llvmpipe: support 8bit subpixel precision

8 bit precision is required by d3d10 but unfortunately
requires 64 bit rasterizer. This commit implements
64 bit rasterization with full support for 8bit subpixel
precision. It's a combination of all individual commits
from the llvmpipe-rast-64 branch.

Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>

1 files changed, 121 insertions, 52 deletions

diff --git a/src/gallium/drivers/llvmpipe/lp_rast_tri.c b/src/gallium/drivers/llvmpipe/lp_rast_tri.c
index 5ef070a..41f6fbf 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast_tri.c
+++ b/src/gallium/drivers/llvmpipe/lp_rast_tri.c
@@ -35,9 +35,6 @@
 #include "lp_perf.h"
 #include "lp_rast_priv.h"
 
-
-
-
 /**
  * Shade all pixels in a 4x4 block.
  */
@@ -66,44 +63,42 @@ block_full_16(struct lp_rasterizer_task *task,
 	 block_full_4(task, tri, x + ix, y + iy);
 }
 
-#if !defined(PIPE_ARCH_SSE)
-
 static INLINE unsigned
-build_mask_linear(int c, int dcdx, int dcdy)
+build_mask_linear(int64_t c, int64_t dcdx, int64_t dcdy)
 {
-   int mask = 0;
-
-   int c0 = c;
-   int c1 = c0 + dcdy;
-   int c2 = c1 + dcdy;
-   int c3 = c2 + dcdy;
-
-   mask |= ((c0 + 0 * dcdx) >> 31) & (1 << 0);
-   mask |= ((c0 + 1 * dcdx) >> 31) & (1 << 1);
-   mask |= ((c0 + 2 * dcdx) >> 31) & (1 << 2);
-   mask |= ((c0 + 3 * dcdx) >> 31) & (1 << 3);
-   mask |= ((c1 + 0 * dcdx) >> 31) & (1 << 4);
-   mask |= ((c1 + 1 * dcdx) >> 31) & (1 << 5);
-   mask |= ((c1 + 2 * dcdx) >> 31) & (1 << 6);
-   mask |= ((c1 + 3 * dcdx) >> 31) & (1 << 7); 
-   mask |= ((c2 + 0 * dcdx) >> 31) & (1 << 8);
-   mask |= ((c2 + 1 * dcdx) >> 31) & (1 << 9);
-   mask |= ((c2 + 2 * dcdx) >> 31) & (1 << 10);
-   mask |= ((c2 + 3 * dcdx) >> 31) & (1 << 11);
-   mask |= ((c3 + 0 * dcdx) >> 31) & (1 << 12);
-   mask |= ((c3 + 1 * dcdx) >> 31) & (1 << 13);
-   mask |= ((c3 + 2 * dcdx) >> 31) & (1 << 14);
-   mask |= ((c3 + 3 * dcdx) >> 31) & (1 << 15);
+   unsigned mask = 0;
+
+   int64_t c0 = c;
+   int64_t c1 = c0 + dcdy;
+   int64_t c2 = c1 + dcdy;
+   int64_t c3 = c2 + dcdy;
+
+   mask |= ((c0 + 0 * dcdx) >> FIXED_SHIFT) & (1 << 0);
+   mask |= ((c0 + 1 * dcdx) >> FIXED_SHIFT) & (1 << 1);
+   mask |= ((c0 + 2 * dcdx) >> FIXED_SHIFT) & (1 << 2);
+   mask |= ((c0 + 3 * dcdx) >> FIXED_SHIFT) & (1 << 3);
+   mask |= ((c1 + 0 * dcdx) >> FIXED_SHIFT) & (1 << 4);
+   mask |= ((c1 + 1 * dcdx) >> FIXED_SHIFT) & (1 << 5);
+   mask |= ((c1 + 2 * dcdx) >> FIXED_SHIFT) & (1 << 6);
+   mask |= ((c1 + 3 * dcdx) >> FIXED_SHIFT) & (1 << 7);
+   mask |= ((c2 + 0 * dcdx) >> FIXED_SHIFT) & (1 << 8);
+   mask |= ((c2 + 1 * dcdx) >> FIXED_SHIFT) & (1 << 9);
+   mask |= ((c2 + 2 * dcdx) >> FIXED_SHIFT) & (1 << 10);
+   mask |= ((c2 + 3 * dcdx) >> FIXED_SHIFT) & (1 << 11);
+   mask |= ((c3 + 0 * dcdx) >> FIXED_SHIFT) & (1 << 12);
+   mask |= ((c3 + 1 * dcdx) >> FIXED_SHIFT) & (1 << 13);
+   mask |= ((c3 + 2 * dcdx) >> FIXED_SHIFT) & (1 << 14);
+   mask |= ((c3 + 3 * dcdx) >> FIXED_SHIFT) & (1 << 15);
   
