#include <limits.h>
#include "util/u_memory.h"
#include "util/u_math.h"
-#include "util/u_cpu_detect.h"
+#include "util/u_rect.h"
#include "util/u_surface.h"
+#include "util/u_pack_color.h"
+#include "util/u_string.h"
+#include "util/u_thread.h"
+
+#include "util/os_time.h"
#include "lp_scene_queue.h"
+#include "lp_context.h"
#include "lp_debug.h"
#include "lp_fence.h"
+#include "lp_perf.h"
+#include "lp_query.h"
#include "lp_rast.h"
#include "lp_rast_priv.h"
-#include "lp_tile_soa.h"
-#include "lp_bld_debug.h"
+#include "gallivm/lp_bld_format.h"
+#include "gallivm/lp_bld_debug.h"
#include "lp_scene.h"
+#include "lp_tex_sample.h"
+
+
+#ifdef DEBUG
+int jit_line = 0;
+const struct lp_rast_state *jit_state = NULL;
+const struct lp_rasterizer_task *jit_task = NULL;
+#endif
/**
- * Begin the rasterization phase.
- * Map the framebuffer surfaces. Initialize the 'rast' state.
+ * Begin rasterizing a scene.
+ * Called once per scene by one thread.
*/
-static boolean
+static void
lp_rast_begin( struct lp_rasterizer *rast,
- const struct pipe_framebuffer_state *fb,
- boolean write_color,
- boolean write_zstencil )
+ struct lp_scene *scene )
{
- struct pipe_screen *screen = rast->screen;
- struct pipe_surface *cbuf, *zsbuf;
- int i;
+ rast->curr_scene = scene;
LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__);
- util_copy_framebuffer_state(&rast->state.fb, fb);
-
- rast->state.write_zstencil = write_zstencil;
- rast->state.write_color = write_color;
-
- rast->check_for_clipped_tiles = (fb->width % TILE_SIZE != 0 ||
- fb->height % TILE_SIZE != 0);
-
-
- for (i = 0; i < rast->state.fb.nr_cbufs; i++) {
- cbuf = rast->state.fb.cbufs[i];
- if (cbuf) {
- rast->cbuf_transfer[i] = screen->get_tex_transfer(rast->screen,
- cbuf->texture,
- cbuf->face,
- cbuf->level,
- cbuf->zslice,
- PIPE_TRANSFER_READ_WRITE,
- 0, 0,
- cbuf->width,
- cbuf->height);
- if (!rast->cbuf_transfer[i])
- goto fail;
-
- rast->cbuf_map[i] = screen->transfer_map(rast->screen,
- rast->cbuf_transfer[i]);
- if (!rast->cbuf_map[i])
- goto fail;
- }
- }
-
- zsbuf = rast->state.fb.zsbuf;
- if (zsbuf) {
- rast->zsbuf_transfer = screen->get_tex_transfer(rast->screen,
- zsbuf->texture,
- zsbuf->face,
- zsbuf->level,
- zsbuf->zslice,
- PIPE_TRANSFER_READ_WRITE,
- 0, 0,
- zsbuf->width,
- zsbuf->height);
- if (!rast->zsbuf_transfer)
- goto fail;
-
- rast->zsbuf_map = screen->transfer_map(rast->screen,
- rast->zsbuf_transfer);
- if (!rast->zsbuf_map)
- goto fail;
- }
-
- return TRUE;
-
-fail:
- /* Unmap and release transfers?
- */
- return FALSE;
+ lp_scene_begin_rasterization( scene );
+ lp_scene_bin_iter_begin( scene );
}
-/**
- * Finish the rasterization phase.
- * Unmap framebuffer surfaces.
- */
static void
lp_rast_end( struct lp_rasterizer *rast )
{
- struct pipe_screen *screen = rast->screen;
- unsigned i;
-
- for (i = 0; i < rast->state.fb.nr_cbufs; i++) {
- if (rast->cbuf_map[i])
- screen->transfer_unmap(screen, rast->cbuf_transfer[i]);
-
- if (rast->cbuf_transfer[i])
- screen->tex_transfer_destroy(rast->cbuf_transfer[i]);
+ lp_scene_end_rasterization( rast->curr_scene );
- rast->cbuf_transfer[i] = NULL;
- rast->cbuf_map[i] = NULL;
- }
-
- if (rast->zsbuf_map)
- screen->transfer_unmap(screen, rast->zsbuf_transfer);
-
- if (rast->zsbuf_transfer)
- screen->tex_transfer_destroy(rast->zsbuf_transfer);
-
- rast->zsbuf_transfer = NULL;
- rast->zsbuf_map = NULL;
+ rast->curr_scene = NULL;
}
/**
- * Begining rasterization of a tile.
+ * Beginning rasterization of a tile.
* \param x window X position of the tile, in pixels
* \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_DBG(DEBUG_RAST, "%s %d,%d\n", __FUNCTION__, x, y);
-
- rast->tasks[thread_index].x = x;
- rast->tasks[thread_index].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 )
+lp_rast_tile_begin(struct lp_rasterizer_task *task,
+ const struct cmd_bin *bin,
+ int x, int y)
{
- const uint8_t *clear_color = arg.clear_color;
- uint8_t **color_tile = rast->tasks[thread_index].tile.color;
unsigned i;
+ struct lp_scene *scene = task->scene;
+
+ LP_DBG(DEBUG_RAST, "%s %d,%d\n", __FUNCTION__, x, y);
- LP_DBG(DEBUG_RAST, "%s 0x%x,0x%x,0x%x,0x%x\n", __FUNCTION__,
- clear_color[0],
- clear_color[1],
- clear_color[2],
- clear_color[3]);
-
- if (clear_color[0] == clear_color[1] &&
- clear_color[1] == clear_color[2] &&
- clear_color[2] == clear_color[3]) {
- /* clear to grayscale value {x, x, x, x} */
- for (i = 0; i < rast->state.fb.nr_cbufs; i++) {
- memset(color_tile[i], clear_color[0], TILE_SIZE * TILE_SIZE * 4);
+ task->bin = bin;
+ task->x = x * TILE_SIZE;
+ task->y = y * TILE_SIZE;
+ task->width = TILE_SIZE + x * TILE_SIZE > task->scene->fb.width ?
