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
-/*#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
-#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"
-
-
-static INLINE void
-transpose4_epi32(const __m128i * restrict a,
- const __m128i * restrict b,
- const __m128i * restrict c,
- const __m128i * restrict d,
- __m128i * restrict o,
- __m128i * restrict p,
- __m128i * restrict q,
- __m128i * restrict r)
-{
- __m128i t0 = _mm_unpacklo_epi32(*a, *b);
- __m128i t1 = _mm_unpacklo_epi32(*c, *d);
- __m128i t2 = _mm_unpackhi_epi32(*a, *b);
- __m128i t3 = _mm_unpackhi_epi32(*c, *d);
-
- *o = _mm_unpacklo_epi64(t0, t1);
- *p = _mm_unpackhi_epi64(t0, t1);
- *q = _mm_unpacklo_epi64(t2, t3);
- *r = _mm_unpackhi_epi64(t2, t3);
-}
-
-
-#define SCALAR_EPI32(m, i) _mm_shuffle_epi32((m), _MM_SHUFFLE(i,i,i,i))
-
-#define NR_PLANES 3
-
-
-
-/* Provide an SSE2 implementation of _mm_mullo_epi32() in terms of
- * _mm_mul_epu32().
- *
- * I suspect this works fine for us because one of our operands is
- * always positive, but not sure that this can be used for general
- * signed integer multiplication.
- *
- * This seems close enough to the speed of SSE4 and the real
- * _mm_mullo_epi32() intrinsic as to not justify adding an sse4
- * dependency at this point.
- */
-static INLINE __m128i mm_mullo_epi32(const __m128i a, const __m128i b)
-{
- __m128i a4 = _mm_srli_epi64(a, 32); /* shift by one dword */
- __m128i b4 = _mm_srli_epi64(b, 32); /* shift by one dword */
- __m128i ba = _mm_mul_epu32(b, a); /* multply dwords 0, 2 */
- __m128i b4a4 = _mm_mul_epu32(b4, a4); /* multiply dwords 1, 3 */
-
- /* Interleave the results, either with shuffles or (slightly
- * faster) direct bit operations:
- */
-#if 0
- __m128i ba8 = _mm_shuffle_epi32(ba, 8);
- __m128i b4a48 = _mm_shuffle_epi32(b4a4, 8);
- __m128i result = _mm_unpacklo_epi32(ba8, b4a48);
-#else
- __m128i mask = _mm_setr_epi32(~0,0,~0,0);
- __m128i ba_mask = _mm_and_si128(ba, mask);
- __m128i b4a4_mask_shift = _mm_slli_epi64(b4a4, 32);
- __m128i result = _mm_or_si128(ba_mask, b4a4_mask_shift);
-#endif
-
- return result;
-}
const union lp_rast_cmd_arg arg)
{
const struct lp_rast_triangle *tri = arg.triangle.tri;
- const struct lp_rast_plane *plane = tri->plane;
+ const struct lp_rast_plane *plane = GET_PLANES(tri);
int x = (arg.triangle.plane_mask & 0xff) + task->x;
int y = (arg.triangle.plane_mask >> 8) + task->y;
unsigned i, j;
struct { unsigned mask:16; unsigned i:8; unsigned j:8; } out[16];
unsigned nr = 0;
- __m128i p0 = _mm_loadu_si128((__m128i *)&plane[0]); /* c, dcdx, dcdy, eo */
- __m128i p1 = _mm_loadu_si128((__m128i *)&plane[1]); /* c, dcdx, dcdy, eo */
- __m128i p2 = _mm_loadu_si128((__m128i *)&plane[2]); /* c, dcdx, dcdy, eo */
+ __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 zero = _mm_setzero_si128();
__m128i c;
const union lp_rast_cmd_arg arg)
{
const struct lp_rast_triangle *tri = arg.triangle.tri;
- const struct lp_rast_plane *plane = tri->plane;
+ const struct lp_rast_plane *plane = GET_PLANES(tri);
int x = (arg.triangle.plane_mask & 0xff) + task->x;
int y = (arg.triangle.plane_mask >> 8) + task->y;
- __m128i p0 = _mm_loadu_si128((__m128i *)&plane[0]); /* c, dcdx, dcdy, eo */
- __m128i p1 = _mm_loadu_si128((__m128i *)&plane[1]); /* c, dcdx, dcdy, eo */
- __m128i p2 = _mm_loadu_si128((__m128i *)&plane[2]); /* c, dcdx, dcdy, eo */
+ __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 zero = _mm_setzero_si128();
__m128i c;
0xffff & ~mask);
}
}
+
+#undef NR_PLANES
+#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
+/*#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
+#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"
+
projects / mesa.git / blobdiff