AlexIndustrial/mesa
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

llvmpipe: asst. clean-ups in lp_rast.c

Browse Source 
Signed-off-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16937>
This commit is contained in:
Brian Paul
2022-05-20 13:08:48 -06:00
committed by Marge Bot
parent e060d7747d
commit b3dc37d22d
1 changed files with 143 additions and 176 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/gallium/drivers/llvmpipe/lp_rast.c
+143 -176
View File
@@ -66,20 +66,20 @@ const float lp_sample_pos_4x[4][2] = { { 0.375, 0.125 },
* Called once per scene by one thread.
*/
static void
lp_rast_begin( struct lp_rasterizer *rast,
struct lp_scene *scene )
lp_rast_begin(struct lp_rasterizer *rast,
struct lp_scene *scene)
{
rast->curr_scene = scene;
LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__);
lp_scene_begin_rasterization( scene );
lp_scene_bin_iter_begin( scene );
lp_scene_begin_rasterization(scene);
lp_scene_bin_iter_begin(scene);
}
static void
lp_rast_end( struct lp_rasterizer *rast )
lp_rast_end(struct lp_rasterizer *rast)
{
rast->curr_scene = NULL;
}
@@ -95,7 +95,6 @@ lp_rast_tile_begin(struct lp_rasterizer_task *task,
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);
@@ -103,22 +102,22 @@ lp_rast_tile_begin(struct lp_rasterizer_task *task,
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->width = TILE_SIZE + x * TILE_SIZE > scene->fb.width ?
scene->fb.width - x * TILE_SIZE : TILE_SIZE;
task->height = TILE_SIZE + y * TILE_SIZE > scene->fb.height ?
scene->fb.height - y * TILE_SIZE : TILE_SIZE;
task->thread_data.vis_counter = 0;
task->thread_data.ps_invocations = 0;
for (i = 0; i < task->scene->fb.nr_cbufs; i++) {
if (task->scene->fb.cbufs[i]) {
for (unsigned i = 0; i < scene->fb.nr_cbufs; i++) {
if (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;
}
}
if (task->scene->fb.zsbuf) {
if (scene->fb.zsbuf) {
task->depth_tile = scene->zsbuf.map +
scene->zsbuf.stride * task->y +
scene->zsbuf.format_bytes * task->x;
@@ -136,26 +135,26 @@ lp_rast_clear_color(struct lp_rasterizer_task *task,
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;
const unsigned cbuf = arg.clear_rb->cbuf;
/* 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;
const enum pipe_format format = scene->fb.cbufs[cbuf]->format;
union util_color uc = arg.clear_rb->color_val;
/*
* this is pretty rough since we have target format (bunch of bytes...) here.
* dump it as raw 4 dwords.
* 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",
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;
void *map = (char *) scene->cbufs[cbuf].map
+ scene->cbufs[cbuf].sample_stride * s;
util_fill_box(map,
format,
scene->cbufs[cbuf].stride,
@@ -191,9 +190,6 @@ lp_rast_clear_zstencil(struct lp_rasterizer_task *task,
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;
LP_DBG(DEBUG_RAST, "%s: value=0x%08x, mask=0x%08x\n",
__FUNCTION__, clear_value, clear_mask);
@@ -203,21 +199,21 @@ lp_rast_clear_zstencil(struct lp_rasterizer_task *task,
*/
if (scene->fb.zsbuf) {
unsigned layer;
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);
uint8_t *dst_layer =
task->depth_tile + (s * scene->zsbuf.sample_stride);
const unsigned 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 (unsigned layer = 0; layer <= scene->fb_max_layer; layer++) {
uint8_t *dst = dst_layer;
switch (block_size) {
case 1:
assert(clear_mask == 0xff);
for (i = 0; i < height; i++) {
for (unsigned i = 0; i < height; i++) {
uint8_t *row = (uint8_t *)dst;
memset(row, (uint8_t) clear_value, width);
dst += dst_stride;
@@ -225,17 +221,17 @@ lp_rast_clear_zstencil(struct lp_rasterizer_task *task,
break;
case 2:
if (clear_mask == 0xffff) {
