diff options
Diffstat (limited to 'src/gallium/drivers/llvmpipe/lp_rast_tri.c')
| -rw-r--r-- | src/gallium/drivers/llvmpipe/lp_rast_tri.c | 241 |
1 files changed, 134 insertions, 107 deletions
diff --git a/src/gallium/drivers/llvmpipe/lp_rast_tri.c b/src/gallium/drivers/llvmpipe/lp_rast_tri.c index dbaa8e023a..a1f309d4b0 100644 --- a/src/gallium/drivers/llvmpipe/lp_rast_tri.c +++ b/src/gallium/drivers/llvmpipe/lp_rast_tri.c @@ -68,36 +68,6 @@ block_full_16(struct lp_rasterizer_task *task, } #if !defined(PIPE_ARCH_SSE) -static INLINE unsigned -build_mask(int c, int dcdx, int dcdy) -{ - int mask = 0; - - int c0 = c; - int c1 = c0 + dcdx; - int c2 = c1 + dcdx; - int c3 = c2 + dcdx; - - mask |= ((c0 + 0 * dcdy) >> 31) & (1 << 0); - mask |= ((c0 + 1 * dcdy) >> 31) & (1 << 2); - mask |= ((c0 + 2 * dcdy) >> 31) & (1 << 8); - mask |= ((c0 + 3 * dcdy) >> 31) & (1 << 10); - mask |= ((c1 + 0 * dcdy) >> 31) & (1 << 1); - mask |= ((c1 + 1 * dcdy) >> 31) & (1 << 3); - mask |= ((c1 + 2 * dcdy) >> 31) & (1 << 9); - mask |= ((c1 + 3 * dcdy) >> 31) & (1 << 11); - mask |= ((c2 + 0 * dcdy) >> 31) & (1 << 4); - mask |= ((c2 + 1 * dcdy) >> 31) & (1 << 6); - mask |= ((c2 + 2 * dcdy) >> 31) & (1 << 12); - mask |= ((c2 + 3 * dcdy) >> 31) & (1 << 14); - mask |= ((c3 + 0 * dcdy) >> 31) & (1 << 5); - mask |= ((c3 + 1 * dcdy) >> 31) & (1 << 7); - mask |= ((c3 + 2 * dcdy) >> 31) & (1 << 13); - mask |= ((c3 + 3 * dcdy) >> 31) & (1 << 15); - - return mask; -} - static INLINE unsigned build_mask_linear(int c, int dcdx, int dcdy) @@ -142,6 +112,23 @@ build_masks(int c, *partmask |= build_mask_linear(c + cdiff, dcdx, dcdy); } +void +lp_rast_triangle_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_3(task, arg2); +} + +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); +} + #else #include <emmintrin.h> #include "util/u_sse.h" @@ -220,79 +207,28 @@ build_mask_linear(int c, int dcdx, int dcdy) } static INLINE unsigned -build_mask(int c, int dcdx, int dcdy) +sign_bits4(const __m128i *cstep, int cdiff) { - __m128i step = _mm_setr_epi32(0, dcdx, dcdy, dcdx + dcdy); - __m128i c0 = _mm_set1_epi32(c); - - /* Get values across the quad - */ - __m128i cstep0 = _mm_add_epi32(c0, step); - - /* Scale up step for moving between quads. - */ - __m128i step4 = _mm_add_epi32(step, step); - /* Get values for the remaining quads: + /* Adjust the step values */ - __m128i cstep1 = _mm_add_epi32(cstep0, - _mm_shuffle_epi32(step4, _MM_SHUFFLE(1,1,1,1))); - __m128i cstep2 = _mm_add_epi32(cstep0, - _mm_shuffle_epi32(step4, _MM_SHUFFLE(2,2,2,2))); - __m128i cstep3 = _mm_add_epi32(cstep2, - _mm_shuffle_epi32(step4, _MM_SHUFFLE(1,1,1,1))); + __m128i cio4 = _mm_set1_epi32(cdiff); + __m128i cstep0 = _mm_add_epi32(cstep[0], cio4); + __m128i cstep1 = _mm_add_epi32(cstep[1], cio4); + __m128i cstep2 = _mm_add_epi32(cstep[2], cio4); + __m128i cstep3 = _mm_add_epi32(cstep[3], cio4); - /* pack pairs of results into epi16 + /* Pack down to epi8 */ __m128i cstep01 = _mm_packs_epi32(cstep0, cstep1); __m128i cstep23 = _mm_packs_epi32(cstep2, cstep3); - - /* pack