#include "lp_scene_queue.h"
#include "lp_debug.h"
#include "lp_fence.h"
+#include "lp_perf.h"
#include "lp_rast.h"
#include "lp_rast_priv.h"
#include "lp_tile_soa.h"
-#include "lp_bld_debug.h"
+#include "gallivm/lp_bld_debug.h"
#include "lp_scene.h"
* \param y window Y position of the tile, in pixels
*/
static void
-lp_rast_start_tile( struct lp_rasterizer *rast,
- unsigned thread_index,
- unsigned x, unsigned y )
+lp_rast_start_tile(struct lp_rasterizer_task *task,
+ unsigned x, unsigned y)
{
LP_DBG(DEBUG_RAST, "%s %d,%d\n", __FUNCTION__, x, y);
- rast->tasks[thread_index].x = x;
- rast->tasks[thread_index].y = y;
+ task->x = x;
+ task->y = y;
}
* Clear the rasterizer's current color tile.
* This is a bin command called during bin processing.
*/
-void lp_rast_clear_color( struct lp_rasterizer *rast,
- unsigned thread_index,
- const union lp_rast_cmd_arg arg )
+void
+lp_rast_clear_color(struct lp_rasterizer_task *task,
+ const union lp_rast_cmd_arg arg)
{
+ struct lp_rasterizer *rast = task->rast;
const uint8_t *clear_color = arg.clear_color;
- uint8_t **color_tile = rast->tasks[thread_index].tile.color;
+ uint8_t **color_tile = task->tile.color;
unsigned i;
LP_DBG(DEBUG_RAST, "%s 0x%x,0x%x,0x%x,0x%x\n", __FUNCTION__,
assert(c - color_tile[i] == TILE_SIZE * TILE_SIZE * 4);
}
}
+
+ LP_COUNT(nr_color_tile_clear);
}
* Clear the rasterizer's current z/stencil tile.
* This is a bin command called during bin processing.
*/
-void lp_rast_clear_zstencil( struct lp_rasterizer *rast,
- unsigned thread_index,
- const union lp_rast_cmd_arg arg)
+void
+lp_rast_clear_zstencil(struct lp_rasterizer_task *task,
+ const union lp_rast_cmd_arg arg)
{
- unsigned i;
- uint32_t *depth_tile = rast->tasks[thread_index].tile.depth;
-
+ struct lp_rasterizer *rast = task->rast;
+ const unsigned tile_x = task->x;
+ const unsigned tile_y = task->y;
+ const unsigned height = TILE_SIZE/TILE_VECTOR_HEIGHT;
+ const unsigned width = TILE_SIZE*TILE_VECTOR_HEIGHT;
+ unsigned block_size = util_format_get_blocksize(rast->zsbuf_transfer->texture->format);
+ uint8_t *dst;
+ unsigned dst_stride = rast->zsbuf_transfer->stride*TILE_VECTOR_HEIGHT;
+ unsigned i, j;
+
LP_DBG(DEBUG_RAST, "%s 0x%x\n", __FUNCTION__, arg.clear_zstencil);
- for (i = 0; i < TILE_SIZE * TILE_SIZE; i++)
- depth_tile[i] = arg.clear_zstencil;
+ assert(rast->zsbuf_map);
+ if (!rast->zsbuf_map)
+ return;
+
+ LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__);
+
+ /*
+ * Clear the aera of the swizzled depth/depth buffer matching this tile, in
+ * stripes of TILE_VECTOR_HEIGHT x TILE_SIZE at a time.
+ *
+ * The swizzled depth format is such that the depths for
+ * TILE_VECTOR_HEIGHT x TILE_VECTOR_WIDTH pixels have consecutive offsets.
+ */
+
+ dst = lp_rast_depth_pointer(rast, tile_x, tile_y);
+
+ switch (block_size) {
+ case 1:
+ memset(dst, (uint8_t) arg.clear_zstencil, height * width);
+ break;
+ case 2:
+ for (i = 0; i < height; i++) {
+ uint16_t *row = (uint16_t *)dst;
+ for (j = 0; j < width; j++)
+ *row++ = (uint16_t) arg.clear_zstencil;
+ dst += dst_stride;
+ }
+ break;
+ case 4:
+ for (i = 0; i < height; i++) {
+ uint32_t *row = (uint32_t *)dst;
+ for (j = 0; j < width; j++)
+ *row++ = arg.clear_zstencil;
+ dst += dst_stride;
+ }
+ break;
+ default:
+ assert(0);
+ break;
+ }
}
* Load tile color from the framebuffer surface.