for (i = 0; i < height; i++) {
for (unsigned i = 0; i < height; i++) {
uint16_t *row = (uint16_t *)dst;
for (j = 0; j < width; j++)
for (unsigned j = 0; j < width; j++)
*row++ = (uint16_t) clear_value;
dst += dst_stride;
}
}
else {
for (i = 0; i < height; i++) {
for (unsigned i = 0; i < height; i++) {
uint16_t *row = (uint16_t *)dst;
for (j = 0; j < width; j++) {
for (unsigned j = 0; j < width; j++) {
uint16_t tmp = ~clear_mask & *row;
*row++ = clear_value | tmp;
}
@@ -245,15 +241,15 @@ lp_rast_clear_zstencil(struct lp_rasterizer_task *task,
break;
case 4:
if (clear_mask == 0xffffffff) {
for (i = 0; i < height; i++) {
for (unsigned i = 0; i < height; i++) {
util_memset32(dst, clear_value, width);
dst += dst_stride;
}
}
else {
for (i = 0; i < height; i++) {
for (unsigned i = 0; i < height; i++) {
uint32_t *row = (uint32_t *)dst;
for (j = 0; j < width; j++) {
for (unsigned j = 0; j < width; j++) {
uint32_t tmp = ~clear_mask & *row;
*row++ = clear_value | tmp;
}
@@ -264,15 +260,15 @@ lp_rast_clear_zstencil(struct lp_rasterizer_task *task,
case 8:
clear_value64 &= clear_mask64;
if (clear_mask64 == 0xffffffffffULL) {
for (i = 0; i < height; i++) {
for (unsigned i = 0; i < height; i++) {
util_memset64(dst, clear_value64, width);
dst += dst_stride;
}
}
else {
for (i = 0; i < height; i++) {
for (unsigned i = 0; i < height; i++) {
uint64_t *row = (uint64_t *)dst;
for (j = 0; j < width; j++) {
for (unsigned j = 0; j < width; j++) {
uint64_t tmp = ~clear_mask64 & *row;
*row++ = clear_value64 | tmp;
}
@@ -292,7 +288,6 @@ lp_rast_clear_zstencil(struct lp_rasterizer_task *task,
}
/**
* Run the shader on all blocks in a tile. This is used when a tile is
* completely contained inside a triangle.
@@ -304,10 +299,7 @@ lp_rast_shade_tile(struct lp_rasterizer_task *task,
{
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 (inputs->disable) {
/* This command was partially binned and has been disabled */
@@ -316,31 +308,28 @@ lp_rast_shade_tile(struct lp_rasterizer_task *task,
LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__);
state = task->state;
const struct lp_rast_state *state = task->state;
assert(state);
if (!state) {
return;
}
variant = state->variant;
const struct lp_fragment_shader_variant *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) {
for (unsigned y = 0; y < task->height; y += 4){
for (unsigned x = 0; x < task->width; x += 4) {
/* color buffer */
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++){
for (unsigned i = 0; i < scene->fb.nr_cbufs; i++){
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 + inputs->view_index);
tile_y + y,
inputs->layer + inputs->view_index);
}
else {
stride[i] = 0;
@@ -350,9 +339,13 @@ lp_rast_shade_tile(struct lp_rasterizer_task *task,
}
/* depth buffer */
uint8_t *depth = NULL;
unsigned depth_stride = 0;
unsigned depth_sample_stride = 0;
if (scene->zsbuf.map) {
depth = lp_rast_get_depth_block_pointer(task, tile_x + x,
tile_y + y, inputs->layer + inputs->view_index);
tile_y + y,
inputs->layer + inputs->view_index);
depth_stride = scene->zsbuf.stride;
depth_sample_stride = scene->zsbuf.sample_stride;
}
@@ -367,7 +360,7 @@ lp_rast_shade_tile(struct lp_rasterizer_task *task,
/* run shader on 4x4 block */
BEGIN_JIT_CALL(state, task);
variant->jit_function[RAST_WHOLE]( &state->jit_context,
variant->jit_function[RAST_WHOLE](&state->jit_context,
tile_x + x, tile_y + y,
inputs->frontfacing,
GET_A0(inputs),
@@ -420,15 +413,8 @@ lp_rast_shade_quads_mask_sample(struct lp_rasterizer_task *task,
uint64_t mask)
{
const struct lp_rast_state *state = task->state;
struct lp_fragment_shader_variant *variant = state->variant;
const 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);
