#include "util/u_surface.h"
#include "util/u_pack_color.h"
+#include "os/os_time.h"
+
#include "lp_scene_queue.h"
#include "lp_debug.h"
#include "lp_fence.h"
#include "lp_query.h"
#include "lp_rast.h"
#include "lp_rast_priv.h"
-#include "lp_tile_soa.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
lp_scene_end_rasterization( rast->curr_scene );
rast->curr_scene = NULL;
-
-#ifdef DEBUG
- if (0)
- debug_printf("Post render scene: tile unswizzle: %u tile swizzle: %u\n",
- lp_tile_unswizzle_count, lp_tile_swizzle_count);
-#endif
}
* and update the tile's layout info.
*/
(void) llvmpipe_get_texture_tile(lpt,
- zsbuf->face + zsbuf->zslice,
- zsbuf->level,
+ zsbuf->u.tex.first_layer,
+ zsbuf->u.tex.level,
usage,
task->x,
task->y);
const union lp_rast_cmd_arg arg)
{
const struct lp_scene *scene = task->scene;
- const uint8_t *clear_color = arg.clear_color;
+ uint8_t clear_color[4];
unsigned i;
- LP_DBG(DEBUG_RAST, "%s 0x%x,0x%x,0x%x,0x%x\n", __FUNCTION__,
+ for (i = 0; i < 4; ++i) {
+ clear_color[i] = float_to_ubyte(arg.clear_color[i]);
+ }
+
+ 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 < scene->fb.nr_cbufs; i++) {
- uint8_t *ptr =
- lp_rast_get_color_tile_pointer(task, i, LP_TEX_USAGE_WRITE_ALL);
- memset(ptr, clear_color[0], TILE_SIZE * TILE_SIZE * 4);
- }
- }
- 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 < scene->fb.nr_cbufs; i++) {
- uint8_t *c =
- lp_rast_get_color_tile_pointer(task, i, LP_TEX_USAGE_WRITE_ALL);
- 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;
- }
- }
+ for (i = 0; i < scene->fb.nr_cbufs; i++) {
+ const struct lp_scene *scene = task->scene;
+ union util_color uc;
+
+ util_pack_color(arg.clear_color,
+ scene->fb.cbufs[i]->format, &uc);
+
+ util_fill_rect(scene->cbufs[i].map,
+ scene->fb.cbufs[i]->format,
+ scene->cbufs[i].stride,
+ task->x,
+ task->y,
+ TILE_SIZE,
+ TILE_SIZE,
+ &uc);
}
LP_COUNT(nr_color_tile_clear);
const union lp_rast_cmd_arg arg)
{
const struct lp_scene *scene = task->scene;
- unsigned clear_value = arg.clear_zstencil.value;
- unsigned clear_mask = arg.clear_zstencil.mask;
+ uint32_t clear_value = arg.clear_zstencil.value;
+ uint32_t clear_mask = arg.clear_zstencil.mask;
const unsigned height = TILE_SIZE / TILE_VECTOR_HEIGHT;
const unsigned width = TILE_SIZE * TILE_VECTOR_HEIGHT;
const unsigned block_size = scene->zsbuf.blocksize;
uint8_t *dst;
unsigned i, j;
- LP_DBG(DEBUG_RAST, "%s 0x%x%x\n", __FUNCTION__, clear_value, clear_mask);
+ LP_DBG(DEBUG_RAST, "%s: value=0x%08x, mask=0x%08x\n",
+ __FUNCTION__, clear_value, clear_mask);
/*
- * Clear the aera of the swizzled depth/depth buffer matching this tile, in
+ * Clear the area 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
dst = task->depth_tile;
+ clear_value &= clear_mask;
+
switch (block_size) {
case 1:
+ assert(clear_mask == 0xff);
memset(dst, (uint8_t) clear_value, 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) clear_value;
- dst += dst_stride;
+ 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:
uint32_t *row = (uint32_t *)dst;
for (j = 0; j < width; j++) {
uint32_t tmp = ~clear_mask & *row;
- *row++ = (clear_value & clear_mask) | tmp;
+ *row++ = clear_value | tmp;
}
dst += dst_stride;
}
-
-/**
- * Convert the color tile from tiled to linear layout.
- * This is generally only done when we're flushing the scene just prior to
- * SwapBuffers. If we didn't do this here, we'd have to convert the entire
- * tiled color buffer to linear layout in the llvmpipe_texture_unmap()
- * function. It's better to do it here to take advantage of
- * threading/parallelism.
- * This is a bin command which is stored in all bins.