* This is a bin command called during bin processing.
*/
-void lp_rast_load_color( struct lp_rasterizer *rast,
- unsigned thread_index,
- const union lp_rast_cmd_arg arg)
+void
+lp_rast_load_color(struct lp_rasterizer_task *task,
+ const union lp_rast_cmd_arg arg)
{
- struct lp_rasterizer_task *task = &rast->tasks[thread_index];
- const unsigned x = task->x;
- const unsigned y = task->y;
+ struct lp_rasterizer *rast = task->rast;
+ const unsigned x = task->x, y = task->y;
unsigned i;
LP_DBG(DEBUG_RAST, "%s at %u, %u\n", __FUNCTION__, x, y);
int w = TILE_SIZE;
int h = TILE_SIZE;
- if (x >= transfer->width)
- continue;
-
- if (y >= transfer->height)
+ if (x >= transfer->width || y >= transfer->height)
continue;
- /* XXX: require tile-size aligned render target dimensions:
- */
- if (x + w > transfer->width)
- w -= x + w - transfer->width;
-
- if (y + h > transfer->height)
- h -= y + h - transfer->height;
assert(w >= 0);
assert(h >= 0);
assert(h <= TILE_SIZE);
lp_tile_read_4ub(transfer->texture->format,
- rast->tasks[thread_index].tile.color[i],
+ task->tile.color[i],
rast->cbuf_map[i],
transfer->stride,
x, y,
w, h);
- }
-}
-
-static void
-lp_tile_read_z32(uint32_t *tile,
- const uint8_t *map,
- unsigned map_stride,
- unsigned x0, unsigned y0, unsigned w, unsigned h)
-{
- unsigned x, y;
- const uint8_t *map_row = map + y0*map_stride;
- for (y = 0; y < h; ++y) {
- const uint32_t *map_pixel = (uint32_t *)(map_row + x0*4);
- for (x = 0; x < w; ++x) {
- *tile++ = *map_pixel++;
- }
- map_row += map_stride;
+ LP_COUNT(nr_color_tile_load);
}
}
-/**
- * Load tile z/stencil from the framebuffer surface.
- * This is a bin command called during bin processing.
- */
-void lp_rast_load_zstencil( struct lp_rasterizer *rast,
- unsigned thread_index,
- const union lp_rast_cmd_arg arg )
-{
- const unsigned x = rast->tasks[thread_index].x;
- const unsigned y = rast->tasks[thread_index].y;
- unsigned w = TILE_SIZE;
- unsigned h = TILE_SIZE;
-
- if (x + w > rast->state.fb.width)
- w -= x + w - rast->state.fb.width;
-
- if (y + h > rast->state.fb.height)
- h -= y + h - rast->state.fb.height;
-
- LP_DBG(DEBUG_RAST, "%s %d,%d %dx%d\n", __FUNCTION__, x, y, w, h);
-
- assert(rast->zsbuf_transfer->texture->format == PIPE_FORMAT_Z32_UNORM);
- lp_tile_read_z32(rast->tasks[thread_index].tile.depth,
- rast->zsbuf_map,
- rast->zsbuf_transfer->stride,
- x, y, w, h);
-}
-
-void lp_rast_set_state( struct lp_rasterizer *rast,
- unsigned thread_index,
- const union lp_rast_cmd_arg arg )
+void
+lp_rast_set_state(struct lp_rasterizer_task *task,
+ const union lp_rast_cmd_arg arg)
{
const struct lp_rast_state *state = arg.set_state;
LP_DBG(DEBUG_RAST, "%s %p\n", __FUNCTION__, (void *) state);
/* just set the current state pointer for this rasterizer */
- rast->tasks[thread_index].current_state = state;
+ task->current_state = state;
}
-/* Within a tile:
- */
-
/**
* Run the shader on all blocks in a tile. This is used when a tile is
* completely contained inside a triangle.
* This is a bin command called during bin processing.
