during translation.
- * These are placed at the end of the constant buffer and grow toward
- * the beginning (eg: slot 31, 30 29, ...)
- * User-provided constants start at 0.
- * This allows both types of constants to co-exist (until there's too many)
- * and doesn't require regenerating/changing the fragment program to
- * shuffle constants around.
- */
- uint num_constants;
- float constants[I915_MAX_CONSTANT][4];
-
- /**
- * Status of each constant
- * if I915_CONSTFLAG_PARAM, the value must be taken from the corresponding
- * slot of the user's constant buffer. (set by pipe->set_constant_buffer())
- * Else, the bitmask indicates which components are occupied by immediates.
- */
- ubyte constant_flags[I915_MAX_CONSTANT];
-};
-
-
-struct i915_cache_context;
-
-/* Use to calculate differences between state emitted to hardware and
- * current driver-calculated state.
- */
-struct i915_state
-{
- unsigned immediate[I915_MAX_IMMEDIATE];
- unsigned dynamic[I915_MAX_DYNAMIC];
-
- float constants[PIPE_SHADER_TYPES][I915_MAX_CONSTANT][4];
- /** number of constants passed in through a constant buffer */
- uint num_user_constants[PIPE_SHADER_TYPES];
-
- /* texture sampler state */
- unsigned sampler[I915_TEX_UNITS][3];
- unsigned sampler_enable_flags;
- unsigned sampler_enable_nr;
-
- /* texture image buffers */
- unsigned texbuffer[I915_TEX_UNITS][2];
-
- /** Describes the current hardware vertex layout */
- struct vertex_info vertex_info;
-
- unsigned id; /* track lost context events */
-};
-
-struct i915_blend_state {
- unsigned iab;
- unsigned modes4;
- unsigned LIS5;
- unsigned LIS6;
-};
-
-struct i915_depth_stencil_state {
- unsigned stencil_modes4;
- unsigned bfo[2];
- unsigned stencil_LIS5;
- unsigned depth_LIS6;
-};
-
-struct i915_rasterizer_state {
- int light_twoside : 1;
- unsigned st;
- enum interp_mode color_interp;
-
- unsigned LIS4;
- unsigned LIS7;
- unsigned sc[1];
-
- const struct pipe_rasterizer_state *templ;
-
- union { float f; unsigned u; } ds[2];
-};
-
-struct i915_sampler_state {
- unsigned state[3];
- const struct pipe_sampler_state *templ;
-};
-
-
-struct i915_texture {
- struct pipe_texture base;
-
- /* Derived from the above:
- */
- unsigned pitch;
- unsigned depth_pitch; /* per-image on i945? */
- unsigned total_height;
-
- unsigned nr_images[PIPE_MAX_TEXTURE_LEVELS];
-
- /* Explicitly store the offset of each image for each cube face or
- * depth value. Pretty much have to accept that hardware formats
- * are going to be so diverse that there is no unified way to
- * compute the offsets of depth/cube images within a mipmap level,
- * so have to store them as a lookup table:
- */
- unsigned *image_offset[PIPE_MAX_TEXTURE_LEVELS]; /**< array [depth] of offsets */
-
- /* Includes image offset tables:
- */
- unsigned level_offset[PIPE_MAX_TEXTURE_LEVELS];
-
- /* The data is held here:
- */
- struct pipe_buffer *buffer;
-};
-
-struct i915_context
-{
- struct pipe_context pipe;
- struct i915_winsys *winsys;
- struct draw_context *draw;
-
- /* The most recent drawing state as set by the driver:
- */
- const struct i915_blend_state *blend;
- const struct i915_sampler_state *sampler[PIPE_MAX_SAMPLERS];
- const struct i915_depth_stencil_state *depth_stencil;
- const struct i915_rasterizer_state *rasterizer;
-
- struct i915_fragment_shader *fs;
-
- struct pipe_blend_color blend_color;
- struct pipe_clip_state clip;
- struct pipe_constant_buffer constants[PIPE_SHADER_TYPES];
- struct pipe_framebuffer_state framebuffer;
- struct pipe_poly_stipple poly_stipple;
- struct pipe_scissor_state scissor;
- struct i915_texture *texture[PIPE_MAX_SAMPLERS];
- struct pipe_viewport_state viewport;
- struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS];
-
- unsigned dirty;
-
- unsigned num_samplers;
- unsigned num_textures;
- unsigned num_vertex_elements;
- unsigned num_vertex_buffers;
-
- unsigned *batch_start;
-
- /** Vertex buffer */
- struct pipe_buffer *vbo;
-
- struct i915_state current;
- unsigned hardware_dirty;
-
- unsigned debug;
-};
-
-/* A flag for each state_tracker state object:
- */
-#define I915_NEW_VIEWPORT 0x1
-#define I915_NEW_RASTERIZER 0x2
-#define I915_NEW_FS 0x4
-#define I915_NEW_BLEND 0x8
-#define I915_NEW_CLIP 0x10
-#define I915_NEW_SCISSOR 0x20
-#define I915_NEW_STIPPLE 0x40
-#define I915_NEW_FRAMEBUFFER 0x80
-#define I915_NEW_ALPHA_TEST 0x100
-#define I915_NEW_DEPTH_STENCIL 0x200
-#define I915_NEW_SAMPLER 0x400
-#define I915_NEW_TEXTURE 0x800
-#define I915_NEW_CONSTANTS 0x1000
-#define I915_NEW_VBO 0x2000
-#define I915_NEW_VS 0x4000
-
-
-/* Driver's internally generated state flags:
- */
-#define I915_NEW_VERTEX_FORMAT 0x10000
-
-
-/* Dirty flags for hardware emit
- */
-#define I915_HW_STATIC (1<<I915_CACHE_STATIC)
-#define I915_HW_DYNAMIC (1<<I915_CACHE_DYNAMIC)
-#define I915_HW_SAMPLER (1<<I915_CACHE_SAMPLER)
-#define I915_HW_MAP (1<<I915_CACHE_MAP)
-#define I915_HW_PROGRAM (1<<I915_CACHE_PROGRAM)
-#define I915_HW_CONSTANTS (1<<I915_CACHE_CONSTANTS)
-#define I915_HW_IMMEDIATE (1<<(I915_MAX_CACHE+0))
-#define I915_HW_INVARIENT (1<<(I915_MAX_CACHE+1))
-
-
-/***********************************************************************
- * i915_prim_emit.c:
- */
-struct draw_stage *i915_draw_render_stage( struct i915_context *i915 );
-
-
-/***********************************************************************
- * i915_prim_vbuf.c:
- */
-struct draw_stage *i915_draw_vbuf_stage( struct i915_context *i915 );
-
-
-/***********************************************************************
- * i915_state_emit.c:
- */
-void i915_emit_hardware_state(struct i915_context *i915 );
-
-
-
-/***********************************************************************
- * i915_clear.c:
- */
-void i915_clear(struct pipe_context *pipe, struct pipe_surface *ps,
- unsigned clearValue);
-
-
-/***********************************************************************
- * i915_surface.c:
- */
-void i915_init_surface_functions( struct i915_context *i915 );
-
-void i915_init_state_functions( struct i915_context *i915 );
-void i915_init_flush_functions( struct i915_context *i915 );
-void i915_init_string_functions( struct i915_context *i915 );
-
-
-
-
-/***********************************************************************
- * Inline conversion functions. These are better-typed than the
- * macros used previously:
- */
-static INLINE struct i915_context *
-i915_context( struct pipe_context *pipe )
-{
- return (struct i915_context *)pipe;
-}
-
-
-
-#endif
+ /************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef I915_CONTEXT_H +#define I915_CONTEXT_H + + +#include "pipe/p_context.h" +#include "pipe/p_defines.h" +#include "pipe/p_state.h" + +#include "draw/draw_vertex.h" + +#include "tgsi/util/tgsi_scan.h" + + +#define I915_TEX_UNITS 8 + +#define I915_DYNAMIC_MODES4 0 +#define I915_DYNAMIC_DEPTHSCALE_0 1 /* just the header */ +#define I915_DYNAMIC_DEPTHSCALE_1 2 +#define I915_DYNAMIC_IAB 3 +#define I915_DYNAMIC_BC_0 4 /* just the header */ +#define I915_DYNAMIC_BC_1 5 +#define I915_DYNAMIC_BFO_0 6 +#define I915_DYNAMIC_BFO_1 7 +#define I915_DYNAMIC_STP_0 8 +#define I915_DYNAMIC_STP_1 9 +#define I915_DYNAMIC_SC_ENA_0 10 +#define I915_DYNAMIC_SC_RECT_0 11 +#define I915_DYNAMIC_SC_RECT_1 12 +#define I915_DYNAMIC_SC_RECT_2 13 +#define I915_MAX_DYNAMIC 14 + + +#define I915_IMMEDIATE_S0 0 +#define I915_IMMEDIATE_S1 1 +#define I915_IMMEDIATE_S2 2 +#define I915_IMMEDIATE_S3 3 +#define I915_IMMEDIATE_S4 4 +#define I915_IMMEDIATE_S5 5 +#define I915_IMMEDIATE_S6 6 +#define I915_IMMEDIATE_S7 7 +#define I915_MAX_IMMEDIATE 8 + +/* These must mach the order of LI0_STATE_* bits, as they will be used + * to generate hardware packets: + */ +#define I915_CACHE_STATIC 0 +#define I915_CACHE_DYNAMIC 1 /* handled specially */ +#define I915_CACHE_SAMPLER 2 +#define I915_CACHE_MAP 3 +#define I915_CACHE_PROGRAM 4 +#define I915_CACHE_CONSTANTS 5 +#define I915_MAX_CACHE 6 + +#define I915_MAX_CONSTANT 32 + + +/** See constant_flags[] below */ +#define I915_CONSTFLAG_USER 0x1f + + +/** + * Subclass of pipe_shader_state + */ +struct i915_fragment_shader +{ + struct pipe_shader_state state; + + struct tgsi_shader_info info; + + uint *program; + uint program_len; + + /** + * constants introduced during translation. + * These are placed at the end of the constant buffer and grow toward + * the beginning (eg: slot 31, 30 29, ...) + * User-provided constants start at 0. + * This allows both types of constants to co-exist (until there's too many) + * and doesn't require regenerating/changing the fragment program to + * shuffle constants around. + */ + uint num_constants; + float constants[I915_MAX_CONSTANT][4]; + + /** + * Status of each constant + * if I915_CONSTFLAG_PARAM, the value must be taken from the corresponding + * slot of the user's constant buffer. (set by pipe->set_constant_buffer()) + * Else, the bitmask indicates which components are occupied by immediates. + */ + ubyte constant_flags[I915_MAX_CONSTANT]; +}; + + +struct i915_cache_context; + +/* Use to calculate differences between state emitted to hardware and + * current driver-calculated state. + */ +struct i915_state +{ + unsigned immediate[I915_MAX_IMMEDIATE]; + unsigned dynamic[I915_MAX_DYNAMIC]; + + float constants[PIPE_SHADER_TYPES][I915_MAX_CONSTANT][4]; + /** number of constants passed in through a constant buffer */ + uint num_user_constants[PIPE_SHADER_TYPES]; + + /* texture sampler state */ + unsigned sampler[I915_TEX_UNITS][3]; + unsigned sampler_enable_flags; + unsigned sampler_enable_nr; + + /* texture image buffers */ + unsigned texbuffer[I915_TEX_UNITS][2]; + + /** Describes the current hardware vertex layout */ + struct vertex_info vertex_info; + + unsigned id; /* track lost context events */ +}; + +struct i915_blend_state { + unsigned iab; + unsigned modes4; + unsigned LIS5; + unsigned LIS6; +}; + +struct i915_depth_stencil_state { + unsigned stencil_modes4; + unsigned bfo[2]; + unsigned stencil_LIS5; + unsigned depth_LIS6; +}; + +struct i915_rasterizer_state { + int light_twoside : 1; + unsigned st; + enum interp_mode color_interp; + + unsigned LIS4; + unsigned LIS7; + unsigned sc[1]; + + const struct pipe_rasterizer_state *templ; + + union { float f; unsigned u; } ds[2]; +}; + +struct i915_sampler_state { + unsigned state[3]; + const struct pipe_sampler_state *templ; +}; + + +struct i915_texture { + struct pipe_texture base; + + /* Derived from the above: + */ + unsigned pitch; + unsigned depth_pitch; /* per-image on i945? */ + unsigned total_height; + + unsigned nr_images[PIPE_MAX_TEXTURE_LEVELS]; + + /* Explicitly store the offset of each image for each cube face or + * depth value. Pretty much have to accept that hardware formats + * are going to be so diverse that there is no unified way to + * compute the offsets of depth/cube images within a mipmap level, + * so have to store them as a lookup table: + */ + unsigned *image_offset[PIPE_MAX_TEXTURE_LEVELS]; /**< array [depth] of offsets */ + + /* Includes image offset tables: + */ + unsigned level_offset[PIPE_MAX_TEXTURE_LEVELS]; + + /* The data is held here: + */ + struct pipe_buffer *buffer; +}; + +struct i915_context +{ + struct pipe_context pipe; + struct i915_winsys *winsys; + struct draw_context *draw; + + /* The most recent drawing state as set by the driver: + */ + const struct i915_blend_state *blend; + const struct i915_sampler_state *sampler[PIPE_MAX_SAMPLERS]; + const struct i915_depth_stencil_state *depth_stencil; + const struct i915_rasterizer_state *rasterizer; + + struct i915_fragment_shader *fs; + + struct pipe_blend_color blend_color; + struct pipe_clip_state clip; + struct pipe_constant_buffer constants[PIPE_SHADER_TYPES]; + struct pipe_framebuffer_state framebuffer; + struct pipe_poly_stipple poly_stipple; + struct pipe_scissor_state scissor; + struct i915_texture *texture[PIPE_MAX_SAMPLERS]; + struct pipe_viewport_state viewport; + struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS]; + + unsigned dirty; + + unsigned num_samplers; + unsigned num_textures; + unsigned num_vertex_elements; + unsigned num_vertex_buffers; + + unsigned *batch_start; + + /** Vertex buffer */ + struct pipe_buffer *vbo; + + struct i915_state current; + unsigned hardware_dirty; + + unsigned debug; +}; + +/* A flag for each state_tracker state object: + */ +#define I915_NEW_VIEWPORT 0x1 +#define I915_NEW_RASTERIZER 0x2 +#define I915_NEW_FS 0x4 +#define I915_NEW_BLEND 0x8 +#define I915_NEW_CLIP 0x10 +#define I915_NEW_SCISSOR 0x20 +#define I915_NEW_STIPPLE 0x40 +#define I915_NEW_FRAMEBUFFER 0x80 +#define I915_NEW_ALPHA_TEST 0x100 +#define I915_NEW_DEPTH_STENCIL 0x200 +#define I915_NEW_SAMPLER 0x400 +#define I915_NEW_TEXTURE 0x800 +#define I915_NEW_CONSTANTS 0x1000 +#define I915_NEW_VBO 0x2000 +#define I915_NEW_VS 0x4000 + + +/* Driver's internally generated state flags: + */ +#define I915_NEW_VERTEX_FORMAT 0x10000 + + +/* Dirty flags for hardware emit + */ +#define I915_HW_STATIC (1<<I915_CACHE_STATIC) +#define I915_HW_DYNAMIC (1<<I915_CACHE_DYNAMIC) +#define I915_HW_SAMPLER (1<<I915_CACHE_SAMPLER) +#define I915_HW_MAP (1<<I915_CACHE_MAP) +#define I915_HW_PROGRAM (1<<I915_CACHE_PROGRAM) +#define I915_HW_CONSTANTS (1<<I915_CACHE_CONSTANTS) +#define I915_HW_IMMEDIATE (1<<(I915_MAX_CACHE+0)) +#define I915_HW_INVARIENT (1<<(I915_MAX_CACHE+1)) + + +/*********************************************************************** + * i915_prim_emit.c: + */ +struct draw_stage *i915_draw_render_stage( struct i915_context *i915 ); + + +/*********************************************************************** + * i915_prim_vbuf.c: + */ +struct draw_stage *i915_draw_vbuf_stage( struct i915_context *i915 ); + + +/*********************************************************************** + * i915_state_emit.c: + */ +void i915_emit_hardware_state(struct i915_context *i915 ); + + + +/*********************************************************************** + * i915_clear.c: + */ +void i915_clear(struct pipe_context *pipe, struct pipe_surface *ps, + unsigned clearValue); + + +/*********************************************************************** + * i915_surface.c: + */ +void i915_init_surface_functions( struct i915_context *i915 ); + +void i915_init_state_functions( struct i915_context *i915 ); +void i915_init_flush_functions( struct i915_context *i915 ); +void i915_init_string_functions( struct i915_context *i915 ); + + + + +/*********************************************************************** + * Inline conversion functions. These are better-typed than the + * macros used previously: + */ +static INLINE struct i915_context * +i915_context( struct pipe_context *pipe ) +{ + return (struct i915_context *)pipe; +} + + + +#endif diff --git a/src/gallium/drivers/i915simple/i915_screen.c b/src/gallium/drivers/i915simple/i915_screen.c index 646cfd9..ba8f183 100644 --- a/src/gallium/drivers/i915simple/i915_screen.c +++ b/src/gallium/drivers/i915simple/i915_screen.c @@ -201,6 +201,35 @@ i915_destroy_screen( struct pipe_screen *screen ) } +static void * +i915_surface_map( struct pipe_screen *screen, + struct pipe_surface *surface, + unsigned flags ) +{ + char *map = screen->winsys->buffer_map( screen->winsys, surface->buffer, flags ); + if (map == NULL) + return NULL; + + if (surface->texture && + (flags & PIPE_BUFFER_USAGE_CPU_WRITE)) + { + /* Do something to notify contexts of a texture change. + */ + /* i915_screen(screen)->timestamp++; */ + } + + return map + surface->offset; +} + +static void +i915_surface_unmap(struct pipe_screen *screen, + struct pipe_surface *surface) +{ + screen->winsys->buffer_unmap( screen->winsys, surface->buffer ); +} + + + /** * Create a new i915_screen object */ @@ -244,6 +273,8 @@ i915_create_screen(struct pipe_winsys *winsys, uint pci_id) i915screen->screen.get_param = i915_get_param; i915screen->screen.get_paramf = i915_get_paramf; i915screen->screen.is_format_supported = i915_is_format_supported; + i915screen->screen.surface_map = i915_surface_map; + i915screen->screen.surface_unmap = i915_surface_unmap; i915_init_screen_texture_functions(&i915screen->screen); diff --git a/src/gallium/drivers/i915simple/i915_state.c b/src/gallium/drivers/i915simple/i915_state.c index 4adeb37..e6c4671 100644 --- a/src/gallium/drivers/i915simple/i915_state.c +++ b/src/gallium/drivers/i915simple/i915_state.c @@ -466,7 +466,7 @@ void i915_delete_fs_state(struct pipe_context *pipe, void *shader) FREE(ifs->program); ifs->program_len = 0; - FREE(ifs->state.tokens); + FREE((struct tgsi_token *)ifs->state.tokens); FREE(ifs); } @@ -717,10 +717,16 @@ static void i915_set_vertex_elements(struct pipe_context *pipe, } +static void i915_set_edgeflags(struct pipe_context *pipe, + const unsigned *bitfield) +{ + /* TODO do something here */ +} void i915_init_state_functions( struct i915_context *i915 ) { + i915->pipe.set_edgeflags = i915_set_edgeflags; i915->pipe.create_blend_state = i915_create_blend_state; i915->pipe.bind_blend_state = i915_bind_blend_state; i915->pipe.delete_blend_state = i915_delete_blend_state; diff --git a/src/gallium/drivers/i915simple/i915_surface.c b/src/gallium/drivers/i915simple/i915_surface.c index f4fbedb..98367ac 100644 --- a/src/gallium/drivers/i915simple/i915_surface.c +++ b/src/gallium/drivers/i915simple/i915_surface.c @@ -51,17 +51,25 @@ i915_surface_copy(struct pipe_context *pipe, assert( dst->cpp == src->cpp ); if (0) { - pipe_copy_rect(pipe_surface_map(dst), + void *dst_map = pipe->screen->surface_map( pipe->screen, + dst, + PIPE_BUFFER_USAGE_CPU_WRITE ); + + const void *src_map = pipe->screen->surface_map( pipe->screen, + src, + PIPE_BUFFER_USAGE_CPU_READ ); + + pipe_copy_rect(dst_map, dst->cpp, dst->pitch, dstx, dsty, width, height, - pipe_surface_map(src), + src_map, do_flip ? -(int) src->pitch : src->pitch, srcx, do_flip ? 1 - srcy - height : srcy); - pipe_surface_unmap(src); - pipe_surface_unmap(dst); + pipe->screen->surface_unmap(pipe->screen, src); + pipe->screen->surface_unmap(pipe->screen, dst); } else { i915_copy_blit( i915_context(pipe), @@ -92,7 +100,10 @@ i915_surface_fill(struct pipe_context *pipe, { if (0) { unsigned i, j; - void *dst_map = pipe_surface_map(dst); + void *dst_map = pipe->screen->surface_map( pipe->screen, + dst, + PIPE_BUFFER_USAGE_CPU_WRITE ); + switch (dst->cpp) { case 1: { @@ -126,7 +137,7 @@ i915_surface_fill(struct pipe_context *pipe, break; } - pipe_surface_unmap( dst ); + pipe->screen->surface_unmap(pipe->screen, dst); } else { i915_fill_blit( i915_context(pipe), diff --git a/src/gallium/drivers/i915simple/i915_texture.c b/src/gallium/drivers/i915simple/i915_texture.c index c39e747..df11ba0 100644 --- a/src/gallium/drivers/i915simple/i915_texture.c +++ b/src/gallium/drivers/i915simple/i915_texture.c @@ -105,6 +105,51 @@ i915_miptree_set_image_offset(struct i915_texture *tex, } +/* Hack it up to use the old winsys->surface_alloc_storage() + * method for now: + */ +static boolean +i915_displaytarget_layout(struct pipe_screen *screen, + struct i915_texture *tex) +{ + struct pipe_winsys *ws = screen->winsys; + struct pipe_surface surf; + unsigned flags = (PIPE_BUFFER_USAGE_CPU_READ | + PIPE_BUFFER_USAGE_CPU_WRITE | + PIPE_BUFFER_USAGE_GPU_READ | + PIPE_BUFFER_USAGE_GPU_WRITE); + + + memset(&surf, 0, sizeof(surf)); + + ws->surface_alloc_storage( ws, + &surf, + tex->base.width[0], + tex->base.height[0], + tex->base.format, + flags, + tex->base.tex_usage); + + /* Now extract the goodies: + */ + i915_miptree_set_level_info( tex, 0, 1, 0, 0, + tex->base.width[0], + tex->base.height[0], + 1 ); + i915_miptree_set_image_offset( tex, 0, 0, 0, 0 ); + + tex->buffer = surf.buffer; + tex->pitch = surf.pitch; + tex->total_height = 0; + + + return tex->buffer != NULL; +} + + + + + static void i945_miptree_layout_2d( struct i915_texture *tex ) { @@ -480,39 +525,54 @@ i945_miptree_layout(struct i915_texture * tex) static struct pipe_texture * -i915_texture_create_screen(struct pipe_screen *screen, - const struct pipe_texture *templat) +i915_texture_create(struct pipe_screen *screen, + const struct pipe_texture *templat) { + struct i915_screen *i915screen = i915_screen(screen); + struct pipe_winsys *ws = screen->winsys; struct i915_texture *tex = CALLOC_STRUCT(i915_texture); - if (tex) { - struct i915_screen *i915screen = i915_screen(screen); - struct pipe_winsys *ws = screen->winsys; - - tex->base = *templat; - tex->base.refcount = 1; - tex->base.screen = screen; + if (!tex) + return NULL; - if (i915screen->is_i945 ? i945_miptree_layout(tex) : - i915_miptree_layout(tex)) - tex->buffer = ws->buffer_create(ws, 64, - PIPE_BUFFER_USAGE_PIXEL, - tex->pitch * tex->base.cpp * - tex->total_height); + tex->base = *templat; + tex->base.refcount = 1; + tex->base.screen = screen; - if (!tex->buffer) { - FREE(tex); - return NULL; + if (tex->base.tex_usage & PIPE_TEXTURE_USAGE_DISPLAY_TARGET) { + if (!i915_displaytarget_layout(screen, tex)) + goto fail; + } + else { + if (i915screen->is_i945) { + if (!i945_miptree_layout(tex)) + goto fail; } + else { + if (!i915_miptree_layout(tex)) + goto fail; + } + + tex->buffer = ws->buffer_create(ws, 64, + PIPE_BUFFER_USAGE_PIXEL, + tex->pitch * tex->base.cpp * + tex->total_height); + + if (!tex->buffer) + goto fail; } return &tex->base; + + fail: + FREE(tex); + return NULL; } static void -i915_texture_release_screen(struct pipe_screen *screen, - struct pipe_texture **pt) +i915_texture_release(struct pipe_screen *screen, + struct pipe_texture **pt) { if (!*pt) return; @@ -541,21 +601,15 @@ i915_texture_release_screen(struct pipe_screen *screen, } -static void -i915_texture_update(struct pipe_context *pipe, struct pipe_texture *texture, - uint face, uint levelsMask) -{ - /* no-op? */ -} - /* * XXX note: same as code in sp_surface.c */ static struct pipe_surface * -i915_get_tex_surface_screen(struct pipe_screen *screen, - struct pipe_texture *pt, - unsigned face, unsigned level, unsigned zslice) +i915_get_tex_surface(struct pipe_screen *screen, + struct pipe_texture *pt, + unsigned face, unsigned level, unsigned zslice, + unsigned flags) { struct i915_texture *tex = (struct i915_texture *)pt; struct pipe_winsys *ws = screen->winsys; @@ -579,6 +633,7 @@ i915_get_tex_surface_screen(struct pipe_screen *screen, if (ps) { assert(ps->refcount); assert(ps->winsys); + pipe_texture_reference(&ps->texture, pt); pipe_buffer_reference(ws, &ps->buffer, tex->buffer); ps->format = pt->format; ps->cpp = pt->cpp; @@ -586,6 +641,7 @@ i915_get_tex_surface_screen(struct pipe_screen *screen, ps->height = pt->height[level]; ps->pitch = tex->pitch; ps->offset = offset; + ps->usage = flags; } return ps; } @@ -594,14 +650,38 @@ i915_get_tex_surface_screen(struct pipe_screen *screen, void i915_init_texture_functions(struct i915_context *i915) { - i915->pipe.texture_update = i915_texture_update; +// i915->pipe.texture_update = i915_texture_update; } +static void +i915_tex_surface_release(struct pipe_screen *screen, + struct pipe_surface **surface) +{ + struct pipe_surface *surf = *surface; + + if (--surf->refcount == 0) { + + /* This really should not be possible, but it's actually + * happening quite a bit... Will fix. + */ + if (surf->status == PIPE_SURFACE_STATUS_CLEAR) { + debug_printf("XXX destroying a surface with pending clears...\n"); + assert(0); + } + + pipe_texture_reference(&surf->texture, NULL); + pipe_buffer_reference(screen->winsys, &surf->buffer, NULL); + FREE(surf); + } + + *surface = NULL; +} void i915_init_screen_texture_functions(struct pipe_screen *screen) { - screen->texture_create = i915_texture_create_screen; - screen->texture_release = i915_texture_release_screen; - screen->get_tex_surface = i915_get_tex_surface_screen; + screen->texture_create = i915_texture_create; + screen->texture_release = i915_texture_release; + screen->get_tex_surface = i915_get_tex_surface; + screen->tex_surface_release = i915_tex_surface_release; } diff --git a/src/gallium/drivers/i965simple/brw_context.h b/src/gallium/drivers/i965simple/brw_context.h index eeccf36..8ac6b4e 100644 --- a/src/gallium/drivers/i965simple/brw_context.h +++ b/src/gallium/drivers/i965simple/brw_context.h @@ -1,684 +1,684 @@ -/*
- Copyright (C) Intel Corp. 2006. All Rights Reserved.
- Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
- develop this 3D driver.
-
- Permission is hereby granted, free of charge, to any person obtaining
- a copy of this software and associated documentation files (the
- "Software"), to deal in the Software without restriction, including
- without limitation the rights to use, copy, modify, merge, publish,
- distribute, sublicense, and/or sell copies of the Software, and to
- permit persons to whom the Software is furnished to do so, subject to
- the following conditions:
-
- The above copyright notice and this permission notice (including the
- next paragraph) shall be included in all copies or substantial
- portions of the Software.
-
- THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
- EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
- IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
- LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
- OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
- WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
- **********************************************************************/
- /*
- * Authors:
- * Keith Whitwell <keith@tungstengraphics.com>
- */
-
-
-#ifndef BRWCONTEXT_INC
-#define BRWCONTEXT_INC
-
-
-#include "pipe/p_context.h"
-#include "pipe/p_defines.h"
-#include "pipe/p_state.h"
-
-#include "tgsi/util/tgsi_scan.h"
-
-#include "brw_structs.h"
-#include "brw_winsys.h"
-
-
-/* Glossary:
- *
- * URB - uniform resource buffer. A mid-sized buffer which is
- * partitioned between the fixed function units and used for passing
- * values (vertices, primitives, constants) between them.
- *
- * CURBE - constant URB entry. An urb region (entry) used to hold
- * constant values which the fixed function units can be instructed to
- * preload into the GRF when spawining a thread.
- *
- * VUE - vertex URB entry. An urb entry holding a vertex and usually
- * a vertex header. The header contains control information and
- * things like primitive type, Begin/end flags and clip codes.
- *
- * PUE - primitive URB entry. An urb entry produced by the setup (SF)
- * unit holding rasterization and interpolation parameters.
- *
- * GRF - general register file. One of several register files
- * addressable by programmed threads. The inputs (r0, payload, curbe,
- * urb) of the thread are preloaded to this area before the thread is
- * spawned. The registers are individually 8 dwords wide and suitable
- * for general usage. Registers holding thread input values are not
- * special and may be overwritten.
- *
- * MRF - message register file. Threads communicate (and terminate)
- * by sending messages. Message parameters are placed in contigous
- * MRF registers. All program output is via these messages. URB
- * entries are populated by sending a message to the shared URB
- * function containing the new data, together with a control word,
- * often an unmodified copy of R0.
- *
- * R0 - GRF register 0. Typically holds control information used when
- * sending messages to other threads.
- *
- * EU or GEN4 EU: The name of the programmable subsystem of the
- * i965 hardware. Threads are executed by the EU, the registers
- * described above are part of the EU architecture.
- *
- * Fixed function units:
- *
- * CS - Command streamer. Notional first unit, little software
- * interaction. Holds the URB entries used for constant data, ie the
- * CURBEs.
- *
- * VF/VS - Vertex Fetch / Vertex Shader. The fixed function part of
- * this unit is responsible for pulling vertices out of vertex buffers
- * in vram and injecting them into the processing pipe as VUEs. If
- * enabled, it first passes them to a VS thread which is a good place
- * for the driver to implement any active vertex shader.
- *
- * GS - Geometry Shader. This corresponds to a new DX10 concept. If
- * enabled, incoming strips etc are passed to GS threads in individual
- * line/triangle/point units. The GS thread may perform arbitary
- * computation and emit whatever primtives with whatever vertices it
- * chooses. This makes GS an excellent place to implement GL's
- * unfilled polygon modes, though of course it is capable of much
- * more. Additionally, GS is used to translate away primitives not
- * handled by latter units, including Quads and Lineloops.
- *
- * CS - Clipper. Mesa's clipping algorithms are imported to run on
- * this unit. The fixed function part performs cliptesting against
- * the 6 fixed clipplanes and makes descisions on whether or not the
- * incoming primitive needs to be passed to a thread for clipping.
- * User clip planes are handled via cooperation with the VS thread.
- *
- * SF - Strips Fans or Setup: Triangles are prepared for
- * rasterization. Interpolation coefficients are calculated.
- * Flatshading and two-side lighting usually performed here.
- *
- * WM - Windower. Interpolation of vertex attributes performed here.
- * Fragment shader implemented here. SIMD aspects of EU taken full
- * advantage of, as pixels are processed in blocks of 16.
- *
- * CC - Color Calculator. No EU threads associated with this unit.
- * Handles blending and (presumably) depth and stencil testing.
- */
-
-#define BRW_MAX_CURBE (32*16)
-
-struct brw_context;
-struct brw_winsys;
-
-
-/* Raised when we receive new state across the pipe interface:
- */
-#define BRW_NEW_VIEWPORT 0x1
-#define BRW_NEW_RASTERIZER 0x2
-#define BRW_NEW_FS 0x4
-#define BRW_NEW_BLEND 0x8
-#define BRW_NEW_CLIP 0x10
-#define BRW_NEW_SCISSOR 0x20
-#define BRW_NEW_STIPPLE 0x40
-#define BRW_NEW_FRAMEBUFFER 0x80
-#define BRW_NEW_ALPHA_TEST 0x100
-#define BRW_NEW_DEPTH_STENCIL 0x200
-#define BRW_NEW_SAMPLER 0x400
-#define BRW_NEW_TEXTURE 0x800
-#define BRW_NEW_CONSTANTS 0x1000
-#define BRW_NEW_VBO 0x2000
-#define BRW_NEW_VS 0x4000
-
-/* Raised for other internal events:
- */
-#define BRW_NEW_URB_FENCE 0x10000
-#define BRW_NEW_PSP 0x20000
-#define BRW_NEW_CURBE_OFFSETS 0x40000
-#define BRW_NEW_REDUCED_PRIMITIVE 0x80000
-#define BRW_NEW_PRIMITIVE 0x100000
-#define BRW_NEW_SCENE 0x200000
-#define BRW_NEW_SF_LINKAGE 0x400000
-
-extern int BRW_DEBUG;
-
-#define DEBUG_TEXTURE 0x1
-#define DEBUG_STATE 0x2
-#define DEBUG_IOCTL 0x4
-#define DEBUG_PRIMS 0x8
-#define DEBUG_VERTS 0x10
-#define DEBUG_FALLBACKS 0x20
-#define DEBUG_VERBOSE 0x40
-#define DEBUG_DRI 0x80
-#define DEBUG_DMA 0x100
-#define DEBUG_SANITY 0x200
-#define DEBUG_SYNC 0x400
-#define DEBUG_SLEEP 0x800
-#define DEBUG_PIXEL 0x1000
-#define DEBUG_STATS 0x2000
-#define DEBUG_TILE 0x4000
-#define DEBUG_SINGLE_THREAD 0x8000
-#define DEBUG_WM 0x10000
-#define DEBUG_URB 0x20000
-#define DEBUG_VS 0x40000
-#define DEBUG_BATCH 0x80000
-#define DEBUG_BUFMGR 0x100000
-#define DEBUG_BLIT 0x200000
-#define DEBUG_REGION 0x400000
-#define DEBUG_MIPTREE 0x800000
-
-#define DBG(...) do { \
- if (BRW_DEBUG & FILE_DEBUG_FLAG) \
- debug_printf(__VA_ARGS__); \
-} while(0)
-
-#define PRINT(...) do { \
- debug_printf(brw->pipe.winsys, __VA_ARGS__); \
-} while(0)
-
-struct brw_state_flags {
- unsigned cache;
- unsigned brw;
-};
-
-
-struct brw_vertex_program {
- struct pipe_shader_state program;
- struct tgsi_shader_info info;
- int id;
-};
-
-
-struct brw_fragment_program {
- struct pipe_shader_state program;
- struct tgsi_shader_info info;
-
- boolean UsesDepth; /* XXX add this to tgsi_shader_info? */
- int id;
-};
-
-
-struct pipe_setup_linkage {
- struct {
- unsigned vp_output:5;
- unsigned interp_mode:4;
- unsigned bf_vp_output:5;
- } fp_input[PIPE_MAX_SHADER_INPUTS];
-
- unsigned fp_input_count:5;
- unsigned max_vp_output:5;
-};
-
-
-
-struct brw_texture {
- struct pipe_texture base;
-
- /* Derived from the above:
- */
- unsigned pitch;
- unsigned depth_pitch; /* per-image on i945? */
- unsigned total_height;
-
- unsigned nr_images[PIPE_MAX_TEXTURE_LEVELS];
-
- /* Explicitly store the offset of each image for each cube face or
- * depth value. Pretty much have to accept that hardware formats
- * are going to be so diverse that there is no unified way to
- * compute the offsets of depth/cube images within a mipmap level,
- * so have to store them as a lookup table:
- */
- unsigned *image_offset[PIPE_MAX_TEXTURE_LEVELS]; /**< array [depth] of offsets */
-
- /* Includes image offset tables:
- */
- unsigned level_offset[PIPE_MAX_TEXTURE_LEVELS];
-
- /* The data is held here:
- */
- struct pipe_buffer *buffer;
-};
-
-/* Data about a particular attempt to compile a program. Note that
- * there can be many of these, each in a different GL state
- * corresponding to a different brw_wm_prog_key struct, with different
- * compiled programs:
- */
-/* Data about a particular attempt to compile a program. Note that
- * there can be many of these, each in a different GL state
- * corresponding to a different brw_wm_prog_key struct, with different
- * compiled programs:
- */
-
-struct brw_wm_prog_data {
- unsigned curb_read_length;
- unsigned urb_read_length;
-
- unsigned first_curbe_grf;
- unsigned total_grf;
- unsigned total_scratch;
-
- /* Internally generated constants for the CURBE. These are loaded
- * ahead of the data from the constant buffer.
- */
- const float internal_const[8];
- unsigned nr_internal_consts;
- unsigned max_const;
-
- boolean error;
-};
-
-struct brw_sf_prog_data {
- unsigned urb_read_length;
- unsigned total_grf;
-
- /* Each vertex may have upto 12 attributes, 4 components each,
- * except WPOS which requires only 2. (11*4 + 2) == 44 ==> 11
- * rows.
- *
- * Actually we use 4 for each, so call it 12 rows.
- */
- unsigned urb_entry_size;
-};
-
-struct brw_clip_prog_data {
- unsigned curb_read_length; /* user planes? */
- unsigned clip_mode;
- unsigned urb_read_length;
- unsigned total_grf;
-};
-
-struct brw_gs_prog_data {
- unsigned urb_read_length;
- unsigned total_grf;
-};
-
-struct brw_vs_prog_data {
- unsigned curb_read_length;
- unsigned urb_read_length;
- unsigned total_grf;
- unsigned outputs_written;
-
- unsigned inputs_read;
-
- unsigned max_const;
-
- float imm_buf[PIPE_MAX_CONSTANT][4];
- unsigned num_imm;
- unsigned num_consts;
-
- /* Used for calculating urb partitions:
- */
- unsigned urb_entry_size;
-};
-
-
-#define BRW_MAX_TEX_UNIT 8
-#define BRW_WM_MAX_SURF BRW_MAX_TEX_UNIT + 1
-
-/* Create a fixed sized struct for caching binding tables:
- */
-struct brw_surface_binding_table {
- unsigned surf_ss_offset[BRW_WM_MAX_SURF];
-};
-
-
-struct brw_cache;
-
-struct brw_mem_pool {
- struct pipe_buffer *buffer;
-
- unsigned size;
- unsigned offset; /* offset of first free byte */
-
- struct brw_context *brw;
-};
-
-struct brw_cache_item {
- unsigned hash;
- unsigned key_size; /* for variable-sized keys */
- const void *key;
-
- unsigned offset; /* offset within pool's buffer */
- unsigned data_size;
-
- struct brw_cache_item *next;
-};
-
-
-
-struct brw_cache {
- unsigned id;
-
- const char *name;
-
- struct brw_context *brw;
- struct brw_mem_pool *pool;
-
- struct brw_cache_item **items;
- unsigned size, n_items;
-
- unsigned key_size; /* for fixed-size keys */
- unsigned aux_size;
-
- unsigned last_addr; /* offset of active item */
-};
-
-
-
-
-/* Considered adding a member to this struct to document which flags
- * an update might raise so that ordering of the state atoms can be
- * checked or derived at runtime. Dropped the idea in favor of having
- * a debug mode where the state is monitored for flags which are
- * raised that have already been tested against.
- */
-struct brw_tracked_state {
- struct brw_state_flags dirty;
- void (*update)( struct brw_context *brw );
-};
-
-
-/* Flags for brw->state.cache.
- */
-#define CACHE_NEW_CC_VP (1<<BRW_CC_VP)
-#define CACHE_NEW_CC_UNIT (1<<BRW_CC_UNIT)
-#define CACHE_NEW_WM_PROG (1<<BRW_WM_PROG)
-#define CACHE_NEW_SAMPLER_DEFAULT_COLOR (1<<BRW_SAMPLER_DEFAULT_COLOR)
-#define CACHE_NEW_SAMPLER (1<<BRW_SAMPLER)
-#define CACHE_NEW_WM_UNIT (1<<BRW_WM_UNIT)
-#define CACHE_NEW_SF_PROG (1<<BRW_SF_PROG)
-#define CACHE_NEW_SF_VP (1<<BRW_SF_VP)
-#define CACHE_NEW_SF_UNIT (1<<BRW_SF_UNIT)
-#define CACHE_NEW_VS_UNIT (1<<BRW_VS_UNIT)
-#define CACHE_NEW_VS_PROG (1<<BRW_VS_PROG)
-#define CACHE_NEW_GS_UNIT (1<<BRW_GS_UNIT)
-#define CACHE_NEW_GS_PROG (1<<BRW_GS_PROG)
-#define CACHE_NEW_CLIP_VP (1<<BRW_CLIP_VP)
-#define CACHE_NEW_CLIP_UNIT (1<<BRW_CLIP_UNIT)
-#define CACHE_NEW_CLIP_PROG (1<<BRW_CLIP_PROG)
-#define CACHE_NEW_SURFACE (1<<BRW_SS_SURFACE)
-#define CACHE_NEW_SURF_BIND (1<<BRW_SS_SURF_BIND)
-
-
-
-
-enum brw_mempool_id {
- BRW_GS_POOL,
- BRW_SS_POOL,
- BRW_MAX_POOL
-};
-
-
-struct brw_cached_batch_item {
- struct header *header;
- unsigned sz;
- struct brw_cached_batch_item *next;
-};
-
-
-
-/* Protect against a future where PIPE_MAX_ATTRIBS > 32. Wouldn't life
- * be easier if C allowed arrays of packed elements?