@@ -442,7 +428,10 @@ lp_rast_shade_quads_mask_sample(struct lp_rasterizer_task *task,
assert((y % 4) == 0);
/* color buffer */
for (i = 0; i < scene->fb.nr_cbufs; i++) {
uint8_t *color[PIPE_MAX_COLOR_BUFS];
unsigned stride[PIPE_MAX_COLOR_BUFS];
unsigned sample_stride[PIPE_MAX_COLOR_BUFS];
for (unsigned i = 0; i < scene->fb.nr_cbufs; i++) {
if (scene->fb.cbufs[i]) {
stride[i] = scene->cbufs[i].stride;
sample_stride[i] = scene->cbufs[i].sample_stride;
@@ -457,6 +446,9 @@ lp_rast_shade_quads_mask_sample(struct lp_rasterizer_task *task,
}
/* depth buffer */
uint8_t *depth = NULL;
unsigned depth_stride = 0;
unsigned depth_sample_stride = 0;
if (scene->zsbuf.map) {
depth_stride = scene->zsbuf.stride;
depth_sample_stride = scene->zsbuf.sample_stride;
@@ -494,6 +486,7 @@ lp_rast_shade_quads_mask_sample(struct lp_rasterizer_task *task,
}
}
void
lp_rast_shade_quads_mask(struct lp_rasterizer_task *task,
const struct lp_rast_shader_inputs *inputs,
@@ -506,6 +499,7 @@ lp_rast_shade_quads_mask(struct lp_rasterizer_task *task,
lp_rast_shade_quads_mask_sample(task, inputs, x, y, new_mask);
}
/**
* Directly copy pixels from a texture to the destination color buffer.
* This is a bin command called during bin processing.
@@ -519,15 +513,10 @@ lp_rast_blit_tile_to_dest(struct lp_rasterizer_task *task,
const struct lp_rast_state *state = task->state;
struct lp_fragment_shader_variant *variant = state->variant;
const struct lp_jit_texture *texture = &state->jit_context.textures[0];
const uint8_t *src;
uint8_t *dst;
unsigned src_stride;
unsigned dst_stride;
struct pipe_surface *cbuf = scene->fb.cbufs[0];
const unsigned face_slice = cbuf->u.tex.first_layer;
const unsigned level = cbuf->u.tex.level;
struct llvmpipe_resource *lpt = llvmpipe_resource(cbuf->texture);
int src_x, src_y;
LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__);
@@ -536,21 +525,20 @@ lp_rast_blit_tile_to_dest(struct lp_rasterizer_task *task,
return;
}
dst = llvmpipe_get_texture_image_address(lpt, face_slice, level);
uint8_t *dst = llvmpipe_get_texture_image_address(lpt, face_slice, level);
if (!dst)
return;
dst_stride = lpt->row_stride[level];
const unsigned dst_stride = lpt->row_stride[level];
src = texture->base;
src_stride = texture->row_stride[0];
const uint8_t *src = texture->base;
const unsigned src_stride = texture->row_stride[0];
src_x = util_iround(GET_A0(inputs)[1][0]*texture->width - 0.5f);
src_y = util_iround(GET_A0(inputs)[1][1]*texture->height - 0.5f);
int src_x = util_iround(GET_A0(inputs)[1][0]*texture->width - 0.5f);
int src_y = util_iround(GET_A0(inputs)[1][1]*texture->height - 0.5f);
src_x = src_x + task->x;
src_y = src_y + task->y;
src_x += task->x;
src_y += task->y;
if (0) {
union util_color uc;
@@ -586,18 +574,16 @@ lp_rast_blit_tile_to_dest(struct lp_rasterizer_task *task,
if (variant->shader->kind == LP_FS_KIND_BLIT_RGB1) {
if (cbuf->format == PIPE_FORMAT_B8G8R8A8_UNORM) {
int x, y;
dst += task->x * 4;
src += src_x * 4;
dst += task->y * dst_stride;
src += src_y * src_stride;
for (y = 0; y < task->height; ++y) {
for (int y = 0; y < task->height; ++y) {
const uint32_t *src_row = (const uint32_t *)src;
uint32_t *dst_row = (uint32_t *)dst;
for (x = 0; x < task->width; ++x) {
for (int x = 0; x < task->width; ++x) {
*dst_row++ = *src_row++ | 0xff000000;
}
dst += dst_stride;
@@ -617,6 +603,7 @@ lp_rast_blit_tile_to_dest(struct lp_rasterizer_task *task,
lp_rast_shade_tile_opaque(task, arg);
}
static void
lp_rast_blit_tile(struct lp_rasterizer_task *task,
const union lp_rast_cmd_arg arg)
@@ -626,6 +613,7 @@ lp_rast_blit_tile(struct lp_rasterizer_task *task,
lp_rast_blit_tile_to_dest(task, arg);
}
/**
* Begin a new occlusion query.
