/**************************************************************************
*
- * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * Copyright 2007 VMware, Inc.
* All Rights Reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
* OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
* ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
#if defined(PIPE_ARCH_SSE)
#include <emmintrin.h>
+#elif defined(_ARCH_PWR8) && UTIL_ARCH_LITTLE_ENDIAN
+#include <altivec.h>
+#include "util/u_pwr8.h"
#endif
-static INLINE int
+#if !defined(PIPE_ARCH_SSE)
+
+static inline int
subpixel_snap(float a)
{
return util_iround(FIXED_ONE * a);
}
-static INLINE float
-fixed_to_float(int a)
-{
- return a * (1.0f / FIXED_ONE);
-}
-
+#endif
/* Position and area in fixed point coordinates */
struct fixed_position {
int32_t x[4];
int32_t y[4];
- int64_t area;
int32_t dx01;
int32_t dy01;
int32_t dx20;
int32_t dy20;
+ int64_t area;
};
unsigned plane_sz = nr_planes * sizeof(struct lp_rast_plane);
struct lp_rast_triangle *tri;
+ STATIC_ASSERT(sizeof(struct lp_rast_plane) % 8 == 0);
+
*tri_size = (sizeof(struct lp_rast_triangle) +
3 * input_array_sz +
plane_sz);
tri = lp_scene_alloc_aligned( scene, *tri_size, 16 );
- if (tri == NULL)
+ if (!tri)
return NULL;
tri->inputs.stride = input_array_sz;
const struct lp_setup_variant_key *key = &setup->setup.variant->key;
struct lp_rast_triangle *tri;
struct lp_rast_plane *plane;
- struct u_rect bbox;
+ const struct u_rect *scissor = NULL;
+ struct u_rect bbox, bboxpos;
+ boolean s_planes[4];
unsigned tri_bytes;
int nr_planes = 3;
unsigned viewport_index = 0;
unsigned layer = 0;
+ const float (*pv)[4];
/* Area should always be positive here */
assert(position->area > 0);
if (0)
lp_setup_print_triangle(setup, v0, v1, v2);
- if (setup->scissor_test) {
- nr_planes = 7;
- if (setup->viewport_index_slot > 0) {
- unsigned *udata = (unsigned*)v0[setup->viewport_index_slot];
- viewport_index = lp_clamp_viewport_idx(*udata);
- }
+ if (setup->flatshade_first) {
+ pv = v0;
}
else {
- nr_planes = 3;
+ pv = v2;
+ }
+ if (setup->viewport_index_slot > 0) {
+ unsigned *udata = (unsigned*)pv[setup->viewport_index_slot];
+ viewport_index = lp_clamp_viewport_idx(*udata);
}
if (setup->layer_slot > 0) {
- layer = *(unsigned*)v1[setup->layer_slot];
+ layer = *(unsigned*)pv[setup->layer_slot];
layer = MIN2(layer, scene->fb_max_layer);
}
* up needing a bottom-left fill convention, which requires
* slightly different rounding.
*/
- int adj = (setup->pixel_offset != 0) ? 1 : 0;
+ int adj = (setup->bottom_edge_rule != 0) ? 1 : 0;
/* Inclusive x0, exclusive x1 */
bbox.x0 = MIN3(position->x[0], position->x[1], position->x[2]) >> FIXED_ORDER;
return TRUE;
}
+ bboxpos = bbox;
+
/* Can safely discard negative regions, but need to keep hold of
* information about when the triangle extends past screen
* boundaries. See trimmed_box in lp_setup_bin_triangle().
*/
- bbox.x0 = MAX2(bbox.x0, 0);
- bbox.y0 = MAX2(bbox.y0, 0);
+ bboxpos.x0 = MAX2(bboxpos.x0, 0);
+ bboxpos.y0 = MAX2(bboxpos.y0, 0);
+
+ nr_planes = 3;
+ /*
+ * Determine how many scissor planes we need, that is drop scissor
+ * edges if the bounding box of the tri is fully inside that edge.