*/
-void lp_rast_shade_tile( struct lp_rasterizer *rast,
- unsigned thread_index,
- const union lp_rast_cmd_arg arg )
+void
+lp_rast_shade_tile(struct lp_rasterizer_task *task,
+ const union lp_rast_cmd_arg arg)
{
- /* Set c1,c2,c3 to large values so the in/out test always passes */
- const int32_t c1 = INT_MIN, c2 = INT_MIN, c3 = INT_MIN;
+ struct lp_rasterizer *rast = task->rast;
+ const struct lp_rast_state *state = task->current_state;
+ struct lp_rast_tile *tile = &task->tile;
const struct lp_rast_shader_inputs *inputs = arg.shade_tile;
- const unsigned tile_x = rast->tasks[thread_index].x;
- const unsigned tile_y = rast->tasks[thread_index].y;
+ const unsigned tile_x = task->x, tile_y = task->y;
unsigned x, y;
LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__);
- /* Use the existing preference for 4x4 (four quads) shading:
- */
- for (y = 0; y < TILE_SIZE; y += 4)
- for (x = 0; x < TILE_SIZE; x += 4)
- lp_rast_shade_quads( rast,
- thread_index,
- inputs,
- tile_x + x,
- tile_y + y,
- c1, c2, c3);
+ /* render the whole 64x64 tile in 4x4 chunks */
+ for (y = 0; y < TILE_SIZE; y += 4){
+ for (x = 0; x < TILE_SIZE; x += 4) {
+ uint8_t *color[PIPE_MAX_COLOR_BUFS];
+ uint32_t *depth;
+ unsigned block_offset, i;
+
+ /* offset of the 16x16 pixel block within the tile */
+ block_offset = ((y / 4) * (16 * 16) + (x / 4) * 16);
+
+ /* color buffer */
+ for (i = 0; i < rast->state.fb.nr_cbufs; i++)
+ color[i] = tile->color[i] + 4 * block_offset;
+
+ /* depth buffer */
+ depth = lp_rast_depth_pointer(rast, tile_x + x, tile_y + y);
+
+ /* run shader */
+ state->jit_function[0]( &state->jit_context,
+ tile_x + x, tile_y + y,
+ inputs->a0,
+ inputs->dadx,
+ inputs->dady,
+ color,
+ depth,
+ INT_MIN, INT_MIN, INT_MIN,
+ NULL, NULL, NULL );
+ }
+ }
}
* Compute shading for a 4x4 block of pixels.
* This is a bin command called during bin processing.
*/
-void lp_rast_shade_quads( struct lp_rasterizer *rast,
- unsigned thread_index,
+void lp_rast_shade_quads( struct lp_rasterizer_task *task,
const struct lp_rast_shader_inputs *inputs,
unsigned x, unsigned y,
int32_t c1, int32_t c2, int32_t c3)
{
- const struct lp_rast_state *state = rast->tasks[thread_index].current_state;
- struct lp_rast_tile *tile = &rast->tasks[thread_index].tile;
+ const struct lp_rast_state *state = task->current_state;
+ struct lp_rasterizer *rast = task->rast;
+ struct lp_rast_tile *tile = &task->tile;
uint8_t *color[PIPE_MAX_COLOR_BUFS];
void *depth;
unsigned i;
iy = y % TILE_SIZE;
/* offset of the 16x16 pixel block within the tile */
- block_offset = ((iy/4)*(16*16) + (ix/4)*16);
+ block_offset = ((iy / 4) * (16 * 16) + (ix / 4) * 16);
/* color buffer */
for (i = 0; i < rast->state.fb.nr_cbufs; i++)
color[i] = tile->color[i] + 4 * block_offset;
/* depth buffer */
- depth = tile->depth + block_offset;
+ depth = lp_rast_depth_pointer(rast, x, y);
#ifdef DEBUG
- assert(lp_check_alignment(tile->depth, 16));
assert(lp_check_alignment(tile->color[0], 16));
assert(lp_check_alignment(state->jit_context.blend_color, 16));
#endif
/* run shader */
- state->jit_function( &state->jit_context,
+ state->jit_function[1]( &state->jit_context,
x, y,
inputs->a0,
inputs->dadx,
}
-/* End of tile:
+/**
+ * Set top row and left column of the tile's pixels to white. For debugging.