- */
-#define ATTRIB_BIT_DWORDS ((PIPE_MAX_ATTRIBS+31)/32)
-
-
-
-
-struct brw_vertex_info {
- unsigned varying; /* varying:1[PIPE_MAX_ATTRIBS] */
- unsigned sizes[ATTRIB_BIT_DWORDS * 2]; /* sizes:2[PIPE_MAX_ATTRIBS] */
-};
-
-
-
-
-
-struct brw_context
-{
- struct pipe_context pipe;
- struct brw_winsys *winsys;
-
- unsigned primitive;
- unsigned reduced_primitive;
-
- boolean emit_state_always;
-
- struct {
- struct brw_state_flags dirty;
- } state;
-
-
- struct {
- const struct pipe_blend_state *Blend;
- const struct pipe_depth_stencil_alpha_state *DepthStencil;
- const struct pipe_poly_stipple *PolygonStipple;
- const struct pipe_rasterizer_state *Raster;
- const struct pipe_sampler_state *Samplers[PIPE_MAX_SAMPLERS];
- const struct brw_vertex_program *VertexProgram;
- const struct brw_fragment_program *FragmentProgram;
-
- struct pipe_clip_state Clip;
- struct pipe_blend_color BlendColor;
- struct pipe_scissor_state Scissor;
- struct pipe_viewport_state Viewport;
- struct pipe_framebuffer_state FrameBuffer;
-
- const struct pipe_constant_buffer *Constants[2];
- const struct brw_texture *Texture[PIPE_MAX_SAMPLERS];
- } attribs;
-
- unsigned num_samplers;
- unsigned num_textures;
-
- struct brw_mem_pool pool[BRW_MAX_POOL];
- struct brw_cache cache[BRW_MAX_CACHE];
- struct brw_cached_batch_item *cached_batch_items;
-
- struct {
-
- /* Arrays with buffer objects to copy non-bufferobj arrays into
- * for upload:
- */
- const struct pipe_vertex_buffer *vbo_array[PIPE_MAX_ATTRIBS];
-
- struct brw_vertex_element_state inputs[PIPE_MAX_ATTRIBS];
-
-#define BRW_NR_UPLOAD_BUFS 17
-#define BRW_UPLOAD_INIT_SIZE (128*1024)
-
- /* Summary of size and varying of active arrays, so we can check
- * for changes to this state:
- */
- struct brw_vertex_info info;
- } vb;
-
-
- unsigned hardware_dirty;
- unsigned dirty;
- unsigned pci_id;
- /* BRW_NEW_URB_ALLOCATIONS:
- */
- struct {
- unsigned vsize; /* vertex size plus header in urb registers */
- unsigned csize; /* constant buffer size in urb registers */
- unsigned sfsize; /* setup data size in urb registers */
-
- boolean constrained;
-
- unsigned nr_vs_entries;
- unsigned nr_gs_entries;
- unsigned nr_clip_entries;
- unsigned nr_sf_entries;
- unsigned nr_cs_entries;
-
-/* unsigned vs_size; */
-/* unsigned gs_size; */
-/* unsigned clip_size; */
-/* unsigned sf_size; */
-/* unsigned cs_size; */
-
- unsigned vs_start;
- unsigned gs_start;
- unsigned clip_start;
- unsigned sf_start;
- unsigned cs_start;
- } urb;
-
-
- /* BRW_NEW_CURBE_OFFSETS:
- */
- struct {
- unsigned wm_start;
- unsigned wm_size;
- unsigned clip_start;
- unsigned clip_size;
- unsigned vs_start;
- unsigned vs_size;
- unsigned total_size;
-
- unsigned gs_offset;
-
- float *last_buf;
- unsigned last_bufsz;
- } curbe;
-
- struct {
- struct brw_vs_prog_data *prog_data;
-
- unsigned prog_gs_offset;
- unsigned state_gs_offset;
- } vs;
-
- struct {
- struct brw_gs_prog_data *prog_data;
-
- boolean prog_active;
- unsigned prog_gs_offset;
- unsigned state_gs_offset;
- } gs;
-
- struct {
- struct brw_clip_prog_data *prog_data;
-
- unsigned prog_gs_offset;
- unsigned vp_gs_offset;
- unsigned state_gs_offset;
- } clip;
-
-
- struct {
- struct brw_sf_prog_data *prog_data;
-
- struct pipe_setup_linkage linkage;
-
- unsigned prog_gs_offset;
- unsigned vp_gs_offset;
- unsigned state_gs_offset;
- } sf;
-
- struct {
- struct brw_wm_prog_data *prog_data;
-
-// struct brw_wm_compiler *compile_data;
-
-
- /**
- * Array of sampler state uploaded at sampler_gs_offset of BRW_SAMPLER
- * cache
- */
- struct brw_sampler_state sampler[BRW_MAX_TEX_UNIT];
-
- unsigned render_surf;
- unsigned nr_surfaces;
-
- unsigned max_threads;
- struct pipe_buffer *scratch_buffer;
- unsigned scratch_buffer_size;
-
- unsigned sampler_count;
- unsigned sampler_gs_offset;
-
- struct brw_surface_binding_table bind;
- unsigned bind_ss_offset;
-
- unsigned prog_gs_offset;
- unsigned state_gs_offset;
- } wm;
-
-
- struct {
- unsigned vp_gs_offset;
- unsigned state_gs_offset;
- } cc;
-
-
- /* Used to give every program string a unique id
- */
- unsigned program_id;
-};
-
-
-#define BRW_PACKCOLOR8888(r,g,b,a) ((r<<24) | (g<<16) | (b<<8) | a)
-
-
-/*======================================================================
- * brw_vtbl.c
- */
-void brw_do_flush( struct brw_context *brw,
- unsigned flags );
-
-
-/*======================================================================
- * brw_state.c
- */
-void brw_validate_state(struct brw_context *brw);
-void brw_init_state(struct brw_context *brw);
-void brw_destroy_state(struct brw_context *brw);
-
-
-/*======================================================================
- * brw_tex.c
- */
-void brwUpdateTextureState( struct brw_context *brw );
-
-
-/* brw_urb.c
- */
-void brw_upload_urb_fence(struct brw_context *brw);
-
-void brw_upload_constant_buffer_state(struct brw_context *brw);
-
-void brw_init_surface_functions(struct brw_context *brw);
-void brw_init_state_functions(struct brw_context *brw);
-void brw_init_flush_functions(struct brw_context *brw);
-void brw_init_string_functions(struct brw_context *brw);
-
-/*======================================================================
- * Inline conversion functions. These are better-typed than the
- * macros used previously:
- */
-static inline struct brw_context *
-brw_context( struct pipe_context *ctx )
-{
- return (struct brw_context *)ctx;
-}
-
-#endif
-
+/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#ifndef BRWCONTEXT_INC +#define BRWCONTEXT_INC + + +#include "pipe/p_context.h" +#include "pipe/p_defines.h" +#include "pipe/p_state.h" + +#include "tgsi/util/tgsi_scan.h" + +#include "brw_structs.h" +#include "brw_winsys.h" + + +/* Glossary: + * + * URB - uniform resource buffer. A mid-sized buffer which is + * partitioned between the fixed function units and used for passing + * values (vertices, primitives, constants) between them. + * + * CURBE - constant URB entry. An urb region (entry) used to hold + * constant values which the fixed function units can be instructed to + * preload into the GRF when spawining a thread. + * + * VUE - vertex URB entry. An urb entry holding a vertex and usually + * a vertex header. The header contains control information and + * things like primitive type, Begin/end flags and clip codes. + * + * PUE - primitive URB entry. An urb entry produced by the setup (SF) + * unit holding rasterization and interpolation parameters. + * + * GRF - general register file. One of several register files + * addressable by programmed threads. The inputs (r0, payload, curbe, + * urb) of the thread are preloaded to this area before the thread is + * spawned. The registers are individually 8 dwords wide and suitable + * for general usage. Registers holding thread input values are not + * special and may be overwritten. + * + * MRF - message register file. Threads communicate (and terminate) + * by sending messages. Message parameters are placed in contigous + * MRF registers. All program output is via these messages. URB + * entries are populated by sending a message to the shared URB + * function containing the new data, together with a control word, + * often an unmodified copy of R0. + * + * R0 - GRF register 0. Typically holds control information used when + * sending messages to other threads. + * + * EU or GEN4 EU: The name of the programmable subsystem of the + * i965 hardware. Threads are executed by the EU, the registers + * described above are part of the EU architecture. + * + * Fixed function units: + * + * CS - Command streamer. Notional first unit, little software + * interaction. Holds the URB entries used for constant data, ie the + * CURBEs. + * + * VF/VS - Vertex Fetch / Vertex Shader. The fixed function part of + * this unit is responsible for pulling vertices out of vertex buffers + * in vram and injecting them into the processing pipe as VUEs. If + * enabled, it first passes them to a VS thread which is a good place + * for the driver to implement any active vertex shader. + * + * GS - Geometry Shader. This corresponds to a new DX10 concept. If + * enabled, incoming strips etc are passed to GS threads in individual + * line/triangle/point units. The GS thread may perform arbitary + * computation and emit whatever primtives with whatever vertices it + * chooses. This makes GS an excellent place to implement GL's + * unfilled polygon modes, though of course it is capable of much + * more. Additionally, GS is used to translate away primitives not + * handled by latter units, including Quads and Lineloops. + * + * CS - Clipper. Mesa's clipping algorithms are imported to run on + * this unit. The fixed function part performs cliptesting against + * the 6 fixed clipplanes and makes descisions on whether or not the + * incoming primitive needs to be passed to a thread for clipping. + * User clip planes are handled via cooperation with the VS thread. + * + * SF - Strips Fans or Setup: Triangles are prepared for + * rasterization. Interpolation coefficients are calculated. + * Flatshading and two-side lighting usually performed here. + * + * WM - Windower. Interpolation of vertex attributes performed here. + * Fragment shader implemented here. SIMD aspects of EU taken full + * advantage of, as pixels are processed in blocks of 16. + * + * CC - Color Calculator. No EU threads associated with this unit. + * Handles blending and (presumably) depth and stencil testing. + */ + +#define BRW_MAX_CURBE (32*16) + +struct brw_context; +struct brw_winsys; + + +/* Raised when we receive new state across the pipe interface: + */ +#define BRW_NEW_VIEWPORT 0x1 +#define BRW_NEW_RASTERIZER 0x2 +#define BRW_NEW_FS 0x4 +#define BRW_NEW_BLEND 0x8 +#define BRW_NEW_CLIP 0x10 +#define BRW_NEW_SCISSOR 0x20 +#define BRW_NEW_STIPPLE 0x40 +#define BRW_NEW_FRAMEBUFFER 0x80 +#define BRW_NEW_ALPHA_TEST 0x100 +#define BRW_NEW_DEPTH_STENCIL 0x200 +#define BRW_NEW_SAMPLER 0x400 +#define BRW_NEW_TEXTURE 0x800 +#define BRW_NEW_CONSTANTS 0x1000 +#define BRW_NEW_VBO 0x2000 +#define BRW_NEW_VS 0x4000 + +/* Raised for other internal events: + */ +#define BRW_NEW_URB_FENCE 0x10000 +#define BRW_NEW_PSP 0x20000 +#define BRW_NEW_CURBE_OFFSETS 0x40000 +#define BRW_NEW_REDUCED_PRIMITIVE 0x80000 +#define BRW_NEW_PRIMITIVE 0x100000 +#define BRW_NEW_SCENE 0x200000 +#define BRW_NEW_SF_LINKAGE 0x400000 + +extern int BRW_DEBUG; + +#define DEBUG_TEXTURE 0x1 +#define DEBUG_STATE 0x2 +#define DEBUG_IOCTL 0x4 +#define DEBUG_PRIMS 0x8 +#define DEBUG_VERTS 0x10 +#define DEBUG_FALLBACKS 0x20 +#define DEBUG_VERBOSE 0x40 +#define DEBUG_DRI 0x80 +#define DEBUG_DMA 0x100 +#define DEBUG_SANITY 0x200 +#define DEBUG_SYNC 0x400 +#define DEBUG_SLEEP 0x800 +#define DEBUG_PIXEL 0x1000 +#define DEBUG_STATS 0x2000 +#define DEBUG_TILE 0x4000 +#define DEBUG_SINGLE_THREAD 0x8000 +#define DEBUG_WM 0x10000 +#define DEBUG_URB 0x20000 +#define DEBUG_VS 0x40000 +#define DEBUG_BATCH 0x80000 +#define DEBUG_BUFMGR 0x100000 +#define DEBUG_BLIT 0x200000 +#define DEBUG_REGION 0x400000 +#define DEBUG_MIPTREE 0x800000 + +#define DBG(...) do { \ + if (BRW_DEBUG & FILE_DEBUG_FLAG) \ + debug_printf(__VA_ARGS__); \ +} while(0) + +#define PRINT(...) do { \ + debug_printf(brw->pipe.winsys, __VA_ARGS__); \ +} while(0) + +struct brw_state_flags { + unsigned cache; + unsigned brw; +}; + + +struct brw_vertex_program { + struct pipe_shader_state program; + struct tgsi_shader_info info; + int id; +}; + + +struct brw_fragment_program { + struct pipe_shader_state program; + struct tgsi_shader_info info; + + boolean UsesDepth; /* XXX add this to tgsi_shader_info? */ + int id; +}; + + +struct pipe_setup_linkage { + struct { + unsigned vp_output:5; + unsigned interp_mode:4; + unsigned bf_vp_output:5; + } fp_input[PIPE_MAX_SHADER_INPUTS]; + + unsigned fp_input_count:5; + unsigned max_vp_output:5; +}; + + + +struct brw_texture { + struct pipe_texture base; + + /* Derived from the above: + */ + unsigned pitch; + unsigned depth_pitch; /* per-image on i945? */ + unsigned total_height; + + unsigned nr_images[PIPE_MAX_TEXTURE_LEVELS]; + + /* Explicitly store the offset of each image for each cube face or + * depth value. Pretty much have to accept that hardware formats + * are going to be so diverse that there is no unified way to + * compute the offsets of depth/cube images within a mipmap level, + * so have to store them as a lookup table: + */ + unsigned *image_offset[PIPE_MAX_TEXTURE_LEVELS]; /**< array [depth] of offsets */ + + /* Includes image offset tables: + */ + unsigned level_offset[PIPE_MAX_TEXTURE_LEVELS]; + + /* The data is held here: + */ + struct pipe_buffer *buffer; +}; + +/* Data about a particular attempt to compile a program. Note that + * there can be many of these, each in a different GL state + * corresponding to a different brw_wm_prog_key struct, with different + * compiled programs: + */ +/* Data about a particular attempt to compile a program. Note that + * there can be many of these, each in a different GL state + * corresponding to a different brw_wm_prog_key struct, with different + * compiled programs: + */ + +struct brw_wm_prog_data { + unsigned curb_read_length; + unsigned urb_read_length; + + unsigned first_curbe_grf; + unsigned total_grf; + unsigned total_scratch; + + /* Internally generated constants for the CURBE. These are loaded + * ahead of the data from the constant buffer. + */ + const float internal_const[8]; + unsigned nr_internal_consts; + unsigned max_const; + + boolean error; +}; + +struct brw_sf_prog_data { + unsigned urb_read_length; + unsigned total_grf; + + /* Each vertex may have upto 12 attributes, 4 components each, + * except WPOS which requires only 2. (11*4 + 2) == 44 ==> 11 + * rows. + * + * Actually we use 4 for each, so call it 12 rows. + */ + unsigned urb_entry_size; +}; + +struct brw_clip_prog_data { + unsigned curb_read_length; /* user planes? */ + unsigned clip_mode; + unsigned urb_read_length; + unsigned total_grf; +}; + +struct brw_gs_prog_data { + unsigned urb_read_length; + unsigned total_grf; +}; + +struct brw_vs_prog_data { + unsigned curb_read_length; + unsigned urb_read_length; + unsigned total_grf; + unsigned outputs_written; + + unsigned inputs_read; + + unsigned max_const; + + float imm_buf[PIPE_MAX_CONSTANT][4]; + unsigned num_imm; + unsigned num_consts; + + /* Used for calculating urb partitions: + */ + unsigned urb_entry_size; +}; + + +#define BRW_MAX_TEX_UNIT 8 +#define BRW_WM_MAX_SURF BRW_MAX_TEX_UNIT + 1 + +/* Create a fixed sized struct for caching binding tables: + */ +struct brw_surface_binding_table { + unsigned surf_ss_offset[BRW_WM_MAX_SURF]; +}; + + +struct brw_cache; + +struct brw_mem_pool { + struct pipe_buffer *buffer; + + unsigned size; + unsigned offset; /* offset of first free byte */ + + struct brw_context *brw; +}; + +struct brw_cache_item { + unsigned hash; + unsigned key_size; /* for variable-sized keys */ + const void *key; + + unsigned offset; /* offset within pool's buffer */ + unsigned data_size; + + struct brw_cache_item *next; +}; + + + +struct brw_cache { + unsigned id; + + const char *name; + + struct brw_context *brw; + struct brw_mem_pool *pool; + + struct brw_cache_item **items; + unsigned size, n_items; + + unsigned key_size; /* for fixed-size keys */ + unsigned aux_size; + + unsigned last_addr; /* offset of active item */ +}; + + + + +/* Considered adding a member to this struct to document which flags + * an update might raise so that ordering of the state atoms can be + * checked or derived at runtime. Dropped the idea in favor of having + * a debug mode where the state is monitored for flags which are + * raised that have already been tested against. + */ +struct brw_tracked_state { + struct brw_state_flags dirty; + void (*update)( struct brw_context *brw ); +}; + + +/* Flags for brw->state.cache. + */ +#define CACHE_NEW_CC_VP (1<<BRW_CC_VP) +#define CACHE_NEW_CC_UNIT (1<<BRW_CC_UNIT) +#define CACHE_NEW_WM_PROG (1<<BRW_WM_PROG) +#define CACHE_NEW_SAMPLER_DEFAULT_COLOR (1<<BRW_SAMPLER_DEFAULT_COLOR) +#define CACHE_NEW_SAMPLER (1<<BRW_SAMPLER) +#define CACHE_NEW_WM_UNIT (1<<BRW_WM_UNIT) +#define CACHE_NEW_SF_PROG (1<<BRW_SF_PROG) +#define CACHE_NEW_SF_VP (1<<BRW_SF_VP) +#define CACHE_NEW_SF_UNIT (1<<BRW_SF_UNIT) +#define CACHE_NEW_VS_UNIT (1<<BRW_VS_UNIT) +#define CACHE_NEW_VS_PROG (1<<BRW_VS_PROG) +#define CACHE_NEW_GS_UNIT (1<<BRW_GS_UNIT) +#define CACHE_NEW_GS_PROG (1<<BRW_GS_PROG) +#define CACHE_NEW_CLIP_VP (1<<BRW_CLIP_VP) +#define CACHE_NEW_CLIP_UNIT (1<<BRW_CLIP_UNIT) +#define CACHE_NEW_CLIP_PROG (1<<BRW_CLIP_PROG) +#define CACHE_NEW_SURFACE (1<<BRW_SS_SURFACE) +#define CACHE_NEW_SURF_BIND (1<<BRW_SS_SURF_BIND) + + + + +enum brw_mempool_id { + BRW_GS_POOL, + BRW_SS_POOL, + BRW_MAX_POOL +}; + + +struct brw_cached_batch_item { + struct header *header; + unsigned sz; + struct brw_cached_batch_item *next; +}; + + + +/* Protect against a future where PIPE_MAX_ATTRIBS > 32. Wouldn't life + * be easier if C allowed arrays of packed elements? + */ +#define ATTRIB_BIT_DWORDS ((PIPE_MAX_ATTRIBS+31)/32) + + + + +struct brw_vertex_info { + unsigned varying; /* varying:1[PIPE_MAX_ATTRIBS] */ + unsigned sizes[ATTRIB_BIT_DWORDS * 2]; /* sizes:2[PIPE_MAX_ATTRIBS] */ +}; + + + + + +struct brw_context +{ + struct pipe_context pipe; + struct brw_winsys *winsys; + + unsigned primitive; + unsigned reduced_primitive; + + boolean emit_state_always; + + struct { + struct brw_state_flags dirty; + } state; + + + struct { + const struct pipe_blend_state *Blend; + const struct pipe_depth_stencil_alpha_state *DepthStencil; + const struct pipe_poly_stipple *PolygonStipple; + const struct pipe_rasterizer_state *Raster; + const struct pipe_sampler_state *Samplers[PIPE_MAX_SAMPLERS]; + const struct brw_vertex_program *VertexProgram; + const struct brw_fragment_program *FragmentProgram; + + struct pipe_clip_state Clip; + struct pipe_blend_color BlendColor; + struct pipe_scissor_state Scissor; + struct pipe_viewport_state Viewport; + struct pipe_framebuffer_state FrameBuffer; + + const struct pipe_constant_buffer *Constants[2]; + const struct brw_texture *Texture[PIPE_MAX_SAMPLERS]; + } attribs; + + unsigned num_samplers; + unsigned num_textures; + + struct brw_mem_pool pool[BRW_MAX_POOL]; + struct brw_cache cache[BRW_MAX_CACHE]; + struct brw_cached_batch_item *cached_batch_items; + + struct { + + /* Arrays with buffer objects to copy non-bufferobj arrays into + * for upload: + */ + const struct pipe_vertex_buffer *vbo_array[PIPE_MAX_ATTRIBS]; + + struct brw_vertex_element_state inputs[PIPE_MAX_ATTRIBS]; + +#define BRW_NR_UPLOAD_BUFS 17 +#define BRW_UPLOAD_INIT_SIZE (128*1024) + + /* Summary of size and varying of active arrays, so we can check + * for changes to this state: + */ + struct brw_vertex_info info; + } vb; + + + unsigned hardware_dirty; + unsigned dirty; + unsigned pci_id; + /* BRW_NEW_URB_ALLOCATIONS: + */ + struct { + unsigned vsize; /* vertex size plus header in urb registers */ + unsigned csize; /* constant buffer size in urb registers */ + unsigned sfsize; /* setup data size in urb registers */ + + boolean constrained; + + unsigned nr_vs_entries; + unsigned nr_gs_entries; + unsigned nr_clip_entries; + unsigned nr_sf_entries; + unsigned nr_cs_entries; + +/* unsigned vs_size; */ +/* unsigned gs_size; */ +/* unsigned clip_size; */ +/* unsigned sf_size; */ +/* unsigned cs_size; */ + + unsigned vs_start; + unsigned gs_start; + unsigned clip_start; + unsigned sf_start; + unsigned cs_start; + } urb; + + + /* BRW_NEW_CURBE_OFFSETS: + */ + struct { + unsigned wm_start; + unsigned wm_size; + unsigned clip_start; + unsigned clip_size; + unsigned vs_start; + unsigned vs_size; + unsigned total_size; + + unsigned gs_offset; + + float *last_buf; + unsigned last_bufsz; + } curbe; + + struct { + struct brw_vs_prog_data *prog_data; + + unsigned prog_gs_offset; + unsigned state_gs_offset; + } vs; + + struct { + struct brw_gs_prog_data *prog_data; + + boolean prog_active; + unsigned prog_gs_offset; + unsigned state_gs_offset; + } gs; + + struct { + struct brw_clip_prog_data *prog_data; + + unsigned prog_gs_offset; + unsigned vp_gs_offset; + unsigned state_gs_offset; + } clip; + + + struct { + struct brw_sf_prog_data *prog_data; + + struct pipe_setup_linkage linkage; + + unsigned prog_gs_offset; + unsigned vp_gs_offset; + unsigned state_gs_offset; + } sf; + + struct { + struct brw_wm_prog_data *prog_data; + +// struct brw_wm_compiler *compile_data; + + + /** + * Array of sampler state uploaded at sampler_gs_offset of BRW_SAMPLER + * cache + */ + struct brw_sampler_state sampler[BRW_MAX_TEX_UNIT]; + + unsigned render_surf; + unsigned nr_surfaces; + + unsigned max_threads; + struct pipe_buffer *scratch_buffer; + unsigned scratch_buffer_size; + + unsigned sampler_count; + unsigned sampler_gs_offset; + + struct brw_surface_binding_table bind; + unsigned bind_ss_offset; + + unsigned prog_gs_offset; + unsigned state_gs_offset; + } wm; + + + struct { + unsigned vp_gs_offset; + unsigned state_gs_offset; + } cc; + + + /* Used to give every program string a unique id + */ + unsigned program_id; +}; + + +#define BRW_PACKCOLOR8888(r,g,b,a) ((r<<24) | (g<<16) | (b<<8) | a) + + +/*====================================================================== + * brw_vtbl.c + */ +void brw_do_flush( struct brw_context *brw, + unsigned flags ); + + +/*====================================================================== + * brw_state.c + */ +void brw_validate_state(struct brw_context *brw); +void brw_init_state(struct brw_context *brw); +void brw_destroy_state(struct brw_context *brw); + + +/*====================================================================== + * brw_tex.c + */ +void brwUpdateTextureState( struct brw_context *brw ); + + +/* brw_urb.c + */ +void brw_upload_urb_fence(struct brw_context *brw); + +void brw_upload_constant_buffer_state(struct brw_context *brw); + +void brw_init_surface_functions(struct brw_context *brw); +void brw_init_state_functions(struct brw_context *brw); +void brw_init_flush_functions(struct brw_context *brw); +void brw_init_string_functions(struct brw_context *brw); + +/*====================================================================== + * Inline conversion functions. These are better-typed than the + * macros used previously: + */ +static inline struct brw_context * +brw_context( struct pipe_context *ctx ) +{ + return (struct brw_context *)ctx; +} + +#endif + diff --git a/src/gallium/drivers/i965simple/brw_state.c b/src/gallium/drivers/i965simple/brw_state.c index ac243b7..caeeba4 100644 --- a/src/gallium/drivers/i965simple/brw_state.c +++ b/src/gallium/drivers/i965simple/brw_state.c @@ -1,469 +1,469 @@ -/**************************************************************************
- *
- * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
- * All Rights Reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- *
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- *
- **************************************************************************/
-
-/* Authors: Zack Rusin <zack@tungstengraphics.com>
- * Keith Whitwell <keith@tungstengraphics.com>
- */
-
-
-#include "pipe/p_winsys.h"
-#include "pipe/p_util.h"
-#include "pipe/p_inlines.h"
-#include "pipe/p_shader_tokens.h"
-#include "tgsi/util/tgsi_dump.h"
-#include "tgsi/util/tgsi_parse.h"
-
-#include "brw_context.h"
-#include "brw_defines.h"
-#include "brw_state.h"
-#include "brw_draw.h"
-
-
-#define DUP( TYPE, VAL ) \
-do { \
- struct TYPE *x = malloc(sizeof(*x)); \
- memcpy(x, VAL, sizeof(*x) ); \
- return x; \
-} while (0)
-
-/************************************************************************
- * Blend
- */
-static void *
-brw_create_blend_state(struct pipe_context *pipe,
- const struct pipe_blend_state *blend)
-{
- DUP( pipe_blend_state, blend );
-}
-
-static void brw_bind_blend_state(struct pipe_context *pipe,
- void *blend)
-{
- struct brw_context *brw = brw_context(pipe);
-
- brw->attribs.Blend = (struct pipe_blend_state*)blend;
- brw->state.dirty.brw |= BRW_NEW_BLEND;
-}
-
-
-static void brw_delete_blend_state(struct pipe_context *pipe, void *blend)
-{
- free(blend);
-}
-
-static void brw_set_blend_color( struct pipe_context *pipe,
- const struct pipe_blend_color *blend_color )
-{
- struct brw_context *brw = brw_context(pipe);
-
- brw->attribs.BlendColor = *blend_color;
-
- brw->state.dirty.brw |= BRW_NEW_BLEND;
-}
-
-/************************************************************************
- * Sampler
- */
-
-static void *
-brw_create_sampler_state(struct pipe_context *pipe,
- const struct pipe_sampler_state *sampler)
-{
- DUP( pipe_sampler_state, sampler );
-}
-
-static void brw_bind_sampler_states(struct pipe_context *pipe,
- unsigned num, void **sampler)
-{
- struct brw_context *brw = brw_context(pipe);
-
- assert(num <= PIPE_MAX_SAMPLERS);
-
- /* Check for no-op */
- if (num == brw->num_samplers &&
- !memcmp(brw->attribs.Samplers, sampler, num * sizeof(void *)))
- return;
-
- memcpy(brw->attribs.Samplers, sampler, num * sizeof(void *));
- memset(&brw->attribs.Samplers[num], 0, (PIPE_MAX_SAMPLERS - num) *
- sizeof(void *));
-
- brw->num_samplers = num;
-
- brw->state.dirty.brw |= BRW_NEW_SAMPLER;
-}
-
-static void brw_delete_sampler_state(struct pipe_context *pipe,
- void *sampler)
-{
- free(sampler);
-}
-
-
-/************************************************************************
- * Depth stencil
- */
-
-static void *
-brw_create_depth_stencil_state(struct pipe_context *pipe,
- const struct pipe_depth_stencil_alpha_state *depth_stencil)
-{
- DUP( pipe_depth_stencil_alpha_state, depth_stencil );
-}
-
-static void brw_bind_depth_stencil_state(struct pipe_context *pipe,
- void *depth_stencil)
-{
- struct brw_context *brw = brw_context(pipe);
-
- brw->attribs.DepthStencil = (const struct pipe_depth_stencil_alpha_state *)depth_stencil;
-
- brw->state.dirty.brw |= BRW_NEW_DEPTH_STENCIL;
-}
-
-static void brw_delete_depth_stencil_state(struct pipe_context *pipe,
- void *depth_stencil)
-{
- free(depth_stencil);
-}
-
-/************************************************************************
- * Scissor
- */
-static void brw_set_scissor_state( struct pipe_context *pipe,
- const struct pipe_scissor_state *scissor )
-{
- struct brw_context *brw = brw_context(pipe);
-
- memcpy( &brw->attribs.Scissor, scissor, sizeof(*scissor) );
- brw->state.dirty.brw |= BRW_NEW_SCISSOR;
-}
-
-
-/************************************************************************
- * Stipple
- */
-
-static void brw_set_polygon_stipple( struct pipe_context *pipe,
- const struct pipe_poly_stipple *stipple )
-{
-}
-
-
-/************************************************************************
- * Fragment shader
- */
-
-static void * brw_create_fs_state(struct pipe_context *pipe,
- const struct pipe_shader_state *shader)
-{
- struct brw_fragment_program *brw_fp = CALLOC_STRUCT(brw_fragment_program);
-
- brw_fp->program.tokens = tgsi_dup_tokens(shader->tokens);
- brw_fp->id = brw_context(pipe)->program_id++;
-
- tgsi_scan_shader(shader->tokens, &brw_fp->info);
-
-#if 0
- brw_shader_info(shader->tokens,
- &brw_fp->info2);
-#endif
-
- tgsi_dump(shader->tokens, 0);
-
-
- return (void *)brw_fp;
-}
-
-static void brw_bind_fs_state(struct pipe_context *pipe, void *shader)
-{
- struct brw_context *brw = brw_context(pipe);
-
- brw->attribs.FragmentProgram = (struct brw_fragment_program *)shader;
- brw->state.dirty.brw |= BRW_NEW_FS;
-}
-
-static void brw_delete_fs_state(struct pipe_context *pipe, void *shader)
-{
- struct brw_fragment_program *brw_fp = (struct brw_fragment_program *) shader;
-
- FREE((void *) brw_fp->program.tokens);
- FREE(brw_fp);
-}
-
-
-/************************************************************************
- * Vertex shader and other TNL state
- */
-
-static void *brw_create_vs_state(struct pipe_context *pipe,
- const struct pipe_shader_state *shader)
-{
- struct brw_vertex_program *brw_vp = CALLOC_STRUCT(brw_vertex_program);
-
- brw_vp->program.tokens = tgsi_dup_tokens(shader->tokens);
- brw_vp->id = brw_context(pipe)->program_id++;
-
- tgsi_scan_shader(shader->tokens, &brw_vp->info);
-
-#if 0
- brw_shader_info(shader->tokens,
- &brw_vp->info2);
-#endif
- tgsi_dump(shader->tokens, 0);
-
- return (void *)brw_vp;
-}
-
-static void brw_bind_vs_state(struct pipe_context *pipe, void *vs)
-{
- struct brw_context *brw = brw_context(pipe);
-
- brw->attribs.VertexProgram = (struct brw_vertex_program *)vs;
- brw->state.dirty.brw |= BRW_NEW_VS;
-
- debug_printf("YYYYYYYYYYYYY BINDING VERTEX SHADER\n");
-}
-
-static void brw_delete_vs_state(struct pipe_context *pipe, void *shader)
-{
- struct brw_vertex_program *brw_vp = (struct brw_vertex_program *) shader;
-
- FREE((void *) brw_vp->program.tokens);
- FREE(brw_vp);
-}
-
-
-static void brw_set_clip_state( struct pipe_context *pipe,
- const struct pipe_clip_state *clip )
-{
- struct brw_context *brw = brw_context(pipe);
-
- brw->attribs.Clip = *clip;
-}
-
-
-static void brw_set_viewport_state( struct pipe_context *pipe,
- const struct pipe_viewport_state *viewport )
-{
- struct brw_context *brw = brw_context(pipe);
-
- brw->attribs.Viewport = *viewport; /* struct copy */
- brw->state.dirty.brw |= BRW_NEW_VIEWPORT;
-
- /* pass the viewport info to the draw module */
- //draw_set_viewport_state(brw->draw, viewport);
-}
-
-
-static void brw_set_vertex_buffers(struct pipe_context *pipe,
- unsigned count,
- const struct pipe_vertex_buffer *buffers)
-{
- struct brw_context *brw = brw_context(pipe);
- memcpy(brw->vb.vbo_array, buffers, count * sizeof(buffers[0]));
-}
-
-static void brw_set_vertex_elements(struct pipe_context *pipe,
- unsigned count,
- const struct pipe_vertex_element *elements)
-{
- /* flush ? */
- struct brw_context *brw = brw_context(pipe);
- uint i;
-
- assert(count <= PIPE_MAX_ATTRIBS);
-
- for (i = 0; i < count; i++) {
- struct brw_vertex_element_state el;
- memset(&el, 0, sizeof(el));
-
- el.ve0.src_offset = elements[i].src_offset;
- el.ve0.src_format = brw_translate_surface_format(elements[i].src_format);
- el.ve0.valid = 1;
- el.ve0.vertex_buffer_index = elements[i].vertex_buffer_index;
-
- el.ve1.dst_offset = i * 4;
-
- el.ve1.vfcomponent3 = BRW_VFCOMPONENT_STORE_SRC;
- el.ve1.vfcomponent2 = BRW_VFCOMPONENT_STORE_SRC;
- el.ve1.vfcomponent1 = BRW_VFCOMPONENT_STORE_SRC;
- el.ve1.vfcomponent0 = BRW_VFCOMPONENT_STORE_SRC;
-
- switch (elements[i].nr_components) {
- case 1: el.ve1.vfcomponent1 = BRW_VFCOMPONENT_STORE_0;
- case 2: el.ve1.vfcomponent2 = BRW_VFCOMPONENT_STORE_0;
- case 3: el.ve1.vfcomponent3 = BRW_VFCOMPONENT_STORE_1_FLT;
- break;
- }
-
- brw->vb.inputs[i] = el;
- }
-}
-
-
-
-/************************************************************************
- * Constant buffers
- */
-
-static void brw_set_constant_buffer(struct pipe_context *pipe,
- uint shader, uint index,
- const struct pipe_constant_buffer *buf)
-{
- struct brw_context *brw = brw_context(pipe);
-
- assert(buf == 0 || index == 0);
-
- brw->attribs.Constants[shader] = buf;
- brw->state.dirty.brw |= BRW_NEW_CONSTANTS;
-}
-
-
-/************************************************************************
- * Texture surfaces
- */
-
-
-static void brw_set_sampler_textures(struct pipe_context *pipe,
- unsigned num,
- struct pipe_texture **texture)
-{
- struct brw_context *brw = brw_context(pipe);
- uint i;
-
- assert(num <= PIPE_MAX_SAMPLERS);
-
- /* Check for no-op */
- if (num == brw->num_textures &&
- !memcmp(brw->attribs.Texture, texture, num *
- sizeof(struct pipe_texture *)))
- return;
-
- for (i = 0; i < num; i++)
- pipe_texture_reference((struct pipe_texture **) &brw->attribs.Texture[i],
- texture[i]);
-
- for (i = num; i < brw->num_textures; i++)
- pipe_texture_reference((struct pipe_texture **) &brw->attribs.Texture[i],
- NULL);
-
- brw->num_textures = num;
-
- brw->state.dirty.brw |= BRW_NEW_TEXTURE;
-}
-
-
-/************************************************************************
- * Render targets, etc
- */
-
-static void brw_set_framebuffer_state(struct pipe_context *pipe,
- const struct pipe_framebuffer_state *fb)
-{
- struct brw_context *brw = brw_context(pipe);
-
- brw->attribs.FrameBuffer = *fb; /* struct copy */
-
- brw->state.dirty.brw |= BRW_NEW_FRAMEBUFFER;
-}
-
-
-
-/************************************************************************
- * Rasterizer state
- */
-
-static void *
-brw_create_rasterizer_state(struct pipe_context *pipe,
- const struct pipe_rasterizer_state *rasterizer)
-{
- DUP(pipe_rasterizer_state, rasterizer);
-}
-
-static void brw_bind_rasterizer_state( struct pipe_context *pipe,
- void *setup )
-{
- struct brw_context *brw = brw_context(pipe);
-
- brw->attribs.Raster = (struct pipe_rasterizer_state *)setup;
-
- /* Also pass-through to draw module:
- */
- //draw_set_rasterizer_state(brw->draw, setup);
-
- brw->state.dirty.brw |= BRW_NEW_RASTERIZER;
-}
-
-static void brw_delete_rasterizer_state(struct pipe_context *pipe,
- void *setup)
-{
- free(setup);
-}
-
-
-
-void
-brw_init_state_functions( struct brw_context *brw )
-{
- brw->pipe.create_blend_state = brw_create_blend_state;
- brw->pipe.bind_blend_state = brw_bind_blend_state;
- brw->pipe.delete_blend_state = brw_delete_blend_state;
-
- brw->pipe.create_sampler_state = brw_create_sampler_state;
- brw->pipe.bind_sampler_states = brw_bind_sampler_states;
- brw->pipe.delete_sampler_state = brw_delete_sampler_state;
-
- brw->pipe.create_depth_stencil_alpha_state = brw_create_depth_stencil_state;
- brw->pipe.bind_depth_stencil_alpha_state = brw_bind_depth_stencil_state;
- brw->pipe.delete_depth_stencil_alpha_state = brw_delete_depth_stencil_state;
-
- brw->pipe.create_rasterizer_state = brw_create_rasterizer_state;
- brw->pipe.bind_rasterizer_state = brw_bind_rasterizer_state;
- brw->pipe.delete_rasterizer_state = brw_delete_rasterizer_state;
- brw->pipe.create_fs_state = brw_create_fs_state;
- brw->pipe.bind_fs_state = brw_bind_fs_state;
- brw->pipe.delete_fs_state = brw_delete_fs_state;
- brw->pipe.create_vs_state = brw_create_vs_state;
- brw->pipe.bind_vs_state = brw_bind_vs_state;
- brw->pipe.delete_vs_state = brw_delete_vs_state;
-
- brw->pipe.set_blend_color = brw_set_blend_color;
- brw->pipe.set_clip_state = brw_set_clip_state;
- brw->pipe.set_constant_buffer = brw_set_constant_buffer;
- brw->pipe.set_framebuffer_state = brw_set_framebuffer_state;
-
-// brw->pipe.set_feedback_state = brw_set_feedback_state;
-// brw->pipe.set_feedback_buffer = brw_set_feedback_buffer;
-
- brw->pipe.set_polygon_stipple = brw_set_polygon_stipple;
- brw->pipe.set_scissor_state = brw_set_scissor_state;
- brw->pipe.set_sampler_textures = brw_set_sampler_textures;
- brw->pipe.set_viewport_state = brw_set_viewport_state;
- brw->pipe.set_vertex_buffers = brw_set_vertex_buffers;
- brw->pipe.set_vertex_elements = brw_set_vertex_elements;
-}
+/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Zack Rusin <zack@tungstengraphics.com> + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "pipe/p_winsys.h" +#include "pipe/p_util.h" +#include "pipe/p_inlines.h" +#include "pipe/p_shader_tokens.h" +#include "tgsi/util/tgsi_dump.h" +#include "tgsi/util/tgsi_parse.h" + +#include "brw_context.h" +#include "brw_defines.h" +#include "brw_state.h" +#include "brw_draw.h" + + +#define DUP( TYPE, VAL ) \ +do { \ + struct TYPE *x = malloc(sizeof(*x)); \ + memcpy(x, VAL, sizeof(*x) ); \ + return x; \ +} while (0) + +/************************************************************************ + * Blend + */ +static void * +brw_create_blend_state(struct pipe_context *pipe, + const struct pipe_blend_state *blend) +{ + DUP( pipe_blend_state, blend ); +} + +static void brw_bind_blend_state(struct pipe_context *pipe, + void *blend) +{ + struct brw_context *brw = brw_context(pipe); + + brw->attribs.Blend = (struct pipe_blend_state*)blend; + brw->state.dirty.brw |= BRW_NEW_BLEND; +} + + +static void brw_delete_blend_state(struct pipe_context *pipe, void *blend) +{ + free(blend); +} + +static void brw_set_blend_color( struct pipe_context *pipe, + const struct pipe_blend_color *blend_color ) +{ + struct brw_context *brw = brw_context(pipe); + + brw->attribs.BlendColor = *blend_color; + + brw->state.dirty.brw |= BRW_NEW_BLEND; +} + +/************************************************************************ + * Sampler + */ + +static void * +brw_create_sampler_state(struct pipe_context *pipe, + const struct pipe_sampler_state *sampler) +{ + DUP( pipe_sampler_state, sampler ); +} + +static void brw_bind_sampler_states(struct pipe_context *pipe, + unsigned num, void **sampler) +{ + struct brw_context *brw = brw_context(pipe); + + assert(num <= PIPE_MAX_SAMPLERS); + + /* Check for no-op */ + if (num == brw->num_samplers && + !memcmp(brw->attribs.Samplers, sampler, num * sizeof(void *))) + return; + + memcpy(brw->attribs.Samplers, sampler, num * sizeof(void *)); + memset(&brw->attribs.Samplers[num], 0, (PIPE_MAX_SAMPLERS - num) * + sizeof(void *)); + + brw->num_samplers = num; + + brw->state.dirty.brw |= BRW_NEW_SAMPLER; +} + +static void brw_delete_sampler_state(struct pipe_context *pipe, + void *sampler) +{ + free(sampler); +} + + +/************************************************************************ + * Depth stencil + */ + +static void * +brw_create_depth_stencil_state(struct pipe_context *pipe, + const struct pipe_depth_stencil_alpha_state *depth_stencil) +{ + DUP( pipe_depth_stencil_alpha_state, depth_stencil ); +} + +static void brw_bind_depth_stencil_state(struct pipe_context *pipe, + void *depth_stencil) +{ + struct brw_context *brw = brw_context(pipe); + + brw->attribs.DepthStencil = (const struct pipe_depth_stencil_alpha_state *)depth_stencil; + + brw->state.dirty.brw |= BRW_NEW_DEPTH_STENCIL; +} + +static void brw_delete_depth_stencil_state(struct pipe_context *pipe, + void *depth_stencil) +{ + free(depth_stencil); +} + +/************************************************************************ + * Scissor + */ +static void brw_set_scissor_state( struct pipe_context *pipe, + const struct pipe_scissor_state *scissor ) +{ + struct brw_context *brw = brw_context(pipe); + + memcpy( &brw->attribs.Scissor, scissor, sizeof(*scissor) ); + brw->state.dirty.brw |= BRW_NEW_SCISSOR; +} + + +/************************************************************************ + * Stipple + */ + +static void brw_set_polygon_stipple( struct pipe_context *pipe, + const struct pipe_poly_stipple *stipple ) +{ +} + + +/************************************************************************ + * Fragment shader + */ + +static void * brw_create_fs_state(struct pipe_context *pipe, + const struct pipe_shader_state *shader) +{ + struct brw_fragment_program *brw_fp = CALLOC_STRUCT(brw_fragment_program); + + brw_fp->program.tokens = tgsi_dup_tokens(shader->tokens); + brw_fp->id = brw_context(pipe)->program_id++; + + tgsi_scan_shader(shader->tokens, &brw_fp->info); + +#if 0 + brw_shader_info(shader->tokens, + &brw_fp->info2); +#endif + + tgsi_dump(shader->tokens, 0); + + + return (void *)brw_fp; +} + +static void brw_bind_fs_state(struct pipe_context *pipe, void *shader) +{ + struct brw_context *brw = brw_context(pipe); + + brw->attribs.FragmentProgram = (struct brw_fragment_program *)shader; + brw->state.dirty.brw |= BRW_NEW_FS; +} + +static void brw_delete_fs_state(struct pipe_context *pipe, void *shader) +{ + struct brw_fragment_program *brw_fp = (struct brw_fragment_program *) shader; + + FREE((void *) brw_fp->program.tokens); + FREE(brw_fp); +} + + +/************************************************************************ + * Vertex shader and other TNL state + */ + +static void *brw_create_vs_state(struct pipe_context *pipe, + const struct pipe_shader_state *shader) +{ + struct brw_vertex_program *brw_vp = CALLOC_STRUCT(brw_vertex_program); + + brw_vp->program.tokens = tgsi_dup_tokens(shader->tokens); + brw_vp->id = brw_context(pipe)->program_id++; + + tgsi_scan_shader(shader->tokens, &brw_vp->info); + +#if 0 + brw_shader_info(shader->tokens, + &brw_vp->info2); +#endif + tgsi_dump(shader->tokens, 0); + + return (void *)brw_vp; +} + +static void brw_bind_vs_state(struct pipe_context *pipe, void *vs) +{ + struct brw_context *brw = brw_context(pipe); + + brw->attribs.VertexProgram = (struct brw_vertex_program *)vs; + brw->state.dirty.brw |= BRW_NEW_VS; + + debug_printf("YYYYYYYYYYYYY BINDING VERTEX SHADER\n"); +} + +static void brw_delete_vs_state(struct pipe_context *pipe, void *shader) +{ + struct brw_vertex_program *brw_vp = (struct brw_vertex_program *) shader; + + FREE((void *) brw_vp->program.tokens); + FREE(brw_vp); +} + + +static void brw_set_clip_state( struct pipe_context *pipe, + const struct pipe_clip_state *clip ) +{ + struct brw_context *brw = brw_context(pipe); + + brw->attribs.Clip = *clip; +} + + +static void brw_set_viewport_state( struct pipe_context *pipe, + const struct pipe_viewport_state *viewport ) +{ + struct brw_context *brw = brw_context(pipe); + + brw->attribs.Viewport = *viewport; /* struct copy */ + brw->state.dirty.brw |= BRW_NEW_VIEWPORT; + + /* pass the viewport info to the draw module */ + //draw_set_viewport_state(brw->draw, viewport); +} + + +static void brw_set_vertex_buffers(struct pipe_context *pipe, + unsigned count, + const struct pipe_vertex_buffer *buffers) +{ + struct brw_context *brw = brw_context(pipe); + memcpy(brw->vb.vbo_array, buffers, count * sizeof(buffers[0])); +} + +static void brw_set_vertex_elements(struct pipe_context *pipe, + unsigned count, + const struct pipe_vertex_element *elements) +{ + /* flush ? */ + struct brw_context *brw = brw_context(pipe); + uint i; + + assert(count <= PIPE_MAX_ATTRIBS); + + for (i = 0; i < count; i++) { + struct brw_vertex_element_state el; + memset(&el, 0, sizeof(el)); + + el.ve0.src_offset = elements[i].src_offset; + el.ve0.src_format = brw_translate_surface_format(elements[i].src_format); + el.ve0.valid = 1; + el.ve0.vertex_buffer_index = elements[i].vertex_buffer_index; + + el.ve1.dst_offset = i * 4; + + el.ve1.vfcomponent3 = BRW_VFCOMPONENT_STORE_SRC; + el.ve1.vfcomponent2 = BRW_VFCOMPONENT_STORE_SRC; + el.ve1.vfcomponent1 = BRW_VFCOMPONENT_STORE_SRC; + el.ve1.vfcomponent0 = BRW_VFCOMPONENT_STORE_SRC; + + switch (elements[i].nr_components) { + case 1: el.ve1.vfcomponent1 = BRW_VFCOMPONENT_STORE_0; + case 2: el.ve1.vfcomponent2 = BRW_VFCOMPONENT_STORE_0; + case 3: el.ve1.vfcomponent3 = BRW_VFCOMPONENT_STORE_1_FLT; + break; + } + + brw->vb.inputs[i] = el; + } +} + + + +/************************************************************************ + * Constant buffers + */ + +static void brw_set_constant_buffer(struct pipe_context *pipe, + uint shader, uint index, + const struct pipe_constant_buffer *buf) +{ + struct brw_context *brw = brw_context(pipe); + + assert(buf == 0 || index == 0); + + brw->attribs.Constants[shader] = buf; + brw->state.dirty.brw |= BRW_NEW_CONSTANTS; +} + + +/************************************************************************ + * Texture surfaces + */ + + +static void brw_set_sampler_textures(struct pipe_context *pipe, + unsigned num, + struct pipe_texture **texture) +{ + struct brw_context *brw = brw_context(pipe); + uint i; + + assert(num <= PIPE_MAX_SAMPLERS); + + /* Check for no-op */ + if (num == brw->num_textures && + !memcmp(brw->attribs.Texture, texture, num * + sizeof(struct pipe_texture *))) + return; + + for (i = 0; i < num; i++) + pipe_texture_reference((struct pipe_texture **) &brw->attribs.Texture[i], + texture[i]); + + for (i = num; i < brw->num_textures; i++) + pipe_texture_reference((struct pipe_texture **) &brw->attribs.Texture[i], + NULL); + + brw->num_textures = num; + + brw->state.dirty.brw |= BRW_NEW_TEXTURE; +} + + +/************************************************************************ + * Render targets, etc + */ + +static void brw_set_framebuffer_state(struct pipe_context *pipe, + const struct pipe_framebuffer_state *fb) +{ + struct brw_context *brw = brw_context(pipe); + + brw->attribs.FrameBuffer = *fb; /* struct copy */ + + brw->state.dirty.brw |= BRW_NEW_FRAMEBUFFER; +} + + + +/************************************************************************ + * Rasterizer state + */ + +static void * +brw_create_rasterizer_state(struct pipe_context *pipe, + const struct pipe_rasterizer_state *rasterizer) +{ + DUP(pipe_rasterizer_state, rasterizer); +} + +static void brw_bind_rasterizer_state( struct pipe_context *pipe, + void *setup ) +{ + struct brw_context *brw = brw_context(pipe); + + brw->attribs.Raster = (struct pipe_rasterizer_state *)setup; + + /* Also pass-through to draw module: + */ + //draw_set_rasterizer_state(brw->draw, setup); + + brw->state.dirty.brw |= BRW_NEW_RASTERIZER; +} + +static void brw_delete_rasterizer_state(struct pipe_context *pipe, + void *setup) +{ + free(setup); +} + + + +void +brw_init_state_functions( struct brw_context *brw ) +{ + brw->pipe.create_blend_state = brw_create_blend_state; + brw->pipe.bind_blend_state = brw_bind_blend_state; + brw->pipe.delete_blend_state = brw_delete_blend_state; + + brw->pipe.create_sampler_state = brw_create_sampler_state; + brw->pipe.bind_sampler_states = brw_bind_sampler_states; + brw->pipe.delete_sampler_state = brw_delete_sampler_state; + + brw->pipe.create_depth_stencil_alpha_state = brw_create_depth_stencil_state; + brw->pipe.bind_depth_stencil_alpha_state = brw_bind_depth_stencil_state; + brw->pipe.delete_depth_stencil_alpha_state = brw_delete_depth_stencil_state; + + brw->pipe.create_rasterizer_state = brw_create_rasterizer_state; + brw->pipe.bind_rasterizer_state = brw_bind_rasterizer_state; + brw->pipe.delete_rasterizer_state = brw_delete_rasterizer_state; + brw->pipe.create_fs_state = brw_create_fs_state; + brw->pipe.bind_fs_state = brw_bind_fs_state; + brw->pipe.delete_fs_state = brw_delete_fs_state; + brw->pipe.create_vs_state = brw_create_vs_state; + brw->pipe.bind_vs_state = brw_bind_vs_state; + brw->pipe.delete_vs_state = brw_delete_vs_state; + + brw->pipe.set_blend_color = brw_set_blend_color; + brw->pipe.set_clip_state = brw_set_clip_state; + brw->pipe.set_constant_buffer = brw_set_constant_buffer; + brw->pipe.set_framebuffer_state = brw_set_framebuffer_state; + +// brw->pipe.set_feedback_state = brw_set_feedback_state; +// brw->pipe.set_feedback_buffer = brw_set_feedback_buffer; + + brw->pipe.set_polygon_stipple = brw_set_polygon_stipple; + brw->pipe.set_scissor_state = brw_set_scissor_state; + brw->pipe.set_sampler_textures = brw_set_sampler_textures; + brw->pipe.set_viewport_state = brw_set_viewport_state; + brw->pipe.set_vertex_buffers = brw_set_vertex_buffers; + brw->pipe.set_vertex_elements = brw_set_vertex_elements; +} diff --git a/src/gallium/drivers/i965simple/brw_surface.c b/src/gallium/drivers/i965simple/brw_surface.c index c99a91d..3e3736b 100644 --- a/src/gallium/drivers/i965simple/brw_surface.c +++ b/src/gallium/drivers/i965simple/brw_surface.c @@ -35,27 +35,6 @@ #include "util/p_tile.h" -/* Upload data to a rectangular sub-region. Lots of choices how to do this: - * - * - memcpy by span to current destination - * - upload data as new buffer and blit - * - * Currently always memcpy. - */ -static void -brw_surface_data(struct pipe_context *pipe, - struct pipe_surface *dst, - unsigned dstx, unsigned dsty, - const void *src, unsigned src_pitch, - unsigned srcx, unsigned srcy, unsigned width, unsigned height) -{ - pipe_copy_rect(pipe_surface_map(dst) + dst->offset, - dst->cpp, dst->pitch, - dstx, dsty, width, height, src, src_pitch, srcx, srcy); - - pipe_surface_unmap(dst); -} - /* Assumes all values are within bounds -- no checking at this level - * do it higher up if required. @@ -72,17 +51,25 @@ brw_surface_copy(struct pipe_context *pipe, assert(dst->cpp == src->cpp); if (0) { - pipe_copy_rect(pipe_surface_map(dst) + dst->offset, + void *dst_map = pipe->screen->surface_map( pipe->screen, + dst, + PIPE_BUFFER_USAGE_CPU_WRITE ); + + const void *src_map = pipe->screen->surface_map( pipe->screen, + src, + PIPE_BUFFER_USAGE_CPU_READ ); + + pipe_copy_rect(dst_map, dst->cpp, dst->pitch, - dstx, dsty, - width, height, - pipe_surface_map(src) + src->offset, - do_flip ? -src->pitch : src->pitch, + dstx, dsty, + width, height, + src_map, + do_flip ? -(int) src->pitch : src->pitch, srcx, do_flip ? 1 - srcy - height : srcy); - pipe_surface_unmap(src); - pipe_surface_unmap(dst); + pipe->screen->surface_unmap(pipe->screen, src); + pipe->screen->surface_unmap(pipe->screen, dst); } else { brw_copy_blit(brw_context(pipe), @@ -113,7 +100,10 @@ brw_surface_fill(struct pipe_context *pipe, { if (0) { unsigned i, j; - void *dst_map = pipe_surface_map(dst); + void *dst_map = pipe->screen->surface_map( pipe->screen, + dst, + PIPE_BUFFER_USAGE_CPU_WRITE ); + switch (dst->cpp) { case 1: { @@ -147,7 +137,7 @@ brw_surface_fill(struct pipe_context *pipe, break; } - pipe_surface_unmap( dst ); + pipe->screen->surface_unmap(pipe->screen, dst); } else { brw_fill_blit(brw_context(pipe), @@ -164,7 +154,6 @@ brw_surface_fill(struct pipe_context *pipe, void brw_init_surface_functions(struct brw_context *brw) { - (void) brw_surface_data; /* silence warning */ brw->pipe.surface_copy = brw_surface_copy; brw->pipe.surface_fill = brw_surface_fill; } diff --git a/src/gallium/drivers/i965simple/brw_tex_layout.c b/src/gallium/drivers/i965simple/brw_tex_layout.c index b580f98..78ae0b1 100644 --- a/src/gallium/drivers/i965simple/brw_tex_layout.c +++ b/src/gallium/drivers/i965simple/brw_tex_layout.c @@ -357,14 +357,6 @@ brw_texture_release_screen(struct pipe_screen *screen, } -static void -brw_texture_update(struct pipe_context *pipe, struct pipe_texture *texture, - uint face, uint levelsMask) -{ - /* no-op? */ -} - - static struct pipe_surface * brw_get_tex_surface_screen(struct pipe_screen *screen, struct pipe_texture *pt, @@ -407,7 +399,7 @@ brw_get_tex_surface_screen(struct pipe_screen *screen, void brw_init_texture_functions(struct brw_context *brw) { - brw->pipe.texture_update = brw_texture_update; +// brw->pipe.texture_update = brw_texture_update; } diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c index a48e546..531bd76 100644 --- a/src/gallium/drivers/softpipe/sp_context.c +++ b/src/gallium/drivers/softpipe/sp_context.c @@ -193,11 +193,11 @@ softpipe_create( struct pipe_screen *screen, * Must be before quad stage setup! */ for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) - softpipe->cbuf_cache[i] = sp_create_tile_cache(); - softpipe->zsbuf_cache = sp_create_tile_cache(); + softpipe->cbuf_cache[i] = sp_create_tile_cache( screen ); + softpipe->zsbuf_cache = sp_create_tile_cache( screen ); for (i = 0; i < PIPE_MAX_SAMPLERS; i++) - softpipe->tex_cache[i] = sp_create_tile_cache(); + softpipe->tex_cache[i] = sp_create_tile_cache( screen ); /* setup quad rendering stages */ diff --git a/src/gallium/drivers/softpipe/sp_flush.c b/src/gallium/drivers/softpipe/sp_flush.c index 0625b69..e03994b6 100644 --- a/src/gallium/drivers/softpipe/sp_flush.c +++ b/src/gallium/drivers/softpipe/sp_flush.c @@ -50,25 +50,28 @@ softpipe_flush( struct pipe_context *pipe, draw_flush(softpipe->draw); - /* - flush the quad pipeline - * - flush the texture cache - * - flush the render cache - */ + if (flags & PIPE_FLUSH_TEXTURE_CACHE) { + for (i = 0; i < softpipe->num_textures; i++) { + sp_flush_tile_cache(softpipe, softpipe->tex_cache[i]); + } + } - for (i = 0; i < softpipe->framebuffer.num_cbufs; i++) - if (softpipe->cbuf_cache[i]) - sp_flush_tile_cache(softpipe, softpipe->cbuf_cache[i]); + if (flags & PIPE_FLUSH_RENDER_CACHE) { + for (i = 0; i < softpipe->framebuffer.num_cbufs; i++) + if (softpipe->cbuf_cache[i]) + sp_flush_tile_cache(softpipe, softpipe->cbuf_cache[i]); - if (softpipe->zsbuf_cache) - sp_flush_tile_cache(softpipe, softpipe->zsbuf_cache); + if (softpipe->zsbuf_cache) + sp_flush_tile_cache(softpipe, softpipe->zsbuf_cache); - /* Need this call for hardware buffers before swapbuffers. - * - * there should probably be another/different flush-type function - * that's called before swapbuffers because we don't always want - * to unmap surfaces when flushing. - */ - softpipe_unmap_surfaces(softpipe); + /* Need this call for hardware buffers before swapbuffers. + * + * there should probably be another/different flush-type function + * that's called before swapbuffers because we don't always want + * to unmap surfaces when flushing. + */ + softpipe_unmap_surfaces(softpipe); + } if (fence) *fence = NULL; diff --git a/src/gallium/drivers/softpipe/sp_screen.c b/src/gallium/drivers/softpipe/sp_screen.c index 7dacb1c..e9926bf 100644 --- a/src/gallium/drivers/softpipe/sp_screen.c +++ b/src/gallium/drivers/softpipe/sp_screen.c @@ -33,6 +33,7 @@ #include "sp_texture.h" #include "sp_winsys.h" +#include "sp_screen.h" static const char * @@ -137,6 +138,7 @@ softpipe_destroy_screen( struct pipe_screen *screen ) } + /** * Create a new pipe_screen object * Note: we're not presently subclassing pipe_screen (no softpipe_screen). @@ -144,22 +146,22 @@ softpipe_destroy_screen( struct pipe_screen *screen ) struct pipe_screen * softpipe_create_screen(struct pipe_winsys *winsys) { - struct pipe_screen *screen = CALLOC_STRUCT(pipe_screen); + struct softpipe_screen *screen = CALLOC_STRUCT(softpipe_screen); if (!screen) return NULL; - screen->winsys = winsys; + screen->base.winsys = winsys; - screen->destroy = softpipe_destroy_screen; + screen->base.destroy = softpipe_destroy_screen; - screen->get_name = softpipe_get_name; - screen->get_vendor = softpipe_get_vendor; - screen->get_param = softpipe_get_param; - screen->get_paramf = softpipe_get_paramf; - screen->is_format_supported = softpipe_is_format_supported; + screen->base.get_name = softpipe_get_name; + screen->base.get_vendor = softpipe_get_vendor; + screen->base.get_param = softpipe_get_param; + screen->base.get_paramf = softpipe_get_paramf; + screen->base.is_format_supported = softpipe_is_format_supported; - softpipe_init_screen_texture_funcs(screen); + softpipe_init_screen_texture_funcs(&screen->base); - return screen; + return &screen->base; } diff --git a/src/gallium/winsys/dri/intel/intel_batchpool.h b/src/gallium/drivers/softpipe/sp_screen.h index f6a9572..3d4bfd3 100644 --- a/src/gallium/winsys/dri/intel/intel_batchpool.h +++ b/src/gallium/drivers/softpipe/sp_screen.h @@ -1,8 +1,8 @@ /************************************************************************** * - * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. * All Rights Reserved. - * + * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the * "Software"), to deal in the Software without restriction, including @@ -25,13 +25,34 @@ * **************************************************************************/ -#ifndef INTEL_BATCHPOOL_H -#define INTEL_BATCHPOOL_H +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ + +#ifndef SP_SCREEN_H +#define SP_SCREEN_H + +#include "pipe/p_screen.h" +#include "pipe/p_defines.h" + + + +struct softpipe_screen { + struct pipe_screen base; + + /* Increments whenever textures are modified. Contexts can track + * this. + */ + unsigned timestamp; +}; + + + -extern struct _DriBufferPool *driBatchPoolInit(int fd, unsigned flags, - unsigned long bufSize, - unsigned numBufs, - unsigned checkDelayed); +static INLINE struct softpipe_screen * +softpipe_screen( struct pipe_screen *pipe ) +{ + return (struct softpipe_screen *)pipe; +} -#endif /* INTEL_BATCHPOOL_H */ +#endif /* SP_SCREEN_H */ diff --git a/src/gallium/drivers/softpipe/sp_surface.c b/src/gallium/drivers/softpipe/sp_surface.c index 653449c..29a1e92 100644 --- a/src/gallium/drivers/softpipe/sp_surface.c +++ b/src/gallium/drivers/softpipe/sp_surface.c @@ -46,19 +46,28 @@ sp_surface_copy(struct pipe_context *pipe, struct pipe_surface *src, unsigned srcx, unsigned srcy, unsigned width, unsigned height) { + void *dst_map = pipe->screen->surface_map( pipe->screen, + dst, + PIPE_BUFFER_USAGE_CPU_WRITE ); + + const void *src_map = pipe->screen->surface_map( pipe->screen, + src, + PIPE_BUFFER_USAGE_CPU_READ ); + assert( dst->cpp == src->cpp ); + assert(src_map && dst_map); - pipe_copy_rect(pipe_surface_map(dst), + pipe_copy_rect(dst_map, dst->cpp, dst->pitch, dstx, dsty, width, height, - pipe_surface_map(src), + src_map, do_flip ? -(int) src->pitch : src->pitch, srcx, do_flip ? 