/**************************************************************************
*
- * 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.
#include "util/u_math.h"
#include "util/u_memory.h"
#include "util/u_rect.h"
+#include "util/u_sse.h"
#include "lp_perf.h"
#include "lp_setup_context.h"
-#include "lp_setup_coef.h"
#include "lp_rast.h"
#include "lp_state_fs.h"
+#include "lp_state_setup.h"
+#include "lp_context.h"
+
+#include <inttypes.h>
#define NUM_CHANNELS 4
+#if defined(PIPE_ARCH_SSE)
+#include <emmintrin.h>
+#endif
-
-static INLINE int
+static inline int
subpixel_snap(float a)
{
return util_iround(FIXED_ONE * a);
}
-static INLINE float
+static inline float
fixed_to_float(int a)
{
- return a * (1.0 / FIXED_ONE);
+ return a * (1.0f / FIXED_ONE);
}
-
-
-
+/* 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;
+};
/**
* immediately after it.
* The memory is allocated from the per-scene pool, not per-tile.
* \param tri_size returns number of bytes allocated
- * \param nr_inputs number of fragment shader inputs
+ * \param num_inputs number of fragment shader inputs
* \return pointer to triangle space
*/
struct lp_rast_triangle *
unsigned *tri_size)
{
unsigned input_array_sz = NUM_CHANNELS * (nr_inputs + 1) * sizeof(float);
+ unsigned plane_sz = nr_planes * sizeof(struct lp_rast_plane);
struct lp_rast_triangle *tri;
- unsigned tri_bytes, bytes;
- char *inputs;
- tri_bytes = align(Offset(struct lp_rast_triangle, plane[nr_planes]), 16);
- bytes = tri_bytes + (3 * input_array_sz);
+ *tri_size = (sizeof(struct lp_rast_triangle) +
+ 3 * input_array_sz +
+ plane_sz);
- tri = lp_scene_alloc_aligned( scene, bytes, 16 );
+ tri = lp_scene_alloc_aligned( scene, *tri_size, 16 );
+ if (!tri)
+ return NULL;
- if (tri) {
- inputs = ((char *)tri) + tri_bytes;
- tri->inputs.a0 = (float (*)[4]) inputs;
- tri->inputs.dadx = (float (*)[4]) (inputs + input_array_sz);
- tri->inputs.dady = (float (*)[4]) (inputs + 2 * input_array_sz);
+ tri->inputs.stride = input_array_sz;
- *tri_size = bytes;
+ {
+ char *a = (char *)tri;
+ char *b = (char *)&GET_PLANES(tri)[nr_planes];
+ assert(b - a == *tri_size);
}
return tri;
const char *name,
const float (*v)[4])
{
+ const struct lp_setup_variant_key *key = &setup->setup.variant->key;
int i, j;
debug_printf(" wpos (%s[0]) xyzw %f %f %f %f\n",
name,
v[0][0], v[0][1], v[0][2], v[0][3]);
- for (i = 0; i < setup->fs.nr_inputs; i++) {
- const float *in = v[setup->fs.input[i].src_index];
+ for (i = 0; i < key->num_inputs; i++) {
+ const float *in = v[key->inputs[i].src_index];
debug_printf(" in[%d] (%s[%d]) %s%s%s%s ",
i,
- name, setup->fs.input[i].src_index,
- (setup->fs.input[i].usage_mask & 0x1) ? "x" : " ",
- (setup->fs.input[i].usage_mask & 0x2) ? "y" : " ",
- (setup->fs.input[i].usage_mask & 0x4) ? "z" : " ",
- (setup->fs.input[i].usage_mask & 0x8) ? "w" : " ");
+ name, key->inputs[i].src_index,
+ (key->inputs[i].usage_mask & 0x1) ? "x" : " ",
+ (key->inputs[i].usage_mask & 0x2) ? "y" : " ",
+ (key->inputs[i].usage_mask & 0x4) ? "z" : " ",
+ (key->inputs[i].usage_mask & 0x8) ? "w" : " ");
for (j = 0; j < 4; j++)
- if (setup->fs.input[i].usage_mask & (1<<j))
+ if (key->inputs[i].usage_mask & (1<<j))
debug_printf("%.5f ", in[j]);
debug_printf("\n");
}
+#define MAX_PLANES 8
static unsigned
-lp_rast_tri_tab[9] = {
+lp_rast_tri_tab[MAX_PLANES+1] = {