+ */
+ if (setup->scissor_test) {
+ /* why not just use draw_regions */
+ scissor = &setup->scissors[viewport_index];
+ scissor_planes_needed(s_planes, &bboxpos, scissor);
+ nr_planes += s_planes[0] + s_planes[1] + s_planes[2] + s_planes[3];
+ }
tri = lp_setup_alloc_triangle(scene,
key->num_inputs,
if (!tri)
return FALSE;
-#if 0
+#ifdef DEBUG
tri->v[0][0] = v0[0][0];
tri->v[1][0] = v1[0][0];
tri->v[2][0] = v2[0][0];
/* Setup parameter interpolants:
*/
- setup->setup.variant->jit_function( v0,
- v1,
- v2,
- frontfacing,
- GET_A0(&tri->inputs),
- GET_DADX(&tri->inputs),
- GET_DADY(&tri->inputs) );
+ setup->setup.variant->jit_function(v0, v1, v2,
+ frontfacing,
+ GET_A0(&tri->inputs),
+ GET_DADX(&tri->inputs),
+ GET_DADY(&tri->inputs));
tri->inputs.frontfacing = frontfacing;
tri->inputs.disable = FALSE;
if (0)
lp_dump_setup_coef(&setup->setup.variant->key,
- (const float (*)[4])GET_A0(&tri->inputs),
- (const float (*)[4])GET_DADX(&tri->inputs),
- (const float (*)[4])GET_DADY(&tri->inputs));
+ (const float (*)[4])GET_A0(&tri->inputs),
+ (const float (*)[4])GET_DADX(&tri->inputs),
+ (const float (*)[4])GET_DADY(&tri->inputs));
plane = GET_PLANES(tri);
#if defined(PIPE_ARCH_SSE)
- if (setup->fb.width <= MAX_FIXED_LENGTH32 &&
- setup->fb.height <= MAX_FIXED_LENGTH32 &&
- (bbox.x1 - bbox.x0) <= MAX_FIXED_LENGTH32 &&
- (bbox.y1 - bbox.y0) <= MAX_FIXED_LENGTH32) {
+ if (1) {
__m128i vertx, verty;
__m128i shufx, shufy;
- __m128i dcdx, dcdy, c;
- __m128i unused;
+ __m128i dcdx, dcdy;
+ __m128i cdx02, cdx13, cdy02, cdy13, c02, c13;
+ __m128i c01, c23, unused;
__m128i dcdx_neg_mask;
__m128i dcdy_neg_mask;
__m128i dcdx_zero_mask;
- __m128i top_left_flag;
- __m128i c_inc_mask, c_inc;
+ __m128i top_left_flag, c_dec;
__m128i eo, p0, p1, p2;
__m128i zero = _mm_setzero_si128();
- PIPE_ALIGN_VAR(16) int32_t temp_vec[4];
- vertx = _mm_loadu_si128((__m128i *)position->x); /* vertex x coords */
- verty = _mm_loadu_si128((__m128i *)position->y); /* vertex y coords */
+ vertx = _mm_load_si128((__m128i *)position->x); /* vertex x coords */
+ verty = _mm_load_si128((__m128i *)position->y); /* vertex y coords */
shufx = _mm_shuffle_epi32(vertx, _MM_SHUFFLE(3,0,2,1));
shufy = _mm_shuffle_epi32(verty, _MM_SHUFFLE(3,0,2,1));
top_left_flag = _mm_set1_epi32((setup->bottom_edge_rule == 0) ? ~0 : 0);
- c_inc_mask = _mm_or_si128(dcdx_neg_mask,
- _mm_and_si128(dcdx_zero_mask,
- _mm_xor_si128(dcdy_neg_mask,
- top_left_flag)));
-
- c_inc = _mm_srli_epi32(c_inc_mask, 31);
-
- c = _mm_sub_epi32(mm_mullo_epi32(dcdx, vertx),
- mm_mullo_epi32(dcdy, verty));
+ c_dec = _mm_or_si128(dcdx_neg_mask,
+ _mm_and_si128(dcdx_zero_mask,
+ _mm_xor_si128(dcdy_neg_mask,
+ top_left_flag)));
- c = _mm_add_epi32(c, c_inc);
+ /*
+ * 64 bit arithmetic.
+ * Note we need _signed_ mul (_mm_mul_epi32) which we emulate.
+ */
+ cdx02 = mm_mullohi_epi32(dcdx, vertx, &cdx13);
+ cdy02 = mm_mullohi_epi32(dcdy, verty, &cdy13);
+ c02 = _mm_sub_epi64(cdx02, cdy02);
+ c13 = _mm_sub_epi64(cdx13, cdy13);
+ c02 = _mm_sub_epi64(c02, _mm_shuffle_epi32(c_dec,
+ _MM_SHUFFLE(2,2,0,0)));
+ c13 = _mm_sub_epi64(c13, _mm_shuffle_epi32(c_dec,
+ _MM_SHUFFLE(3,3,1,1)));
+
+ /*
+ * Useful for very small fbs/tris (or fewer subpixel bits) only:
+ * c = _mm_sub_epi32(mm_mullo_epi32(dcdx, vertx),
+ * mm_mullo_epi32(dcdy, verty));
+ *
+ * c = _mm_sub_epi32(c, c_dec);
+ */
/* Scale up to match c:
*/
dcdx = _mm_slli_epi32(dcdx, FIXED_ORDER);
dcdy = _mm_slli_epi32(dcdy, FIXED_ORDER);
- /* Calculate trivial reject values:
+ /*
+ * Calculate trivial reject values:
+ * Note eo cannot overflow even if dcdx/dcdy would already have
+ * 31 bits (which they shouldn't have). This is because eo
+ * is never negative (albeit if we rely on that need to be careful...)