+ task->scene->fb.width - x * TILE_SIZE : TILE_SIZE;
+ task->height = TILE_SIZE + y * TILE_SIZE > task->scene->fb.height ?
+ task->scene->fb.height - y * TILE_SIZE : TILE_SIZE;
+
+ task->thread_data.vis_counter = 0;
+ task->ps_invocations = 0;
+
+ for (i = 0; i < task->scene->fb.nr_cbufs; i++) {
+ if (task->scene->fb.cbufs[i]) {
+ task->color_tiles[i] = scene->cbufs[i].map +
+ scene->cbufs[i].stride * task->y +
+ scene->cbufs[i].format_bytes * task->x;
}
}
- else {
- /* Non-gray color.
- * Note: if the swizzled tile layout changes (see TILE_PIXEL) this code
- * will need to change. It'll be pretty obvious when clearing no longer
- * works.
- */
- const unsigned chunk = TILE_SIZE / 4;
- for (i = 0; i < rast->state.fb.nr_cbufs; i++) {
- uint8_t *c = color_tile[i];
- unsigned j;
- for (j = 0; j < 4 * TILE_SIZE; j++) {
- memset(c, clear_color[0], chunk);
- c += chunk;
- memset(c, clear_color[1], chunk);
- c += chunk;
- memset(c, clear_color[2], chunk);
- c += chunk;
- memset(c, clear_color[3], chunk);
- c += chunk;
- }
- assert(c - color_tile[i] == TILE_SIZE * TILE_SIZE * 4);
- }
+ if (task->scene->fb.zsbuf) {
+ task->depth_tile = scene->zsbuf.map +
+ scene->zsbuf.stride * task->y +
+ scene->zsbuf.format_bytes * task->x;
}
}
/**
- * Clear the rasterizer's current z/stencil tile.
+ * Clear the rasterizer's current color tile.
* This is a bin command called during bin processing.
+ * Clear commands always clear all bound layers.
*/
-void lp_rast_clear_zstencil( struct lp_rasterizer *rast,
- unsigned thread_index,
- const union lp_rast_cmd_arg arg)
+static void
+lp_rast_clear_color(struct lp_rasterizer_task *task,
+ const union lp_rast_cmd_arg arg)
{
- unsigned i;
- uint32_t *depth_tile = rast->tasks[thread_index].tile.depth;
-
- LP_DBG(DEBUG_RAST, "%s 0x%x\n", __FUNCTION__, arg.clear_zstencil);
+ const struct lp_scene *scene = task->scene;
+ unsigned cbuf = arg.clear_rb->cbuf;
+ union util_color uc;
+ enum pipe_format format;
+
+ /* we never bin clear commands for non-existing buffers */
+ assert(cbuf < scene->fb.nr_cbufs);
+ assert(scene->fb.cbufs[cbuf]);
+
+ format = scene->fb.cbufs[cbuf]->format;
+ uc = arg.clear_rb->color_val;
- for (i = 0; i < TILE_SIZE * TILE_SIZE; i++)
- depth_tile[i] = arg.clear_zstencil;
+ /*
+ * this is pretty rough since we have target format (bunch of bytes...) here.
+ * dump it as raw 4 dwords.
+ */
+ LP_DBG(DEBUG_RAST, "%s clear value (target format %d) raw 0x%x,0x%x,0x%x,0x%x\n",
+ __FUNCTION__, format, uc.ui[0], uc.ui[1], uc.ui[2], uc.ui[3]);
+
+
+ util_fill_box(scene->cbufs[cbuf].map,
+ format,
+ scene->cbufs[cbuf].stride,
+ scene->cbufs[cbuf].layer_stride,
+ task->x,
+ task->y,
+ 0,
+ task->width,
+ task->height,
+ scene->fb_max_layer + 1,
+ &uc);
+
+ /* this will increase for each rb which probably doesn't mean much */
+ LP_COUNT(nr_color_tile_clear);
}
/**
- * Load tile color from the framebuffer surface.
+ * Clear the rasterizer's current z/stencil tile.
* This is a bin command called during bin processing.
+ * Clear commands always clear all bound layers.