0, /* should be impossible */
LP_RAST_OP_TRIANGLE_1,
LP_RAST_OP_TRIANGLE_2,
LP_RAST_OP_TRIANGLE_8
};
+static unsigned
+lp_rast_32_tri_tab[MAX_PLANES+1] = {
+ 0, /* should be impossible */
+ LP_RAST_OP_TRIANGLE_32_1,
+ LP_RAST_OP_TRIANGLE_32_2,
+ LP_RAST_OP_TRIANGLE_32_3,
+ LP_RAST_OP_TRIANGLE_32_4,
+ LP_RAST_OP_TRIANGLE_32_5,
+ LP_RAST_OP_TRIANGLE_32_6,
+ LP_RAST_OP_TRIANGLE_32_7,
+ LP_RAST_OP_TRIANGLE_32_8
+};
+
/**
/* if variant is opaque and scissor doesn't effect the tile */
if (inputs->opaque) {
- if (!scene->fb.zsbuf) {
+ /* Several things prevent this optimization from working:
+ * - For layered rendering we can't determine if this covers the same layer
+ * as previous rendering (or in case of clears those actually always cover
+ * all layers so optimization is impossible). Need to use fb_max_layer and
+ * not setup->layer_slot to determine this since even if there's currently
+ * no slot assigned previous rendering could have used one.
+ * - If there were any Begin/End query commands in the scene then those
+ * would get removed which would be very wrong. Furthermore, if queries
+ * were just active we also can't do the optimization since to get
+ * accurate query results we unfortunately need to execute the rendering
+ * commands.
+ */
+ if (!scene->fb.zsbuf && scene->fb_max_layer == 0 && !scene->had_queries) {
/*
* All previous rendering will be overwritten so reset the bin.
*/
}
LP_COUNT(nr_shade_opaque_64);
- return lp_scene_bin_command( scene, tx, ty,
- LP_RAST_OP_SHADE_TILE_OPAQUE,
- lp_rast_arg_inputs(inputs) );
+ return lp_scene_bin_cmd_with_state( scene, tx, ty,
+ setup->fs.stored,
+ LP_RAST_OP_SHADE_TILE_OPAQUE,
+ lp_rast_arg_inputs(inputs) );
} else {
LP_COUNT(nr_shade_64);
- return lp_scene_bin_command( scene, tx, ty,
- LP_RAST_OP_SHADE_TILE,
- lp_rast_arg_inputs(inputs) );
+ return lp_scene_bin_cmd_with_state( scene, tx, ty,
+ setup->fs.stored,
+ LP_RAST_OP_SHADE_TILE,
+ lp_rast_arg_inputs(inputs) );
}
}
*/
static boolean
do_triangle_ccw(struct lp_setup_context *setup,
- const float (*v0)[4],
- const float (*v1)[4],
- const float (*v2)[4],
- boolean frontfacing )
+ struct fixed_position* position,
+ const float (*v0)[4],
+ const float (*v1)[4],
+ const float (*v2)[4],
+ boolean frontfacing )
{
struct lp_scene *scene = setup->scene;
+ const struct lp_setup_variant_key *key = &setup->setup.variant->key;
struct lp_rast_triangle *tri;
- int x[3];
- int y[3];
- int area;
+ struct lp_rast_plane *plane;
struct u_rect bbox;
unsigned tri_bytes;
- int i;
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->flatshade_first) {
+ pv = v0;
+ }
+ else {
+ 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*)pv[setup->layer_slot];
+ layer = MIN2(layer, scene->fb_max_layer);
+ }
+
if (setup->scissor_test) {
nr_planes = 7;
}
nr_planes = 3;
}
- /* x/y positions in fixed point */
- x[0] = subpixel_snap(v0[0][0] - setup->pixel_offset);
- x[1] = subpixel_snap(v1[0][0] - setup->pixel_offset);
- x[2] = subpixel_snap(v2[0][0] - setup->pixel_offset);
- y[0] = subpixel_snap(v0[0][1] - setup->pixel_offset);
- y[1] = subpixel_snap(v1[0][1] - setup->pixel_offset);
- y[2] = subpixel_snap(v2[0][1] - setup->pixel_offset);
-
-
/* Bounding rectangle (in pixels) */
{
/* Yes this is necessary to accurately calculate bounding boxes