*/
eo = _mm_sub_epi32(_mm_andnot_si128(dcdy_neg_mask, dcdy),
_mm_and_si128(dcdx_neg_mask, dcdx));
/* ei = _mm_sub_epi32(_mm_sub_epi32(dcdy, dcdx), eo); */
+ /*
+ * Pointless transpose which gets undone immediately in
+ * rasterization.
+ * It is actually difficult to do away with it - would essentially
+ * need GET_PLANES_DX, GET_PLANES_DY etc., but the calculations
+ * for this then would need to depend on the number of planes.
+ * The transpose is quite special here due to c being 64bit...
+ * The store has to be unaligned (unless we'd make the plane size
+ * a multiple of 128), and of course storing eo separately...
+ */
+ c01 = _mm_unpacklo_epi64(c02, c13);
+ c23 = _mm_unpackhi_epi64(c02, c13);
+ transpose2_64_2_32(&c01, &c23, &dcdx, &dcdy,
+ &p0, &p1, &p2, &unused);
+ _mm_storeu_si128((__m128i *)&plane[0], p0);
+ plane[0].eo = (uint32_t)_mm_cvtsi128_si32(eo);
+ _mm_storeu_si128((__m128i *)&plane[1], p1);
+ eo = _mm_shuffle_epi32(eo, _MM_SHUFFLE(3,2,0,1));
+ plane[1].eo = (uint32_t)_mm_cvtsi128_si32(eo);
+ _mm_storeu_si128((__m128i *)&plane[2], p2);
+ eo = _mm_shuffle_epi32(eo, _MM_SHUFFLE(0,0,0,2));
+ plane[2].eo = (uint32_t)_mm_cvtsi128_si32(eo);
+ } else
+#elif defined(_ARCH_PWR8) && UTIL_ARCH_LITTLE_ENDIAN
+ /*
+ * XXX this code is effectively disabled for all practical purposes,
+ * as the allowed fb size is tiny if FIXED_ORDER is 8.
+ */
+ if (setup->fb.width <= MAX_FIXED_LENGTH32 &&
+ setup->fb.height <= MAX_FIXED_LENGTH32 &&
+ (bbox.x1 - bbox.x0) <= MAX_FIXED_LENGTH32 &&
+ (bbox.y1 - bbox.y0) <= MAX_FIXED_LENGTH32) {
+ unsigned int bottom_edge;
+ __m128i vertx, verty;
+ __m128i shufx, shufy;
+ __m128i dcdx, dcdy, c;
+ __m128i unused;
+ __m128i dcdx_neg_mask;
+ __m128i dcdy_neg_mask;
+ __m128i dcdx_zero_mask;
+ __m128i top_left_flag;
+ __m128i c_inc_mask, c_inc;
+ __m128i eo, p0, p1, p2;
+ __m128i_union vshuf_mask;
+ __m128i zero = vec_splats((unsigned char) 0);
+ PIPE_ALIGN_VAR(16) int32_t temp_vec[4];
+
+#if UTIL_ARCH_LITTLE_ENDIAN
+ vshuf_mask.i[0] = 0x07060504;
+ vshuf_mask.i[1] = 0x0B0A0908;
+ vshuf_mask.i[2] = 0x03020100;
+ vshuf_mask.i[3] = 0x0F0E0D0C;
+#else
+ vshuf_mask.i[0] = 0x00010203;
+ vshuf_mask.i[1] = 0x0C0D0E0F;
+ vshuf_mask.i[2] = 0x04050607;
+ vshuf_mask.i[3] = 0x08090A0B;
+#endif
+
+ /* vertex x coords */
+ vertx = vec_load_si128((const uint32_t *) position->x);
+ /* vertex y coords */
+ verty = vec_load_si128((const uint32_t *) position->y);
+
+ shufx = vec_perm (vertx, vertx, vshuf_mask.m128i);
+ shufy = vec_perm (verty, verty, vshuf_mask.m128i);
+
+ dcdx = vec_sub_epi32(verty, shufy);
+ dcdy = vec_sub_epi32(vertx, shufx);
+
+ dcdx_neg_mask = vec_srai_epi32(dcdx, 31);
+ dcdx_zero_mask = vec_cmpeq_epi32(dcdx, zero);
+ dcdy_neg_mask = vec_srai_epi32(dcdy, 31);