*/
-void lp_rast_load_color( struct lp_rasterizer *rast,
- unsigned thread_index,
- const union lp_rast_cmd_arg arg)
+static void
+lp_rast_clear_zstencil(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;
- unsigned i;
-
- LP_DBG(DEBUG_RAST, "%s at %u, %u\n", __FUNCTION__, x, y);
+ const struct lp_scene *scene = task->scene;
+ uint64_t clear_value64 = arg.clear_zstencil.value;
+ uint64_t clear_mask64 = arg.clear_zstencil.mask;
+ uint32_t clear_value = (uint32_t) clear_value64;
+ uint32_t clear_mask = (uint32_t) clear_mask64;
+ const unsigned height = task->height;
+ const unsigned width = task->width;
+ const unsigned dst_stride = scene->zsbuf.stride;
+ uint8_t *dst;
+ unsigned i, j;
+ unsigned block_size;
- for (i = 0; i < rast->state.fb.nr_cbufs; i++) {
- struct pipe_transfer *transfer = rast->cbuf_transfer[i];
- int w = TILE_SIZE;
- int h = TILE_SIZE;
+ LP_DBG(DEBUG_RAST, "%s: value=0x%08x, mask=0x%08x\n",
+ __FUNCTION__, clear_value, clear_mask);
- if (x >= transfer->width)
- continue;
+ /*
+ * Clear the area of the depth/depth buffer matching this tile.
+ */
- if (y >= transfer->height)
- continue;
+ if (scene->fb.zsbuf) {
+ unsigned layer;
+ uint8_t *dst_layer = task->depth_tile;
+ block_size = util_format_get_blocksize(scene->fb.zsbuf->format);
- assert(w >= 0);
- assert(h >= 0);
- assert(w <= TILE_SIZE);
- assert(h <= TILE_SIZE);
+ clear_value &= clear_mask;
- lp_tile_read_4ub(transfer->texture->format,
- task->tile.color[i],
- rast->cbuf_map[i],
- transfer->stride,
- x, y,
- w, h);
- }
-}
+ for (layer = 0; layer <= scene->fb_max_layer; layer++) {
+ dst = dst_layer;
+ switch (block_size) {
+ case 1:
+ assert(clear_mask == 0xff);
+ memset(dst, (uint8_t) clear_value, height * width);
+ break;
+ case 2:
+ if (clear_mask == 0xffff) {
+ for (i = 0; i < height; i++) {
+ uint16_t *row = (uint16_t *)dst;
+ for (j = 0; j < width; j++)
+ *row++ = (uint16_t) clear_value;
+ dst += dst_stride;
+ }
+ }
+ else {
+ for (i = 0; i < height; i++) {
+ uint16_t *row = (uint16_t *)dst;
+ for (j = 0; j < width; j++) {
+ uint16_t tmp = ~clear_mask & *row;
+ *row++ = clear_value | tmp;
+ }
+ dst += dst_stride;
+ }
+ }
+ break;
+ case 4:
+ if (clear_mask == 0xffffffff) {
+ for (i = 0; i < height; i++) {
+ uint32_t *row = (uint32_t *)dst;
+ for (j = 0; j < width; j++)
+ *row++ = clear_value;
+ dst += dst_stride;
+ }
+ }
+ else {
+ for (i = 0; i < height; i++) {
+ uint32_t *row = (uint32_t *)dst;
+ for (j = 0; j < width; j++) {
+ uint32_t tmp = ~clear_mask & *row;
+ *row++ = clear_value | tmp;
+ }
+ dst += dst_stride;
+ }
+ }
+ break;
+ case 8:
+ clear_value64 &= clear_mask64;
+ if (clear_mask64 == 0xffffffffffULL) {
+ for (i = 0; i < height; i++) {
+ uint64_t *row = (uint64_t *)dst;
+ for (j = 0; j < width; j++)
+ *row++ = clear_value64;
+ dst += dst_stride;
+ }
+ }
+ else {
+ for (i = 0; i < height; i++) {
+ uint64_t *row = (uint64_t *)dst;
+ for (j = 0; j < width; j++) {
+ uint64_t tmp = ~clear_mask64 & *row;
+ *row++ = clear_value64 | tmp;
+ }
+ dst += dst_stride;
+ }
+ }
+ break;
-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++;
+ default:
+ assert(0);
+ break;
+ }
+ dst_layer += scene->zsbuf.layer_stride;
}
- map_row += map_stride;
}
}
+
+
/**
- * Load tile z/stencil from the framebuffer surface.
+ * 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_load_zstencil( struct lp_rasterizer *rast,
- unsigned thread_index,
- const union lp_rast_cmd_arg arg )
+static void
+lp_rast_shade_tile(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;
- unsigned w = TILE_SIZE;
- unsigned h = TILE_SIZE;
-
- if (x + w > rast->state.fb.width)
- w -= x + w - rast->state.fb.width;
+ const struct lp_scene *scene = task->scene;
+ const struct lp_rast_shader_inputs *inputs = arg.shade_tile;
+ const struct lp_rast_state *state;
+ struct lp_fragment_shader_variant *variant;
+ const unsigned tile_x = task->x, tile_y = task->y;
+ unsigned x, y;
- if (y + h > rast->state.fb.height)
- h -= y + h - rast->state.fb.height;
+ if (inputs->disable) {
+ /* This command was partially binned and has been disabled */
+ return;
+ }
- LP_DBG(DEBUG_RAST, "%s %d,%d %dx%d\n", __FUNCTION__, x, y, w, h);
+ LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__);
- assert(rast->zsbuf_transfer->texture->format == PIPE_FORMAT_Z32_UNORM);
- lp_tile_read_z32(task->tile.depth,
- rast->zsbuf_map,
- rast->zsbuf_transfer->stride,
- x, y, w, h);
-}
+ state = task->state;
+ assert(state);
+ if (!state) {
+ return;
+ }
+ variant = state->variant;
+ /* render the whole 64x64 tile in 4x4 chunks */
+ for (y = 0; y < task->height; y += 4){
+ for (x = 0; x < task->width; x += 4) {
+ uint8_t *color[PIPE_MAX_COLOR_BUFS];
+ unsigned stride[PIPE_MAX_COLOR_BUFS];
+ uint8_t *depth = NULL;
+ unsigned depth_stride = 0;