* 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;
- bbox.x0 = (MIN3(x[0], x[1], x[2]) + (FIXED_ONE-1)) >> FIXED_ORDER;
- bbox.x1 = (MAX3(x[0], x[1], x[2]) + (FIXED_ONE-1)) >> FIXED_ORDER;
- bbox.y0 = (MIN3(y[0], y[1], y[2]) + (FIXED_ONE-1) + adj) >> FIXED_ORDER;
- bbox.y1 = (MAX3(y[0], y[1], y[2]) + (FIXED_ONE-1) + adj) >> FIXED_ORDER;
+ /* Inclusive x0, exclusive x1 */
+ bbox.x0 = MIN3(position->x[0], position->x[1], position->x[2]) >> FIXED_ORDER;
+ bbox.x1 = (MAX3(position->x[0], position->x[1], position->x[2]) - 1) >> FIXED_ORDER;
- /* Inclusive coordinates:
- */
- bbox.x1--;
- bbox.y1--;
+ /* Inclusive / exclusive depending upon adj (bottom-left or top-right) */
+ bbox.y0 = (MIN3(position->y[0], position->y[1], position->y[2]) + adj) >> FIXED_ORDER;
+ bbox.y1 = (MAX3(position->y[0], position->y[1], position->y[2]) - 1 + adj) >> FIXED_ORDER;
}
if (bbox.x1 < bbox.x0 ||
return TRUE;
}
- if (!u_rect_test_intersection(&setup->draw_region, &bbox)) {
+ if (!u_rect_test_intersection(&setup->draw_regions[viewport_index], &bbox)) {
if (0) debug_printf("offscreen\n");
LP_COUNT(nr_culled_tris);
return TRUE;
}
- u_rect_find_intersection(&setup->draw_region, &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);
tri = lp_setup_alloc_triangle(scene,
- setup->fs.nr_inputs,
+ key->num_inputs,
nr_planes,
&tri_bytes);
if (!tri)
return FALSE;
-#ifdef DEBUG
+#if 0
tri->v[0][0] = v0[0][0];
tri->v[1][0] = v1[0][0];
tri->v[2][0] = v2[0][0];
tri->v[2][1] = v2[0][1];
#endif
- tri->plane[0].dcdy = x[0] - x[1];
- tri->plane[1].dcdy = x[1] - x[2];
- tri->plane[2].dcdy = x[2] - x[0];
-
- tri->plane[0].dcdx = y[0] - y[1];
- tri->plane[1].dcdx = y[1] - y[2];
- tri->plane[2].dcdx = y[2] - y[0];
-
- area = (tri->plane[0].dcdy * tri->plane[2].dcdx -
- tri->plane[2].dcdy * tri->plane[0].dcdx);
-
LP_COUNT(nr_tris);
- /* Cull non-ccw and zero-sized triangles.
- *
- * XXX: subject to overflow??
- */
- if (area <= 0) {
- lp_scene_putback_data( scene, tri_bytes );
- LP_COUNT(nr_culled_tris);
- return TRUE;
- }
-
/* Setup parameter interpolants:
*/
- lp_setup_tri_coef( setup, &tri->inputs, v0, v1, v2, frontfacing );
-
- tri->inputs.facing = frontfacing ? 1.0F : -1.0F;
+ 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;
tri->inputs.opaque = setup->fs.current.variant->opaque;
- tri->inputs.state = setup->fs.stored;
+ tri->inputs.layer = layer;
+ tri->inputs.viewport_index = viewport_index;
-
- for (i = 0; i < 3; i++) {
- struct lp_rast_plane *plane = &tri->plane[i];
+ 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));
+
+ 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) {
+ __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 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 */
+
+ shufx = _mm_shuffle_epi32(vertx, _MM_SHUFFLE(3,0,2,1));
+ shufy = _mm_shuffle_epi32(verty, _MM_SHUFFLE(3,0,2,1));
+
+ dcdx = _mm_sub_epi32(verty, shufy);
+ dcdy = _mm_sub_epi32(vertx, shufx);
+
+ dcdx_neg_mask = _mm_srai_epi32(dcdx, 31);
+ dcdx_zero_mask = _mm_cmpeq_epi32(dcdx, zero);
+ dcdy_neg_mask = _mm_srai_epi32(dcdy, 31);
+
+ 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 = _mm_add_epi32(c, c_inc);
+
+ /* Scale up to match c:
+ */
+ dcdx = _mm_slli_epi32(dcdx, FIXED_ORDER);
+ dcdy = _mm_slli_epi32(dcdy, FIXED_ORDER);
- /* half-edge constants, will be interated over the whole render
- * target.