+ */
+static void
+outline_tile(uint8_t *tile)
+{
+ const uint8_t val = 0xff;
+ unsigned i;
+
+ for (i = 0; i < TILE_SIZE; i++) {
+ TILE_PIXEL(tile, i, 0, 0) = val;
+ TILE_PIXEL(tile, i, 0, 1) = val;
+ TILE_PIXEL(tile, i, 0, 2) = val;
+ TILE_PIXEL(tile, i, 0, 3) = val;
+
+ TILE_PIXEL(tile, 0, i, 0) = val;
+ TILE_PIXEL(tile, 0, i, 1) = val;
+ TILE_PIXEL(tile, 0, i, 2) = val;
+ TILE_PIXEL(tile, 0, i, 3) = val;
+ }
+}
+
+
+/**
+ * Draw grid of gray lines at 16-pixel intervals across the tile to
+ * show the sub-tile boundaries. For debugging.
*/
+static void
+outline_subtiles(uint8_t *tile)
+{
+ const uint8_t val = 0x80;
+ const unsigned step = 16;
+ unsigned i, j;
+
+ for (i = 0; i < TILE_SIZE; i += step) {
+ for (j = 0; j < TILE_SIZE; j++) {
+ TILE_PIXEL(tile, i, j, 0) = val;
+ TILE_PIXEL(tile, i, j, 1) = val;
+ TILE_PIXEL(tile, i, j, 2) = val;
+ TILE_PIXEL(tile, i, j, 3) = val;
+
+ TILE_PIXEL(tile, j, i, 0) = val;
+ TILE_PIXEL(tile, j, i, 1) = val;
+ TILE_PIXEL(tile, j, i, 2) = val;
+ TILE_PIXEL(tile, j, i, 3) = val;
+ }
+ }
+
+ outline_tile(tile);
+}
+
/**
* Write the rasterizer's color tile to the framebuffer.
*/
-static void lp_rast_store_color( struct lp_rasterizer *rast,
- unsigned thread_index)
+static void
+lp_rast_store_color(struct lp_rasterizer_task *task)
{
- const unsigned x = rast->tasks[thread_index].x;
- const unsigned y = rast->tasks[thread_index].y;
+ struct lp_rasterizer *rast = task->rast;
+ const unsigned x = task->x, y = task->y;
unsigned i;
for (i = 0; i < rast->state.fb.nr_cbufs; i++) {
if (y >= transfer->height)
continue;
- /* XXX: require tile-size aligned render target dimensions:
- */
- if (x + w > transfer->width)
- w -= x + w - transfer->width;
-
- if (y + h > transfer->height)
- h -= y + h - transfer->height;
-
- assert(w >= 0);
- assert(h >= 0);
- assert(w <= TILE_SIZE);
- assert(h <= TILE_SIZE);
-
LP_DBG(DEBUG_RAST, "%s [%u] %d,%d %dx%d\n", __FUNCTION__,
- thread_index, x, y, w, h);
+ task->thread_index, x, y, w, h);
+
+ if (LP_DEBUG & DEBUG_SHOW_SUBTILES)
+ outline_subtiles(task->tile.color[i]);
+ else if (LP_DEBUG & DEBUG_SHOW_TILES)
+ outline_tile(task->tile.color[i]);
lp_tile_write_4ub(transfer->texture->format,
- rast->tasks[thread_index].tile.color[i],
+ task->tile.color[i],
rast->cbuf_map[i],
transfer->stride,
x, y,
w, h);
- }
-}
-
-static void
-lp_tile_write_z32(const uint32_t *src, uint8_t *dst, unsigned dst_stride,
- unsigned x0, unsigned y0, unsigned w, unsigned h)
-{
- unsigned x, y;
- uint8_t *dst_row = dst + y0*dst_stride;
- for (y = 0; y < h; ++y) {
- uint32_t *dst_pixel = (uint32_t *)(dst_row + x0*4);
- for (x = 0; x < w; ++x) {
- *dst_pixel++ = *src++;
- }
- dst_row += dst_stride;
+ LP_COUNT(nr_color_tile_store);
}
}
-/**
- * Write the rasterizer's z/stencil tile to the framebuffer.
- */
-static void lp_rast_store_zstencil( struct lp_rasterizer *rast,
- unsigned thread_index )
-{
- const unsigned x = rast->tasks[thread_index].x;
- const unsigned y = rast->tasks[thread_index].y;
- unsigned w = TILE_SIZE;
- unsigned h = TILE_SIZE;
-
- if (x + w > rast->state.fb.width)
- w -= x + w - rast->state.fb.width;
-
- if (y + h > rast->state.fb.height)
- h -= y + h - rast->state.fb.height;
-
- LP_DBG(DEBUG_RAST, "%s %d,%d %dx%d\n", __FUNCTION__, x, y, w, h);
-
- assert(rast->zsbuf_transfer->texture->format == PIPE_FORMAT_Z32_UNORM);
- lp_tile_write_z32(rast->tasks[thread_index].tile.depth,
- rast->zsbuf_map,
- rast->zsbuf_transfer->stride,
- x, y, w, h);
-}
-
/**
* Write the rasterizer's tiles to the framebuffer.
