#include "util/u_rect.h"
#include "util/u_surface.h"
#include "util/u_pack_color.h"
-
-#include "os/os_time.h"
+#include "util/u_string.h"
+#include "util/u_thread.h"
+#include "util/u_memset.h"
+#include "util/os_time.h"
#include "lp_scene_queue.h"
#include "lp_context.h"
#include "lp_query.h"
#include "lp_rast.h"
#include "lp_rast_priv.h"
+#include "gallivm/lp_bld_format.h"
#include "gallivm/lp_bld_debug.h"
#include "lp_scene.h"
#include "lp_tex_sample.h"
const struct lp_rasterizer_task *jit_task = NULL;
#endif
+const float lp_sample_pos_4x[4][2] = { { 0.375, 0.125 },
+ { 0.875, 0.375 },
+ { 0.125, 0.625 },
+ { 0.625, 0.875 } };
/**
* Begin rasterizing a scene.
/**
- * 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
*/
const struct cmd_bin *bin,
int x, int y)
{
+ unsigned i;
+ struct lp_scene *scene = task->scene;
+
LP_DBG(DEBUG_RAST, "%s %d,%d\n", __FUNCTION__, x, y);
task->bin = bin;
task->scene->fb.height - y * TILE_SIZE : TILE_SIZE;
task->thread_data.vis_counter = 0;
- task->ps_invocations = 0;
-
- /* reset pointers to color and depth tile(s) */
- memset(task->color_tiles, 0, sizeof(task->color_tiles));
- task->depth_tile = NULL;
-}
+ task->thread_data.ps_invocations = 0;
-
-/**
- * Examine a framebuffer object to determine if any of the colorbuffers
- * use a pure integer format.
- * XXX this could be a gallium utility function if useful elsewhere.
- */
-static boolean
-is_fb_pure_integer(const struct pipe_framebuffer_state *fb)
-{
- unsigned i;
- for (i = 0; i < fb->nr_cbufs; i++) {
- if (fb->cbufs[i] &&
- util_format_is_pure_integer(fb->cbufs[i]->format)) {
- return TRUE;
+ 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;
}
}
- return FALSE;
+ if (task->scene->fb.zsbuf) {
+ task->depth_tile = scene->zsbuf.map +
+ scene->zsbuf.stride * task->y +
+ scene->zsbuf.format_bytes * task->x;
+ }
}
const union lp_rast_cmd_arg arg)
{
const struct lp_scene *scene = task->scene;
+ unsigned cbuf = arg.clear_rb->cbuf;
+ union util_color uc;
+ enum pipe_format format;
- if (scene->fb.nr_cbufs) {
- unsigned i;
- union util_color uc;
-
- if (is_fb_pure_integer(&scene->fb)) {
- /*
- * We expect int/uint clear values here, though some APIs
- * might disagree (but in any case util_pack_color()
- * couldn't handle it)...
- */
- LP_DBG(DEBUG_RAST, "%s pure int 0x%x,0x%x,0x%x,0x%x\n", __FUNCTION__,
- arg.clear_color.ui[0],
- arg.clear_color.ui[1],
- arg.clear_color.ui[2],
- arg.clear_color.ui[3]);
-
- for (i = 0; i < scene->fb.nr_cbufs; i++) {
- enum pipe_format format = scene->fb.cbufs[i]->format;
+ /* we never bin clear commands for non-existing buffers */
+ assert(cbuf < scene->fb.nr_cbufs);
+ assert(scene->fb.cbufs[cbuf]);
- if (util_format_is_pure_sint(format)) {
- util_format_write_4i(format, arg.clear_color.i, 0, &uc, 0, 0, 0, 1, 1);
- }
- else {
- assert(util_format_is_pure_uint(format));
- util_format_write_4ui(format, arg.clear_color.ui, 0, &uc, 0, 0, 0, 1, 1);
- }
-
- util_fill_box(scene->cbufs[i].map,
- format,
- scene->cbufs[i].stride,
- scene->cbufs[i].layer_stride,
- task->x,
- task->y,
- 0,
- task->width,
- task->height,
- scene->fb_max_layer + 1,
- &uc);
- }
- }
- else {
- uint8_t clear_color[4];
-
- for (i = 0; i < 4; ++i) {
- clear_color[i] = float_to_ubyte(arg.clear_color.f[i]);
- }
+ format = scene->fb.cbufs[cbuf]->format;
+ uc = arg.clear_rb->color_val;
- 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]);
-
- for (i = 0; i < scene->fb.nr_cbufs; i++) {
- if (scene->fb.cbufs[i]) {
- util_pack_color(arg.clear_color.f,
- scene->fb.cbufs[i]->format, &uc);
-
- util_fill_box(scene->cbufs[i].map,
- scene->fb.cbufs[i]->format,
- scene->cbufs[i].stride,
- scene->cbufs[i].layer_stride,
- task->x,
- task->y,
- 0,
- task->width,
- task->height,
- scene->fb_max_layer + 1,
- &uc);
- }
- }
- }
+ /*
+ * 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]);
+
+ for (unsigned s = 0; s < scene->cbufs[cbuf].nr_samples; s++) {
+ void *map = (char *)scene->cbufs[cbuf].map + scene->cbufs[cbuf].sample_stride * s;
+ util_fill_box(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);
}
-
-
/**
* Clear the rasterizer's current z/stencil tile.