into epi8, preserving sign bits - */ __m128i result = _mm_packs_epi16(cstep01, cstep23); - /* extract sign bits to create mask + /* Extract the sign bits */ return _mm_movemask_epi8(result); } -#endif - - - - -#define TAG(x) x##_1 -#define NR_PLANES 1 -#include "lp_rast_tri_tmp.h" - -#define TAG(x) x##_2 -#define NR_PLANES 2 -#include "lp_rast_tri_tmp.h" - -#define TAG(x) x##_3 -#define NR_PLANES 3 -#include "lp_rast_tri_tmp.h" - -#define TAG(x) x##_4 -#define NR_PLANES 4 -#include "lp_rast_tri_tmp.h" - -#define TAG(x) x##_5 -#define NR_PLANES 5 -#include "lp_rast_tri_tmp.h" - -#define TAG(x) x##_6 -#define NR_PLANES 6 -#include "lp_rast_tri_tmp.h" - -#define TAG(x) x##_7 -#define NR_PLANES 7 -#include "lp_rast_tri_tmp.h" - -#define TAG(x) x##_8 -#define NR_PLANES 8 -#include "lp_rast_tri_tmp.h" - /* Special case for 3 plane triangle which is contained entirely * within a 16x16 block. @@ -304,29 +240,32 @@ lp_rast_triangle_3_16(struct lp_rasterizer_task *task, 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 & 0xf) * 16; - const int y = task->y + (mask >> 4) * 16; + const int x = task->x + (mask & 0xff); + const int y = task->y + (mask >> 8); unsigned outmask, inmask, partmask, partial_mask; unsigned j; - int c[3]; + __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++) { - c[j] = plane[j].c + plane[j].dcdy * y - plane[j].dcdx * x; + 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 dcdx = -plane[j].dcdx * 4; - const int dcdy = plane[j].dcdy * 4; + 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; - build_masks(c[j] + cox, - cio - cox, - dcdx, dcdy, - &outmask, /* sign bits from c[i][0..15] + cox */ - &partmask); /* sign bits from c[i][0..15] + cio */ + outmask |= sign_bits4(cstep4[j], c + cox); + partmask |= sign_bits4(cstep4[j], c + cio); } } @@ -352,16 +291,20 @@ lp_rast_triangle_3_16(struct lp_rasterizer_task *task, int iy = (i >> 2) * 4; int px = x + ix; int py = y + iy; - int cx[3]; + unsigned mask = 0xffff; partial_mask &= ~(1 << i); - for (j = 0; j < 3; j++) - cx[j] = (c[j] - - plane[j].dcdx * ix - + plane[j].dcdy * iy); + 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); + } - do_block_4_3(task, tri, plane, px, py, cx); + if (mask) + lp_rast_shade_quads_mask(task, &tri->inputs, px, py, mask); } /* Iterate over fulls: @@ -378,3 +321,87 @@ lp_rast_triangle_3_16(struct lp_rasterizer_task *task, 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 + + + + +#define TAG(x) x##_1 +#define NR_PLANES 1 +#include "lp_rast_tri_tmp.h" + +#define TAG(x) x##_2 +#define NR_PLANES 2 +#include "lp_rast_tri_tmp.h" + +#define TAG(x) x##_3 +#define NR_PLANES 3 +#include "lp_rast_tri_tmp.h" + +#define TAG(x) x##_4 +#define NR_PLANES 4 +#include "lp_rast_tri_tmp.h" + +#define TAG(x) x##_5 +#define NR_PLANES 5 +#include "lp_rast_tri_tmp.h" + +#define TAG(x) x##_6 +#define NR_PLANES 6 +#include "lp_rast_tri_tmp.h" + +#define TAG(x) x##_7 +#define NR_PLANES 7 +#include "lp_rast_tri_tmp.h" + +#define TAG(x) x##_8 +#define NR_PLANES 8 +#include "lp_rast_tri_tmp.h" + |