1 - srcy - height : srcy); - pipe_surface_unmap(src); - pipe_surface_unmap(dst); + pipe->screen->surface_unmap(pipe->screen, src); + pipe->screen->surface_unmap(pipe->screen, dst); } @@ -83,7 +92,9 @@ sp_surface_fill(struct pipe_context *pipe, unsigned width, unsigned height, unsigned value) { unsigned i, j; - void *dst_map = pipe_surface_map(dst); + void *dst_map = pipe->screen->surface_map( pipe->screen, + dst, + PIPE_BUFFER_USAGE_CPU_WRITE ); assert(dst->pitch > 0); assert(width <= dst->pitch); @@ -147,7 +158,7 @@ sp_surface_fill(struct pipe_context *pipe, break; } - pipe_surface_unmap( dst ); + pipe->screen->surface_unmap(pipe->screen, dst); } diff --git a/src/gallium/drivers/softpipe/sp_texture.c b/src/gallium/drivers/softpipe/sp_texture.c index 256586e..1d7a1ff 100644 --- a/src/gallium/drivers/softpipe/sp_texture.c +++ b/src/gallium/drivers/softpipe/sp_texture.c @@ -40,6 +40,7 @@ #include "sp_state.h" #include "sp_texture.h" #include "sp_tile_cache.h" +#include "sp_screen.h" /* Simple, maximally packed layout. @@ -51,40 +52,88 @@ static unsigned minify( unsigned d ) } -static void -softpipe_texture_layout(struct softpipe_texture * spt) +/* Conventional allocation path for non-display textures: + */ +static boolean +softpipe_texture_layout(struct pipe_screen *screen, + struct softpipe_texture * spt) { + struct pipe_winsys *ws = screen->winsys; struct pipe_texture *pt = &spt->base; unsigned level; unsigned width = pt->width[0]; unsigned height = pt->height[0]; unsigned depth = pt->depth[0]; - spt->buffer_size = 0; + unsigned buffer_size = 0; for (level = 0; level <= pt->last_level; level++) { pt->width[level] = width; pt->height[level] = height; pt->depth[level] = depth; + spt->pitch[level] = width; - spt->level_offset[level] = spt->buffer_size; + spt->level_offset[level] = buffer_size; - spt->buffer_size += ((pt->compressed) ? MAX2(1, height/4) : height) * - ((pt->target == PIPE_TEXTURE_CUBE) ? 6 : depth) * - width * pt->cpp; + buffer_size += (((pt->compressed) ? MAX2(1, height/4) : height) * + ((pt->target == PIPE_TEXTURE_CUBE) ? 6 : depth) * + width * pt->cpp); width = minify(width); height = minify(height); depth = minify(depth); } + + spt->buffer = ws->buffer_create(ws, 32, + PIPE_BUFFER_USAGE_PIXEL, + buffer_size); + + return spt->buffer != NULL; +} + + + +/* Hack it up to use the old winsys->surface_alloc_storage() + * method for now: + */ +static boolean +softpipe_displaytarget_layout(struct pipe_screen *screen, + struct softpipe_texture * spt) +{ + struct pipe_winsys *ws = screen->winsys; + struct pipe_surface surf; + unsigned flags = (PIPE_BUFFER_USAGE_CPU_READ | + PIPE_BUFFER_USAGE_CPU_WRITE | + PIPE_BUFFER_USAGE_GPU_READ | + PIPE_BUFFER_USAGE_GPU_WRITE); + + + memset(&surf, 0, sizeof(surf)); + + ws->surface_alloc_storage( ws, + &surf, + spt->base.width[0], + spt->base.height[0], + spt->base.format, + flags, + spt->base.tex_usage); + + /* Now extract the goodies: + */ + spt->buffer = surf.buffer; + spt->pitch[0] = surf.pitch; + + return spt->buffer != NULL; } + + + static struct pipe_texture * softpipe_texture_create(struct pipe_screen *screen, const struct pipe_texture *templat) { - struct pipe_winsys *ws = screen->winsys; struct softpipe_texture *spt = CALLOC_STRUCT(softpipe_texture); if (!spt) return NULL; @@ -93,19 +142,21 @@ softpipe_texture_create(struct pipe_screen *screen, spt->base.refcount = 1; spt->base.screen = screen; - softpipe_texture_layout(spt); - - spt->buffer = ws->buffer_create(ws, 32, - PIPE_BUFFER_USAGE_PIXEL, - spt->buffer_size); - if (!spt->buffer) { - FREE(spt); - return NULL; + if (spt->base.tex_usage & PIPE_TEXTURE_USAGE_DISPLAY_TARGET) { + if (!softpipe_displaytarget_layout(screen, spt)) + goto fail; } - + else { + if (!softpipe_texture_layout(screen, spt)) + goto fail; + } + assert(spt->base.refcount == 1); - return &spt->base; + + fail: + FREE(spt); + return NULL; } @@ -116,19 +167,10 @@ softpipe_texture_release(struct pipe_screen *screen, if (!*pt) return; - /* - DBG("%s %p refcount will be %d\n", - __FUNCTION__, (void *) *pt, (*pt)->refcount - 1); - */ if (--(*pt)->refcount <= 0) { struct softpipe_texture *spt = softpipe_texture(*pt); - /* - DBG("%s deleting %p\n", __FUNCTION__, (void *) spt); - */ - pipe_buffer_reference(screen->winsys, &spt->buffer, NULL); - FREE(spt); } *pt = NULL; @@ -138,7 +180,8 @@ softpipe_texture_release(struct pipe_screen *screen, static struct pipe_surface * softpipe_get_tex_surface(struct pipe_screen *screen, struct pipe_texture *pt, - unsigned face, unsigned level, unsigned zslice) + unsigned face, unsigned level, unsigned zslice, + unsigned usage) { struct pipe_winsys *ws = screen->winsys; struct softpipe_texture *spt = softpipe_texture(pt); @@ -157,6 +200,24 @@ softpipe_get_tex_surface(struct pipe_screen *screen, ps->height = pt->height[level]; ps->pitch = ps->width; ps->offset = spt->level_offset[level]; + ps->usage = usage; + + /* Because we are softpipe, anything that the state tracker + * thought was going to be done with the GPU will actually get + * done with the CPU. Let's adjust the flags to take that into + * account. + */ + if (ps->usage & PIPE_BUFFER_USAGE_GPU_WRITE) + ps->usage |= PIPE_BUFFER_USAGE_CPU_WRITE; + + if (ps->usage & PIPE_BUFFER_USAGE_GPU_READ) + ps->usage |= PIPE_BUFFER_USAGE_CPU_READ; + + + pipe_texture_reference(&ps->texture, pt); + ps->face = face; + ps->level = level; + ps->zslice = zslice; if (pt->target == PIPE_TEXTURE_CUBE || pt->target == PIPE_TEXTURE_3D) { ps->offset += ((pt->target == PIPE_TEXTURE_CUBE) ? face : zslice) * @@ -172,25 +233,64 @@ softpipe_get_tex_surface(struct pipe_screen *screen, } -static void -softpipe_texture_update(struct pipe_context *pipe, - struct pipe_texture *texture, - uint face, uint levelsMask) +static void +softpipe_tex_surface_release(struct pipe_screen *screen, + struct pipe_surface **s) { - struct softpipe_context *softpipe = softpipe_context(pipe); - uint unit; - for (unit = 0; unit < softpipe->num_textures; unit++) { - if (softpipe->texture[unit] == texture) { - sp_flush_tile_cache(softpipe, softpipe->tex_cache[unit]); - } + /* Effectively do the texture_update work here - if texture images + * needed post-processing to put them into hardware layout, this is + * where it would happen. For softpipe, nothing to do. + */ + assert ((*s)->texture); + pipe_texture_reference(&(*s)->texture, NULL); + + screen->winsys->surface_release(screen->winsys, s); +} + + +static void * +softpipe_surface_map( struct pipe_screen *screen, + struct pipe_surface *surface, + unsigned flags ) +{ + ubyte *map; + + if (flags & ~surface->usage) { + assert(0); + return NULL; + } + + map = screen->winsys->buffer_map( screen->winsys, surface->buffer, flags ); + if (map == NULL) + return NULL; + + /* May want to different things here depending on read/write nature + * of the map: + */ + if (surface->texture && + (flags & PIPE_BUFFER_USAGE_CPU_WRITE)) + { + /* Do something to notify sharing contexts of a texture change. + * In softpipe, that would mean flushing the texture cache. + */ + softpipe_screen(screen)->timestamp++; } + + return map + surface->offset; +} + + +static void +softpipe_surface_unmap(struct pipe_screen *screen, + struct pipe_surface *surface) +{ + screen->winsys->buffer_unmap( screen->winsys, surface->buffer ); } void -softpipe_init_texture_funcs( struct softpipe_context *softpipe ) +softpipe_init_texture_funcs(struct softpipe_context *sp) { - softpipe->pipe.texture_update = softpipe_texture_update; } @@ -199,5 +299,10 @@ softpipe_init_screen_texture_funcs(struct pipe_screen *screen) { screen->texture_create = softpipe_texture_create; screen->texture_release = softpipe_texture_release; + screen->get_tex_surface = softpipe_get_tex_surface; + screen->tex_surface_release = softpipe_tex_surface_release; + + screen->surface_map = softpipe_surface_map; + screen->surface_unmap = softpipe_surface_unmap; } diff --git a/src/gallium/drivers/softpipe/sp_texture.h b/src/gallium/drivers/softpipe/sp_texture.h index a732214..779a9d8 100644 --- a/src/gallium/drivers/softpipe/sp_texture.h +++ b/src/gallium/drivers/softpipe/sp_texture.h @@ -42,11 +42,11 @@ struct softpipe_texture struct pipe_texture base; unsigned long level_offset[PIPE_MAX_TEXTURE_LEVELS]; + unsigned long pitch[PIPE_MAX_TEXTURE_LEVELS]; /* The data is held here: */ struct pipe_buffer *buffer; - unsigned long buffer_size; }; @@ -61,7 +61,6 @@ softpipe_texture(struct pipe_texture *pt) extern void softpipe_init_texture_funcs( struct softpipe_context *softpipe ); - extern void softpipe_init_screen_texture_funcs(struct pipe_screen *screen); diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.c b/src/gallium/drivers/softpipe/sp_tile_cache.c index 1117c0a..28c29da 100644 --- a/src/gallium/drivers/softpipe/sp_tile_cache.c +++ b/src/gallium/drivers/softpipe/sp_tile_cache.c @@ -49,6 +49,7 @@ struct softpipe_tile_cache { + struct pipe_screen *screen; struct pipe_surface *surface; /**< the surface we're caching */ void *surface_map; struct pipe_texture *texture; /**< if caching a texture */ @@ -109,13 +110,14 @@ clear_clear_flag(uint *bitvec, int x, int y) struct softpipe_tile_cache * -sp_create_tile_cache(void) +sp_create_tile_cache( struct pipe_screen *screen ) { struct softpipe_tile_cache *tc; uint pos; tc = CALLOC_STRUCT( softpipe_tile_cache ); if (tc) { + tc->screen = screen; for (pos = 0; pos < NUM_ENTRIES; pos++) { tc->entries[pos].x = tc->entries[pos].y = -1; @@ -154,16 +156,17 @@ sp_tile_cache_set_surface(struct softpipe_tile_cache *tc, assert(!tc->texture); if (tc->surface_map) { - /*assert(tc->surface != ps);*/ - pipe_surface_unmap(tc->surface); + tc->screen->surface_unmap(tc->screen, tc->surface); tc->surface_map = NULL; } pipe_surface_reference(&tc->surface, ps); - if (ps) { - if (tc->surface_map) - tc->surface_map = pipe_surface_map(ps); + if (tc->surface) { + if (tc->surface_map) /* XXX: this is always NULL!? */ + tc->surface_map = tc->screen->surface_map(tc->screen, tc->surface, + PIPE_BUFFER_USAGE_CPU_READ | + PIPE_BUFFER_USAGE_CPU_WRITE); tc->depth_stencil = (ps->format == PIPE_FORMAT_S8Z24_UNORM || ps->format == PIPE_FORMAT_Z16_UNORM || @@ -187,10 +190,13 @@ void sp_tile_cache_map_surfaces(struct softpipe_tile_cache *tc) { if (tc->surface && !tc->surface_map) - tc->surface_map = pipe_surface_map(tc->surface); + tc->surface_map = tc->screen->surface_map(tc->screen, tc->surface, + PIPE_BUFFER_USAGE_CPU_WRITE | + PIPE_BUFFER_USAGE_CPU_READ); if (tc->tex_surf && !tc->tex_surf_map) - tc->tex_surf_map = pipe_surface_map(tc->tex_surf); + tc->tex_surf_map = tc->screen->surface_map(tc->screen, tc->tex_surf, + PIPE_BUFFER_USAGE_CPU_READ); } @@ -198,12 +204,12 @@ void sp_tile_cache_unmap_surfaces(struct softpipe_tile_cache *tc) { if (tc->surface_map) { - pipe_surface_unmap(tc->surface); + tc->screen->surface_unmap(tc->screen, tc->surface); tc->surface_map = NULL; } if (tc->tex_surf_map) { - pipe_surface_unmap(tc->tex_surf); + tc->screen->surface_unmap(tc->screen, tc->tex_surf); tc->tex_surf_map = NULL; } } @@ -224,7 +230,7 @@ sp_tile_cache_set_texture(struct pipe_context *pipe, pipe_texture_reference(&tc->texture, texture); if (tc->tex_surf_map) { - pipe_surface_unmap(tc->tex_surf); + tc->screen->surface_unmap(tc->screen, tc->tex_surf); tc->tex_surf_map = NULL; } pipe_surface_reference(&tc->tex_surf, NULL); @@ -514,10 +520,12 @@ sp_get_cached_tile_tex(struct pipe_context *pipe, /* get new surface (view into texture) */ if (tc->tex_surf_map) - pipe_surface_unmap(tc->tex_surf); + tc->screen->surface_unmap(tc->screen, tc->tex_surf); - tc->tex_surf = screen->get_tex_surface(screen, tc->texture, face, level, z); - tc->tex_surf_map = pipe_surface_map(tc->tex_surf); + tc->tex_surf = screen->get_tex_surface(screen, tc->texture, face, level, z, + PIPE_BUFFER_USAGE_CPU_READ); + tc->tex_surf_map = screen->surface_map(screen, tc->tex_surf, + PIPE_BUFFER_USAGE_CPU_READ); tc->tex_face = face; tc->tex_level = level; diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.h b/src/gallium/drivers/softpipe/sp_tile_cache.h index 2631e29..bc96c94 100644 --- a/src/gallium/drivers/softpipe/sp_tile_cache.h +++ b/src/gallium/drivers/softpipe/sp_tile_cache.h @@ -61,7 +61,7 @@ struct softpipe_cached_tile extern struct softpipe_tile_cache * -sp_create_tile_cache(void); +sp_create_tile_cache( struct pipe_screen *screen ); extern void sp_destroy_tile_cache(struct softpipe_tile_cache *tc); diff --git a/src/gallium/include/pipe/p_compiler.h b/src/gallium/include/pipe/p_compiler.h index 01d1807..29b7f9c 100644 --- a/src/gallium/include/pipe/p_compiler.h +++ b/src/gallium/include/pipe/p_compiler.h @@ -52,39 +52,55 @@ #endif /* __MSC__ */ -typedef unsigned int uint; -typedef unsigned char ubyte; -typedef unsigned char boolean; -typedef unsigned short ushort; -typedef unsigned long long uint64; - - #if defined(__MSC__) -typedef char int8_t; -typedef unsigned char uint8_t; -typedef short int16_t; -typedef unsigned short uint16_t; -typedef long int32_t; -typedef unsigned long uint32_t; -typedef long long int64_t; -typedef unsigned long long uint64_t; +typedef __int8 int8_t; +typedef unsigned __int8 uint8_t; +typedef __int16 int16_t; +typedef unsigned __int16 uint16_t; +typedef __int32 int32_t; +typedef unsigned __int32 uint32_t; +typedef __int64 int64_t; +typedef unsigned __int64 uint64_t; #if defined(_WIN64) typedef __int64 intptr_t; typedef unsigned __int64 uintptr_t; #else -typedef int intptr_t; -typedef unsigned int uintptr_t; +typedef __int32 intptr_t; +typedef unsigned __int32 uintptr_t; #endif +#ifndef __cplusplus +#define false 0 +#define true 1 +#define bool _Bool +typedef int _Bool; +#define __bool_true_false_are_defined 1 +#endif /* !__cplusplus */ + #else #include <stdint.h> +#include <stdbool.h> #endif -#define TRUE 1 -#define FALSE 0 +typedef unsigned int uint; +typedef unsigned char ubyte; +typedef unsigned short ushort; +typedef uint64_t uint64; + +#if 0 +#define boolean bool +#else +typedef unsigned char boolean; +#endif +#ifndef TRUE +#define TRUE true +#endif +#ifndef FALSE +#define FALSE false +#endif /* Function inlining */ diff --git a/src/gallium/include/pipe/p_config.h b/src/gallium/include/pipe/p_config.h index 5c030bd..d2d2ae1 100644 --- a/src/gallium/include/pipe/p_config.h +++ b/src/gallium/include/pipe/p_config.h @@ -35,6 +35,10 @@ * this file is auto-generated by an autoconf-like tool at some point, as some * things cannot be determined by existing defines alone. * + * See also: + * - http://gcc.gnu.org/onlinedocs/cpp/Common-Predefined-Macros.html + * - echo | gcc -dM -E - | sort + * - http://msdn.microsoft.com/en-us/library/b0084kay.aspx * @author José Fonseca <jrfonseca@tungstengraphics.com> */ @@ -63,11 +67,11 @@ * Processor architecture */ -#if defined(_X86_) || defined(__i386__) || defined(__386__) || defined(i386) +#if defined(__i386__) /* gcc */ || defined(_M_IX86) /* msvc */ || defined(_X86_) || defined(__386__) || defined(i386) #define PIPE_ARCH_X86 #endif -#if 0 /* FIXME */ +#if defined(__x86_64__) /* gcc */ || defined(_M_X64) /* msvc */ || defined(_M_AMD64) /* msvc */ #define PIPE_ARCH_X86_64 #endif @@ -100,6 +104,7 @@ #endif #if defined(PIPE_OS_WINDOWS) +#ifndef _WIN32_WCE #if !defined(PIPE_SUBSYSTEM_USER) && !defined(PIPE_SUBSYSTEM_KERNEL) #error Neither PIPE_SUBSYSTEM_USER or PIPE_SUBSYSTEM_KERNEL defined. #endif @@ -109,12 +114,12 @@ #if 0 /* FIXME */ #define PIPE_SUBSYSTEM_WINDOWS_MINIPORT #endif -#if 0 /* FIXME */ -#define PIPE_SUBSYSTEM_WINDOWS_CE -#endif #if defined(PIPE_SUBSYSTEM_USER) #define PIPE_SUBSYSTEM_WINDOWS_USER #endif +#else /* _WIN32_WCE */ +#define PIPE_SUBSYSTEM_WINDOWS_CE +#endif /* _WIN32_WCE */ #endif /* PIPE_OS_WINDOWS */ diff --git a/src/gallium/include/pipe/p_context.h b/src/gallium/include/pipe/p_context.h index f3a9c2c..0f68f59 100644 --- a/src/gallium/include/pipe/p_context.h +++ b/src/gallium/include/pipe/p_context.h @@ -198,12 +198,6 @@ struct pipe_context { /*@}*/ - /** Called when texture data is changed */ - void (*texture_update)(struct pipe_context *pipe, - struct pipe_texture *texture, - uint face, uint dirtyLevelsMask); - - /** Flush rendering (flags = bitmask of PIPE_FLUSH_x tokens) */ void (*flush)( struct pipe_context *pipe, unsigned flags, diff --git a/src/gallium/include/pipe/p_debug.h b/src/gallium/include/pipe/p_debug.h index 7a7312e..05eca75 100644 --- a/src/gallium/include/pipe/p_debug.h +++ b/src/gallium/include/pipe/p_debug.h @@ -59,6 +59,13 @@ extern "C" { #endif #endif + +/* MSVC bebore VC7 does not have the __FUNCTION__ macro */ +#if defined(_MSC_VER) && _MSC_VER < 1300 +#define __FUNCTION__ "???" +#endif + + void _debug_vprintf(const char *format, va_list ap); @@ -127,8 +134,8 @@ void _debug_break(void); #ifdef DEBUG #if (defined(__i386__) || defined(__386__)) && defined(__GNUC__) #define debug_break() __asm("int3") -#elif (defined(__i386__) || defined(__386__)) && defined(__MSC__) -#define debug_break() _asm {int 3} +#elif defined(_M_IX86) && defined(_MSC_VER) +#define debug_break() do { _asm {int 3} } while(0) #else #define debug_break() _debug_break() #endif @@ -313,6 +320,28 @@ void debug_memory_end(unsigned long beginning); +#if defined(PROFILE) && defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY) + +void +debug_profile_start(void); + +void +debug_profile_stop(void); + +#endif + + +#ifdef DEBUG +void debug_dump_image(const char *prefix, + unsigned format, unsigned cpp, + unsigned width, unsigned height, + unsigned pitch, + const void *data); +#else +#define debug_dump_image(prefix, format, cpp, width, height, pitch, data) ((void)0) +#endif + + #ifdef __cplusplus } #endif diff --git a/src/gallium/include/pipe/p_defines.h b/src/gallium/include/pipe/p_defines.h index 46a355d..83330ef 100644 --- a/src/gallium/include/pipe/p_defines.h +++ b/src/gallium/include/pipe/p_defines.h @@ -174,6 +174,12 @@ enum pipe_texture_target { /** + * Surface layout + */ +#define PIPE_SURFACE_LAYOUT_LINEAR 0 + + +/** * Surface status */ #define PIPE_SURFACE_STATUS_UNDEFINED 0 diff --git a/src/gallium/include/pipe/p_format.h b/src/gallium/include/pipe/p_format.h index a4bd23c..9ba00f8 100644 --- a/src/gallium/include/pipe/p_format.h +++ b/src/gallium/include/pipe/p_format.h @@ -307,6 +307,9 @@ enum pipe_format { PIPE_FORMAT_R8G8B8_SNORM = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGB1, 1, 1, 1, 0, PIPE_FORMAT_TYPE_SNORM ), PIPE_FORMAT_R8G8B8A8_SNORM = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGBA, 1, 1, 1, 1, PIPE_FORMAT_TYPE_SNORM ), PIPE_FORMAT_R8G8B8X8_SNORM = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGB1, 1, 1, 1, 1, PIPE_FORMAT_TYPE_SNORM ), + PIPE_FORMAT_B6G5R5_SNORM = _PIPE_FORMAT_RGBAZS_1 ( _PIPE_FORMAT_BGR1, 6, 5, 5, 0, PIPE_FORMAT_TYPE_SNORM ), + PIPE_FORMAT_A8B8G8R8_SNORM = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_BGRA, 1, 1, 1, 1, PIPE_FORMAT_TYPE_SNORM ), + PIPE_FORMAT_X8B8G8R8_SNORM = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGB1, 1, 1, 1, 1, PIPE_FORMAT_TYPE_SNORM ), PIPE_FORMAT_R8_SSCALED = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_R001, 1, 0, 0, 0, PIPE_FORMAT_TYPE_SSCALED ), PIPE_FORMAT_R8G8_SSCALED = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RG01, 1, 1, 0, 0, PIPE_FORMAT_TYPE_SSCALED ), PIPE_FORMAT_R8G8B8_SSCALED = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGB1, 1, 1, 1, 0, PIPE_FORMAT_TYPE_SSCALED ), diff --git a/src/gallium/include/pipe/p_inlines.h b/src/gallium/include/pipe/p_inlines.h index 8eb604e..1e4b98e 100644 --- a/src/gallium/include/pipe/p_inlines.h +++ b/src/gallium/include/pipe/p_inlines.h @@ -39,21 +39,40 @@ extern "C" { #endif +/* XXX: these are a kludge. will fix when all surfaces are views into + * textures, and free-floating winsys surfaces go away. + */ static INLINE void * -pipe_surface_map(struct pipe_surface *surface) +pipe_surface_map( struct pipe_surface *surf, unsigned flags ) { - return (char *)surface->winsys->buffer_map( surface->winsys, surface->buffer, - PIPE_BUFFER_USAGE_CPU_WRITE | - PIPE_BUFFER_USAGE_CPU_READ ) - + surface->offset; + if (surf->texture) { + struct pipe_screen *screen = surf->texture->screen; + return surf->texture->screen->surface_map( screen, surf, flags ); + } + else { + struct pipe_winsys *winsys = surf->winsys; + char *map = (char *)winsys->buffer_map( winsys, surf->buffer, flags ); + if (map == NULL) + return NULL; + return (void *)(map + surf->offset); + } } static INLINE void -pipe_surface_unmap(struct pipe_surface *surface) +pipe_surface_unmap( struct pipe_surface *surf ) { - surface->winsys->buffer_unmap( surface->winsys, surface->buffer ); + if (surf->texture) { + struct pipe_screen *screen = surf->texture->screen; + surf->texture->screen->surface_unmap( screen, surf ); + } + else { + struct pipe_winsys *winsys = surf->winsys; + winsys->buffer_unmap( winsys, surf->buffer ); + } } + + /** * Set 'ptr' to point to 'surf' and update reference counting. * The old thing pointed to, if any, will be unreferenced first. @@ -66,9 +85,20 @@ pipe_surface_reference(struct pipe_surface **ptr, struct pipe_surface *surf) if (surf) surf->refcount++; - if (*ptr /* && --(*ptr)->refcount == 0 */) { - struct pipe_winsys *winsys = (*ptr)->winsys; - winsys->surface_release(winsys, ptr); + if (*ptr) { + + /* There are currently two sorts of surfaces... This needs to be + * fixed so that all surfaces are views into a texture. + */ + if ((*ptr)->texture) { + struct pipe_screen *screen = (*ptr)->texture->screen; + screen->tex_surface_release( screen, ptr ); + } + else { + struct pipe_winsys *winsys = (*ptr)->winsys; + winsys->surface_release(winsys, ptr); + } + assert(!*ptr); } diff --git a/src/gallium/include/pipe/p_screen.h b/src/gallium/include/pipe/p_screen.h index 26ac99d..cc8430d 100644 --- a/src/gallium/include/pipe/p_screen.h +++ b/src/gallium/include/pipe/p_screen.h @@ -89,6 +89,18 @@ struct pipe_screen { struct pipe_texture * (*texture_create)(struct pipe_screen *, const struct pipe_texture *templat); + /** + * Create a new texture object, using the given template info, but on top of + * existing memory. + * + * It is assumed that the buffer data is layed out according to the expected + * by the hardware. NULL will be returned if any inconsistency is found. + */ + struct pipe_texture * (*texture_blanket)(struct pipe_screen *, + const struct pipe_texture *templat, + const unsigned *pitch, + struct pipe_buffer *buffer); + void (*texture_release)(struct pipe_screen *, struct pipe_texture **pt); @@ -96,7 +108,22 @@ struct pipe_screen { struct pipe_surface *(*get_tex_surface)(struct pipe_screen *, struct pipe_texture *texture, unsigned face, unsigned level, - unsigned zslice); + unsigned zslice, + unsigned usage ); + + /* Surfaces allocated by the above must be released here: + */ + void (*tex_surface_release)( struct pipe_screen *, + struct pipe_surface ** ); + + + void *(*surface_map)( struct pipe_screen *, + struct pipe_surface *surface, + unsigned flags ); + + void (*surface_unmap)( struct pipe_screen *, + struct pipe_surface *surface ); + }; diff --git a/src/gallium/include/pipe/p_state.h b/src/gallium/include/pipe/p_state.h index 912d84e..e7ee8c9 100644 --- a/src/gallium/include/pipe/p_state.h +++ b/src/gallium/include/pipe/p_state.h @@ -271,12 +271,26 @@ struct pipe_surface unsigned width; unsigned height; unsigned pitch; /**< in pixels */ + unsigned layout; /**< PIPE_SURFACE_LAYOUT_x */ unsigned offset; /**< offset from start of buffer, in bytes */ unsigned refcount; + unsigned usage; /**< PIPE_BUFFER_USAGE_* */ + struct pipe_winsys *winsys; /**< winsys which owns/created the surface */ + + struct pipe_texture *texture; /**< optional texture into which this is a view */ + unsigned face; + unsigned level; + unsigned zslice; }; +#define PIPE_TEXTURE_USAGE_RENDER_TARGET 0x1 +#define PIPE_TEXTURE_USAGE_DISPLAY_TARGET 0x2 /* ie a backbuffer */ +#define PIPE_TEXTURE_USAGE_PRIMARY 0x4 /* ie a frontbuffer */ +#define PIPE_TEXTURE_USAGE_DEPTH_STENCIL 0x8 +#define PIPE_TEXTURE_USAGE_SAMPLER 0x10 + /** * Texture object. */ @@ -292,6 +306,8 @@ struct pipe_texture unsigned cpp:8; unsigned last_level:8; /**< Index of last mipmap level present/defined */ unsigned compressed:1; + + unsigned tex_usage; /* PIPE_TEXTURE_USAGE_* */ /* These are also refcounted: */ diff --git a/src/gallium/include/pipe/p_util.h b/src/gallium/include/pipe/p_util.h index 0e7e246..597354c 100644 --- a/src/gallium/include/pipe/p_util.h +++ b/src/gallium/include/pipe/p_util.h @@ -204,7 +204,9 @@ mem_dup(const void *src, uint size) #define MIN2( A, B ) ( (A)<(B) ? (A) : (B) ) #define MAX2( A, B ) ( (A)>(B) ? (A) : (B) ) +#ifndef Elements #define Elements(x) (sizeof(x)/sizeof((x)[0])) +#endif #define Offset(TYPE, MEMBER) ((unsigned)&(((TYPE *)NULL)->MEMBER)) /** diff --git a/src/gallium/include/pipe/p_winsys.h b/src/gallium/include/pipe/p_winsys.h index 3005ec2..7ebc285 100644 --- a/src/gallium/include/pipe/p_winsys.h +++ b/src/gallium/include/pipe/p_winsys.h @@ -86,11 +86,12 @@ struct pipe_winsys struct pipe_surface *surf, unsigned width, unsigned height, enum pipe_format format, - unsigned flags); + unsigned flags, + unsigned tex_usage); void (*surface_release)(struct pipe_winsys *ws, struct pipe_surface **s); - + /** * Buffer management. Buffer attributes are mostly fixed over its lifetime. * diff --git a/src/gallium/winsys/dri/intel/Makefile b/src/gallium/winsys/dri/intel/Makefile index 40654bb..c0ce2f9 100644 --- a/src/gallium/winsys/dri/intel/Makefile +++ b/src/gallium/winsys/dri/intel/Makefile @@ -2,7 +2,7 @@ TOP = ../../../../.. include $(TOP)/configs/current -LIBNAME = i915tex_dri.so +LIBNAME = i915_dri.so MINIGLX_SOURCES = server/intel_dri.c @@ -19,11 +19,14 @@ DRIVER_SOURCES = \ intel_context.c \ intel_lock.c \ intel_screen.c \ - intel_batchpool.c + ws_dri_bufmgr.c \ + ws_dri_drmpool.c \ + ws_dri_fencemgr.c \ + ws_dri_mallocpool.c \ + ws_dri_slabpool.c C_SOURCES = \ $(COMMON_GALLIUM_SOURCES) \ - $(COMMON_BM_SOURCES) \ $(DRIVER_SOURCES) ASM_SOURCES = diff --git a/src/gallium/winsys/dri/intel/intel_batchbuffer.c b/src/gallium/winsys/dri/intel/intel_batchbuffer.c index 5830b88..aa2eed5 100644 --- a/src/gallium/winsys/dri/intel/intel_batchbuffer.c +++ b/src/gallium/winsys/dri/intel/intel_batchbuffer.c @@ -25,108 +25,95 @@ * **************************************************************************/ -#include <errno.h> -#include <stdio.h> #include "intel_batchbuffer.h" #include "intel_context.h" -#include "intel_screen.h" -#include "intel_reg.h" -#include "drm.h" - -/* Relocations in kernel space: - * - pass dma buffer seperately - * - memory manager knows how to patch - * - pass list of dependent buffers - * - pass relocation list - * - * Either: - * - get back an offset for buffer to fire - * - memory manager knows how to fire buffer - * - * Really want the buffer to be AGP and pinned. - * - */ - -/* Cliprect fence: The highest fence protecting a dma buffer - * containing explicit cliprect information. Like the old drawable - * lock but irq-driven. X server must wait for this fence to expire - * before changing cliprects [and then doing sw rendering?]. For - * other dma buffers, the scheduler will grab current cliprect info - * and mix into buffer. X server must hold the lock while changing - * cliprects??? Make per-drawable. Need cliprects in shared memory - * -- beats storing them with every cmd buffer in the queue. - * - * ==> X server must wait for this fence to expire before touching the - * framebuffer with new cliprects. - * - * ==> Cliprect-dependent buffers associated with a - * cliprect-timestamp. All of the buffers associated with a timestamp - * must go to hardware before any buffer with a newer timestamp. - * - * ==> Dma should be queued per-drawable for correct X/GL - * synchronization. Or can fences be used for this? - * - * Applies to: Blit operations, metaops, X server operations -- X - * server automatically waits on its own dma to complete before - * modifying cliprects ??? - */ +#include <errno.h> +#if 0 static void -intel_dump_batchbuffer(uint offset, uint * ptr, uint count) +intel_dump_batchbuffer(GLuint offset, GLuint * ptr, GLuint count) { int i; - printf("\n\n\nSTART BATCH (%d dwords):\n", count / 4); - for (i = 0; i < count / 4; i += 1) - printf("\t0x%08x\n", ptr[i]); - printf("END BATCH\n\n\n"); + fprintf(stderr, "\n\n\nSTART BATCH (%d dwords):\n", count / 4); + for (i = 0; i < count / 4; i += 4) + fprintf(stderr, "0x%x:\t0x%08x 0x%08x 0x%08x 0x%08x\n", + offset + i * 4, ptr[i], ptr[i + 1], ptr[i + 2], ptr[i + 3]); + fprintf(stderr, "END BATCH\n\n\n"); +} +#endif + +static void +intel_realloc_relocs(struct intel_batchbuffer *batch, int num_relocs) +{ + unsigned long size = num_relocs * I915_RELOC0_STRIDE + I915_RELOC_HEADER; + + size *= sizeof(uint32_t); + batch->reloc = realloc(batch->reloc, size); + batch->reloc_size = num_relocs; } void intel_batchbuffer_reset(struct intel_batchbuffer *batch) { - int i; - - if (batch->map) { - driBOUnmap(batch->buffer); - batch->map = NULL; - } - /* * Get a new, free batchbuffer. */ - batch->size = BATCH_SZ; - driBOData(batch->buffer, batch->size, NULL, 0); + drmBO *bo; + struct drm_bo_info_req *req; + + driBOUnrefUserList(batch->list); + driBOResetList(batch->list); - driBOResetList(&batch->list); + batch->size = 4 * 4096; // ZZZ JB batch->intel->intelScreen->maxBatchSize; + driBOData(batch->buffer, batch->size, NULL, NULL, 0); /* - * Unreference buffers previously on the relocation list. + * Add the batchbuffer to the validate list. */ - for (i = 0; i < batch->nr_relocs; i++) { - struct buffer_reloc *r = &batch->reloc[i]; - driBOUnReference(r->buf); - } - batch->list_count = 0; - batch->nr_relocs = 0; - batch->flags = 0; + driBOAddListItem(batch->list, batch->buffer, + DRM_BO_FLAG_EXE | DRM_BO_FLAG_MEM_TT, + DRM_BO_FLAG_EXE | DRM_BO_MASK_MEM, + &batch->dest_location, &batch->node); + + req = &batch->node->bo_arg.d.req.bo_req; /* - * We don't refcount the batchbuffer itself since we can't destroy it - * while it's on the list. + * Set up information needed for us to make relocations + * relative to the underlying drm buffer objects. */ - driBOAddListItem(&batch->list, batch->buffer, - DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_EXE, - DRM_BO_MASK_MEM | DRM_BO_FLAG_EXE); + driReadLockKernelBO(); + bo = driBOKernel(batch->buffer); + req->presumed_offset = (uint64_t) bo->offset; + req->hint = DRM_BO_HINT_PRESUMED_OFFSET; + batch->drmBOVirtual = (uint8_t *) bo->virtual; + driReadUnlockKernelBO(); + /* + * Adjust the relocation buffer size. + */ + + if (batch->reloc_size > INTEL_MAX_RELOCS || + batch->reloc == NULL) + intel_realloc_relocs(batch, INTEL_DEFAULT_RELOCS); + + assert(batch->reloc != NULL); + batch->reloc[0] = 0; /* No relocs yet. */ + batch->reloc[1] = 1; /* Reloc type 1 */ + batch->reloc[2] = 0; /* Only a single relocation list. */ + batch->reloc[3] = 0; /* Only a single relocation list. */ batch->map = driBOMap(batch->buffer, DRM_BO_FLAG_WRITE, 0); + batch->poolOffset = driBOPoolOffset(batch->buffer); batch->ptr = batch->map; + batch->dirty_state = ~0; + batch->nr_relocs = 0; + batch->flags = 0; + batch->id = 0;//batch->intel->intelScreen->batch_id++; } - /*====================================================================== * Public functions */ @@ -141,121 +128,253 @@ intel_batchbuffer_alloc(struct intel_context *intel) &batch->buffer, 4096, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_EXE, 0); batch->last_fence = NULL; - driBOCreateList(20, &batch->list); + batch->list = driBOCreateList(20); + batch->reloc = NULL; intel_batchbuffer_reset(batch); return batch; } - void intel_batchbuffer_free(struct intel_batchbuffer *batch) { if (batch->last_fence) { driFenceFinish(batch->last_fence, - DRM_FENCE_TYPE_EXE | DRM_I915_FENCE_TYPE_RW, - GL_FALSE); - driFenceUnReference(batch->last_fence); - batch->last_fence = NULL; + DRM_FENCE_TYPE_EXE, GL_FALSE); + driFenceUnReference(&batch->last_fence); } if (batch->map) { driBOUnmap(batch->buffer); batch->map = NULL; } driBOUnReference(batch->buffer); + driBOFreeList(batch->list); + if (batch->reloc) + free(batch->reloc); batch->buffer = NULL; free(batch); } +void +intel_offset_relocation(struct intel_batchbuffer *batch, + unsigned pre_add, + struct _DriBufferObject *driBO, + uint64_t val_flags, + uint64_t val_mask) +{ + int itemLoc; + struct _drmBONode *node; + uint32_t *reloc; + struct drm_bo_info_req *req; + + driBOAddListItem(batch->list, driBO, val_flags, val_mask, + &itemLoc, &node); + req = &node->bo_arg.d.req.bo_req; + + if (!(req->hint & DRM_BO_HINT_PRESUMED_OFFSET)) { + + /* + * Stop other threads from tampering with the underlying + * drmBO while we're reading its offset. + */ + + driReadLockKernelBO(); + req->presumed_offset = (uint64_t) driBOKernel(driBO)->offset; + driReadUnlockKernelBO(); + req->hint = DRM_BO_HINT_PRESUMED_OFFSET; + } + + pre_add += driBOPoolOffset(driBO); + + if (batch->nr_relocs == batch->reloc_size) + intel_realloc_relocs(batch, batch->reloc_size * 2); + + reloc = batch->reloc + + (I915_RELOC_HEADER + batch->nr_relocs * I915_RELOC0_STRIDE); + + reloc[0] = ((uint8_t *)batch->ptr - batch->drmBOVirtual); + intel_batchbuffer_emit_dword(batch, req->presumed_offset + pre_add); + reloc[1] = pre_add; + reloc[2] = itemLoc; + reloc[3] = batch->dest_location; + batch->nr_relocs++; +} static void -intel_batch_ioctl(struct intel_context *intel, - uint start_offset, uint used, boolean allow_unlock) +i915_drm_copy_reply(const struct drm_bo_info_rep * rep, drmBO * buf) { - drmI830BatchBuffer batch; - - batch.start = start_offset; - batch.used = used; - batch.cliprects = NULL; /* unused */ - batch.num_cliprects = 0; - batch.DR1 = 0; - batch.DR4 = 0; /* still need this ? */ - - DBG(IOCTL, "%s: 0x%x..0x%x DR4: %x cliprects: %d\n", - __FUNCTION__, - batch.start, - batch.start + batch.used * 4, batch.DR4, batch.num_cliprects); - - if (drmCommandWrite(intel->driFd, DRM_I830_BATCHBUFFER, &batch, - sizeof(batch))) { - printf("DRM_I830_BATCHBUFFER: %d\n", -errno); - UNLOCK_HARDWARE(intel); - exit(1); - } + buf->handle = rep->handle; + buf->flags = rep->flags; + buf->size = rep->size; + buf->offset = rep->offset; + buf->mapHandle = rep->arg_handle; + buf->proposedFlags = rep->proposed_flags; + buf->start = rep->buffer_start; + buf->fenceFlags = rep->fence_flags; + buf->replyFlags = rep->rep_flags; + buf->pageAlignment = rep->page_alignment; } +static int +i915_execbuf(struct intel_batchbuffer *batch, + GLuint used, + GLboolean ignore_cliprects, + drmBOList *list, + struct drm_i915_execbuffer *ea) +{ + struct intel_context *intel = batch->intel; + drmBONode *node; + drmMMListHead *l; + struct drm_i915_op_arg *arg, *first; + struct drm_bo_op_req *req; + struct drm_bo_info_rep *rep; + uint64_t *prevNext = NULL; + drmBO *buf; + int ret = 0; + uint32_t count = 0; + + first = NULL; + for (l = list->list.next; l != &list->list; l = l->next) { + node = DRMLISTENTRY(drmBONode, l, head); + + arg = &node->bo_arg; + req = &arg->d.req; + + if (!first) + first = arg; + + if (prevNext) + *prevNext = (unsigned long)arg; + + prevNext = &arg->next; + req->bo_req.handle = node->buf->handle; + req->op = drm_bo_validate; + req->bo_req.flags = node->arg0; + req->bo_req.mask = node->arg1; + req->bo_req.hint |= 0; + count++; + } + + memset(ea, 0, sizeof(*ea)); + ea->num_buffers = count; + ea->batch.start = batch->poolOffset; + ea->batch.used = used; +#if 0 /* ZZZ JB: no cliprects used */ + ea->batch.cliprects = intel->pClipRects; + ea->batch.num_cliprects = ignore_cliprects ? 0 : intel->numClipRects; + ea->batch.DR1 = 0; + ea->batch.DR4 = 0;((((GLuint) intel->drawX) & 0xffff) | + (((GLuint) intel->drawY) << 16)); +#else + ea->batch.cliprects = NULL; + ea->batch.num_cliprects = 0; + ea->batch.DR1 = 0; + ea->batch.DR4 = 0; +#endif + ea->fence_arg.flags = DRM_I915_FENCE_FLAG_FLUSHED; + ea->ops_list = (unsigned long) first; + first->reloc_ptr = (unsigned long) batch->reloc; + batch->reloc[0] = batch->nr_relocs; + + //return -EFAULT; + do { + ret = drmCommandWriteRead(intel->driFd, DRM_I915_EXECBUFFER, ea, + sizeof(*ea)); + } while (ret == -EAGAIN); + + if (ret != 0) + return ret; + + for (l = list->list.next; l != &list->list; l = l->next) { + node = DRMLISTENTRY(drmBONode, l, head); + arg = &node->bo_arg; + rep = &arg->d.rep.bo_info; + + if (!arg->handled) { + return -EFAULT; + } + if (arg->d.rep.ret) + return arg->d.rep.ret; + + buf = node->buf; + i915_drm_copy_reply(rep, buf); + } + return 0; +} /* TODO: Push this whole function into bufmgr. */ -static void +static struct _DriFenceObject * do_flush_locked(struct intel_batchbuffer *batch, - uint used, boolean allow_unlock) + GLuint used, + GLboolean ignore_cliprects, GLboolean allow_unlock) { - uint *ptr; - uint i, fenceFlags; + struct intel_context *intel = batch->intel; struct _DriFenceObject *fo; + drmFence fence; + drmBOList *boList; + struct drm_i915_execbuffer ea; + int ret = 0; + + driBOValidateUserList(batch->list); + boList = driGetdrmBOList(batch->list); + +#if 0 /* ZZZ JB Allways run */ + if (!(intel->numClipRects == 0 && !ignore_cliprects)) { +#else + if (1) { +#endif + ret = i915_execbuf(batch, used, ignore_cliprects, boList, &ea); + } else { + driPutdrmBOList(batch->list); + fo = NULL; + goto out; + } + driPutdrmBOList(batch->list); + if (ret) + abort(); - driBOValidateList(batch->intel->driFd, &batch->list); - - /* Apply the relocations. This nasty map indicates to me that the - * whole task should be done internally by the memory manager, and - * that dma buffers probably need to be pinned within agp space. - */ - ptr = (uint *) driBOMap(batch->buffer, DRM_BO_FLAG_WRITE, - DRM_BO_HINT_ALLOW_UNFENCED_MAP); + if (ea.fence_arg.error != 0) { - for (i = 0; i < batch->nr_relocs; i++) { - struct buffer_reloc *r = &batch->reloc[i]; + /* + * The hardware has been idled by the kernel. + * Don't fence the driBOs. + */ - ptr[r->offset / 4] = driBOOffset(r->buf) + r->delta; + if (batch->last_fence) + driFenceUnReference(&batch->last_fence); +#if 0 /* ZZZ JB: no _mesa_* funcs in gallium */ + _mesa_printf("fence error\n"); +#endif + batch->last_fence = NULL; + fo = NULL; + goto out; } - if (0) - intel_dump_batchbuffer(0, ptr, used); + fence.handle = ea.fence_arg.handle; + fence.fence_class = ea.fence_arg.fence_class; + fence.type = ea.fence_arg.type; + fence.flags = ea.fence_arg.flags; + fence.signaled = ea.fence_arg.signaled; - driBOUnmap(batch->buffer); - batch->map = NULL; - - intel_batch_ioctl(batch->intel, - driBOOffset(batch->buffer), - used, allow_unlock); - - /* - * Kernel fencing. The flags tells the kernel that we've - * programmed an MI_FLUSH. - */ - fenceFlags = DRM_I915_FENCE_FLAG_FLUSHED; - fo = driFenceBuffers(batch->intel->driFd, "Batch fence", fenceFlags); + fo = driBOFenceUserList(batch->intel->intelScreen->mgr, batch->list, + "SuperFence", &fence); + if (driFenceType(fo) & DRM_I915_FENCE_TYPE_RW) { + if (batch->last_fence) + driFenceUnReference(&batch->last_fence); /* - * User space fencing. - */ - driBOFence(batch->buffer, fo); - - if (driFenceType(fo) == DRM_FENCE_TYPE_EXE) { - /* - * Oops. We only validated a batch buffer. This means we - * didn't do any proper rendering. Discard this fence object. - */ - driFenceUnReference(fo); - } - else { - driFenceUnReference(batch->last_fence); - batch->last_fence = fo; - for (i = 0; i < batch->nr_relocs; i++) { - struct buffer_reloc *r = &batch->reloc[i]; - driBOFence(r->buf, fo); - } - } + * FIXME: Context last fence?? + */ + batch->last_fence = fo; + driFenceReference(fo); + } + out: +#if 0 /* ZZZ JB: fix this */ + intel->vtbl.lost_hardware(intel); +#else + (void)intel; +#endif + return fo; } @@ -263,29 +382,43 @@ struct _DriFenceObject * intel_batchbuffer_flush(struct intel_batchbuffer *batch) { struct intel_context *intel = batch->intel; - uint used = batch->ptr - batch->map; - const boolean was_locked = intel->locked; + GLuint used = batch->ptr - batch->map; + GLboolean was_locked = intel->locked; + struct _DriFenceObject *fence; - if (used == 0) + if (used == 0) { + driFenceReference(batch->last_fence); return batch->last_fence; - -#define MI_FLUSH ((0 << 29) | (4 << 23)) + } /* Add the MI_BATCH_BUFFER_END. Always add an MI_FLUSH - this is a * performance drain that we would like to avoid. */ +#if 0 /* ZZZ JB: what should we do here? */ if (used & 4) { - ((int *) batch->ptr)[0] = MI_FLUSH; + ((int *) batch->ptr)[0] = intel->vtbl.flush_cmd(); ((int *) batch->ptr)[1] = 0; ((int *) batch->ptr)[2] = MI_BATCH_BUFFER_END; used += 12; } else { - ((int *) batch->ptr)[0] = MI_FLUSH; + ((int *) batch->ptr)[0] = intel->vtbl.flush_cmd(); ((int *) batch->ptr)[1] = MI_BATCH_BUFFER_END; used += 8; } - +#else + if (used & 4) { + ((int *) batch->ptr)[0] = ((0<<29)|(4<<23)); // MI_FLUSH; + ((int *) batch->ptr)[1] = 0; + ((int *) batch->ptr)[2] = (0xA<<23); // MI_BATCH_BUFFER_END; + used += 12; + } + else { + ((int *) batch->ptr)[0] = ((0<<29)|(4<<23)); // MI_FLUSH; + ((int *) batch->ptr)[1] = (0xA<<23); // MI_BATCH_BUFFER_END; + used += 8; + } +#endif driBOUnmap(batch->buffer); batch->ptr = NULL; batch->map = NULL; @@ -296,7 +429,8 @@ intel_batchbuffer_flush(struct intel_batchbuffer *batch) if (!was_locked) LOCK_HARDWARE(intel); - do_flush_locked(batch, used, GL_FALSE); + fence = do_flush_locked(batch, used, !(batch->flags & INTEL_BATCH_CLIPRECTS), + GL_FALSE); if (!was_locked) UNLOCK_HARDWARE(intel); @@ -304,52 +438,20 @@ intel_batchbuffer_flush(struct intel_batchbuffer *batch) /* Reset the buffer: */ intel_batchbuffer_reset(batch); - return batch->last_fence; + return fence; } - void intel_batchbuffer_finish(struct intel_batchbuffer *batch) { struct _DriFenceObject *fence = intel_batchbuffer_flush(batch); - if (fence) { - driFenceReference(fence); - driFenceFinish(fence, - DRM_FENCE_TYPE_EXE | DRM_I915_FENCE_TYPE_RW, - GL_FALSE); - driFenceUnReference(fence); - } + driFenceFinish(fence, driFenceType(fence), GL_FALSE); + driFenceUnReference(&fence); } - -/* This is the only way buffers get added to the validate list. - */ -boolean -intel_batchbuffer_emit_reloc(struct intel_batchbuffer *batch, - struct _DriBufferObject *buffer, - uint flags, uint mask, uint delta) -{ - assert(batch->nr_relocs < MAX_RELOCS); - - driBOAddListItem(&batch->list, buffer, flags, mask); - - { - struct buffer_reloc *r = &batch->reloc[batch->nr_relocs++]; - driBOReference(buffer); - r->buf = buffer; - r->offset = batch->ptr - batch->map; - r->delta = delta; - *(uint *) batch->ptr = 0x12345678; - } - - batch->ptr += 4; - return GL_TRUE; -} - - void intel_batchbuffer_data(struct intel_batchbuffer *batch, - const void *data, uint bytes, uint flags) + const void *data, GLuint bytes, GLuint flags) { assert((bytes & 3) == 0); intel_batchbuffer_require_space(batch, bytes, flags); diff --git a/src/gallium/winsys/dri/intel/intel_batchbuffer.h b/src/gallium/winsys/dri/intel/intel_batchbuffer.h index caf6870..9e4b804 100644 --- a/src/gallium/winsys/dri/intel/intel_batchbuffer.h +++ b/src/gallium/winsys/dri/intel/intel_batchbuffer.h @@ -1,54 +1,20 @@ -/************************************************************************** - * - * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. - * All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the - * "Software"), to deal in the Software without restriction, including - * without limitation the rights to use, copy, modify, merge, publish, - * distribute, sub license, and/or sell copies of the Software, and to - * permit persons to whom the Software is furnished to do so, subject to - * the following conditions: - * - * The above copyright notice and this permission notice (including the - * next paragraph) shall be included in all copies or substantial portions - * of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF - * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. - * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR - * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, - * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE - * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - * - **************************************************************************/ - #ifndef INTEL_BATCHBUFFER_H #define INTEL_BATCHBUFFER_H -#include "pipe/p_debug.h" -#include "pipe/p_compiler.h" -#include "dri_bufmgr.h" +#include "mtypes.h" +#include "ws_dri_bufmgr.h" struct intel_context; #define BATCH_SZ 16384 #define BATCH_RESERVED 16 -#define MAX_RELOCS 4096 +#define INTEL_DEFAULT_RELOCS 100 +#define INTEL_MAX_RELOCS 400 #define INTEL_BATCH_NO_CLIPRECTS 0x1 #define INTEL_BATCH_CLIPRECTS 0x2 -struct buffer_reloc -{ - struct _DriBufferObject *buf; - uint offset; - uint delta; /* not needed? */ -}; - struct intel_batchbuffer { struct bufmgr *bm; @@ -56,19 +22,30 @@ struct intel_batchbuffer struct _DriBufferObject *buffer; struct _DriFenceObject *last_fence; - uint flags; + GLuint flags; - drmBOList list; - uint list_count; - ubyte *map; - ubyte *ptr; + struct _DriBufferList *list; + GLuint list_count; + GLubyte *map; + GLubyte *ptr; - struct buffer_reloc reloc[MAX_RELOCS]; - uint nr_relocs; - uint size; + uint32_t *reloc; + GLuint reloc_size; + GLuint nr_relocs; + + GLuint size; + + GLuint dirty_state; + GLuint id; + + uint32_t poolOffset; + uint8_t *drmBOVirtual; + struct _drmBONode *node; /* Validation list node for this buffer */ + int dest_location; /* Validation list sequence for this buffer */ }; -struct intel_batchbuffer *intel_batchbuffer_alloc(struct intel_context *intel); +struct intel_batchbuffer *intel_batchbuffer_alloc(struct intel_context + *intel); void intel_batchbuffer_free(struct intel_batchbuffer *batch); @@ -82,26 +59,28 @@ void intel_batchbuffer_reset(struct intel_batchbuffer *batch); /* Unlike bmBufferData, this currently requires the buffer be mapped. - * Consider it a convenience function wrapping multiple + * Consider it a convenience function wrapping multple * intel_buffer_dword() calls. */ void intel_batchbuffer_data(struct intel_batchbuffer *batch, - const void *data, uint bytes, uint flags); + const void *data, GLuint bytes, GLuint flags); void intel_batchbuffer_release_space(struct intel_batchbuffer *batch, - uint bytes); + GLuint bytes); -boolean intel_batchbuffer_emit_reloc(struct intel_batchbuffer *batch, - struct _DriBufferObject *buffer, - uint flags, - uint mask, uint offset); +void +intel_offset_relocation(struct intel_batchbuffer *batch, + unsigned pre_add, + struct _DriBufferObject *driBO, + uint64_t val_flags, + uint64_t val_mask); /* Inline functions - might actually be better off with these * non-inlined. Certainly better off switching all command packets to * be passed as structs rather than dwords, but that's a little bit of * work... */ -static INLINE uint +static INLINE GLuint intel_batchbuffer_space(struct intel_batchbuffer *batch) { return (batch->size - BATCH_RESERVED) - (batch->ptr - batch->map); @@ -109,22 +88,26 @@ intel_batchbuffer_space(struct intel_batchbuffer *batch) static INLINE void -intel_batchbuffer_emit_dword(struct intel_batchbuffer *batch, uint dword) +intel_batchbuffer_emit_dword(struct intel_batchbuffer *batch, GLuint dword) { assert(batch->map); assert(intel_batchbuffer_space(batch) >= 4); - *(uint *) (batch->ptr) = dword; + *(GLuint *) (batch->ptr) = dword; batch->ptr += 4; } static INLINE void intel_batchbuffer_require_space(struct intel_batchbuffer *batch, - uint sz, uint flags) + GLuint sz, GLuint flags) { + struct _DriFenceObject *fence; + assert(sz < batch->size - 8); if (intel_batchbuffer_space(batch) < sz || - (batch->flags != 0 && flags != 0 && batch->flags != flags)) - intel_batchbuffer_flush(batch); + (batch->flags != 0 && flags != 0 && batch->flags != flags)) { + fence = intel_batchbuffer_flush(batch); + driFenceUnReference(&fence); + } batch->flags |= flags; } @@ -134,14 +117,15 @@ intel_batchbuffer_require_space(struct intel_batchbuffer *batch, #define BATCH_LOCALS #define BEGIN_BATCH(n, flags) do { \ + assert(!intel->prim.flush); \ intel_batchbuffer_require_space(intel->batch, (n)*4, flags); \ } while (0) #define OUT_BATCH(d) intel_batchbuffer_emit_dword(intel->batch, d) #define OUT_RELOC(buf,flags,mask,delta) do { \ - assert((delta) >= 0); \ - intel_batchbuffer_emit_reloc(intel->batch, buf, flags, mask, delta); \ + assert((delta) >= 0); \ + intel_offset_relocation(intel->batch, delta, buf, flags, mask); \ } while (0) #define ADVANCE_BATCH() do { } while(0) diff --git a/src/gallium/winsys/dri/intel/intel_batchpool.c b/src/gallium/winsys/dri/intel/intel_batchpool.c deleted file mode 100644 index ce154c7..0000000 --- a/src/gallium/winsys/dri/intel/intel_batchpool.c +++ /dev/null @@ -1,427 +0,0 @@ -/************************************************************************** - * - * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA - * All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the - * "Software"), to deal in the Software without restriction, including - * without limitation the rights to use, copy, modify, merge, publish, - * distribute, sub license, and/or sell copies of the Software, and to - * permit persons to whom the Software is furnished to do so, subject to - * the following conditions: - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL - * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, - * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR - * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE - * USE OR OTHER DEALINGS IN THE SOFTWARE. - * - * The above copyright notice and this permission notice (including the - * next paragraph) shall be included in all copies or substantial portions - * of the Software. - * - * - **************************************************************************/ -/* - * Authors: Thomas Hellström <thomas-at-tungstengraphics-dot-com> - */ - -/** - * XXX NOTE: there are no intel dependencies in this file. - * Rename to dri_batchpool.c? - */ - -#include <xf86drm.h> -#include <stdlib.h> -#include <stdio.h> -#include <errno.h> - -#include "pipe/p_compiler.h" -#include "pipe/p_thread.h" - -#include "dri_bufpool.h" -#include "dri_bufmgr.h" -#include "intel_batchpool.h" - - -typedef struct -{ - drmMMListHead head; - struct _BPool *parent; - struct _DriFenceObject *fence; - unsigned long start; - int unfenced; - int mapped; -} BBuf; - -typedef struct _BPool -{ - _glthread_Mutex mutex; - unsigned long bufSize; - unsigned poolSize; - unsigned numFree; - unsigned numTot; - unsigned numDelayed; - unsigned checkDelayed; - drmMMListHead free; - drmMMListHead delayed; - drmMMListHead head; - drmBO kernelBO; - void *virtual; - BBuf *bufs; -} BPool; - - -static BPool * -createBPool(int fd, unsigned long bufSize, unsigned numBufs, unsigned flags, - unsigned checkDelayed) -{ - BPool *p = (BPool *) malloc(sizeof(*p)); - BBuf *buf; - int i; - - if (!p) - return NULL; - - p->bufs = (BBuf *) malloc(numBufs * sizeof(*p->bufs)); - if (!p->bufs) { - free(p); - return NULL; - } - - DRMINITLISTHEAD(&p->free); - DRMINITLISTHEAD(&p->head); - DRMINITLISTHEAD(&p->delayed); - - p->numTot = numBufs; - p->numFree = numBufs; - p->bufSize = bufSize; - p->numDelayed = 0; - p->checkDelayed = checkDelayed; - - _glthread_INIT_MUTEX(p->mutex); - - if (drmBOCreate(fd, 0, numBufs * bufSize, 0, NULL, drm_bo_type_dc, - flags, DRM_BO_HINT_DONT_FENCE, &p->kernelBO)) { - free(p->bufs); - free(p); - return NULL; - } - if (drmBOMap(fd, &p->kernelBO, DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE, 0, - &p->virtual)) { - drmBODestroy(fd, &p->kernelBO); - free(p->bufs); - free(p); - return NULL; - } - - /* - * We unmap the buffer so that we can validate it later. Note that this is - * just a synchronizing operation. The buffer will have a virtual mapping - * until it is destroyed. - */ - - drmBOUnmap(fd, &p->kernelBO); - - buf = p->bufs; - for (i = 0; i < numBufs; ++i) { - buf->parent = p; - buf->fence = NULL; - buf->start = i * bufSize; - buf->mapped = 0; - buf->unfenced = 0; - DRMLISTADDTAIL(&buf->head, &p->free); - buf++; - } - - return p; -} - - -static void -pool_checkFree(BPool * p, int wait) -{ - drmMMListHead *list, *prev; - BBuf *buf; - int signaled = 0; - int i; - - list = p->delayed.next; - - if (p->numDelayed > 3) { - for (i = 0; i < p->numDelayed; i += 3) { - list = list->next; - } - } - - prev = list->prev; - for (; list != &p->delayed; list = prev, prev = list->prev) { - - buf = DRMLISTENTRY(BBuf, list, head); - - if (!signaled) { - if (wait) { - driFenceFinish(buf->fence, DRM_FENCE_TYPE_EXE, 1); - signaled = 1; - } - else { - signaled = driFenceSignaled(buf->fence, DRM_FENCE_TYPE_EXE); - } - } - - if (!signaled) - break; - - driFenceUnReference(buf->fence); - buf->fence = NULL; - DRMLISTDEL(list); - p->numDelayed--; - DRMLISTADD(list, &p->free); - p->numFree++; - } -} - -static void * -pool_create(struct _DriBufferPool *pool, - unsigned long size, unsigned flags, unsigned hint, - unsigned alignment) -{ - BPool *p = (BPool *) pool->data; - - drmMMListHead *item; - - if (alignment && (alignment != 4096)) - return NULL; - - _glthread_LOCK_MUTEX(p->mutex); - - if (p->numFree == 0) - pool_checkFree(p, TRUE); - - if (p->numFree == 0) { - fprintf(stderr, "Out of fixed size buffer objects\n"); - BM_CKFATAL(-ENOMEM); - } - - item = p->free.next; - - if (item == &p->free) { - fprintf(stderr, "Fixed size buffer pool corruption\n"); - } - - DRMLISTDEL(item); - --p->numFree; - - _glthread_UNLOCK_MUTEX(p->mutex); - return (void *) DRMLISTENTRY(BBuf, item, head); -} - - -static int -pool_destroy(struct _DriBufferPool *pool, void *private) -{ - BBuf *buf = (BBuf *) private; - BPool *p = buf->parent; - - _glthread_LOCK_MUTEX(p->mutex); - - if (buf->fence) { - DRMLISTADDTAIL(&buf->head, &p->delayed); - p->numDelayed++; - } - else { - buf->unfenced = 0; - DRMLISTADD(&buf->head, &p->free); - p->numFree++; - } - - if ((p->numDelayed % p->checkDelayed) == 0) - pool_checkFree(p, 0); - - _glthread_UNLOCK_MUTEX(p->mutex); - return 0; -} - - -static int -pool_map(struct _DriBufferPool *pool, void *private, unsigned flags, - int hint, void **virtual) -{ - - BBuf *buf = (BBuf *) private; - BPool *p = buf->parent; - - _glthread_LOCK_MUTEX(p->mutex); - - /* - * Currently Mesa doesn't have any condition variables to resolve this - * cleanly in a multithreading environment. - * We bail out instead. - */ - - if (buf->mapped) { - fprintf(stderr, "Trying to map already mapped buffer object\n"); - BM_CKFATAL(-EINVAL); - } - -#if 0 - if (buf->unfenced && !