+ /* Calculate trivial reject values:
*/
- plane->c = plane->dcdx * x[i] - plane->dcdy * y[i];
-
- /* correct for top-left vs. bottom-left fill convention.
- *
- * note that we're overloading gl_rasterization_rules to mean
- * both (0.5,0.5) pixel centers *and* bottom-left filling
- * convention.
- *
- * GL actually has a top-left filling convention, but GL's
- * notion of "top" differs from gallium's...
- *
- * Also, sometimes (in FBO cases) GL will render upside down
- * to its usual method, in which case it will probably want
- * to use the opposite, top-left convention.
- */
- if (plane->dcdx < 0) {
- /* both fill conventions want this - adjust for left edges */
- plane->c++;
- }
- else if (plane->dcdx == 0) {
- if (setup->pixel_offset == 0) {
- /* correct for top-left fill convention:
- */
- if (plane->dcdy > 0) plane->c++;
- }
- else {
- /* correct for bottom-left fill convention:
- */
- if (plane->dcdy < 0) plane->c++;
- }
- }
+ eo = _mm_sub_epi32(_mm_andnot_si128(dcdy_neg_mask, dcdy),
+ _mm_and_si128(dcdx_neg_mask, dcdx));
- plane->dcdx *= FIXED_ONE;
- plane->dcdy *= FIXED_ONE;
+ /* ei = _mm_sub_epi32(_mm_sub_epi32(dcdy, dcdx), eo); */
- /* find trivial reject offsets for each edge for a single-pixel
- * sized block. These will be scaled up at each recursive level to
- * match the active blocksize. Scaling in this way works best if
- * the blocks are square.
+ /* Pointless transpose which gets undone immediately in
+ * rasterization:
*/
- plane->eo = 0;
- if (plane->dcdx < 0) plane->eo -= plane->dcdx;
- if (plane->dcdy > 0) plane->eo += plane->dcdy;
+ 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]; \
+ } while(0)
+
+ STORE_PLANE(plane[0], p0);
+ STORE_PLANE(plane[1], p1);
+ STORE_PLANE(plane[2], p2);
+#undef STORE_PLANE
+ } else
+#endif
+ {
+ int i;
+ plane[0].dcdy = position->dx01;
+ plane[1].dcdy = position->x[1] - position->x[2];
+ plane[2].dcdy = position->dx20;
+ plane[0].dcdx = position->dy01;
+ plane[1].dcdx = position->y[1] - position->y[2];
+ plane[2].dcdx = position->dy20;
+
+ for (i = 0; i < 3; i++) {
+ /* half-edge constants, will be interated over the whole render
+ * target.
+ */
+ plane[i].c = IMUL64(plane[i].dcdx, position->x[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++;
+ }
+ else if (plane[i].dcdx == 0) {
+ if (setup->bottom_edge_rule == 0){
+ /* correct for top-left fill convention:
+ */
+ if (plane[i].dcdy > 0) plane[i].c++;
+ }
+ else {
+ /* correct for bottom-left fill convention:
+ */
+ if (plane[i].dcdy < 0) plane[i].c++;
+ }
+ }
- /* Calculate trivial accept offsets from the above.
- */
- plane->ei = plane->dcdy - plane->dcdx - plane->eo;
+ /* Scale up to match c:
+ */
+ assert((plane[i].dcdx << FIXED_ORDER) >> FIXED_ORDER == plane[i].dcdx);
+ assert((plane[i].dcdy << FIXED_ORDER) >> FIXED_ORDER == plane[i].dcdy);
+ plane[i].dcdx <<= FIXED_ORDER;
+ plane[i].dcdy <<= FIXED_ORDER;
+
+ /* find trivial reject offsets for each edge for a single-pixel
+ * sized block. These will be scaled up at each recursive level to
+ * match the active blocksize. Scaling in this way works best if
+ * the blocks are square.