+ unsigned i;
-void lp_rast_set_state( struct lp_rasterizer *rast,
- unsigned thread_index,
- const union lp_rast_cmd_arg arg )
-{
- const struct lp_rast_state *state = arg.set_state;
+ /* color buffer */
+ for (i = 0; i < scene->fb.nr_cbufs; i++){
+ if (scene->fb.cbufs[i]) {
+ stride[i] = scene->cbufs[i].stride;
+ color[i] = lp_rast_get_color_block_pointer(task, i, tile_x + x,
+ tile_y + y, inputs->layer);
+ }
+ else {
+ stride[i] = 0;
+ color[i] = NULL;
+ }
+ }
- LP_DBG(DEBUG_RAST, "%s %p\n", __FUNCTION__, (void *) state);
+ /* depth buffer */
+ if (scene->zsbuf.map) {
+ depth = lp_rast_get_depth_block_pointer(task, tile_x + x,
+ tile_y + y, inputs->layer);
+ depth_stride = scene->zsbuf.stride;
+ }
- /* just set the current state pointer for this rasterizer */
- rast->tasks[thread_index].current_state = state;
+ /* Propagate non-interpolated raster state. */
+ task->thread_data.raster_state.viewport_index = inputs->viewport_index;
+
+ /* run shader on 4x4 block */
+ BEGIN_JIT_CALL(state, task);
+ variant->jit_function[RAST_WHOLE]( &state->jit_context,
+ tile_x + x, tile_y + y,
+ inputs->frontfacing,
+ GET_A0(inputs),
+ GET_DADX(inputs),
+ GET_DADY(inputs),
+ color,
+ depth,
+ 0xffff,
+ &task->thread_data,
+ stride,
+ depth_stride);
+ END_JIT_CALL();
+ }
+ }
}
-
/**
* Run the shader on all blocks in a tile. This is used when a tile is
- * completely contained inside a triangle.
+ * completely contained inside a triangle, and the shader is opaque.
* 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 )
+static void
+lp_rast_shade_tile_opaque(struct lp_rasterizer_task *task,
+ const union lp_rast_cmd_arg arg)
{
- struct lp_rasterizer_task *task = &rast->tasks[thread_index];
- 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 = task->x;
- const unsigned tile_y = task->y;
- unsigned x, y;
-
LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__);
- /* 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 = tile->depth + block_offset;
-
- /* 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 );
- }
+ assert(task->state);
+ if (!task->state) {
+ return;
}
+
+ lp_rast_shade_tile(task, arg);
}
/**
- * Compute shading for a 4x4 block of pixels.
+ * Compute shading for a 4x4 block of pixels inside a triangle.
* This is a bin command called during bin processing.
+ * \param x X position of quad in window coords
+ * \param y Y position of quad in window coords
*/
-void lp_rast_shade_quads( struct lp_rasterizer *rast,
- unsigned thread_index,
- const struct lp_rast_shader_inputs *inputs,
- unsigned x, unsigned y,
- int32_t c1, int32_t c2, int32_t c3)
+void
+lp_rast_shade_quads_mask(struct lp_rasterizer_task *task,
+ const struct lp_rast_shader_inputs *inputs,
+ unsigned x, unsigned y,
+ unsigned mask)
{
- struct lp_rasterizer_task *task = &rast->tasks[thread_index];
- const struct lp_rast_state *state = task->current_state;
- struct lp_rast_tile *tile = &task->tile;
+ const struct lp_rast_state *state = task->state;
+ struct lp_fragment_shader_variant *variant = state->variant;
+ const struct lp_scene *scene = task->scene;
uint8_t *color[PIPE_MAX_COLOR_BUFS];
- void *depth;
+ unsigned stride[PIPE_MAX_COLOR_BUFS];
+ uint8_t *depth = NULL;
+ unsigned depth_stride = 0;
unsigned i;
- unsigned ix, iy;
- int block_offset;
-#ifdef DEBUG
assert(state);
/* Sanity checks */
+ assert(x < scene->tiles_x * TILE_SIZE);
+ assert(y < scene->tiles_y * TILE_SIZE);
assert(x % TILE_VECTOR_WIDTH == 0);
assert(y % TILE_VECTOR_HEIGHT == 0);
assert((x % 4) == 0);
assert((y % 4) == 0);
-#endif
-
- ix = x % TILE_SIZE;
- iy = y % TILE_SIZE;
-
- /* offset of the 16x16 pixel block within the tile */
- 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;
+ for (i = 0; i < scene->fb.nr_cbufs; i++) {
+ if (scene->fb.cbufs[i]) {
+ stride[i] = scene->cbufs[i].stride;
+ color[i] = lp_rast_get_color_block_pointer(task, i, x, y,
+ inputs->layer);
+ }
+ else {
+ stride[i] = 0;
+ color[i] = NULL;
+ }
+ }
/* depth buffer */
- depth = tile->depth + block_offset;
-
+ if (scene->zsbuf.map) {
+ depth_stride = scene->zsbuf.stride;
+ depth = lp_rast_get_depth_block_pointer(task, x, y, inputs->layer);
+ }
+ assert(lp_check_alignment(state->jit_context.u8_blend_color, 16));
-#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));
-
- assert(lp_check_alignment(inputs->step[0], 16));
- assert(lp_check_alignment(inputs->step[1], 16));
- assert(lp_check_alignment(inputs->step[2], 16));
-#endif
-
- /* run shader */
- state->jit_function[1]( &state->jit_context,
- x, y,
- inputs->a0,
- inputs->dadx,
- inputs->dady,
- color,
- depth,
- c1, c2, c3,
- inputs->step[0], inputs->step[1], inputs->step[2]);
+ /*
+ * The rasterizer may produce fragments outside our
+ * allocated 4x4 blocks hence need to filter them out here.