* This is a bin command put in all bins.
@@ -699,37 +687,31 @@ lp_rast_set_state(struct lp_rasterizer_task *task,
}
/**
* Called when we're done writing to a color tile.
*/
static void
lp_rast_tile_end(struct lp_rasterizer_task *task)
{
unsigned i;
for (i = 0; i < task->scene->num_active_queries; ++i) {
lp_rast_end_query(task, lp_rast_arg_query(task->scene->active_queries[i]));
for (unsigned i = 0; i < task->scene->num_active_queries; ++i) {
lp_rast_end_query(task,
lp_rast_arg_query(task->scene->active_queries[i]));
}
/* debug */
memset(task->color_tiles, 0, sizeof(task->color_tiles));
task->depth_tile = NULL;
task->bin = NULL;
}
/* Currently have two rendering paths only - the general case triangle
* path and the super-specialized blit/clear path.
*/
#define TRI ((LP_RAST_FLAGS_TRI <<1)-1) /* general case */
#define RECT ((LP_RAST_FLAGS_RECT<<1)-1) /* direct rectangle rasterizer */
#define BLIT ((LP_RAST_FLAGS_BLIT<<1)-1) /* write direct-to-dest */
#define TRI ((LP_RAST_FLAGS_TRI <<1)-1) /* general case */
#define RECT ((LP_RAST_FLAGS_RECT<<1)-1) /* direct rectangle rasterizer */
#define BLIT ((LP_RAST_FLAGS_BLIT<<1)-1) /* write direct-to-dest */
static const unsigned
rast_flags[] = {
@@ -773,12 +755,11 @@ rast_flags[] = {
TRI, /* lp_rast_triangle_ms_3_4 */
TRI, /* lp_rast_triangle_ms_3_16 */
TRI, /* lp_rast_triangle_ms_4_16 */
RECT, /* rectangle */
BLIT, /* blit */
};
/*
/*
*/
static const lp_rast_cmd_func
dispatch_blit[] = {
@@ -822,7 +803,6 @@ dispatch_blit[] = {
NULL, /* lp_rast_triangle_ms_3_4 */
NULL, /* lp_rast_triangle_ms_3_16 */
NULL, /* lp_rast_triangle_ms_4_16 */
NULL, /* rectangle */
lp_rast_blit_tile_to_dest,
};
@@ -926,27 +906,25 @@ dispatch_tri_debug[] =
lp_rast_triangle_ms_3_4,
lp_rast_triangle_ms_3_16,
lp_rast_triangle_ms_4_16,
lp_rast_rectangle,
lp_rast_shade_tile,
};
struct lp_bin_info
lp_characterize_bin(const struct cmd_bin *bin)
{
struct cmd_block *block;
struct lp_bin_info info;
unsigned andflags = ~0;
unsigned k, j = 0;
unsigned andflags = ~0, j = 0;
STATIC_ASSERT(ARRAY_SIZE(rast_flags) == LP_RAST_OP_MAX);
for (block = bin->head; block; block = block->next) {
for (k = 0; k < block->count; k++, j++) {
for (const struct cmd_block *block = bin->head; block; block = block->next) {
for (unsigned k = 0; k < block->count; k++, j++) {
andflags &= rast_flags[block->cmd[k]];
}
}
struct lp_bin_info info;
info.type = andflags;
info.count = j;
@@ -958,48 +936,41 @@ static void
blit_rasterize_bin(struct lp_rasterizer_task *task,
const struct cmd_bin *bin)
{
const struct cmd_block *block;
unsigned k;
STATIC_ASSERT(ARRAY_SIZE(dispatch_blit) == LP_RAST_OP_MAX);
if (0) debug_printf("%s\n", __FUNCTION__);
for (block = bin->head; block; block = block->next) {
for (k = 0; k < block->count; k++) {
dispatch_blit[block->cmd[k]]( task, block->arg[k] );
for (const struct cmd_block *block = bin->head; block; block = block->next) {
for (unsigned k = 0; k < block->count; k++) {
dispatch_blit[block->cmd[k]](task, block->arg[k]);
}
}
}
static void
tri_rasterize_bin(struct lp_rasterizer_task *task,
const struct cmd_bin *bin,
int x, int y)
{
const struct cmd_block *block;
unsigned k;
STATIC_ASSERT(ARRAY_SIZE(dispatch_tri) == LP_RAST_OP_MAX);
for (block = bin->head; block; block = block->next) {
for (k = 0; k < block->count; k++) {
dispatch_tri[block->cmd[k]]( task, block->arg[k] );
for (const struct cmd_block *block = bin->head; block; block = block->next) {
for (unsigned k = 0; k < block->count; k++) {
dispatch_tri[block->cmd[k]](task, block->arg[k]);
}
}
}
static void
debug_rasterize_bin(struct lp_rasterizer_task *task,
const struct cmd_bin *bin)
