Mesa (master): llvmpipe: add rast_tri_4_16 for small lines and points
Keith Whitwell
keithw at kemper.freedesktop.org
Fri Oct 8 16:32:05 UTC 2010
Module: Mesa
Branch: master
Commit: 0ff132e5a633170afaed0aea54d01438c895b8ab
URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=0ff132e5a633170afaed0aea54d01438c895b8ab
Author: Keith Whitwell <keithw at vmware.com>
Date: Fri Oct 8 17:21:03 2010 +0100
llvmpipe: add rast_tri_4_16 for small lines and points
---
src/gallium/drivers/llvmpipe/lp_rast.c | 1 +
src/gallium/drivers/llvmpipe/lp_rast.h | 11 +-
src/gallium/drivers/llvmpipe/lp_rast_debug.c | 1 +
src/gallium/drivers/llvmpipe/lp_rast_priv.h | 4 +
src/gallium/drivers/llvmpipe/lp_rast_tri.c | 152 ++----------------------
src/gallium/drivers/llvmpipe/lp_rast_tri_tmp.h | 127 ++++++++++++++++++++
src/gallium/drivers/llvmpipe/lp_setup_tri.c | 13 ++-
7 files changed, 161 insertions(+), 148 deletions(-)
diff --git a/src/gallium/drivers/llvmpipe/lp_rast.c b/src/gallium/drivers/llvmpipe/lp_rast.c
index 790d88a..db9b2f9 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast.c
+++ b/src/gallium/drivers/llvmpipe/lp_rast.c
@@ -597,6 +597,7 @@ static lp_rast_cmd_func dispatch[LP_RAST_OP_MAX] =
lp_rast_triangle_8,
lp_rast_triangle_3_4,
lp_rast_triangle_3_16,
+ lp_rast_triangle_4_16,
lp_rast_shade_tile,
lp_rast_shade_tile_opaque,
lp_rast_begin_query,
diff --git a/src/gallium/drivers/llvmpipe/lp_rast.h b/src/gallium/drivers/llvmpipe/lp_rast.h
index 0f62377..df0bea0 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast.h
+++ b/src/gallium/drivers/llvmpipe/lp_rast.h
@@ -238,12 +238,13 @@ lp_rast_arg_null( void )
#define LP_RAST_OP_TRIANGLE_8 0x9
#define LP_RAST_OP_TRIANGLE_3_4 0xa
#define LP_RAST_OP_TRIANGLE_3_16 0xb
-#define LP_RAST_OP_SHADE_TILE 0xc
-#define LP_RAST_OP_SHADE_TILE_OPAQUE 0xd
-#define LP_RAST_OP_BEGIN_QUERY 0xe
-#define LP_RAST_OP_END_QUERY 0xf
+#define LP_RAST_OP_TRIANGLE_4_16 0xc
+#define LP_RAST_OP_SHADE_TILE 0xd
+#define LP_RAST_OP_SHADE_TILE_OPAQUE 0xe
+#define LP_RAST_OP_BEGIN_QUERY 0xf
+#define LP_RAST_OP_END_QUERY 0x10
-#define LP_RAST_OP_MAX 0x10
+#define LP_RAST_OP_MAX 0x11
#define LP_RAST_OP_MASK 0xff
void
diff --git a/src/gallium/drivers/llvmpipe/lp_rast_debug.c b/src/gallium/drivers/llvmpipe/lp_rast_debug.c
index 9fc7864..6f4ba1c 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast_debug.c
+++ b/src/gallium/drivers/llvmpipe/lp_rast_debug.c
@@ -42,6 +42,7 @@ static const char *cmd_names[LP_RAST_OP_MAX] =
"triangle_8",
"triangle_3_4",
"triangle_3_16",
+ "triangle_4_16",
"shade_tile",
"shade_tile_opaque",
"begin_query",
diff --git a/src/gallium/drivers/llvmpipe/lp_rast_priv.h b/src/gallium/drivers/llvmpipe/lp_rast_priv.h
index 7370119..104000a 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast_priv.h
+++ b/src/gallium/drivers/llvmpipe/lp_rast_priv.h
@@ -293,6 +293,10 @@ void lp_rast_triangle_3_4(struct lp_rasterizer_task *,
void lp_rast_triangle_3_16( struct lp_rasterizer_task *,
const union lp_rast_cmd_arg );
+
+void lp_rast_triangle_4_16( struct lp_rasterizer_task *,
+ const union lp_rast_cmd_arg );
+
void
lp_debug_bin( const struct cmd_bin *bin );
diff --git a/src/gallium/drivers/llvmpipe/lp_rast_tri.c b/src/gallium/drivers/llvmpipe/lp_rast_tri.c
index a1f309d..f870a18 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast_tri.c
+++ b/src/gallium/drivers/llvmpipe/lp_rast_tri.c
@@ -123,6 +123,16 @@ lp_rast_triangle_3_16(struct lp_rasterizer_task *task,
}
void
+lp_rast_triangle_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_3(task, arg2);
+}
+
+void
lp_rast_triangle_3_4(struct lp_rasterizer_task *task,
const union lp_rast_cmd_arg arg)
{
@@ -229,145 +239,6 @@ sign_bits4(const __m128i *cstep, int cdiff)
return _mm_movemask_epi8(result);
}
-
-/* Special case for 3 plane triangle which is contained entirely
- * within a 16x16 block.