+ */
+ if ((x % TILE_SIZE) < task->width && (y % TILE_SIZE) < task->height) {
+ /* not very accurate would need a popcount on the mask */
+ /* always count this not worth bothering? */
+ task->ps_invocations += 1 * variant->ps_inv_multiplier;
+
+ /* Propagate non-interpolated raster state. */
+ task->thread_data.raster_state.viewport_index = inputs->viewport_index;
+
+ /* run shader on 4x4 block */
+ BEGIN_JIT_CALL(state, task);
+ variant->jit_function[RAST_EDGE_TEST](&state->jit_context,
+ x, y,
+ inputs->frontfacing,
+ GET_A0(inputs),
+ GET_DADX(inputs),
+ GET_DADY(inputs),
+ color,
+ depth,
+ mask,
+ &task->thread_data,
+ stride,
+ depth_stride);
+ END_JIT_CALL();
+ }
}
+
/**
- * Set top row and left column of the tile's pixels to white. For debugging.
+ * Begin a new occlusion query.
+ * This is a bin command put in all bins.
+ * Called per thread.
*/
static void
-outline_tile(uint8_t *tile)
+lp_rast_begin_query(struct lp_rasterizer_task *task,
+ const union lp_rast_cmd_arg arg)
{
- 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;
+ struct llvmpipe_query *pq = arg.query_obj;
+
+ switch (pq->type) {
+ case PIPE_QUERY_OCCLUSION_COUNTER:
+ case PIPE_QUERY_OCCLUSION_PREDICATE:
+ case PIPE_QUERY_OCCLUSION_PREDICATE_CONSERVATIVE:
+ pq->start[task->thread_index] = task->thread_data.vis_counter;
+ break;
+ case PIPE_QUERY_PIPELINE_STATISTICS:
+ pq->start[task->thread_index] = task->ps_invocations;
+ break;
+ default:
+ assert(0);
+ break;
}
}
/**
- * Draw grid of gray lines at 16-pixel intervals across the tile to
- * show the sub-tile boundaries. For debugging.
+ * End the current occlusion query.
+ * This is a bin command put in all bins.
+ * Called per thread.
*/
static void
-outline_subtiles(uint8_t *tile)
+lp_rast_end_query(struct lp_rasterizer_task *task,
+ const union lp_rast_cmd_arg arg)
{
- 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;
- }
+ struct llvmpipe_query *pq = arg.query_obj;
+
+ switch (pq->type) {
+ case PIPE_QUERY_OCCLUSION_COUNTER:
+ case PIPE_QUERY_OCCLUSION_PREDICATE:
+ case PIPE_QUERY_OCCLUSION_PREDICATE_CONSERVATIVE:
+ pq->end[task->thread_index] +=
+ task->thread_data.vis_counter - pq->start[task->thread_index];
+ pq->start[task->thread_index] = 0;
+ break;
+ case PIPE_QUERY_TIMESTAMP:
+ pq->end[task->thread_index] = os_time_get_nano();
+ break;
+ case PIPE_QUERY_PIPELINE_STATISTICS:
+ pq->end[task->thread_index] +=
+ task->ps_invocations - pq->start[task->thread_index];
+ pq->start[task->thread_index] = 0;
+ break;
+ default:
+ assert(0);
+ break;
}
-
- 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)
+void
+lp_rast_set_state(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;
- unsigned i;
-
- for (i = 0; i < rast->state.fb.nr_cbufs; i++) {
- struct pipe_transfer *transfer = rast->cbuf_transfer[i];
- int w = TILE_SIZE;
- int h = TILE_SIZE;
-
- if (x >= transfer->width)
- continue;
-
- if (y >= transfer->height)
- continue;
-
- LP_DBG(DEBUG_RAST, "%s [%u] %d,%d %dx%d\n", __FUNCTION__,
- 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,
- task->tile.color[i],
- rast->cbuf_map[i],
- transfer->stride,
- x, y,
- w, h);
- }
+ task->state = arg.state;
}
-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;
- }
-}
/**
- * Write the rasterizer's z/stencil tile to the framebuffer.
+ * Called when we're done writing to a color tile.
*/
-static void lp_rast_store_zstencil( struct lp_rasterizer *rast,
- unsigned thread_index )
+static void
+lp_rast_tile_end(struct lp_rasterizer_task *task)
{
- struct lp_rasterizer_task *task = &rast->tasks[thread_index];
- const unsigned x = task->x;
- const unsigned y = task->y;
- unsigned w = TILE_SIZE;
- unsigned h = TILE_SIZE;
-
- if (x + w > rast->state.fb.width)
- w -= x + w - rast->state.fb.width;
+ unsigned i;
- if (y + h > rast->state.fb.height)
- h -= y + h - rast->state.fb.height;
+ for (i = 0; i < task->scene->num_active_queries; ++i) {
+ lp_rast_end_query(task, lp_rast_arg_query(task->scene->active_queries[i]));
+ }
- LP_DBG(DEBUG_RAST, "%s %d,%d %dx%d\n", __FUNCTION__, x, y, w, h);
+ /* debug */
+ memset(task->color_tiles, 0, sizeof(task->color_tiles));
+ task->depth_tile = NULL;
- assert(rast->zsbuf_transfer->texture->format == PIPE_FORMAT_Z32_UNORM);
- lp_tile_write_z32(task->tile.depth,
- rast->zsbuf_map,
- rast->zsbuf_transfer->stride,
- x, y, w, h);
+ task->bin = NULL;
}
-
-/**
- * Write the rasterizer's tiles to the framebuffer.
