/**************************************************************************
*
- * 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 "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 {
- int x[4];
- int y[4];
- int area;
- int dx01;
- int dy01;
- int dx20;
- int dy20;
+ int32_t x[4];
+ int32_t y[4];
+ int64_t area;
+ int32_t dx01;
+ int32_t dy01;
+ int32_t dx20;
+ int32_t dy20;
};
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.
*/
struct u_rect bbox;
unsigned tri_bytes;
int nr_planes = 3;
+ unsigned viewport_index = 0;
+ unsigned layer = 0;
/* Area should always be positive here */
assert(position->area > 0);
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);
+ }
}
else {
nr_planes = 3;
}
+ if (setup->layer_slot > 0) {
+ layer = *(unsigned*)v1[setup->layer_slot];
+ layer = MIN2(layer, scene->fb_max_layer);
+ }
/* Bounding rectangle (in pixels) */
{
* 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;
}
- 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;
tri->inputs.frontfacing = frontfacing;
tri->inputs.disable = FALSE;
tri->inputs.opaque = setup->fs.current.variant->opaque;
+ tri->inputs.layer = layer;
+ tri->inputs.viewport_index = viewport_index;
if (0)
lp_dump_setup_coef(&setup->setup.variant->key,
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 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 */
dcdx_zero_mask = _mm_cmpeq_epi32(dcdx, zero);
dcdy_neg_mask = _mm_srai_epi32(dcdy, 31);
- top_left_flag = _mm_set1_epi32((setup->pixel_offset == 0) ? ~0 : 0);
+ 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,
transpose4_epi32(&c, &dcdx, &dcdy, &eo,
&p0, &p1, &p2, &unused);
- _mm_store_si128((__m128i *)&plane[0], p0);
- _mm_store_si128((__m128i *)&plane[1], p1);
- _mm_store_si128((__m128i *)&plane[2], p2);
- }
-#else
+#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;
/* half-edge constants, will be interated over the whole render
* target.
*/
- plane[i].c = plane[i].dcdx * position->x[i] - plane[i].dcdy * position->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.
+ 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->pixel_offset == 0) {
+ if (setup->bottom_edge_rule == 0){
/* correct for top-left fill convention:
*/
if (plane[i].dcdy > 0) plane[i].c++;
}
}
- plane[i].dcdx *= FIXED_ONE;
- plane[i].dcdy *= FIXED_ONE;
+ /* 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
if (plane[i].dcdy > 0) plane[i].eo += plane[i].dcdy;
}
}
-#endif
if (0) {
- debug_printf("p0: %08x/%08x/%08x/%08x\n",
+ debug_printf("p0: %"PRIx64"/%08x/%08x/%"PRIx64"\n",
plane[0].c,
plane[0].dcdx,
plane[0].dcdy,
plane[0].eo);
- debug_printf("p1: %08x/%08x/%08x/%08x\n",
+ debug_printf("p1: %"PRIx64"/%08x/%08x/%"PRIx64"\n",
plane[1].c,
plane[1].dcdx,
plane[1].dcdy,
plane[1].eo);
- debug_printf("p0: %08x/%08x/%08x/%08x\n",
+ debug_printf("p2: %"PRIx64"/%08x/%08x/%"PRIx64"\n",
plane[2].c,
plane[2].dcdx,
plane[2].dcdy,
* these planes elsewhere.
*/
if (nr_planes == 7) {
- const struct u_rect *scissor = &setup->scissor;
+ const struct u_rect *scissor = &setup->scissors[viewport_index];
plane[3].dcdx = -1;
plane[3].dcdy = 0;
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)));
+ 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_region, &trimmed_box);
+ u_rect_find_intersection(&setup->draw_regions[viewport_index],
+ &trimmed_box);
/* Determine which tile(s) intersect the triangle's bounding box
*/
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) );
}
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) );
}
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_cmd_with_state( scene, ix0, iy0, setup->fs.stored,
- 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
{
struct lp_rast_plane *plane = GET_PLANES(tri);
- int c[MAX_PLANES];
- int ei[MAX_PLANES];
+ int64_t c[MAX_PLANES];
+ int64_t ei[MAX_PLANES];
- int eo[MAX_PLANES];
- int xstep[MAX_PLANES];
- int ystep[MAX_PLANES];
+ int64_t eo[MAX_PLANES];
+ int64_t xstep[MAX_PLANES];
+ int64_t ystep[MAX_PLANES];
int x, y;
int ix0 = trimmed_box.x0 / TILE_SIZE;
for (i = 0; i < nr_planes; i++) {
c[i] = (plane[i].c +
- plane[i].dcdy * iy0 * TILE_SIZE -
- plane[i].dcdx * ix0 * TILE_SIZE);
+ 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] = -(plane[i].dcdx << TILE_ORDER);
- ystep[i] = plane[i].dcdy << 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[MAX_PLANES];
+ 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);
if (!lp_scene_bin_cmd_with_state( scene, x, y,
setup->fs.stored,
- lp_rast_tri_tab[count],
+ use_32bits ?
+ lp_rast_32_tri_tab[count] :
+ lp_rast_tri_tab[count],
lp_rast_arg_triangle(tri, partial) ))
goto fail;
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];
}
}
}
-
/**
* Calculate fixed position data for a triangle
*/
-static INLINE void
+static inline void
calc_fixed_position( struct lp_setup_context *setup,
struct fixed_position* position,
const float (*v0)[4],
position->dx20 = position->x[2] - position->x[0];
position->dy20 = position->y[2] - position->y[0];
- position->area = position->dx01 * position->dy20 - position->dx20 * position->dy01;
+ 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
+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( struct fixed_position* position )
+static inline void
+rotate_fixed_position_12( struct fixed_position* position )
{
int x, y;
const float (*v2)[4] )
{
struct fixed_position position;
+
calc_fixed_position(setup, &position, v0, v1, v2);
if (position.area < 0) {
- rotate_fixed_position(&position);
- retry_triangle_ccw(setup, &position, v0, v2, v1, !setup->ccw_is_frontface);
+ 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);
+ }
}
}
const float (*v2)[4])
{
struct fixed_position position;
+
calc_fixed_position(setup, &position, v0, v1, v2);
if (position.area > 0)
const float (*v2)[4] )
{
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) {
if (position.area > 0)
retry_triangle_ccw( setup, &position, v0, v1, v2, setup->ccw_is_frontface );
else if (position.area < 0) {
- rotate_fixed_position( &position );
- retry_triangle_ccw( setup, &position, v0, v2, v1, !setup->ccw_is_frontface );
+ 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