- */
-void
-lp_rast_triangle_3_16(struct lp_rasterizer_task *task,
- const union lp_rast_cmd_arg arg)
-{
- const struct lp_rast_triangle *tri = arg.triangle.tri;
- const struct lp_rast_plane *plane = tri->plane;
- unsigned mask = arg.triangle.plane_mask;
- const int x = task->x + (mask & 0xff);
- const int y = task->y + (mask >> 8);
- unsigned outmask, inmask, partmask, partial_mask;
- unsigned j;
- __m128i cstep4[3][4];
-
- outmask = 0; /* outside one or more trivial reject planes */
- partmask = 0; /* outside one or more trivial accept planes */
-
- for (j = 0; j < 3; j++) {
- const int dcdx = -plane[j].dcdx * 4;
- const int dcdy = plane[j].dcdy * 4;
- __m128i xdcdy = _mm_set1_epi32(dcdy);
-
- cstep4[j][0] = _mm_setr_epi32(0, dcdx, dcdx*2, dcdx*3);
- cstep4[j][1] = _mm_add_epi32(cstep4[j][0], xdcdy);
- cstep4[j][2] = _mm_add_epi32(cstep4[j][1], xdcdy);
- cstep4[j][3] = _mm_add_epi32(cstep4[j][2], xdcdy);
-
- {
- const int c = plane[j].c + plane[j].dcdy * y - plane[j].dcdx * x;
- const int cox = plane[j].eo * 4;
- const int cio = plane[j].ei * 4 - 1;
-
- outmask |= sign_bits4(cstep4[j], c + cox);
- partmask |= sign_bits4(cstep4[j], c + cio);
- }
- }
-
- if (outmask == 0xffff)
- return;
-
- /* Mask of sub-blocks which are inside all trivial accept planes:
- */
- inmask = ~partmask & 0xffff;
-
- /* Mask of sub-blocks which are inside all trivial reject planes,
- * but outside at least one trivial accept plane:
- */
- partial_mask = partmask & ~outmask;
-
- assert((partial_mask & inmask) == 0);
-
- /* Iterate over partials:
- */
- while (partial_mask) {
- int i = ffs(partial_mask) - 1;
- int ix = (i & 3) * 4;
- int iy = (i >> 2) * 4;
- int px = x + ix;
- int py = y + iy;
- unsigned mask = 0xffff;
-
- partial_mask &= ~(1 << i);
-
- for (j = 0; j < 3; j++) {
- const int cx = (plane[j].c
- - plane[j].dcdx * px
- + plane[j].dcdy * py) * 4;
-
- mask &= ~sign_bits4(cstep4[j], cx);
- }
-
- if (mask)
- lp_rast_shade_quads_mask(task, &tri->inputs, px, py, mask);
- }
-
- /* Iterate over fulls:
- */
- while (inmask) {
- int i = ffs(inmask) - 1;
- int ix = (i & 3) * 4;
- int iy = (i >> 2) * 4;
- int px = x + ix;
- int py = y + iy;
-
- inmask &= ~(1 << i);
-
- block_full_4(task, tri, px, py);
- }
-}
-
-
-void
-lp_rast_triangle_3_4(struct lp_rasterizer_task *task,
- const union lp_rast_cmd_arg arg)
-{
- const struct lp_rast_triangle *tri = arg.triangle.tri;
- const struct lp_rast_plane *plane = tri->plane;
- unsigned mask = arg.triangle.plane_mask;
- const int x = task->x + (mask & 0xff);
- const int y = task->y + (mask >> 8);
- unsigned j;
-
- /* Iterate over partials:
- */
- {
- unsigned mask = 0xffff;
-
- for (j = 0; j < 3; j++) {
- const int cx = (plane[j].c
- - plane[j].dcdx * x
- + plane[j].dcdy * y);
-