    return mask;
 }
 
 
 static INLINE void
-build_masks(int c, 
-	    int cdiff,
-	    int dcdx,
-	    int dcdy,
+build_masks(int64_t c,
+            int64_t cdiff,
+            int64_t dcdx,
+            int64_t dcdy,
 	    unsigned *outmask,
 	    unsigned *partmask)
 {
@@ -122,6 +117,13 @@ lp_rast_triangle_3_16(struct lp_rasterizer_task *task,
 }
 
 void
+lp_rast_triangle_3_4(struct lp_rasterizer_task *task,
+                      const union lp_rast_cmd_arg arg)
+{
+   lp_rast_triangle_3_16(task, arg);
+}
+
+void
 lp_rast_triangle_4_16(struct lp_rasterizer_task *task,
                       const union lp_rast_cmd_arg arg)
 {
@@ -131,11 +133,33 @@ lp_rast_triangle_4_16(struct lp_rasterizer_task *task,
    lp_rast_triangle_4(task, arg2);
 }
 
+#if !defined(PIPE_ARCH_SSE)
+
 void
-lp_rast_triangle_3_4(struct lp_rasterizer_task *task,
+lp_rast_triangle_32_3_16(struct lp_rasterizer_task *task,
+                         const union lp_rast_cmd_arg arg)
+{
+   union lp_rast_cmd_arg arg2;
+   arg2.triangle.tri = arg.triangle.tri;
+   arg2.triangle.plane_mask = (1<<3)-1;
+   lp_rast_triangle_32_3(task, arg2);
+}
+
+void
+lp_rast_triangle_32_4_16(struct lp_rasterizer_task *task,
+                         const union lp_rast_cmd_arg arg)
+{
+   union lp_rast_cmd_arg arg2;
+   arg2.triangle.tri = arg.triangle.tri;
+   arg2.triangle.plane_mask = (1<<4)-1;
+   lp_rast_triangle_32_4(task, arg2);
+}
+
+void
+lp_rast_triangle_32_3_4(struct lp_rasterizer_task *task,
                       const union lp_rast_cmd_arg arg)
 {
-   lp_rast_triangle_3_16(task, arg);
+   lp_rast_triangle_32_3_16(task, arg);
 }
 
 #else
@@ -144,12 +168,12 @@ lp_rast_triangle_3_4(struct lp_rasterizer_task *task,
 
 
 static INLINE void
-build_masks(int c, 
-	    int cdiff,
-	    int dcdx,
-	    int dcdy,
-	    unsigned *outmask,
-	    unsigned *partmask)
+build_masks_32(int c, 
+               int cdiff,
+               int dcdx,
+               int dcdy,
+               unsigned *outmask,
+               unsigned *partmask)
 {
    __m128i cstep0 = _mm_setr_epi32(c, c+dcdx, c+dcdx*2, c+dcdx*3);
    __m128i xdcdy = _mm_set1_epi32(dcdy);
@@ -190,7 +214,7 @@ build_masks(int c,
 
 
 static INLINE unsigned
-build_mask_linear(int c, int dcdx, int dcdy)
+build_mask_linear_32(int c, int dcdx, int dcdy)
 {
    __m128i cstep0 = _mm_setr_epi32(c, c+dcdx, c+dcdx*2, c+dcdx*3);
    __m128i xdcdy = _mm_set1_epi32(dcdy);
@@ -248,7 +272,7 @@ sign_bits4(const __m128i *cstep, int cdiff)
 
 
 void
-lp_rast_triangle_3_16(struct lp_rasterizer_task *task,
+lp_rast_triangle_32_3_16(struct lp_rasterizer_task *task,
                       const union lp_rast_cmd_arg arg)
 {
    const struct lp_rast_triangle *tri = arg.triangle.tri;
@@ -260,9 +284,9 @@ lp_rast_triangle_3_16(struct lp_rasterizer_task *task,
    struct { unsigned mask:16; unsigned i:8; unsigned j:8; } out[16];
    unsigned nr = 0;
 
-   __m128i p0 = _mm_load_si128((__m128i *)&plane[0]); /* c, dcdx, dcdy, eo */
-   __m128i p1 = _mm_load_si128((__m128i *)&plane[1]); /* c, dcdx, dcdy, eo */
-   __m128i p2 = _mm_load_si128((__m128i *)&plane[2]); /* c, dcdx, dcdy, eo */
+   __m128i p0 = lp_plane_to_m128i(&plane[0]); /* c, dcdx, dcdy, eo */
+   __m128i p1 = lp_plane_to_m128i(&plane[1]); /* c, dcdx, dcdy, eo */
+   __m128i p2 = lp_plane_to_m128i(&plane[2]); /* c, dcdx, dcdy, eo */
    __m128i zero = _mm_setzero_si128();
 
