winsys/radeon: move managing GEM domains back to drivers

This partially reverts commit 363ff844753c46ac9c13866627e096b091ea81f8.

It caused severe performance drops in Nexuiz. Reported by Phoronix.

Tested by me on r300g and by IRC people on r600g.

4 files changed, 40 insertions, 56 deletions

diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_bo.c b/src/gallium/winsys/radeon/drm/radeon_drm_bo.c
index ccf9c4f..d4746ff 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_bo.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_bo.c
@@ -346,11 +346,9 @@ static struct pb_buffer *radeon_bomgr_create_bo(struct pb_manager *_mgr,
 
     memset(&args, 0, sizeof(args));
 
-    assert(rdesc->initial_domains && rdesc->reloc_domains);
+    assert(rdesc->initial_domains);
     assert((rdesc->initial_domains &
             ~(RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM)) == 0);
-    assert((rdesc->reloc_domains &
-            ~(RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM)) == 0);
 
     args.size = size;
     args.alignment = desc->alignment;
@@ -377,7 +375,6 @@ static struct pb_buffer *radeon_bomgr_create_bo(struct pb_manager *_mgr,
     bo->mgr = mgr;
     bo->rws = mgr->rws;
     bo->handle = args.handle;
-    bo->reloc_domains = rdesc->reloc_domains;
     pipe_mutex_init(bo->map_mutex);
 
     return &bo->base;
@@ -526,7 +523,8 @@ static struct pb_buffer *
 radeon_winsys_bo_create(struct radeon_winsys *rws,
                         unsigned size,
                         unsigned alignment,
-                        unsigned bind, unsigned usage)
+                        unsigned bind,
+                        enum radeon_bo_domain domain)
 {
     struct radeon_drm_winsys *ws = radeon_drm_winsys(rws);
     struct radeon_bo_desc desc;
@@ -536,31 +534,9 @@ radeon_winsys_bo_create(struct radeon_winsys *rws,
     memset(&desc, 0, sizeof(desc));
     desc.base.alignment = alignment;
 
-    /* Determine the memory domains. */
-    switch (usage) {
-    case PIPE_USAGE_STAGING:
-    case PIPE_USAGE_STREAM:
-    case PIPE_USAGE_DYNAMIC:
-            desc.initial_domains = RADEON_GEM_DOMAIN_GTT;
-            desc.reloc_domains = RADEON_GEM_DOMAIN_GTT;
-            break;
-    case PIPE_USAGE_IMMUTABLE:
-    case PIPE_USAGE_STATIC:
-            desc.initial_domains = RADEON_GEM_DOMAIN_VRAM;
-            desc.reloc_domains = RADEON_GEM_DOMAIN_VRAM;
-            break;
-    default:
-            if (bind & (PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER |
-                        PIPE_BIND_CONSTANT_BUFFER)) {
-                desc.initial_domains = RADEON_GEM_DOMAIN_GTT;
-            } else {
-                desc.initial_domains = RADEON_GEM_DOMAIN_VRAM;
-            }
-            desc.reloc_domains = RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM;
-    }
-
     /* Additional criteria for the cache manager. */
-    desc.base.usage = desc.initial_domains;
+    desc.base.usage = domain;
+    desc.initial_domains = domain;
 
     /* Assign a buffer manager. */
     if (bind & (PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER |
@@ -618,7 +594,6 @@ static struct pb_buffer *radeon_winsys_bo_from_handle(struct radeon_winsys *rws,
     }
     bo->handle = open_arg.handle;
     bo->name = whandle->handle;
-    bo->reloc_domains = RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM;
 
     /* Initialize it. */
     pipe_reference_init(&bo->base.reference, 1);
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_bo.h b/src/gallium/winsys/radeon/drm/radeon_drm_bo.h
index ba71cfb..35d25e8 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_bo.h
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_bo.h
@@ -42,7 +42,6 @@ struct radeon_bo_desc {
     struct pb_desc base;
 
     unsigned initial_domains;
-    unsigned reloc_domains;
 };
 
 struct radeon_bo {
@@ -58,7 +57,6 @@ struct radeon_bo {
     void *ptr;
     pipe_mutex map_mutex;
 