+ */
+ plane[i].eo = 0;
+ if (plane[i].dcdx < 0) plane[i].eo -= plane[i].dcdx;
+ if (plane[i].dcdy > 0) plane[i].eo += plane[i].dcdy;
+ }
+ }
+
+ if (0) {
+ debug_printf("p0: %"PRIx64"/%08x/%08x/%"PRIx64"\n",
+ plane[0].c,
+ plane[0].dcdx,
+ plane[0].dcdy,
+ plane[0].eo);
+
+ debug_printf("p1: %"PRIx64"/%08x/%08x/%"PRIx64"\n",
+ plane[1].c,
+ plane[1].dcdx,
+ plane[1].dcdy,
+ plane[1].eo);
+
+ debug_printf("p2: %"PRIx64"/%08x/%08x/%"PRIx64"\n",
+ plane[2].c,
+ plane[2].dcdx,
+ plane[2].dcdy,
+ plane[2].eo);
}
* these planes elsewhere.
*/
if (nr_planes == 7) {
- tri->plane[3].dcdx = -1;
- tri->plane[3].dcdy = 0;
- tri->plane[3].c = 1-bbox.x0;
- tri->plane[3].ei = 0;
- tri->plane[3].eo = 1;
-
- tri->plane[4].dcdx = 1;
- tri->plane[4].dcdy = 0;
- tri->plane[4].c = bbox.x1+1;
- tri->plane[4].ei = -1;
- tri->plane[4].eo = 0;
-
- tri->plane[5].dcdx = 0;
- tri->plane[5].dcdy = 1;
- tri->plane[5].c = 1-bbox.y0;
- tri->plane[5].ei = 0;
- tri->plane[5].eo = 1;
-
- tri->plane[6].dcdx = 0;
- tri->plane[6].dcdy = -1;
- tri->plane[6].c = bbox.y1+1;
- tri->plane[6].ei = -1;
- tri->plane[6].eo = 0;
+ 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;
}
- return lp_setup_bin_triangle( setup, tri, &bbox, nr_planes );
+ return lp_setup_bin_triangle(setup, tri, &bbox, 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)
lp_setup_bin_triangle( struct lp_setup_context *setup,
struct lp_rast_triangle *tri,
const struct u_rect *bbox,
- int nr_planes )
+ int nr_planes,
+ unsigned viewport_index )
{
struct lp_scene *scene = setup->scene;
+ struct u_rect trimmed_box = *bbox;
int i;
-
/* What is the largest power-of-two boundary this triangle crosses:
*/
int dx = floor_pot((bbox->x0 ^ bbox->x1) |
/* The largest dimension of the rasterized area of the triangle
* (aligned to a 4x4 grid), rounded down to the nearest power of two:
*/
- int sz = floor_pot((bbox->x1 - (bbox->x0 & ~3)) |
- (bbox->y1 - (bbox->y0 & ~3)));
-
- if (nr_planes == 3) {
- if (sz < 4 && dx < 64)
- {
- /* Triangle is contained in a single 4x4 stamp:
- */
- int mask = (bbox->x0 & 63 & ~3) | ((bbox->y0 & 63 & ~3) << 8);
-
- return lp_scene_bin_command( scene,
- bbox->x0/64, bbox->y0/64,
- LP_RAST_OP_TRIANGLE_3_4,
- lp_rast_arg_triangle(tri, mask) );
- }
-
- if (sz < 16 && dx < 64)
- {
- int mask = (bbox->x0 & 63 & ~3) | ((bbox->y0 & 63 & ~3) << 8);
-
- /* Triangle is contained in a single 16x16 block:
- */
- return lp_scene_bin_command( scene,
- bbox->x0/64, bbox->y0/64,
- LP_RAST_OP_TRIANGLE_3_16,
- lp_rast_arg_triangle(tri, mask) );
- }
- }
-
+ 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;
+
+ /* Now apply scissor, etc to the bounding box. Could do this
+ * earlier, but it confuses the logic for tri-16 and would force
+ * the rasterizer to also respect scissor, etc, just for the rare
+ * cases where a small triangle extends beyond the scissor.
+ */
+ u_rect_find_intersection(&setup->draw_regions[viewport_index],
+ &trimmed_box);
/* Determine which tile(s) intersect the triangle's bounding box
*/
{
int ix0 = bbox->x0 / TILE_SIZE;
int iy0 = bbox->y0 / TILE_SIZE;
+ unsigned px = bbox->x0 & 63 & ~3;
+ unsigned py = bbox->y0 & 63 & ~3;
assert(iy0 == bbox->y1 / TILE_SIZE &&
ix0 == bbox->x1 / TILE_SIZE);
+ if (nr_planes == 3) {
+ if (sz < 4)
+ {
+ /* Triangle is contained in a single 4x4 stamp:
+ */
+ assert(px + 4 <= TILE_SIZE);
+ assert(py + 4 <= TILE_SIZE);
+ return lp_scene_bin_cmd_with_state( scene, ix0, iy0,
+ setup->fs.stored,
+ use_32bits ?