(hint & DRM_BO_HINT_ALLOW_UNFENCED_MAP)) { - fprintf(stderr, "Trying to map an unfenced buffer object 0x%08x" - " 0x%08x %d\n", hint, flags, buf->start); - BM_CKFATAL(-EINVAL); - } - -#endif - - if (buf->fence) { - _glthread_UNLOCK_MUTEX(p->mutex); - return -EBUSY; - } - - buf->mapped = TRUE; - *virtual = (unsigned char *) p->virtual + buf->start; - _glthread_UNLOCK_MUTEX(p->mutex); - return 0; -} - -static int -pool_waitIdle(struct _DriBufferPool *pool, void *private, int lazy) -{ - BBuf *buf = (BBuf *) private; - driFenceFinish(buf->fence, 0x0, lazy); - return 0; -} - -static int -pool_unmap(struct _DriBufferPool *pool, void *private) -{ - BBuf *buf = (BBuf *) private; - - buf->mapped = 0; - return 0; -} - -static unsigned long -pool_offset(struct _DriBufferPool *pool, void *private) -{ - BBuf *buf = (BBuf *) private; - BPool *p = buf->parent; - - return p->kernelBO.offset + buf->start; -} - -static unsigned -pool_flags(struct _DriBufferPool *pool, void *private) -{ - BPool *p = (BPool *) pool->data; - - return p->kernelBO.flags; -} - -static unsigned long -pool_size(struct _DriBufferPool *pool, void *private) -{ - BPool *p = (BPool *) pool->data; - - return p->bufSize; -} - - -static int -pool_fence(struct _DriBufferPool *pool, void *private, - struct _DriFenceObject *fence) -{ - BBuf *buf = (BBuf *) private; - BPool *p = buf->parent; - - _glthread_LOCK_MUTEX(p->mutex); - if (buf->fence) { - driFenceUnReference(buf->fence); - } - buf->fence = fence; - buf->unfenced = 0; - driFenceReference(buf->fence); - _glthread_UNLOCK_MUTEX(p->mutex); - - return 0; -} - -static drmBO * -pool_kernel(struct _DriBufferPool *pool, void *private) -{ - BBuf *buf = (BBuf *) private; - BPool *p = buf->parent; - - return &p->kernelBO; -} - -static int -pool_validate(struct _DriBufferPool *pool, void *private) -{ - BBuf *buf = (BBuf *) private; - BPool *p = buf->parent; - _glthread_LOCK_MUTEX(p->mutex); - buf->unfenced = TRUE; - _glthread_UNLOCK_MUTEX(p->mutex); - return 0; -} - -static void -pool_takedown(struct _DriBufferPool *pool) -{ - BPool *p = (BPool *) pool->data; - - /* - * Wait on outstanding fences. - */ - - _glthread_LOCK_MUTEX(p->mutex); - while ((p->numFree < p->numTot) && p->numDelayed) { - _glthread_UNLOCK_MUTEX(p->mutex); - sched_yield(); - pool_checkFree(p, TRUE); - _glthread_LOCK_MUTEX(p->mutex); - } - - drmBODestroy(pool->fd, &p->kernelBO); - free(p->bufs); - _glthread_UNLOCK_MUTEX(p->mutex); - free(p); - free(pool); -} - - -struct _DriBufferPool * -driBatchPoolInit(int fd, unsigned flags, - unsigned long bufSize, - unsigned numBufs, unsigned checkDelayed) -{ - struct _DriBufferPool *pool; - - pool = (struct _DriBufferPool *) malloc(sizeof(*pool)); - if (!pool) - return NULL; - - pool->data = createBPool(fd, bufSize, numBufs, flags, checkDelayed); - if (!pool->data) - return NULL; - - pool->fd = fd; - pool->map = &pool_map; - pool->unmap = &pool_unmap; - pool->destroy = &pool_destroy; - pool->offset = &pool_offset; - pool->flags = &pool_flags; - pool->size = &pool_size; - pool->create = &pool_create; - pool->fence = &pool_fence; - pool->kernel = &pool_kernel; - pool->validate = &pool_validate; - pool->waitIdle = &pool_waitIdle; - pool->setstatic = NULL; - pool->takeDown = &pool_takedown; - return pool; -} diff --git a/src/gallium/winsys/dri/intel/intel_context.c b/src/gallium/winsys/dri/intel/intel_context.c index 8eba33c..cc2fc11 100644 --- a/src/gallium/winsys/dri/intel/intel_context.c +++ b/src/gallium/winsys/dri/intel/intel_context.c @@ -162,6 +162,7 @@ intelCreateContext(const __GLcontextModes * visual, * memory pools */ DRM_LIGHT_LOCK(sPriv->fd, &sPriv->pSAREA->lock, driContextPriv->hHWContext); + // ZZZ JB should be per screen and not be done per context havePools = intelCreatePools(sPriv); DRM_UNLOCK(sPriv->fd, &sPriv->pSAREA->lock, driContextPriv->hHWContext); if (!havePools) @@ -234,12 +235,12 @@ intelDestroyContext(__DRIcontextPrivate * driContextPriv) if (intel->last_swap_fence) { driFenceFinish(intel->last_swap_fence, DRM_FENCE_TYPE_EXE, GL_TRUE); - driFenceUnReference(intel->last_swap_fence); + driFenceUnReference(&intel->last_swap_fence); intel->last_swap_fence = NULL; } if (intel->first_swap_fence) { driFenceFinish(intel->first_swap_fence, DRM_FENCE_TYPE_EXE, GL_TRUE); - driFenceUnReference(intel->first_swap_fence); + driFenceUnReference(&intel->first_swap_fence); intel->first_swap_fence = NULL; } diff --git a/src/gallium/winsys/dri/intel/intel_context.h b/src/gallium/winsys/dri/intel/intel_context.h index 4543098..597dc13 100644 --- a/src/gallium/winsys/dri/intel/intel_context.h +++ b/src/gallium/winsys/dri/intel/intel_context.h @@ -28,7 +28,7 @@ #ifndef INTEL_CONTEXT_H #define INTEL_CONTEXT_H - +#include <stdint.h> #include "drm.h" #include "pipe/p_debug.h" diff --git a/src/gallium/winsys/dri/intel/intel_screen.c b/src/gallium/winsys/dri/intel/intel_screen.c index 9e31c01..58bebbe 100644 --- a/src/gallium/winsys/dri/intel/intel_screen.c +++ b/src/gallium/winsys/dri/intel/intel_screen.c @@ -32,12 +32,12 @@ #include "intel_context.h" #include "intel_screen.h" #include "intel_batchbuffer.h" -#include "intel_batchpool.h" +//#include "intel_batchpool.h" #include "intel_swapbuffers.h" #include "intel_winsys.h" #include "i830_dri.h" -#include "dri_bufpool.h" +#include "ws_dri_bufpool.h" #include "pipe/p_context.h" #include "state_tracker/st_public.h" @@ -132,6 +132,7 @@ intelUpdateScreenRotation(__DRIscreenPrivate * sPriv, drmI830Sarea * sarea) assert( sarea->front_size >= intelScreen->front.pitch * intelScreen->front.height ); +#if 0 /* JB not important */ if (!sarea->front_handle) return; @@ -142,30 +143,41 @@ intelUpdateScreenRotation(__DRIscreenPrivate * sPriv, drmI830Sarea * sarea) fprintf(stderr, "drmMap(frontbuffer) failed!\n"); return; } +#endif +#if 0 /* JB */ if (intelScreen->staticPool) { driGenBuffers(intelScreen->staticPool, "static region", 1, &intelScreen->front.buffer, 64, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_NO_MOVE | DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE, 0); - + driBOSetStatic(intelScreen->front.buffer, intelScreen->front.offset, intelScreen->front.pitch * intelScreen->front.height, intelScreen->front.map, 0); } +#else + if (intelScreen->staticPool) { + if (intelScreen->front.buffer) + driBOUnReference(intelScreen->front.buffer); + driGenBuffers(intelScreen->staticPool, "front", 1, &intelScreen->front.buffer, 0, 0, 0); + driBOSetReferenced(intelScreen->front.buffer, sarea->front_bo_handle); + } +#endif } boolean intelCreatePools(__DRIscreenPrivate * sPriv) { - unsigned batchPoolSize = 1024*1024; + //unsigned batchPoolSize = 1024*1024; struct intel_screen *intelScreen = intel_screen(sPriv); if (intelScreen->havePools) return GL_TRUE; +#if 0 /* ZZZ JB fix this */ intelScreen->staticPool = driDRMStaticPoolInit(sPriv->fd); if (!intelScreen->staticPool) return GL_FALSE; @@ -181,7 +193,17 @@ intelCreatePools(__DRIscreenPrivate * sPriv) fprintf(stderr, "Failed to initialize batch pool - possible incorrect agpgart installed\n"); return GL_FALSE; } - +#else + intelScreen->staticPool = driDRMPoolInit(sPriv->fd); + intelScreen->batchPool = driSlabPoolInit(sPriv->fd, + DRM_BO_FLAG_EXE | + DRM_BO_FLAG_MEM_TT, + DRM_BO_FLAG_EXE | + DRM_BO_FLAG_MEM_TT, + 4 * 4096, //intelScreen->maxBatchSize, + 1, 40, 16*16384, 0, + intelScreen->fMan); +#endif intelScreen->havePools = GL_TRUE; intelUpdateScreenRotation(sPriv, intelScreen->sarea); @@ -240,7 +262,26 @@ intelInitDriver(__DRIscreenPrivate * sPriv) (*glx_enable_extension) (psc, "GLX_SGI_make_current_read"); } - intelScreen->winsys = intel_create_pipe_winsys(sPriv->fd); + + +#if 1 // ZZZ JB + intelScreen->mgr = driFenceMgrTTMInit(sPriv->fd); + if (!intelScreen->mgr) { + fprintf(stderr, "Failed to create fence manager.\n"); + return GL_FALSE; + } + + intelScreen->fMan = driInitFreeSlabManager(10, 10); + if (!intelScreen->fMan) { + fprintf(stderr, "Failed to create free slab manager.\n"); + return GL_FALSE; + } + + if (!intelCreatePools(sPriv)) + return GL_FALSE; +#endif + + intelScreen->winsys = intel_create_pipe_winsys(sPriv->fd, intelScreen->fMan); return GL_TRUE; } diff --git a/src/gallium/winsys/dri/intel/intel_screen.h b/src/gallium/winsys/dri/intel/intel_screen.h index 3396f9e..c17f020 100644 --- a/src/gallium/winsys/dri/intel/intel_screen.h +++ b/src/gallium/winsys/dri/intel/intel_screen.h @@ -31,7 +31,7 @@ #include "dri_util.h" #include "i830_common.h" #include "xmlconfig.h" -#include "dri_bufpool.h" +#include "ws_dri_bufpool.h" #include "pipe/p_compiler.h" @@ -74,6 +74,13 @@ struct intel_screen */ struct intel_context *dummyContext; + /* + * New stuff form the i915tex integration + */ + struct _DriFenceMgr *mgr; + struct _DriFreeSlabManager *fMan; + unsigned batch_id; + struct pipe_winsys *winsys; }; diff --git a/src/gallium/winsys/dri/intel/intel_swapbuffers.c b/src/gallium/winsys/dri/intel/intel_swapbuffers.c index 56b86d6..923b542 100644 --- a/src/gallium/winsys/dri/intel/intel_swapbuffers.c +++ b/src/gallium/winsys/dri/intel/intel_swapbuffers.c @@ -63,7 +63,7 @@ intelDisplaySurface(__DRIdrawablePrivate *dPriv, if (intel->last_swap_fence) { driFenceFinish(intel->last_swap_fence, DRM_FENCE_TYPE_EXE, TRUE); - driFenceUnReference(intel->last_swap_fence); + driFenceUnReference(&intel->last_swap_fence); intel->last_swap_fence = NULL; } intel->last_swap_fence = intel->first_swap_fence; @@ -178,9 +178,8 @@ intelDisplaySurface(__DRIdrawablePrivate *dPriv, } if (intel->first_swap_fence) - driFenceUnReference(intel->first_swap_fence); + driFenceUnReference(&intel->first_swap_fence); intel->first_swap_fence = intel_batchbuffer_flush(intel->batch); - driFenceReference(intel->first_swap_fence); } UNLOCK_HARDWARE(intel); diff --git a/src/gallium/winsys/dri/intel/intel_winsys.h b/src/gallium/winsys/dri/intel/intel_winsys.h index ffc4078..3d32db1 100644 --- a/src/gallium/winsys/dri/intel/intel_winsys.h +++ b/src/gallium/winsys/dri/intel/intel_winsys.h @@ -37,7 +37,7 @@ struct pipe_buffer; struct _DriBufferObject; struct pipe_winsys * -intel_create_pipe_winsys( int fd ); +intel_create_pipe_winsys( int fd, struct _DriFreeSlabManager *fMan ); void intel_destroy_pipe_winsys( struct pipe_winsys *winsys ); @@ -53,6 +53,7 @@ intel_create_i915simple( struct intel_context *intel, struct intel_buffer { struct pipe_buffer base; + struct _DriBufferPool *pool; struct _DriBufferObject *driBO; }; @@ -69,5 +70,4 @@ dri_bo( struct pipe_buffer *buf ) } - #endif diff --git a/src/gallium/winsys/dri/intel/intel_winsys_i915.c b/src/gallium/winsys/dri/intel/intel_winsys_i915.c index 4d183db..a35825d 100644 --- a/src/gallium/winsys/dri/intel/intel_winsys_i915.c +++ b/src/gallium/winsys/dri/intel/intel_winsys_i915.c @@ -31,8 +31,8 @@ #include <stdlib.h> #include <xf86drm.h> -#include "dri_bufpool.h" -#include "dri_bufmgr.h" +#include "ws_dri_bufpool.h" +#include "ws_dri_bufmgr.h" #include "intel_context.h" #include "intel_batchbuffer.h" @@ -106,10 +106,18 @@ static void intel_i915_batch_reloc( struct i915_winsys *sws, mask |= DRM_BO_FLAG_READ; } - intel_batchbuffer_emit_reloc( intel->batch, +#if 0 /* JB old */ + intel_batchbuffer_emit_reloc( intel->batch, dri_bo( buf ), - flags, mask, + flags, mask, delta ); +#else /* new */ + intel_offset_relocation( intel->batch, + delta, + dri_bo( buf ), + flags, + mask ); +#endif } @@ -124,12 +132,24 @@ static void intel_i915_batch_flush( struct i915_winsys *sws, struct pipe_fence_handle *pipe; } fu; + if (fence) + assert(!*fence); + fu.dri = intel_batchbuffer_flush( intel->batch ); - if (fu.dri) - iws->pws->fence_reference(iws->pws, fence, fu.pipe); + if (!fu.dri) { + assert(0); + *fence = NULL; + return; + } + + if (fu.dri) { + if (fence) + *fence = fu.pipe; + else + driFenceUnReference(&fu.dri); + } -// if (0) intel_i915_batch_wait_idle( sws ); } diff --git a/src/gallium/winsys/dri/intel/intel_winsys_pipe.c b/src/gallium/winsys/dri/intel/intel_winsys_pipe.c index 77dec94..10e1a84 100644 --- a/src/gallium/winsys/dri/intel/intel_winsys_pipe.c +++ b/src/gallium/winsys/dri/intel/intel_winsys_pipe.c @@ -31,8 +31,8 @@ #include <stdlib.h> #include <xf86drm.h> -#include "dri_bufpool.h" -#include "dri_bufmgr.h" +//#include "dri_bufpool.h" +//#include "dri_bufmgr.h" #include "intel_context.h" #include "intel_winsys.h" @@ -50,6 +50,9 @@ struct intel_pipe_winsys { struct pipe_winsys winsys; struct _DriBufferPool *regionPool; + struct _DriBufferPool *mallocPool; + struct _DriBufferPool *vertexPool; + struct _DriFreeSlabManager *fMan; /** shared between all pipes */ }; @@ -92,6 +95,7 @@ intel_buffer_destroy(struct pipe_winsys *winsys, struct pipe_buffer *buf) { driBOUnReference( dri_bo(buf) ); + FREE(buf); } @@ -108,16 +112,23 @@ intel_buffer_create(struct pipe_winsys *winsys, struct intel_buffer *buffer = CALLOC_STRUCT( intel_buffer ); struct intel_pipe_winsys *iws = intel_pipe_winsys(winsys); unsigned flags = 0; + struct _DriBufferPool *pool; buffer->base.refcount = 1; buffer->base.alignment = alignment; buffer->base.usage = usage; buffer->base.size = size; - if (usage & (PIPE_BUFFER_USAGE_VERTEX /*| IWS_BUFFER_USAGE_LOCAL*/)) { + if (usage & (PIPE_BUFFER_USAGE_VERTEX | PIPE_BUFFER_USAGE_CONSTANT)) { flags |= DRM_BO_FLAG_MEM_LOCAL | DRM_BO_FLAG_CACHED; + pool = iws->mallocPool; + } else if (usage & PIPE_BUFFER_USAGE_CUSTOM) { + /* For vertex buffers */ + flags |= DRM_BO_FLAG_MEM_VRAM | DRM_BO_FLAG_MEM_TT; + pool = iws->vertexPool; } else { flags |= DRM_BO_FLAG_MEM_VRAM | DRM_BO_FLAG_MEM_TT; + pool = iws->regionPool; } if (usage & PIPE_BUFFER_USAGE_GPU_READ) @@ -139,10 +150,11 @@ intel_buffer_create(struct pipe_winsys *winsys, flags |= DRM_BO_FLAG_CACHED; #endif - driGenBuffers( iws->regionPool, + buffer->pool = pool; + driGenBuffers( buffer->pool, "pipe buffer", 1, &buffer->driBO, alignment, flags, 0 ); - driBOData( buffer->driBO, size, NULL, 0 ); + driBOData( buffer->driBO, size, NULL, buffer->pool, 0 ); return &buffer->base; } @@ -155,7 +167,9 @@ intel_user_buffer_create(struct pipe_winsys *winsys, void *ptr, unsigned bytes) struct intel_pipe_winsys *iws = intel_pipe_winsys(winsys); driGenUserBuffer( iws->regionPool, - "pipe user buffer", &buffer->driBO, ptr, bytes); + "pipe user buffer", &buffer->driBO, ptr, bytes ); + + buffer->base.refcount = 1; return &buffer->base; } @@ -206,10 +220,11 @@ intel_i915_surface_alloc_storage(struct pipe_winsys *winsys, struct pipe_surface *surf, unsigned width, unsigned height, enum pipe_format format, - unsigned flags) + unsigned flags, + unsigned tex_usage) { const unsigned alignment = 64; - int ret; + //int ret; surf->width = width; surf->height = height; @@ -249,9 +264,36 @@ intel_get_name( struct pipe_winsys *winsys ) return "Intel/DRI/ttm"; } +static void +intel_fence_reference( struct pipe_winsys *sws, + struct pipe_fence_handle **ptr, + struct pipe_fence_handle *fence ) +{ + if (*ptr) + driFenceUnReference((struct _DriFenceObject **)ptr); + + if (fence) + *ptr = (struct pipe_fence_handle *)driFenceReference((struct _DriFenceObject *)fence); +} + +static int +intel_fence_signalled( struct pipe_winsys *sws, + struct pipe_fence_handle *fence, + unsigned flag ) +{ + return driFenceSignaled((struct _DriFenceObject *)fence, flag); +} + +static int +intel_fence_finish( struct pipe_winsys *sws, + struct pipe_fence_handle *fence, + unsigned flag ) +{ + return driFenceFinish((struct _DriFenceObject *)fence, flag, 0); +} struct pipe_winsys * -intel_create_pipe_winsys( int fd ) +intel_create_pipe_winsys( int fd, struct _DriFreeSlabManager *fMan ) { struct intel_pipe_winsys *iws = CALLOC_STRUCT( intel_pipe_winsys ); @@ -273,8 +315,25 @@ intel_create_pipe_winsys( int fd ) iws->winsys.surface_alloc_storage = intel_i915_surface_alloc_storage; iws->winsys.surface_release = intel_i915_surface_release; - if (fd) - iws->regionPool = driDRMPoolInit(fd); + iws->winsys.fence_reference = intel_fence_reference; + iws->winsys.fence_signalled = intel_fence_signalled; + iws->winsys.fence_finish = intel_fence_finish; + + if (fd) { + iws->regionPool = driDRMPoolInit(fd); + iws->vertexPool = driSlabPoolInit(fd, + DRM_BO_FLAG_READ | + DRM_BO_FLAG_WRITE | + DRM_BO_FLAG_MEM_TT, + DRM_BO_FLAG_READ | + DRM_BO_FLAG_WRITE | + DRM_BO_FLAG_MEM_TT, + 32 * 4096, + 1, 40, 32 * 4096 * 2, 0, + fMan); + } + + iws->mallocPool = driMallocPoolInit(); return &iws->winsys; } @@ -287,6 +346,9 @@ intel_destroy_pipe_winsys( struct pipe_winsys *winsys ) if (iws->regionPool) { driPoolTakeDown(iws->regionPool); } + if (iws->mallocPool) { + driPoolTakeDown(iws->mallocPool); + } free(iws); } diff --git a/src/gallium/winsys/dri/intel/server/i830_common.h b/src/gallium/winsys/dri/intel/server/i830_common.h index d4d5888..f84f453 100644 --- a/src/gallium/winsys/dri/intel/server/i830_common.h +++ b/src/gallium/winsys/dri/intel/server/i830_common.h @@ -26,7 +26,6 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. **************************************************************************/ -/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/i810/i830_common.h,v 1.1 2002/09/11 00:29:32 dawes Exp $ */ #ifndef _I830_COMMON_H_ #define _I830_COMMON_H_ @@ -54,6 +53,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. #define DRM_I830_DESTROY_HEAP 0x0c #define DRM_I830_SET_VBLANK_PIPE 0x0d #define DRM_I830_GET_VBLANK_PIPE 0x0e +#define DRM_I830_MMIO 0x10 typedef struct { enum { @@ -85,6 +85,7 @@ typedef struct { int last_enqueue; /* last time a buffer was enqueued */ int last_dispatch; /* age of the most recently dispatched buffer */ int ctxOwner; /* last context to upload state */ + /** Last context that used the buffer manager. */ int texAge; int pf_enabled; /* is pageflipping allowed? */ int pf_active; @@ -121,20 +122,29 @@ typedef struct { unsigned int rotated_tiled; unsigned int rotated2_tiled; - int pipeA_x; - int pipeA_y; - int pipeA_w; - int pipeA_h; - int pipeB_x; - int pipeB_y; - int pipeB_w; - int pipeB_h; + int planeA_x; + int planeA_y; + int planeA_w; + int planeA_h; + int planeB_x; + int planeB_y; + int planeB_w; + int planeB_h; /* Triple buffering */ drm_handle_t third_handle; int third_offset; int third_size; unsigned int third_tiled; + + /* buffer object handles for the static buffers. May change + * over the lifetime of the client, though it doesn't in our current + * implementation. + */ + unsigned int front_bo_handle; + unsigned int back_bo_handle; + unsigned int third_bo_handle; + unsigned int depth_bo_handle; } drmI830Sarea; /* Flags for perf_boxes @@ -223,4 +233,23 @@ typedef struct { int pipe; } drmI830VBlankPipe; +#define MMIO_READ 0 +#define MMIO_WRITE 1 + +#define MMIO_REGS_IA_PRIMATIVES_COUNT 0 +#define MMIO_REGS_IA_VERTICES_COUNT 1 +#define MMIO_REGS_VS_INVOCATION_COUNT 2 +#define MMIO_REGS_GS_PRIMITIVES_COUNT 3 +#define MMIO_REGS_GS_INVOCATION_COUNT 4 +#define MMIO_REGS_CL_PRIMITIVES_COUNT 5 +#define MMIO_REGS_CL_INVOCATION_COUNT 6 +#define MMIO_REGS_PS_INVOCATION_COUNT 7 +#define MMIO_REGS_PS_DEPTH_COUNT 8 + +typedef struct { + unsigned int read_write:1; + unsigned int reg:31; + void __user *data; +} drmI830MMIO; + #endif /* _I830_DRM_H_ */ diff --git a/src/gallium/winsys/dri/intel/server/i830_dri.h b/src/gallium/winsys/dri/intel/server/i830_dri.h index c2a3af8..685de4a 100644 --- a/src/gallium/winsys/dri/intel/server/i830_dri.h +++ b/src/gallium/winsys/dri/intel/server/i830_dri.h @@ -1,4 +1,3 @@ -/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/i810/i830_dri.h,v 1.4 2002/10/30 12:52:18 alanh Exp $ */ #ifndef _I830_DRI_H #define _I830_DRI_H diff --git a/src/gallium/winsys/dri/intel/ws_dri_bufmgr.c b/src/gallium/winsys/dri/intel/ws_dri_bufmgr.c new file mode 100644 index 0000000..1bc1089 --- /dev/null +++ b/src/gallium/winsys/dri/intel/ws_dri_bufmgr.c @@ -0,0 +1,953 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * + **************************************************************************/ +/* + * Authors: Thomas Hellström <thomas-at-tungstengraphics-dot-com> + * Keith Whitwell <keithw-at-tungstengraphics-dot-com> + */ + +#include <xf86drm.h> +#include <stdlib.h> +#include "glthread.h" +#include "errno.h" +#include "ws_dri_bufmgr.h" +#include "string.h" +#include "imports.h" +#include "ws_dri_bufpool.h" +#include "ws_dri_fencemgr.h" + +/* + * This lock is here to protect drmBO structs changing underneath us during a + * validate list call, since validatelist cannot take individiual locks for + * each drmBO. Validatelist takes this lock in write mode. Any access to an + * individual drmBO should take this lock in read mode, since in that case, the + * driBufferObject mutex will protect the access. Locking order is + * driBufferObject mutex - > this rw lock. + */ + +_glthread_DECLARE_STATIC_MUTEX(bmMutex); +_glthread_DECLARE_STATIC_COND(bmCond); + +static int kernelReaders = 0; +static int num_buffers = 0; +static int num_user_buffers = 0; + +static drmBO *drmBOListBuf(void *iterator) +{ + drmBONode *node; + drmMMListHead *l = (drmMMListHead *) iterator; + node = DRMLISTENTRY(drmBONode, l, head); + return node->buf; +} + +static void *drmBOListIterator(drmBOList *list) +{ + void *ret = list->list.next; + + if (ret == &list->list) + return NULL; + return ret; +} + +static void *drmBOListNext(drmBOList *list, void *iterator) +{ + void *ret; + + drmMMListHead *l = (drmMMListHead *) iterator; + ret = l->next; + if (ret == &list->list) + return NULL; + return ret; +} + +static drmBONode *drmAddListItem(drmBOList *list, drmBO *item, + uint64_t arg0, + uint64_t arg1) +{ + drmBONode *node; + drmMMListHead *l; + + l = list->free.next; + if (l == &list->free) { + node = (drmBONode *) malloc(sizeof(*node)); + if (!node) { + return NULL; + } + list->numCurrent++; + } + else { + DRMLISTDEL(l); + node = DRMLISTENTRY(drmBONode, l, head); + } + node->buf = item; + node->arg0 = arg0; + node->arg1 = arg1; + DRMLISTADD(&node->head, &list->list); + list->numOnList++; + return node; +} + +static int drmAddValidateItem(drmBOList *list, drmBO *buf, uint64_t flags, + uint64_t mask, int *newItem) +{ + drmBONode *node, *cur; + drmMMListHead *l; + + *newItem = 0; + cur = NULL; + + for (l = list->list.next; l != &list->list; l = l->next) { + node = DRMLISTENTRY(drmBONode, l, head); + if (node->buf == buf) { + cur = node; + break; + } + } + if (!cur) { + cur = drmAddListItem(list, buf, flags, mask); + if (!cur) { + return -ENOMEM; + } + *newItem = 1; + cur->arg0 = flags; + cur->arg1 = mask; + } + else { + uint64_t memFlags = cur->arg0 & flags & DRM_BO_MASK_MEM; + uint64_t accFlags = (cur->arg0 | flags) & ~DRM_BO_MASK_MEM; + + if (mask & cur->arg1 & ~DRM_BO_MASK_MEM & (cur->arg0 ^ flags)) { + return -EINVAL; + } + + cur->arg1 |= mask; + cur->arg0 = (cur->arg0 & ~mask) | ((memFlags | accFlags) & mask); + + if (((cur->arg1 & DRM_BO_MASK_MEM) != 0) && + (cur->arg0 & DRM_BO_MASK_MEM) == 0) { + return -EINVAL; + } + } + return 0; +} + +static void drmBOFreeList(drmBOList *list) +{ + drmBONode *node; + drmMMListHead *l; + + l = list->list.next; + while(l != &list->list) { + DRMLISTDEL(l); + node = DRMLISTENTRY(drmBONode, l, head); + free(node); + l = list->list.next; + list->numCurrent--; + list->numOnList--; + } + + l = list->free.next; + while(l != &list->free) { + DRMLISTDEL(l); + node = DRMLISTENTRY(drmBONode, l, head); + free(node); + l = list->free.next; + list->numCurrent--; + } +} + +static int drmAdjustListNodes(drmBOList *list) +{ + drmBONode *node; + drmMMListHead *l; + int ret = 0; + + while(list->numCurrent < list->numTarget) { + node = (drmBONode *) malloc(sizeof(*node)); + if (!node) { + ret = -ENOMEM; + break; + } + list->numCurrent++; + DRMLISTADD(&node->head, &list->free); + } + + while(list->numCurrent > list->numTarget) { + l = list->free.next; + if (l == &list->free) + break; + DRMLISTDEL(l); + node = DRMLISTENTRY(drmBONode, l, head); + free(node); + list->numCurrent--; + } + return ret; +} + +static int drmBOCreateList(int numTarget, drmBOList *list) +{ + DRMINITLISTHEAD(&list->list); + DRMINITLISTHEAD(&list->free); + list->numTarget = numTarget; + list->numCurrent = 0; + list->numOnList = 0; + return drmAdjustListNodes(list); +} + +static int drmBOResetList(drmBOList *list) +{ + drmMMListHead *l; + int ret; + + ret = drmAdjustListNodes(list); + if (ret) + return ret; + + l = list->list.next; + while (l != &list->list) { + DRMLISTDEL(l); + DRMLISTADD(l, &list->free); + list->numOnList--; + l = list->list.next; + } + return drmAdjustListNodes(list); +} + +void driWriteLockKernelBO(void) +{ + _glthread_LOCK_MUTEX(bmMutex); + while(kernelReaders != 0) + _glthread_COND_WAIT(bmCond, bmMutex); +} + +void driWriteUnlockKernelBO(void) +{ + _glthread_UNLOCK_MUTEX(bmMutex); +} + +void driReadLockKernelBO(void) +{ + _glthread_LOCK_MUTEX(bmMutex); + kernelReaders++; + _glthread_UNLOCK_MUTEX(bmMutex); +} + +void driReadUnlockKernelBO(void) +{ + _glthread_LOCK_MUTEX(bmMutex); + if (--kernelReaders == 0) + _glthread_COND_BROADCAST(bmCond); + _glthread_UNLOCK_MUTEX(bmMutex); +} + + + + +/* + * TODO: Introduce fence pools in the same way as + * buffer object pools. + */ + +typedef struct _DriBufferObject +{ + DriBufferPool *pool; + _glthread_Mutex mutex; + int refCount; + const char *name; + uint64_t flags; + unsigned hint; + unsigned alignment; + unsigned createdByReference; + void *private; + /* user-space buffer: */ + unsigned userBuffer; + void *userData; + unsigned userSize; +} DriBufferObject; + +typedef struct _DriBufferList { + drmBOList drmBuffers; /* List of kernel buffers needing validation */ + drmBOList driBuffers; /* List of user-space buffers needing validation */ +} DriBufferList; + + +void +bmError(int val, const char *file, const char *function, int line) +{ + _mesa_printf("Fatal video memory manager error \"%s\".\n" + "Check kernel logs or set the LIBGL_DEBUG\n" + "environment variable to \"verbose\" for more info.\n" + "Detected in file %s, line %d, function %s.\n", + strerror(-val), file, line, function); +#ifndef NDEBUG + abort(); +#else + abort(); +#endif +} + +extern drmBO * +driBOKernel(struct _DriBufferObject *buf) +{ + drmBO *ret; + + driReadLockKernelBO(); + _glthread_LOCK_MUTEX(buf->mutex); + assert(buf->private != NULL); + ret = buf->pool->kernel(buf->pool, buf->private); + if (!ret) + BM_CKFATAL(-EINVAL); + _glthread_UNLOCK_MUTEX(buf->mutex); + driReadUnlockKernelBO(); + + return ret; +} + +void +driBOWaitIdle(struct _DriBufferObject *buf, int lazy) +{ + + /* + * This function may block. Is it sane to keep the mutex held during + * that time?? + */ + + _glthread_LOCK_MUTEX(buf->mutex); + BM_CKFATAL(buf->pool->waitIdle(buf->pool, buf->private, &buf->mutex, lazy)); + _glthread_UNLOCK_MUTEX(buf->mutex); +} + +void * +driBOMap(struct _DriBufferObject *buf, unsigned flags, unsigned hint) +{ + void *virtual; + int retval; + + if (buf->userBuffer) { + return buf->userData; + } + + _glthread_LOCK_MUTEX(buf->mutex); + assert(buf->private != NULL); + retval = buf->pool->map(buf->pool, buf->private, flags, hint, + &buf->mutex, &virtual); + _glthread_UNLOCK_MUTEX(buf->mutex); + + return retval == 0 ? virtual : NULL; +} + +void +driBOUnmap(struct _DriBufferObject *buf) +{ + if (buf->userBuffer) + return; + + assert(buf->private != NULL); + _glthread_LOCK_MUTEX(buf->mutex); + BM_CKFATAL(buf->pool->unmap(buf->pool, buf->private)); + _glthread_UNLOCK_MUTEX(buf->mutex); +} + +unsigned long +driBOOffset(struct _DriBufferObject *buf) +{ + unsigned long ret; + + assert(buf->private != NULL); + + _glthread_LOCK_MUTEX(buf->mutex); + ret = buf->pool->offset(buf->pool, buf->private); + _glthread_UNLOCK_MUTEX(buf->mutex); + return ret; +} + +unsigned long +driBOPoolOffset(struct _DriBufferObject *buf) +{ + unsigned long ret; + + assert(buf->private != NULL); + + _glthread_LOCK_MUTEX(buf->mutex); + ret = buf->pool->poolOffset(buf->pool, buf->private); + _glthread_UNLOCK_MUTEX(buf->mutex); + return ret; +} + +uint64_t +driBOFlags(struct _DriBufferObject *buf) +{ + uint64_t ret; + + assert(buf->private != NULL); + + driReadLockKernelBO(); + _glthread_LOCK_MUTEX(buf->mutex); + ret = buf->pool->flags(buf->pool, buf->private); + _glthread_UNLOCK_MUTEX(buf->mutex); + driReadUnlockKernelBO(); + return ret; +} + +struct _DriBufferObject * +driBOReference(struct _DriBufferObject *buf) +{ + _glthread_LOCK_MUTEX(buf->mutex); + if (++buf->refCount == 1) { + _glthread_UNLOCK_MUTEX(buf->mutex); + BM_CKFATAL(-EINVAL); + } + _glthread_UNLOCK_MUTEX(buf->mutex); + return buf; +} + +void +driBOUnReference(struct _DriBufferObject *buf) +{ + int tmp; + + if (!buf) + return; + + _glthread_LOCK_MUTEX(buf->mutex); + tmp = --buf->refCount; + if (!tmp) { + _glthread_UNLOCK_MUTEX(buf->mutex); + if (buf->private) { + if (buf->createdByReference) + buf->pool->unreference(buf->pool, buf->private); + else + buf->pool->destroy(buf->pool, buf->private); + } + if (buf->userBuffer) + num_user_buffers--; + else + num_buffers--; + free(buf); + } else + _glthread_UNLOCK_MUTEX(buf->mutex); + +} + + +int +driBOData(struct _DriBufferObject *buf, + unsigned size, const void *data, + DriBufferPool *newPool, + uint64_t flags) +{ + void *virtual = NULL; + int newBuffer; + int retval = 0; + struct _DriBufferPool *pool; + + assert(!buf->userBuffer); /* XXX just do a memcpy? */ + + _glthread_LOCK_MUTEX(buf->mutex); + pool = buf->pool; + + if (pool == NULL && newPool != NULL) { + buf->pool = newPool; + pool = newPool; + } + if (newPool == NULL) + newPool = pool; + + if (!pool->create) { + _mesa_error(NULL, GL_INVALID_OPERATION, + "driBOData called on invalid buffer\n"); + BM_CKFATAL(-EINVAL); + } + + newBuffer = (!buf->private || pool != newPool || + pool->size(pool, buf->private) < size); + + if (!flags) + flags = buf->flags; + + if (newBuffer) { + + if (buf->createdByReference) { + _mesa_error(NULL, GL_INVALID_OPERATION, + "driBOData requiring resizing called on " + "shared buffer.\n"); + BM_CKFATAL(-EINVAL); + } + + if (buf->private) + buf->pool->destroy(buf->pool, buf->private); + + pool = newPool; + buf->pool = newPool; + buf->private = pool->create(pool, size, flags, DRM_BO_HINT_DONT_FENCE, + buf->alignment); + if (!buf->private) + retval = -ENOMEM; + + if (retval == 0) + retval = pool->map(pool, buf->private, + DRM_BO_FLAG_WRITE, + DRM_BO_HINT_DONT_BLOCK, &buf->mutex, &virtual); + } else if (pool->map(pool, buf->private, DRM_BO_FLAG_WRITE, + DRM_BO_HINT_DONT_BLOCK, &buf->mutex, &virtual)) { + /* + * Buffer is busy. need to create a new one. + */ + + void *newBuf; + + newBuf = pool->create(pool, size, flags, DRM_BO_HINT_DONT_FENCE, + buf->alignment); + if (newBuf) { + buf->pool->destroy(buf->pool, buf->private); + buf->private = newBuf; + } + + retval = pool->map(pool, buf->private, + DRM_BO_FLAG_WRITE, 0, &buf->mutex, &virtual); + } else { + uint64_t flag_diff = flags ^ buf->flags; + + /* + * We might need to change buffer flags. + */ + + if (flag_diff){ + assert(pool->setStatus != NULL); + BM_CKFATAL(pool->unmap(pool, buf->private)); + BM_CKFATAL(pool->setStatus(pool, buf->private, flag_diff, + buf->flags)); + if (!data) + goto out; + + retval = pool->map(pool, buf->private, + DRM_BO_FLAG_WRITE, 0, &buf->mutex, &virtual); + } + } + + if (retval == 0) { + if (data) + memcpy(virtual, data, size); + + BM_CKFATAL(pool->unmap(pool, buf->private)); + } + + out: + _glthread_UNLOCK_MUTEX(buf->mutex); + + return retval; +} + +void +driBOSubData(struct _DriBufferObject *buf, + unsigned long offset, unsigned long size, const void *data) +{ + void *virtual; + + assert(!buf->userBuffer); /* XXX just do a memcpy? */ + + _glthread_LOCK_MUTEX(buf->mutex); + if (size && data) { + BM_CKFATAL(buf->pool->map(buf->pool, buf->private, + DRM_BO_FLAG_WRITE, 0, &buf->mutex, + &virtual)); + memcpy((unsigned char *) virtual + offset, data, size); + BM_CKFATAL(buf->pool->unmap(buf->pool, buf->private)); + } + _glthread_UNLOCK_MUTEX(buf->mutex); +} + +void +driBOGetSubData(struct _DriBufferObject *buf, + unsigned long offset, unsigned long size, void *data) +{ + void *virtual; + + assert(!buf->userBuffer); /* XXX just do a memcpy? */ + + _glthread_LOCK_MUTEX(buf->mutex); + if (size && data) { + BM_CKFATAL(buf->pool->map(buf->pool, buf->private, + DRM_BO_FLAG_READ, 0, &buf->mutex, &virtual)); + memcpy(data, (unsigned char *) virtual + offset, size); + BM_CKFATAL(buf->pool->unmap(buf->pool, buf->private)); + } + _glthread_UNLOCK_MUTEX(buf->mutex); +} + +void +driBOSetReferenced(struct _DriBufferObject *buf, + unsigned long handle) +{ + _glthread_LOCK_MUTEX(buf->mutex); + if (buf->private != NULL) { + _mesa_error(NULL, GL_INVALID_OPERATION, + "Invalid buffer for setReferenced\n"); + BM_CKFATAL(-EINVAL); + + } + if (buf->pool->reference == NULL) { + _mesa_error(NULL, GL_INVALID_OPERATION, + "Invalid buffer pool for setReferenced\n"); + BM_CKFATAL(-EINVAL); + } + buf->private = buf->pool->reference(buf->pool, handle); + if (!buf->private) { + _mesa_error(NULL, GL_OUT_OF_MEMORY, + "Invalid buffer pool for setStatic\n"); + BM_CKFATAL(-ENOMEM); + } + buf->createdByReference = GL_TRUE; + buf->flags = buf->pool->kernel(buf->pool, buf->private)->flags; + _glthread_UNLOCK_MUTEX(buf->mutex); +} + +int +driGenBuffers(struct _DriBufferPool *pool, + const char *name, + unsigned n, + struct _DriBufferObject *buffers[], + unsigned alignment, uint64_t flags, unsigned hint) +{ + struct _DriBufferObject *buf; + int i; + + flags = (flags) ? flags : DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_MEM_VRAM | + DRM_BO_FLAG_MEM_LOCAL | DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE; + + ++num_buffers; + + assert(pool); + + for (i = 0; i < n; ++i) { + buf = (struct _DriBufferObject *) calloc(1, sizeof(*buf)); + if (!buf) + return -ENOMEM; + + _glthread_INIT_MUTEX(buf->mutex); + _glthread_LOCK_MUTEX(buf->mutex); + buf->refCount = 1; + buf->flags = flags; + buf->hint = hint; + buf->name = name; + buf->alignment = alignment; + buf->pool = pool; + buf->createdByReference = 0; + _glthread_UNLOCK_MUTEX(buf->mutex); + buffers[i] = buf; + } + return 0; +} + +void +driGenUserBuffer(struct _DriBufferPool *pool, + const char *name, + struct _DriBufferObject **buffers, + void *ptr, unsigned bytes) +{ + const unsigned alignment = 1, flags = 0, hint = 0; + + --num_buffers; /* JB: is inced in GenBuffes */ + driGenBuffers(pool, name, 1, buffers, alignment, flags, hint); + ++num_user_buffers; + + (*buffers)->userBuffer = 1; + (*buffers)->userData = ptr; + (*buffers)->userSize = bytes; +} + +void +driDeleteBuffers(unsigned n, struct _DriBufferObject *buffers[]) +{ + int i; + + for (i = 0; i < n; ++i) { + driBOUnReference(buffers[i]); + } +} + + +void +driInitBufMgr(int fd) +{ + ; +} + +/* + * Note that lists are per-context and don't need mutex protection. + */ + +struct _DriBufferList * +driBOCreateList(int target) +{ + struct _DriBufferList *list = calloc(sizeof(*list), 1); + + BM_CKFATAL(drmBOCreateList(target, &list->drmBuffers)); + BM_CKFATAL(drmBOCreateList(target, &list->driBuffers)); + return list; +} + +int +driBOResetList(struct _DriBufferList * list) +{ + int ret; + ret = drmBOResetList(&list->drmBuffers); + if (ret) + return ret; + ret = drmBOResetList(&list->driBuffers); + return ret; +} + +void +driBOFreeList(struct _DriBufferList * list) +{ + drmBOFreeList(&list->drmBuffers); + drmBOFreeList(&list->driBuffers); + free(list); +} + + +/* + * Copied from libdrm, because it is needed by driAddValidateItem. + */ + +static drmBONode * +driAddListItem(drmBOList * list, drmBO * item, + uint64_t arg0, uint64_t arg1) +{ + drmBONode *node; + drmMMListHead *l; + + l = list->free.next; + if (l == &list->free) { + node = (drmBONode *) malloc(sizeof(*node)); + if (!node) { + return NULL; + } + list->numCurrent++; + } else { + DRMLISTDEL(l); + node = DRMLISTENTRY(drmBONode, l, head); + } + memset(&node->bo_arg, 0, sizeof(node->bo_arg)); + node->buf = item; + node->arg0 = arg0; + node->arg1 = arg1; + DRMLISTADDTAIL(&node->head, &list->list); + list->numOnList++; + return node; +} + +/* + * Slightly modified version compared to the libdrm version. + * This one returns the list index of the buffer put on the list. + */ + +static int +driAddValidateItem(drmBOList * list, drmBO * buf, uint64_t flags, + uint64_t mask, int *itemLoc, + struct _drmBONode **pnode) +{ + drmBONode *node, *cur; + drmMMListHead *l; + int count = 0; + + cur = NULL; + + for (l = list->list.next; l != &list->list; l = l->next) { + node = DRMLISTENTRY(drmBONode, l, head); + if (node->buf == buf) { + cur = node; + break; + } + count++; + } + if (!cur) { + cur = driAddListItem(list, buf, flags, mask); + if (!cur) + return -ENOMEM; + + cur->arg0 = flags; + cur->arg1 = mask; + } else { + uint64_t memFlags = cur->arg0 & flags & DRM_BO_MASK_MEM; + uint64_t accFlags = (cur->arg0 | flags) & ~DRM_BO_MASK_MEM; + + if (mask & cur->arg1 & ~DRM_BO_MASK_MEM & (cur->arg0 ^ flags)) { + return -EINVAL; + } + + cur->arg1 |= mask; + cur->arg0 = (cur->arg0 & ~mask) | ((memFlags | accFlags) & mask); + + if (((cur->arg1 & DRM_BO_MASK_MEM) != 0) && + (cur->arg0 & DRM_BO_MASK_MEM) == 0) { + return -EINVAL; + } + } + *itemLoc = count; + *pnode = cur; + return 0; +} + + +void +driBOAddListItem(struct _DriBufferList * list, struct _DriBufferObject *buf, + uint64_t flags, uint64_t mask, int *itemLoc, + struct _drmBONode **node) +{ + int newItem; + + _glthread_LOCK_MUTEX(buf->mutex); + BM_CKFATAL(driAddValidateItem(&list->drmBuffers, + buf->pool->kernel(buf->pool, buf->private), + flags, mask, itemLoc, node)); + BM_CKFATAL(drmAddValidateItem(&list->driBuffers, (drmBO *) buf, + flags, mask, &newItem)); + if (newItem) + buf->refCount++; + + _glthread_UNLOCK_MUTEX(buf->mutex); +} + +drmBOList *driGetdrmBOList(struct _DriBufferList *list) +{ + driWriteLockKernelBO(); + return &list->drmBuffers; +} + +void driPutdrmBOList(struct _DriBufferList *list) +{ + driWriteUnlockKernelBO(); +} + + +void +driBOFence(struct _DriBufferObject *buf, struct _DriFenceObject *fence) +{ + _glthread_LOCK_MUTEX(buf->mutex); + if (buf->pool->fence) + BM_CKFATAL(buf->pool->fence(buf->pool, buf->private, fence)); + _glthread_UNLOCK_MUTEX(buf->mutex); + +} + +void +driBOUnrefUserList(struct _DriBufferList *list) +{ + struct _DriBufferObject *buf; + void *curBuf; + + curBuf = drmBOListIterator(&list->driBuffers); + while (curBuf) { + buf = (struct _DriBufferObject *)drmBOListBuf(curBuf); + driBOUnReference(buf); + curBuf = drmBOListNext(&list->driBuffers, curBuf); + } +} + +struct _DriFenceObject * +driBOFenceUserList(struct _DriFenceMgr *mgr, + struct _DriBufferList *list, const char *name, + drmFence *kFence) +{ + struct _DriFenceObject *fence; + struct _DriBufferObject *buf; + void *curBuf; + + fence = driFenceCreate(mgr, kFence->fence_class, kFence->type, + kFence, sizeof(*kFence)); + curBuf = drmBOListIterator(&list->driBuffers); + + /* + * User-space fencing callbacks. + */ + + while (curBuf) { + buf = (struct _DriBufferObject *) drmBOListBuf(curBuf); + driBOFence(buf, fence); + driBOUnReference(buf); + curBuf = drmBOListNext(&list->driBuffers, curBuf); + } + + driBOResetList(list); + return fence; +} + +void +driBOValidateUserList(struct _DriBufferList * list) +{ + void *curBuf; + struct _DriBufferObject *buf; + + curBuf = drmBOListIterator(&list->driBuffers); + + /* + * User-space validation callbacks. + */ + + while (curBuf) { + buf = (struct _DriBufferObject *) drmBOListBuf(curBuf); + _glthread_LOCK_MUTEX(buf->mutex); + if (buf->pool->validate) + BM_CKFATAL(buf->pool->validate(buf->pool, buf->private, &buf->mutex)); + _glthread_UNLOCK_MUTEX(buf->mutex); + curBuf = drmBOListNext(&list->driBuffers, curBuf); + } +} + + +void +driPoolTakeDown(struct _DriBufferPool *pool) +{ + pool->takeDown(pool); + +} + +unsigned long +driBOSize(struct _DriBufferObject *buf) +{ + unsigned long size; + + _glthread_LOCK_MUTEX(buf->mutex); + size = buf->pool->size(buf->pool, buf->private); + _glthread_UNLOCK_MUTEX(buf->mutex); + + return size; + +} + +drmBOList *driBOGetDRMBuffers(struct _DriBufferList *list) +{ + return &list->drmBuffers; +} + +drmBOList *driBOGetDRIBuffers(struct _DriBufferList *list) +{ + return &list->driBuffers; +} + diff --git a/src/gallium/winsys/dri/intel/ws_dri_bufmgr.h b/src/gallium/winsys/dri/intel/ws_dri_bufmgr.h new file mode 100644 index 0000000..fdaf5ee --- /dev/null +++ b/src/gallium/winsys/dri/intel/ws_dri_bufmgr.h @@ -0,0 +1,138 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * + **************************************************************************/ +/* + * Authors: Thomas Hellström <thomas-at-tungstengraphics-dot-com> + * Keith Whitwell <keithw-at-tungstengraphics-dot-com> + */ + +#ifndef _PSB_BUFMGR_H_ +#define _PSB_BUFMGR_H_ +#include <xf86mm.h> +#include "i915_drm.h" +#include "ws_dri_fencemgr.h" + +typedef struct _drmBONode +{ + drmMMListHead head; + drmBO *buf; + struct drm_i915_op_arg bo_arg; + uint64_t arg0; + uint64_t arg1; +} drmBONode; + +typedef struct _drmBOList { + unsigned numTarget; + unsigned numCurrent; + unsigned numOnList; + drmMMListHead list; + drmMMListHead free; +} drmBOList; + + +struct _DriFenceObject; +struct _DriBufferObject; +struct _DriBufferPool; +struct _DriBufferList; + +/* + * Return a pointer to the libdrm buffer object this DriBufferObject + * uses. + */ + +extern drmBO *driBOKernel(struct _DriBufferObject *buf); +extern void *driBOMap(struct _DriBufferObject *buf, unsigned flags, + unsigned hint); +extern void driBOUnmap(struct _DriBufferObject *buf); +extern unsigned long driBOOffset(struct _DriBufferObject *buf); +extern unsigned long driBOPoolOffset(struct _DriBufferObject *buf); + +extern uint64_t driBOFlags(struct _DriBufferObject *buf); +extern struct _DriBufferObject *driBOReference(struct _DriBufferObject *buf); +extern void driBOUnReference(struct _DriBufferObject *buf); + +extern int driBOData(struct _DriBufferObject *r_buf, + unsigned size, const void *data, + struct _DriBufferPool *pool, uint64_t flags); + +extern void driBOSubData(struct _DriBufferObject *buf, + unsigned long offset, unsigned long size, + const void *data); +extern void driBOGetSubData(struct _DriBufferObject *buf, + unsigned long offset, unsigned long size, + void *data); +extern int driGenBuffers(struct _DriBufferPool *pool, + const char *name, + unsigned n, + struct _DriBufferObject *buffers[], + unsigned alignment, uint64_t flags, unsigned hint); +extern void driGenUserBuffer(struct _DriBufferPool *pool, + const char *name, + struct _DriBufferObject *buffers[], + void *ptr, unsigned bytes); +extern void driDeleteBuffers(unsigned n, struct _DriBufferObject *buffers[]); +extern void driInitBufMgr(int fd); +extern struct _DriBufferList *driBOCreateList(int target); +extern int driBOResetList(struct _DriBufferList * list); +extern void driBOAddListItem(struct _DriBufferList * list, + struct _DriBufferObject *buf, + uint64_t flags, uint64_t mask, int *itemLoc, + struct _drmBONode **node); + +extern void driBOValidateList(int fd, struct _DriBufferList * list); +extern void driBOFreeList(struct _DriBufferList * list); +extern struct _DriFenceObject *driBOFenceUserList(struct _DriFenceMgr *mgr, + struct _DriBufferList *list, + const char *name, + drmFence *kFence); +extern void driBOUnrefUserList(struct _DriBufferList *list); +extern void driBOValidateUserList(struct _DriBufferList * list); +extern drmBOList *driGetdrmBOList(struct _DriBufferList *list); +extern void driPutdrmBOList(struct _DriBufferList *list); + +extern void driBOFence(struct _DriBufferObject *buf, + struct _DriFenceObject *fence); + +extern void driPoolTakeDown(struct _DriBufferPool *pool); +extern void driBOSetReferenced(struct _DriBufferObject *buf, + unsigned long handle); +unsigned long driBOSize(struct _DriBufferObject *buf); +extern void driBOWaitIdle(struct _DriBufferObject *buf, int lazy); +extern void driPoolTakeDown(struct _DriBufferPool *pool); + +extern void driReadLockKernelBO(void); +extern void driReadUnlockKernelBO(void); +extern void driWriteLockKernelBO(void); +extern void driWriteUnlockKernelBO(void); + +/* + * For debugging purposes. + */ + +extern drmBOList *driBOGetDRMBuffers(struct _DriBufferList *list); +extern drmBOList *driBOGetDRIBuffers(struct _DriBufferList *list); +#endif diff --git a/src/gallium/winsys/dri/intel/ws_dri_bufpool.h b/src/gallium/winsys/dri/intel/ws_dri_bufpool.h new file mode 100644 index 0000000..3a302e1 --- /dev/null +++ b/src/gallium/winsys/dri/intel/ws_dri_bufpool.h @@ -0,0 +1,102 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * + **************************************************************************/ +/* + * Authors: Thomas Hellström <thomas-at-tungstengraphics-dot-com> + */ + +#ifndef _PSB_BUFPOOL_H_ +#define _PSB_BUFPOOL_H_ + +#include <xf86drm.h> +#include <glthread.h> +struct _DriFenceObject; + +typedef struct _DriBufferPool +{ + int fd; + int (*map) (struct _DriBufferPool * pool, void *private, + unsigned flags, int hint, _glthread_Mutex *mutex, + void **virtual); + int (*unmap) (struct _DriBufferPool * pool, void *private); + int (*destroy) (struct _DriBufferPool * pool, void *private); + unsigned long (*offset) (struct _DriBufferPool * pool, void *private); + unsigned long (*poolOffset) (struct _DriBufferPool * pool, void *private); + uint64_t (*flags) (struct _DriBufferPool * pool, void *private); + unsigned long (*size) (struct _DriBufferPool * pool, void *private); + void *(*create) (struct _DriBufferPool * pool, unsigned long size, + uint64_t flags, unsigned hint, unsigned alignment); + void *(*reference) (struct _DriBufferPool * pool, unsigned handle); + int (*unreference) (struct _DriBufferPool * pool, void *private); + int (*fence) (struct _DriBufferPool * pool, void *private, + struct _DriFenceObject * fence); + drmBO *(*kernel) (struct _DriBufferPool * pool, void *private); + int (*validate) (struct _DriBufferPool * pool, void *private, _glthread_Mutex *mutex); + int (*waitIdle) (struct _DriBufferPool *pool, void *private, _glthread_Mutex *mutex, + int lazy); + int (*setStatus) (struct _DriBufferPool *pool, void *private, + uint64_t flag_diff, uint64_t old_flags); + void (*takeDown) (struct _DriBufferPool * pool); + void *data; +} DriBufferPool; + +extern void bmError(int val, const char *file, const char *function, + int line); +#define BM_CKFATAL(val) \ + do{ \ + int tstVal = (val); \ + if (tstVal) \ + bmError(tstVal, __FILE__, __FUNCTION__, __LINE__); \ + } while(0); + + +/* + * Builtin pools. + */ + +/* + * Kernel buffer objects. Size in multiples of page size. Page size aligned. + */ + +extern struct _DriBufferPool *driDRMPoolInit(int fd); +extern struct _DriBufferPool *driMallocPoolInit(void); + +struct _DriFreeSlabManager; +extern struct _DriBufferPool * driSlabPoolInit(int fd, uint64_t flags, + uint64_t validMask, + uint32_t smallestSize, + uint32_t numSizes, + uint32_t desiredNumBuffers, + uint32_t maxSlabSize, + uint32_t pageAlignment, + struct _DriFreeSlabManager *fMan); +extern void driFinishFreeSlabManager(struct _DriFreeSlabManager *fMan); +extern struct _DriFreeSlabManager * +driInitFreeSlabManager(uint32_t checkIntervalMsec, uint32_t slabTimeoutMsec); + + +#endif diff --git a/src/gallium/winsys/dri/intel/ws_dri_drmpool.c b/src/gallium/winsys/dri/intel/ws_dri_drmpool.c new file mode 100644 index 0000000..7c55dbc --- /dev/null +++ b/src/gallium/winsys/dri/intel/ws_dri_drmpool.c @@ -0,0 +1,268 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * + **************************************************************************/ +/* + * Authors: Thomas Hellström <thomas-at-tungstengraphics-dot-com> + */ + +#include <xf86drm.h> +#include <stdlib.h> +#include <unistd.h> +#include "ws_dri_bufpool.h" +#include "ws_dri_bufmgr.h" +#include "assert.h" + +/* + * Buffer pool implementation using DRM buffer objects as DRI buffer objects. + */ + +static void * +pool_create(struct _DriBufferPool *pool, + unsigned long size, uint64_t flags, unsigned hint, + unsigned alignment) +{ + drmBO *buf = (drmBO *) malloc(sizeof(*buf)); + int ret; + unsigned pageSize = getpagesize(); + + if (!buf) + return NULL; + + if ((alignment > pageSize) && (alignment % pageSize)) { + free(buf); + return NULL; + } + + ret = drmBOCreate(pool->fd, size, alignment / pageSize, + NULL, + flags, hint, buf); + if (ret) { + free(buf); + return NULL; + } + + return (void *) buf; +} + +static void * +pool_reference(struct _DriBufferPool *pool, unsigned handle) +{ + drmBO *buf = (drmBO *) malloc(sizeof(*buf)); + int ret; + + if (!buf) + return NULL; + + ret = drmBOReference(pool->fd, handle, buf); + + if (ret) { + free(buf); + return NULL; + } + + return (void *) buf; +} + +static int +pool_destroy(struct _DriBufferPool *pool, void *private) +{ + int ret; + drmBO *buf = (drmBO *) private; + driReadLockKernelBO(); + ret = drmBOUnreference(pool->fd, buf); + free(buf); + driReadUnlockKernelBO(); + return ret; +} + +static int +pool_unreference(struct _DriBufferPool *pool, void *private) +{ + int ret; + drmBO *buf = (drmBO *) private; + driReadLockKernelBO(); + ret = drmBOUnreference(pool->fd, buf); + free(buf); + driReadUnlockKernelBO(); + return ret; +} + +static int +pool_map(struct _DriBufferPool *pool, void *private, unsigned flags, + int hint, _glthread_Mutex *mutex, void **virtual) +{ + drmBO *buf = (drmBO *) private; + int ret; + + driReadLockKernelBO(); + ret = drmBOMap(pool->fd, buf, flags, hint, virtual); + driReadUnlockKernelBO(); + return ret; +} + +static int +pool_unmap(struct _DriBufferPool *pool, void *private) +{ + drmBO *buf = (drmBO *) private; + int ret; + + driReadLockKernelBO(); + ret = drmBOUnmap(pool->fd, buf); + driReadUnlockKernelBO(); + + return ret; +} + +static unsigned long +pool_offset(struct _DriBufferPool *pool, void *private) +{ + drmBO *buf = (drmBO *) private; + unsigned long offset; + + driReadLockKernelBO(); + assert(buf->flags & DRM_BO_FLAG_NO_MOVE); + offset = buf->offset; + driReadUnlockKernelBO(); + + return buf->offset; +} + +static unsigned long +pool_poolOffset(struct _DriBufferPool *pool, void *private) +{ + return 0; +} + +static uint64_t +pool_flags(struct _DriBufferPool *pool, void *private) +{ + drmBO *buf = (drmBO *) private; + uint64_t flags; + + driReadLockKernelBO(); + flags = buf->flags; + driReadUnlockKernelBO(); + + return flags; +} + + +static unsigned long +pool_size(struct _DriBufferPool *pool, void *private) +{ + drmBO *buf = (drmBO *) private; + unsigned long size; + + driReadLockKernelBO(); + size = buf->size; + driReadUnlockKernelBO(); + + return buf->size; +} + +static int +pool_fence(struct _DriBufferPool *pool, void *private, + struct _DriFenceObject *fence) +{ + /* + * Noop. The kernel handles all fencing. + */ + + return 0; +} + +static drmBO * +pool_kernel(struct _DriBufferPool *pool, void *private) +{ + return (drmBO *) private; +} + +static int +pool_waitIdle(struct _DriBufferPool *pool, void *private, _glthread_Mutex *mutex, + int lazy) +{ + drmBO *buf = (drmBO *) private; + int ret; + + driReadLockKernelBO(); + ret = drmBOWaitIdle(pool->fd, buf, (lazy) ? DRM_BO_HINT_WAIT_LAZY:0); + driReadUnlockKernelBO(); + + return ret; +} + + +static void +pool_takedown(struct _DriBufferPool *pool) +{ + free(pool); +} + +/*static int +pool_setStatus(struct _DriBufferPool *pool, void *private, + uint64_t flag_diff, uint64_t old_flags) +{ + drmBO *buf = (drmBO *) private; + uint64_t new_flags = old_flags ^ flag_diff; + int ret; + + driReadLockKernelBO(); + ret = drmBOSetStatus(pool->fd, buf, new_flags, flag_diff, + 0, 0, 0); + driReadUnlockKernelBO(); + return ret; +}*/ + +struct _DriBufferPool * +driDRMPoolInit(int fd) +{ + struct _DriBufferPool *pool; + + pool = (struct _DriBufferPool *) malloc(sizeof(*pool)); + + if (!pool) + return NULL; + + pool->fd = fd; + pool->map = &pool_map; + pool->unmap = &pool_unmap; + pool->destroy = &pool_destroy; + pool->offset = &pool_offset; + pool->poolOffset = &pool_poolOffset; + pool->flags = &pool_flags; + pool->size = &pool_size; + pool->create = &pool_create; + pool->fence = &pool_fence; + pool->kernel = &pool_kernel; + pool->validate = NULL; + pool->waitIdle = &pool_waitIdle; + pool->takeDown = &pool_takedown; + pool->reference = &pool_reference; + pool->unreference = &pool_unreference; + pool->data = NULL; + return pool; +} diff --git a/src/gallium/winsys/dri/intel/ws_dri_fencemgr.c b/src/gallium/winsys/dri/intel/ws_dri_fencemgr.c new file mode 100644 index 0000000..1f893b4 --- /dev/null +++ b/src/gallium/winsys/dri/intel/ws_dri_fencemgr.c @@ -0,0 +1,377 @@ +#include "ws_dri_fencemgr.h" +#include "glthread.h" +#include <xf86mm.h> +#include <string.h> +#include <unistd.h> + +/* + * Note: Locking order is + * _DriFenceObject::mutex + * _DriFenceMgr::mutex + */ + +struct _DriFenceMgr { + /* + * Constant members. Need no mutex protection. + */ + struct _DriFenceMgrCreateInfo info; + void *private; + + /* + * These members are protected by this->mutex + */ + _glthread_Mutex mutex; + int refCount; + drmMMListHead *heads; + int num_fences; +}; + +struct _DriFenceObject { + + /* + * These members are constant and need no mutex protection. + */ + struct _DriFenceMgr *mgr; + uint32_t fence_class; + uint32_t fence_type; + + /* + * These members are protected by mgr->mutex. + */ + drmMMListHead head; + int refCount; + + /* + * These members are protected by this->mutex. + */ + _glthread_Mutex mutex; + uint32_t signaled_type; + void *private; +}; + +uint32_t +driFenceType(struct _DriFenceObject *fence) +{ + return fence->fence_type; +} + +struct _DriFenceMgr * +driFenceMgrCreate(const struct _DriFenceMgrCreateInfo *info) +{ + struct _DriFenceMgr *tmp; + uint32_t i; + + tmp = calloc(1, sizeof(*tmp)); + if (!tmp) + return NULL; + + _glthread_INIT_MUTEX(tmp->mutex); + _glthread_LOCK_MUTEX(tmp->mutex); + tmp->refCount = 1; + tmp->info = *info; + tmp->num_fences = 0; + tmp->heads = calloc(tmp->info.num_classes, sizeof(*tmp->heads)); + if (!tmp->heads) + goto out_err; + + for (i=0; i<tmp->info.num_classes; ++i) { + DRMINITLISTHEAD(&tmp->heads[i]); + } + _glthread_UNLOCK_MUTEX(tmp->mutex); + return tmp; + + out_err: + if (tmp) + free(tmp); + return NULL; +} + +static void +driFenceMgrUnrefUnlock(struct _DriFenceMgr **pMgr) +{ + struct _DriFenceMgr *mgr = *pMgr; + + *pMgr = NULL; + if (--mgr->refCount == 0) + free(mgr); + else + _glthread_UNLOCK_MUTEX(mgr->mutex); +} + +void +driFenceMgrUnReference(struct _DriFenceMgr **pMgr) +{ + _glthread_LOCK_MUTEX((*pMgr)->mutex); + driFenceMgrUnrefUnlock(pMgr); +} + +static void +driFenceUnReferenceLocked(struct _DriFenceObject **pFence) +{ + struct _DriFenceObject *fence = *pFence; + struct _DriFenceMgr *mgr = fence->mgr; + + *pFence = NULL; + if (--fence->refCount == 0) { + DRMLISTDELINIT(&fence->head); + if (fence->private) + mgr->info.unreference(mgr, &fence->private); + --mgr->num_fences; + fence->mgr = NULL; + --mgr->refCount; + free(fence); + + } +} + + +static void +driSignalPreviousFencesLocked(struct _DriFenceMgr *mgr, + drmMMListHead *list, + uint32_t fence_class, + uint32_t fence_type) +{ + struct _DriFenceObject *entry; + drmMMListHead *prev; + + while(list != &mgr->heads[fence_class]) { + entry = DRMLISTENTRY(struct _DriFenceObject, list, head); + + /* + * Up refcount so that entry doesn't disappear from under us + * when we unlock-relock mgr to get the correct locking order. + */ + + ++entry->refCount; + _glthread_UNLOCK_MUTEX(mgr->mutex); + _glthread_LOCK_MUTEX(entry->mutex); + _glthread_LOCK_MUTEX(mgr->mutex); + + prev = list->prev; + + + + if (list->prev == list) { + + /* + * Somebody else removed the entry from the list. + */ + + _glthread_UNLOCK_MUTEX(entry->mutex); + driFenceUnReferenceLocked(&entry); + return; + } + + entry->signaled_type |= (fence_type & entry->fence_type); + if (entry->signaled_type == entry->fence_type) { + DRMLISTDELINIT(list); + mgr->info.unreference(mgr, &entry->private); + } + _glthread_UNLOCK_MUTEX(entry->mutex); + driFenceUnReferenceLocked(&entry); + list = prev; + } +} + + +int +driFenceFinish(struct _DriFenceObject *fence, uint32_t fence_type, + int lazy_hint) +{ + struct _DriFenceMgr *mgr = fence->mgr; + int ret = 0; + + _glthread_LOCK_MUTEX(fence->mutex); + + if ((fence->signaled_type & fence_type) == fence_type) + goto out0; + + ret = mgr->info.finish(mgr, fence->private, fence_type, lazy_hint); + if (ret) + goto out0; + + _glthread_LOCK_MUTEX(mgr->mutex); + _glthread_UNLOCK_MUTEX(fence->mutex); + + driSignalPreviousFencesLocked(mgr, &fence->head, fence->fence_class, + fence_type); + _glthread_UNLOCK_MUTEX(mgr->mutex); + return 0; + + out0: + _glthread_UNLOCK_MUTEX(fence->mutex); + return ret; +} + +uint32_t driFenceSignaledTypeCached(struct _DriFenceObject *fence) +{ + uint32_t ret; + + _glthread_LOCK_MUTEX(fence->mutex); + ret = fence->signaled_type; + _glthread_UNLOCK_MUTEX(fence->mutex); + + return ret; +} + +int +driFenceSignaledType(struct _DriFenceObject *fence, uint32_t flush_type, + uint32_t *signaled) +{ + int ret = 0; + struct _DriFenceMgr *mgr; + + _glthread_LOCK_MUTEX(fence->mutex); + mgr = fence->mgr; + *signaled = fence->signaled_type; + if ((fence->signaled_type & flush_type) == flush_type) + goto out0; + + ret = mgr->info.signaled(mgr, fence->private, flush_type, signaled); + if (ret) { + *signaled = fence->signaled_type; + goto out0; + } + + if ((fence->signaled_type | *signaled) == fence->signaled_type) + goto out0; + + _glthread_LOCK_MUTEX(mgr->mutex); + _glthread_UNLOCK_MUTEX(fence->mutex); + + driSignalPreviousFencesLocked(mgr, &fence->head, fence->fence_class, + *signaled); + + _glthread_UNLOCK_MUTEX(mgr->mutex); + return 0; + out0: + _glthread_UNLOCK_MUTEX(fence->mutex); + return ret; +} + +struct _DriFenceObject * +driFenceReference(struct _DriFenceObject *fence) +{ + _glthread_LOCK_MUTEX(fence->mgr->mutex); + ++fence->refCount; + _glthread_UNLOCK_MUTEX(fence->mgr->mutex); + return fence; +} + +void +driFenceUnReference(struct _DriFenceObject **pFence) +{ + struct _DriFenceMgr *mgr; + + if (*pFence == NULL) + return; + + mgr = (*pFence)->mgr; + _glthread_LOCK_MUTEX(mgr->mutex); + ++mgr->refCount; + driFenceUnReferenceLocked(pFence); + driFenceMgrUnrefUnlock(&mgr); +} + +struct _DriFenceObject +*driFenceCreate(struct _DriFenceMgr *mgr, uint32_t fence_class, + uint32_t fence_type, void *private, size_t private_size) +{ + struct _DriFenceObject *fence; + size_t fence_size = sizeof(*fence); + + if (private_size) + fence_size = ((fence_size + 15) & ~15); + + fence = calloc(1, fence_size + private_size); + + if (!fence) { + int ret = mgr->info.finish(mgr, private, fence_type, 0); + + if (ret) + usleep(10000000); + + return NULL; + } + + _glthread_INIT_MUTEX(fence->mutex); + _glthread_LOCK_MUTEX(fence->mutex); + _glthread_LOCK_MUTEX(mgr->mutex); + fence->refCount = 1; + DRMLISTADDTAIL(&fence->head, &mgr->heads[fence_class]); + fence->mgr = mgr; + ++mgr->refCount; + ++mgr->num_fences; + _glthread_UNLOCK_MUTEX(mgr->mutex); + fence->fence_class = fence_class; + fence->fence_type = fence_type; + fence->signaled_type = 0; + fence->private = private; + if (private_size) { + fence->private = (void *)(((uint8_t *) fence) + fence_size); + memcpy(fence->private, private, private_size); + } + + _glthread_UNLOCK_MUTEX(fence->mutex); + return fence; +} + + +static int +tSignaled(struct _DriFenceMgr *mgr, void *private, uint32_t flush_type, + uint32_t *signaled_type) +{ + long fd = (long) mgr->private; + int dummy; + drmFence *fence = (drmFence *) private; + int ret; + + *signaled_type = 0; + ret = drmFenceSignaled((int) fd, fence, flush_type, &dummy); + if (ret) + return ret; + + *signaled_type = fence->signaled; + + return 0; +} + +static int +tFinish(struct _DriFenceMgr *mgr, void *private, uint32_t fence_type, + int lazy_hint) +{ + long fd = (long) mgr->private; + unsigned flags = lazy_hint ? DRM_FENCE_FLAG_WAIT_LAZY : 0; + + return drmFenceWait((int)fd, flags, (drmFence *) private, fence_type); +} + +static int +tUnref(struct _DriFenceMgr *mgr, void **private) +{ + long fd = (long) mgr->private; + drmFence *fence = (drmFence *) *private; + *private = NULL; + + return drmFenceUnreference(fd, fence); +} + +struct _DriFenceMgr *driFenceMgrTTMInit(int fd) +{ + struct _DriFenceMgrCreateInfo info; + struct _DriFenceMgr *mgr; + + info.flags = DRI_FENCE_CLASS_ORDERED; + info.num_classes = 4; + info.signaled = tSignaled; + info.finish = tFinish; + info.unreference = tUnref; + + mgr = driFenceMgrCreate(&info); + if (mgr == NULL) + return NULL; + + mgr->private = (void *) (long) fd; + return mgr; +} + diff --git a/src/gallium/winsys/dri/intel/ws_dri_fencemgr.h b/src/gallium/winsys/dri/intel/ws_dri_fencemgr.h new file mode 100644 index 0000000..4ea58df --- /dev/null +++ b/src/gallium/winsys/dri/intel/ws_dri_fencemgr.h @@ -0,0 +1,115 @@ +#ifndef DRI_FENCEMGR_H +#define DRI_FENCEMGR_H + +#include <stdint.h> +#include <stdlib.h> + +struct _DriFenceObject; +struct _DriFenceMgr; + +/* + * Do a quick check to see if the fence manager has registered the fence + * object as signaled. Note that this function may return a false negative + * answer. + */ +extern uint32_t driFenceSignaledTypeCached(struct _DriFenceObject *fence); + +/* + * Check if the fence object is signaled. This function can be substantially + * more expensive to call than the above function, but will not return a false + * negative answer. The argument "flush_type" sets the types that the + * underlying mechanism must make sure will eventually signal. + */ +extern int driFenceSignaledType(struct _DriFenceObject *fence, + uint32_t flush_type, uint32_t *signaled); + +/* + * Convenience functions. + */ + +static inline int driFenceSignaled(struct _DriFenceObject *fence, + uint32_t flush_type) +{ + uint32_t signaled_types; + int ret = driFenceSignaledType(fence, flush_type, &signaled_types); + if (ret) + return 0; + return ((signaled_types & flush_type) == flush_type); +} + +static inline int driFenceSignaledCached(struct _DriFenceObject *fence, + uint32_t flush_type) +{ + uint32_t signaled_types = + driFenceSignaledTypeCached(fence); + + return ((signaled_types & flush_type) == flush_type); +} + +/* + * Reference a fence object. + */ +extern struct _DriFenceObject *driFenceReference(struct _DriFenceObject *fence); + +/* + * Unreference a fence object. The fence object pointer will be reset to NULL. + */ + +extern void driFenceUnReference(struct _DriFenceObject **pFence); + + +/* + * Wait for a fence to signal the indicated fence_type. + * If "lazy_hint" is true, it indicates that the wait may sleep to avoid + * busy-wait polling. + */ +extern int driFenceFinish(struct _DriFenceObject *fence, uint32_t fence_type, + int lazy_hint); + +/* + * Create a DriFenceObject for manager "mgr". + * + * "private" is a pointer that should be used for the callbacks in + * struct _DriFenceMgrCreateInfo. + * + * if private_size is nonzero, then the info stored at *private, with size + * private size will be copied and the fence manager will instead use a + * pointer to the copied data for the callbacks in + * struct _DriFenceMgrCreateInfo. In that case, the object pointed to by + * "private" may be destroyed after the call to driFenceCreate. + */ +extern struct _DriFenceObject *driFenceCreate(struct _DriFenceMgr *mgr, + uint32_t fence_class, + uint32_t fence_type, + void *private, + size_t private_size); + +extern uint32_t driFenceType(struct _DriFenceObject *fence); + +/* + * Fence creations are ordered. If a fence signals a fence_type, + * it is safe to assume that all fences of the same class that was + * created before that fence has signaled the same type. + */ + +#define DRI_FENCE_CLASS_ORDERED (1 << 0) + +struct _DriFenceMgrCreateInfo { + uint32_t flags; + uint32_t num_classes; + int (*signaled) (struct _DriFenceMgr *mgr, void *private, uint32_t flush_type, + uint32_t *signaled_type); + int (*finish) (struct _DriFenceMgr *mgr, void *private, uint32_t fence_type, int lazy_hint); + int (*unreference) (struct _DriFenceMgr *mgr, void **private); +}; + +extern struct _DriFenceMgr * +driFenceMgrCreate(const struct _DriFenceMgrCreateInfo *info); + +void +driFenceMgrUnReference(struct _DriFenceMgr **pMgr); + +extern struct _DriFenceMgr * +driFenceMgrTTMInit(int fd); + +#endif diff --git a/src/gallium/winsys/dri/intel/ws_dri_mallocpool.c b/src/gallium/winsys/dri/intel/ws_dri_mallocpool.c new file mode 100644 index 0000000..bf97d7e --- /dev/null +++ b/src/gallium/winsys/dri/intel/ws_dri_mallocpool.c @@ -0,0 +1,162 @@ +/************************************************************************** + * + * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, TX., USA + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * + **************************************************************************/ +/* + * Authors: Thomas Hellström <thomas-at-tungstengraphics-dot-com> + */ + +#include <xf86drm.h> +#include <stdlib.h> +#include <errno.h> +#include "imports.h" +#include "glthread.h" +#include "ws_dri_bufpool.h" +#include "ws_dri_bufmgr.h" +#include "intel_screen.h" + +static void * +pool_create(struct _DriBufferPool *pool, + unsigned long size, uint64_t flags, unsigned hint, + unsigned alignment) +{ + unsigned long *private = malloc(size + 2*sizeof(unsigned long)); + if ((flags & DRM_BO_MASK_MEM) != DRM_BO_FLAG_MEM_LOCAL) + abort(); + + *private = size; + return (void *)private; +} + + +static int +pool_destroy(struct _DriBufferPool *pool, void *private) +{ + free(private); + return 0; +} + +static int +pool_waitIdle(struct _DriBufferPool *pool, void *private, + _glthread_Mutex *mutex, int lazy) +{ + return 0; +} + +static int +pool_map(struct _DriBufferPool *pool, void *private, unsigned flags, + int hint, _glthread_Mutex *mutex, void **virtual) +{ + *virtual = (void *)((unsigned long *)private + 2); + return 0; +} + +static int +pool_unmap(struct _DriBufferPool *pool, void *private) +{ + return 0; +} + +static unsigned long +pool_offset(struct _DriBufferPool *pool, void *private) +{ + /* + * BUG + */ + abort(); + return 0UL; +} + +static unsigned long +pool_poolOffset(struct _DriBufferPool *pool, void *private) +{ + /* + * BUG + */ + abort(); +} + +static uint64_t +pool_flags(struct _DriBufferPool *pool, void *private) +{ + return DRM_BO_FLAG_MEM_LOCAL | DRM_BO_FLAG_CACHED; +} + +static unsigned long +pool_size(struct _DriBufferPool *pool, void *private) +{ + return *(unsigned long *) private; +} + + +static int +pool_fence(struct _DriBufferPool *pool, void *private, + struct _DriFenceObject *fence) +{ + abort(); + return 0UL; +} + +static drmBO * +pool_kernel(struct _DriBufferPool *pool, void *private) +{ + abort(); + return NULL; +} + +static void +pool_takedown(struct _DriBufferPool *pool) +{ + free(pool); +} + + +struct _DriBufferPool * +driMallocPoolInit(void) +{ + struct _DriBufferPool *pool; + + pool = (struct _DriBufferPool *) malloc(sizeof(*pool)); + if (!pool) + return NULL; + + pool->data = NULL; + pool->fd = -1; + pool->map = &pool_map; + pool->unmap = &pool_unmap; + pool->destroy = &pool_destroy; + pool->offset = &pool_offset; + pool->poolOffset = &pool_poolOffset; + pool->flags = &pool_flags; + pool->size = &pool_size; + pool->create = &pool_create; + pool->fence = &pool_fence; + pool->kernel = &pool_kernel; + pool->validate = NULL; + pool->waitIdle = &pool_waitIdle; + pool->takeDown = &pool_takedown; + return pool; +} diff --git a/src/gallium/winsys/dri/intel/ws_dri_slabpool.c b/src/gallium/winsys/dri/intel/ws_dri_slabpool.c new file mode 100644 index 0000000..62d82bb --- /dev/null +++ b/src/gallium/winsys/dri/intel/ws_dri_slabpool.c @@ -0,0 +1,968 @@ +/************************************************************************** + * + * Copyright 2006-2008 Tungsten Graphics, Inc., Cedar Park, TX., USA + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * + **************************************************************************/ +/* + * Authors: Thomas Hellstrom <thomas-at-tungstengraphics-dot-com> + */ + +#include <stdint.h> +#include <sys/time.h> +#include <errno.h> +#include <unistd.h> +#include <assert.h> +#include "ws_dri_bufpool.h" +#include "ws_dri_fencemgr.h" +#include "ws_dri_bufmgr.h" +#include "glthread.h" + +#define DRI_SLABPOOL_ALLOC_RETRIES 100 + +struct _DriSlab; + +struct _DriSlabBuffer { + int isSlabBuffer; + drmBO *bo; + struct _DriFenceObject *fence; + struct _DriSlab *parent; + drmMMListHead head; + uint32_t mapCount; + uint32_t start; + uint32_t fenceType; + int unFenced; + _glthread_Cond event; +}; + +struct _DriKernelBO { + int fd; + drmBO bo; + drmMMListHead timeoutHead; + drmMMListHead head; + struct timeval timeFreed; + uint32_t pageAlignment; + void *virtual; +}; + +struct _DriSlab{ + drmMMListHead head; + drmMMListHead freeBuffers; + uint32_t numBuffers; + uint32_t numFree; + struct _DriSlabBuffer *buffers; + struct _DriSlabSizeHeader *header; + struct _DriKernelBO *kbo; +}; + + +struct _DriSlabSizeHeader { + drmMMListHead slabs; + drmMMListHead freeSlabs; + drmMMListHead delayedBuffers; + uint32_t numDelayed; + struct _DriSlabPool *slabPool; + uint32_t bufSize; + _glthread_Mutex mutex; +}; + +struct _DriFreeSlabManager { + struct timeval slabTimeout; + struct timeval checkInterval; + struct timeval nextCheck; + drmMMListHead timeoutList; + drmMMListHead unCached; + drmMMListHead cached; + _glthread_Mutex mutex; +}; + + +struct _DriSlabPool { + + /* + * The data of this structure remains constant after + * initialization and thus needs no mutex protection. + */ + + struct _DriFreeSlabManager *fMan; + uint64_t proposedFlags; + uint64_t validMask; + uint32_t *bucketSizes; + uint32_t numBuckets; + uint32_t pageSize; + int fd; + int pageAlignment; + int maxSlabSize; + int desiredNumBuffers; + struct _DriSlabSizeHeader *headers; +}; + +/* + * FIXME: Perhaps arrange timeout slabs in size buckets for fast + * retreival?? + */ + + +static inline int +driTimeAfterEq(struct timeval *arg1, struct timeval *arg2) +{ + return ((arg1->tv_sec > arg2->tv_sec) || + ((arg1->tv_sec == arg2->tv_sec) && + (arg1->tv_usec > arg2->tv_usec))); +} + +static inline void +driTimeAdd(struct timeval *arg, struct timeval *add) +{ + unsigned int sec; + + arg->tv_sec += add->tv_sec; + arg->tv_usec += add->tv_usec; + sec = arg->tv_usec / 1000000; + arg->tv_sec += sec; + arg->tv_usec -= sec*1000000; +} + +static void +driFreeKernelBO(struct _DriKernelBO *kbo) +{ + if (!kbo) + return; + + (void) drmBOUnreference(kbo->fd, &kbo->bo); + free(kbo); +} + + +static void +driFreeTimeoutKBOsLocked(struct _DriFreeSlabManager *fMan, + struct timeval *time) +{ + drmMMListHead *list, *next; + struct _DriKernelBO *kbo; + + if (!driTimeAfterEq(time, &fMan->nextCheck)) + return; + + for (list = fMan->timeoutList.next, next = list->next; + list != &fMan->timeoutList; + list = next, next = list->next) { + + kbo = DRMLISTENTRY(struct _DriKernelBO, list, timeoutHead); + + if (!driTimeAfterEq(time, &kbo->timeFreed)) + break; + + DRMLISTDELINIT(&kbo->timeoutHead); + DRMLISTDELINIT(&kbo->head); + driFreeKernelBO(kbo); + } + + fMan->nextCheck = *time; + driTimeAdd(&fMan->nextCheck, &fMan->checkInterval); +} + + +/* + * Add a _DriKernelBO to the free slab manager. + * This means that it is available for reuse, but if it's not + * reused in a while, it will be freed. + */ + +static void +driSetKernelBOFree(struct _DriFreeSlabManager *fMan, + struct _DriKernelBO *kbo) +{ + struct timeval time; + + _glthread_LOCK_MUTEX(fMan->mutex); + gettimeofday(&time, NULL); + driTimeAdd(&time, &fMan->slabTimeout); + + kbo->timeFreed = time; + + if (kbo->bo.flags & DRM_BO_FLAG_CACHED) + DRMLISTADD(&kbo->head, &fMan->cached); + else + DRMLISTADD(&kbo->head, &fMan->unCached); + + DRMLISTADDTAIL(&kbo->timeoutHead, &fMan->timeoutList); + driFreeTimeoutKBOsLocked(fMan, &time); + + _glthread_UNLOCK_MUTEX(fMan->mutex); +} + +/* + * Get a _DriKernelBO for us to use as storage for a slab. + * + */ + +static struct _DriKernelBO * +driAllocKernelBO(struct _DriSlabSizeHeader *header) + +{ + struct _DriSlabPool *slabPool = header->slabPool; + struct _DriFreeSlabManager *fMan = slabPool->fMan; + drmMMListHead *list, *next, *head; + uint32_t size = header->bufSize * slabPool->desiredNumBuffers; + struct _DriKernelBO *kbo; + struct _DriKernelBO *kboTmp; + int ret; + + /* + * FIXME: We should perhaps allow some variation in slabsize in order + * to efficiently reuse slabs. + */ + + size = (size <= slabPool->maxSlabSize) ? size : slabPool->maxSlabSize; + size = (size + slabPool->pageSize - 1) & ~(slabPool->pageSize - 1); + _glthread_LOCK_MUTEX(fMan->mutex); + + kbo = NULL; + + retry: + head = (slabPool->proposedFlags & DRM_BO_FLAG_CACHED) ? + &fMan->cached : &fMan->unCached; + + for (list = head->next, next = list->next; + list != head; + list = next, next = list->next) { + + kboTmp = DRMLISTENTRY(struct _DriKernelBO, list, head); + + if ((kboTmp->bo.size == size) && + (slabPool->pageAlignment == 0 || + (kboTmp->pageAlignment % slabPool->pageAlignment) == 0)) { + + if (!kbo) + kbo = kboTmp; + + if ((kbo->bo.proposedFlags ^ slabPool->proposedFlags) == 0) + break; + + } + } + + if (kbo) { + DRMLISTDELINIT(&kbo->head); + DRMLISTDELINIT(&kbo->timeoutHead); + } + + _glthread_UNLOCK_MUTEX(fMan->mutex); + + if (kbo) { + uint64_t new_mask = kbo->bo.proposedFlags ^ slabPool->proposedFlags; + + ret = 0; + if (new_mask) { + ret = drmBOSetStatus(kbo->fd, &kbo->bo, slabPool->proposedFlags, + new_mask, DRM_BO_HINT_DONT_FENCE, 0, 0); + } + if (ret == 0) + return kbo; + + driFreeKernelBO(kbo); + kbo = NULL; + goto retry; + } + + kbo = calloc(1, sizeof(struct _DriKernelBO)); + if (!kbo) + return NULL; + + kbo->fd = slabPool->fd; + DRMINITLISTHEAD(&kbo->head); + DRMINITLISTHEAD(&kbo->timeoutHead); + ret = drmBOCreate(kbo->fd, size, slabPool->pageAlignment, NULL, + slabPool->proposedFlags, + DRM_BO_HINT_DONT_FENCE, &kbo->bo); + if (ret) + goto out_err0; + + ret = drmBOMap(kbo->fd, &kbo->bo, + DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE, + 0, &kbo->virtual); + + if (ret) + goto out_err1; + + ret = drmBOUnmap(kbo->fd, &kbo->bo); + if (ret) + goto out_err1; + + return kbo; + + out_err1: + drmBOUnreference(kbo->fd, &kbo->bo); + out_err0: + free(kbo); + return NULL; +} + + +static int +driAllocSlab(struct _DriSlabSizeHeader *header) +{ + struct _DriSlab *slab; + struct _DriSlabBuffer *buf; + uint32_t numBuffers; + int ret; + int i; + + slab = calloc(1, sizeof(*slab)); + if (!slab) + return -ENOMEM; + + slab->kbo = driAllocKernelBO(header); + if (!slab->kbo) { + ret = -ENOMEM; + goto out_err0; + } + + numBuffers = slab->kbo->bo.size / header->bufSize; + + slab->buffers = calloc(numBuffers, sizeof(*slab->buffers)); + if (!slab->buffers) { + ret = -ENOMEM; + goto out_err1; + } + + DRMINITLISTHEAD(&slab->head); + DRMINITLISTHEAD(&slab->freeBuffers); + slab->numBuffers = numBuffers; + slab->numFree = 0; + slab->header = header; + + buf = slab->buffers; + for (i=0; i < numBuffers; ++i) { + buf->parent = slab; + buf->start = i* header->bufSize; + buf->mapCount = 0; + buf->isSlabBuffer = 1; + _glthread_INIT_COND(buf->event); + DRMLISTADDTAIL(&buf->head, &slab->freeBuffers); + slab->numFree++; + buf++; + } + + DRMLISTADDTAIL(&slab->head, &header->slabs); + + return 0; + + out_err1: + driSetKernelBOFree(header->slabPool->fMan, slab->kbo); + free(slab->buffers); + out_err0: + free(slab); + return ret; +} + +/* + * Delete a buffer from the slab header delayed list and put + * it on the slab free list. + */ + +static void +driSlabFreeBufferLocked(struct _DriSlabBuffer *buf) +{ + struct _DriSlab *slab = buf->parent; + struct _DriSlabSizeHeader *header = slab->header; + drmMMListHead *list = &buf->head; + + DRMLISTDEL(list); + DRMLISTADDTAIL(list, &slab->freeBuffers); + slab->numFree++; + + if (slab->head.next == &slab->head) + DRMLISTADDTAIL(&slab->head, &header->slabs); + + if (slab->numFree == slab->numBuffers) { + list = &slab->head; + DRMLISTDEL(list); + DRMLISTADDTAIL(list, &header->freeSlabs); + } + + if (header->slabs.next == &header->slabs || + slab->numFree != slab->numBuffers) { + + drmMMListHead *next; + struct _DriFreeSlabManager *fMan = header->slabPool->fMan; + + for (list = header->freeSlabs.next, next = list->next; + list != &header->freeSlabs; + list = next, next = list->next) { + + slab = DRMLISTENTRY(struct _DriSlab, list, head); + + DRMLISTDELINIT(list); + driSetKernelBOFree(fMan, slab->kbo); + free(slab->buffers); + free(slab); + } + } +} + +static void +driSlabCheckFreeLocked(struct _DriSlabSizeHeader *header, int wait) +{ + drmMMListHead *list, *prev, *first; + struct _DriSlabBuffer *buf; + struct _DriSlab *slab; + int firstWasSignaled = 1; + int signaled; + int i; + int ret; + + /* + * Rerun the freeing test if the youngest tested buffer + * was signaled, since there might be more idle buffers + * in the delay list. + */ + + while (firstWasSignaled) { + firstWasSignaled = 0; + signaled = 0; + first = header->delayedBuffers.next; + + /* Only examine the oldest 1/3 of delayed buffers: + */ + if (header->numDelayed > 3) { + for (i = 0; i < header->numDelayed; i += 3) { + first = first->next; + } + } + + for (list = first, prev = list->prev; + list != &header->delayedBuffers; + list = prev, prev = list->prev) { + buf = DRMLISTENTRY(struct _DriSlabBuffer, list, head); + slab = buf->parent; + + if (!signaled) { + if (wait) { + ret = driFenceFinish(buf->fence, buf->fenceType, 0); + if (ret) + break; + signaled = 1; + wait = 0; + } else { + signaled = driFenceSignaled(buf->fence, buf->fenceType); + } + if (signaled) { + if (list == first) + firstWasSignaled = 1; + driFenceUnReference(&buf->fence); + header->numDelayed--; + driSlabFreeBufferLocked(buf); + } + } else if (driFenceSignaledCached(buf->fence, buf->fenceType)) { + driFenceUnReference(&buf->fence); + header->numDelayed--; + driSlabFreeBufferLocked(buf); + } + } + } +} + + +static struct _DriSlabBuffer * +driSlabAllocBuffer(struct _DriSlabSizeHeader *header) +{ + static struct _DriSlabBuffer *buf; + struct _DriSlab *slab; + drmMMListHead *list; + int count = DRI_SLABPOOL_ALLOC_RETRIES; + + _glthread_LOCK_MUTEX(header->mutex); + while(header->slabs.next == &header->slabs && count > 0) { + driSlabCheckFreeLocked(header, 0); + if (header->slabs.next != &header->slabs) + break; + + _glthread_UNLOCK_MUTEX(header->mutex); + if (count != DRI_SLABPOOL_ALLOC_RETRIES) + usleep(1); + _glthread_LOCK_MUTEX(header->mutex); + (void) driAllocSlab(header); + count--; + } + + list = header->slabs.next; + if (list == &header->slabs) { + _glthread_UNLOCK_MUTEX(header->mutex); + return NULL; + } + slab = DRMLISTENTRY(struct _DriSlab, list, head); + if (--slab->numFree == 0) + DRMLISTDELINIT(list); + + list = slab->freeBuffers.next; + DRMLISTDELINIT(list); + + _glthread_UNLOCK_MUTEX(header->mutex); + buf = DRMLISTENTRY(struct _DriSlabBuffer, list, head); + return buf; +} + +static void * +pool_create(struct _DriBufferPool *driPool, unsigned long size, + uint64_t flags, unsigned hint, unsigned alignment) +{ + struct _DriSlabPool *pool = (struct _DriSlabPool *) driPool->data; + struct _DriSlabSizeHeader *header; + struct _DriSlabBuffer *buf; + void *dummy; + int i; + int ret; + + /* + * FIXME: Check for compatibility. + */ + + header = pool->headers; + for (i=0; i<pool->numBuckets; ++i) { + if (header->bufSize >= size) + break; + header++; + } + + if (i < pool->numBuckets) + return driSlabAllocBuffer(header); + + + /* + * Fall back to allocate a buffer object directly from DRM. + * and wrap it in a driBO structure. + */ + + + buf = calloc(1, sizeof(*buf)); + + if (!buf) + return NULL; + + buf->bo = calloc(1, sizeof(*buf->bo)); + if (!buf->bo) + goto out_err0; + + if (alignment) { + if ((alignment < pool->pageSize) && (pool->pageSize % alignment)) + goto out_err1; + if ((alignment > pool->pageSize) && (alignment % pool->pageSize)) + goto out_err1; + } + + ret = drmBOCreate(pool->fd, size, alignment / pool->pageSize, NULL, + flags, hint, buf->bo); + if (ret) + goto out_err1; + + ret = drmBOMap(pool->fd, buf->bo, DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE, + 0, &dummy); + if (ret) + goto out_err2; + + ret = drmBOUnmap(pool->fd, buf->bo); + if (ret) + goto out_err2; + + return buf; + out_err2: + drmBOUnreference(pool->fd, buf->bo); + out_err1: + free(buf->bo); + out_err0: + free(buf); + return NULL; +} + +static int +pool_destroy(struct _DriBufferPool *driPool, void *private) +{ + struct _DriSlabBuffer *buf = + (struct _DriSlabBuffer *) private; + struct _DriSlab *slab; + struct _DriSlabSizeHeader *header; + + if (!buf->isSlabBuffer) { + struct _DriSlabPool *pool = (struct _DriSlabPool *) driPool->data; + int ret; + + ret = drmBOUnreference(pool->fd, buf->bo); + free(buf->bo); + free(buf); + return ret; + } + + slab = buf->parent; + header = slab->header; + + _glthread_LOCK_MUTEX(header->mutex); + buf->unFenced = 0; + buf->mapCount = 0; + + if (buf->fence && !driFenceSignaledCached(buf->fence, buf->fenceType)) { + DRMLISTADDTAIL(&buf->head, &header->delayedBuffers); + header->numDelayed++; + } else { + if (buf->fence) + driFenceUnReference(&buf->fence); + driSlabFreeBufferLocked(buf); + } + + _glthread_UNLOCK_MUTEX(header->mutex); + return 0; +} + +static int +pool_waitIdle(struct _DriBufferPool *driPool, void *private, + _glthread_Mutex *mutex, int lazy) +{ + struct _DriSlabBuffer *buf = (struct _DriSlabBuffer *) private; + + while(buf->unFenced) + _glthread_COND_WAIT(buf->event, *mutex); + + if (!buf->fence) + return 0; + + driFenceFinish(buf->fence, buf->fenceType, lazy); + driFenceUnReference(&buf->fence); + + return 0; +} + +static int +pool_map(struct _DriBufferPool *pool, void *private, unsigned flags, + int hint, _glthread_Mutex *mutex, void **virtual) +{ + struct _DriSlabBuffer *buf = (struct _DriSlabBuffer *) private; + int busy; + + if (buf->isSlabBuffer) + busy = buf->unFenced || (buf->fence && !driFenceSignaledCached(buf->fence, buf->fenceType)); + else + busy = buf->fence && !driFenceSignaled(buf->fence, buf->fenceType); + + + if (busy) { + if (hint & DRM_BO_HINT_DONT_BLOCK) + return -EBUSY; + else { + (void) pool_waitIdle(pool, private, mutex, 0); + } + } + + ++buf->mapCount; + *virtual = (buf->isSlabBuffer) ? + (void *) ((uint8_t *) buf->parent->kbo->virtual + buf->start) : + (void *) buf->bo->virtual; + + return 0; +} + +static int +pool_unmap(struct _DriBufferPool *pool, void *private) +{ + struct _DriSlabBuffer *buf = (struct _DriSlabBuffer *) private; + + --buf->mapCount; + if (buf->mapCount == 0 && buf->isSlabBuffer) + _glthread_COND_BROADCAST(buf->event); + + return 0; +} + +static unsigned long +pool_offset(struct _DriBufferPool *pool, void *private) +{ + struct _DriSlabBuffer *buf = (struct _DriSlabBuffer *) private; + struct _DriSlab *slab; + struct _DriSlabSizeHeader *header; + + if (!buf->isSlabBuffer) { + assert(buf->bo->proposedFlags & DRM_BO_FLAG_NO_MOVE); + return buf->bo->offset; + } + + slab = buf->parent; + header = slab->header; + + (void) header; + assert(header->slabPool->proposedFlags & DRM_BO_FLAG_NO_MOVE); + return slab->kbo->bo.offset + buf->start; +} + +static unsigned long +pool_poolOffset(struct _DriBufferPool *pool, void *private) +{ + struct _DriSlabBuffer *buf = (struct _DriSlabBuffer *) private; + + return buf->start; +} + +static uint64_t +pool_flags(struct _DriBufferPool *pool, void *private) +{ + struct _DriSlabBuffer *buf = (struct _DriSlabBuffer *) private; + + if (!buf->isSlabBuffer) + return buf->bo->flags; + + return buf->parent->kbo->bo.flags; +} + +static unsigned long +pool_size(struct _DriBufferPool *pool, void *private) +{ + struct _DriSlabBuffer *buf = (struct _DriSlabBuffer *) private; + if (!buf->isSlabBuffer) + return buf->bo->size; + + return buf->parent->header->bufSize; +} + +static int +pool_fence(struct _DriBufferPool *pool, void *private, + struct _DriFenceObject *fence) +{ + struct _DriSlabBuffer *buf = (struct _DriSlabBuffer *) private; + drmBO *bo; + + if (buf->fence) + driFenceUnReference(&buf->fence); + + buf->fence = driFenceReference(fence); + bo = (buf->isSlabBuffer) ? + &buf->parent->kbo->bo: + buf->bo; + buf->fenceType = bo->fenceFlags; + + buf->unFenced = 0; + _glthread_COND_BROADCAST(buf->event); + + return 0; +} + +static drmBO * +pool_kernel(struct _DriBufferPool *pool, void *private) +{ + struct _DriSlabBuffer *buf = (struct _DriSlabBuffer *) private; + + return (buf->isSlabBuffer) ? &buf->parent->kbo->bo : buf->bo; +} + +static int +pool_validate(struct _DriBufferPool *pool, void *private, + _glthread_Mutex *mutex) +{ + struct _DriSlabBuffer *buf = (struct _DriSlabBuffer *) private; + + if (!buf->isSlabBuffer) + return 0; + + while(buf->mapCount != 0) + _glthread_COND_WAIT(buf->event, *mutex); + + buf->unFenced = 1; + return 0; +} + + +struct _DriFreeSlabManager * +driInitFreeSlabManager(uint32_t checkIntervalMsec, uint32_t slabTimeoutMsec) +{ + struct _DriFreeSlabManager *tmp; + + tmp = calloc(1, sizeof(*tmp)); + if (!tmp) + return NULL; + + _glthread_INIT_MUTEX(tmp->mutex); + _glthread_LOCK_MUTEX(tmp->mutex); + tmp->slabTimeout.tv_usec = slabTimeoutMsec*1000; + tmp->slabTimeout.tv_sec = tmp->slabTimeout.tv_usec / 1000000; + tmp->slabTimeout.tv_usec -= tmp->slabTimeout.tv_sec*1000000; + + tmp->checkInterval.tv_usec = checkIntervalMsec*1000; + tmp->checkInterval.tv_sec = tmp->checkInterval.tv_usec / 1000000; + tmp->checkInterval.tv_usec -= tmp->checkInterval.tv_sec*1000000; + + gettimeofday(&tmp->nextCheck, NULL); + driTimeAdd(&tmp->nextCheck, &tmp->checkInterval); + DRMINITLISTHEAD(&tmp->timeoutList); + DRMINITLISTHEAD(&tmp->unCached); + DRMINITLISTHEAD(&tmp->cached); + _glthread_UNLOCK_MUTEX(tmp->mutex); + + return tmp; +} + +void +driFinishFreeSlabManager(struct _DriFreeSlabManager *fMan) +{ + struct timeval time; + + time = fMan->nextCheck; + driTimeAdd(&time, &fMan->checkInterval); + + _glthread_LOCK_MUTEX(fMan->mutex); + driFreeTimeoutKBOsLocked(fMan, &time); + _glthread_UNLOCK_MUTEX(fMan->mutex); + + assert(fMan->timeoutList.next == &fMan->timeoutList); + assert(fMan->unCached.next == &fMan->unCached); + assert(fMan->cached.next == &fMan->cached); + + free(fMan); +} + +static void +driInitSizeHeader(struct _DriSlabPool *pool, uint32_t size, + struct _DriSlabSizeHeader *header) +{ + _glthread_INIT_MUTEX(header->mutex); + _glthread_LOCK_MUTEX(header->mutex); + + DRMINITLISTHEAD(&header->slabs); + DRMINITLISTHEAD(&header->freeSlabs); + DRMINITLISTHEAD(&header->delayedBuffers); + + header->numDelayed = 0; + header->slabPool = pool; + header->bufSize = size; + + _glthread_UNLOCK_MUTEX(header->mutex); +} + +static void +driFinishSizeHeader(struct _DriSlabSizeHeader *header) +{ + drmMMListHead *list, *next; + struct _DriSlabBuffer *buf; + + _glthread_LOCK_MUTEX(header->mutex); + for (list = header->delayedBuffers.next, next = list->next; + list != &header->delayedBuffers; + list = next, next = list->next) { + + buf = DRMLISTENTRY(struct _DriSlabBuffer, list , head); + if (buf->fence) { + (void) driFenceFinish(buf->fence, buf->fenceType, 0); + driFenceUnReference(&buf->fence); + } + header->numDelayed--; + driSlabFreeBufferLocked(buf); + } + _glthread_UNLOCK_MUTEX(header->mutex); +} + +static void +pool_takedown(struct _DriBufferPool *driPool) +{ + struct _DriSlabPool *pool = driPool->data; + int i; + + for (i=0; i<pool->numBuckets; ++i) { + driFinishSizeHeader(&pool->headers[i]); + } + + free(pool->headers); + free(pool->bucketSizes); + free(pool); + free(driPool); +} + +struct _DriBufferPool * +driSlabPoolInit(int fd, uint64_t flags, + uint64_t validMask, + uint32_t smallestSize, + uint32_t numSizes, + uint32_t desiredNumBuffers, + uint32_t maxSlabSize, + uint32_t pageAlignment, + struct _DriFreeSlabManager *fMan) +{ + struct _DriBufferPool *driPool; + struct _DriSlabPool *pool; + uint32_t i; + + driPool = calloc(1, sizeof(*driPool)); + if (!driPool) + return NULL; + + pool = calloc(1, sizeof(*pool)); + if (!pool) + goto out_err0; + + pool->bucketSizes = calloc(numSizes, sizeof(*pool->bucketSizes)); + if (!pool->bucketSizes) + goto out_err1; + + pool->headers = calloc(numSizes, sizeof(*pool->headers)); + if (!pool->headers) + goto out_err2; + + pool->fMan = fMan; + pool->proposedFlags = flags; + pool->validMask = validMask; + pool->numBuckets = numSizes; + pool->pageSize = getpagesize(); + pool->fd = fd; + pool->pageAlignment = pageAlignment; + pool->maxSlabSize = maxSlabSize; + pool->desiredNumBuffers = desiredNumBuffers; + + for (i=0; i<pool->numBuckets; ++i) { + pool->bucketSizes[i] = (smallestSize << i); + driInitSizeHeader(pool, pool->bucketSizes[i], + &pool->headers[i]); + } + + driPool->data = (void *) pool; + driPool->map = &pool_map; + driPool->unmap = &pool_unmap; + driPool->destroy = &pool_destroy; + driPool->offset = &pool_offset; + driPool->poolOffset = &pool_poolOffset; + driPool->flags = &pool_flags; + driPool->size = &pool_size; + driPool->create = &pool_create; + driPool->fence = &pool_fence; + driPool->kernel = &pool_kernel; + driPool->validate = &pool_validate; + driPool->waitIdle = &pool_waitIdle; + driPool->takeDown = &pool_takedown; + + return driPool; + + out_err2: + free(pool->bucketSizes); + out_err1: + free(pool); + out_err0: + free(driPool); + + return NULL; +} diff --git a/src/gallium/winsys/egl_drm/Makefile b/src/gallium/winsys/egl_drm/Makefile new file mode 100644 index 0000000..4139d9e --- /dev/null +++ b/src/gallium/winsys/egl_drm/Makefile @@ -0,0 +1,38 @@ +# src/mesa/drivers/egl_drm/Makefile + +TOP = ../../../.. + +include $(TOP)/configs/current + + + +default: $(TOP)/$(LIB_DIR) subdirs + + +$(TOP)/$(LIB_DIR): + -mkdir $(TOP)/$(LIB_DIR) + + +subdirs: + @for dir in $(DRI_DIRS) ; do \ + if [ -d $$dir ] ; then \ + (cd $$dir && $(MAKE)) || exit 1 ; \ + fi \ + done + + +install: + @for dir in $(DRI_DIRS) ; do \ + if [ -d $$dir ] ; then \ + (cd $$dir && $(MAKE) install) || exit 1 ; \ + fi \ + done + + +clean: + @for dir in $(DRI_DIRS) ; do \ + if [ -d $$dir ] ; then \ + (cd $$dir && $(MAKE) clean) ; \ + fi \ + done + -rm -f common/*.o diff --git a/src/gallium/winsys/egl_drm/Makefile.template b/src/gallium/winsys/egl_drm/Makefile.template new file mode 100644 index 0000000..3bc1fdd --- /dev/null +++ b/src/gallium/winsys/egl_drm/Makefile.template @@ -0,0 +1,116 @@ +# -*-makefile-*- + +MESA_MODULES = \ + $(TOP)/src/mesa/libmesa.a \ + $(GALLIUM_AUXILIARIES) + +COMMON_GALLIUM_SOURCES = \ + $(TOP)/src/mesa/drivers/dri/common/utils.c \ + $(TOP)/src/mesa/drivers/dri/common/vblank.c \ + $(TOP)/src/mesa/drivers/dri/common/dri_util.c \ + $(TOP)/src/mesa/drivers/dri/common/xmlconfig.c + +COMMON_SOURCES = $(COMMON_GALLIUM_SOURCES) \ + $(TOP)/src/mesa/drivers/common/driverfuncs.c \ + $(TOP)/src/mesa/drivers/dri/common/texmem.c \ + $(TOP)/src/mesa/drivers/dri/common/drirenderbuffer.c + +COMMON_BM_SOURCES = \ + $(TOP)/src/mesa/drivers/dri/common/dri_bufmgr.c \ + $(TOP)/src/mesa/drivers/dri/common/dri_drmpool.c + + +ifeq ($(WINDOW_SYSTEM),dri) +WINOBJ= +WINLIB= +INCLUDES = $(SHARED_INCLUDES) $(EXPAT_INCLUDES) + +OBJECTS = \ + $(C_SOURCES:.c=.o) \ + $(ASM_SOURCES:.S=.o) + +else +# miniglx +WINOBJ= +WINLIB=-L$(MESA)/src/glx/mini +MINIGLX_INCLUDES = -I$(TOP)/src/glx/mini +INCLUDES = $(MINIGLX_INCLUDES) \ + $(SHARED_INCLUDES) \ + $(PCIACCESS_CFLAGS) + +OBJECTS = $(C_SOURCES:.c=.o) \ + $(MINIGLX_SOURCES:.c=.o) \ + $(ASM_SOURCES:.S=.o) +endif + + +### Include directories +SHARED_INCLUDES = \ + -I. \ + -I$(TOP)/src/mesa/drivers/dri/common \ + -Iserver \ + -I$(TOP)/include \ + -I$(TOP)/include/GL/internal \ + -I$(TOP)/src/gallium/include \ + -I$(TOP)/src/gallium/auxiliary \ + -I$(TOP)/src/gallium/drivers \ + -I$(TOP)/src/mesa \ + -I$(TOP)/src/mesa/main \ + -I$(TOP)/src/mesa/glapi \ + -I$(TOP)/src/mesa/math \ + -I$(TOP)/src/mesa/transform \ + -I$(TOP)/src/mesa/shader \ + -I$(TOP)/src/mesa/swrast \ + -I$(TOP)/src/mesa/swrast_setup \ + -I$(TOP)/src/egl/main \ + -I$(TOP)/src/egl/drivers/dri \ + $(LIBDRM_CFLAGS) + + +##### RULES ##### + +.c.o: + $(CC) -c $(INCLUDES) $(CFLAGS) $(DRIVER_DEFINES) $< -o $@ + +.S.o: + $(CC) -c $(INCLUDES) $(CFLAGS) $(DRIVER_DEFINES) $< -o $@ + + +##### TARGETS ##### + +default: depend symlinks $(LIBNAME) $(TOP)/$(LIB_DIR)/$(LIBNAME) + + +$(LIBNAME): $(OBJECTS) $(MESA_MODULES) $(PIPE_DRIVERS) $(WINOBJ) Makefile $(TOP)/src/mesa/drivers/dri/Makefile.template + $(TOP)/bin/mklib -noprefix -o $@ \ + $(OBJECTS) $(PIPE_DRIVERS) $(MESA_MODULES) $(WINOBJ) $(DRI_LIB_DEPS) + + +$(TOP)/$(LIB_DIR)/$(LIBNAME): $(LIBNAME) + $(INSTALL) $(LIBNAME) $(TOP)/$(LIB_DIR) + + +depend: $(C_SOURCES) $(ASM_SOURCES) $(SYMLINKS) + rm -f depend + touch depend + $(MKDEP) $(MKDEP_OPTIONS) $(DRIVER_DEFINES) $(INCLUDES) $(C_SOURCES) \ + $(ASM_SOURCES) 2> /dev/null + + +# Emacs tags +tags: + etags `find . -name \*.[ch]` `find ../include` + + +# Remove .o and backup files +clean: + -rm -f *.o */*.o *~ *.so *~ server/*.o $(SYMLINKS) + -rm -f depend depend.bak + + +install: $(LIBNAME) + $(INSTALL) -d $(DRI_DRIVER_INSTALL_DIR) + $(INSTALL) -m 755 $(LIBNAME) $(DRI_DRIVER_INSTALL_DIR) + + +include depend diff --git a/src/gallium/winsys/egl_drm/intel/Makefile b/src/gallium/winsys/egl_drm/intel/Makefile new file mode 100644 index 0000000..5778ba7 --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/Makefile @@ -0,0 +1,40 @@ + +TOP = ../../../../.. +include $(TOP)/configs/current + +LIBNAME = EGL_i915.so + +PIPE_DRIVERS = \ + $(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \ + $(TOP)/src/gallium/drivers/i915simple/libi915simple.a + +DRIVER_SOURCES = \ + intel_winsys_pipe.c \ + intel_winsys_softpipe.c \ + intel_winsys_i915.c \ + intel_batchbuffer.c \ + intel_swapbuffers.c \ + intel_context.c \ + intel_lock.c \ + intel_screen.c \ + ws_dri_bufmgr.c \ + ws_dri_drmpool.c \ + ws_dri_fencemgr.c \ + ws_dri_mallocpool.c \ + ws_dri_slabpool.c \ + intel_egl.c + +C_SOURCES = \ + $(COMMON_GALLIUM_SOURCES) \ + $(DRIVER_SOURCES) + +ASM_SOURCES = + +DRIVER_DEFINES = -I$(TOP)/src/mesa/drivers/dri/intel $(shell pkg-config libdrm --atleast-version=2.3.1 \ + && echo "-DDRM_VBLANK_FLIP=DRM_VBLANK_FLIP") + +include ../Makefile.template + +intel_tex_layout.o: $(TOP)/src/mesa/drivers/dri/intel/intel_tex_layout.c + +symlinks: diff --git a/src/gallium/winsys/egl_drm/intel/SConscript b/src/gallium/winsys/egl_drm/intel/SConscript new file mode 100644 index 0000000..0ad19d4 --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/SConscript @@ -0,0 +1,39 @@ +Import('*') + +env = drienv.Clone() + +env.Append(CPPPATH = [ + '../intel', + 'server' +]) + +#MINIGLX_SOURCES = server/intel_dri.c + +DRIVER_SOURCES = [ + 'intel_winsys_pipe.c', + 'intel_winsys_softpipe.c', + 'intel_winsys_i915.c', + 'intel_batchbuffer.c', + 'intel_swapbuffers.c', + 'intel_context.c', + 'intel_lock.c', + 'intel_screen.c', + 'intel_batchpool.c', +] + +sources = \ + COMMON_GALLIUM_SOURCES + \ + COMMON_BM_SOURCES + \ + DRIVER_SOURCES + +drivers = [ + softpipe, + i915simple +] + +# TODO: write a wrapper function http://www.scons.org/wiki/WrapperFunctions +env.SharedLibrary( + target ='i915tex_dri.so', + source = sources, + LIBS = drivers + mesa + auxiliaries + env['LIBS'], +)