-    uint32_t reloc_domains;
     uint32_t handle;
     uint32_t name;
 
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_cs.c b/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
index 2239059..e6109af 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
@@ -181,13 +181,14 @@ static struct radeon_winsys_cs *radeon_drm_cs_create(struct radeon_winsys *rws)
 #define OUT_CS(cs, value) (cs)->buf[(cs)->cdw++] = (value)
 
 static INLINE void update_reloc_domains(struct drm_radeon_cs_reloc *reloc,
-                                        enum radeon_bo_usage usage,
-                                        unsigned domains)
+                                        enum radeon_bo_domain rd,
+                                        enum radeon_bo_domain wd,
+                                        enum radeon_bo_domain *added_domains)
 {
-    if (usage & RADEON_USAGE_READ)
-        reloc->read_domains |= domains;
-    if (usage & RADEON_USAGE_WRITE)
-        reloc->write_domain |= domains;
+    *added_domains = (rd | wd) & ~(reloc->read_domains | reloc->write_domain);
+
+    reloc->read_domains |= rd;
+    reloc->write_domain |= wd;
 }
 
 int radeon_get_reloc(struct radeon_cs_context *csc, struct radeon_bo *bo)
@@ -209,7 +210,7 @@ int radeon_get_reloc(struct radeon_cs_context *csc, struct radeon_bo *bo)
             if (reloc->handle == bo->handle) {
                 /* Put this reloc in the hash list.
                  * This will prevent additional hash collisions if there are
-                 * several subsequent get_reloc calls of the same buffer.
+                 * several consecutive get_reloc calls for the same buffer.
                  *
                  * Example: Assuming buffers A,B,C collide in the hash list,
                  * the following sequence of relocs:
@@ -230,16 +231,19 @@ int radeon_get_reloc(struct radeon_cs_context *csc, struct radeon_bo *bo)
 static unsigned radeon_add_reloc(struct radeon_cs_context *csc,
                                  struct radeon_bo *bo,
                                  enum radeon_bo_usage usage,
-                                 unsigned *added_domains)
+                                 enum radeon_bo_domain domains,
+                                 enum radeon_bo_domain *added_domains)
 {
     struct drm_radeon_cs_reloc *reloc;
     unsigned i;
     unsigned hash = bo->handle & (sizeof(csc->is_handle_added)-1);
+    enum radeon_bo_domain rd = usage & RADEON_USAGE_READ ? domains : 0;
+    enum radeon_bo_domain wd = usage & RADEON_USAGE_WRITE ? domains : 0;
 
     if (csc->is_handle_added[hash]) {
         reloc = csc->relocs_hashlist[hash];
         if (reloc->handle == bo->handle) {
-            update_reloc_domains(reloc, usage, bo->reloc_domains);
+            update_reloc_domains(reloc, rd, wd, added_domains);
             return csc->reloc_indices_hashlist[hash];
         }
 
@@ -248,7 +252,7 @@ static unsigned radeon_add_reloc(struct radeon_cs_context *csc,
             --i;
             reloc = &csc->relocs[i];
             if (reloc->handle == bo->handle) {
-                update_reloc_domains(reloc, usage, bo->reloc_domains);
+                update_reloc_domains(reloc, rd, wd, added_domains);
 
                 csc->relocs_hashlist[hash] = reloc;
                 csc->reloc_indices_hashlist[hash] = i;
@@ -278,10 +282,8 @@ static unsigned radeon_add_reloc(struct radeon_cs_context *csc,
     p_atomic_inc(&bo->num_cs_references);
     reloc = &csc->relocs[csc->crelocs];
     reloc->handle = bo->handle;
-    if (usage & RADEON_USAGE_READ)
-        reloc->read_domains = bo->reloc_domains;
-    if (usage & RADEON_USAGE_WRITE)
-        reloc->write_domain = bo->reloc_domains;
+    reloc->read_domains = rd;
+    reloc->write_domain = wd;
     reloc->flags = 0;
 
     csc->is_handle_added[hash] = TRUE;
@@ -290,23 +292,24 @@ static unsigned radeon_add_reloc(struct radeon_cs_context *csc,
 
     csc->chunks[1].length_dw += RELOC_DWORDS;
 