+
+ bottom_edge = (setup->bottom_edge_rule == 0) ? ~0 : 0;
+ top_left_flag = (__m128i) vec_splats(bottom_edge);
+
+ c_inc_mask = vec_or(dcdx_neg_mask,
+ vec_and(dcdx_zero_mask,
+ vec_xor(dcdy_neg_mask,
+ top_left_flag)));
+
+ c_inc = vec_srli_epi32(c_inc_mask, 31);
+
+ c = vec_sub_epi32(vec_mullo_epi32(dcdx, vertx),
+ vec_mullo_epi32(dcdy, verty));
+
+ c = vec_add_epi32(c, c_inc);
+
+ /* Scale up to match c:
+ */
+ dcdx = vec_slli_epi32(dcdx, FIXED_ORDER);
+ dcdy = vec_slli_epi32(dcdy, FIXED_ORDER);
+
+ /* Calculate trivial reject values:
+ */
+ eo = vec_sub_epi32(vec_andnot_si128(dcdy_neg_mask, dcdy),
+ vec_and(dcdx_neg_mask, dcdx));
+
+ /* ei = _mm_sub_epi32(_mm_sub_epi32(dcdy, dcdx), eo); */
+
/* Pointless transpose which gets undone immediately in
* rasterization:
*/
transpose4_epi32(&c, &dcdx, &dcdy, &eo,
&p0, &p1, &p2, &unused);
-#define STORE_PLANE(plane, vec) do { \
- _mm_store_si128((__m128i *)&temp_vec, vec); \
- plane.c = (int64_t)temp_vec[0]; \
- plane.dcdx = temp_vec[1]; \
- plane.dcdy = temp_vec[2]; \
- plane.eo = temp_vec[3]; \
+#define STORE_PLANE(plane, vec) do { \
+ vec_store_si128((uint32_t *)&temp_vec, vec); \
+ plane.c = (int64_t)temp_vec[0]; \
+ plane.dcdx = temp_vec[1]; \
+ plane.dcdy = temp_vec[2]; \
+ plane.eo = temp_vec[3]; \
} while(0)
STORE_PLANE(plane[0], p0);
plane[2].dcdx = position->dy20;
for (i = 0; i < 3; i++) {
- /* half-edge constants, will be interated over the whole render
+ /* half-edge constants, will be iterated over the whole render
* target.
*/
plane[i].c = IMUL64(plane[i].dcdx, position->x[i]) -
- IMUL64(plane[i].dcdy, position->y[i]);
+ IMUL64(plane[i].dcdy, position->y[i]);
/* correct for top-left vs. bottom-left fill convention.
- */
+ */
if (plane[i].dcdx < 0) {
/* both fill conventions want this - adjust for left edges */
- plane[i].c++;
+ plane[i].c++;
}
else if (plane[i].dcdx == 0) {
if (setup->bottom_edge_rule == 0){
}
if (0) {
- debug_printf("p0: %"PRIx64"/%08x/%08x/%"PRIx64"\n",
+ debug_printf("p0: %"PRIx64"/%08x/%08x/%08x\n",
plane[0].c,
plane[0].dcdx,
plane[0].dcdy,
plane[0].eo);
-
- debug_printf("p1: %"PRIx64"/%08x/%08x/%"PRIx64"\n",
+
+ debug_printf("p1: %"PRIx64"/%08x/%08x/%08x\n",
plane[1].c,
plane[1].dcdx,
plane[1].dcdy,
plane[1].eo);
-
- debug_printf("p2: %"PRIx64"/%08x/%08x/%"PRIx64"\n",
+
+ debug_printf("p2: %"PRIx64"/%08x/%08x/%08x\n",
plane[2].c,
plane[2].dcdx,
plane[2].dcdy,
* Note that otherwise, the scissor planes only vary in 'C' value,
* and even then only on state-changes. Could alternatively store
* these planes elsewhere.
+ * (Or only store the c value together with a bit indicating which
+ * scissor edge this is, so rasterization would treat them differently
+ * (easier to evaluate) to ordinary planes.)