{
const struct cmd_block *block;
unsigned k;
STATIC_ASSERT(ARRAY_SIZE(dispatch_tri_debug) == LP_RAST_OP_MAX);
for (block = bin->head; block; block = block->next) {
for (k = 0; k < block->count; k++) {
dispatch_tri_debug[block->cmd[k]]( task, block->arg[k] );
for (const struct cmd_block *block = bin->head; block; block = block->next) {
for (unsigned k = 0; k < block->count; k++) {
dispatch_tri_debug[block->cmd[k]](task, block->arg[k]);
}
}
}
@@ -1013,22 +984,23 @@ debug_rasterize_bin(struct lp_rasterizer_task *task,
*/
static void
rasterize_bin(struct lp_rasterizer_task *task,
const struct cmd_bin *bin, int x, int y )
const struct cmd_bin *bin, int x, int y)
{
struct lp_bin_info info = lp_characterize_bin(bin);
lp_rast_tile_begin( task, bin, x, y );
lp_rast_tile_begin(task, bin, x, y);
if (LP_DEBUG & DEBUG_NO_FASTPATH)
if (LP_DEBUG & DEBUG_NO_FASTPATH) {
debug_rasterize_bin(task, bin);
else if (info.type & LP_RAST_FLAGS_BLIT)
} else if (info.type & LP_RAST_FLAGS_BLIT) {
blit_rasterize_bin(task, bin);
else if (task->scene->permit_linear_rasterizer &&
} else if (task->scene->permit_linear_rasterizer &&
!(LP_PERF & PERF_NO_RAST_LINEAR) &&
(info.type & LP_RAST_FLAGS_RECT))
(info.type & LP_RAST_FLAGS_RECT)) {
lp_linear_rasterize_bin(task, bin);
else
} else {
tri_rasterize_bin(task, bin, x, y);
}
lp_rast_tile_end(task);
@@ -1051,11 +1023,11 @@ rasterize_bin(struct lp_rasterizer_task *task,
* stores them again unchanged. This typically happens when bins have
* been flushed for some reason in the middle of a frame, or when
* incremental updates are being made to a render target.
*
*
* Try to avoid doing pointless work in this case.
*/
static boolean
is_empty_bin( const struct cmd_bin *bin )
is_empty_bin(const struct cmd_bin *bin)
{
return bin->head == NULL;
}
@@ -1090,7 +1062,7 @@ rasterize_scene(struct lp_rasterizer_task *task,
assert(scene);
while ((bin = lp_scene_bin_iter_next(scene, &i, &j))) {
if (!is_empty_bin( bin ))
if (!is_empty_bin(bin))
rasterize_bin(task, bin, i, j);
}
}
@@ -1123,8 +1095,8 @@ rasterize_scene(struct lp_rasterizer_task *task,
* Called by setup module when it has something for us to render.
*/
void
lp_rast_queue_scene( struct lp_rasterizer *rast,
struct lp_scene *scene)
lp_rast_queue_scene(struct lp_rasterizer *rast,
struct lp_scene *scene)
{
LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);
@@ -1132,16 +1104,16 @@ lp_rast_queue_scene( struct lp_rasterizer *rast,
/* no threading */
unsigned fpstate = util_fpstate_get();
/* Make sure that denorms are treated like zeros. This is
/* 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_rast_begin( rast, scene );
lp_rast_begin(rast, scene);
rasterize_scene( &rast->tasks[0], scene );
rasterize_scene(&rast->tasks[0], scene);
lp_rast_end( rast );
lp_rast_end(rast);
util_fpstate_set(fpstate);
@@ -1151,7 +1123,7 @@ lp_rast_queue_scene( struct lp_rasterizer *rast,
/* threaded rendering! */
unsigned i;
lp_scene_enqueue( rast->full_scenes, scene );
lp_scene_enqueue(rast->full_scenes, scene);
/* signal the threads that there's work to do */
for (i = 0; i < rast->num_threads; i++) {
@@ -1164,7 +1136,7 @@ lp_rast_queue_scene( struct lp_rasterizer *rast,
void
lp_rast_finish( struct lp_rasterizer *rast )
lp_rast_finish(struct lp_rasterizer *rast)
{
if (rast->num_threads == 0) {
/* nothing to do */
@@ -1194,15 +1166,14 @@ thread_function(void *init_data)
struct lp_rasterizer *rast = task->rast;
boolean debug = false;
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
/* Make sure that denorms are treated like zeros. This is
* the behavior required by D3D10. OpenGL doesn't care.