+ LP_RAST_OP_TRIANGLE_32_3_4 :
+ LP_RAST_OP_TRIANGLE_3_4,
+ lp_rast_arg_triangle_contained(tri, px, py) );
+ }
+
+ if (sz < 16)
+ {
+ /* Triangle is contained in a single 16x16 block:
+ */
+
+ /*
+ * The 16x16 block is only 4x4 aligned, and can exceed the tile
+ * dimensions if the triangle is 16 pixels in one dimension but 4
+ * in the other. So budge the 16x16 back inside the tile.
+ */
+ px = MIN2(px, TILE_SIZE - 16);
+ py = MIN2(py, TILE_SIZE - 16);
+
+ assert(px + 16 <= TILE_SIZE);
+ assert(py + 16 <= TILE_SIZE);
+
+ return lp_scene_bin_cmd_with_state( scene, ix0, iy0,
+ setup->fs.stored,
+ use_32bits ?
+ LP_RAST_OP_TRIANGLE_32_3_16 :
+ LP_RAST_OP_TRIANGLE_3_16,
+ lp_rast_arg_triangle_contained(tri, px, py) );
+ }
+ }
+ else if (nr_planes == 4 && sz < 16)
+ {
+ px = MIN2(px, TILE_SIZE - 16);
+ py = MIN2(py, TILE_SIZE - 16);
+
+ assert(px + 16 <= TILE_SIZE);
+ assert(py + 16 <= TILE_SIZE);
+
+ return lp_scene_bin_cmd_with_state(scene, ix0, iy0,
+ setup->fs.stored,
+ use_32bits ?
+ LP_RAST_OP_TRIANGLE_32_4_16 :
+ LP_RAST_OP_TRIANGLE_4_16,
+ lp_rast_arg_triangle_contained(tri, px, py));
+ }
+
+
/* Triangle is contained in a single tile:
*/
- return lp_scene_bin_command( scene, ix0, iy0,
- lp_rast_tri_tab[nr_planes],
- lp_rast_arg_triangle(tri, (1<<nr_planes)-1) );
+ return lp_scene_bin_cmd_with_state(
+ scene, ix0, iy0, setup->fs.stored,
+ use_32bits ? lp_rast_32_tri_tab[nr_planes] : lp_rast_tri_tab[nr_planes],
+ lp_rast_arg_triangle(tri, (1<<nr_planes)-1));
}
else
{
- int c[7];
- int ei[7];
- int eo[7];
- int xstep[7];
- int ystep[7];
+ struct lp_rast_plane *plane = GET_PLANES(tri);
+ int64_t c[MAX_PLANES];
+ int64_t ei[MAX_PLANES];
+
+ int64_t eo[MAX_PLANES];
+ int64_t xstep[MAX_PLANES];
+ int64_t ystep[MAX_PLANES];
int x, y;
- int ix0 = bbox->x0 / TILE_SIZE;
- int iy0 = bbox->y0 / TILE_SIZE;
- int ix1 = bbox->x1 / TILE_SIZE;
- int iy1 = bbox->y1 / TILE_SIZE;
+ int ix0 = trimmed_box.x0 / TILE_SIZE;
+ int iy0 = trimmed_box.y0 / TILE_SIZE;
+ int ix1 = trimmed_box.x1 / TILE_SIZE;
+ int iy1 = trimmed_box.y1 / TILE_SIZE;
for (i = 0; i < nr_planes; i++) {
- c[i] = (tri->plane[i].c +
- tri->plane[i].dcdy * iy0 * TILE_SIZE -
- tri->plane[i].dcdx * ix0 * TILE_SIZE);
-
- ei[i] = tri->plane[i].ei << TILE_ORDER;
- eo[i] = tri->plane[i].eo << TILE_ORDER;
- xstep[i] = -(tri->plane[i].dcdx << TILE_ORDER);
- ystep[i] = tri->plane[i].dcdy << TILE_ORDER;
+ c[i] = (plane[i].c +
+ IMUL64(plane[i].dcdy, iy0) * TILE_SIZE -
+ IMUL64(plane[i].dcdx, ix0) * TILE_SIZE);
+
+ ei[i] = (plane[i].dcdy -
+ plane[i].dcdx -
+ plane[i].eo) << TILE_ORDER;
+
+ eo[i] = plane[i].eo << TILE_ORDER;
+ xstep[i] = -(((int64_t)plane[i].dcdx) << TILE_ORDER);
+ ystep[i] = ((int64_t)plane[i].dcdy) << TILE_ORDER;
}
*/
for (y = iy0; y <= iy1; y++)
{