*/
static void
-lp_rast_end_tile( struct lp_rasterizer *rast,
- unsigned thread_index )
+lp_rast_end_tile(struct lp_rasterizer_task *task)
{
+ struct lp_rasterizer *rast = task->rast;
+
LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__);
if (rast->state.write_color)
- lp_rast_store_color(rast, thread_index);
-
- if (rast->state.write_zstencil)
- lp_rast_store_zstencil(rast, thread_index);
+ lp_rast_store_color(task);
}
* Signal on a fence. This is called during bin execution/rasterization.
* Called per thread.
*/
-void lp_rast_fence( struct lp_rasterizer *rast,
- unsigned thread_index,
- const union lp_rast_cmd_arg arg )
+void
+lp_rast_fence(struct lp_rasterizer_task *task,
+ const union lp_rast_cmd_arg arg)
{
struct lp_fence *fence = arg.fence;
util_unreference_framebuffer_state( &scene->fb );
lp_scene_reset( scene );
+
+ assert(lp_scene_is_empty(scene));
+
lp_scene_enqueue( rast->empty_scenes, scene );
rast->curr_scene = NULL;
}
* Called per thread.
*/
static void
-rasterize_bin( struct lp_rasterizer *rast,
- unsigned thread_index,
- const struct cmd_bin *bin,
- int x, int y)
+rasterize_bin(struct lp_rasterizer_task *task,
+ const struct cmd_bin *bin,
+ int x, int y)
{
const struct cmd_block_list *commands = &bin->commands;
struct cmd_block *block;
unsigned k;
- lp_rast_start_tile( rast, thread_index, x, y );
+ lp_rast_start_tile( task, x, y );
/* simply execute each of the commands in the block list */
for (block = commands->head; block; block = block->next) {
for (k = 0; k < block->count; k++) {
- block->cmd[k]( rast, thread_index, block->arg[k] );
+ block->cmd[k]( task, block->arg[k] );
}
}
- lp_rast_end_tile( rast, thread_index );
+ lp_rast_end_tile( task );
}
} cmd_names[] =
{
RAST(load_color),
- RAST(load_zstencil),
RAST(clear_color),
RAST(clear_zstencil),
RAST(triangle),
for (i = 0; i < head->count; i++)
if (head->cmd[i] != lp_rast_load_color &&
- head->cmd[i] != lp_rast_load_zstencil &&
head->cmd[i] != lp_rast_set_state) {
return FALSE;
}
* Called per thread.
*/
static void
-rasterize_scene( struct lp_rasterizer *rast,
- unsigned thread_index,
+rasterize_scene(struct lp_rasterizer_task *task,
struct lp_scene *scene,
- bool write_depth )
+ bool write_depth)
{
/* loop over scene bins, rasterize each */
#if 0
unsigned i, j;
for (i = 0; i < scene->tiles_x; i++) {
for (j = 0; j < scene->tiles_y; j++) {
- struct cmd_bin *bin = lp_get_bin(scene, i, j);
- rasterize_bin( rast, thread_index,
- bin, i * TILE_SIZE, j * TILE_SIZE );
+ struct cmd_bin *bin = lp_scene_get_bin(scene, i, j);
+ rasterize_bin(task, bin, i * TILE_SIZE, j * TILE_SIZE);
}
}
}
assert(scene);
while ((bin = lp_scene_bin_iter_next(scene, &x, &y))) {
if (!is_empty_bin( bin ))
- rasterize_bin( rast, thread_index, bin, x * TILE_SIZE, y * TILE_SIZE);
+ rasterize_bin(task, bin, x * TILE_SIZE, y * TILE_SIZE);
}
}
#endif
if (debug) {
unsigned x, y;
- printf("rasterize scene:\n");
- printf(" data size: %u\n", lp_scene_data_size(scene));
+ debug_printf("rasterize scene:\n");
+ debug_printf(" data size: %u\n", lp_scene_data_size(scene));
for (y = 0; y < scene->tiles_y; y++) {
for (x = 0; x < scene->tiles_x; x++) {
- printf(" bin %u, %u size: %u\n", x, y,
- lp_scene_bin_size(scene, x, y));