*/
fpstate = util_fpstate_get();
unsigned fpstate = util_fpstate_get();
util_fpstate_set_denorms_to_zero(fpstate);
while (1) {
@@ -1219,29 +1190,27 @@ thread_function(void *init_data)
* - get next scene to rasterize
* - map the framebuffer surfaces
*/
lp_rast_begin( rast,
lp_scene_dequeue( rast->full_scenes, TRUE ) );
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.
*/
util_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(task,
rast->curr_scene);
rasterize_scene(task, rast->curr_scene);
/* wait for all threads to finish with this scene */
util_barrier_wait( &rast->barrier );
util_barrier_wait(&rast->barrier);
/* XXX: shouldn't be necessary:
*/
if (task->thread_index == 0) {
lp_rast_end( rast );
lp_rast_end(rast);
}
/* signal done with work */
@@ -1265,10 +1234,8 @@ thread_function(void *init_data)
static void
create_rast_threads(struct lp_rasterizer *rast)
{
unsigned i;
/* NOTE: if num_threads is zero, we won't use any threads */
for (i = 0; i < rast->num_threads; i++) {
for (unsigned 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] = u_thread_create(thread_function,
@@ -1281,14 +1248,13 @@ create_rast_threads(struct lp_rasterizer *rast)
}
/**
* 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( unsigned num_threads )
lp_rast_create(unsigned num_threads)
{
struct lp_rasterizer *rast;
unsigned i;
@@ -1307,8 +1273,8 @@ lp_rast_create( unsigned num_threads )
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);
task->thread_data.cache =
align_malloc(sizeof(struct lp_build_format_cache), 16);
if (!task->thread_data.cache) {
goto no_thread_data_cache;
}
@@ -1322,7 +1288,7 @@ lp_rast_create( unsigned num_threads )
/* for synchronizing rasterization threads */
if (rast->num_threads > 0) {
util_barrier_init( &rast->barrier, rast->num_threads );
util_barrier_init(&rast->barrier, rast->num_threads);
}
memset(lp_dummy_tile, 0, sizeof lp_dummy_tile);
@@ -1346,47 +1312,48 @@ no_rast:
/* Shutdown:
*/
void lp_rast_destroy( struct lp_rasterizer *rast )
void
lp_rast_destroy(struct lp_rasterizer *rast)
{
unsigned i;
/* 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++) {
for (unsigned 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++) {
for (unsigned i = 0; i < rast->num_threads; i++) {
#ifdef _WIN32
/* Threads might already be dead - Windows apparently terminates other threads when
* returning from main.
/* Threads might already be dead - Windows apparently terminates
* other threads when returning from main.
*/
DWORD exit_code = STILL_ACTIVE;
if (GetExitCodeThread(rast->threads[i], &exit_code) && exit_code == STILL_ACTIVE)
if (GetExitCodeThread(rast->threads[i], &exit_code) &&
exit_code == STILL_ACTIVE) {
pipe_semaphore_wait(&rast->tasks[i].work_done);
}
#else
thrd_join(rast->threads[i], NULL);
#endif
}
/* Clean up per-thread data */
for (i = 0; i < rast->num_threads; i++) {
for (unsigned 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++) {
for (unsigned i = 0; i < MAX2(1, rast->num_threads); i++) {
align_free(rast->tasks[i].thread_data.cache);
}
/* for synchronizing rasterization threads */
if (rast->num_threads > 0) {
util_barrier_destroy( &rast->barrier );
util_barrier_destroy(&rast->barrier);
}
lp_scene_queue_destroy(rast->full_scenes);