-    *added_domains = bo->reloc_domains;
+    *added_domains = rd | wd;
     return csc->crelocs++;
 }
 
 static unsigned radeon_drm_cs_add_reloc(struct radeon_winsys_cs *rcs,
                                         struct radeon_winsys_cs_handle *buf,
-                                        enum radeon_bo_usage usage)
+                                        enum radeon_bo_usage usage,
+                                        enum radeon_bo_domain domains)
 {
     struct radeon_drm_cs *cs = radeon_drm_cs(rcs);
     struct radeon_bo *bo = (struct radeon_bo*)buf;
-    unsigned added_domains = 0;
+    enum radeon_bo_domain added_domains;
 
-    unsigned index = radeon_add_reloc(cs->csc, bo, usage, &added_domains);
+    unsigned index = radeon_add_reloc(cs->csc, bo, usage, domains, &added_domains);
 
-    if (added_domains & RADEON_GEM_DOMAIN_GTT)
+    if (added_domains & RADEON_DOMAIN_GTT)
         cs->csc->used_gart += bo->base.size;
-    if (added_domains & RADEON_GEM_DOMAIN_VRAM)
+    if (added_domains & RADEON_DOMAIN_VRAM)
         cs->csc->used_vram += bo->base.size;
 
     return index;
diff --git a/src/gallium/winsys/radeon/drm/radeon_winsys.h b/src/gallium/winsys/radeon/drm/radeon_winsys.h
index ea335d8..59c1aad 100644
--- a/src/gallium/winsys/radeon/drm/radeon_winsys.h
+++ b/src/gallium/winsys/radeon/drm/radeon_winsys.h
@@ -58,6 +58,11 @@ enum radeon_bo_layout {
     RADEON_LAYOUT_UNKNOWN
 };
 
+enum radeon_bo_domain { /* bitfield */
+    RADEON_DOMAIN_GTT  = 2,
+    RADEON_DOMAIN_VRAM = 4
+};
+
 enum radeon_bo_usage { /* bitfield */
     RADEON_USAGE_READ = 2,
     RADEON_USAGE_WRITE = 4,
@@ -137,13 +142,14 @@ struct radeon_winsys {
      * \param size      The size to allocate.
      * \param alignment An alignment of the buffer in memory.
      * \param bind      A bitmask of the PIPE_BIND_* flags.
-     * \param usage     A bitmask of the PIPE_USAGE_* flags.
+     * \param domain    A bitmask of the RADEON_DOMAIN_* flags.
      * \return          The created buffer object.
      */
     struct pb_buffer *(*buffer_create)(struct radeon_winsys *ws,
                                        unsigned size,
                                        unsigned alignment,
-                                       unsigned bind, unsigned usage);
+                                       unsigned bind,
+                                       enum radeon_bo_domain domain);
 
     struct radeon_winsys_cs_handle *(*buffer_get_cs_handle)(
             struct pb_buffer *buf);
@@ -271,12 +277,14 @@ struct radeon_winsys {
      *
      * \param cs  A command stream to add buffer for validation against.
      * \param buf A winsys buffer to validate.
-     * \param usage  Whether the buffer is used for read and/or write.
+     * \param usage   Whether the buffer is used for read and/or write.
+     * \param domain  Bitmask of the RADEON_DOMAIN_* flags.
      * \return Relocation index.
      */
     unsigned (*cs_add_reloc)(struct radeon_winsys_cs *cs,
                              struct radeon_winsys_cs_handle *buf,
-                             enum radeon_bo_usage usage);
+                             enum radeon_bo_usage usage,
+                             enum radeon_bo_domain domain);
 
     /**
      * Return TRUE if there is enough memory in VRAM and GTT for the relocs