}
static inline unsigned
-build_mask_linear(int64_t c, int64_t dcdx, int64_t dcdy)
+build_mask_linear(int32_t c, int32_t dcdx, int32_t dcdy)
{
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);
+ int32_t c0 = c;
+ int32_t c1 = c0 + dcdy;
+ int32_t c2 = c1 + dcdy;
+ int32_t 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);
return mask;
}
static inline void
-build_masks(int64_t c,
- int64_t cdiff,
- int64_t dcdx,
- int64_t dcdy,
- unsigned *outmask,
- unsigned *partmask)
+build_masks(int32_t c,
+ int32_t cdiff,
+ int32_t dcdx,
+ int32_t dcdy,
+ unsigned *outmask,
+ unsigned *partmask)
{
*outmask |= build_mask_linear(c, dcdx, dcdy);
*partmask |= build_mask_linear(c + cdiff, dcdx, dcdy);
static inline void
-build_masks_32(int c,
- int cdiff,
- int dcdx,
- int dcdy,
- unsigned *outmask,
- unsigned *partmask)
+build_masks_sse(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);
static inline unsigned
-build_mask_linear_32(int c, int dcdx, int dcdy)
+build_mask_linear_sse(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);
void
lp_rast_triangle_32_3_16(struct lp_rasterizer_task *task,
- const union lp_rast_cmd_arg arg)
+ const union lp_rast_cmd_arg arg)
{
const struct lp_rast_triangle *tri = arg.triangle.tri;
const struct lp_rast_plane *plane = GET_PLANES(tri);
struct { unsigned mask:16; unsigned i:8; unsigned j:8; } out[16];
unsigned nr = 0;
- __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 */
+ /* p0 and p2 are aligned, p1 is not (plane size 24 bytes). */
+ __m128i p0 = _mm_load_si128((__m128i *)&plane[0]); /* clo, chi, dcdx, dcdy */
+ __m128i p1 = _mm_loadu_si128((__m128i *)&plane[1]);
+ __m128i p2 = _mm_load_si128((__m128i *)&plane[2]);
__m128i zero = _mm_setzero_si128();
- __m128i c;
- __m128i dcdx;
- __m128i dcdy;
- __m128i rej4;
-
- __m128i dcdx2;
- __m128i dcdx3;
+ __m128i c, dcdx, dcdy, rej4;
+ __m128i dcdx_neg_mask, dcdy_neg_mask;
+ __m128i dcdx2, dcdx3;
__m128i span_0; /* 0,dcdx,2dcdx,3dcdx for plane 0 */
__m128i span_1; /* 0,dcdx,2dcdx,3dcdx for plane 1 */
__m128i span_2; /* 0,dcdx,2dcdx,3dcdx for plane 2 */
__m128i unused;
-
+
transpose4_epi32(&p0, &p1, &p2, &zero,
- &c, &dcdx, &dcdy, &rej4);
+ &c, &unused, &dcdx, &dcdy);
+
+ /* recalc eo - easier than trying to load as scalars / shuffle... */
+ dcdx_neg_mask = _mm_srai_epi32(dcdx, 31);
+ dcdy_neg_mask = _mm_srai_epi32(dcdy, 31);
+ rej4 = _mm_sub_epi32(_mm_andnot_si128(dcdy_neg_mask, dcdy),
+ _mm_and_si128(dcdx_neg_mask, dcdx));
/* Adjust dcdx;
*/
void
lp_rast_triangle_32_3_4(struct lp_rasterizer_task *task,
- const union lp_rast_cmd_arg arg)
+ const union lp_rast_cmd_arg arg)
{
const struct lp_rast_triangle *tri = arg.triangle.tri;
const struct lp_rast_plane *plane = GET_PLANES(tri);
unsigned x = (arg.triangle.plane_mask & 0xff) + task->x;
unsigned y = (arg.triangle.plane_mask >> 8) + task->y;
- __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 */
+ /* p0 and p2 are aligned, p1 is not (plane size 24 bytes). */
+ __m128i p0 = _mm_load_si128((__m128i *)&plane[0]); /* clo, chi, dcdx, dcdy */
+ __m128i p1 = _mm_loadu_si128((__m128i *)&plane[1]);
+ __m128i p2 = _mm_load_si128((__m128i *)&plane[2]);
__m128i zero = _mm_setzero_si128();
- __m128i c;
- __m128i dcdx;
- __m128i dcdy;
+ __m128i c, dcdx, dcdy;
+ __m128i dcdx2, dcdx3;
- __m128i dcdx2;
- __m128i dcdx3;
-
__m128i span_0; /* 0,dcdx,2dcdx,3dcdx for plane 0 */
__m128i span_1; /* 0,dcdx,2dcdx,3dcdx for plane 1 */
__m128i span_2; /* 0,dcdx,2dcdx,3dcdx for plane 2 */
__m128i unused;
transpose4_epi32(&p0, &p1, &p2, &zero,
- &c, &dcdx, &dcdy, &unused);
+ &c, &unused, &dcdx, &dcdy);
/* Adjust dcdx;
*/
#include "util/u_pwr8.h"
static inline void
-build_masks_32(int c,
- int cdiff,
- int dcdx,
- int dcdy,
- unsigned *outmask,
- unsigned *partmask)
+build_masks_ppc(int c,
+ int cdiff,
+ int dcdx,
+ int dcdy,
+ unsigned *outmask,
+ unsigned *partmask)
{
__m128i cstep0 = vec_setr_epi32(c, c+dcdx, c+dcdx*2, c+dcdx*3);
__m128i xdcdy = (__m128i) vec_splats(dcdy);
}
static inline unsigned
-build_mask_linear_32(int c, int dcdx, int dcdy)
+build_mask_linear_ppc(int c, int dcdx, int dcdy)
{
__m128i cstep0 = vec_setr_epi32(c, c+dcdx, c+dcdx*2, c+dcdx*3);
__m128i xdcdy = (__m128i) vec_splats(dcdy);
#endif
+#if defined PIPE_ARCH_SSE
+#define BUILD_MASKS(c, cdiff, dcdx, dcdy, omask, pmask) build_masks_sse((int)c, (int)cdiff, dcdx, dcdy, omask, pmask)
+#define BUILD_MASK_LINEAR(c, dcdx, dcdy) build_mask_linear_sse((int)c, dcdx, dcdy)
+#elif (defined(_ARCH_PWR8) && defined(PIPE_ARCH_LITTLE_ENDIAN))
+#define BUILD_MASKS(c, cdiff, dcdx, dcdy, omask, pmask) build_masks_ppc((int)c, (int)cdiff, dcdx, dcdy, omask, pmask)
+#define BUILD_MASK_LINEAR(c, dcdx, dcdy) build_mask_linear_ppc((int)c, dcdx, dcdy)
+#else
#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)
+#endif
+
+#define RASTER_64 1
#define TAG(x) x##_1
#define NR_PLANES 1
#define NR_PLANES 8
#include "lp_rast_tri_tmp.h"
-#if defined(PIPE_ARCH_SSE) || (defined(_ARCH_PWR8) && defined(PIPE_ARCH_LITTLE_ENDIAN))
-#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
+#undef RASTER_64
#define TAG(x) x##_32_1
#define NR_PLANES 1