- */
-static void
-lp_rast_end_tile( struct lp_rasterizer *rast,
- unsigned thread_index )
+static lp_rast_cmd_func dispatch[LP_RAST_OP_MAX] =
{
- 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_clear_color,
+ lp_rast_clear_zstencil,
+ lp_rast_triangle_1,
+ lp_rast_triangle_2,
+ lp_rast_triangle_3,
+ lp_rast_triangle_4,
+ lp_rast_triangle_5,
+ lp_rast_triangle_6,
+ lp_rast_triangle_7,
+ lp_rast_triangle_8,
+ lp_rast_triangle_3_4,
+ lp_rast_triangle_3_16,
+ lp_rast_triangle_4_16,
+ lp_rast_shade_tile,
+ lp_rast_shade_tile_opaque,
+ lp_rast_begin_query,
+ lp_rast_end_query,
+ lp_rast_set_state,
+ lp_rast_triangle_32_1,
+ lp_rast_triangle_32_2,
+ lp_rast_triangle_32_3,
+ lp_rast_triangle_32_4,
+ lp_rast_triangle_32_5,
+ lp_rast_triangle_32_6,
+ lp_rast_triangle_32_7,
+ lp_rast_triangle_32_8,
+ lp_rast_triangle_32_3_4,
+ lp_rast_triangle_32_3_16,
+ lp_rast_triangle_32_4_16
+};
-/**
- * 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 )
+static void
+do_rasterize_bin(struct lp_rasterizer_task *task,
+ const struct cmd_bin *bin,
+ int x, int y)
{
- struct lp_fence *fence = arg.fence;
-
- pipe_mutex_lock( fence->mutex );
-
- fence->count++;
- assert(fence->count <= fence->rank);
-
- LP_DBG(DEBUG_RAST, "%s count=%u rank=%u\n", __FUNCTION__,
- fence->count, fence->rank);
+ const struct cmd_block *block;
+ unsigned k;
- pipe_condvar_signal( fence->signalled );
+ if (0)
+ lp_debug_bin(bin, x, y);
- pipe_mutex_unlock( fence->mutex );
+ for (block = bin->head; block; block = block->next) {
+ for (k = 0; k < block->count; k++) {
+ dispatch[block->cmd[k]]( task, block->arg[k] );
+ }
+ }
}
-/**
- * When all the threads are done rasterizing a scene, one thread will
- * call this function to reset the scene and put it onto the empty queue.
- */
-static void
-release_scene( struct lp_rasterizer *rast,
- struct lp_scene *scene )
-{
- util_unreference_framebuffer_state( &scene->fb );
-
- lp_scene_reset( scene );
- lp_scene_enqueue( rast->empty_scenes, scene );
- rast->curr_scene = NULL;
-}
-
/**
* Rasterize commands for a single bin.
* 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 );
-
- /* 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] );
- }
- }
+ lp_rast_tile_begin( task, bin, x, y );
- lp_rast_end_tile( rast, thread_index );
-}
+ do_rasterize_bin(task, bin, x, y);
+ lp_rast_tile_end(task);
-#define RAST(x) { lp_rast_##x, #x }
-static struct {
- lp_rast_cmd cmd;
- const char *name;
-} cmd_names[] =
-{
- RAST(load_color),
- RAST(load_zstencil),
- RAST(clear_color),
- RAST(clear_zstencil),
- RAST(triangle),
- RAST(shade_tile),
- RAST(set_state),
- RAST(fence),
-};
-
-static void
-debug_bin( const struct cmd_bin *bin )
-{
- const struct cmd_block *head = bin->commands.head;
- int i, j;
-
- for (i = 0; i < head->count; i++) {
- debug_printf("%d: ", i);
- for (j = 0; j < Elements(cmd_names); j++) {
- if (head->cmd[i] == cmd_names[j].cmd) {
- debug_printf("%s\n", cmd_names[j].name);
- break;
- }
- }
- if (j == Elements(cmd_names))
- debug_printf("...other\n");
+ /* Debug/Perf flags:
+ */
+ if (bin->head->count == 1) {
+ if (bin->head->cmd[0] == LP_RAST_OP_SHADE_TILE_OPAQUE)
+ LP_COUNT(nr_pure_shade_opaque_64);
+ else if (bin->head->cmd[0] == LP_RAST_OP_SHADE_TILE)
+ LP_COUNT(nr_pure_shade_64);
}
-
}
+
/* An empty bin is one that just loads the contents of the tile and
* stores them again unchanged. This typically happens when bins have
* been flushed for some reason in the middle of a frame, or when
static boolean
is_empty_bin( const struct cmd_bin *bin )
{
- const struct cmd_block *head = bin->commands.head;
- int i;
-
- if (0)
- debug_bin(bin);
-
- /* We emit at most two load-tile commands at the start of the first
- * command block. In addition we seem to emit a couple of
- * set-state commands even in empty bins.
- *
- * As a heuristic, if a bin has more than 4 commands, consider it
- * non-empty.
- */
- if (head->next != NULL ||
- head->count > 4) {
- return FALSE;
- }
-
- 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;
- }
-
- return TRUE;
+ return bin->head == NULL;
}
-
/**
* Rasterize/execute all bins within a scene.