*/
- if (nr_planes == 7) {
- const struct u_rect *scissor = &setup->scissors[viewport_index];
-
- plane[3].dcdx = -1;
- plane[3].dcdy = 0;
- plane[3].c = 1-scissor->x0;
- plane[3].eo = 1;
-
- plane[4].dcdx = 1;
- plane[4].dcdy = 0;
- plane[4].c = scissor->x1+1;
- plane[4].eo = 0;
-
- plane[5].dcdx = 0;
- plane[5].dcdy = 1;
- plane[5].c = 1-scissor->y0;
- plane[5].eo = 1;
-
- plane[6].dcdx = 0;
- plane[6].dcdy = -1;
- plane[6].c = scissor->y1+1;
- plane[6].eo = 0;
+ if (nr_planes > 3) {
+ /* why not just use draw_regions */
+ struct lp_rast_plane *plane_s = &plane[3];
+
+ if (s_planes[0]) {
+ plane_s->dcdx = ~0U << 8;
+ plane_s->dcdy = 0;
+ plane_s->c = (1-scissor->x0) << 8;
+ plane_s->eo = 1 << 8;
+ plane_s++;
+ }
+ if (s_planes[1]) {
+ plane_s->dcdx = 1 << 8;
+ plane_s->dcdy = 0;
+ plane_s->c = (scissor->x1+1) << 8;
+ plane_s->eo = 0 << 8;
+ plane_s++;
+ }
+ if (s_planes[2]) {
+ plane_s->dcdx = 0;
+ plane_s->dcdy = 1 << 8;
+ plane_s->c = (1-scissor->y0) << 8;
+ plane_s->eo = 1 << 8;
+ plane_s++;
+ }
+ if (s_planes[3]) {
+ plane_s->dcdx = 0;
+ plane_s->dcdy = ~0U << 8;
+ plane_s->c = (scissor->y1+1) << 8;
+ plane_s->eo = 0;
+ plane_s++;
+ }
+ assert(plane_s == &plane[nr_planes]);
}
- return lp_setup_bin_triangle(setup, tri, &bbox, nr_planes, viewport_index);
+ return lp_setup_bin_triangle(setup, tri, &bbox, &bboxpos, nr_planes, viewport_index);
}
/*
*
* Undefined if no bit set exists, so code should check against 0 first.
*/
-static INLINE uint32_t
+static inline uint32_t
floor_pot(uint32_t n)
{
-#if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86)
+#if defined(PIPE_CC_GCC) && (defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64))
if (n == 0)
return 0;
__asm__("bsr %1,%0"
: "=r" (n)
- : "rm" (n));
+ : "rm" (n)
+ : "cc");
return 1 << n;
#else
n |= (n >> 1);
boolean
-lp_setup_bin_triangle( struct lp_setup_context *setup,
- struct lp_rast_triangle *tri,
- const struct u_rect *bbox,
- int nr_planes,
- unsigned viewport_index )
+lp_setup_bin_triangle(struct lp_setup_context *setup,
+ struct lp_rast_triangle *tri,
+ const struct u_rect *bboxorig,
+ const struct u_rect *bbox,
+ int nr_planes,
+ unsigned viewport_index)
{
struct lp_scene *scene = setup->scene;
struct u_rect trimmed_box = *bbox;
int max_sz = ((bbox->x1 - (bbox->x0 & ~3)) |
(bbox->y1 - (bbox->y0 & ~3)));
int sz = floor_pot(max_sz);
- boolean use_32bits = max_sz <= MAX_FIXED_LENGTH32;
+
+ /*
+ * NOTE: It is important to use the original bounding box
+ * which might contain negative values here, because if the
+ * plane math may overflow or not with the 32bit rasterization
+ * functions depends on the original extent of the triangle.
+ */
+ int max_szorig = ((bboxorig->x1 - (bboxorig->x0 & ~3)) |
+ (bboxorig->y1 - (bboxorig->y0 & ~3)));
+ boolean use_32bits = max_szorig <= MAX_FIXED_LENGTH32;
/* Now apply scissor, etc to the bounding box. Could do this
* earlier, but it confuses the logic for tri-16 and would force
ei[i] = (plane[i].dcdy -
plane[i].dcdx -
- plane[i].eo) << TILE_ORDER;
+ (int64_t)plane[i].eo) << TILE_ORDER;
- eo[i] = plane[i].eo << TILE_ORDER;
+ eo[i] = (int64_t)plane[i].eo << TILE_ORDER;
xstep[i] = -(((int64_t)plane[i].dcdx) << TILE_ORDER);
ystep[i] = ((int64_t)plane[i].dcdy) << TILE_ORDER;
}
for (i = 0; i < nr_planes; i++) {
int64_t planeout = cx[i] + eo[i];
int64_t planepartial = cx[i] + ei[i] - 1;
- out |= (planeout >> 63);
- partial |= (planepartial >> 63) & (1<<i);
+ out |= (int) (planeout >> 63);
+ partial |= ((int) (planepartial >> 63)) & (1<<i);
}
if (out) {
/**
* Calculate fixed position data for a triangle
+ * It is unfortunate we need to do that here (as we need area
+ * calculated in fixed point), as there's quite some code duplication
+ * to what is done in the jit setup prog.