- boolean in = FALSE; /* are we inside the triangle? */
- int cx[7];
+ boolean in = FALSE; /* are we inside the triangle? */
+ int64_t cx[MAX_PLANES];
for (i = 0; i < nr_planes; i++)
cx[i] = c[i];
- for (x = ix0; x <= ix1; x++)
- {
+ for (x = ix0; x <= ix1; x++)
+ {
int out = 0;
int partial = 0;
for (i = 0; i < nr_planes; i++) {
- int planeout = cx[i] + eo[i];
- int planepartial = cx[i] + ei[i] - 1;
- out |= (planeout >> 31);
- partial |= (planepartial >> 31) & (1<<i);
+ int64_t planeout = cx[i] + eo[i];
+ int64_t planepartial = cx[i] + ei[i] - 1;
+ out |= (int) (planeout >> 63);
+ partial |= ((int) (planepartial >> 63)) & (1<<i);
}
if (out) {
LP_COUNT(nr_empty_64);
}
else if (partial) {
- /* Not trivially accepted by at least one plane -
+ /* Not trivially accepted by at least one plane -
* rasterize/shade partial tile
*/
int count = util_bitcount(partial);
in = TRUE;
- if (!lp_scene_bin_command( scene, x, y,
- lp_rast_tri_tab[count],
- lp_rast_arg_triangle(tri, partial) ))
+
+ if (!lp_scene_bin_cmd_with_state( scene, x, y,
+ setup->fs.stored,
+ use_32bits ?
+ lp_rast_32_tri_tab[count] :
+ lp_rast_tri_tab[count],
+ lp_rast_arg_triangle(tri, partial) ))
goto fail;
LP_COUNT(nr_partially_covered_64);
goto fail;
}
- /* Iterate cx values across the region:
- */
+ /* Iterate cx values across the region: */
for (i = 0; i < nr_planes; i++)
cx[i] += xstep[i];
- }
-
- /* Iterate c values down the region:
- */
+ }
+
+ /* Iterate c values down the region: */
for (i = 0; i < nr_planes; i++)
c[i] += ystep[i];
}
/**
- * Draw triangle if it's CW, cull otherwise.
+ * Try to draw the triangle, restart the scene on failure.
*/
-static void triangle_cw( struct lp_setup_context *setup,
- const float (*v0)[4],
- const float (*v1)[4],
- const float (*v2)[4] )
+static void retry_triangle_ccw( struct lp_setup_context *setup,
+ struct fixed_position* position,
+ const float (*v0)[4],
+ const float (*v1)[4],
+ const float (*v2)[4],
+ boolean front)
{
- if (!do_triangle_ccw( setup, v1, v0, v2, !setup->ccw_is_frontface ))
+ if (!do_triangle_ccw( setup, position, v0, v1, v2, front ))
{
- lp_setup_flush_and_restart(setup);
+ if (!lp_setup_flush_and_restart(setup))
+ return;
- if (!do_triangle_ccw( setup, v1, v0, v2, !setup->ccw_is_frontface ))
- assert(0);
+ if (!do_triangle_ccw( setup, position, v0, v1, v2, front ))
+ return;
}
}
+/**
+ * Calculate fixed position data for a triangle
+ */
+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])
+{
+ 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->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->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];
+
+ position->area = IMUL64(position->dx01, position->dy20) -
+ IMUL64(position->dx20, position->dy01);
+}
+
/**
- * Draw triangle if it's CCW, cull otherwise.