* Called per thread.
*/
static void
-rasterize_scene( struct lp_rasterizer *rast,
- unsigned thread_index,
- struct lp_scene *scene,
- bool write_depth )
+rasterize_scene(struct lp_rasterizer_task *task,
+ struct lp_scene *scene)
{
- /* 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 );
+ task->scene = scene;
+
+ /* Clear the cache tags. This should not always be necessary but
+ simpler for now. */
+#if LP_USE_TEXTURE_CACHE
+ memset(task->thread_data.cache->cache_tags, 0,
+ sizeof(task->thread_data.cache->cache_tags));
+#if LP_BUILD_FORMAT_CACHE_DEBUG
+ task->thread_data.cache->cache_access_total = 0;
+ task->thread_data.cache->cache_access_miss = 0;
+#endif
+#endif
+
+ if (!task->rast->no_rast && !scene->discard) {
+ /* loop over scene bins, rasterize each */
+ {
+ struct cmd_bin *bin;
+ int i, j;
+
+ assert(scene);
+ while ((bin = lp_scene_bin_iter_next(scene, &i, &j))) {
+ if (!is_empty_bin( bin ))
+ rasterize_bin(task, bin, i, j);
}
}
}
-#else
- {
- struct cmd_bin *bin;
- int x, y;
- 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);
+
+#if LP_BUILD_FORMAT_CACHE_DEBUG
+ {
+ uint64_t total, miss;
+ total = task->thread_data.cache->cache_access_total;
+ miss = task->thread_data.cache->cache_access_miss;
+ if (total) {
+ debug_printf("thread %d cache access %llu miss %llu hit rate %f\n",
+ task->thread_index, (long long unsigned)total,
+ (long long unsigned)miss,
+ (float)(total - miss)/(float)total);
}
}
#endif
+
+ if (scene->fence) {
+ lp_fence_signal(scene->fence);
+ }
+
+ task->scene = NULL;
}
* Called by setup module when it has something for us to render.
*/
void
-lp_rasterize_scene( struct lp_rasterizer *rast,
- struct lp_scene *scene,
- const struct pipe_framebuffer_state *fb,
- bool write_depth )
+lp_rast_queue_scene( struct lp_rasterizer *rast,
+ struct lp_scene *scene)
{
- boolean debug = false;
-
LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);
- if (debug) {
- unsigned x, y;
- 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++) {
- debug_printf(" bin %u, %u size: %u\n", x, y,
- lp_scene_bin_size(scene, x, y));
- }
- }
- }
-
- /* save framebuffer state in the bin */
- util_copy_framebuffer_state(&scene->fb, fb);
- scene->write_depth = write_depth;
-
if (rast->num_threads == 0) {
/* no threading */
+ unsigned fpstate = util_fpstate_get();
- lp_rast_begin( rast, fb,
- fb->nr_cbufs != 0, /* always write color if cbufs present */
- fb->zsbuf != NULL && write_depth );
+ /* Make sure that denorms are treated like zeros. This is
+ * the behavior required by D3D10. OpenGL doesn't care.
+ */
+ util_fpstate_set_denorms_to_zero(fpstate);
- lp_scene_bin_iter_begin( scene );
- rasterize_scene( rast, 0, scene, write_depth );
+ lp_rast_begin( rast, scene );
- release_scene( rast, scene );
+ rasterize_scene( &rast->tasks[0], scene );
lp_rast_end( rast );
+
+ util_fpstate_set(fpstate);
+
+ rast->curr_scene = NULL;
}
else {
/* threaded rendering! */
for (i = 0; i < rast->num_threads; i++) {
pipe_semaphore_signal(&rast->tasks[i].work_ready);
}
+ }
+
+ LP_DBG(DEBUG_SETUP, "%s done \n", __FUNCTION__);
+}
+
+
+void
+lp_rast_finish( struct lp_rasterizer *rast )
+{
+ if (rast->num_threads == 0) {
+ /* nothing to do */
+ }
+ else {
+ int i;
/* wait for work to complete */
for (i = 0; i < rast->num_threads; i++) {
pipe_semaphore_wait(&rast->tasks[i].work_done);
}
}
-
- LP_DBG(DEBUG_SETUP, "%s done \n", __FUNCTION__);
}
* 2. do work
* 3. signal that we're done
*/
-static PIPE_THREAD_ROUTINE( thread_func, init_data )
+static int
+thread_function(void *init_data)
{
struct lp_rasterizer_task *task = (struct lp_rasterizer_task *) init_data;
struct lp_rasterizer *rast = task->rast;
boolean debug = false;
+ char thread_name[16];
+ unsigned fpstate;
+
+ util_snprintf(thread_name, sizeof thread_name, "llvmpipe-%u", task->thread_index);
+ u_thread_setname(thread_name);
+
+ /* Make sure that denorms are treated like zeros. This is
+ * the behavior required by D3D10. OpenGL doesn't care.
+ */
+ fpstate = util_fpstate_get();
+ util_fpstate_set_denorms_to_zero(fpstate);
while (1) {
/* wait for work */
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
* - map the framebuffer surfaces
*/
- const struct pipe_framebuffer_state *fb;
- boolean write_depth;
-
- rast->curr_scene = lp_scene_dequeue( rast->full_scenes, TRUE );
-
- lp_scene_bin_iter_begin( rast->curr_scene );
-
- fb = &rast->curr_scene->fb;
- write_depth = rast->curr_scene->write_depth;
-
- lp_rast_begin( rast, fb,
- fb->nr_cbufs != 0,
- fb->zsbuf != NULL && write_depth );
+ lp_rast_begin( rast,
+ lp_scene_dequeue( rast->full_scenes, TRUE ) );
}
/* Wait for all threads to get here so that threads[1+] don't
* get a null rast->curr_scene pointer.