\ No newline at end of file diff --git a/src/gallium/winsys/egl_drm/intel/intel_batchbuffer.c b/src/gallium/winsys/egl_drm/intel/intel_batchbuffer.c new file mode 100644 index 0000000..7ffa05a --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/intel_batchbuffer.c @@ -0,0 +1,465 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "intel_batchbuffer.h" +#include "intel_context.h" +#include "intel_egl.h" + +#include <errno.h> + +#if 0 +static void +intel_dump_batchbuffer(GLuint offset, GLuint * ptr, GLuint count) +{ + int i; + fprintf(stderr, "\n\n\nSTART BATCH (%d dwords):\n", count / 4); + for (i = 0; i < count / 4; i += 4) + fprintf(stderr, "0x%x:\t0x%08x 0x%08x 0x%08x 0x%08x\n", + offset + i * 4, ptr[i], ptr[i + 1], ptr[i + 2], ptr[i + 3]); + fprintf(stderr, "END BATCH\n\n\n"); +} +#endif + +static void +intel_realloc_relocs(struct intel_batchbuffer *batch, int num_relocs) +{ + unsigned long size = num_relocs * I915_RELOC0_STRIDE + I915_RELOC_HEADER; + + size *= sizeof(uint32_t); + batch->reloc = realloc(batch->reloc, size); + batch->reloc_size = num_relocs; +} + + +void +intel_batchbuffer_reset(struct intel_batchbuffer *batch) +{ + /* + * Get a new, free batchbuffer. + */ + drmBO *bo; + struct drm_bo_info_req *req; + int ret; + + driBOUnrefUserList(batch->list); + driBOResetList(batch->list); + + batch->size = 4096; // ZZZ JB batch->intel->intelScreen->maxBatchSize; + ret = driBOData(batch->buffer, batch->size, NULL, NULL, 0); + assert(!ret); + + /* + * Add the batchbuffer to the validate list. + */ + + driBOAddListItem(batch->list, batch->buffer, + DRM_BO_FLAG_EXE | DRM_BO_FLAG_MEM_TT, + DRM_BO_FLAG_EXE | DRM_BO_MASK_MEM, + &batch->dest_location, &batch->node); + + req = &batch->node->bo_arg.d.req.bo_req; + + /* + * Set up information needed for us to make relocations + * relative to the underlying drm buffer objects. + */ + + driReadLockKernelBO(); + bo = driBOKernel(batch->buffer); + req->presumed_offset = (uint64_t) bo->offset; + req->hint = DRM_BO_HINT_PRESUMED_OFFSET; + batch->drmBOVirtual = (uint8_t *) bo->virtual; + driReadUnlockKernelBO(); + + /* + * Adjust the relocation buffer size. + */ + + if (batch->reloc_size > INTEL_MAX_RELOCS || + batch->reloc == NULL) + intel_realloc_relocs(batch, INTEL_DEFAULT_RELOCS); + + assert(batch->reloc != NULL); + batch->reloc[0] = 0; /* No relocs yet. */ + batch->reloc[1] = 1; /* Reloc type 1 */ + batch->reloc[2] = 0; /* Only a single relocation list. */ + batch->reloc[3] = 0; /* Only a single relocation list. */ + + batch->map = driBOMap(batch->buffer, DRM_BO_FLAG_WRITE, 0); + batch->poolOffset = driBOPoolOffset(batch->buffer); + batch->ptr = batch->map; + batch->dirty_state = ~0; + batch->nr_relocs = 0; + batch->flags = 0; + batch->id = 0;//batch->intel->intelScreen->batch_id++; +} + +/*====================================================================== + * Public functions + */ +struct intel_batchbuffer * +intel_batchbuffer_alloc(struct intel_screen *intel_screen) +{ + struct intel_batchbuffer *batch = calloc(sizeof(*batch), 1); + + batch->intel_screen = intel_screen; + + driGenBuffers(intel_screen->batchPool, "batchbuffer", 1, + &batch->buffer, 4096, + DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_EXE, 0); + batch->last_fence = NULL; + batch->list = driBOCreateList(20); + batch->reloc = NULL; + intel_batchbuffer_reset(batch); + return batch; +} + +void +intel_batchbuffer_free(struct intel_batchbuffer *batch) +{ + if (batch->last_fence) { + driFenceFinish(batch->last_fence, + DRM_FENCE_TYPE_EXE, GL_FALSE); + driFenceUnReference(&batch->last_fence); + } + if (batch->map) { + driBOUnmap(batch->buffer); + batch->map = NULL; + } + driBOUnReference(batch->buffer); + driBOFreeList(batch->list); + if (batch->reloc) + free(batch->reloc); + batch->buffer = NULL; + free(batch); +} + +void +intel_offset_relocation(struct intel_batchbuffer *batch, + unsigned pre_add, + struct _DriBufferObject *driBO, + uint64_t val_flags, + uint64_t val_mask) +{ + int itemLoc; + struct _drmBONode *node; + uint32_t *reloc; + struct drm_bo_info_req *req; + + driBOAddListItem(batch->list, driBO, val_flags, val_mask, + &itemLoc, &node); + req = &node->bo_arg.d.req.bo_req; + + if (!(req->hint & DRM_BO_HINT_PRESUMED_OFFSET)) { + + /* + * Stop other threads from tampering with the underlying + * drmBO while we're reading its offset. + */ + + driReadLockKernelBO(); + req->presumed_offset = (uint64_t) driBOKernel(driBO)->offset; + driReadUnlockKernelBO(); + req->hint = DRM_BO_HINT_PRESUMED_OFFSET; + } + + pre_add += driBOPoolOffset(driBO); + + if (batch->nr_relocs == batch->reloc_size) + intel_realloc_relocs(batch, batch->reloc_size * 2); + + reloc = batch->reloc + + (I915_RELOC_HEADER + batch->nr_relocs * I915_RELOC0_STRIDE); + + reloc[0] = ((uint8_t *)batch->ptr - batch->drmBOVirtual); + intel_batchbuffer_emit_dword(batch, req->presumed_offset + pre_add); + reloc[1] = pre_add; + reloc[2] = itemLoc; + reloc[3] = batch->dest_location; + batch->nr_relocs++; +} + +static void +i915_drm_copy_reply(const struct drm_bo_info_rep * rep, drmBO * buf) +{ + buf->handle = rep->handle; + buf->flags = rep->flags; + buf->size = rep->size; + buf->offset = rep->offset; + buf->mapHandle = rep->arg_handle; + buf->proposedFlags = rep->proposed_flags; + buf->start = rep->buffer_start; + buf->fenceFlags = rep->fence_flags; + buf->replyFlags = rep->rep_flags; + buf->pageAlignment = rep->page_alignment; +} + +static int +i915_execbuf(struct intel_batchbuffer *batch, + GLuint used, + GLboolean ignore_cliprects, + drmBOList *list, + struct drm_i915_execbuffer *ea) +{ + struct intel_screen *intel_screen = batch->intel_screen; + drmBONode *node; + drmMMListHead *l; + struct drm_i915_op_arg *arg, *first; + struct drm_bo_op_req *req; + struct drm_bo_info_rep *rep; + uint64_t *prevNext = NULL; + drmBO *buf; + int ret = 0; + uint32_t count = 0; + + first = NULL; + for (l = list->list.next; l != &list->list; l = l->next) { + node = DRMLISTENTRY(drmBONode, l, head); + + arg = &node->bo_arg; + req = &arg->d.req; + + if (!first) + first = arg; + + if (prevNext) + *prevNext = (unsigned long)arg; + + prevNext = &arg->next; + req->bo_req.handle = node->buf->handle; + req->op = drm_bo_validate; + req->bo_req.flags = node->arg0; + req->bo_req.mask = node->arg1; + req->bo_req.hint |= 0; + count++; + } + + memset(ea, 0, sizeof(*ea)); + ea->num_buffers = count; + ea->batch.start = batch->poolOffset; + ea->batch.used = used; +#if 0 /* ZZZ JB: no cliprects used */ + ea->batch.cliprects = intel->pClipRects; + ea->batch.num_cliprects = ignore_cliprects ? 0 : intel->numClipRects; + ea->batch.DR1 = 0; + ea->batch.DR4 = 0;((((GLuint) intel->drawX) & 0xffff) | + (((GLuint) intel->drawY) << 16)); +#else + ea->batch.cliprects = NULL; + ea->batch.num_cliprects = 0; + ea->batch.DR1 = 0; + ea->batch.DR4 = 0; +#endif + ea->fence_arg.flags = DRM_I915_FENCE_FLAG_FLUSHED; + ea->ops_list = (unsigned long) first; + first->reloc_ptr = (unsigned long) batch->reloc; + batch->reloc[0] = batch->nr_relocs; + + //return -EFAULT; + do { + ret = drmCommandWriteRead(intel_screen->device->drmFD, DRM_I915_EXECBUFFER, ea, + sizeof(*ea)); + } while (ret == -EAGAIN); + + if (ret != 0) { + printf("%s somebody set us up the bomb\n", __FUNCTION__); + return ret; + } + + for (l = list->list.next; l != &list->list; l = l->next) { + node = DRMLISTENTRY(drmBONode, l, head); + arg = &node->bo_arg; + rep = &arg->d.rep.bo_info; + + if (!arg->handled) { + return -EFAULT; + } + if (arg->d.rep.ret) + return arg->d.rep.ret; + + buf = node->buf; + i915_drm_copy_reply(rep, buf); + } + return 0; +} + +/* TODO: Push this whole function into bufmgr. + */ +static struct _DriFenceObject * +do_flush_locked(struct intel_batchbuffer *batch, + GLuint used, + GLboolean ignore_cliprects, GLboolean allow_unlock) +{ + struct intel_screen *intel_screen = batch->intel_screen; + struct _DriFenceObject *fo; + drmFence fence; + drmBOList *boList; + struct drm_i915_execbuffer ea; + int ret = 0; + + driBOValidateUserList(batch->list); + boList = driGetdrmBOList(batch->list); + +#if 0 /* ZZZ JB Allways run */ + if (!(intel->numClipRects == 0 && !ignore_cliprects)) { +#else + if (1) { +#endif + ret = i915_execbuf(batch, used, ignore_cliprects, boList, &ea); + } else { + driPutdrmBOList(batch->list); + fo = NULL; + goto out; + } + driPutdrmBOList(batch->list); + if (ret) + abort(); + + if (ea.fence_arg.error != 0) { + + /* + * The hardware has been idled by the kernel. + * Don't fence the driBOs. + */ + + if (batch->last_fence) + driFenceUnReference(&batch->last_fence); +#if 0 /* ZZZ JB: no _mesa_* funcs in gallium */ + _mesa_printf("fence error\n"); +#endif + batch->last_fence = NULL; + fo = NULL; + goto out; + } + + fence.handle = ea.fence_arg.handle; + fence.fence_class = ea.fence_arg.fence_class; + fence.type = ea.fence_arg.type; + fence.flags = ea.fence_arg.flags; + fence.signaled = ea.fence_arg.signaled; + + fo = driBOFenceUserList(intel_screen->mgr, batch->list, + "SuperFence", &fence); + + if (driFenceType(fo) & DRM_I915_FENCE_TYPE_RW) { + if (batch->last_fence) + driFenceUnReference(&batch->last_fence); + /* + * FIXME: Context last fence?? + */ + batch->last_fence = fo; + driFenceReference(fo); + } + out: +#if 0 /* ZZZ JB: fix this */ + intel->vtbl.lost_hardware(intel); +#else +#endif + return fo; +} + + +struct _DriFenceObject * +intel_batchbuffer_flush(struct intel_batchbuffer *batch) +{ + //struct intel_context *intel = batch->intel; + GLuint used = batch->ptr - batch->map; + //GLboolean was_locked = 1; + struct _DriFenceObject *fence; + + if (used == 0) { + driFenceReference(batch->last_fence); + return batch->last_fence; + } + + /* Add the MI_BATCH_BUFFER_END. Always add an MI_FLUSH - this is a + * performance drain that we would like to avoid. + */ +#if 0 /* ZZZ JB: what should we do here? */ + if (used & 4) { + ((int *) batch->ptr)[0] = intel->vtbl.flush_cmd(); + ((int *) batch->ptr)[1] = 0; + ((int *) batch->ptr)[2] = MI_BATCH_BUFFER_END; + used += 12; + } + else { + ((int *) batch->ptr)[0] = intel->vtbl.flush_cmd(); + ((int *) batch->ptr)[1] = MI_BATCH_BUFFER_END; + used += 8; + } +#else + if (used & 4) { + ((int *) batch->ptr)[0] = ((0<<29)|(4<<23)); // MI_FLUSH; + ((int *) batch->ptr)[1] = 0; + ((int *) batch->ptr)[2] = (0xA<<23); // MI_BATCH_BUFFER_END; + used += 12; + } + else { + ((int *) batch->ptr)[0] = ((0<<29)|(4<<23)); // MI_FLUSH; + ((int *) batch->ptr)[1] = (0xA<<23); // MI_BATCH_BUFFER_END; + used += 8; + } +#endif + driBOUnmap(batch->buffer); + batch->ptr = NULL; + batch->map = NULL; + + /* TODO: Just pass the relocation list and dma buffer up to the + * kernel. + */ +/* if (!was_locked) + LOCK_HARDWARE(intel);*/ + + fence = do_flush_locked(batch, used, !(batch->flags & INTEL_BATCH_CLIPRECTS), + GL_FALSE); + +/* if (!was_locked) + UNLOCK_HARDWARE(intel);*/ + + /* Reset the buffer: + */ + intel_batchbuffer_reset(batch); + return fence; +} + +void +intel_batchbuffer_finish(struct intel_batchbuffer *batch) +{ + struct _DriFenceObject *fence = intel_batchbuffer_flush(batch); + driFenceFinish(fence, driFenceType(fence), GL_FALSE); + driFenceUnReference(&fence); +} + +void +intel_batchbuffer_data(struct intel_batchbuffer *batch, + const void *data, GLuint bytes, GLuint flags) +{ + assert((bytes & 3) == 0); + intel_batchbuffer_require_space(batch, bytes, flags); + memcpy(batch->ptr, data, bytes); + batch->ptr += bytes; +} diff --git a/src/gallium/winsys/egl_drm/intel/intel_batchbuffer.h b/src/gallium/winsys/egl_drm/intel/intel_batchbuffer.h new file mode 100644 index 0000000..6d35cf8 --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/intel_batchbuffer.h @@ -0,0 +1,133 @@ +#ifndef INTEL_BATCHBUFFER_H +#define INTEL_BATCHBUFFER_H + +#include "mtypes.h" +#include "ws_dri_bufmgr.h" + +struct intel_screen; + +#define BATCH_SZ 16384 +#define BATCH_RESERVED 16 + +#define INTEL_DEFAULT_RELOCS 100 +#define INTEL_MAX_RELOCS 400 + +#define INTEL_BATCH_NO_CLIPRECTS 0x1 +#define INTEL_BATCH_CLIPRECTS 0x2 + +struct intel_batchbuffer +{ + struct bufmgr *bm; + struct intel_screen *intel_screen; + + struct _DriBufferObject *buffer; + struct _DriFenceObject *last_fence; + GLuint flags; + + struct _DriBufferList *list; + GLuint list_count; + GLubyte *map; + GLubyte *ptr; + + uint32_t *reloc; + GLuint reloc_size; + GLuint nr_relocs; + + GLuint size; + + GLuint dirty_state; + GLuint id; + + uint32_t poolOffset; + uint8_t *drmBOVirtual; + struct _drmBONode *node; /* Validation list node for this buffer */ + int dest_location; /* Validation list sequence for this buffer */ +}; + +struct intel_batchbuffer *intel_batchbuffer_alloc(struct intel_screen + *intel); + +void intel_batchbuffer_free(struct intel_batchbuffer *batch); + + +void intel_batchbuffer_finish(struct intel_batchbuffer *batch); + +struct _DriFenceObject *intel_batchbuffer_flush(struct intel_batchbuffer + *batch); + +void intel_batchbuffer_reset(struct intel_batchbuffer *batch); + + +/* Unlike bmBufferData, this currently requires the buffer be mapped. + * Consider it a convenience function wrapping multple + * intel_buffer_dword() calls. + */ +void intel_batchbuffer_data(struct intel_batchbuffer *batch, + const void *data, GLuint bytes, GLuint flags); + +void intel_batchbuffer_release_space(struct intel_batchbuffer *batch, + GLuint bytes); + +void +intel_offset_relocation(struct intel_batchbuffer *batch, + unsigned pre_add, + struct _DriBufferObject *driBO, + uint64_t val_flags, + uint64_t val_mask); + +/* Inline functions - might actually be better off with these + * non-inlined. Certainly better off switching all command packets to + * be passed as structs rather than dwords, but that's a little bit of + * work... + */ +static INLINE GLuint +intel_batchbuffer_space(struct intel_batchbuffer *batch) +{ + return (batch->size - BATCH_RESERVED) - (batch->ptr - batch->map); +} + + +static INLINE void +intel_batchbuffer_emit_dword(struct intel_batchbuffer *batch, GLuint dword) +{ + assert(batch->map); + assert(intel_batchbuffer_space(batch) >= 4); + *(GLuint *) (batch->ptr) = dword; + batch->ptr += 4; +} + +static INLINE void +intel_batchbuffer_require_space(struct intel_batchbuffer *batch, + GLuint sz, GLuint flags) +{ + struct _DriFenceObject *fence; + + assert(sz < batch->size - 8); + if (intel_batchbuffer_space(batch) < sz || + (batch->flags != 0 && flags != 0 && batch->flags != flags)) { + fence = intel_batchbuffer_flush(batch); + driFenceUnReference(&fence); + } + + batch->flags |= flags; +} + +/* Here are the crusty old macros, to be removed: + */ +#define BATCH_LOCALS + +#define BEGIN_BATCH(n, flags) do { \ + assert(!intel->prim.flush); \ + intel_batchbuffer_require_space(intel->batch, (n)*4, flags); \ +} while (0) + +#define OUT_BATCH(d) intel_batchbuffer_emit_dword(intel->batch, d) + +#define OUT_RELOC(buf,flags,mask,delta) do { \ + assert((delta) >= 0); \ + intel_offset_relocation(intel->batch, delta, buf, flags, mask); \ +} while (0) + +#define ADVANCE_BATCH() do { } while(0) + +#endif diff --git a/src/gallium/winsys/egl_drm/intel/intel_context.c b/src/gallium/winsys/egl_drm/intel/intel_context.c new file mode 100644 index 0000000..394a40d --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/intel_context.c @@ -0,0 +1,366 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "intel_screen.h" +#include "intel_context.h" +#include "intel_swapbuffers.h" +#include "intel_winsys.h" +#include "intel_batchbuffer.h" + +#include "state_tracker/st_public.h" +#include "pipe/p_defines.h" +#include "pipe/p_context.h" +#include "intel_egl.h" +#include "utils.h" + +#ifdef DEBUG +int __intel_debug = 0; +#endif + + +#define need_GL_ARB_multisample +#define need_GL_ARB_point_parameters +#define need_GL_ARB_texture_compression +#define need_GL_ARB_vertex_buffer_object +#define need_GL_ARB_vertex_program +#define need_GL_ARB_window_pos +#define need_GL_EXT_blend_color +#define need_GL_EXT_blend_equation_separate +#define need_GL_EXT_blend_func_separate +#define need_GL_EXT_blend_minmax +#define need_GL_EXT_cull_vertex +#define need_GL_EXT_fog_coord +#define need_GL_EXT_framebuffer_object +#define need_GL_EXT_multi_draw_arrays +#define need_GL_EXT_secondary_color +#define need_GL_NV_vertex_program +#include "extension_helper.h" + + +/** + * Extension strings exported by the intel driver. + * + * \note + * It appears that ARB_texture_env_crossbar has "disappeared" compared to the + * old i830-specific driver. + */ +const struct dri_extension card_extensions[] = { + {"GL_ARB_multisample", GL_ARB_multisample_functions}, + {"GL_ARB_multitexture", NULL}, + {"GL_ARB_point_parameters", GL_ARB_point_parameters_functions}, + {"GL_ARB_texture_border_clamp", NULL}, + {"GL_ARB_texture_compression", GL_ARB_texture_compression_functions}, + {"GL_ARB_texture_cube_map", NULL}, + {"GL_ARB_texture_env_add", NULL}, + {"GL_ARB_texture_env_combine", NULL}, + {"GL_ARB_texture_env_dot3", NULL}, + {"GL_ARB_texture_mirrored_repeat", NULL}, + {"GL_ARB_texture_rectangle", NULL}, + {"GL_ARB_vertex_buffer_object", GL_ARB_vertex_buffer_object_functions}, + {"GL_ARB_pixel_buffer_object", NULL}, + {"GL_ARB_vertex_program", GL_ARB_vertex_program_functions}, + {"GL_ARB_window_pos", GL_ARB_window_pos_functions}, + {"GL_EXT_blend_color", GL_EXT_blend_color_functions}, + {"GL_EXT_blend_equation_separate", GL_EXT_blend_equation_separate_functions}, + {"GL_EXT_blend_func_separate", GL_EXT_blend_func_separate_functions}, + {"GL_EXT_blend_minmax", GL_EXT_blend_minmax_functions}, + {"GL_EXT_blend_subtract", NULL}, + {"GL_EXT_cull_vertex", GL_EXT_cull_vertex_functions}, + {"GL_EXT_fog_coord", GL_EXT_fog_coord_functions}, + {"GL_EXT_framebuffer_object", GL_EXT_framebuffer_object_functions}, + {"GL_EXT_multi_draw_arrays", GL_EXT_multi_draw_arrays_functions}, + {"GL_EXT_packed_depth_stencil", NULL}, + {"GL_EXT_pixel_buffer_object", NULL}, + {"GL_EXT_secondary_color", GL_EXT_secondary_color_functions}, + {"GL_EXT_stencil_wrap", NULL}, + {"GL_EXT_texture_edge_clamp", NULL}, + {"GL_EXT_texture_env_combine", NULL}, + {"GL_EXT_texture_env_dot3", NULL}, + {"GL_EXT_texture_filter_anisotropic", NULL}, + {"GL_EXT_texture_lod_bias", NULL}, + {"GL_3DFX_texture_compression_FXT1", NULL}, + {"GL_APPLE_client_storage", NULL}, + {"GL_MESA_pack_invert", NULL}, + {"GL_MESA_ycbcr_texture", NULL}, + {"GL_NV_blend_square", NULL}, + {"GL_NV_vertex_program", GL_NV_vertex_program_functions}, + {"GL_NV_vertex_program1_1", NULL}, + {"GL_SGIS_generate_mipmap", NULL }, + {NULL, NULL} +}; + +int +intel_create_context(struct egl_drm_context *eglCon, const __GLcontextModes *visual, void *sharedContextPrivate) +{ + struct intel_context *iCon = CALLOC_STRUCT(intel_context); + struct intel_screen *iScrn = (struct intel_screen *)eglCon->device->priv; + struct pipe_context *pipe; + struct st_context *st_share = NULL; + + eglCon->priv = iCon; + + iCon->intel_screen = iScrn; + iCon->egl_context = eglCon; + iCon->egl_device = eglCon->device; + + iCon->batch = intel_batchbuffer_alloc(iScrn); + iCon->last_swap_fence = NULL; + iCon->first_swap_fence = NULL; + + pipe = intel_create_i915simple(iCon, iScrn->winsys); +// pipe = intel_create_softpipe(iCon, iScrn->winsys); + + pipe->priv = iCon; + + iCon->st = st_create_context(pipe, visual, st_share); + return TRUE; +} + +void +intel_make_current(struct egl_drm_context *context, struct egl_drm_drawable *draw, struct egl_drm_drawable *read) +{ + if (context) { + struct intel_context *intel = (struct intel_context *)context->priv; + struct intel_framebuffer *draw_fb = (struct intel_framebuffer *)draw->priv; + struct intel_framebuffer *read_fb = (struct intel_framebuffer *)read->priv; + + assert(draw_fb->stfb); + assert(read_fb->stfb); + + st_make_current(intel->st, draw_fb->stfb, read_fb->stfb); + + intel->egl_drawable = draw; + + st_resize_framebuffer(draw_fb->stfb, draw->w, draw->h); + + if (draw != read) + st_resize_framebuffer(read_fb->stfb, read->w, read->h); + + //intelUpdateWindowSize(driDrawPriv); + } else { + st_make_current(NULL, NULL, NULL); + } +} + +void +intel_bind_frontbuffer(struct egl_drm_drawable *draw, struct egl_drm_frontbuffer *front) +{ + struct intel_screen *intelScreen = (struct intel_screen *)draw->device->priv; + struct intel_framebuffer *draw_fb = (struct intel_framebuffer *)draw->priv; + + driBOUnReference(draw_fb->front_buffer); + draw_fb->front_buffer = NULL; + draw_fb->front = NULL; + + /* to unbind just call this function with front == NULL */ + if (!front) + return; + + draw_fb->front = front; + + driGenBuffers(intelScreen->staticPool, "front", 1, &draw_fb->front_buffer, 0, 0, 0); + driBOSetReferenced(draw_fb->front_buffer, front->handle); + + st_resize_framebuffer(draw_fb->stfb, draw->w, draw->h); +} + +#if 0 +GLboolean +intelCreateContext(const __GLcontextModes * visual, + __DRIcontextPrivate * driContextPriv, + void *sharedContextPrivate) +{ + struct intel_context *intel = CALLOC_STRUCT(intel_context); + __DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv; + struct intel_screen *intelScreen = intel_screen(sPriv); + drmI830Sarea *saPriv = intelScreen->sarea; + int fthrottle_mode; + GLboolean havePools; + struct pipe_context *pipe; + struct st_context *st_share = NULL; + + if (sharedContextPrivate) { + st_share = ((struct intel_context *) sharedContextPrivate)->st; + } + + driContextPriv->driverPrivate = intel; + intel->intelScreen = intelScreen; + intel->driScreen = sPriv; + intel->sarea = saPriv; + + driParseConfigFiles(&intel->optionCache, &intelScreen->optionCache, + intel->driScreen->myNum, "i915"); + + + /* + * memory pools + */ + DRM_LIGHT_LOCK(sPriv->fd, &sPriv->pSAREA->lock, driContextPriv->hHWContext); + // ZZZ JB should be per screen and not be done per context + havePools = intelCreatePools(sPriv); + DRM_UNLOCK(sPriv->fd, &sPriv->pSAREA->lock, driContextPriv->hHWContext); + if (!havePools) + return GL_FALSE; + + + /* Dri stuff */ + intel->hHWContext = driContextPriv->hHWContext; + intel->driFd = sPriv->fd; + intel->driHwLock = (drmLock *) & sPriv->pSAREA->lock; + + fthrottle_mode = driQueryOptioni(&intel->optionCache, "fthrottle_mode"); + intel->iw.irq_seq = -1; + intel->irqsEmitted = 0; + + intel->batch = intel_batchbuffer_alloc(intel); + intel->last_swap_fence = NULL; + intel->first_swap_fence = NULL; + +#ifdef DEBUG + __intel_debug = driParseDebugString(getenv("INTEL_DEBUG"), debug_control); +#endif + + /* + * Pipe-related setup + */ + if (getenv("INTEL_SP")) { + /* use softpipe driver instead of hw */ + pipe = intel_create_softpipe( intel, intelScreen->winsys ); + } + else { + switch (intel->intelScreen->deviceID) { + case PCI_CHIP_I945_G: + case PCI_CHIP_I945_GM: + case PCI_CHIP_I945_GME: + case PCI_CHIP_G33_G: + case PCI_CHIP_Q33_G: + case PCI_CHIP_Q35_G: + case PCI_CHIP_I915_G: + case PCI_CHIP_I915_GM: + pipe = intel_create_i915simple( intel, intelScreen->winsys ); + break; + default: + fprintf(stderr, "Unknown PCIID %x in %s, using software driver\n", + intel->intelScreen->deviceID, __FUNCTION__); + + pipe = intel_create_softpipe( intel, intelScreen->winsys ); + break; + } + } + + pipe->priv = intel; + + intel->st = st_create_context(pipe, visual, st_share); + + return GL_TRUE; +} + + +void +intelDestroyContext(__DRIcontextPrivate * driContextPriv) +{ + struct intel_context *intel = intel_context(driContextPriv); + + assert(intel); /* should never be null */ + if (intel) { + st_finish(intel->st); + + intel_batchbuffer_free(intel->batch); + + if (intel->last_swap_fence) { + driFenceFinish(intel->last_swap_fence, DRM_FENCE_TYPE_EXE, GL_TRUE); + driFenceUnReference(&intel->last_swap_fence); + intel->last_swap_fence = NULL; + } + if (intel->first_swap_fence) { + driFenceFinish(intel->first_swap_fence, DRM_FENCE_TYPE_EXE, GL_TRUE); + driFenceUnReference(&intel->first_swap_fence); + intel->first_swap_fence = NULL; + } + + if (intel->intelScreen->dummyContext == intel) + intel->intelScreen->dummyContext = NULL; + + st_destroy_context(intel->st); + free(intel); + } +} + + +GLboolean +intelUnbindContext(__DRIcontextPrivate * driContextPriv) +{ + struct intel_context *intel = intel_context(driContextPriv); + st_flush(intel->st, PIPE_FLUSH_RENDER_CACHE, NULL); + /* XXX make_current(NULL)? */ + return GL_TRUE; +} + + +GLboolean +intelMakeCurrent(__DRIcontextPrivate * driContextPriv, + __DRIdrawablePrivate * driDrawPriv, + __DRIdrawablePrivate * driReadPriv) +{ + if (driContextPriv) { + struct intel_context *intel = intel_context(driContextPriv); + struct intel_framebuffer *draw_fb = intel_framebuffer(driDrawPriv); + struct intel_framebuffer *read_fb = intel_framebuffer(driReadPriv); + + assert(draw_fb->stfb); + assert(read_fb->stfb); + + /* This is for situations in which we need a rendering context but + * there may not be any currently bound. + */ + intel->intelScreen->dummyContext = intel; + + st_make_current(intel->st, draw_fb->stfb, read_fb->stfb); + + if ((intel->driDrawable != driDrawPriv) || + (intel->lastStamp != driDrawPriv->lastStamp)) { + intel->driDrawable = driDrawPriv; + intelUpdateWindowSize(driDrawPriv); + intel->lastStamp = driDrawPriv->lastStamp; + } + + /* The size of the draw buffer will have been updated above. + * If the readbuffer is a different window, check/update its size now. + */ + if (driReadPriv != driDrawPriv) { + intelUpdateWindowSize(driReadPriv); + } + + } + else { + st_make_current(NULL, NULL, NULL); + } + + return GL_TRUE; +} +#endif diff --git a/src/gallium/winsys/egl_drm/intel/intel_context.h b/src/gallium/winsys/egl_drm/intel/intel_context.h new file mode 100644 index 0000000..aa9903f --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/intel_context.h @@ -0,0 +1,162 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef INTEL_CONTEXT_H +#define INTEL_CONTEXT_H + +#include <stdint.h> +#include "drm.h" + +#include "pipe/p_debug.h" + +#include "intel_screen.h" +#include "i915_drm.h" + + +struct pipe_context; +struct intel_context; +struct _DriBufferObject; +struct st_context; +struct egl_drm_device; +struct egl_drm_context; +struct egl_drm_frontbuffer; + + +#define INTEL_MAX_FIXUP 64 + +/** + * Intel rendering context, contains a state tracker and intel-specific info. + */ +struct intel_context +{ + struct st_context *st; + + struct _DriFenceObject *last_swap_fence; + struct _DriFenceObject *first_swap_fence; + + struct intel_batchbuffer *batch; + +#if 0 + boolean locked; + char *prevLockFile; + int prevLockLine; +#endif + + /* pick this up from the screen instead + int drmFd; + */ + + struct intel_screen *intel_screen; + + uint lastStamp; + /* new egl stuff */ + struct egl_drm_device *egl_device; + struct egl_drm_context *egl_context; + struct egl_drm_drawable *egl_drawable; +}; + + + +/** + * Intel framebuffer. + */ +struct intel_framebuffer +{ + struct st_framebuffer *stfb; + + /* other fields TBD */ + int other; + struct _DriBufferObject *front_buffer; + struct egl_drm_frontbuffer *front; +}; + + + + +/* These are functions now: + */ +void LOCK_HARDWARE( struct intel_context *intel ); +void UNLOCK_HARDWARE( struct intel_context *intel ); + +extern char *__progname; + + + +/* ================================================================ + * Debugging: + */ +#ifdef DEBUG +extern int __intel_debug; + +#define DEBUG_SWAP 0x1 +#define DEBUG_LOCK 0x2 +#define DEBUG_IOCTL 0x4 +#define DEBUG_BATCH 0x8 + +#define DBG(flag, ...) do { \ + if (__intel_debug & (DEBUG_##flag)) \ + printf(__VA_ARGS__); \ +} while(0) + +#else +#define DBG(flag, ...) +#endif + + + +#define PCI_CHIP_845_G 0x2562 +#define PCI_CHIP_I830_M 0x3577 +#define PCI_CHIP_I855_GM 0x3582 +#define PCI_CHIP_I865_G 0x2572 +#define PCI_CHIP_I915_G 0x2582 +#define PCI_CHIP_I915_GM 0x2592 +#define PCI_CHIP_I945_G 0x2772 +#define PCI_CHIP_I945_GM 0x27A2 +#define PCI_CHIP_I945_GME 0x27AE +#define PCI_CHIP_G33_G 0x29C2 +#define PCI_CHIP_Q35_G 0x29B2 +#define PCI_CHIP_Q33_G 0x29D2 + + +#if 0 +/** Cast wrapper */ +static INLINE struct intel_context * +intel_context(__DRIcontextPrivate *driContextPriv) +{ + return (struct intel_context *) driContextPriv->driverPrivate; +} + + +/** Cast wrapper */ +static INLINE struct intel_framebuffer * +intel_framebuffer(__DRIdrawablePrivate * driDrawPriv) +{ + return (struct intel_framebuffer *) driDrawPriv->driverPrivate; +} +#endif + +#endif diff --git a/src/gallium/winsys/egl_drm/intel/intel_egl.c b/src/gallium/winsys/egl_drm/intel/intel_egl.c new file mode 100644 index 0000000..809371f --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/intel_egl.c @@ -0,0 +1,621 @@ + +#include <assert.h> +#include <stdlib.h> +#include <string.h> +#include <stdio.h> +#include <stdint.h> + +#include "eglconfig.h" +#include "eglcontext.h" +#include "egldisplay.h" +#include "egldriver.h" +#include "eglglobals.h" +#include "eglmode.h" +#include "eglscreen.h" +#include "eglsurface.h" + +#include "intel_egl.h" + +#include "xf86drm.h" +#include "xf86drmMode.h" + +#include "intel_context.h" + +#include "ws_dri_bufmgr.h" + +#include "intel_winsys.h" +#include "state_tracker/st_public.h" + +struct egl_drm_device* egl_drm_create_device(int drmFD); + +struct egl_drm_device* +egl_drm_create_device(int drmFD) +{ + struct egl_drm_device *device = malloc(sizeof(*device)); + memset(device, 0, sizeof(*device)); + device->drmFD = drmFD; + + if (!intel_init_driver(device)) { + printf("EGL: failed to initalize device\n"); + free(device); + } + + return device; +} + +__GLcontextModes* _gl_context_modes_create( unsigned count, size_t minimum_size ); + +struct drm_driver +{ + _EGLDriver base; /* base class/object */ + + drmModeResPtr res; + struct egl_drm_device *device; +}; + +struct drm_surface +{ + _EGLSurface base; /* base class/object */ + + struct egl_drm_drawable *drawable; +}; + +struct drm_context +{ + _EGLContext base; /* base class/object */ + + struct egl_drm_context *context; +}; + +struct drm_screen +{ + _EGLScreen base; + + /* backing buffer and crtc */ + drmBO buffer; + drmModeFBPtr fb; + uint32_t fbID; + drmModeCrtcPtr crtc; + + /* currently only support one output */ + drmModeOutputPtr output; + uint32_t outputID; + + struct drm_mode_modeinfo *mode; + + /* geometry of the screen */ + struct egl_drm_frontbuffer front; +}; + +static void +drm_update_res(struct drm_driver *drm_drv) +{ + drmModeFreeResources(drm_drv->res); + drm_drv->res = drmModeGetResources(drm_drv->device->drmFD); +} + +static void +drm_add_modes_from_output(_EGLScreen *screen, drmModeOutputPtr output) +{ + struct drm_mode_modeinfo *m; + int i; + + for (i = 0; i < output->count_modes; i++) { + m = &output->modes[i]; + _eglAddNewMode(screen, m->hdisplay, m->vdisplay, m->vrefresh, m->name); + } +} + +static EGLBoolean +drm_initialize(_EGLDriver *drv, EGLDisplay dpy, EGLint *major, EGLint *minor) +{ + _EGLDisplay *disp = _eglLookupDisplay(dpy); + struct drm_driver *drm_drv = (struct drm_driver *)drv; + struct drm_screen *screen = NULL; + drmModeOutputPtr output = NULL; + drmModeResPtr res = NULL; + unsigned count_outputs = 0; + + EGLint i; + int fd; + + fd = drmOpen("i915", NULL); + if (fd < 0) { + return EGL_FALSE; + } + + drm_drv->device = egl_drm_create_device(fd); + if (!drm_drv->device) { + drmClose(fd); + return EGL_FALSE; + } + + drm_update_res(drm_drv); + res = drm_drv->res; + if (res) + count_outputs = res->count_outputs; + + for(i = 0; i < count_outputs; i++) { + output = drmModeGetOutput(fd, res->outputs[i]); + + if (!output) + continue; + + if (output->connection == DRM_MODE_DISCONNECTED) { + drmModeFreeOutput(output); + continue; + } + + screen = malloc(sizeof(struct drm_screen)); + memset(screen, 0, sizeof(*screen)); + screen->outputID = res->outputs[i]; + screen->output = output; + _eglInitScreen(&screen->base); + _eglAddScreen(disp, &screen->base); + drm_add_modes_from_output(&screen->base, output); + } + + /* for now we only have one config */ + _EGLConfig config; + _eglInitConfig(&config, i + 1); + _eglSetConfigAttrib(&config, EGL_RED_SIZE, 8); + _eglSetConfigAttrib(&config, EGL_GREEN_SIZE, 8); + _eglSetConfigAttrib(&config, EGL_BLUE_SIZE, 8); + _eglSetConfigAttrib(&config, EGL_ALPHA_SIZE, 8); + _eglSetConfigAttrib(&config, EGL_BUFFER_SIZE, 32); + _eglSetConfigAttrib(&config, EGL_DEPTH_SIZE, 24); + _eglSetConfigAttrib(&config, EGL_STENCIL_SIZE, 8); + _eglSetConfigAttrib(&config, EGL_SURFACE_TYPE, EGL_PBUFFER_BIT); + _eglAddConfig(disp, &config); + + drv->Initialized = EGL_TRUE; + + *major = 1; + *minor = 0; + + return EGL_TRUE; +} + + +static EGLBoolean +drm_terminate(_EGLDriver *drv, EGLDisplay dpy) +{ + /* TODO: clean up */ + free(drv); + return EGL_TRUE; +} + + +static struct drm_context * +lookup_drm_context(EGLContext context) +{ + _EGLContext *c = _eglLookupContext(context); + return (struct drm_context *) c; +} + + +static struct drm_surface * +lookup_drm_surface(EGLSurface surface) +{ + _EGLSurface *s = _eglLookupSurface(surface); + return (struct drm_surface *) s; +} + +static struct drm_screen * +lookup_drm_screen(EGLDisplay dpy, EGLScreenMESA screen) +{ + _EGLScreen *s = _eglLookupScreen(dpy, screen); + return (struct drm_screen *) s; +} + +static __GLcontextModes* +visual_from_config(_EGLConfig *conf) +{ + __GLcontextModes *visual; + (void)conf; + + visual = _gl_context_modes_create(1, sizeof(*visual)); + visual->redBits = 8; + visual->greenBits = 8; + visual->blueBits = 8; + visual->alphaBits = 8; + + visual->rgbBits = 32; + visual->doubleBufferMode = 1; + + visual->depthBits = 24; + visual->stencilBits = 8; + + return visual; +} + + + +static EGLContext +drm_create_context(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, EGLContext share_list, const EGLint *attrib_list) +{ + struct drm_driver *drm_drv = (struct drm_driver *)drv; + struct drm_context *c; + struct drm_egl_context *share = NULL; + _EGLConfig *conf; + int i; + int ret; + __GLcontextModes *visual; + struct egl_drm_context *context; + + conf = _eglLookupConfig(drv, dpy, config); + if (!conf) { + _eglError(EGL_BAD_CONFIG, "eglCreateContext"); + return EGL_NO_CONTEXT; + } + + for (i = 0; attrib_list && attrib_list[i] != EGL_NONE; i++) { + switch (attrib_list[i]) { + /* no attribs defined for now */ + default: + _eglError(EGL_BAD_ATTRIBUTE, "eglCreateContext"); + return EGL_NO_CONTEXT; + } + } + + c = (struct drm_context *) calloc(1, sizeof(struct drm_context)); + if (!c) + return EGL_NO_CONTEXT; + + _eglInitContext(drv, dpy, &c->base, config, attrib_list); + + context = malloc(sizeof(*context)); + memset(context, 0, sizeof(*context)); + + if (!context) + goto err_c; + + context->device = drm_drv->device; + visual = visual_from_config(conf); + + ret = intel_create_context(context, visual, share); + free(visual); + + if (!ret) + goto err_gl; + + c->context = context; + + /* generate handle and insert into hash table */ + _eglSaveContext(&c->base); + assert(_eglGetContextHandle(&c->base)); + + return _eglGetContextHandle(&c->base); +err_gl: + free(context); +err_c: + free(c); + return EGL_NO_CONTEXT; +} + +static EGLBoolean +drm_destroy_context(_EGLDriver *drv, EGLDisplay dpy, EGLContext context) +{ + struct drm_context *fc = lookup_drm_context(context); + _eglRemoveContext(&fc->base); + if (fc->base.IsBound) { + fc->base.DeletePending = EGL_TRUE; + } else { + free(fc); + } + return EGL_TRUE; +} + + +static EGLSurface +drm_create_window_surface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, NativeWindowType window, const EGLint *attrib_list) +{ + return EGL_NO_SURFACE; +} + + +static EGLSurface +drm_create_pixmap_surface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, NativePixmapType pixmap, const EGLint *attrib_list) +{ + return EGL_NO_SURFACE; +} + +static EGLSurface +drm_create_pbuffer_surface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, + const EGLint *attrib_list) +{ + struct drm_driver *drm_drv = (struct drm_driver *)drv; + int i; + int ret; + int width = -1; + int height = -1; + struct drm_surface *surf = NULL; + struct egl_drm_drawable *drawable = NULL; + __GLcontextModes *visual; + _EGLConfig *conf; + + conf = _eglLookupConfig(drv, dpy, config); + if (!conf) { + _eglError(EGL_BAD_CONFIG, "eglCreatePbufferSurface"); + return EGL_NO_CONTEXT; + } + + for (i = 0; attrib_list && attrib_list[i] != EGL_NONE; i++) { + switch (attrib_list[i]) { + case EGL_WIDTH: + width = attrib_list[++i]; + break; + case EGL_HEIGHT: + height = attrib_list[++i]; + break; + default: + _eglError(EGL_BAD_ATTRIBUTE, "eglCreatePbufferSurface"); + return EGL_NO_SURFACE; + } + } + + if (width < 1 || height < 1) { + _eglError(EGL_BAD_ATTRIBUTE, "eglCreatePbufferSurface"); + return EGL_NO_SURFACE; + } + + surf = (struct drm_surface *) calloc(1, sizeof(struct drm_surface)); + if (!surf) + goto err; + + if (!_eglInitSurface(drv, dpy, &surf->base, EGL_PBUFFER_BIT, config, attrib_list)) + goto err_surf; + + drawable = malloc(sizeof(*drawable)); + memset(drawable, 0, sizeof(*drawable)); + + drawable->w = width; + drawable->h = height; + + visual = visual_from_config(conf); + + drawable->device = drm_drv->device; + ret = intel_create_drawable(drawable, visual); + free(visual); + + if (!ret) + goto err_draw; + + surf->drawable = drawable; + + _eglSaveSurface(&surf->base); + return surf->base.Handle; + +err_draw: + free(drawable); +err_surf: + free(surf); +err: + return EGL_NO_SURFACE; +} + +static EGLSurface +drm_create_screen_surface_mesa(_EGLDriver *drv, EGLDisplay dpy, EGLConfig cfg, + const EGLint *attrib_list) +{ + EGLSurface surf = drm_create_pbuffer_surface(drv, dpy, cfg, attrib_list); + + return surf; +} + +static struct drm_mode_modeinfo * +drm_find_mode(drmModeOutputPtr output, _EGLMode *mode) +{ + int i; + struct drm_mode_modeinfo *m; + + for (i = 0; i < output->count_modes; i++) { + m = &output->modes[i]; + if (m->hdisplay == mode->Width && m->vdisplay == mode->Height && m->vrefresh == mode->RefreshRate) + break; + m = NULL; + } + + return m; +} +static void +draw(size_t x, size_t y, size_t w, size_t h, size_t pitch, size_t v, unsigned int *ptr) +{ + int i, j; + + for (i = x; i < x + w; i++) + for(j = y; j < y + h; j++) + ptr[(i * pitch / 4) + j] = v; + +} + +static void +prettyColors(int fd, unsigned int handle, size_t pitch) +{ + drmBO bo; + unsigned int *ptr; + int i; + + drmBOReference(fd, handle, &bo); + drmBOMap(fd, &bo, DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE, 0, (void**)&ptr); + + for (i = 0; i < (bo.size / 4); i++) + ptr[i] = 0xFFFFFFFF; + + for (i = 0; i < 4; i++) + draw(i * 40, i * 40, 40, 40, pitch, 0, ptr); + + + draw(200, 100, 40, 40, pitch, 0xff00ff, ptr); + draw(100, 200, 40, 40, pitch, 0xff00ff, ptr); + + drmBOUnmap(fd, &bo); +} + +static EGLBoolean +drm_show_screen_surface_mesa(_EGLDriver *drv, EGLDisplay dpy, + EGLScreenMESA screen, + EGLSurface surface, EGLModeMESA m) +{ + struct drm_driver *drm_drv = (struct drm_driver *)drv; + struct drm_surface *surf = lookup_drm_surface(surface); + struct drm_screen *scrn = lookup_drm_screen(dpy, screen); + //struct intel_framebuffer *intel_fb = NULL; + //struct pipe_surface *front_surf = NULL; + _EGLMode *mode = _eglLookupMode(dpy, m); + size_t pitch = 2048 * 4; + size_t size = mode->Height * pitch; + int ret; + + /* TODO if allready shown take down */ + + printf("setting mode to %i x %i\n", mode->Width, mode->Height); + + ret = drmBOCreate(drm_drv->device->drmFD, size, 0, 0, + DRM_BO_FLAG_READ | + DRM_BO_FLAG_WRITE | + DRM_BO_FLAG_MEM_TT | + DRM_BO_FLAG_MEM_VRAM | + DRM_BO_FLAG_NO_EVICT, + DRM_BO_HINT_DONT_FENCE, &scrn->buffer); + + if (ret) { + printf("failed to create framebuffer (ret %d)\n", ret); + return EGL_FALSE; + } + + prettyColors(drm_drv->device->drmFD, scrn->buffer.handle, pitch); + + ret = drmModeAddFB(drm_drv->device->drmFD, mode->Width, mode->Height, + 32, 32, pitch, + scrn->buffer.handle, + &scrn->fbID); + + if (ret) + goto err_bo; + + scrn->fb = drmModeGetFB(drm_drv->device->drmFD, scrn->fbID); + if (!scrn->fb) + goto err_bo; + + scrn->mode = drm_find_mode(scrn->output, mode); + if (!scrn->mode) { + printf("oh noes, no matching mode found\n"); + goto err_fb; + } + + ret = drmModeSetCrtc( + drm_drv->device->drmFD, + drm_drv->res->crtcs[1], + scrn->fbID, + 0, 0, + &scrn->outputID, 1, + scrn->mode); + + + scrn->front.handle = scrn->buffer.handle; + scrn->front.pitch = pitch; + scrn->front.width = mode->Width; + scrn->front.height = mode->Height; + + intel_bind_frontbuffer(surf->drawable, &scrn->front); + + return EGL_TRUE; + +err_fb: + /* TODO remove fb */ + +err_bo: + drmBOUnreference(drm_drv->device->drmFD, &scrn->buffer); + return EGL_FALSE; +} + +static EGLBoolean +drm_destroy_surface(_EGLDriver *drv, EGLDisplay dpy, EGLSurface surface) +{ + struct drm_surface *fs = lookup_drm_surface(surface); + _eglRemoveSurface(&fs->base); + if (fs->base.IsBound) { + fs->base.DeletePending = EGL_TRUE; + } else { + free(fs); + } + return EGL_TRUE; +} + + + + +static EGLBoolean +drm_make_current(_EGLDriver *drv, EGLDisplay dpy, EGLSurface draw, EGLSurface read, EGLContext context) +{ + struct drm_surface *readSurf = lookup_drm_surface(read); + struct drm_surface *drawSurf = lookup_drm_surface(draw); + struct drm_context *ctx = lookup_drm_context(context); + EGLBoolean b; + + b = _eglMakeCurrent(drv, dpy, draw, read, context); + if (!b) + return EGL_FALSE; + + /* XXX this is where we'd do the hardware context switch */ + (void) drawSurf; + (void) readSurf; + (void) ctx; + + intel_make_current(ctx->context, drawSurf->drawable, readSurf->drawable); + return EGL_TRUE; +} + +static EGLBoolean +drm_swap_buffers(_EGLDriver *drv, EGLDisplay dpy, EGLSurface draw) +{ + struct drm_surface *surf = lookup_drm_surface(draw); + if (!surf) + return EGL_FALSE; + + /* error checking */ + if (!_eglSwapBuffers(drv, dpy, draw)) + return EGL_FALSE; + + intel_swap_buffers(surf->drawable); + return EGL_TRUE; +} + + +/** + * The bootstrap function. Return a new drm_driver object and + * plug in API functions. + */ +_EGLDriver * +_eglMain(_EGLDisplay *dpy) +{ + struct drm_driver *drm; + + drm = (struct drm_driver *) calloc(1, sizeof(struct drm_driver)); + if (!drm) { + return NULL; + } + + /* First fill in the dispatch table with defaults */ + _eglInitDriverFallbacks(&drm->base); + /* then plug in our Drm-specific functions */ + drm->base.API.Initialize = drm_initialize; + drm->base.API.Terminate = drm_terminate; + drm->base.API.CreateContext = drm_create_context; + drm->base.API.MakeCurrent = drm_make_current; + drm->base.API.CreateWindowSurface = drm_create_window_surface; + drm->base.API.CreatePixmapSurface = drm_create_pixmap_surface; + drm->base.API.CreatePbufferSurface = drm_create_pbuffer_surface; + drm->base.API.DestroySurface = drm_destroy_surface; + drm->base.API.DestroyContext = drm_destroy_context; + drm->base.API.CreateScreenSurfaceMESA = drm_create_screen_surface_mesa; + drm->base.API.ShowScreenSurfaceMESA = drm_show_screen_surface_mesa; + drm->base.API.SwapBuffers = drm_swap_buffers; + + /* enable supported extensions */ + drm->base.Extensions.MESA_screen_surface = EGL_TRUE; + drm->base.Extensions.MESA_copy_context = EGL_TRUE; + + return &drm->base; +} diff --git a/src/gallium/winsys/egl_drm/intel/intel_egl.h b/src/gallium/winsys/egl_drm/intel/intel_egl.h new file mode 100644 index 0000000..18e84e8 --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/intel_egl.h @@ -0,0 +1,42 @@ + +#ifndef _INTEL_EGL_H_ +#define _INTEL_EGL_H_ + +struct egl_drm_device +{ + void *priv; + int drmFD; +}; + +struct egl_drm_context +{ + void *priv; + struct egl_drm_device *device; +}; + +struct egl_drm_drawable +{ + void *priv; + struct egl_drm_device *device; + size_t h; + size_t w; +}; + +struct egl_drm_frontbuffer +{ + uint32_t handle; + uint32_t pitch; + uint32_t width; + uint32_t height; +}; + +#include "GL/internal/glcore.h" + +int intel_init_driver(struct egl_drm_device *device); +int intel_create_context(struct egl_drm_context *context, const __GLcontextModes *visual, void *sharedContextPrivate); +int intel_create_drawable(struct egl_drm_drawable *drawable, const __GLcontextModes * visual); +void intel_make_current(struct egl_drm_context *context, struct egl_drm_drawable *draw, struct egl_drm_drawable *read); +void intel_swap_buffers(struct egl_drm_drawable *draw); +void intel_bind_frontbuffer(struct egl_drm_drawable *draw, struct egl_drm_frontbuffer *front); + +#endif diff --git a/src/gallium/winsys/egl_drm/intel/intel_lock.c b/src/gallium/winsys/egl_drm/intel/intel_lock.c new file mode 100644 index 0000000..cec83c7 --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/intel_lock.c @@ -0,0 +1,102 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "main/glheader.h" +#include "glapi/glthread.h" +#include <GL/internal/glcore.h> +#include "state_tracker/st_public.h" +#include "intel_context.h" + +#if 0 + +_glthread_DECLARE_STATIC_MUTEX( lockMutex ); + + +static void +intelContendedLock(struct intel_context *intel, uint flags) +{ + __DRIdrawablePrivate *dPriv = intel->driDrawable; + __DRIscreenPrivate *sPriv = intel->driScreen; + struct intel_screen *intelScreen = intel_screen(sPriv); + drmI830Sarea *sarea = intel->sarea; + + drmGetLock(intel->driFd, intel->hHWContext, flags); + + DBG(LOCK, "%s - got contended lock\n", __progname); + + /* If the window moved, may need to set a new cliprect now. + * + * NOTE: This releases and regains the hw lock, so all state + * checking must be done *after* this call: + */ + if (dPriv) + DRI_VALIDATE_DRAWABLE_INFO(sPriv, dPriv); + + if (sarea->width != intelScreen->front.width || + sarea->height != intelScreen->front.height) { + + intelUpdateScreenRotation(sPriv, sarea); + } +} + + +/* Lock the hardware and validate our state. + */ +void LOCK_HARDWARE( struct intel_context *intel ) +{ + char __ret = 0; + + _glthread_LOCK_MUTEX(lockMutex); + assert(!intel->locked); + + DRM_CAS(intel->driHwLock, intel->hHWContext, + (DRM_LOCK_HELD|intel->hHWContext), __ret); + + if (__ret) + intelContendedLock( intel, 0 ); + + DBG(LOCK, "%s - locked\n", __progname); + + intel->locked = 1; +} + + +/* Unlock the hardware using the global current context + */ +void UNLOCK_HARDWARE( struct intel_context *intel ) +{ + assert(intel->locked); + intel->locked = 0; + + DRM_UNLOCK(intel->driFd, intel->driHwLock, intel->hHWContext); + + _glthread_UNLOCK_MUTEX(lockMutex); + + DBG(LOCK, "%s - unlocked\n", __progname); +} +#endif diff --git a/src/gallium/winsys/egl_drm/intel/intel_reg.h b/src/gallium/winsys/egl_drm/intel/intel_reg.h new file mode 100644 index 0000000..f37c24f --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/intel_reg.h @@ -0,0 +1,53 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#ifndef _INTEL_REG_H_ +#define _INTEL_REG_H_ + + +#define BR00_BITBLT_CLIENT 0x40000000 +#define BR00_OP_COLOR_BLT 0x10000000 +#define BR00_OP_SRC_COPY_BLT 0x10C00000 +#define BR13_SOLID_PATTERN 0x80000000 + +#define XY_COLOR_BLT_CMD ((2<<29)|(0x50<<22)|0x4) +#define XY_COLOR_BLT_WRITE_ALPHA (1<<21) +#define XY_COLOR_BLT_WRITE_RGB (1<<20) + +#define XY_SRC_COPY_BLT_CMD ((2<<29)|(0x53<<22)|6) +#define XY_SRC_COPY_BLT_WRITE_ALPHA (1<<21) +#define XY_SRC_COPY_BLT_WRITE_RGB (1<<20) + +#define MI_WAIT_FOR_EVENT ((0x3<<23)) +#define MI_WAIT_FOR_PLANE_B_FLIP (1<<6) +#define MI_WAIT_FOR_PLANE_A_FLIP (1<<2) + +#define MI_BATCH_BUFFER_END (0xA<<23) + + +#endif diff --git a/src/gallium/winsys/egl_drm/intel/intel_screen.c b/src/gallium/winsys/egl_drm/intel/intel_screen.c new file mode 100644 index 0000000..38c4098 --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/intel_screen.c @@ -0,0 +1,680 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "utils.h" +#include "vblank.h" +#include "xmlpool.h" + +#include "intel_context.h" +#include "intel_screen.h" +#include "intel_batchbuffer.h" +//#include "intel_batchpool.h" +#include "intel_swapbuffers.h" +#include "intel_winsys.h" + +#include "ws_dri_bufpool.h" + +#include "pipe/p_context.h" +#include "state_tracker/st_public.h" +#include "state_tracker/st_cb_fbo.h" +#include "intel_egl.h" + +static boolean +intel_create_pools(struct intel_screen *intel_screen) +{ + if (intel_screen->havePools) + return GL_TRUE; + + intel_screen->mgr = driFenceMgrTTMInit(intel_screen->device->drmFD); + if (!intel_screen->mgr) { + fprintf(stderr, "Failed to create fence manager.\n"); + return FALSE; + } + + intel_screen->fMan = driInitFreeSlabManager(10, 10); + if (!intel_screen->fMan) { + fprintf(stderr, "Failed to create free slab manager.\n"); + return FALSE; + } + + intel_screen->staticPool = driDRMPoolInit(intel_screen->device->drmFD); + intel_screen->batchPool = driSlabPoolInit(intel_screen->device->drmFD, + DRM_BO_FLAG_EXE | + DRM_BO_FLAG_MEM_TT, + DRM_BO_FLAG_EXE | + DRM_BO_FLAG_MEM_TT, + 4096, //intelScreen->maxBatchSize, + 1, 40, 16*16384, 0, + intel_screen->fMan); + + intel_screen->havePools = GL_TRUE; + + return GL_TRUE; +} + +extern const struct dri_extension card_extensions[]; + +int +intel_init_driver(struct egl_drm_device *device) +{ + struct intel_screen *intel_screen; + + /* Allocate the private area */ + intel_screen = CALLOC_STRUCT(intel_screen); + if (!intel_screen) + return FALSE; + + device->priv = (void *)intel_screen; + intel_screen->device = device; + + if (!intel_create_pools(intel_screen)) + return FALSE; + + intel_screen->batch = intel_batchbuffer_alloc(intel_screen); + + intel_screen->winsys = intel_create_pipe_winsys(device->drmFD, intel_screen->fMan); + + /* hack */ + driInitExtensions(NULL, card_extensions, GL_FALSE); + + return TRUE; +} + +int +intel_create_drawable(struct egl_drm_drawable *drawable, + const __GLcontextModes * visual) +{ + enum pipe_format colorFormat, depthFormat, stencilFormat; + struct intel_framebuffer *intelfb = CALLOC_STRUCT(intel_framebuffer); + + if (!intelfb) + return GL_FALSE; + + if (visual->redBits == 5) + colorFormat = PIPE_FORMAT_R5G6B5_UNORM; + else + colorFormat = PIPE_FORMAT_A8R8G8B8_UNORM; + + if (visual->depthBits == 16) + depthFormat = PIPE_FORMAT_Z16_UNORM; + else if (visual->depthBits == 24) + depthFormat = PIPE_FORMAT_S8Z24_UNORM; + else + depthFormat = PIPE_FORMAT_NONE; + + if (visual->stencilBits == 8) + stencilFormat = PIPE_FORMAT_S8Z24_UNORM; + else + stencilFormat = PIPE_FORMAT_NONE; + + intelfb->stfb = st_create_framebuffer(visual, + colorFormat, + depthFormat, + stencilFormat, + drawable->w, + drawable->h, + (void*) intelfb); + + if (!intelfb->stfb) { + free(intelfb); + return GL_FALSE; + } + + drawable->priv = (void *) intelfb; + return GL_TRUE; +} + +#if 0 +PUBLIC const char __driConfigOptions[] = + DRI_CONF_BEGIN DRI_CONF_SECTION_PERFORMANCE + DRI_CONF_FTHROTTLE_MODE(DRI_CONF_FTHROTTLE_IRQS) + DRI_CONF_VBLANK_MODE(DRI_CONF_VBLANK_DEF_INTERVAL_0) + DRI_CONF_SECTION_END DRI_CONF_SECTION_QUALITY + DRI_CONF_FORCE_S3TC_ENABLE(false) + DRI_CONF_ALLOW_LARGE_TEXTURES(1) + DRI_CONF_SECTION_END DRI_CONF_END; + +const uint __driNConfigOptions = 4; + +#ifdef USE_NEW_INTERFACE +static PFNGLXCREATECONTEXTMODES create_context_modes = NULL; +#endif /*USE_NEW_INTERFACE */ + +extern const struct dri_extension card_extensions[]; + + + + +static void +intelPrintDRIInfo(struct intel_screen * intelScreen, + __DRIscreenPrivate * sPriv, I830DRIPtr gDRIPriv) +{ + fprintf(stderr, "*** Front size: 0x%x offset: 0x%x pitch: %d\n", + intelScreen->front.size, intelScreen->front.offset, + intelScreen->front.pitch); + fprintf(stderr, "*** Memory : 0x%x\n", gDRIPriv->mem); +} + +static void +intelPrintSAREA(const drmI830Sarea * sarea) +{ + fprintf(stderr, "SAREA: sarea width %d height %d\n", sarea->width, + sarea->height); + fprintf(stderr, "SAREA: pitch: %d\n", sarea->pitch); + fprintf(stderr, + "SAREA: front offset: 0x%08x size: 0x%x handle: 0x%x\n", + sarea->front_offset, sarea->front_size, + (unsigned) sarea->front_handle); + fprintf(stderr, + "SAREA: back offset: 0x%08x size: 0x%x handle: 0x%x\n", + sarea->back_offset, sarea->back_size, + (unsigned) sarea->back_handle); + fprintf(stderr, "SAREA: depth offset: 0x%08x size: 0x%x handle: 0x%x\n", + sarea->depth_offset, sarea->depth_size, + (unsigned) sarea->depth_handle); + fprintf(stderr, "SAREA: tex offset: 0x%08x size: 0x%x handle: 0x%x\n", + sarea->tex_offset, sarea->tex_size, (unsigned) sarea->tex_handle); + fprintf(stderr, "SAREA: rotation: %d\n", sarea->rotation); + fprintf(stderr, + "SAREA: rotated offset: 0x%08x size: 0x%x\n", + sarea->rotated_offset, sarea->rotated_size); + fprintf(stderr, "SAREA: rotated pitch: %d\n", sarea->rotated_pitch); +} + +/** + * Use the information in the sarea to update the screen parameters + * related to screen rotation. Needs to be called locked. + */ +void +intelUpdateScreenRotation(__DRIscreenPrivate * sPriv, drmI830Sarea * sarea) +{ + struct intel_screen *intelScreen = intel_screen(sPriv); + + if (intelScreen->front.map) { + drmUnmap(intelScreen->front.map, intelScreen->front.size); + intelScreen->front.map = NULL; + } + + if (intelScreen->front.buffer) + driDeleteBuffers(1, &intelScreen->front.buffer); + + intelScreen->front.width = sarea->width; + intelScreen->front.height = sarea->height; + intelScreen->front.offset = sarea->front_offset; + intelScreen->front.pitch = sarea->pitch * intelScreen->front.cpp; + intelScreen->front.size = sarea->front_size; + intelScreen->front.handle = sarea->front_handle; + + assert( sarea->front_size >= + intelScreen->front.pitch * intelScreen->front.height ); + +#if 0 /* JB not important */ + if (!sarea->front_handle) + return; + + if (drmMap(sPriv->fd, + sarea->front_handle, + intelScreen->front.size, + (drmAddress *) & intelScreen->front.map) != 0) { + fprintf(stderr, "drmMap(frontbuffer) failed!\n"); + return; + } +#endif + +#if 0 /* JB */ + if (intelScreen->staticPool) { + driGenBuffers(intelScreen->staticPool, "static region", 1, + &intelScreen->front.buffer, 64, + DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_NO_MOVE | + DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE, 0); + + driBOSetStatic(intelScreen->front.buffer, + intelScreen->front.offset, + intelScreen->front.pitch * intelScreen->front.height, + intelScreen->front.map, 0); + } +#else + if (intelScreen->staticPool) { + if (intelScreen->front.buffer) + driBOUnReference(intelScreen->front.buffer); + driGenBuffers(intelScreen->staticPool, "front", 1, &intelScreen->front.buffer, 0, 0, 0); + driBOSetReferenced(intelScreen->front.buffer, sarea->front_bo_handle); + } +#endif +} + + +boolean +intelCreatePools(__DRIscreenPrivate * sPriv) +{ + //unsigned batchPoolSize = 1024*1024; + struct intel_screen *intelScreen = intel_screen(sPriv); + + if (intelScreen->havePools) + return GL_TRUE; + +#if 0 /* ZZZ JB fix this */ + intelScreen->staticPool = driDRMStaticPoolInit(sPriv->fd); + if (!intelScreen->staticPool) + return GL_FALSE; + + batchPoolSize /= BATCH_SZ; + intelScreen->batchPool = driBatchPoolInit(sPriv->fd, + DRM_BO_FLAG_EXE | + DRM_BO_FLAG_MEM_TT | + DRM_BO_FLAG_MEM_LOCAL, + BATCH_SZ, + batchPoolSize, 5); + if (!intelScreen->batchPool) { + fprintf(stderr, "Failed to initialize batch pool - possible incorrect agpgart installed\n"); + return GL_FALSE; + } +#else + intelScreen->staticPool = driDRMPoolInit(sPriv->fd); + intelScreen->batchPool = driSlabPoolInit(sPriv->fd, + DRM_BO_FLAG_EXE | + DRM_BO_FLAG_MEM_TT, + DRM_BO_FLAG_EXE | + DRM_BO_FLAG_MEM_TT, + 4096, //intelScreen->maxBatchSize, + 1, 40, 16*16384, 0, + intelScreen->fMan); +#endif + intelScreen->havePools = GL_TRUE; + + //intelUpdateScreenRotation(sPriv, intelScreen->sarea); + + return GL_TRUE; +} + + +static boolean +intelInitDriver(__DRIscreenPrivate * sPriv) +{ + struct intel_screen *intelScreen; + I830DRIPtr gDRIPriv = (I830DRIPtr) sPriv->pDevPriv; + + PFNGLXSCRENABLEEXTENSIONPROC glx_enable_extension = + (PFNGLXSCRENABLEEXTENSIONPROC) (*dri_interface-> + getProcAddress("glxEnableExtension")); + void *const psc = sPriv->psc->screenConfigs; + + if (sPriv->devPrivSize != sizeof(I830DRIRec)) { + fprintf(stderr, + "\nERROR! sizeof(I830DRIRec) does not match passed size from device driver\n"); + return GL_FALSE; + } + + /* Allocate the private area */ + intelScreen = CALLOC_STRUCT(intel_screen); + if (!intelScreen) + return GL_FALSE; + + /* parse information in __driConfigOptions */ + driParseOptionInfo(&intelScreen->optionCache, + __driConfigOptions, __driNConfigOptions); + + sPriv->private = (void *) intelScreen; + + intelScreen->sarea = (drmI830Sarea *) (((GLubyte *) sPriv->pSAREA) + + gDRIPriv->sarea_priv_offset); + intelScreen->deviceID = gDRIPriv->deviceID; + intelScreen->front.cpp = gDRIPriv->cpp; + intelScreen->drmMinor = sPriv->drmMinor; + + + assert(gDRIPriv->bitsPerPixel == 16 || + gDRIPriv->bitsPerPixel == 32); + + + intelUpdateScreenRotation(sPriv, intelScreen->sarea); + + if (0) + intelPrintDRIInfo(intelScreen, sPriv, gDRIPriv); + + if (glx_enable_extension != NULL) { + (*glx_enable_extension) (psc, "GLX_SGI_swap_control"); + (*glx_enable_extension) (psc, "GLX_SGI_video_sync"); + (*glx_enable_extension) (psc, "GLX_MESA_swap_control"); + (*glx_enable_extension) (psc, "GLX_MESA_swap_frame_usage"); + (*glx_enable_extension) (psc, "GLX_SGI_make_current_read"); + } + + + +#if 1 // ZZZ JB + intelScreen->mgr = driFenceMgrTTMInit(sPriv->fd); + if (!intelScreen->mgr) { + fprintf(stderr, "Failed to create fence manager.\n"); + return GL_FALSE; + } + + intelScreen->fMan = driInitFreeSlabManager(10, 10); + if (!intelScreen->fMan) { + fprintf(stderr, "Failed to create free slab manager.