+ * Rotate a triangle, flipping its clockwise direction,
+ * Swaps values for xy[0] and xy[1]
*/
-static void triangle_ccw( struct lp_setup_context *setup,
+static inline void
+rotate_fixed_position_01( struct fixed_position* position )
+{
+ int x, y;
+
+ x = position->x[1];
+ y = position->y[1];
+ position->x[1] = position->x[0];
+ position->y[1] = position->y[0];
+ position->x[0] = x;
+ position->y[0] = y;
+
+ position->dx01 = -position->dx01;
+ position->dy01 = -position->dy01;
+ position->dx20 = position->x[2] - position->x[0];
+ position->dy20 = position->y[2] - position->y[0];
+
+ position->area = -position->area;
+}
+
+
+/**
+ * Rotate a triangle, flipping its clockwise direction,
+ * Swaps values for xy[1] and xy[2]
+ */
+static inline void
+rotate_fixed_position_12( struct fixed_position* position )
+{
+ int x, y;
+
+ x = position->x[2];
+ y = position->y[2];
+ position->x[2] = position->x[1];
+ position->y[2] = position->y[1];
+ position->x[1] = x;
+ position->y[1] = y;
+
+ x = position->dx01;
+ y = position->dy01;
+ position->dx01 = -position->dx20;
+ position->dy01 = -position->dy20;
+ position->dx20 = -x;
+ position->dy20 = -y;
+
+ position->area = -position->area;
+}
+
+
+/**
+ * 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] )
{
- if (!do_triangle_ccw( setup, v0, v1, v2, setup->ccw_is_frontface ))
- {
- lp_setup_flush_and_restart(setup);
- if (!do_triangle_ccw( setup, v0, v1, v2, setup->ccw_is_frontface ))
- assert(0);
+ struct fixed_position position;
+
+ calc_fixed_position(setup, &position, v0, v1, v2);
+
+ if (position.area < 0) {
+ if (setup->flatshade_first) {
+ rotate_fixed_position_12(&position);
+ retry_triangle_ccw(setup, &position, v0, v2, v1, !setup->ccw_is_frontface);
+ } else {
+ rotate_fixed_position_01(&position);
+ retry_triangle_ccw(setup, &position, v1, v0, v2, !setup->ccw_is_frontface);
+ }
}
}
+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;
+
+ calc_fixed_position(setup, &position, v0, v1, v2);
+
+ if (position.area > 0)
+ retry_triangle_ccw(setup, &position, v0, v1, v2, setup->ccw_is_frontface);
+}
/**
* Draw triangle whether it's CW or CCW.
const float (*v1)[4],
const float (*v2)[4] )
{
- /* edge vectors e = v0 - v2, f = v1 - v2 */
- const float ex = v0[0][0] - v2[0][0];
- const float ey = v0[0][1] - v2[0][1];
- const float fx = v1[0][0] - v2[0][0];
- const float fy = v1[0][1] - v2[0][1];
-
- /* det = cross(e,f).z */
- const float det = ex * fy - ey * fx;
- if (det < 0.0f)
- triangle_ccw( setup, v0, v1, v2 );
- else if (det > 0.0f)
- triangle_cw( setup, v0, v1, v2 );
+ struct fixed_position position;
+ struct llvmpipe_context *lp_context = (struct llvmpipe_context *)setup->pipe;
+
+ if (lp_context->active_statistics_queries &&
+ !llvmpipe_rasterization_disabled(lp_context)) {
+ lp_context->pipeline_statistics.c_primitives++;
+ }
+
+ calc_fixed_position(setup, &position, v0, v1, v2);
+
+ if (0) {
+ assert(!util_is_inf_or_nan(v0[0][0]));
+ assert(!util_is_inf_or_nan(v0[0][1]));
+ assert(!util_is_inf_or_nan(v1[0][0]));
+ assert(!util_is_inf_or_nan(v1[0][1]));
+ assert(!util_is_inf_or_nan(v2[0][0]));
+ assert(!util_is_inf_or_nan(v2[0][1]));
+ }
+
+ if (position.area > 0)
+ retry_triangle_ccw( setup, &position, v0, v1, v2, setup->ccw_is_frontface );
+ else if (position.area < 0) {
+ if (setup->flatshade_first) {
+ rotate_fixed_position_12( &position );
+ retry_triangle_ccw( setup, &position, v0, v2, v1, !setup->ccw_is_frontface );
+ } else {
+ rotate_fixed_position_01( &position );
+ retry_triangle_ccw( setup, &position, v1, v0, v2, !setup->ccw_is_frontface );
+ }
+ }
}
projects / mesa.git / blobdiff