- */
-static void
-lp_rast_store_linear_color( struct lp_rasterizer_task *task )
-{
- const struct lp_scene *scene = task->scene;
- unsigned buf;
-
- for (buf = 0; buf < scene->fb.nr_cbufs; buf++) {
- struct pipe_surface *cbuf = scene->fb.cbufs[buf];
- const unsigned face_slice = cbuf->face + cbuf->zslice;
- const unsigned level = cbuf->level;
- struct llvmpipe_resource *lpt = llvmpipe_resource(cbuf->texture);
-
- if (!task->color_tiles[buf])
- continue;
-
- llvmpipe_unswizzle_cbuf_tile(lpt,
- face_slice,
- level,
- task->x, task->y,
- task->color_tiles[buf]);
- }
-}
-
-
-
/**
* Run the shader on all blocks in a tile. This is used when a tile is
* completely contained inside a triangle.
{
const struct lp_scene *scene = task->scene;
const struct lp_rast_shader_inputs *inputs = arg.shade_tile;
- const struct lp_rast_state *state = inputs->state;
- struct lp_fragment_shader_variant *variant = state->variant;
+ 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;
LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__);
+ state = task->state;
+ assert(state);
+ if (!state) {
+ return;
+ }
+ variant = state->variant;
+
/* 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];
+ unsigned stride[PIPE_MAX_COLOR_BUFS];
uint32_t *depth;
unsigned i;
/* color buffer */
- for (i = 0; i < scene->fb.nr_cbufs; i++)
- color[i] = lp_rast_get_color_block_pointer(task, i,
- tile_x + x, tile_y + y);
+ 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);
+ }
/* depth buffer */
depth = lp_rast_get_depth_block_pointer(task, tile_x + x, tile_y + y);
/* run shader on 4x4 block */
- BEGIN_JIT_CALL(state);
+ BEGIN_JIT_CALL(state, task);
variant->jit_function[RAST_WHOLE]( &state->jit_context,
tile_x + x, tile_y + y,
- inputs->facing,
- inputs->a0,
- inputs->dadx,
- inputs->dady,
+ inputs->frontfacing,
+ GET_A0(inputs),
+ GET_DADX(inputs),
+ GET_DADY(inputs),
color,
depth,
0xffff,
- &task->vis_counter);
+ &task->vis_counter,
+ stride);
END_JIT_CALL();
}
}
lp_rast_shade_tile_opaque(struct lp_rasterizer_task *task,
const union lp_rast_cmd_arg arg)
{
- const struct lp_scene *scene = task->scene;
- unsigned i;
-
LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__);
- /* this will prevent converting the layout from tiled to linear */
- for (i = 0; i < scene->fb.nr_cbufs; i++) {
- (void)lp_rast_get_color_tile_pointer(task, i, LP_TEX_USAGE_WRITE_ALL);
+ assert(task->state);
+ if (!task->state) {
+ return;
}
lp_rast_shade_tile(task, arg);
unsigned x, unsigned y,
unsigned mask)
{
- const struct lp_rast_state *state = inputs->state;
+ 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];
void *depth;
unsigned i;
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);
/* color buffer */
for (i = 0; i < scene->fb.nr_cbufs; i++) {
- color[i] = lp_rast_get_color_block_pointer(task, i, x, y);
- assert(lp_check_alignment(color[i], 16));
+ stride[i] = scene->cbufs[i].stride;
+
+ color[i] = lp_rast_get_unswizzled_color_block_pointer(task, i, x, y);
}
/* depth buffer */
depth = lp_rast_get_depth_block_pointer(task, x, y);
- assert(lp_check_alignment(state->jit_context.blend_color, 16));
+ assert(lp_check_alignment(state->jit_context.u8_blend_color, 16));
/* run shader on 4x4 block */
- BEGIN_JIT_CALL(state);
+ BEGIN_JIT_CALL(state, task);
variant->jit_function[RAST_EDGE_TEST](&state->jit_context,
x, y,
- inputs->facing,
- inputs->a0,
- inputs->dadx,
- inputs->dady,
+ inputs->frontfacing,
+ GET_A0(inputs),
+ GET_DADX(inputs),
+ GET_DADY(inputs),
color,
depth,
mask,
- &task->vis_counter);
+ &task->vis_counter,
+ stride);
END_JIT_CALL();
}
{
struct llvmpipe_query *pq = arg.query_obj;
- assert(task->query == NULL);
- task->vis_counter = 0;
- task->query = pq;
+ assert(task->query[pq->type] == NULL);
+
+ switch (pq->type) {
+ case PIPE_QUERY_OCCLUSION_COUNTER:
+ task->vis_counter = 0;
+ break;
+ case PIPE_QUERY_TIME_ELAPSED:
+ task->query_start = os_time_get_nano();
+ break;
+ default:
+ assert(0);
+ break;
+ }
+
+ task->query[pq->type] = pq;
}
lp_rast_end_query(struct lp_rasterizer_task *task,
const union lp_rast_cmd_arg arg)
{
- task->query->count[task->thread_index] += task->vis_counter;
- task->query = NULL;
-}
-
-
-
-/**
- * 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;
+ struct llvmpipe_query *pq = arg.query_obj;
+ assert(task->query[pq->type] == pq || pq->type == PIPE_QUERY_TIMESTAMP);
- 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;
+ switch (pq->type) {
+ case PIPE_QUERY_OCCLUSION_COUNTER:
+ pq->count[task->thread_index] += task->vis_counter;
+ break;
+ case PIPE_QUERY_TIME_ELAPSED:
+ pq->count[task->thread_index] = os_time_get_nano() - task->query_start;
+ break;
+ case PIPE_QUERY_TIMESTAMP:
+ pq->count[task->thread_index] = os_time_get_nano();
+ break;
+ default:
+ assert(0);
+ break;
+ }
- 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;
+ if (task->query[pq->type] == pq) {
+ task->query[pq->type] = NULL;
}
}
-/**
- * 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)
+void
+lp_rast_set_state(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;
- }
- }
-
- outline_tile(tile);
+ task->state = arg.state;
}
static void
lp_rast_tile_end(struct lp_rasterizer_task *task)
{
-#ifdef DEBUG
- if (LP_DEBUG & (DEBUG_SHOW_SUBTILES | DEBUG_SHOW_TILES)) {
- const struct lp_scene *scene = task->scene;
- unsigned buf;
-
- for (buf = 0; buf < scene->fb.nr_cbufs; buf++) {
- uint8_t *color = lp_rast_get_color_block_pointer(task, buf,
- task->x, task->y);
+ unsigned i;
- if (LP_DEBUG & DEBUG_SHOW_SUBTILES)
- outline_subtiles(color);
- else if (LP_DEBUG & DEBUG_SHOW_TILES)
- outline_tile(color);
+ for (i = 0; i < PIPE_QUERY_TYPES; ++i) {
+ if (task->query[i]) {
+ lp_rast_end_query(task, lp_rast_arg_query(task->query[i]));
}
}
-#else
- (void) outline_subtiles;
-#endif
-
- lp_rast_store_linear_color(task);
-
- if (task->query) {
- union lp_rast_cmd_arg dummy = {0};
- lp_rast_end_query(task, dummy);
- }
/* debug */
memset(task->color_tiles, 0, sizeof(task->color_tiles));
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,
};
struct lp_scene *scene)
{
task->scene = scene;
- /* loop over scene bins, rasterize each */
+
+ if (!task->rast->no_rast) {
+ /* 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_scene_get_bin(scene, i, j);
- rasterize_bin(task, bin, i, j);
+ {
+ unsigned i, j;
+ for (i = 0; i < scene->tiles_x; i++) {
+ for (j = 0; j < scene->tiles_y; j++) {
+ struct cmd_bin *bin = lp_scene_get_bin(scene, i, j);
+ rasterize_bin(task, bin, i, j);
+ }
}
}
- }
#else
- {
- struct cmd_bin *bin;
+ {
+ struct cmd_bin *bin;
- assert(scene);
- while ((bin = lp_scene_bin_iter_next(scene))) {
- if (!is_empty_bin( bin ))
- rasterize_bin(task, bin);
+ assert(scene);
+ while ((bin = lp_scene_bin_iter_next(scene))) {
+ if (!is_empty_bin( bin ))
+ rasterize_bin(task, bin);
+ }
}
- }
#endif
+ }
+
if (scene->fence) {
lp_fence_signal(scene->fence);
* 2. do work
* 3. signal that we're done
*/
-static PIPE_THREAD_ROUTINE( thread_func, init_data )
+static PIPE_THREAD_ROUTINE( thread_function, init_data )
{
struct lp_rasterizer_task *task = (struct lp_rasterizer_task *) init_data;
struct lp_rasterizer *rast = task->rast;
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] = pipe_thread_create(thread_function,
(void *) &rast->tasks[i]);
}
}
unsigned i;
rast = CALLOC_STRUCT(lp_rasterizer);
- if(!rast)
- return NULL;
+ if (!rast) {
+ goto no_rast;
+ }
rast->full_scenes = lp_scene_queue_create();
+ if (!rast->full_scenes) {
+ goto no_full_scenes;
+ }
for (i = 0; i < Elements(rast->tasks); i++) {
struct lp_rasterizer_task *task = &rast->tasks[i];
rast->num_threads = num_threads;
+ rast->no_rast = debug_get_bool_option("LP_NO_RAST", FALSE);
+
create_rast_threads(rast);
/* for synchronizing rasterization threads */
memset(lp_dummy_tile, 0, sizeof lp_dummy_tile);
return rast;
+
+no_full_scenes:
+ FREE(rast);
+no_rast:
+ return NULL;
}