- const int dcdx = -plane[j].dcdx;
- const int dcdy = plane[j].dcdy;
- __m128i xdcdy = _mm_set1_epi32(dcdy);
-
- __m128i cstep0 = _mm_setr_epi32(cx, cx + dcdx, cx + dcdx*2, cx + dcdx*3);
- __m128i cstep1 = _mm_add_epi32(cstep0, xdcdy);
- __m128i cstep2 = _mm_add_epi32(cstep1, xdcdy);
- __m128i cstep3 = _mm_add_epi32(cstep2, xdcdy);
-
- __m128i cstep01 = _mm_packs_epi32(cstep0, cstep1);
- __m128i cstep23 = _mm_packs_epi32(cstep2, cstep3);
- __m128i result = _mm_packs_epi16(cstep01, cstep23);
-
- /* Extract the sign bits
- */
- mask &= ~_mm_movemask_epi8(result);
- }
-
- if (mask)
- lp_rast_shade_quads_mask(task, &tri->inputs, x, y, mask);
- }
-}
-
-
#endif
@@ -383,10 +254,13 @@ lp_rast_triangle_3_4(struct lp_rasterizer_task *task,
#define TAG(x) x##_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##_4
#define NR_PLANES 4
+#define TRI_16 lp_rast_triangle_4_16
#include "lp_rast_tri_tmp.h"
#define TAG(x) x##_5
diff --git a/src/gallium/drivers/llvmpipe/lp_rast_tri_tmp.h b/src/gallium/drivers/llvmpipe/lp_rast_tri_tmp.h
index 9830a43..c8f9956 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast_tri_tmp.h
+++ b/src/gallium/drivers/llvmpipe/lp_rast_tri_tmp.h
@@ -245,6 +245,133 @@ TAG(lp_rast_triangle)(struct lp_rasterizer_task *task,
}
}
+#if defined(PIPE_ARCH_SSE) && defined(TRI_16)
+/* XXX: special case this when intersection is not required.
+ * - tile completely within bbox,
+ * - bbox completely within tile.
+ */
+void
+TRI_16(struct lp_rasterizer_task *task,
+ const union lp_rast_cmd_arg arg)
+{
+ const struct lp_rast_triangle *tri = arg.triangle.tri;
+ const struct lp_rast_plane *plane = tri->plane;
+ unsigned mask = arg.triangle.plane_mask;
+ unsigned outmask, partial_mask;
+ unsigned j;
+ __m128i cstep4[NR_PLANES][4];
+
+ int x = (mask & 0xff);
+ int y = (mask >> 8);
+
+ outmask = 0; /* outside one or more trivial reject planes */
+
+ x += task->x;
+ y += task->y;
+
+ for (j = 0; j < NR_PLANES; j++) {
+ const int dcdx = -plane[j].dcdx * 4;
+ const int dcdy = plane[j].dcdy * 4;
+ __m128i xdcdy = _mm_set1_epi32(dcdy);
+
+ cstep4[j][0] = _mm_setr_epi32(0, dcdx, dcdx*2, dcdx*3);
+ cstep4[j][1] = _mm_add_epi32(cstep4[j][0], xdcdy);
+ cstep4[j][2] = _mm_add_epi32(cstep4[j][1], xdcdy);
+ cstep4[j][3] = _mm_add_epi32(cstep4[j][2], xdcdy);
+
+ {
+ const int c = plane[j].c + plane[j].dcdy * y - plane[j].dcdx * x;
+ const int cox = plane[j].eo * 4;
+
+ outmask |= sign_bits4(cstep4[j], c + cox);
+ }
+ }
+
+ if (outmask == 0xffff)
+ return;
+
+
+ /* Mask of sub-blocks which are inside all trivial reject planes,
+ * but outside at least one trivial accept plane:
+ */
+ partial_mask = 0xffff & ~outmask;
+
+ /* Iterate over partials:
+ */
+ while (partial_mask) {
+ int i = ffs(partial_mask) - 1;