\n"); + return GL_FALSE; + } + + if (!intelCreatePools(sPriv)) + return GL_FALSE; +#endif + + intelScreen->winsys = intel_create_pipe_winsys(sPriv->fd, intelScreen->fMan); + + return GL_TRUE; +} + + +static void +intelDestroyScreen(__DRIscreenPrivate * sPriv) +{ + struct intel_screen *intelScreen = intel_screen(sPriv); + + /* intelUnmapScreenRegions(intelScreen); */ + + if (intelScreen->havePools) { + driPoolTakeDown(intelScreen->staticPool); + driPoolTakeDown(intelScreen->batchPool); + } + FREE(intelScreen); + sPriv->private = NULL; +} + + +/** + * This is called when we need to set up GL rendering to a new X window. + */ +static boolean +intelCreateBuffer(__DRIscreenPrivate * driScrnPriv, + __DRIdrawablePrivate * driDrawPriv, + const __GLcontextModes * visual, boolean isPixmap) +{ + if (isPixmap) { + return GL_FALSE; /* not implemented */ + } + else { + enum pipe_format colorFormat, depthFormat, stencilFormat; + struct intel_framebuffer *intelfb = CALLOC_STRUCT(intel_framebuffer); + + if (!intelfb) + return GL_FALSE; + + if (visual->redBits == 5) + colorFormat = PIPE_FORMAT_R5G6B5_UNORM; + else + colorFormat = PIPE_FORMAT_A8R8G8B8_UNORM; + + if (visual->depthBits == 16) + depthFormat = PIPE_FORMAT_Z16_UNORM; + else if (visual->depthBits == 24) + depthFormat = PIPE_FORMAT_S8Z24_UNORM; + else + depthFormat = PIPE_FORMAT_NONE; + + if (visual->stencilBits == 8) + stencilFormat = PIPE_FORMAT_S8Z24_UNORM; + else + stencilFormat = PIPE_FORMAT_NONE; + + intelfb->stfb = st_create_framebuffer(visual, + colorFormat, + depthFormat, + stencilFormat, + driDrawPriv->w, + driDrawPriv->h, + (void*) intelfb); + if (!intelfb->stfb) { + free(intelfb); + return GL_FALSE; + } + + driDrawPriv->driverPrivate = (void *) intelfb; + return GL_TRUE; + } +} + +static void +intelDestroyBuffer(__DRIdrawablePrivate * driDrawPriv) +{ + struct intel_framebuffer *intelfb = intel_framebuffer(driDrawPriv); + assert(intelfb->stfb); + st_unreference_framebuffer(&intelfb->stfb); + free(intelfb); +} + + +/** + * Get information about previous buffer swaps. + */ +static int +intelGetSwapInfo(__DRIdrawablePrivate * dPriv, __DRIswapInfo * sInfo) +{ + if ((dPriv == NULL) || (dPriv->driverPrivate == NULL) + || (sInfo == NULL)) { + return -1; + } + + return 0; +} + + +static void +intelSetTexOffset(__DRIcontext *pDRICtx, int texname, + unsigned long long offset, int depth, uint pitch) +{ + abort(); +#if 0 + struct intel_context *intel = (struct intel_context*) + ((__DRIcontextPrivate*)pDRICtx->private)->driverPrivate; + struct gl_texture_object *tObj = _mesa_lookup_texture(&intel->ctx, texname); + struct st_texture_object *stObj = st_texture_object(tObj); + + if (!stObj) + return; + + if (stObj->pt) + st->pipe->texture_release(intel->st->pipe, &stObj->pt); + + stObj->imageOverride = GL_TRUE; + stObj->depthOverride = depth; + stObj->pitchOverride = pitch; + + if (offset) + stObj->textureOffset = offset; +#endif +} + + +static const struct __DriverAPIRec intelAPI = { + .InitDriver = intelInitDriver, + .DestroyScreen = intelDestroyScreen, + .CreateContext = intelCreateContext, + .DestroyContext = intelDestroyContext, + .CreateBuffer = intelCreateBuffer, + .DestroyBuffer = intelDestroyBuffer, + .SwapBuffers = intelSwapBuffers, + .MakeCurrent = intelMakeCurrent, + .UnbindContext = intelUnbindContext, + .GetSwapInfo = intelGetSwapInfo, + .GetMSC = driGetMSC32, + .WaitForMSC = driWaitForMSC32, + .WaitForSBC = NULL, + .SwapBuffersMSC = NULL, + .CopySubBuffer = intelCopySubBuffer, + .setTexOffset = intelSetTexOffset, +}; + + +static __GLcontextModes * +intelFillInModes(unsigned pixel_bits, unsigned depth_bits, + unsigned stencil_bits, boolean have_back_buffer) +{ + __GLcontextModes *modes; + __GLcontextModes *m; + unsigned num_modes; + unsigned depth_buffer_factor; + unsigned back_buffer_factor; + GLenum fb_format; + GLenum fb_type; + + /* GLX_SWAP_COPY_OML is only supported because the Intel driver doesn't + * support pageflipping at all. + */ + static const GLenum back_buffer_modes[] = { + GLX_NONE, GLX_SWAP_UNDEFINED_OML, GLX_SWAP_COPY_OML + }; + + u_int8_t depth_bits_array[3]; + u_int8_t stencil_bits_array[3]; + + + depth_bits_array[0] = 0; + depth_bits_array[1] = depth_bits; + depth_bits_array[2] = depth_bits; + + /* Just like with the accumulation buffer, always provide some modes + * with a stencil buffer. It will be a sw fallback, but some apps won't + * care about that. + */ + stencil_bits_array[0] = 0; + stencil_bits_array[1] = 0; + if (depth_bits == 24) + stencil_bits_array[1] = (stencil_bits == 0) ? 8 : stencil_bits; + + stencil_bits_array[2] = (stencil_bits == 0) ? 8 : stencil_bits; + + depth_buffer_factor = ((depth_bits != 0) || (stencil_bits != 0)) ? 3 : 1; + back_buffer_factor = (have_back_buffer) ? 3 : 1; + + num_modes = depth_buffer_factor * back_buffer_factor * 4; + + if (pixel_bits == 16) { + fb_format = GL_RGB; + fb_type = GL_UNSIGNED_SHORT_5_6_5; + } + else { + fb_format = GL_BGRA; + fb_type = GL_UNSIGNED_INT_8_8_8_8_REV; + } + + modes = + (*dri_interface->createContextModes) (num_modes, + sizeof(__GLcontextModes)); + m = modes; + if (!driFillInModes(&m, fb_format, fb_type, + depth_bits_array, stencil_bits_array, + depth_buffer_factor, back_buffer_modes, + back_buffer_factor, GLX_TRUE_COLOR)) { + fprintf(stderr, "[%s:%u] Error creating FBConfig!\n", __func__, + __LINE__); + return NULL; + } + if (!driFillInModes(&m, fb_format, fb_type, + depth_bits_array, stencil_bits_array, + depth_buffer_factor, back_buffer_modes, + back_buffer_factor, GLX_DIRECT_COLOR)) { + fprintf(stderr, "[%s:%u] Error creating FBConfig!\n", __func__, + __LINE__); + return NULL; + } + + /* Mark the visual as slow if there are "fake" stencil bits. + */ + for (m = modes; m != NULL; m = m->next) { + if ((m->stencilBits != 0) && (m->stencilBits != stencil_bits)) { + m->visualRating = GLX_SLOW_CONFIG; + } + } + + return modes; +} + + +/** + * This is the bootstrap function for the driver. libGL supplies all of the + * requisite information about the system, and the driver initializes itself. + * This routine also fills in the linked list pointed to by \c driver_modes + * with the \c __GLcontextModes that the driver can support for windows or + * pbuffers. + * + * \return A pointer to a \c __DRIscreenPrivate on success, or \c NULL on + * failure. + */ +PUBLIC void * +__driCreateNewScreen_20050727(__DRInativeDisplay * dpy, int scrn, + __DRIscreen * psc, + const __GLcontextModes * modes, + const __DRIversion * ddx_version, + const __DRIversion * dri_version, + const __DRIversion * drm_version, + const __DRIframebuffer * frame_buffer, + drmAddress pSAREA, int fd, + int internal_api_version, + const __DRIinterfaceMethods * interface, + __GLcontextModes ** driver_modes) +{ + __DRIscreenPrivate *psp; + static const __DRIversion ddx_expected = { 1, 7, 0 }; + static const __DRIversion dri_expected = { 4, 0, 0 }; + static const __DRIversion drm_expected = { 1, 7, 0 }; + + dri_interface = interface; + + if (!driCheckDriDdxDrmVersions2("i915", + dri_version, &dri_expected, + ddx_version, &ddx_expected, + drm_version, &drm_expected)) { + return NULL; + } + + psp = __driUtilCreateNewScreen(dpy, scrn, psc, NULL, + ddx_version, dri_version, drm_version, + frame_buffer, pSAREA, fd, + internal_api_version, &intelAPI); + + if (psp != NULL) { + I830DRIPtr dri_priv = (I830DRIPtr) psp->pDevPriv; + *driver_modes = intelFillInModes(dri_priv->cpp * 8, + (dri_priv->cpp == 2) ? 16 : 24, + (dri_priv->cpp == 2) ? 0 : 8, 1); + + /* Calling driInitExtensions here, with a NULL context pointer, + * does not actually enable the extensions. It just makes sure + * that all the dispatch offsets for all the extensions that + * *might* be enables are known. This is needed because the + * dispatch offsets need to be known when _mesa_context_create + * is called, but we can't enable the extensions until we have a + * context pointer. + * + * Hello chicken. Hello egg. How are you two today? + */ + driInitExtensions(NULL, card_extensions, GL_FALSE); + } + + return (void *) psp; +} +#endif diff --git a/src/gallium/winsys/egl_drm/intel/intel_screen.h b/src/gallium/winsys/egl_drm/intel/intel_screen.h new file mode 100644 index 0000000..e8c1cdf --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/intel_screen.h @@ -0,0 +1,133 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef _INTEL_SCREEN_H_ +#define _INTEL_SCREEN_H_ + +#include "ws_dri_bufpool.h" + +#include "pipe/p_compiler.h" + +struct egl_drm_device *device; + +struct intel_screen +{ +#if 0 + struct { + drm_handle_t handle; + + /* We create a static dri buffer for the frontbuffer. + */ + struct _DriBufferObject *buffer; + + char *map; /* memory map */ + int offset; /* from start of video mem, in bytes */ + int pitch; /* row stride, in bytes */ + int width; + int height; + int size; + int cpp; /* for front and back buffers */ + } front; +#endif + + int drmFB; + +#if 0 + int deviceID; + int drmMinor; + + + drmI830Sarea *sarea;*/ + + + /** + * Configuration cache with default values for all contexts + */ + driOptionCache optionCache; +#endif + + struct _DriBufferPool *batchPool; + struct _DriBufferPool *staticPool; /** for the X screen/framebuffer */ + boolean havePools; + +#if 0 + /** + * Temporary(?) context to use for SwapBuffers or other situations in + * which we need a rendering context, but none is currently bound. + */ + struct intel_context *dummyContext; +#endif + + /* + * New stuff form the i915tex integration + */ + struct _DriFenceMgr *mgr; + struct _DriFreeSlabManager *fMan; + unsigned batch_id; + + struct pipe_winsys *winsys; + struct egl_drm_device *device; + + /* batch buffer used for swap buffers */ + struct intel_batchbuffer *batch; +}; + + + +/** cast wrapper */ +#if 0 +static INLINE struct intel_screen * +intel_screen(__DRIscreenPrivate *sPriv) +{ + return (struct intel_screen *) sPriv->private; +} + + +extern void +intelUpdateScreenRotation(__DRIscreenPrivate * sPriv, drmI830Sarea * sarea); + + +extern void intelDestroyContext(__DRIcontextPrivate * driContextPriv); + +extern boolean intelUnbindContext(__DRIcontextPrivate * driContextPriv); + +extern boolean +intelMakeCurrent(__DRIcontextPrivate * driContextPriv, + __DRIdrawablePrivate * driDrawPriv, + __DRIdrawablePrivate * driReadPriv); + + +extern boolean +intelCreatePools(__DRIscreenPrivate *sPriv); + +extern boolean +intelCreateContext(const __GLcontextModes * visual, + __DRIcontextPrivate * driContextPriv, + void *sharedContextPrivate); + +#endif +#endif diff --git a/src/gallium/winsys/egl_drm/intel/intel_swapbuffers.c b/src/gallium/winsys/egl_drm/intel/intel_swapbuffers.c new file mode 100644 index 0000000..1ce4b27 --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/intel_swapbuffers.c @@ -0,0 +1,327 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "intel_screen.h" +#include "intel_context.h" +#include "intel_swapbuffers.h" +#include "intel_batchbuffer.h" +#include "intel_reg.h" +#include "intel_winsys.h" + +#include "pipe/p_context.h" +#include "state_tracker/st_public.h" +#include "state_tracker/st_context.h" +#include "state_tracker/st_cb_fbo.h" +#include "intel_egl.h" + + +static void +intel_display_surface(struct egl_drm_drawable *draw, + struct pipe_surface *surf); + +void intel_swap_buffers(struct egl_drm_drawable *draw) +{ + struct intel_framebuffer *intel_fb = (struct intel_framebuffer *)draw->priv; + struct pipe_surface *back_surf; + + assert(intel_fb); + assert(intel_fb->stfb); + + back_surf = st_get_framebuffer_surface(intel_fb->stfb, ST_SURFACE_BACK_LEFT); + if (back_surf) { + st_notify_swapbuffers(intel_fb->stfb); + intel_display_surface(draw, back_surf); + st_notify_swapbuffers_complete(intel_fb->stfb); + } +} + +static void +intel_display_surface(struct egl_drm_drawable *draw, + struct pipe_surface *surf) +{ + struct intel_screen *intel = (struct intel_screen *)draw->device->priv; + struct intel_framebuffer *intel_fb = (struct intel_framebuffer *)draw->priv; + struct _DriFenceObject *fence; + + //const int srcWidth = surf->width; + //const int srcHeight = surf->height; + const int srcPitch = surf->pitch; + + const int dstWidth = intel_fb->front->width; + const int dstHeight = intel_fb->front->height; + const int dstPitch = intel_fb->front->pitch / 4;//draw->front.cpp; + + const int cpp = 4;//intel_fb->front->cpp; + + int BR13, CMD; + //int i; + + + BR13 = (dstPitch * cpp) | (0xCC << 16) | (1 << 24) | (1 << 25); + CMD = (XY_SRC_COPY_BLT_CMD | XY_SRC_COPY_BLT_WRITE_ALPHA | + XY_SRC_COPY_BLT_WRITE_RGB); + + printf("srcPitch: %u, dstWidth: %u, dstHeight: %u, dstPitch: %u, cpp: %u\n", srcPitch, dstWidth, dstHeight, dstPitch, cpp); + BEGIN_BATCH(8, INTEL_BATCH_NO_CLIPRECTS); + OUT_BATCH(CMD); + OUT_BATCH(BR13); + OUT_BATCH((0 << 16) | 0); + OUT_BATCH((dstHeight << 16) | dstWidth); + + OUT_RELOC(intel_fb->front_buffer, + DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE, + DRM_BO_MASK_MEM | DRM_BO_FLAG_WRITE, 0); + + OUT_BATCH((0 << 16) | 0); + OUT_BATCH((srcPitch * cpp) & 0xffff); + OUT_RELOC(dri_bo(surf->buffer), + DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ, + DRM_BO_MASK_MEM | DRM_BO_FLAG_READ, 0); + + ADVANCE_BATCH(); + + fence = intel_batchbuffer_flush(intel->batch); + driFenceUnReference(&fence); + intel_batchbuffer_finish(intel->batch); +} + +#if 0 +/** + * Display a colorbuffer surface in an X window. + * Used for SwapBuffers and flushing front buffer rendering. + * + * \param dPriv the window/drawable to display into + * \param surf the surface to display + * \param rect optional subrect of surface to display (may be NULL). + */ +void +intelDisplaySurface(__DRIdrawablePrivate *dPriv, + struct pipe_surface *surf, + const drm_clip_rect_t *rect) +{ + struct intel_screen *intelScreen = intel_screen(dPriv->driScreenPriv); + //struct intel_context *intel = intelScreen->dummyContext; + + DBG(SWAP, "%s\n", __FUNCTION__); + +#if 0 + if (!intel) { + /* XXX this is where some kind of extra/meta context could be useful */ + return; + } +#endif + + if (intel->last_swap_fence) { + driFenceFinish(intel->last_swap_fence, DRM_FENCE_TYPE_EXE, TRUE); + driFenceUnReference(&intel->last_swap_fence); + intel->last_swap_fence = NULL; + } + intel->last_swap_fence = intel->first_swap_fence; + intel->first_swap_fence = NULL; + + /* The LOCK_HARDWARE is required for the cliprects. Buffer offsets + * should work regardless. + */ + LOCK_HARDWARE(intel); + /* if this drawable isn't currently bound the LOCK_HARDWARE done on the + * current context (which is what intelScreenContext should return) might + * not get a contended lock and thus cliprects not updated (tests/manywin) + */ + if (intel_context(dPriv->driContextPriv) != intel) + DRI_VALIDATE_DRAWABLE_INFO(intel->driScreen, dPriv); + + + if (dPriv && dPriv->numClipRects) { + const int srcWidth = surf->width; + const int srcHeight = surf->height; + const int nbox = dPriv->numClipRects; + const drm_clip_rect_t *pbox = dPriv->pClipRects; + const int pitch = intelScreen->front.pitch / intelScreen->front.cpp; + const int cpp = intelScreen->front.cpp; + const int srcpitch = surf->pitch; + int BR13, CMD; + int i; + + ASSERT(surf->buffer); + ASSERT(surf->cpp == cpp); + + DBG(SWAP, "screen pitch %d src surface pitch %d\n", + pitch, surf->pitch); + + if (cpp == 2) { + BR13 = (pitch * cpp) | (0xCC << 16) | (1 << 24); + CMD = XY_SRC_COPY_BLT_CMD; + } + else { + BR13 = (pitch * cpp) | (0xCC << 16) | (1 << 24) | (1 << 25); + CMD = (XY_SRC_COPY_BLT_CMD | XY_SRC_COPY_BLT_WRITE_ALPHA | + XY_SRC_COPY_BLT_WRITE_RGB); + } + + for (i = 0; i < nbox; i++, pbox++) { + drm_clip_rect_t box; + drm_clip_rect_t sbox; + + if (pbox->x1 > pbox->x2 || + pbox->y1 > pbox->y2 || + pbox->x2 > intelScreen->front.width || + pbox->y2 > intelScreen->front.height) { + /* invalid cliprect, skip it */ + continue; + } + + box = *pbox; + + if (rect) { + /* intersect cliprect with user-provided src rect */ + drm_clip_rect_t rrect; + + rrect.x1 = dPriv->x + rect->x1; + rrect.y1 = (dPriv->h - rect->y1 - rect->y2) + dPriv->y; + rrect.x2 = rect->x2 + rrect.x1; + rrect.y2 = rect->y2 + rrect.y1; + if (rrect.x1 > box.x1) + box.x1 = rrect.x1; + if (rrect.y1 > box.y1) + box.y1 = rrect.y1; + if (rrect.x2 < box.x2) + box.x2 = rrect.x2; + if (rrect.y2 < box.y2) + box.y2 = rrect.y2; + + if (box.x1 > box.x2 || box.y1 > box.y2) + continue; + } + + /* restrict blit to size of actually rendered area */ + if (box.x2 - box.x1 > srcWidth) + box.x2 = srcWidth + box.x1; + if (box.y2 - box.y1 > srcHeight) + box.y2 = srcHeight + box.y1; + + DBG(SWAP, "box x1 x2 y1 y2 %d %d %d %d\n", + box.x1, box.x2, box.y1, box.y2); + + sbox.x1 = box.x1 - dPriv->x; + sbox.y1 = box.y1 - dPriv->y; + + assert(box.x1 < box.x2); + assert(box.y1 < box.y2); + + /* XXX this could be done with pipe->surface_copy() */ + BEGIN_BATCH(8, INTEL_BATCH_NO_CLIPRECTS); + OUT_BATCH(CMD); + OUT_BATCH(BR13); + OUT_BATCH((box.y1 << 16) | box.x1); + OUT_BATCH((box.y2 << 16) | box.x2); + + OUT_RELOC(intelScreen->front.buffer, + DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE, + DRM_BO_MASK_MEM | DRM_BO_FLAG_WRITE, 0); + OUT_BATCH((sbox.y1 << 16) | sbox.x1); + OUT_BATCH((srcpitch * cpp) & 0xffff); + OUT_RELOC(dri_bo(surf->buffer), + DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ, + DRM_BO_MASK_MEM | DRM_BO_FLAG_READ, 0); + + ADVANCE_BATCH(); + } + + if (intel->first_swap_fence) + driFenceUnReference(&intel->first_swap_fence); + intel->first_swap_fence = intel_batchbuffer_flush(intel->batch); + } + + UNLOCK_HARDWARE(intel); + + if (intel->lastStamp != dPriv->lastStamp) { + intelUpdateWindowSize(dPriv); + intel->lastStamp = dPriv->lastStamp; + } +} + + + +/** + * This will be called whenever the currently bound window is moved/resized. + */ +void +intelUpdateWindowSize(__DRIdrawablePrivate *dPriv) +{ + struct intel_framebuffer *intelfb = intel_framebuffer(dPriv); + assert(intelfb->stfb); + st_resize_framebuffer(intelfb->stfb, dPriv->w, dPriv->h); +} + + + +void +intelSwapBuffers(__DRIdrawablePrivate * dPriv) +{ + struct intel_framebuffer *intel_fb = intel_framebuffer(dPriv); + struct pipe_surface *back_surf; + + assert(intel_fb); + assert(intel_fb->stfb); + + back_surf = st_get_framebuffer_surface(intel_fb->stfb, + ST_SURFACE_BACK_LEFT); + if (back_surf) { + st_notify_swapbuffers(intel_fb->stfb); + intelDisplaySurface(dPriv, back_surf, NULL); + st_notify_swapbuffers_complete(intel_fb->stfb); + } +} + + +/** + * Called via glXCopySubBufferMESA() to copy a subrect of the back + * buffer to the front buffer/screen. + */ +void +intelCopySubBuffer(__DRIdrawablePrivate * dPriv, int x, int y, int w, int h) +{ + struct intel_framebuffer *intel_fb = intel_framebuffer(dPriv); + struct pipe_surface *back_surf; + + assert(intel_fb); + assert(intel_fb->stfb); + + back_surf = st_get_framebuffer_surface(intel_fb->stfb, + ST_SURFACE_BACK_LEFT); + if (back_surf) { + drm_clip_rect_t rect; + rect.x1 = x; + rect.y1 = y; + rect.x2 = w; + rect.y2 = h; + + st_notify_swapbuffers(intel_fb->stfb); + intelDisplaySurface(dPriv, back_surf, &rect); + } +} +#endif diff --git a/src/gallium/winsys/egl_drm/intel/intel_swapbuffers.h b/src/gallium/winsys/egl_drm/intel/intel_swapbuffers.h new file mode 100644 index 0000000..904f267 --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/intel_swapbuffers.h @@ -0,0 +1,47 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef INTEL_SWAPBUFFERS_H +#define INTEL_SWAPBUFFERS_H + + +struct pipe_surface; + +#if 0 +extern void intelDisplaySurface(__DRIdrawablePrivate * dPriv, + struct pipe_surface *surf, + const drm_clip_rect_t * rect); + +extern void intelSwapBuffers(__DRIdrawablePrivate * dPriv); + +extern void intelCopySubBuffer(__DRIdrawablePrivate * dPriv, + int x, int y, int w, int h); + +extern void intelUpdateWindowSize(__DRIdrawablePrivate *dPriv); +#endif + +#endif /* INTEL_SWAPBUFFERS_H */ diff --git a/src/gallium/winsys/egl_drm/intel/intel_winsys.h b/src/gallium/winsys/egl_drm/intel/intel_winsys.h new file mode 100644 index 0000000..d0a319f --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/intel_winsys.h @@ -0,0 +1,73 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef INTEL_WINSYS_H +#define INTEL_WINSYS_H + +#include "pipe/p_state.h" + +struct intel_context; +struct pipe_context; +struct pipe_winsys; +struct pipe_buffer; +struct _DriBufferObject; + +struct pipe_winsys * +intel_create_pipe_winsys( int fd, struct _DriFreeSlabManager *fMan ); + +void +intel_destroy_pipe_winsys( struct pipe_winsys *winsys ); + +struct pipe_context * +intel_create_softpipe( struct intel_context *intel, + struct pipe_winsys *winsys ); + +struct pipe_context * +intel_create_i915simple( struct intel_context *intel, + struct pipe_winsys *winsys ); + + +struct intel_buffer { + struct pipe_buffer base; + struct _DriBufferObject *driBO; +}; + +static INLINE struct intel_buffer * +intel_buffer( struct pipe_buffer *buf ) +{ + return (struct intel_buffer *)buf; +} + +static INLINE struct _DriBufferObject * +dri_bo( struct pipe_buffer *buf ) +{ + return intel_buffer(buf)->driBO; +} + + + +#endif diff --git a/src/gallium/winsys/egl_drm/intel/intel_winsys_i915.c b/src/gallium/winsys/egl_drm/intel/intel_winsys_i915.c new file mode 100644 index 0000000..8ec5c7e --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/intel_winsys_i915.c @@ -0,0 +1,184 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * + **************************************************************************/ +/* + * Authors: Keith Whitwell <keithw-at-tungstengraphics-dot-com> + */ + +#include <stdlib.h> +#include <xf86drm.h> +#include "ws_dri_bufpool.h" +#include "ws_dri_bufmgr.h" + +#include "intel_context.h" +#include "intel_batchbuffer.h" +#include "intel_winsys.h" + +#include "pipe/p_util.h" +#include "pipe/p_winsys.h" +#include "i915simple/i915_winsys.h" +#include "i915simple/i915_screen.h" + + +struct intel_i915_winsys { + struct i915_winsys winsys; /**< batch buffer funcs */ + struct pipe_winsys *pws; + struct intel_context *intel; +}; + + +/* Turn a i915simple winsys into an intel/i915simple winsys: + */ +static inline struct intel_i915_winsys * +intel_i915_winsys( struct i915_winsys *sws ) +{ + return (struct intel_i915_winsys *)sws; +} + + +/* Simple batchbuffer interface: + */ + +static unsigned *intel_i915_batch_start( struct i915_winsys *sws, + unsigned dwords, + unsigned relocs ) +{ + struct intel_context *intel = intel_i915_winsys(sws)->intel; + + /* XXX: check relocs. + */ + if (intel_batchbuffer_space( intel->batch ) >= dwords * 4) { + /* XXX: Hmm, the driver can't really do much with this pointer: + */ + return (unsigned *)intel->batch->ptr; + } + else + return NULL; +} + +static void intel_i915_batch_dword( struct i915_winsys *sws, + unsigned dword ) +{ + struct intel_context *intel = intel_i915_winsys(sws)->intel; + intel_batchbuffer_emit_dword( intel->batch, dword ); +} + +static void intel_i915_batch_reloc( struct i915_winsys *sws, + struct pipe_buffer *buf, + unsigned access_flags, + unsigned delta ) +{ + struct intel_context *intel = intel_i915_winsys(sws)->intel; + unsigned flags = DRM_BO_FLAG_MEM_TT; + unsigned mask = DRM_BO_MASK_MEM; + + if (access_flags & I915_BUFFER_ACCESS_WRITE) { + flags |= DRM_BO_FLAG_WRITE; + mask |= DRM_BO_FLAG_WRITE; + } + + if (access_flags & I915_BUFFER_ACCESS_READ) { + flags |= DRM_BO_FLAG_READ; + mask |= DRM_BO_FLAG_READ; + } + +#if 0 /* JB old */ + intel_batchbuffer_emit_reloc( intel->batch, + dri_bo( buf ), + flags, mask, + delta ); +#else /* new */ + intel_offset_relocation( intel->batch, + delta, + dri_bo( buf ), + flags, + mask ); +#endif +} + + + +static void intel_i915_batch_flush( struct i915_winsys *sws, + struct pipe_fence_handle **fence ) +{ + struct intel_i915_winsys *iws = intel_i915_winsys(sws); + struct intel_context *intel = iws->intel; + union { + struct _DriFenceObject *dri; + struct pipe_fence_handle *pipe; + } fu; + + if (fence) + assert(!*fence); + + fu.dri = intel_batchbuffer_flush( intel->batch ); + + if (!fu.dri) { + assert(0); + *fence = NULL; + return; + } + + if (fu.dri) { + if (fence) + *fence = fu.pipe; + else + driFenceUnReference(&fu.dri); + } + +} + + +/** + * Create i915 hardware rendering context. + */ +struct pipe_context * +intel_create_i915simple( struct intel_context *intel, + struct pipe_winsys *winsys ) +{ + struct intel_i915_winsys *iws = CALLOC_STRUCT( intel_i915_winsys ); + struct pipe_screen *screen; + + /* Fill in this struct with callbacks that i915simple will need to + * communicate with the window system, buffer manager, etc. + */ + iws->winsys.batch_start = intel_i915_batch_start; + iws->winsys.batch_dword = intel_i915_batch_dword; + iws->winsys.batch_reloc = intel_i915_batch_reloc; + iws->winsys.batch_flush = intel_i915_batch_flush; + iws->pws = winsys; + iws->intel = intel; + + screen = i915_create_screen(winsys, PCI_CHIP_I945_GM); + assert(screen); + + /* Create the i915simple context: + */ + return i915_create_context( screen, + winsys, + &iws->winsys ); +} diff --git a/src/gallium/winsys/egl_drm/intel/intel_winsys_pipe.c b/src/gallium/winsys/egl_drm/intel/intel_winsys_pipe.c new file mode 100644 index 0000000..8bf8c21 --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/intel_winsys_pipe.c @@ -0,0 +1,338 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * + **************************************************************************/ +/* + * Authors: Keith Whitwell <keithw-at-tungstengraphics-dot-com> + */ + +#include <stdlib.h> +#include <xf86drm.h> +//#include "dri_bufpool.h" +//#include "dri_bufmgr.h" + +#include "intel_context.h" +#include "intel_winsys.h" +#include "intel_swapbuffers.h" +#include "intel_batchbuffer.h" + +#include "pipe/p_winsys.h" +#include "pipe/p_defines.h" +#include "pipe/p_state.h" +#include "pipe/p_util.h" +#include "pipe/p_inlines.h" + + + +struct intel_pipe_winsys { + struct pipe_winsys winsys; + struct _DriBufferPool *regionPool; + struct _DriFreeSlabManager *fMan; +}; + + + +/* Turn a pipe winsys into an intel/pipe winsys: + */ +static inline struct intel_pipe_winsys * +intel_pipe_winsys( struct pipe_winsys *winsys ) +{ + return (struct intel_pipe_winsys *)winsys; +} + + +/* Most callbacks map direcly onto dri_bufmgr operations: + */ +static void *intel_buffer_map(struct pipe_winsys *winsys, + struct pipe_buffer *buf, + unsigned flags ) +{ + unsigned drm_flags = 0; + + if (flags & PIPE_BUFFER_USAGE_CPU_WRITE) + drm_flags |= DRM_BO_FLAG_WRITE; + + if (flags & PIPE_BUFFER_USAGE_CPU_READ) + drm_flags |= DRM_BO_FLAG_READ; + + return driBOMap( dri_bo(buf), drm_flags, 0 ); +} + +static void intel_buffer_unmap(struct pipe_winsys *winsys, + struct pipe_buffer *buf) +{ + driBOUnmap( dri_bo(buf) ); +} + + +static void +intel_buffer_destroy(struct pipe_winsys *winsys, + struct pipe_buffer *buf) +{ + driBOUnReference( dri_bo(buf) ); + FREE(buf); +} + + +/* Pipe has no concept of pools. We choose the tex/region pool + * for all buffers. + * Grabs the hardware lock! + */ +static struct pipe_buffer * +intel_buffer_create(struct pipe_winsys *winsys, + unsigned alignment, + unsigned usage, + unsigned size ) +{ + struct intel_buffer *buffer = CALLOC_STRUCT( intel_buffer ); + struct intel_pipe_winsys *iws = intel_pipe_winsys(winsys); + unsigned flags = 0; + + buffer->base.refcount = 1; + buffer->base.alignment = alignment; + buffer->base.usage = usage; + buffer->base.size = size; + + if (usage & (PIPE_BUFFER_USAGE_VERTEX /*| IWS_BUFFER_USAGE_LOCAL*/)) { + flags |= DRM_BO_FLAG_MEM_LOCAL | DRM_BO_FLAG_CACHED; + } else { + flags |= DRM_BO_FLAG_MEM_VRAM | DRM_BO_FLAG_MEM_TT; + } + + if (usage & PIPE_BUFFER_USAGE_GPU_READ) + flags |= DRM_BO_FLAG_READ; + + if (usage & PIPE_BUFFER_USAGE_GPU_WRITE) + flags |= DRM_BO_FLAG_WRITE; + + /* drm complains if we don't set any read/write flags. + */ + if ((flags & (DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE)) == 0) + flags |= DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE; + +#if 0 + if (flags & IWS_BUFFER_USAGE_EXE) + flags |= DRM_BO_FLAG_EXE; + + if (usage & IWS_BUFFER_USAGE_CACHED) + flags |= DRM_BO_FLAG_CACHED; +#endif + + driGenBuffers( iws->regionPool, + "pipe buffer", 1, &buffer->driBO, alignment, flags, 0 ); + + driBOData( buffer->driBO, size, NULL, iws->regionPool, 0 ); + + return &buffer->base; +} + + +static struct pipe_buffer * +intel_user_buffer_create(struct pipe_winsys *winsys, void *ptr, unsigned bytes) +{ + struct intel_buffer *buffer = CALLOC_STRUCT( intel_buffer ); + struct intel_pipe_winsys *iws = intel_pipe_winsys(winsys); + + driGenUserBuffer( iws->regionPool, + "pipe user buffer", &buffer->driBO, ptr, bytes ); + + buffer->base.refcount = 1; + + return &buffer->base; +} + + +/* The state tracker (should!) keep track of whether the fake + * frontbuffer has been touched by any rendering since the last time + * we copied its contents to the real frontbuffer. Our task is easy: + */ +static void +intel_flush_frontbuffer( struct pipe_winsys *winsys, + struct pipe_surface *surf, + void *context_private) +{ +#if 0 + struct intel_context *intel = (struct intel_context *) context_private; + __DRIdrawablePrivate *dPriv = intel->driDrawable; + + intelDisplaySurface(dPriv, surf, NULL); +#endif +} + + +static struct pipe_surface * +intel_i915_surface_alloc(struct pipe_winsys *winsys) +{ + struct pipe_surface *surf = CALLOC_STRUCT(pipe_surface); + if (surf) { + surf->refcount = 1; + surf->winsys = winsys; + } + return surf; +} + + +/** + * Round n up to next multiple. + */ +static INLINE unsigned +round_up(unsigned n, unsigned multiple) +{ + return (n + multiple - 1) & ~(multiple - 1); +} + +/** + * Copied from xm_winsys.c + */ +static int +intel_i915_surface_alloc_storage(struct pipe_winsys *winsys, + struct pipe_surface *surf, + unsigned width, unsigned height, + enum pipe_format format, + unsigned flags) +{ + const unsigned alignment = 64; + //int ret; + + surf->width = width; + surf->height = height; + surf->format = format; + surf->cpp = pf_get_size(format); + surf->pitch = round_up(width, alignment / surf->cpp); + + assert(!surf->buffer); + surf->buffer = winsys->buffer_create(winsys, alignment, + PIPE_BUFFER_USAGE_PIXEL, + surf->pitch * surf->cpp * height); + if(!surf->buffer) + return -1; + + return 0; +} + + +static void +intel_i915_surface_release(struct pipe_winsys *winsys, struct pipe_surface **s) +{ + struct pipe_surface *surf = *s; + surf->refcount--; + if (surf->refcount == 0) { + if (surf->buffer) + pipe_buffer_reference(winsys, &surf->buffer, NULL); + free(surf); + } + *s = NULL; +} + + + +static const char * +intel_get_name( struct pipe_winsys *winsys ) +{ + return "Intel/EGL/ttm"; +} + +static void +intel_fence_reference( struct pipe_winsys *sws, + struct pipe_fence_handle **ptr, + struct pipe_fence_handle *fence ) +{ + if (*ptr) + driFenceUnReference((struct _DriFenceObject **)ptr); + + if (fence) + *ptr = (struct pipe_fence_handle *)driFenceReference((struct _DriFenceObject *)fence); +} + +static int +intel_fence_signalled( struct pipe_winsys *sws, + struct pipe_fence_handle *fence, + unsigned flag ) +{ + return driFenceSignaled((struct _DriFenceObject *)fence, flag); +} + +static int +intel_fence_finish( struct pipe_winsys *sws, + struct pipe_fence_handle *fence, + unsigned flag ) +{ + /* JB: Lets allways lazy wait */ + return driFenceFinish((struct _DriFenceObject *)fence, flag, 1); +} + +struct pipe_winsys * +intel_create_pipe_winsys( int fd, struct _DriFreeSlabManager *fMan ) +{ + struct intel_pipe_winsys *iws = CALLOC_STRUCT( intel_pipe_winsys ); + + /* Fill in this struct with callbacks that pipe will need to + * communicate with the window system, buffer manager, etc. + * + * Pipe would be happy with a malloc based memory manager, but + * the SwapBuffers implementation in this winsys driver requires + * that rendering be done to an appropriate _DriBufferObject. + */ + iws->winsys.buffer_create = intel_buffer_create; + iws->winsys.user_buffer_create = intel_user_buffer_create; + iws->winsys.buffer_map = intel_buffer_map; + iws->winsys.buffer_unmap = intel_buffer_unmap; + iws->winsys.buffer_destroy = intel_buffer_destroy; + iws->winsys.flush_frontbuffer = intel_flush_frontbuffer; + iws->winsys.get_name = intel_get_name; + iws->winsys.surface_alloc = intel_i915_surface_alloc; + iws->winsys.surface_alloc_storage = intel_i915_surface_alloc_storage; + iws->winsys.surface_release = intel_i915_surface_release; + + iws->winsys.fence_reference = intel_fence_reference; + iws->winsys.fence_signalled = intel_fence_signalled; + iws->winsys.fence_finish = intel_fence_finish; + + if (fd) + iws->regionPool = driSlabPoolInit(fd, + DRM_BO_FLAG_READ | + DRM_BO_FLAG_WRITE | + DRM_BO_FLAG_MEM_TT, + DRM_BO_FLAG_READ | + DRM_BO_FLAG_WRITE | + DRM_BO_FLAG_MEM_TT, + 64, 6, 16, 4096, 0, + fMan); + + return &iws->winsys; +} + + +void +intel_destroy_pipe_winsys( struct pipe_winsys *winsys ) +{ + struct intel_pipe_winsys *iws = intel_pipe_winsys(winsys); + if (iws->regionPool) { + driPoolTakeDown(iws->regionPool); + } + free(iws); +} + diff --git a/src/gallium/winsys/egl_drm/intel/intel_winsys_softpipe.c b/src/gallium/winsys/egl_drm/intel/intel_winsys_softpipe.c new file mode 100644 index 0000000..0bc2dc4 --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/intel_winsys_softpipe.c @@ -0,0 +1,82 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * + **************************************************************************/ +/* + * Authors: Keith Whitwell <keithw-at-tungstengraphics-dot-com> + */ + +#include "intel_context.h" +#include "intel_winsys.h" +#include "pipe/p_defines.h" +#include "pipe/p_util.h" +#include "pipe/p_format.h" +#include "softpipe/sp_winsys.h" + + +struct intel_softpipe_winsys { + struct softpipe_winsys sws; + struct intel_context *intel; +}; + +/** + * Return list of surface formats supported by this driver. + */ +static boolean +intel_is_format_supported(struct softpipe_winsys *sws, + enum pipe_format format) +{ + switch(format) { + case PIPE_FORMAT_A8R8G8B8_UNORM: + case PIPE_FORMAT_R5G6B5_UNORM: + case PIPE_FORMAT_S8Z24_UNORM: + return TRUE; + default: + return FALSE; + } +} + + +/** + * Create rendering context which uses software rendering. + */ +struct pipe_context * +intel_create_softpipe( struct intel_context *intel, + struct pipe_winsys *winsys ) +{ + struct intel_softpipe_winsys *isws = CALLOC_STRUCT( intel_softpipe_winsys ); + struct pipe_screen *screen = softpipe_create_screen(winsys); + + /* Fill in this struct with callbacks that softpipe will need to + * communicate with the window system, buffer manager, etc. + */ + isws->sws.is_format_supported = intel_is_format_supported; + isws->intel = intel; + + /* Create the softpipe context: + */ + return softpipe_create( screen, winsys, &isws->sws ); +} diff --git a/src/gallium/winsys/egl_drm/intel/ws_dri_bufmgr.c b/src/gallium/winsys/egl_drm/intel/ws_dri_bufmgr.c new file mode 100644 index 0000000..1bc1089 --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/ws_dri_bufmgr.c @@ -0,0 +1,953 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * + **************************************************************************/ +/* + * Authors: Thomas Hellström <thomas-at-tungstengraphics-dot-com> + * Keith Whitwell <keithw-at-tungstengraphics-dot-com> + */ + +#include <xf86drm.h> +#include <stdlib.h> +#include "glthread.h" +#include "errno.h" +#include "ws_dri_bufmgr.h" +#include "string.h" +#include "imports.h" +#include "ws_dri_bufpool.h" +#include "ws_dri_fencemgr.h" + +/* + * This lock is here to protect drmBO structs changing underneath us during a + * validate list call, since validatelist cannot take individiual locks for + * each drmBO. Validatelist takes this lock in write mode. Any access to an + * individual drmBO should take this lock in read mode, since in that case, the + * driBufferObject mutex will protect the access. Locking order is + * driBufferObject mutex - > this rw lock. + */ + +_glthread_DECLARE_STATIC_MUTEX(bmMutex); +_glthread_DECLARE_STATIC_COND(bmCond); + +static int kernelReaders = 0; +static int num_buffers = 0; +static int num_user_buffers = 0; + +static drmBO *drmBOListBuf(void *iterator) +{ + drmBONode *node; + drmMMListHead *l = (drmMMListHead *) iterator; + node = DRMLISTENTRY(drmBONode, l, head); + return node->buf; +} + +static void *drmBOListIterator(drmBOList *list) +{ + void *ret = list->list.next; + + if (ret == &list->list) + return NULL; + return ret; +} + +static void *drmBOListNext(drmBOList *list, void *iterator) +{ + void *ret; + + drmMMListHead *l = (drmMMListHead *) iterator; + ret = l->next; + if (ret == &list->list) + return NULL; + return ret; +} + +static drmBONode *drmAddListItem(drmBOList *list, drmBO *item, + uint64_t arg0, + uint64_t arg1) +{ + drmBONode *node; + drmMMListHead *l; + + l = list->free.next; + if (l == &list->free) { + node = (drmBONode *) malloc(sizeof(*node)); + if (!node) { + return NULL; + } + list->numCurrent++; + } + else { + DRMLISTDEL(l); + node = DRMLISTENTRY(drmBONode, l, head); + } + node->buf = item; + node->arg0 = arg0; + node->arg1 = arg1; + DRMLISTADD(&node->head, &list->list); + list->numOnList++; + return node; +} + +static int drmAddValidateItem(drmBOList *list, drmBO *buf, uint64_t flags, + uint64_t mask, int *newItem) +{ + drmBONode *node, *cur; + drmMMListHead *l; + + *newItem = 0; + cur = NULL; + + for (l = list->list.next; l != &list->list; l = l->next) { + node = DRMLISTENTRY(drmBONode, l, head); + if (node->buf == buf) { + cur = node; + break; + } + } + if (!cur) { + cur = drmAddListItem(list, buf, flags, mask); + if (!cur) { + return -ENOMEM; + } + *newItem = 1; + cur->arg0 = flags; + cur->arg1 = mask; + } + else { + uint64_t memFlags = cur->arg0 & flags & DRM_BO_MASK_MEM; + uint64_t accFlags = (cur->arg0 | flags) & ~DRM_BO_MASK_MEM; + + if (mask & cur->arg1 & ~DRM_BO_MASK_MEM & (cur->arg0 ^ flags)) { + return -EINVAL; + } + + cur->arg1 |= mask; + cur->arg0 = (cur->arg0 & ~mask) | ((memFlags | accFlags) & mask); + + if (((cur->arg1 & DRM_BO_MASK_MEM) != 0) && + (cur->arg0 & DRM_BO_MASK_MEM) == 0) { + return -EINVAL; + } + } + return 0; +} + +static void drmBOFreeList(drmBOList *list) +{ + drmBONode *node; + drmMMListHead *l; + + l = list->list.next; + while(l != &list->list) { + DRMLISTDEL(l); + node = DRMLISTENTRY(drmBONode, l, head); + free(node); + l = list->list.next; + list->numCurrent--; + list->numOnList--; + } + + l = list->free.next; + while(l != &list->free) { + DRMLISTDEL(l); + node = DRMLISTENTRY(drmBONode, l, head); + free(node); + l = list->free.next; + list->numCurrent--; + } +} + +static int drmAdjustListNodes(drmBOList *list) +{ + drmBONode *node; + drmMMListHead *l; + int ret = 0; + + while(list->numCurrent < list->numTarget) { + node = (drmBONode *) malloc(sizeof(*node)); + if (!node) { + ret = -ENOMEM; + break; + } + list->numCurrent++; + DRMLISTADD(&node->head, &list->free); + } + + while(list->numCurrent > list->numTarget) { + l = list->free.next; + if (l == &list->free) + break; + DRMLISTDEL(l); + node = DRMLISTENTRY(drmBONode, l, head); + free(node); + list->numCurrent--; + } + return ret; +} + +static int drmBOCreateList(int numTarget, drmBOList *list) +{ + DRMINITLISTHEAD(&list->list); + DRMINITLISTHEAD(&list->free); + list->numTarget = numTarget; + list->numCurrent = 0; + list->numOnList = 0; + return drmAdjustListNodes(list); +} + +static int drmBOResetList(drmBOList *list) +{ + drmMMListHead *l; + int ret; + + ret = drmAdjustListNodes(list); + if (ret) + return ret; + + l = list->list.next; + while (l != &list->list) { + DRMLISTDEL(l); + DRMLISTADD(l, &list->free); + list->numOnList--; + l = list->list.next; + } + return drmAdjustListNodes(list); +} + +void driWriteLockKernelBO(void) +{ + _glthread_LOCK_MUTEX(bmMutex); + while(kernelReaders != 0) + _glthread_COND_WAIT(bmCond, bmMutex); +} + +void driWriteUnlockKernelBO(void) +{ + _glthread_UNLOCK_MUTEX(bmMutex); +} + +void driReadLockKernelBO(void) +{ + _glthread_LOCK_MUTEX(bmMutex); + kernelReaders++; + _glthread_UNLOCK_MUTEX(bmMutex); +} + +void driReadUnlockKernelBO(void) +{ + _glthread_LOCK_MUTEX(bmMutex); + if (--kernelReaders == 0) + _glthread_COND_BROADCAST(bmCond); + _glthread_UNLOCK_MUTEX(bmMutex); +} + + + + +/* + * TODO: Introduce fence pools in the same way as + * buffer object pools. + */ + +typedef struct _DriBufferObject +{ + DriBufferPool *pool; + _glthread_Mutex mutex; + int refCount; + const char *name; + uint64_t flags; + unsigned hint; + unsigned alignment; + unsigned createdByReference; + void *private; + /* user-space buffer: */ + unsigned userBuffer; + void *userData; + unsigned userSize; +} DriBufferObject; + +typedef struct _DriBufferList { + drmBOList drmBuffers; /* List of kernel buffers needing validation */ + drmBOList driBuffers; /* List of user-space buffers needing validation */ +} DriBufferList; + + +void +bmError(int val, const char *file, const char *function, int line) +{ + _mesa_printf("Fatal video memory manager error \"%s\".\n" + "Check kernel logs or set the LIBGL_DEBUG\n" + "environment variable to \"verbose\" for more info.\n" + "Detected in file %s, line %d, function %s.\n", + strerror(-val), file, line, function); +#ifndef NDEBUG + abort(); +#else + abort(); +#endif +} + +extern drmBO * +driBOKernel(struct _DriBufferObject *buf) +{ + drmBO *ret; + + driReadLockKernelBO(); + _glthread_LOCK_MUTEX(buf->mutex); + assert(buf->private != NULL); + ret = buf->pool->kernel(buf->pool, buf->private); + if (!ret) + BM_CKFATAL(-EINVAL); + _glthread_UNLOCK_MUTEX(buf->mutex); + driReadUnlockKernelBO(); + + return ret; +} + +void +driBOWaitIdle(struct _DriBufferObject *buf, int lazy) +{ + + /* + * This function may block. Is it sane to keep the mutex held during + * that time?? + */ + + _glthread_LOCK_MUTEX(buf->mutex); + BM_CKFATAL(buf->pool->waitIdle(buf->pool, buf->private, &buf->mutex, lazy)); + _glthread_UNLOCK_MUTEX(buf->mutex); +} + +void * +driBOMap(struct _DriBufferObject *buf, unsigned flags, unsigned hint) +{ + void *virtual; + int retval; + + if (buf->userBuffer) { + return buf->userData; + } + + _glthread_LOCK_MUTEX(buf->mutex); + assert(buf->private != NULL); + retval = buf->pool->map(buf->pool, buf->private, flags, hint, + &buf->mutex, &virtual); + _glthread_UNLOCK_MUTEX(buf->mutex); + + return retval == 0 ? virtual : NULL; +} + +void +driBOUnmap(struct _DriBufferObject *buf) +{ + if (buf->userBuffer) + return; + + assert(buf->private != NULL); + _glthread_LOCK_MUTEX(buf->mutex); + BM_CKFATAL(buf->pool->unmap(buf->pool, buf->private)); + _glthread_UNLOCK_MUTEX(buf->mutex); +} + +unsigned long +driBOOffset(struct _DriBufferObject *buf) +{ + unsigned long ret; + + assert(buf->private != NULL); + + _glthread_LOCK_MUTEX(buf->mutex); + ret = buf->pool->offset(buf->pool, buf->private); + _glthread_UNLOCK_MUTEX(buf->mutex); + return ret; +} + +unsigned long +driBOPoolOffset(struct _DriBufferObject *buf) +{ + unsigned long ret; + + assert(buf->private != NULL); + + _glthread_LOCK_MUTEX(buf->mutex); + ret = buf->pool->poolOffset(buf->pool, buf->private); + _glthread_UNLOCK_MUTEX(buf->mutex); + return ret; +} + +uint64_t +driBOFlags(struct _DriBufferObject *buf) +{ + uint64_t ret; + + assert(buf->private != NULL); + + driReadLockKernelBO(); + _glthread_LOCK_MUTEX(buf->mutex); + ret = buf->pool->flags(buf->pool, buf->private); + _glthread_UNLOCK_MUTEX(buf->mutex); + driReadUnlockKernelBO(); + return ret; +} + +struct _DriBufferObject * +driBOReference(struct _DriBufferObject *buf) +{ + _glthread_LOCK_MUTEX(buf->mutex); + if (++buf->refCount == 1) { + _glthread_UNLOCK_MUTEX(buf->mutex); + BM_CKFATAL(-EINVAL); + } + _glthread_UNLOCK_MUTEX(buf->mutex); + return buf; +} + +void +driBOUnReference(struct _DriBufferObject *buf) +{ + int tmp; + + if (!buf) + return; + + _glthread_LOCK_MUTEX(buf->mutex); + tmp = --buf->refCount; + if (!tmp) { + _glthread_UNLOCK_MUTEX(buf->mutex); + if (buf->private) { + if (buf->createdByReference) + buf->pool->unreference(buf->pool, buf->private); + else + buf->pool->destroy(buf->pool, buf->private); + } + if (buf->userBuffer) + num_user_buffers--; + else + num_buffers--; + free(buf); + } else + _glthread_UNLOCK_MUTEX(buf->mutex); + +} + + +int +driBOData(struct _DriBufferObject *buf, + unsigned size, const void *data, + DriBufferPool *newPool, + uint64_t flags) +{ + void *virtual = NULL; + int newBuffer; + int retval = 0; + struct _DriBufferPool *pool; + + assert(!buf->userBuffer); /* XXX just do a memcpy? */ + + _glthread_LOCK_MUTEX(buf->mutex); + pool = buf->pool; + + if (pool == NULL && newPool != NULL) { + buf->pool = newPool; + pool = newPool; + } + if (newPool == NULL) + newPool = pool; + + if (!pool->create) { + _mesa_error(NULL, GL_INVALID_OPERATION, + "driBOData called on invalid buffer\n"); + BM_CKFATAL(-EINVAL); + } + + newBuffer = (!buf->private || pool != newPool || + pool->size(pool, buf->private) < size); + + if (!flags) + flags = buf->flags; + + if (newBuffer) { + + if (buf->createdByReference) { + _mesa_error(NULL, GL_INVALID_OPERATION, + "driBOData requiring resizing called on " + "shared buffer.\n"); + BM_CKFATAL(-EINVAL); + } + + if (buf->private) + buf->pool->destroy(buf->pool, buf->private); + + pool = newPool; + buf->pool = newPool; + buf->private = pool->create(pool, size, flags, DRM_BO_HINT_DONT_FENCE, + buf->alignment); + if (!buf->private) + retval = -ENOMEM; + + if (retval == 0) + retval = pool->map(pool, buf->private, + DRM_BO_FLAG_WRITE, + DRM_BO_HINT_DONT_BLOCK, &buf->mutex, &virtual); + } else if (pool->map(pool, buf->private, DRM_BO_FLAG_WRITE, + DRM_BO_HINT_DONT_BLOCK, &buf->mutex, &virtual)) { + /* + * Buffer is busy. need to create a new one. + */ + + void *newBuf; + + newBuf = pool->create(pool, size, flags, DRM_BO_HINT_DONT_FENCE, + buf->alignment); + if (newBuf) { + buf->pool->destroy(buf->pool, buf->private); + buf->private = newBuf; + } + + retval = pool->map(pool, buf->private, + DRM_BO_FLAG_WRITE, 0, &buf->mutex, &virtual); + } else { + uint64_t flag_diff = flags ^ buf->flags; + + /* + * We might need to change buffer flags. + */ + + if (flag_diff){ + assert(pool->setStatus != NULL); + BM_CKFATAL(pool->unmap(pool, buf->private)); + BM_CKFATAL(pool->setStatus(pool, buf->private, flag_diff, + buf->flags)); + if (!data) + goto out; + + retval = pool->map(pool, buf->private, + DRM_BO_FLAG_WRITE, 0, &buf->mutex, &virtual); + } + } + + if (retval == 0) { + if (data) + memcpy(virtual, data, size); + + BM_CKFATAL(pool->unmap(pool, buf->private)); + } + + out: + _glthread_UNLOCK_MUTEX(buf->mutex); + + return retval; +} + +void +driBOSubData(struct _DriBufferObject *buf, + unsigned long offset, unsigned long size, const void *data) +{ + void *virtual; + + assert(!buf->userBuffer); /* XXX just do a memcpy? */ + + _glthread_LOCK_MUTEX(buf->mutex); + if (size && data) { + BM_CKFATAL(buf->pool->map(buf->pool, buf->private, + DRM_BO_FLAG_WRITE, 0, &buf->mutex, + &virtual)); + memcpy((unsigned char *) virtual + offset, data, size); + BM_CKFATAL(buf->pool->unmap(buf->pool, buf->private)); + } + _glthread_UNLOCK_MUTEX(buf->mutex); +} + +void +driBOGetSubData(struct _DriBufferObject *buf, + unsigned long offset, unsigned long size, void *data) +{ + void *virtual; + + assert(!buf->userBuffer); /* XXX just do a memcpy? */ + + _glthread_LOCK_MUTEX(buf->mutex); + if (size && data) { + BM_CKFATAL(buf->pool->map(buf->pool, buf->private, + DRM_BO_FLAG_READ, 0, &buf->mutex, &virtual)); + memcpy(data, (unsigned char *) virtual + offset, size); + BM_CKFATAL(buf->pool->unmap(buf->pool, buf->private)); + } + _glthread_UNLOCK_MUTEX(buf->mutex); +} + +void +driBOSetReferenced(struct _DriBufferObject *buf, + unsigned long handle) +{ + _glthread_LOCK_MUTEX(buf->mutex); + if (buf->private != NULL) { + _mesa_error(NULL, GL_INVALID_OPERATION, + "Invalid buffer for setReferenced\n"); + BM_CKFATAL(-EINVAL); + + } + if (buf->pool->reference == NULL) { + _mesa_error(NULL, GL_INVALID_OPERATION, + "Invalid buffer pool for setReferenced\n"); + BM_CKFATAL(-EINVAL); + } + buf->private = buf->pool->reference(buf->pool, handle); + if (!buf->private) { + _mesa_error(NULL, GL_OUT_OF_MEMORY, + "Invalid buffer pool for setStatic\n"); + BM_CKFATAL(-ENOMEM); + } + buf->createdByReference = GL_TRUE; + buf->flags = buf->pool->kernel(buf->pool, buf->private)->flags; + _glthread_UNLOCK_MUTEX(buf->mutex); +} + +int +driGenBuffers(struct _DriBufferPool *pool, + const char *name, + unsigned n, + struct _DriBufferObject *buffers[], + unsigned alignment, uint64_t flags, unsigned hint) +{ + struct _DriBufferObject *buf; + int i; + + flags = (flags) ? flags : DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_MEM_VRAM | + DRM_BO_FLAG_MEM_LOCAL | DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE; + + ++num_buffers; + + assert(pool); + + for (i = 0; i < n; ++i) { + buf = (struct _DriBufferObject *) calloc(1, sizeof(*buf)); + if (!buf) + return -ENOMEM; + + _glthread_INIT_MUTEX(buf->mutex); + _glthread_LOCK_MUTEX(buf->mutex); + buf->refCount = 1; + buf->flags = flags; + buf->hint = hint; + buf->name = name; + buf->alignment = alignment; + buf->pool = pool; + buf->createdByReference = 0; + _glthread_UNLOCK_MUTEX(buf->mutex); + buffers[i] = buf; + } + return 0; +} + +void +driGenUserBuffer(struct _DriBufferPool *pool, + const char *name, + struct _DriBufferObject **buffers, + void *ptr, unsigned bytes) +{ + const unsigned alignment = 1, flags = 0, hint = 0; + + --num_buffers; /* JB: is inced in GenBuffes */ + driGenBuffers(pool, name, 1, buffers, alignment, flags, hint); + ++num_user_buffers; + + (*buffers)->userBuffer = 1; + (*buffers)->userData = ptr; + (*buffers)->userSize = bytes; +} + +void +driDeleteBuffers(unsigned n, struct _DriBufferObject *buffers[]) +{ + int i; + + for (i = 0; i < n; ++i) { + driBOUnReference(buffers[i]); + } +} + + +void +driInitBufMgr(int fd) +{ + ; +} + +/* + * Note that lists are per-context and don't need mutex protection. + */ + +struct _DriBufferList * +driBOCreateList(int target) +{ + struct _DriBufferList *list = calloc(sizeof(*list), 1); + + BM_CKFATAL(drmBOCreateList(target, &list->drmBuffers)); + BM_CKFATAL(drmBOCreateList(target, &list->driBuffers)); + return list; +} + +int +driBOResetList(struct _DriBufferList * list) +{ + int ret; + ret = drmBOResetList(&list->drmBuffers); + if (ret) + return ret; + ret = drmBOResetList(&list->driBuffers); + return ret; +} + +void +driBOFreeList(struct _DriBufferList * list) +{ + drmBOFreeList(&list->drmBuffers); + drmBOFreeList(&list->driBuffers); + free(list); +} + + +/* + * Copied from libdrm, because it is needed by driAddValidateItem. + */ + +static drmBONode * +driAddListItem(drmBOList * list, drmBO * item, + uint64_t arg0, uint64_t arg1) +{ + drmBONode *node; + drmMMListHead *l; + + l = list->free.next; + if (l == &list->free) { + node = (drmBONode *) malloc(sizeof(*node)); + if (!node) { + return NULL; + } + list->numCurrent++; + } else { + DRMLISTDEL(l); + node = DRMLISTENTRY(drmBONode, l, head); + } + memset(&node->bo_arg, 0, sizeof(node->bo_arg)); + node->buf = item; + node->arg0 = arg0; + node->arg1 = arg1; + DRMLISTADDTAIL(&node->head, &list->list); + list->numOnList++; + return node; +} + +/* + * Slightly modified version compared to the libdrm version. + * This one returns the list index of the buffer put on the list. + */ + +static int +driAddValidateItem(drmBOList * list, drmBO * buf, uint64_t flags, + uint64_t mask, int *itemLoc, + struct _drmBONode **pnode) +{ + drmBONode *node, *cur; + drmMMListHead *l; + int count = 0; + + cur = NULL; + + for (l = list->list.next; l != &list->list; l = l->next) { + node = DRMLISTENTRY(drmBONode, l, head); + if (node->buf == buf) { + cur = node; + break; + } + count++; + } + if (!cur) { + cur = driAddListItem(list, buf, flags, mask); + if (!cur) + return -ENOMEM; + + cur->arg0 = flags; + cur->arg1 = mask; + } else { + uint64_t memFlags = cur->arg0 & flags & DRM_BO_MASK_MEM; + uint64_t accFlags = (cur->arg0 | flags) & ~DRM_BO_MASK_MEM; + + if (mask & cur->arg1 & ~DRM_BO_MASK_MEM & (cur->arg0 ^ flags)) { + return -EINVAL; + } + + cur->arg1 |= mask; + cur->arg0 = (cur->arg0 & ~mask) | ((memFlags | accFlags) & mask); + + if (((cur->arg1 & DRM_BO_MASK_MEM) != 0) && + (cur->arg0 & DRM_BO_MASK_MEM) == 0) { + return -EINVAL; + } + } + *itemLoc = count; + *pnode = cur; + return 0; +} + + +void +driBOAddListItem(struct _DriBufferList * list, struct _DriBufferObject *buf, + uint64_t flags, uint64_t mask, int *itemLoc, + struct _drmBONode **node) +{ + int newItem; + + _glthread_LOCK_MUTEX(buf->mutex); + BM_CKFATAL(driAddValidateItem(&list->drmBuffers, + buf->pool->kernel(buf->pool, buf->private), + flags, mask, itemLoc, node)); + BM_CKFATAL(drmAddValidateItem(&list->driBuffers, (drmBO *) buf, + flags, mask, &newItem)); + if (newItem) + buf->refCount++; + + _glthread_UNLOCK_MUTEX(buf->mutex); +} + +drmBOList *driGetdrmBOList(struct _DriBufferList *list) +{ + driWriteLockKernelBO(); + return &list->drmBuffers; +} + +void driPutdrmBOList(struct _DriBufferList *list) +{ + driWriteUnlockKernelBO(); +} + + +void +driBOFence(struct _DriBufferObject *buf, struct _DriFenceObject *fence) +{ + _glthread_LOCK_MUTEX(buf->mutex); + if (buf->pool->fence) + BM_CKFATAL(buf->pool->fence(buf->pool, buf->private, fence)); + _glthread_UNLOCK_MUTEX(buf->mutex); + +} + +void +driBOUnrefUserList(struct _DriBufferList *list) +{ + struct _DriBufferObject *buf; + void *curBuf; + + curBuf = drmBOListIterator(&list->driBuffers); + while (curBuf) { + buf = (struct _DriBufferObject *)drmBOListBuf(curBuf); + driBOUnReference(buf); + curBuf = drmBOListNext(&list->driBuffers, curBuf); + } +} + +struct _DriFenceObject * +driBOFenceUserList(struct _DriFenceMgr *mgr, + struct _DriBufferList *list, const char *name, + drmFence *kFence) +{ + struct _DriFenceObject *fence; + struct _DriBufferObject *buf; + void *curBuf; + + fence = driFenceCreate(mgr, kFence->fence_class, kFence->type, + kFence, sizeof(*kFence)); + curBuf = drmBOListIterator(&list->driBuffers); + + /* + * User-space fencing callbacks. + */ + + while (curBuf) { + buf = (struct _DriBufferObject *) drmBOListBuf(curBuf); + driBOFence(buf, fence); + driBOUnReference(buf); + curBuf = drmBOListNext(&list->driBuffers, curBuf); + } + + driBOResetList(list); + return fence; +} + +void +driBOValidateUserList(struct _DriBufferList * list) +{ + void *curBuf; + struct _DriBufferObject *buf; + + curBuf = drmBOListIterator(&list->driBuffers); + + /* + * User-space validation callbacks. + */ + + while (curBuf) { + buf = (struct _DriBufferObject *) drmBOListBuf(curBuf); + _glthread_LOCK_MUTEX(buf->mutex); + if (buf->pool->validate) + BM_CKFATAL(buf->pool->validate(buf->pool, buf->private, &buf->mutex)); + _glthread_UNLOCK_MUTEX(buf->mutex); + curBuf = drmBOListNext(&list->driBuffers, curBuf); + } +} + + +void +driPoolTakeDown(struct _DriBufferPool *pool) +{ + pool->takeDown(pool); + +} + +unsigned long +driBOSize(struct _DriBufferObject *buf) +{ + unsigned long size; + + _glthread_LOCK_MUTEX(buf->mutex); + size = buf->pool->size(buf->pool, buf->private); + _glthread_UNLOCK_MUTEX(buf->mutex); + + return size; + +} + +drmBOList *driBOGetDRMBuffers(struct _DriBufferList *list) +{ + return &list->drmBuffers; +} + +drmBOList *driBOGetDRIBuffers(struct _DriBufferList *list) +{ + return &list->driBuffers; +} + diff --git a/src/gallium/winsys/egl_drm/intel/ws_dri_bufmgr.h b/src/gallium/winsys/egl_drm/intel/ws_dri_bufmgr.h new file mode 100644 index 0000000..fdaf5ee --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/ws_dri_bufmgr.h @@ -0,0 +1,138 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * + **************************************************************************/ +/* + * Authors: Thomas Hellström <thomas-at-tungstengraphics-dot-com> + * Keith Whitwell <keithw-at-tungstengraphics-dot-com> + */ + +#ifndef _PSB_BUFMGR_H_ +#define _PSB_BUFMGR_H_ +#include <xf86mm.h> +#include "i915_drm.h" +#include "ws_dri_fencemgr.h" + +typedef struct _drmBONode +{ + drmMMListHead head; + drmBO *buf; + struct drm_i915_op_arg bo_arg; + uint64_t arg0; + uint64_t arg1; +} drmBONode; + +typedef struct _drmBOList { + unsigned numTarget; + unsigned numCurrent; + unsigned numOnList; + drmMMListHead list; + drmMMListHead free; +} drmBOList; + + +struct _DriFenceObject; +struct _DriBufferObject; +struct _DriBufferPool; +struct _DriBufferList; + +/* + * Return a pointer to the libdrm buffer object this DriBufferObject + * uses. + */ + +extern drmBO *driBOKernel(struct _DriBufferObject *buf); +extern void *driBOMap(struct _DriBufferObject *buf, unsigned flags, + unsigned hint); +extern void driBOUnmap(struct _DriBufferObject *buf); +extern unsigned long driBOOffset(struct _DriBufferObject *buf); +extern unsigned long driBOPoolOffset(struct _DriBufferObject *buf); + +extern uint64_t driBOFlags(struct _DriBufferObject *buf); +extern struct _DriBufferObject *driBOReference(struct _DriBufferObject *buf); +extern void driBOUnReference(struct _DriBufferObject *buf); + +extern int driBOData(struct _DriBufferObject *r_buf, + unsigned size, const void *data, + struct _DriBufferPool *pool, uint64_t flags); + +extern void driBOSubData(struct _DriBufferObject *buf, + unsigned long offset, unsigned long size, + const void *data); +extern void driBOGetSubData(struct _DriBufferObject *buf, + unsigned long offset, unsigned long size, + void *data); +extern int driGenBuffers(struct _DriBufferPool *pool, + const char *name, + unsigned n, + struct _DriBufferObject *buffers[], + unsigned alignment, uint64_t flags, unsigned hint); +extern void driGenUserBuffer(struct _DriBufferPool *pool, + const char *name, + struct _DriBufferObject *buffers[], + void *ptr, unsigned bytes); +extern void driDeleteBuffers(unsigned n, struct _DriBufferObject *buffers[]); +extern void driInitBufMgr(int fd); +extern struct _DriBufferList *driBOCreateList(int target); +extern int driBOResetList(struct _DriBufferList * list); +extern void driBOAddListItem(struct _DriBufferList * list, + struct _DriBufferObject *buf, + uint64_t flags, uint64_t mask, int *itemLoc, + struct _drmBONode **node); + +extern void driBOValidateList(int fd, struct _DriBufferList * list); +extern void driBOFreeList(struct _DriBufferList * list); +extern struct _DriFenceObject *driBOFenceUserList(struct _DriFenceMgr *mgr, + struct _DriBufferList *list, + const char *name, + drmFence *kFence); +extern void driBOUnrefUserList(struct _DriBufferList *list); +extern void driBOValidateUserList(struct _DriBufferList * list); +extern drmBOList *driGetdrmBOList(struct _DriBufferList *list); +extern void driPutdrmBOList(struct _DriBufferList *list); + +extern void driBOFence(struct _DriBufferObject *buf, + struct _DriFenceObject *fence); + +extern void driPoolTakeDown(struct _DriBufferPool *pool); +extern void driBOSetReferenced(struct _DriBufferObject *buf, + unsigned long handle); +unsigned long driBOSize(struct _DriBufferObject *buf); +extern void driBOWaitIdle(struct _DriBufferObject *buf, int lazy); +extern void driPoolTakeDown(struct _DriBufferPool *pool); + +extern void driReadLockKernelBO(void); +extern void driReadUnlockKernelBO(void); +extern void driWriteLockKernelBO(void); +extern void driWriteUnlockKernelBO(void); + +/* + * For debugging purposes. + */ + +extern drmBOList *driBOGetDRMBuffers(struct _DriBufferList *list); +extern drmBOList *driBOGetDRIBuffers(struct _DriBufferList *list); +#endif diff --git a/src/gallium/winsys/egl_drm/intel/ws_dri_bufpool.h b/src/gallium/winsys/egl_drm/intel/ws_dri_bufpool.h new file mode 100644 index 0000000..3a302e1 --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/ws_dri_bufpool.h @@ -0,0 +1,102 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * + **************************************************************************/ +/* + * Authors: Thomas Hellström <thomas-at-tungstengraphics-dot-com> + */ + +#ifndef _PSB_BUFPOOL_H_ +#define _PSB_BUFPOOL_H_ + +#include <xf86drm.h> +#include <glthread.h> +struct _DriFenceObject; + +typedef struct _DriBufferPool +{ + int fd; + int (*map) (struct _DriBufferPool * pool, void *private, + unsigned flags, int hint, _glthread_Mutex *mutex, + void **virtual); + int (*unmap) (struct _DriBufferPool * pool, void *private); + int (*destroy) (struct _DriBufferPool * pool, void *private); + unsigned long (*offset) (struct _DriBufferPool * pool, void *private); + unsigned long (*poolOffset) (struct _DriBufferPool * pool, void *private); + uint64_t (*flags) (struct _DriBufferPool * pool, void *private); + unsigned long (*size) (struct _DriBufferPool * pool, void *private); + void *(*create) (struct _DriBufferPool * pool, unsigned long size, + uint64_t flags, unsigned hint, unsigned alignment); + void *(*reference) (struct _DriBufferPool * pool, unsigned handle); + int (*unreference) (struct _DriBufferPool * pool, void *private); + int (*fence) (struct _DriBufferPool * pool, void *private, + struct _DriFenceObject * fence); + drmBO *(*kernel) (struct _DriBufferPool * pool, void *private); + int (*validate) (struct _DriBufferPool * pool, void *private, _glthread_Mutex *mutex); + int (*waitIdle) (struct _DriBufferPool *pool, void *private, _glthread_Mutex *mutex, + int lazy); + int (*setStatus) (struct _DriBufferPool *pool, void *private, + uint64_t flag_diff, uint64_t old_flags); + void (*takeDown) (struct _DriBufferPool * pool); + void *data; +} DriBufferPool; + +extern void bmError(int val, const char *file, const char *function, + int line); +#define BM_CKFATAL(val) \ + do{ \ + int tstVal = (val); \ + if (tstVal) \ + bmError(tstVal, __FILE__, __FUNCTION__, __LINE__); \ + } while(0); + + +/* + * Builtin pools. + */ + +/* + * Kernel buffer objects. Size in multiples of page size. Page size aligned. + */ + +extern struct _DriBufferPool *driDRMPoolInit(int fd); +extern struct _DriBufferPool *driMallocPoolInit(void); + +struct _DriFreeSlabManager; +extern struct _DriBufferPool * driSlabPoolInit(int fd, uint64_t flags, + uint64_t validMask, + uint32_t smallestSize, + uint32_t numSizes, + uint32_t desiredNumBuffers, + uint32_t maxSlabSize, + uint32_t pageAlignment, + struct _DriFreeSlabManager *fMan); +extern void driFinishFreeSlabManager(struct _DriFreeSlabManager *fMan); +extern struct _DriFreeSlabManager * +driInitFreeSlabManager(uint32_t checkIntervalMsec, uint32_t slabTimeoutMsec); + + +#endif diff --git a/src/gallium/winsys/egl_drm/intel/ws_dri_drmpool.c b/src/gallium/winsys/egl_drm/intel/ws_dri_drmpool.c new file mode 100644 index 0000000..7c55dbc --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/ws_dri_drmpool.c @@ -0,0 +1,268 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * + **************************************************************************/ +/* + * Authors: Thomas Hellström <thomas-at-tungstengraphics-dot-com> + */ + +#include <xf86drm.h> +#include <stdlib.h> +#include <unistd.h> +#include "ws_dri_bufpool.h" +#include "ws_dri_bufmgr.h" +#include "assert.h" + +/* + * Buffer pool implementation using DRM buffer objects as DRI buffer objects. + */ + +static void * +pool_create(struct _DriBufferPool *pool, + unsigned long size, uint64_t flags, unsigned hint, + unsigned alignment) +{ + drmBO *buf = (drmBO *) malloc(sizeof(*buf)); + int ret; + unsigned pageSize = getpagesize(); + + if (!buf) + return NULL; + + if ((alignment > pageSize) && (alignment % pageSize)) { + free(buf); + return NULL; + } + + ret = drmBOCreate(pool->fd, size, alignment / pageSize, + NULL, + flags, hint, buf); + if (ret) { + free(buf); + return NULL; + } + + return (void *) buf; +} + +static void * +pool_reference(struct _DriBufferPool *pool, unsigned handle) +{ + drmBO *buf = (drmBO *) malloc(sizeof(*buf)); + int ret; + + if (!buf) + return NULL; + + ret = drmBOReference(pool->fd, handle, buf); + + if (ret) { + free(buf); + return NULL; + } + + return (void *) buf; +} + +static int +pool_destroy(struct _DriBufferPool *pool, void *private) +{ + int ret; + drmBO *buf = (drmBO *) private; + driReadLockKernelBO(); + ret = drmBOUnreference(pool->fd, buf); + free(buf); + driReadUnlockKernelBO(); + return ret; +} + +static int +pool_unreference(struct _DriBufferPool *pool, void *private) +{ + int ret; + drmBO *buf = (drmBO *) private; + driReadLockKernelBO(); + ret = drmBOUnreference(pool->fd, buf); + free(buf); + driReadUnlockKernelBO(); + return ret; +} + +static int +pool_map(struct _DriBufferPool *pool, void *private, unsigned flags, + int hint, _glthread_Mutex *mutex, void **virtual) +{ + drmBO *buf = (drmBO *) private; + int ret; + + driReadLockKernelBO(); + ret = drmBOMap(pool->fd, buf, flags, hint, virtual); + driReadUnlockKernelBO(); + return ret; +} + +static int +pool_unmap(struct _DriBufferPool *pool, void *private) +{ + drmBO *buf = (drmBO *) private; + int ret; + + driReadLockKernelBO(); + ret = drmBOUnmap(pool->fd, buf); + driReadUnlockKernelBO(); + + return ret; +} + +static unsigned long +pool_offset(struct _DriBufferPool *pool, void *private) +{ + drmBO *buf = (drmBO *) private; + unsigned long offset; + + driReadLockKernelBO(); + assert(buf->flags & DRM_BO_FLAG_NO_MOVE); + offset = buf->offset; + driReadUnlockKernelBO(); + + return buf->offset; +} + +static unsigned long +pool_poolOffset(struct _DriBufferPool *pool, void *private) +{ + return 0; +} + +static uint64_t +pool_flags(struct _DriBufferPool *pool, void *private) +{ + drmBO *buf = (drmBO *) private; + uint64_t flags; + + driReadLockKernelBO(); + flags = buf->flags; + driReadUnlockKernelBO(); + + return flags; +} + + +static unsigned long +pool_size(struct _DriBufferPool *pool, void *private) +{ + drmBO *buf = (drmBO *) private; + unsigned long size; + + driReadLockKernelBO(); + size = buf->size; + driReadUnlockKernelBO(); + + return buf->size; +} + +static int +pool_fence(struct _DriBufferPool *pool, void *private, + struct _DriFenceObject *fence) +{ + /* + * Noop. The kernel handles all fencing. + */ + + return 0; +} + +static drmBO * +pool_kernel(struct _DriBufferPool *pool, void *private) +{ + return (drmBO *) private; +} + +static int +pool_waitIdle(struct _DriBufferPool *pool, void *private, _glthread_Mutex *mutex, + int lazy) +{ + drmBO *buf = (drmBO *) private; + int ret; + + driReadLockKernelBO(); + ret = drmBOWaitIdle(pool->fd, buf, (lazy) ? DRM_BO_HINT_WAIT_LAZY:0); + driReadUnlockKernelBO(); + + return ret; +} + + +static void +pool_takedown(struct _DriBufferPool *pool) +{ + free(pool); +} + +/*static int +pool_setStatus(struct _DriBufferPool *pool, void *private, + uint64_t flag_diff, uint64_t old_flags) +{ + drmBO *buf = (drmBO *) private; + uint64_t new_flags = old_flags ^ flag_diff; + int ret; + + driReadLockKernelBO(); + ret = drmBOSetStatus(pool->fd, buf, new_flags, flag_diff, + 0, 0, 0); + driReadUnlockKernelBO(); + return ret; +}*/ + +struct _DriBufferPool * +driDRMPoolInit(int fd) +{ + struct _DriBufferPool *pool; + + pool = (struct _DriBufferPool *) malloc(sizeof(*pool)); + + if (!pool) + return NULL; + + pool->fd = fd; + pool->map = &pool_map; + pool->unmap = &pool_unmap; + pool->destroy = &pool_destroy; + pool->offset = &pool_offset; + pool->poolOffset = &pool_poolOffset; + pool->flags = &pool_flags; + pool->size = &pool_size; + pool->create = &pool_create; + pool->fence = &pool_fence; + pool->kernel = &pool_kernel; + pool->validate = NULL; + pool->waitIdle = &pool_waitIdle; + pool->takeDown = &pool_takedown; + pool->reference = &pool_reference; + pool->unreference = &pool_unreference; + pool->data = NULL; + return pool; +} diff --git a/src/gallium/winsys/egl_drm/intel/ws_dri_fencemgr.c b/src/gallium/winsys/egl_drm/intel/ws_dri_fencemgr.c new file mode 100644 index 0000000..1f893b4 --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/ws_dri_fencemgr.c @@ -0,0 +1,377 @@ +#include "ws_dri_fencemgr.h" +#include "glthread.h" +#include <xf86mm.h> +#include <string.h> +#include <unistd.h> + +/* + * Note: Locking order is + * _DriFenceObject::mutex + * _DriFenceMgr::mutex + */ + +struct _DriFenceMgr { + /* + * Constant members. Need no mutex protection. + */ + struct _DriFenceMgrCreateInfo info; + void *private; + + /* + * These members are protected by this->mutex + */ + _glthread_Mutex mutex; + int refCount; + drmMMListHead *heads; + int num_fences; +}; + +struct _DriFenceObject { + + /* + * These members are constant and need no mutex protection. + */ + struct _DriFenceMgr *mgr; + uint32_t fence_class; + uint32_t fence_type; + + /* + * These members are protected by mgr->mutex. + */ + drmMMListHead head; + int refCount; + + /* + * These members are protected by this->mutex. + */ + _glthread_Mutex mutex; + uint32_t signaled_type; + void *private; +}; + +uint32_t +driFenceType(struct _DriFenceObject *fence) +{ + return fence->fence_type; +} + +struct _DriFenceMgr * +driFenceMgrCreate(const struct _DriFenceMgrCreateInfo *info) +{ + struct _DriFenceMgr *tmp; + uint32_t i; + + tmp = calloc(1, sizeof(*tmp)); + if (!tmp) + return NULL; + + _glthread_INIT_MUTEX(tmp->mutex); + _glthread_LOCK_MUTEX(tmp->mutex); + tmp->refCount = 1; + tmp->info = *info; + tmp->num_fences = 0; + tmp->heads = calloc(tmp->info.num_classes, sizeof(*tmp->heads)); + if (!tmp->heads) + goto out_err; + + for (i=0; i<tmp->info.num_classes; ++i) { + DRMINITLISTHEAD(&tmp->heads[i]); + } + _glthread_UNLOCK_MUTEX(tmp->mutex); + return tmp; + + out_err: + if (tmp) + free(tmp); + return NULL; +} + +static void +driFenceMgrUnrefUnlock(struct _DriFenceMgr **pMgr) +{ + struct _DriFenceMgr *mgr = *pMgr; + + *pMgr = NULL; + if (--mgr->refCount == 0) + free(mgr); + else + _glthread_UNLOCK_MUTEX(mgr->mutex); +} + +void +driFenceMgrUnReference(struct _DriFenceMgr **pMgr) +{ + _glthread_LOCK_MUTEX((*pMgr)->mutex); + driFenceMgrUnrefUnlock(pMgr); +} + +static void +driFenceUnReferenceLocked(struct _DriFenceObject **pFence) +{ + struct _DriFenceObject *fence = *pFence; + struct _DriFenceMgr *mgr = fence->mgr; + + *pFence = NULL; + if (--fence->refCount == 0) { + DRMLISTDELINIT(&fence->head); + if (fence->private) + mgr->info.unreference(mgr, &fence->private); + --mgr->num_fences; + fence->mgr = NULL; + --mgr->refCount; + free(fence); + + } +} + + +static void +driSignalPreviousFencesLocked(struct _DriFenceMgr *mgr, + drmMMListHead *list, + uint32_t fence_class, + uint32_t fence_type) +{ + struct _DriFenceObject *entry; + drmMMListHead *prev; + + while(list != &mgr->heads[fence_class]) { + entry = DRMLISTENTRY(struct _DriFenceObject, list, head); + + /* + * Up refcount so that entry doesn't disappear from under us + * when we unlock-relock mgr to get the correct locking order. + */ + + ++entry->refCount; + _glthread_UNLOCK_MUTEX(mgr->mutex); + _glthread_LOCK_MUTEX(entry->mutex); + _glthread_LOCK_MUTEX(mgr->mutex); + + prev = list->prev; + + + + if (list->prev == list) { + + /* + * Somebody else removed the entry from the list. + */ + + _glthread_UNLOCK_MUTEX(entry->mutex); + driFenceUnReferenceLocked(&entry); + return; + } + + entry->signaled_type |= (fence_type & entry->fence_type); + if (entry->signaled_type == entry->fence_type) { + DRMLISTDELINIT(list); + mgr->info.unreference(mgr, &entry->private); + } + _glthread_UNLOCK_MUTEX(entry->mutex); + driFenceUnReferenceLocked(&entry); + list = prev; + } +} + + +int +driFenceFinish(struct _DriFenceObject *fence, uint32_t fence_type, + int lazy_hint) +{ + struct _DriFenceMgr *mgr = fence->mgr; + int ret = 0; + + _glthread_LOCK_MUTEX(fence->mutex); + + if ((fence->signaled_type & fence_type) == fence_type) + goto out0; + + ret = mgr->info.finish(mgr, fence->private, fence_type, lazy_hint); + if (ret) + goto out0; + + _glthread_LOCK_MUTEX(mgr->mutex); + _glthread_UNLOCK_MUTEX(fence->mutex); + + driSignalPreviousFencesLocked(mgr, &fence->head, fence->fence_class, + fence_type); + _glthread_UNLOCK_MUTEX(mgr->mutex); + return 0; + + out0: + _glthread_UNLOCK_MUTEX(fence->mutex); + return ret; +} + +uint32_t driFenceSignaledTypeCached(struct _DriFenceObject *fence) +{ + uint32_t ret; + + _glthread_LOCK_MUTEX(fence->mutex); + ret = fence->signaled_type; + _glthread_UNLOCK_MUTEX(fence->mutex); + + return ret; +} + +int +driFenceSignaledType(struct _DriFenceObject *fence, uint32_t flush_type, + uint32_t *signaled) +{ + int ret = 0; + struct _DriFenceMgr *mgr; + + _glthread_LOCK_MUTEX(fence->mutex); + mgr = fence->mgr; + *signaled = fence->signaled_type; + if ((fence->signaled_type & flush_type) == flush_type) + goto out0; + + ret = mgr->info.signaled(mgr, fence->private, flush_type, signaled); + if (ret) { + *signaled = fence->signaled_type; + goto out0; + } + + if ((fence->signaled_type | *signaled) == fence->signaled_type) + goto out0; + + _glthread_LOCK_MUTEX(mgr->mutex); + _glthread_UNLOCK_MUTEX(fence->mutex); + + driSignalPreviousFencesLocked(mgr, &fence->head, fence->fence_class, + *signaled); + + _glthread_UNLOCK_MUTEX(mgr->mutex); + return 0; + out0: + _glthread_UNLOCK_MUTEX(fence->mutex); + return ret; +} + +struct _DriFenceObject * +driFenceReference(struct _DriFenceObject *fence) +{ + _glthread_LOCK_MUTEX(fence->mgr->mutex); + ++fence->refCount; + _glthread_UNLOCK_MUTEX(fence->mgr->mutex); + return fence; +} + +void +driFenceUnReference(struct _DriFenceObject **pFence) +{ + struct _DriFenceMgr *mgr; + + if (*pFence == NULL) + return; + + mgr = (*pFence)->mgr; + _glthread_LOCK_MUTEX(mgr->mutex); + ++mgr->refCount; + driFenceUnReferenceLocked(pFence); + driFenceMgrUnrefUnlock(&mgr); +} + +struct _DriFenceObject +*driFenceCreate(struct _DriFenceMgr *mgr, uint32_t fence_class, + uint32_t fence_type, void *private, size_t private_size) +{ + struct _DriFenceObject *fence; + size_t fence_size = sizeof(*fence); + + if (private_size) + fence_size = ((fence_size + 15) & ~15); + + fence = calloc(1, fence_size + private_size); + + if (!fence) { + int ret = mgr->info.finish(mgr, private, fence_type, 0); + + if (ret) + usleep(10000000); + + return NULL; + } + + _glthread_INIT_MUTEX(fence->mutex); + _glthread_LOCK_MUTEX(fence->mutex); + _glthread_LOCK_MUTEX(mgr->mutex); + fence->refCount = 1; + DRMLISTADDTAIL(&fence->head, &mgr->heads[fence_class]); + fence->mgr = mgr; + ++mgr->refCount; + ++mgr->num_fences; + _glthread_UNLOCK_MUTEX(mgr->mutex); + fence->fence_class = fence_class; + fence->fence_type = fence_type; + fence->signaled_type = 0; + fence->private = private; + if (private_size) { + fence->private = (void *)(((uint8_t *) fence) + fence_size); + memcpy(fence->private, private, private_size); + } + + _glthread_UNLOCK_MUTEX(fence->mutex); + return fence; +} + + +static int +tSignaled(struct _DriFenceMgr *mgr, void *private, uint32_t flush_type, + uint32_t *signaled_type) +{ + long fd = (long) mgr->private; + int dummy; + drmFence *fence = (drmFence *) private; + int ret; + + *signaled_type = 0; + ret = drmFenceSignaled((int) fd, fence, flush_type, &dummy); + if (ret) + return ret; + + *signaled_type = fence->signaled; + + return 0; +} + +static int +tFinish(struct _DriFenceMgr *mgr, void *private, uint32_t fence_type, + int lazy_hint) +{ + long fd = (long) mgr->private; + unsigned flags = lazy_hint ? DRM_FENCE_FLAG_WAIT_LAZY : 0; + + return drmFenceWait((int)fd, flags, (drmFence *) private, fence_type); +} + +static int +tUnref(struct _DriFenceMgr *mgr, void **private) +{ + long fd = (long) mgr->private; + drmFence *fence = (drmFence *) *private; + *private = NULL; + + return drmFenceUnreference(fd, fence); +} + +struct _DriFenceMgr *driFenceMgrTTMInit(int fd) +{ + struct _DriFenceMgrCreateInfo info; + struct _DriFenceMgr *mgr; + + info.flags = DRI_FENCE_CLASS_ORDERED; + info.num_classes = 4; + info.signaled = tSignaled; + info.finish = tFinish; + info.unreference = tUnref; + + mgr = driFenceMgrCreate(&info); + if (mgr == NULL) + return NULL; + + mgr->private = (void *) (long) fd; + return mgr; +} + diff --git a/src/gallium/winsys/egl_drm/intel/ws_dri_fencemgr.h b/src/gallium/winsys/egl_drm/intel/ws_dri_fencemgr.h new file mode 100644 index 0000000..4ea58df --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/ws_dri_fencemgr.h @@ -0,0 +1,115 @@ +#ifndef DRI_FENCEMGR_H +#define DRI_FENCEMGR_H + +#include <stdint.h> +#include <stdlib.h> + +struct _DriFenceObject; +struct _DriFenceMgr; + +/* + * Do a quick check to see if the fence manager has registered the fence + * object as signaled. Note that this function may return a false negative + * answer. + */ +extern uint32_t driFenceSignaledTypeCached(struct _DriFenceObject *fence); + +/* + * Check if the fence object is signaled. This function can be substantially + * more expensive to call than the above function, but will not return a false + * negative answer. The argument "flush_type" sets the types that the + * underlying mechanism must make sure will eventually signal. + */ +extern int driFenceSignaledType(struct _DriFenceObject *fence, + uint32_t flush_type, uint32_t *signaled); + +/* + * Convenience functions. + */ + +static inline int driFenceSignaled(struct _DriFenceObject *fence, + uint32_t flush_type) +{ + uint32_t signaled_types; + int ret = driFenceSignaledType(fence, flush_type, &signaled_types); + if (ret) + return 0; + return ((signaled_types & flush_type) == flush_type); +} + +static inline int driFenceSignaledCached(struct _DriFenceObject *fence, + uint32_t flush_type) +{ + uint32_t signaled_types = + driFenceSignaledTypeCached(fence); + + return ((signaled_types & flush_type) == flush_type); +} + +/* + * Reference a fence object. + */ +extern struct _DriFenceObject *driFenceReference(struct _DriFenceObject *fence); + +/* + * Unreference a fence object. The fence object pointer will be reset to NULL. + */ + +extern void driFenceUnReference(struct _DriFenceObject **pFence); + + +/* + * Wait for a fence to signal the indicated fence_type. + * If "lazy_hint" is true, it indicates that the wait may sleep to avoid + * busy-wait polling. + */ +extern int driFenceFinish(struct _DriFenceObject *fence, uint32_t fence_type, + int lazy_hint); + +/* + * Create a DriFenceObject for manager "mgr". + * + * "private" is a pointer that should be used for the callbacks in + * struct _DriFenceMgrCreateInfo. + * + * if private_size is nonzero, then the info stored at *private, with size + * private size will be copied and the fence manager will instead use a + * pointer to the copied data for the callbacks in + * struct _DriFenceMgrCreateInfo. In that case, the object pointed to by + * "private" may be destroyed after the call to driFenceCreate. + */ +extern struct _DriFenceObject *driFenceCreate(struct _DriFenceMgr *mgr, + uint32_t fence_class, + uint32_t fence_type, + void *private, + size_t private_size); + +extern uint32_t driFenceType(struct _DriFenceObject *fence); + +/* + * Fence creations are ordered. If a fence signals a fence_type, + * it is safe to assume that all fences of the same class that was + * created before that fence has signaled the same type. + */ + +#define DRI_FENCE_CLASS_ORDERED (1 << 0) + +struct _DriFenceMgrCreateInfo { + uint32_t flags; + uint32_t num_classes; + int (*signaled) (struct _DriFenceMgr *mgr, void *private, uint32_t flush_type, + uint32_t *signaled_type); + int (*finish) (struct _DriFenceMgr *mgr, void *private, uint32_t fence_type, int lazy_hint); + int (*unreference) (struct _DriFenceMgr *mgr, void **private); +}; + +extern struct _DriFenceMgr * +driFenceMgrCreate(const struct _DriFenceMgrCreateInfo *info); + +void +driFenceMgrUnReference(struct _DriFenceMgr **pMgr); + +extern struct _DriFenceMgr * +driFenceMgrTTMInit(int fd); + +#endif diff --git a/src/gallium/winsys/egl_drm/intel/ws_dri_mallocpool.c b/src/gallium/winsys/egl_drm/intel/ws_dri_mallocpool.c new file mode 100644 index 0000000..bf97d7e --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/ws_dri_mallocpool.c @@ -0,0 +1,162 @@ +/************************************************************************** + * + * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, TX., USA + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * + **************************************************************************/ +/* + * Authors: Thomas Hellström <thomas-at-tungstengraphics-dot-com> + */ + +#include <xf86drm.h> +#include <stdlib.h> +#include <errno.h> +#include "imports.h" +#include "glthread.h" +#include "ws_dri_bufpool.h" +#include "ws_dri_bufmgr.h" +#include "intel_screen.h" + +static void * +pool_create(struct _DriBufferPool *pool, + unsigned long size, uint64_t flags, unsigned hint, + unsigned alignment) +{ + unsigned long *private = malloc(size + 2*sizeof(unsigned long)); + if ((flags & DRM_BO_MASK_MEM) != DRM_BO_FLAG_MEM_LOCAL) + abort(); + + *private = size; + return (void *)private; +} + + +static int +pool_destroy(struct _DriBufferPool *pool, void *private) +{ + free(private); + return 0; +} + +static int +pool_waitIdle(struct _DriBufferPool *pool, void *private, + _glthread_Mutex *mutex, int lazy) +{ + return 0; +} + +static int +pool_map(struct _DriBufferPool *pool, void *private, unsigned flags, + int hint, _glthread_Mutex *mutex, void **virtual) +{ + *virtual = (void *)((unsigned long *)private + 2); + return 0; +} + +static int +pool_unmap(struct _DriBufferPool *pool, void *private) +{ + return 0; +} + +static unsigned long +pool_offset(struct _DriBufferPool *pool, void *private) +{ + /* + * BUG + */ + abort(); + return 0UL; +} + +static unsigned long +pool_poolOffset(struct _DriBufferPool *pool, void *private) +{ + /* + * BUG + */ + abort(); +} + +static uint64_t +pool_flags(struct _DriBufferPool *pool, void *private) +{ + return DRM_BO_FLAG_MEM_LOCAL | DRM_BO_FLAG_CACHED; +} + +static unsigned long +pool_size(struct _DriBufferPool *pool, void *private) +{ + return *(unsigned long *) private; +} + + +static int +pool_fence(struct _DriBufferPool *pool, void *private, + struct _DriFenceObject *fence) +{ + abort(); + return 0UL; +} + +static drmBO * +pool_kernel(struct _DriBufferPool *pool, void *private) +{ + abort(); + return NULL; +} + +static void +pool_takedown(struct _DriBufferPool *pool) +{ + free(pool); +} + + +struct _DriBufferPool * +driMallocPoolInit(void) +{ + struct _DriBufferPool *pool; + + pool = (struct _DriBufferPool *) malloc(sizeof(*pool)); + if (!pool) + return NULL; + + pool->data = NULL; + pool->fd = -1; + pool->map = &pool_map; + pool->unmap = &pool_unmap; + pool->destroy = &pool_destroy; + pool->offset = &pool_offset; + pool->poolOffset = &pool_poolOffset; + pool->flags = &pool_flags; + pool->size = &pool_size; + pool->create = &pool_create; + pool->fence = &pool_fence; + pool->kernel = &pool_kernel; + pool->validate = NULL; + pool->waitIdle = &pool_waitIdle; + pool->takeDown = &pool_takedown; + return pool; +} diff --git a/src/gallium/winsys/egl_drm/intel/ws_dri_slabpool.c b/src/gallium/winsys/egl_drm/intel/ws_dri_slabpool.c new file mode 100644 index 0000000..e69519a --- /dev/null +++ b/src/gallium/winsys/egl_drm/intel/ws_dri_slabpool.c @@ -0,0 +1,970 @@ +/************************************************************************** + * + * Copyright 2006-2008 Tungsten Graphics, Inc., Cedar Park, TX., USA + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * + **************************************************************************/ +/* + * Authors: Thomas Hellstrom <thomas-at-tungstengraphics-dot-com> + */ + +#include <stdint.h> +#include <sys/time.h> +#include <errno.h> +#include <unistd.h> +#include <assert.h> +#include "ws_dri_bufpool.h" +#include "ws_dri_fencemgr.h" +#include "ws_dri_bufmgr.h" +#include "glthread.h" + +#define DRI_SLABPOOL_ALLOC_RETRIES 100 + +struct _DriSlab; + +struct _DriSlabBuffer { + int isSlabBuffer; + drmBO *bo; + struct _DriFenceObject *fence; + struct _DriSlab *parent; + drmMMListHead head; + uint32_t mapCount; + uint32_t start; + uint32_t fenceType; + int unFenced; + _glthread_Cond event; +}; + +struct _DriKernelBO { + int fd; + drmBO bo; + drmMMListHead timeoutHead; + drmMMListHead head; + struct timeval timeFreed; + uint32_t pageAlignment; + void *virtual; +}; + +struct _DriSlab{ + drmMMListHead head; + drmMMListHead freeBuffers; + uint32_t numBuffers; + uint32_t numFree; + struct _DriSlabBuffer *buffers; + struct _DriSlabSizeHeader *header; + struct _DriKernelBO *kbo; +}; + + +struct _DriSlabSizeHeader { + drmMMListHead slabs; + drmMMListHead freeSlabs; + drmMMListHead delayedBuffers; + uint32_t numDelayed; + struct _DriSlabPool *slabPool; + uint32_t bufSize; + _glthread_Mutex mutex; +}; + +struct _DriFreeSlabManager { + struct timeval slabTimeout; + struct timeval checkInterval; + struct timeval nextCheck; + drmMMListHead timeoutList; + drmMMListHead unCached; + drmMMListHead cached; + _glthread_Mutex mutex; +}; + + +struct _DriSlabPool { + + /* + * The data of this structure remains constant after + * initialization and thus needs no mutex protection. + */ + + struct _DriFreeSlabManager *fMan; + uint64_t proposedFlags; + uint64_t validMask; + uint32_t *bucketSizes; + uint32_t numBuckets; + uint32_t pageSize; + int fd; + int pageAlignment; + int maxSlabSize; + int desiredNumBuffers; + struct _DriSlabSizeHeader *headers; +}; + +/* + * FIXME: Perhaps arrange timeout slabs in size buckets for fast + * retreival?? + */ + + +static inline int +driTimeAfterEq(struct timeval *arg1, struct timeval *arg2) +{ + return ((arg1->tv_sec > arg2->tv_sec) || + ((arg1->tv_sec == arg2->tv_sec) && + (arg1->tv_usec > arg2->tv_usec))); +} + +static inline void +driTimeAdd(struct timeval *arg, struct timeval *add) +{ + unsigned int sec; + + arg->tv_sec += add->tv_sec; + arg->tv_usec += add->tv_usec; + sec = arg->tv_usec / 1000000; + arg->tv_sec += sec; + arg->tv_usec -= sec*1000000; +} + +static void +driFreeKernelBO(struct _DriKernelBO *kbo) +{ + if (!kbo) + return; + + (void) drmBOUnreference(kbo->fd, &kbo->bo); + free(kbo); +} + + +static void +driFreeTimeoutKBOsLocked(struct _DriFreeSlabManager *fMan, + struct timeval *time) +{ + drmMMListHead *list, *next; + struct _DriKernelBO *kbo; + + if (!driTimeAfterEq(time, &fMan->nextCheck)) + return; + + for (list = fMan->timeoutList.next, next = list->next; + list != &fMan->timeoutList; + list = next, next = list->next) { + + kbo = DRMLISTENTRY(struct _DriKernelBO, list, timeoutHead); + + if (!driTimeAfterEq(time, &kbo->timeFreed)) + break; + + DRMLISTDELINIT(&kbo->timeoutHead); + DRMLISTDELINIT(&kbo->head); + driFreeKernelBO(kbo); + } + + fMan->nextCheck = *time; + driTimeAdd(&fMan->nextCheck, &fMan->checkInterval); +} + + +/* + * Add a _DriKernelBO to the free slab manager. + * This means that it is available for reuse, but if it's not + * reused in a while, it will be freed. + */ + +static void +driSetKernelBOFree(struct _DriFreeSlabManager *fMan, + struct _DriKernelBO *kbo) +{ + struct timeval time; + + _glthread_LOCK_MUTEX(fMan->mutex); + gettimeofday(&time, NULL); + driTimeAdd(&time, &fMan->slabTimeout); + + kbo->timeFreed = time; + + if (kbo->bo.flags & DRM_BO_FLAG_CACHED) + DRMLISTADD(&kbo->head, &fMan->cached); + else + DRMLISTADD(&kbo->head, &fMan->unCached); + + DRMLISTADDTAIL(&kbo->timeoutHead, &fMan->timeoutList); + driFreeTimeoutKBOsLocked(fMan, &time); + + _glthread_UNLOCK_MUTEX(fMan->mutex); +} + +/* + * Get a _DriKernelBO for us to use as storage for a slab. + * + */ + +static struct _DriKernelBO * +driAllocKernelBO(struct _DriSlabSizeHeader *header) + +{ + struct _DriSlabPool *slabPool = header->slabPool; + struct _DriFreeSlabManager *fMan = slabPool->fMan; + drmMMListHead *list, *next, *head; + uint32_t size = header->bufSize * slabPool->desiredNumBuffers; + struct _DriKernelBO *kbo; + struct _DriKernelBO *kboTmp; + int ret; + + /* + * FIXME: We should perhaps allow some variation in slabsize in order + * to efficiently reuse slabs. + */ + + size = (size <= slabPool->maxSlabSize) ? size : slabPool->maxSlabSize; + size = (size + slabPool->pageSize - 1) & ~(slabPool->pageSize - 1); + _glthread_LOCK_MUTEX(fMan->mutex); + + kbo = NULL; + + retry: + head = (slabPool->proposedFlags & DRM_BO_FLAG_CACHED) ? + &fMan->cached : &fMan->unCached; + + for (list = head->next, next = list->next; + list != head; + list = next, next = list->next) { + + kboTmp = DRMLISTENTRY(struct _DriKernelBO, list, head); + + if ((kboTmp->bo.size == size) && + (slabPool->pageAlignment == 0 || + (kboTmp->pageAlignment % slabPool->pageAlignment) == 0)) { + + if (!kbo) + kbo = kboTmp; + + if ((kbo->bo.proposedFlags ^ slabPool->proposedFlags) == 0) + break; + + } + } + + if (kbo) { + DRMLISTDELINIT(&kbo->head); + DRMLISTDELINIT(&kbo->timeoutHead); + } + + _glthread_UNLOCK_MUTEX(fMan->mutex); + + if (kbo) { + uint64_t new_mask = kbo->bo.proposedFlags ^ slabPool->proposedFlags; + + ret = 0; + if (new_mask) { + ret = drmBOSetStatus(kbo->fd, &kbo->bo, slabPool->proposedFlags, + new_mask, DRM_BO_HINT_DONT_FENCE, 0, 0); + } + if (ret == 0) + return kbo; + + driFreeKernelBO(kbo); + kbo = NULL; + goto retry; + } + + kbo = calloc(1, sizeof(struct _DriKernelBO)); + if (!kbo) + return NULL; + + kbo->fd = slabPool->fd; + DRMINITLISTHEAD(&kbo->head); + DRMINITLISTHEAD(&kbo->timeoutHead); + ret = drmBOCreate(kbo->fd, size, slabPool->pageAlignment, NULL, + slabPool->proposedFlags, + DRM_BO_HINT_DONT_FENCE, &kbo->bo); + if (ret) + goto out_err0; + + ret = drmBOMap(kbo->fd, &kbo->bo, + DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE, + 0, &kbo->virtual); + + if (ret) + goto out_err1; + + ret = drmBOUnmap(kbo->fd, &kbo->bo); + if (ret) + goto out_err1; + + return kbo; + + out_err1: + drmBOUnreference(kbo->fd, &kbo->bo); + out_err0: + free(kbo); + return NULL; +} + + +static int +driAllocSlab(struct _DriSlabSizeHeader *header) +{ + struct _DriSlab *slab; + struct _DriSlabBuffer *buf; + uint32_t numBuffers; + int ret; + int i; + + slab = calloc(1, sizeof(*slab)); + if (!slab) + return -ENOMEM; + + slab->kbo = driAllocKernelBO(header); + if (!slab->kbo) { + ret = -ENOMEM; + goto out_err0; + } + + numBuffers = slab->kbo->bo.size / header->bufSize; + + slab->buffers = calloc(numBuffers, sizeof(*slab->buffers)); + if (!slab->buffers) { + ret = -ENOMEM; + goto out_err1; + } + + DRMINITLISTHEAD(&slab->head); + DRMINITLISTHEAD(&slab->freeBuffers); + slab->numBuffers = numBuffers; + slab->numFree = 0; + slab->header = header; + + buf = slab->buffers; + for (i=0; i < numBuffers; ++i) { + buf->parent = slab; + buf->start = i* header->bufSize; + buf->mapCount = 0; + buf->isSlabBuffer = 1; + _glthread_INIT_COND(buf->event); + DRMLISTADDTAIL(&buf->head, &slab->freeBuffers); + slab->numFree++; + buf++; + } + + DRMLISTADDTAIL(&slab->head, &header->slabs); + + return 0; + + out_err1: + driSetKernelBOFree(header->slabPool->fMan, slab->kbo); + free(slab->buffers); + out_err0: + free(slab); + return ret; +} + +/* + * Delete a buffer from the slab header delayed list and put + * it on the slab free list. + */ + +static void +driSlabFreeBufferLocked(struct _DriSlabBuffer *buf) +{ + struct _DriSlab *slab = buf->parent; + struct _DriSlabSizeHeader *header = slab->header; + drmMMListHead *list = &buf->head; + + DRMLISTDEL(list); + DRMLISTADDTAIL(list, &slab->freeBuffers); + slab->numFree++; + + if (slab->head.next == &slab->head) + DRMLISTADDTAIL(&slab->head, &header->slabs); + + if (slab->numFree == slab->numBuffers) { + list = &slab->head; + DRMLISTDEL(list); + DRMLISTADDTAIL(list, &header->freeSlabs); + } + + if (header->slabs.next == &header->slabs || + slab->numFree != slab->numBuffers) { + + drmMMListHead *next; + struct _DriFreeSlabManager *fMan = header->slabPool->fMan; + + for (list = header->freeSlabs.next, next = list->next; + list != &header->freeSlabs; + list = next, next = list->next) { + + slab = DRMLISTENTRY(struct _DriSlab, list, head); + + DRMLISTDELINIT(list); + driSetKernelBOFree(fMan, slab->kbo); + free(slab->buffers); + free(slab); + } + } +} + +static void +driSlabCheckFreeLocked(struct _DriSlabSizeHeader *header, int wait) +{ + drmMMListHead *list, *prev, *first; + struct _DriSlabBuffer *buf; + struct _DriSlab *slab; + int firstWasSignaled = 1; + int signaled; + int i; + int ret; + + /* + * Rerun the freeing test if the youngest tested buffer + * was signaled, since there might be more idle buffers + * in the delay list. + */ + + while (firstWasSignaled) { + firstWasSignaled = 0; + signaled = 0; + first = header->delayedBuffers.next; + + /* Only examine the oldest 1/3 of delayed buffers: + */ + if (header->numDelayed > 3) { + for (i = 0; i < header->numDelayed; i += 3) { + first = first->next; + } + } + + for (list = first, prev = list->prev; + list != &header->delayedBuffers; + list = prev, prev = list->prev) { + buf = DRMLISTENTRY(struct _DriSlabBuffer, list, head); + slab = buf->parent; + + if (!signaled) { + if (wait) { + ret = driFenceFinish(buf->fence, buf->fenceType, 0); + if (ret) + break; + signaled = 1; + wait = 0; + } else { + signaled = driFenceSignaled(buf->fence, buf->fenceType); + } + if (signaled) { + if (list == first) + firstWasSignaled = 1; + driFenceUnReference(&buf->fence); + header->numDelayed--; + driSlabFreeBufferLocked(buf); + } + } else if (driFenceSignaledCached(buf->fence, buf->fenceType)) { + driFenceUnReference(&buf->fence); + header->numDelayed--; + driSlabFreeBufferLocked(buf); + } + } + } +} + + +static struct _DriSlabBuffer * +driSlabAllocBuffer(struct _DriSlabSizeHeader *header) +{ + static struct _DriSlabBuffer *buf; + struct _DriSlab *slab; + drmMMListHead *list; + int count = DRI_SLABPOOL_ALLOC_RETRIES; + + _glthread_LOCK_MUTEX(header->mutex); + while(header->slabs.next == &header->slabs && count > 0) { + driSlabCheckFreeLocked(header, 0); + if (header->slabs.next != &header->slabs) + break; + + _glthread_UNLOCK_MUTEX(header->mutex); + if (count != DRI_SLABPOOL_ALLOC_RETRIES) + usleep(1); + _glthread_LOCK_MUTEX(header->mutex); + (void) driAllocSlab(header); + count--; + } + + list = header->slabs.next; + if (list == &header->slabs) { + _glthread_UNLOCK_MUTEX(header->mutex); + return NULL; + } + slab = DRMLISTENTRY(struct _DriSlab, list, head); + if (--slab->numFree == 0) + DRMLISTDELINIT(list); + + list = slab->freeBuffers.next; + DRMLISTDELINIT(list); + + _glthread_UNLOCK_MUTEX(header->mutex); + buf = DRMLISTENTRY(struct _DriSlabBuffer, list, head); + return buf; +} + +static void * +pool_create(struct _DriBufferPool *driPool, unsigned long size, + uint64_t flags, unsigned hint, unsigned alignment) +{ + struct _DriSlabPool *pool = (struct _DriSlabPool *) driPool->data; + struct _DriSlabSizeHeader *header; + struct _DriSlabBuffer *buf; + void *dummy; + int i; + int ret; + + /* + * FIXME: Check for compatibility. + */ + header = pool->headers; + for (i=0; i<pool->numBuckets; ++i) { + if (header->bufSize >= size) + break; + header++; + } + + if (i < pool->numBuckets) + return driSlabAllocBuffer(header); + + + /* + * Fall back to allocate a buffer object directly from DRM. + * and wrap it in a driBO structure. + */ + + + buf = calloc(1, sizeof(*buf)); + + if (!buf) + return NULL; + + buf->bo = calloc(1, sizeof(*buf->bo)); + if (!buf->bo) + goto out_err0; + + if (alignment) { + if ((alignment < pool->pageSize) && (pool->pageSize % alignment)) + goto out_err1; + if ((alignment > pool->pageSize) && (alignment % pool->pageSize)) + goto out_err1; + } + + ret = drmBOCreate(pool->fd, size, alignment / pool->pageSize, NULL, + flags, hint, buf->bo); + if (ret) + goto out_err1; + + ret = drmBOMap(pool->fd, buf->bo, DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE, + 0, &dummy); + if (ret) + goto out_err2; + + ret = drmBOUnmap(pool->fd, buf->bo); + if (ret) + goto out_err2; + + return buf; + out_err2: + drmBOUnreference(pool->fd, buf->bo); + out_err1: + free(buf->bo); + out_err0: + free(buf); + return NULL; +} + +static int +pool_destroy(struct _DriBufferPool *driPool, void *private) +{ + struct _DriSlabBuffer *buf = + (struct _DriSlabBuffer *) private; + struct _DriSlab *slab; + struct _DriSlabSizeHeader *header; + + if (!buf->isSlabBuffer) { + struct _DriSlabPool *pool = (struct _DriSlabPool *) driPool->data; + int ret; + + ret = drmBOUnreference(pool->fd, buf->bo); + free(buf->bo); + free(buf); + return ret; + } + + slab = buf->parent; + header = slab->header; + + _glthread_LOCK_MUTEX(header->mutex); + buf->unFenced = 0; + buf->mapCount = 0; + + if (buf->fence && !driFenceSignaledCached(buf->fence, buf->fenceType)) { + DRMLISTADDTAIL(&buf->head, &header->delayedBuffers); + header->numDelayed++; + } else { + if (buf->fence) + driFenceUnReference(&buf->fence); + driSlabFreeBufferLocked(buf); + } + + _glthread_UNLOCK_MUTEX(header->mutex); + return 0; +} + +static int +pool_waitIdle(struct _DriBufferPool *driPool, void *private, + _glthread_Mutex *mutex, int lazy) +{ + struct _DriSlabBuffer *buf = (struct _DriSlabBuffer *) private; + + while(buf->unFenced) + _glthread_COND_WAIT(buf->event, *mutex); + + if (!buf->fence) + return 0; + + driFenceFinish(buf->fence, buf->fenceType, lazy); + driFenceUnReference(&buf->fence); + + return 0; +} + +static int +pool_map(struct _DriBufferPool *pool, void *private, unsigned flags, + int hint, _glthread_Mutex *mutex, void **virtual) +{ + struct _DriSlabBuffer *buf = (struct _DriSlabBuffer *) private; + int busy; + + if (buf->isSlabBuffer) + busy = buf->unFenced || (buf->fence && !driFenceSignaledCached(buf->fence, buf->fenceType)); + else + busy = buf->fence && !driFenceSignaled(buf->fence, buf->fenceType); + + + if (busy) { + if (hint & DRM_BO_HINT_DONT_BLOCK) + return -EBUSY; + else { + (void) pool_waitIdle(pool, private, mutex, 0); + } + } + + ++buf->mapCount; + *virtual = (buf->isSlabBuffer) ? + (void *) ((uint8_t *) buf->parent->kbo->virtual + buf->start) : + (void *) buf->bo->virtual; + + return 0; +} + +static int +pool_unmap(struct _DriBufferPool *pool, void *private) +{ + struct _DriSlabBuffer *buf = (struct _DriSlabBuffer *) private; + + --buf->mapCount; + if (buf->mapCount == 0 && buf->isSlabBuffer) + _glthread_COND_BROADCAST(buf->event); + + return 0; +} + +static unsigned long +pool_offset(struct _DriBufferPool *pool, void *private) +{ + struct _DriSlabBuffer *buf = (struct _DriSlabBuffer *) private; + struct _DriSlab *slab; + struct _DriSlabSizeHeader *header; + + if (!buf->isSlabBuffer) { + assert(buf->bo->proposedFlags & DRM_BO_FLAG_NO_MOVE); + return buf->bo->offset; + } + + slab = buf->parent; + header = slab->header; + + (void) header; + assert(header->slabPool->proposedFlags & DRM_BO_FLAG_NO_MOVE); + return slab->kbo->bo.offset + buf->start; +} + +static unsigned long +pool_poolOffset(struct _DriBufferPool *pool, void *private) +{ + struct _DriSlabBuffer *buf = (struct _DriSlabBuffer *) private; + + return buf->start; +} + +static uint64_t +pool_flags(struct _DriBufferPool *pool, void *private) +{ + struct _DriSlabBuffer *buf = (struct _DriSlabBuffer *) private; + + if (!buf->isSlabBuffer) + return buf->bo->flags; + + return buf->parent->kbo->bo.flags; +} + +static unsigned long +pool_size(struct _DriBufferPool *pool, void *private) +{ + struct _DriSlabBuffer *buf = (struct _DriSlabBuffer *) private; + if (!buf->isSlabBuffer) + return buf->bo->size; + + return buf->parent->header->bufSize; +} + +static int +pool_fence(struct _DriBufferPool *pool, void *private, + struct _DriFenceObject *fence) +{ + struct _DriSlabBuffer *buf = (struct _DriSlabBuffer *) private; + drmBO *bo; + + if (buf->fence) + driFenceUnReference(&buf->fence); + + buf->fence = driFenceReference(fence); + bo = (buf->isSlabBuffer) ? + &buf->parent->kbo->bo: + buf->bo; + buf->fenceType = bo->fenceFlags; + + buf->unFenced = 0; + _glthread_COND_BROADCAST(buf->event); + + return 0; +} + +static drmBO * +pool_kernel(struct _DriBufferPool *pool, void *private) +{ + struct _DriSlabBuffer *buf = (struct _DriSlabBuffer *) private; + + if (!buf) + return NULL; + + return (buf->isSlabBuffer) ? &buf->parent->kbo->bo : buf->bo; +} + +static int +pool_validate(struct _DriBufferPool *pool, void *private, + _glthread_Mutex *mutex) +{ + struct _DriSlabBuffer *buf = (struct _DriSlabBuffer *) private; + + if (!buf->isSlabBuffer) + return 0; + + while(buf->mapCount != 0) + _glthread_COND_WAIT(buf->event, *mutex); + + buf->unFenced = 1; + return 0; +} + + +struct _DriFreeSlabManager * +driInitFreeSlabManager(uint32_t checkIntervalMsec, uint32_t slabTimeoutMsec) +{ + struct _DriFreeSlabManager *tmp; + + tmp = calloc(1, sizeof(*tmp)); + if (!tmp) + return NULL; + + _glthread_INIT_MUTEX(tmp->mutex); + _glthread_LOCK_MUTEX(tmp->mutex); + tmp->slabTimeout.tv_usec = slabTimeoutMsec*1000; + tmp->slabTimeout.tv_sec = tmp->slabTimeout.tv_usec / 1000000; + tmp->slabTimeout.tv_usec -= tmp->slabTimeout.tv_sec*1000000; + + tmp->checkInterval.tv_usec = checkIntervalMsec*1000; + tmp->checkInterval.tv_sec = tmp->checkInterval.tv_usec / 1000000; + tmp->checkInterval.tv_usec -= tmp->checkInterval.tv_sec*1000000; + + gettimeofday(&tmp->nextCheck, NULL); + driTimeAdd(&tmp->nextCheck, &tmp->checkInterval); + DRMINITLISTHEAD(&tmp->timeoutList); + DRMINITLISTHEAD(&tmp->unCached); + DRMINITLISTHEAD(&tmp->cached); + _glthread_UNLOCK_MUTEX(tmp->mutex); + + return tmp; +} + +void +driFinishFreeSlabManager(struct _DriFreeSlabManager *fMan) +{ + struct timeval time; + + time = fMan->nextCheck; + driTimeAdd(&time, &fMan->checkInterval); + + _glthread_LOCK_MUTEX(fMan->mutex); + driFreeTimeoutKBOsLocked(fMan, &time); + _glthread_UNLOCK_MUTEX(fMan->mutex); + + assert(fMan->timeoutList.next == &fMan->timeoutList); + assert(fMan->unCached.next == &fMan->unCached); + assert(fMan->cached.next == &fMan->cached); + + free(fMan); +} + +static void +driInitSizeHeader(struct _DriSlabPool *pool, uint32_t size, + struct _DriSlabSizeHeader *header) +{ + _glthread_INIT_MUTEX(header->mutex); + _glthread_LOCK_MUTEX(header->mutex); + + DRMINITLISTHEAD(&header->slabs); + DRMINITLISTHEAD(&header->freeSlabs); + DRMINITLISTHEAD(&header->delayedBuffers); + + header->numDelayed = 0; + header->slabPool = pool; + header->bufSize = size; + + _glthread_UNLOCK_MUTEX(header->mutex); +} + +static void +driFinishSizeHeader(struct _DriSlabSizeHeader *header) +{ + drmMMListHead *list, *next; + struct _DriSlabBuffer *buf; + + _glthread_LOCK_MUTEX(header->mutex); + for (list = header->delayedBuffers.next, next = list->next; + list != &header->delayedBuffers; + list = next, next = list->next) { + + buf = DRMLISTENTRY(struct _DriSlabBuffer, list , head); + if (buf->fence) { + (void) driFenceFinish(buf->fence, buf->fenceType, 0); + driFenceUnReference(&buf->fence); + } + header->numDelayed--; + driSlabFreeBufferLocked(buf); + } + _glthread_UNLOCK_MUTEX(header->mutex); +} + +static void +pool_takedown(struct _DriBufferPool *driPool) +{ + struct _DriSlabPool *pool = driPool->data; + int i; + + for (i=0; i<pool->numBuckets; ++i) { + driFinishSizeHeader(&pool->headers[i]); + } + + free(pool->headers); + free(pool->bucketSizes); + free(pool); + free(driPool); +} + +struct _DriBufferPool * +driSlabPoolInit(int fd, uint64_t flags, + uint64_t validMask, + uint32_t smallestSize, + uint32_t numSizes, + uint32_t desiredNumBuffers, + uint32_t maxSlabSize, + uint32_t pageAlignment, + struct _DriFreeSlabManager *fMan) +{ + struct _DriBufferPool *driPool; + struct _DriSlabPool *pool; + uint32_t i; + + driPool = calloc(1, sizeof(*driPool)); + if (!driPool) + return NULL; + + pool = calloc(1, sizeof(*pool)); + if (!pool) + goto out_err0; + + pool->bucketSizes = calloc(numSizes, sizeof(*pool->bucketSizes)); + if (!pool->bucketSizes) + goto out_err1; + + pool->headers = calloc(numSizes, sizeof(*pool->headers)); + if (!pool->headers) + goto out_err2; + + pool->fMan = fMan; + pool->proposedFlags = flags; + pool->validMask = validMask; + pool->numBuckets = numSizes; + pool->pageSize = getpagesize(); + pool->fd = fd; + pool->pageAlignment = pageAlignment; + pool->maxSlabSize = maxSlabSize; + pool->desiredNumBuffers = desiredNumBuffers; + + for (i=0; i<pool->numBuckets; ++i) { + pool->bucketSizes[i] = (smallestSize << i); + driInitSizeHeader(pool, pool->bucketSizes[i], + &pool->headers[i]); + } + + driPool->data = (void *) pool; + driPool->map = &pool_map; + driPool->unmap = &pool_unmap; + driPool->destroy = &pool_destroy; + driPool->offset = &pool_offset; + driPool->poolOffset = &pool_poolOffset; + driPool->flags = &pool_flags; + driPool->size = &pool_size; + driPool->create = &pool_create; + driPool->fence = &pool_fence; + driPool->kernel = &pool_kernel; + driPool->validate = &pool_validate; + driPool->waitIdle = &pool_waitIdle; + driPool->takeDown = &pool_takedown; + + return driPool; + + out_err2: + free(pool->bucketSizes); + out_err1: + free(pool); + out_err0: + free(driPool); + + return NULL; +} diff --git a/src/gallium/winsys/xlib/xm_winsys.c b/src/gallium/winsys/xlib/xm_winsys.c index 14c3892..b14758f 100644 --- a/src/gallium/winsys/xlib/xm_winsys.c +++ b/src/gallium/winsys/xlib/xm_winsys.c @@ -489,7 +489,8 @@ xm_surface_alloc_storage(struct pipe_winsys *winsys, struct pipe_surface *surf, unsigned width, unsigned height, enum pipe_format format, - unsigned flags) + unsigned flags, + unsigned tex_usage) { const unsigned alignment = 64; @@ -498,6 +499,7 @@ xm_surface_alloc_storage(struct pipe_winsys *winsys, surf->format = format; surf->cpp = pf_get_size(format); surf->pitch = round_up(width, alignment / surf->cpp); + surf->usage = flags; #ifdef GALLIUM_CELL /* XXX a bit of a hack */ height = round_up(height, TILE_SIZE); @@ -551,6 +553,7 @@ static void xm_surface_release(struct pipe_winsys *winsys, struct pipe_surface **s) { struct pipe_surface *surf = *s; + assert(!surf->texture); surf->refcount--; if (surf->refcount == 0) { if (surf->buffer) diff --git a/src/gallium/winsys/xlib/xm_winsys_aub.c b/src/gallium/winsys/xlib/xm_winsys_aub.c index f42f7fc..7737609 100644 --- a/src/gallium/winsys/xlib/xm_winsys_aub.c +++ b/src/gallium/winsys/xlib/xm_winsys_aub.c @@ -276,7 +276,8 @@ aub_i915_surface_alloc_storage(struct pipe_winsys *winsys, struct pipe_surface *surf, unsigned width, unsigned height, enum pipe_format format, - unsigned flags) + unsigned flags, + unsigned tex_usage) { const unsigned alignment = 64; |