* This is a bin command called during bin processing.
if (scene->fb.zsbuf) {
unsigned layer;
- uint8_t *dst_layer = lp_rast_get_unswizzled_depth_tile_pointer(task, LP_TEX_USAGE_READ_WRITE);
- block_size = util_format_get_blocksize(scene->fb.zsbuf->format);
- clear_value &= clear_mask;
+ for (unsigned s = 0; s < scene->zsbuf.nr_samples; s++) {
+ uint8_t *dst_layer = task->depth_tile + (s * scene->zsbuf.sample_stride);
+ block_size = util_format_get_blocksize(scene->fb.zsbuf->format);
+
+ clear_value &= clear_mask;
- for (layer = 0; layer <= scene->fb_max_layer; layer++) {
- dst = dst_layer;
+ 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) {
+ switch (block_size) {
+ case 1:
+ assert(clear_mask == 0xff);
for (i = 0; i < height; i++) {
- uint16_t *row = (uint16_t *)dst;
- for (j = 0; j < width; j++)
- *row++ = (uint16_t) clear_value;
+ uint8_t *row = (uint8_t *)dst;
+ memset(row, (uint8_t) clear_value, width);
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;
+ 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;
}
- 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++) {
+ 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;
+ }
}
- }
- 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;
+ break;
+ case 4:
+ if (clear_mask == 0xffffffff) {
+ for (i = 0; i < height; i++) {
+ util_memset32(dst, clear_value, width);
+ dst += dst_stride;
}
- 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++) {
+ 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;
+ }
}
- }
- 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;
+ 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;
}
- dst += dst_stride;
}
- }
- break;
+ 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;
- default:
- assert(0);
- break;
+ default:
+ assert(0);
+ break;
+ }
+ dst_layer += scene->zsbuf.layer_stride;
}
- dst_layer += scene->zsbuf.layer_stride;
}
}
}
for (x = 0; x < task->width; x += 4) {
uint8_t *color[PIPE_MAX_COLOR_BUFS];
unsigned stride[PIPE_MAX_COLOR_BUFS];
+ unsigned sample_stride[PIPE_MAX_COLOR_BUFS];
uint8_t *depth = NULL;
unsigned depth_stride = 0;
+ unsigned depth_sample_stride = 0;
unsigned i;
/* color buffer */
for (i = 0; i < scene->fb.nr_cbufs; i++){
- stride[i] = scene->cbufs[i].stride;
- color[i] = lp_rast_get_unswizzled_color_block_pointer(task, i, tile_x + x,
- tile_y + y, inputs->layer);
+ if (scene->fb.cbufs[i]) {
+ stride[i] = scene->cbufs[i].stride;
+ sample_stride[i] = scene->cbufs[i].sample_stride;
+ color[i] = lp_rast_get_color_block_pointer(task, i, tile_x + x,
+ tile_y + y, inputs->layer);
+ }
+ else {
+ stride[i] = 0;
+ sample_stride[i] = 0;
+ color[i] = NULL;
+ }
}
/* depth buffer */
if (scene->zsbuf.map) {
- depth = lp_rast_get_unswizzled_depth_block_pointer(task, tile_x + x,
- tile_y + y, inputs->layer);
+ depth = lp_rast_get_depth_block_pointer(task, tile_x + x,
+ tile_y + y, inputs->layer);
depth_stride = scene->zsbuf.stride;
+ depth_sample_stride = scene->zsbuf.sample_stride;
}
+ uint64_t mask = 0;
+ for (unsigned i = 0; i < scene->fb_max_samples; i++)
+ mask |= (uint64_t)(0xffff) << (16 * i);
+
/* Propagate non-interpolated raster state. */
task->thread_data.raster_state.viewport_index = inputs->viewport_index;
GET_DADY(inputs),
color,
depth,
- 0xffff,
+ mask,
&task->thread_data,
stride,
- depth_stride);
+ depth_stride,
+ sample_stride,
+ depth_sample_stride);
END_JIT_CALL();
}
}
* \param y Y position of quad in window coords
*/
void
-lp_rast_shade_quads_mask(struct lp_rasterizer_task *task,
- const struct lp_rast_shader_inputs *inputs,
- unsigned x, unsigned y,
- unsigned mask)
+lp_rast_shade_quads_mask_sample(struct lp_rasterizer_task *task,
+ const struct lp_rast_shader_inputs *inputs,
+ unsigned x, unsigned y,
+ uint64_t mask)
{
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];
unsigned stride[PIPE_MAX_COLOR_BUFS];
+ unsigned sample_stride[PIPE_MAX_COLOR_BUFS];
uint8_t *depth = NULL;
unsigned depth_stride = 0;
+ unsigned depth_sample_stride = 0;
unsigned i;