+ debug_printf(" bin %u, %u size: %u\n", x, y,
+ lp_scene_bin_size(scene, x, y));
}
}
}
fb->zsbuf != NULL && write_depth );
lp_scene_bin_iter_begin( scene );
- rasterize_scene( rast, 0, scene, write_depth );
+ rasterize_scene( &rast->tasks[0], scene, write_depth );
release_scene( rast, scene );
* 2. do work
* 3. signal that we're done
*/
-static void *
-thread_func( void *init_data )
+static PIPE_THREAD_ROUTINE( thread_func, init_data )
{
struct lp_rasterizer_task *task = (struct lp_rasterizer_task *) init_data;
struct lp_rasterizer *rast = task->rast;
debug_printf("thread %d waiting for work\n", task->thread_index);
pipe_semaphore_wait(&task->work_ready);
+ if (rast->exit_flag)
+ break;
+
if (task->thread_index == 0) {
/* thread[0]:
* - get next scene to rasterize
const struct pipe_framebuffer_state *fb;
boolean write_depth;
- rast->curr_scene = lp_scene_dequeue( rast->full_scenes );
+ rast->curr_scene = lp_scene_dequeue( rast->full_scenes, TRUE );
lp_scene_bin_iter_begin( rast->curr_scene );
/* do work */
if (debug)
debug_printf("thread %d doing work\n", task->thread_index);
- rasterize_scene(rast,
- task->thread_index,
- rast->curr_scene,
- rast->curr_scene->write_depth);
+ rasterize_scene(task,
+ rast->curr_scene,
+ rast->curr_scene->write_depth);
/* wait for all threads to finish with this scene */
pipe_barrier_wait( &rast->barrier );
{
unsigned i;
+#ifdef PIPE_OS_WINDOWS
+ /* Multithreading not supported on windows until conditions and barriers are
+ * properly implemented. */
+ rast->num_threads = 0;
+#else
rast->num_threads = util_cpu_caps.nr_cpus;
rast->num_threads = debug_get_num_option("LP_NUM_THREADS", rast->num_threads);
rast->num_threads = MIN2(rast->num_threads, MAX_THREADS);
+#endif
/* NOTE: if num_threads is zero, we won't use any threads */
for (i = 0; i < rast->num_threads; i++) {
rast->full_scenes = lp_scene_queue_create();
for (i = 0; i < Elements(rast->tasks); i++) {
+ struct lp_rasterizer_task *task = &rast->tasks[i];
+
for (cbuf = 0; cbuf < PIPE_MAX_COLOR_BUFS; cbuf++ )
- rast->tasks[i].tile.color[cbuf] = align_malloc( TILE_SIZE*TILE_SIZE*4, 16 );
+ task->tile.color[cbuf] = align_malloc(TILE_SIZE * TILE_SIZE * 4, 16);
- rast->tasks[i].tile.depth = align_malloc( TILE_SIZE*TILE_SIZE*4, 16 );
- rast->tasks[i].rast = rast;
- rast->tasks[i].thread_index = i;
+ task->rast = rast;
+ task->thread_index = i;
}
create_rast_threads(rast);
util_unreference_framebuffer_state(&rast->state.fb);
for (i = 0; i < Elements(rast->tasks); i++) {
- align_free(rast->tasks[i].tile.depth);
for (cbuf = 0; cbuf < PIPE_MAX_COLOR_BUFS; cbuf++ )
align_free(rast->tasks[i].tile.color[cbuf]);
}
+ /* Set exit_flag and signal each thread's work_ready semaphore.
+ * Each thread will be woken up, notice that the exit_flag is set and
+ * break out of its main loop. The thread will then exit.
+ */
+ rast->exit_flag = TRUE;
+ for (i = 0; i < rast->num_threads; i++) {
+ pipe_semaphore_signal(&rast->tasks[i].work_ready);
+ }
+
+ for (i = 0; i < rast->num_threads; i++) {
+ pipe_semaphore_destroy(&rast->tasks[i].work_ready);
+ pipe_semaphore_destroy(&rast->tasks[i].work_done);
+ }
+
/* for synchronizing rasterization threads */
pipe_barrier_destroy( &rast->barrier );