*/
-static INLINE void
-calc_fixed_position( struct lp_setup_context *setup,
- struct fixed_position* position,
- const float (*v0)[4],
- const float (*v1)[4],
- const float (*v2)[4])
+static inline void
+calc_fixed_position(struct lp_setup_context *setup,
+ struct fixed_position* position,
+ const float (*v0)[4],
+ const float (*v1)[4],
+ const float (*v2)[4])
{
+ /*
+ * The rounding may not be quite the same with PIPE_ARCH_SSE
+ * (util_iround right now only does nearest/even on x87,
+ * otherwise nearest/away-from-zero).
+ * Both should be acceptable, I think.
+ */
+#if defined(PIPE_ARCH_SSE)
+ __m128 v0r, v1r;
+ __m128 vxy0xy2, vxy1xy0;
+ __m128i vxy0xy2i, vxy1xy0i;
+ __m128i dxdy0120, x0x2y0y2, x1x0y1y0, x0120, y0120;
+ __m128 pix_offset = _mm_set1_ps(setup->pixel_offset);
+ __m128 fixed_one = _mm_set1_ps((float)FIXED_ONE);
+ v0r = _mm_castpd_ps(_mm_load_sd((double *)v0[0]));
+ vxy0xy2 = _mm_loadh_pi(v0r, (__m64 *)v2[0]);
+ v1r = _mm_castpd_ps(_mm_load_sd((double *)v1[0]));
+ vxy1xy0 = _mm_movelh_ps(v1r, vxy0xy2);
+ vxy0xy2 = _mm_sub_ps(vxy0xy2, pix_offset);
+ vxy1xy0 = _mm_sub_ps(vxy1xy0, pix_offset);
+ vxy0xy2 = _mm_mul_ps(vxy0xy2, fixed_one);
+ vxy1xy0 = _mm_mul_ps(vxy1xy0, fixed_one);
+ vxy0xy2i = _mm_cvtps_epi32(vxy0xy2);
+ vxy1xy0i = _mm_cvtps_epi32(vxy1xy0);
+ dxdy0120 = _mm_sub_epi32(vxy0xy2i, vxy1xy0i);
+ _mm_store_si128((__m128i *)&position->dx01, dxdy0120);
+ /*
+ * For the mul, would need some more shuffles, plus emulation
+ * for the signed mul (without sse41), so don't bother.
+ */
+ x0x2y0y2 = _mm_shuffle_epi32(vxy0xy2i, _MM_SHUFFLE(3,1,2,0));
+ x1x0y1y0 = _mm_shuffle_epi32(vxy1xy0i, _MM_SHUFFLE(3,1,2,0));
+ x0120 = _mm_unpacklo_epi32(x0x2y0y2, x1x0y1y0);
+ y0120 = _mm_unpackhi_epi32(x0x2y0y2, x1x0y1y0);
+ _mm_store_si128((__m128i *)&position->x[0], x0120);
+ _mm_store_si128((__m128i *)&position->y[0], y0120);
+
+#else
position->x[0] = subpixel_snap(v0[0][0] - setup->pixel_offset);
position->x[1] = subpixel_snap(v1[0][0] - setup->pixel_offset);
position->x[2] = subpixel_snap(v2[0][0] - setup->pixel_offset);
- position->x[3] = 0;
+ position->x[3] = 0; // should be unused
position->y[0] = subpixel_snap(v0[0][1] - setup->pixel_offset);
position->y[1] = subpixel_snap(v1[0][1] - setup->pixel_offset);
position->y[2] = subpixel_snap(v2[0][1] - setup->pixel_offset);
- position->y[3] = 0;
+ position->y[3] = 0; // should be unused
position->dx01 = position->x[0] - position->x[1];
position->dy01 = position->y[0] - position->y[1];
position->dx20 = position->x[2] - position->x[0];
position->dy20 = position->y[2] - position->y[0];
+#endif
position->area = IMUL64(position->dx01, position->dy20) -
IMUL64(position->dx20, position->dy01);
* Rotate a triangle, flipping its clockwise direction,
* Swaps values for xy[0] and xy[1]
*/
-static INLINE void
+static inline void
rotate_fixed_position_01( struct fixed_position* position )
{
int x, y;
* Rotate a triangle, flipping its clockwise direction,
* Swaps values for xy[1] and xy[2]
*/
-static INLINE void
+static inline void
rotate_fixed_position_12( struct fixed_position* position )
{
int x, y;
}
-typedef void (*triangle_func_t)(struct lp_setup_context *setup,
- const float (*v0)[4],
- const float (*v1)[4],
- const float (*v2)[4]);
-
-
-/**
- * Subdivide this triangle by bisecting edge (v0, v1).
- * \param pv the provoking vertex (must = v0 or v1 or v2)
- * TODO: should probably think about non-overflowing arithmetic elsewhere.