+ int ix = (i & 3) * 4;
+ int iy = (i >> 2) * 4;
+ int px = x + ix;
+ int py = y + iy;
+ unsigned mask = 0xffff;
+
+ partial_mask &= ~(1 << i);
+
+ for (j = 0; j < NR_PLANES; j++) {
+ const int cx = (plane[j].c
+ - plane[j].dcdx * px
+ + plane[j].dcdy * py) * 4;
+
+ mask &= ~sign_bits4(cstep4[j], cx);
+ }
+
+ if (mask)
+ lp_rast_shade_quads_mask(task, &tri->inputs, px, py, mask);
+ }
+}
+#endif
+
+#if defined(PIPE_ARCH_SSE) && defined(TRI_4)
+void
+TRI_4(struct lp_rasterizer_task *task,
+ const union lp_rast_cmd_arg arg)
+{
+ const struct lp_rast_triangle *tri = arg.triangle.tri;
+ const struct lp_rast_plane *plane = tri->plane;
+ unsigned mask = arg.triangle.plane_mask;
+ const int x = task->x + (mask & 0xff);
+ const int y = task->y + (mask >> 8);
+ unsigned j;
+
+ /* Iterate over partials:
+ */
+ {
+ unsigned mask = 0xffff;
+
+ for (j = 0; j < NR_PLANES; j++) {
+ const int cx = (plane[j].c
+ - plane[j].dcdx * x
+ + plane[j].dcdy * y);
+
+ const int dcdx = -plane[j].dcdx;
+ const int dcdy = plane[j].dcdy;
+ __m128i xdcdy = _mm_set1_epi32(dcdy);
+
+ __m128i cstep0 = _mm_setr_epi32(cx, cx + dcdx, cx + dcdx*2, cx + dcdx*3);
+ __m128i cstep1 = _mm_add_epi32(cstep0, xdcdy);
+ __m128i cstep2 = _mm_add_epi32(cstep1, xdcdy);
+ __m128i cstep3 = _mm_add_epi32(cstep2, xdcdy);
+
+ __m128i cstep01 = _mm_packs_epi32(cstep0, cstep1);
+ __m128i cstep23 = _mm_packs_epi32(cstep2, cstep3);
+ __m128i result = _mm_packs_epi16(cstep01, cstep23);
+
+ /* Extract the sign bits
+ */
+ mask &= ~_mm_movemask_epi8(result);
+ }
+
+ if (mask)
+ lp_rast_shade_quads_mask(task, &tri->inputs, x, y, mask);
+ }
+}
+#endif
+
+
+
#undef TAG
+#undef TRI_4
+#undef TRI_16
#undef NR_PLANES
diff --git a/src/gallium/drivers/llvmpipe/lp_setup_tri.c b/src/gallium/drivers/llvmpipe/lp_setup_tri.c
index 9f87101..8fd0346 100644
--- a/src/gallium/drivers/llvmpipe/lp_setup_tri.c
+++ b/src/gallium/drivers/llvmpipe/lp_setup_tri.c
@@ -479,15 +479,14 @@ lp_setup_bin_triangle( struct lp_setup_context *setup,
{
int ix0 = bbox->x0 / TILE_SIZE;
int iy0 = bbox->y0 / TILE_SIZE;
+ int px = bbox->x0 & 63 & ~3;
+ int py = bbox->y0 & 63 & ~3;
+ int mask = px | (py << 8);
assert(iy0 == bbox->y1 / TILE_SIZE &&
ix0 == bbox->x1 / TILE_SIZE);
if (nr_planes == 3) {
- int px = bbox->x0 & 63 & ~3;
- int py = bbox->y0 & 63 & ~3;
- int mask = px | (py << 8);
-
if (sz < 4)
{
/* Triangle is contained in a single 4x4 stamp:
@@ -507,6 +506,12 @@ lp_setup_bin_triangle( struct lp_setup_context *setup,
lp_rast_arg_triangle(tri, mask) );
}
}
+ else if (nr_planes == 4 && sz < 16)
+ {
+ return lp_scene_bin_command( scene, ix0, iy0,
+ LP_RAST_OP_TRIANGLE_4_16,
+ lp_rast_arg_triangle(tri, mask) );
+ }
/* Triangle is contained in a single tile:
More information about the mesa-commit
mailing list