    __m128i c;
@@ -362,7 +386,7 @@ lp_rast_triangle_3_16(struct lp_rasterizer_task *task,
 
 
 void
-lp_rast_triangle_3_4(struct lp_rasterizer_task *task,
+lp_rast_triangle_32_3_4(struct lp_rasterizer_task *task,
                      const union lp_rast_cmd_arg arg)
 {
    const struct lp_rast_triangle *tri = arg.triangle.tri;
@@ -370,9 +394,9 @@ lp_rast_triangle_3_4(struct lp_rasterizer_task *task,
    unsigned x = (arg.triangle.plane_mask & 0xff) + task->x;
    unsigned y = (arg.triangle.plane_mask >> 8) + task->y;
 
-   __m128i p0 = _mm_load_si128((__m128i *)&plane[0]); /* c, dcdx, dcdy, eo */
-   __m128i p1 = _mm_load_si128((__m128i *)&plane[1]); /* c, dcdx, dcdy, eo */
-   __m128i p2 = _mm_load_si128((__m128i *)&plane[2]); /* c, dcdx, dcdy, eo */
+   __m128i p0 = lp_plane_to_m128i(&plane[0]); /* c, dcdx, dcdy, eo */
+   __m128i p1 = lp_plane_to_m128i(&plane[1]); /* c, dcdx, dcdy, eo */
+   __m128i p2 = lp_plane_to_m128i(&plane[2]); /* c, dcdx, dcdy, eo */
    __m128i zero = _mm_setzero_si128();
 
    __m128i c;
@@ -450,7 +474,8 @@ lp_rast_triangle_3_4(struct lp_rasterizer_task *task,
 #endif
 
 
-
+#define BUILD_MASKS(c, cdiff, dcdx, dcdy, omask, pmask) build_masks(c, cdiff, dcdx, dcdy, omask, pmask)
+#define BUILD_MASK_LINEAR(c, dcdx, dcdy) build_mask_linear(c, dcdx, dcdy)
 
 #define TAG(x) x##_1
 #define NR_PLANES 1
@@ -468,7 +493,7 @@ lp_rast_triangle_3_4(struct lp_rasterizer_task *task,
 
 #define TAG(x) x##_4
 #define NR_PLANES 4
-#define TRI_16 lp_rast_triangle_4_16
+/*#define TRI_16 lp_rast_triangle_4_16*/
 #include "lp_rast_tri_tmp.h"
 
 #define TAG(x) x##_5
@@ -487,3 +512,47 @@ lp_rast_triangle_3_4(struct lp_rasterizer_task *task,
 #define NR_PLANES 8
 #include "lp_rast_tri_tmp.h"
 
+#ifdef PIPE_ARCH_SSE
+#undef BUILD_MASKS
+#undef BUILD_MASK_LINEAR
+#define BUILD_MASKS(c, cdiff, dcdx, dcdy, omask, pmask) build_masks_32((int)c, (int)cdiff, dcdx, dcdy, omask, pmask)
+#define BUILD_MASK_LINEAR(c, dcdx, dcdy) build_mask_linear_32((int)c, dcdx, dcdy)
+#endif
+
+#define TAG(x) x##_32_1
+#define NR_PLANES 1
+#include "lp_rast_tri_tmp.h"
+
+#define TAG(x) x##_32_2
+#define NR_PLANES 2
+#include "lp_rast_tri_tmp.h"
+
+#define TAG(x) x##_32_3
+#define NR_PLANES 3
+/*#define TRI_4 lp_rast_triangle_3_4*/
+/*#define TRI_16 lp_rast_triangle_3_16*/
+#include "lp_rast_tri_tmp.h"
+
+#define TAG(x) x##_32_4
+#define NR_PLANES 4
+#ifdef PIPE_ARCH_SSE
+#define TRI_16 lp_rast_triangle_32_4_16
+#endif
+#include "lp_rast_tri_tmp.h"
+
+#define TAG(x) x##_32_5
+#define NR_PLANES 5
+#include "lp_rast_tri_tmp.h"
+
+#define TAG(x) x##_32_6
+#define NR_PLANES 6
+#include "lp_rast_tri_tmp.h"
+
+#define TAG(x) x##_32_7
+#define NR_PLANES 7
+#include "lp_rast_tri_tmp.h"
+
+#define TAG(x) x##_32_8
+#define NR_PLANES 8
+#include "lp_rast_tri_tmp.h"
+