- * This will definitely screw with pipeline counters for instance.
- */
-static void
-subdiv_tri(struct lp_setup_context *setup,
- const float (*v0)[4],
- const float (*v1)[4],
- const float (*v2)[4],
- const float (*pv)[4],
- triangle_func_t tri)
-{
- unsigned n = setup->fs.current.variant->shader->info.base.num_inputs + 1;
- const struct lp_shader_input *inputs =
- setup->fs.current.variant->shader->inputs;
- PIPE_ALIGN_VAR(LP_MIN_VECTOR_ALIGN) float vmid[PIPE_MAX_ATTRIBS][4];
- const float (*vm)[4] = (const float (*)[4]) vmid;
- unsigned i;
- float w0, w1, wm;
- boolean flatshade = setup->fs.current.variant->key.flatshade;
-
- /* find position midpoint (attrib[0] = position) */
- vmid[0][0] = 0.5f * (v1[0][0] + v0[0][0]);
- vmid[0][1] = 0.5f * (v1[0][1] + v0[0][1]);
- vmid[0][2] = 0.5f * (v1[0][2] + v0[0][2]);
- vmid[0][3] = 0.5f * (v1[0][3] + v0[0][3]);
-
- w0 = v0[0][3];
- w1 = v1[0][3];
- wm = vmid[0][3];
-
- /* interpolate other attributes */
- for (i = 1; i < n; i++) {
- if ((inputs[i - 1].interp == LP_INTERP_COLOR && flatshade) ||
- inputs[i - 1].interp == LP_INTERP_CONSTANT) {
- /* copy the provoking vertex's attribute */
- vmid[i][0] = pv[i][0];
- vmid[i][1] = pv[i][1];
- vmid[i][2] = pv[i][2];
- vmid[i][3] = pv[i][3];
- }
- else {
- /* interpolate with perspective correction (for linear too) */
- vmid[i][0] = 0.5f * (v1[i][0] * w1 + v0[i][0] * w0) / wm;
- vmid[i][1] = 0.5f * (v1[i][1] * w1 + v0[i][1] * w0) / wm;
- vmid[i][2] = 0.5f * (v1[i][2] * w1 + v0[i][2] * w0) / wm;
- vmid[i][3] = 0.5f * (v1[i][3] * w1 + v0[i][3] * w0) / wm;
- }
- }
-
- /* handling flat shading and first vs. last provoking vertex is a
- * little tricky...
- */
- if (pv == v0) {
- if (setup->flatshade_first) {
- /* first vertex must be v0 or vm */
- tri(setup, v0, vm, v2);
- tri(setup, vm, v1, v2);
- }
- else {
- /* last vertex must be v0 or vm */
- tri(setup, vm, v2, v0);
- tri(setup, v1, v2, vm);
- }
- }
- else if (pv == v1) {
- if (setup->flatshade_first) {
- tri(setup, vm, v2, v0);
- tri(setup, v1, v2, vm);
- }
- else {
- tri(setup, v2, v0, vm);
- tri(setup, v2, vm, v1);
- }
- }
- else {
- if (setup->flatshade_first) {
- tri(setup, v2, v0, vm);
- tri(setup, v2, vm, v1);
- }
- else {
- tri(setup, v0, vm, v2);
- tri(setup, vm, v1, v2);
- }
- }
-}
-
-
-/**
- * Check the lengths of the edges of the triangle. If any edge is too
- * long, subdivide the longest edge and draw two sub-triangles.
- * Note: this may be called recursively.
- * \return TRUE if triangle was subdivided, FALSE otherwise
- */
-static boolean
-check_subdivide_triangle(struct lp_setup_context *setup,
- const float (*v0)[4],
- const float (*v1)[4],
- const float (*v2)[4],
- triangle_func_t tri)
-{
- const float maxLen = (float) MAX_FIXED_LENGTH; /* longest permissible edge, in pixels */
- float dx10, dy10, len10;
- float dx21, dy21, len21;
- float dx02, dy02, len02;
- const float (*pv)[4] = setup->flatshade_first ? v0 : v2;
-
- /* compute lengths of triangle edges, squared */
- dx10 = v1[0][0] - v0[0][0];
- dy10 = v1[0][1] - v0[0][1];
- len10 = dx10 * dx10 + dy10 * dy10;
-
- dx21 = v2[0][0] - v1[0][0];
- dy21 = v2[0][1] - v1[0][1];
- len21 = dx21 * dx21 + dy21 * dy21;
-
- dx02 = v0[0][0] - v2[0][0];
- dy02 = v0[0][1] - v2[0][1];
- len02 = dx02 * dx02 + dy02 * dy02;
-
- /* Look for longest the edge that's longer than maxLen. If we find
- * such an edge, split the triangle using the midpoint of that edge.