*/
- pipe_barrier_wait( &rast->barrier );
+ util_barrier_wait( &rast->barrier );
/* 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);
/* wait for all threads to finish with this scene */
- pipe_barrier_wait( &rast->barrier );
+ util_barrier_wait( &rast->barrier );
+ /* XXX: shouldn't be necessary:
+ */
if (task->thread_index == 0) {
- /* thread[0]:
- * - release the scene object
- * - unmap the framebuffer surfaces
- */
- release_scene( rast, rast->curr_scene );
lp_rast_end( rast );
}
/* signal done with work */
if (debug)
debug_printf("thread %d done working\n", task->thread_index);
+
pipe_semaphore_signal(&task->work_done);
}
- return NULL;
+#ifdef _WIN32
+ pipe_semaphore_signal(&task->work_done);
+#endif
+
+ return 0;
}
{
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++) {
pipe_semaphore_init(&rast->tasks[i].work_ready, 0);
pipe_semaphore_init(&rast->tasks[i].work_done, 0);
- rast->threads[i] = pipe_thread_create(thread_func,
+ rast->threads[i] = u_thread_create(thread_function,
(void *) &rast->tasks[i]);
}
}
/**
- * Create new lp_rasterizer.
- * \param empty the queue to put empty scenes on after we've finished
- * processing them.
+ * Create new lp_rasterizer. If num_threads is zero, don't create any
+ * new threads, do rendering synchronously.
+ * \param num_threads number of rasterizer threads to create
*/
struct lp_rasterizer *
-lp_rast_create( struct pipe_screen *screen, struct lp_scene_queue *empty )
+lp_rast_create( unsigned num_threads )
{
struct lp_rasterizer *rast;
- unsigned i, cbuf;
+ unsigned i;
rast = CALLOC_STRUCT(lp_rasterizer);
- if(!rast)
- return NULL;
-
- rast->screen = screen;
+ if (!rast) {
+ goto no_rast;
+ }
- rast->empty_scenes = empty;
rast->full_scenes = lp_scene_queue_create();
+ if (!rast->full_scenes) {
+ goto no_full_scenes;
+ }
- for (i = 0; i < Elements(rast->tasks); i++) {
+ for (i = 0; i < MAX2(1, num_threads); i++) {
struct lp_rasterizer_task *task = &rast->tasks[i];
-
- for (cbuf = 0; cbuf < PIPE_MAX_COLOR_BUFS; cbuf++ )
- task->tile.color[cbuf] = align_malloc(TILE_SIZE * TILE_SIZE * 4, 16);
-
- task->tile.depth = align_malloc(TILE_SIZE * TILE_SIZE * 4, 16);
task->rast = rast;
task->thread_index = i;
+ task->thread_data.cache = align_malloc(sizeof(struct lp_build_format_cache),
+ 16);
+ if (!task->thread_data.cache) {
+ goto no_thread_data_cache;
+ }
}
+ rast->num_threads = num_threads;
+
+ rast->no_rast = debug_get_bool_option("LP_NO_RAST", FALSE);
+
create_rast_threads(rast);
/* for synchronizing rasterization threads */
- pipe_barrier_init( &rast->barrier, rast->num_threads );
+ if (rast->num_threads > 0) {
+ util_barrier_init( &rast->barrier, rast->num_threads );
+ }
+
+ memset(lp_dummy_tile, 0, sizeof lp_dummy_tile);
return rast;
+
+no_thread_data_cache:
+ for (i = 0; i < MAX2(1, rast->num_threads); i++) {
+ if (rast->tasks[i].thread_data.cache) {
+ align_free(rast->tasks[i].thread_data.cache);
+ }
+ }
+
+ lp_scene_queue_destroy(rast->full_scenes);
+no_full_scenes:
+ FREE(rast);
+no_rast:
+ return NULL;
}
*/
void lp_rast_destroy( struct lp_rasterizer *rast )
{
- unsigned i, cbuf;
+ unsigned i;
- util_unreference_framebuffer_state(&rast->state.fb);
+ /* 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);
+ }
+
+ /* Wait for threads to terminate before cleaning up per-thread data.
+ * We don't actually call pipe_thread_wait to avoid dead lock on Windows
+ * per https://bugs.freedesktop.org/show_bug.cgi?id=76252 */
+ for (i = 0; i < rast->num_threads; i++) {
+#ifdef _WIN32
+ pipe_semaphore_wait(&rast->tasks[i].work_done);
+#else
+ thrd_join(rast->threads[i], NULL);
+#endif
+ }
- 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]);
+ /* Clean up per-thread data */
+ 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 (i = 0; i < MAX2(1, rast->num_threads); i++) {
+ align_free(rast->tasks[i].thread_data.cache);
}
/* for synchronizing rasterization threads */
- pipe_barrier_destroy( &rast->barrier );
+ if (rast->num_threads > 0) {
+ util_barrier_destroy( &rast->barrier );
+ }
+
+ lp_scene_queue_destroy(rast->full_scenes);
FREE(rast);
}
-/** Return number of rasterization threads */
-unsigned
-lp_rast_get_num_threads( struct lp_rasterizer *rast )
-{
- return rast->num_threads;
-}