assert(state);
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_unswizzled_color_block_pointer(task, i, x, y,
- inputs->layer);
+ sample_stride[i] = scene->cbufs[i].sample_stride;
+ color[i] = lp_rast_get_color_block_pointer(task, i, x, y,
+ inputs->layer);
}
else {
stride[i] = 0;
+ sample_stride[i] = 0;
color[i] = NULL;
}
}
/* depth buffer */
if (scene->zsbuf.map) {
depth_stride = scene->zsbuf.stride;
- depth = lp_rast_get_unswizzled_depth_block_pointer(task, x, y, inputs->layer);
+ depth_sample_stride = scene->zsbuf.sample_stride;
+ depth = lp_rast_get_depth_block_pointer(task, x, y, inputs->layer);
}
assert(lp_check_alignment(state->jit_context.u8_blend_color, 16));
* 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;
mask,
&task->thread_data,
stride,
- depth_stride);
+ depth_stride,
+ sample_stride,
+ depth_sample_stride);
END_JIT_CALL();
}
}
-
+void
+lp_rast_shade_quads_mask(struct lp_rasterizer_task *task,
+ const struct lp_rast_shader_inputs *inputs,
+ unsigned x, unsigned y,
+ unsigned mask)
+{
+ uint64_t new_mask = 0;
+ for (unsigned i = 0; i < task->scene->fb_max_samples; i++)
+ new_mask |= ((uint64_t)mask) << (16 * i);
+ lp_rast_shade_quads_mask_sample(task, inputs, x, y, new_mask);
+}
/**
* Begin a new occlusion query.
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;
+ pq->start[task->thread_index] = task->thread_data.ps_invocations;
break;
default:
assert(0);
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_PIPELINE_STATISTICS:
pq->end[task->thread_index] +=
- task->ps_invocations - pq->start[task->thread_index];
+ task->thread_data.ps_invocations - pq->start[task->thread_index];
pq->start[task->thread_index] = 0;
break;
default:
lp_rast_triangle_32_8,
lp_rast_triangle_32_3_4,
lp_rast_triangle_32_3_16,
- lp_rast_triangle_32_4_16
+ lp_rast_triangle_32_4_16,
+ lp_rast_triangle_ms_1,
+ lp_rast_triangle_ms_2,
+ lp_rast_triangle_ms_3,
+ lp_rast_triangle_ms_4,
+ lp_rast_triangle_ms_5,
+ lp_rast_triangle_ms_6,
+ lp_rast_triangle_ms_7,
+ lp_rast_triangle_ms_8,
+ lp_rast_triangle_ms_3_4,
+ lp_rast_triangle_ms_3_16,
+ lp_rast_triangle_ms_4_16,
};
lp_rast_tile_end(task);
-
+#ifdef DEBUG
/* Debug/Perf flags:
*/
if (bin->head->count == 1) {
else if (bin->head->cmd[0] == LP_RAST_OP_SHADE_TILE)
LP_COUNT(nr_pure_shade_64);
}
+#endif
}
{
task->scene = scene;
- if (!task->rast->no_rast && !scene->discard) {
+ /* 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) {
/* loop over scene bins, rasterize each */
{
struct cmd_bin *bin;
}
+#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);
}
* 2. do work
* 3. signal that we're done
*/
-static PIPE_THREAD_ROUTINE( thread_function, 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;
- unsigned fpstate = util_fpstate_get();
+ char thread_name[16];
+ unsigned fpstate;
+
+ 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 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)
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:
*/
pipe_semaphore_signal(&task->work_done);
}
- return NULL;
+#ifdef _WIN32
+ pipe_semaphore_signal(&task->work_done);
+#endif
+
+ return 0;
}
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_function,
+ rast->threads[i] = u_thread_create(thread_function,
(void *) &rast->tasks[i]);
+ if (!rast->threads[i]) {
+ rast->num_threads = i; /* previous thread is max */
+ break;
+ }
}
}
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];
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;
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:
pipe_semaphore_signal(&rast->tasks[i].work_ready);
}
- /* Wait for threads to terminate before cleaning up per-thread data */
+ /* 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++) {
- pipe_thread_wait(rast->threads[i]);
+#ifdef _WIN32
+ pipe_semaphore_wait(&rast->tasks[i].work_done);
+#else
+ thrd_join(rast->threads[i], NULL);
+#endif
}
/* Clean up per-thread data */
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);
projects / mesa.git / blobdiff