- * Note: it's important to split the longest edge, not just any edge
- * that's longer than maxLen. Otherwise, we can get into a degenerate
- * situation and recurse indefinitely.
- */
- if (len10 > maxLen * maxLen &&
- len10 >= len21 &&
- len10 >= len02) {
- /* subdivide v0, v1 edge */
- subdiv_tri(setup, v0, v1, v2, pv, tri);
- return TRUE;
- }
-
- if (len21 > maxLen * maxLen &&
- len21 >= len10 &&
- len21 >= len02) {
- /* subdivide v1, v2 edge */
- subdiv_tri(setup, v1, v2, v0, pv, tri);
- return TRUE;
- }
-
- if (len02 > maxLen * maxLen &&
- len02 >= len21 &&
- len02 >= len10) {
- /* subdivide v2, v0 edge */
- subdiv_tri(setup, v2, v0, v1, pv, tri);
- return TRUE;
- }
-
- return FALSE;
-}
-
-
/**
* Draw triangle if it's CW, cull otherwise.
*/
-static void triangle_cw( struct lp_setup_context *setup,
- const float (*v0)[4],
- const float (*v1)[4],
- const float (*v2)[4] )
+static void triangle_cw(struct lp_setup_context *setup,
+ const float (*v0)[4],
+ const float (*v1)[4],
+ const float (*v2)[4])
{
- struct fixed_position position;
+ PIPE_ALIGN_VAR(16) struct fixed_position position;
+ struct llvmpipe_context *lp_context = (struct llvmpipe_context *)setup->pipe;
- if (setup->subdivide_large_triangles &&
- check_subdivide_triangle(setup, v0, v1, v2, triangle_cw))
- return;
+ if (lp_context->active_statistics_queries) {
+ lp_context->pipeline_statistics.c_primitives++;
+ }
calc_fixed_position(setup, &position, v0, v1, v2);
}
-static void triangle_ccw( struct lp_setup_context *setup,
- const float (*v0)[4],
- const float (*v1)[4],
- const float (*v2)[4])
+static void triangle_ccw(struct lp_setup_context *setup,
+ const float (*v0)[4],
+ const float (*v1)[4],
+ const float (*v2)[4])
{
- struct fixed_position position;
+ PIPE_ALIGN_VAR(16) struct fixed_position position;
+ struct llvmpipe_context *lp_context = (struct llvmpipe_context *)setup->pipe;
- if (setup->subdivide_large_triangles &&
- check_subdivide_triangle(setup, v0, v1, v2, triangle_ccw))
- return;
+ if (lp_context->active_statistics_queries) {
+ lp_context->pipeline_statistics.c_primitives++;
+ }
calc_fixed_position(setup, &position, v0, v1, v2);
/**
* Draw triangle whether it's CW or CCW.
*/
-static void triangle_both( struct lp_setup_context *setup,
- const float (*v0)[4],
- const float (*v1)[4],
- const float (*v2)[4] )
+static void triangle_both(struct lp_setup_context *setup,
+ const float (*v0)[4],
+ const float (*v1)[4],
+ const float (*v2)[4])
{
- struct fixed_position position;
+ PIPE_ALIGN_VAR(16) struct fixed_position position;
struct llvmpipe_context *lp_context = (struct llvmpipe_context *)setup->pipe;
- if (setup->subdivide_large_triangles &&
- check_subdivide_triangle(setup, v0, v1, v2, triangle_both))
- return;
-
- if (lp_context->active_statistics_queries &&
- !llvmpipe_rasterization_disabled(lp_context)) {
+ if (lp_context->active_statistics_queries) {
lp_context->pipeline_statistics.c_primitives++;
}
}
-static void triangle_nop( struct lp_setup_context *setup,
- const float (*v0)[4],
- const float (*v1)[4],
- const float (*v2)[4] )
+static void triangle_noop(struct lp_setup_context *setup,
+ const float (*v0)[4],
+ const float (*v1)[4],
+ const float (*v2)[4])
{
}
void
-lp_setup_choose_triangle( struct lp_setup_context *setup )
+lp_setup_choose_triangle(struct lp_setup_context *setup)
{
+ if (setup->rasterizer_discard) {
+ setup->triangle = triangle_noop;
+ return;
+ }
switch (setup->cullmode) {
case PIPE_FACE_NONE:
setup->triangle = triangle_both;
setup->triangle = setup->ccw_is_frontface ? triangle_cw : triangle_ccw;
break;
default:
- setup->triangle = triangle_nop;
+ setup->triangle = triangle_noop;
break;
}
}
projects / mesa.git / blobdiff