joejulian/gio
1
0
Fork 1
mirror of https://git.sr.ht/~eliasnaur/gio synced 2026-07-04 08:55:35 +00:00
Code Issues Packages Projects Releases Wiki Activity

gpu/shaders: ensure dynamically uniform barriers when malloc fails

Browse Source 
GPU APIs require that barrier() calls are dynamically uniform, that is
for every barrier in the code, every shader invocation in a workgroup
must all call it, or all not call it.

Signed-off-by: Elias Naur <mail@eliasnaur.com>
This commit is contained in:
Elias Naur
2021-04-11 16:34:33 +02:00
parent f930e3f3d7
commit ee519351f7
9 changed files with 564 additions and 580 deletions
Download Patch File Download Diff File Expand all files Collapse all files
gpu/shaders/coarse.comp
+14 -17
View File
@@ -56,7 +56,7 @@ void write_tile_alloc(uint el_ix, Alloc a) {
sh_tile_alloc[el_ix] = a;
}
Alloc read_tile_alloc(uint el_ix) {
Alloc read_tile_alloc(uint el_ix, bool mem_ok) {
return sh_tile_alloc[el_ix];
}
#else
@@ -64,9 +64,9 @@ void write_tile_alloc(uint el_ix, Alloc a) {
// No-op
}
Alloc read_tile_alloc(uint el_ix) {
Alloc read_tile_alloc(uint el_ix, bool mem_ok) {
// All memory.
return new_alloc(0, memory.length()*4);
return new_alloc(0, memory.length()*4, mem_ok);
}
#endif
@@ -109,10 +109,6 @@ void write_fill(Alloc alloc, inout CmdRef cmd_ref, uint flags, Tile tile, float
}
void main() {
if (mem_error != NO_ERROR) {
return;
}
// Could use either linear or 2d layouts for both dispatch and
// invocations within the workgroup. We'll use variables to abstract.
uint width_in_bins = (conf.width_in_tiles + N_TILE_X - 1)/N_TILE_X;
@@ -158,6 +154,7 @@ void main() {
uint num_begin_slots = 0;
uint begin_slot = 0;
bool mem_ok = mem_error == NO_ERROR;
while (true) {
for (uint i = 0; i < N_SLICE; i++) {
sh_bitmaps[i][th_ix] = 0;
@@ -172,7 +169,7 @@ void main() {
uint in_ix = (conf.bin_alloc.offset >> 2) + ((partition_ix + th_ix) * N_TILE + bin_ix) * 2;
count = read_mem(conf.bin_alloc, in_ix);
uint offset = read_mem(conf.bin_alloc, in_ix + 1);
sh_part_elements[th_ix] = new_alloc(offset, count*BinInstance_size);
sh_part_elements[th_ix] = new_alloc(offset, count*BinInstance_size, mem_ok);
}
// prefix sum of counts
for (uint i = 0; i < LG_N_PART_READ; i++) {
@@ -196,7 +193,7 @@ void main() {
}
// use binary search to find element to read
uint ix = rd_ix + th_ix;
if (ix >= wr_ix && ix < ready_ix) {
if (ix >= wr_ix && ix < ready_ix && mem_ok) {
uint part_ix = 0;
for (uint i = 0; i < LG_N_PART_READ; i++) {
uint probe = part_ix + ((N_PART_READ / 2) >> i);
@@ -253,7 +250,7 @@ void main() {
// base relative to bin
uint base = path.tiles.offset - uint(dy * stride + dx) * Tile_size;
sh_tile_base[th_ix] = base;
Alloc path_alloc = new_alloc(path.tiles.offset, (path.bbox.z - path.bbox.x) * (path.bbox.w - path.bbox.y) * Tile_size);
Alloc path_alloc = new_alloc(path.tiles.offset, (path.bbox.z - path.bbox.x) * (path.bbox.w - path.bbox.y) * Tile_size, mem_ok);
write_tile_alloc(th_ix, path_alloc);
break;
default:
@@ -288,11 +285,11 @@ void main() {
uint width = sh_tile_width[el_ix];
uint x = sh_tile_x0[el_ix] + seq_ix % width;
uint y = sh_tile_y0[el_ix] + seq_ix / width;
bool include_tile;
bool include_tile = false;
if (tag == Annotated_BeginClip || tag == Annotated_EndClip) {
include_tile = true;
} else {
Tile tile = Tile_read(read_tile_alloc(el_ix), TileRef(sh_tile_base[el_ix] + (sh_tile_stride[el_ix] * y + x) * Tile_size));
} else if (mem_ok) {
Tile tile = Tile_read(read_tile_alloc(el_ix, mem_ok), TileRef(sh_tile_base[el_ix] + (sh_tile_stride[el_ix] * y + x) * Tile_size));
// Include the path in the tile if
// - the tile contains at least a segment (tile offset non-zero)
// - the tile is completely covered (backdrop non-zero)
@@ -311,7 +308,7 @@ void main() {
// through the non-segment elements.
uint slice_ix = 0;
uint bitmap = sh_bitmaps[0][th_ix];
while (true) {
while (mem_ok) {
if (bitmap == 0) {
slice_ix++;
if (slice_ix == N_SLICE) {
@@ -337,7 +334,7 @@ void main() {
if (clip_zero_depth == 0) {
switch (tag.tag) {
case Annotated_Color:
Tile tile = Tile_read(read_tile_alloc(element_ref_ix), TileRef(sh_tile_base[element_ref_ix]
Tile tile = Tile_read(read_tile_alloc(element_ref_ix, mem_ok), TileRef(sh_tile_base[element_ref_ix]
+ (sh_tile_stride[element_ref_ix] * tile_y + tile_x) * Tile_size));
AnnoColor fill = Annotated_Color_read(conf.anno_alloc, ref);
if (!alloc_cmd(cmd_alloc, cmd_ref, cmd_limit)) {
@@ -348,7 +345,7 @@ void main() {
cmd_ref.offset += 4 + CmdColor_size;
break;
case Annotated_Image:
tile = Tile_read(read_tile_alloc(element_ref_ix), TileRef(sh_tile_base[element_ref_ix]
tile = Tile_read(read_tile_alloc(element_ref_ix, mem_ok), TileRef(sh_tile_base[element_ref_ix]
+ (sh_tile_stride[element_ref_ix] * tile_y + tile_x) * Tile_size));
AnnoImage fill_img = Annotated_Image_read(conf.anno_alloc, ref);
if (!alloc_cmd(cmd_alloc, cmd_ref, cmd_limit)) {
@@ -359,7 +356,7 @@ void main() {
cmd_ref.offset += 4 + CmdImage_size;
break;
case Annotated_BeginClip:
tile = Tile_read(read_tile_alloc(element_ref_ix), TileRef(sh_tile_base[element_ref_ix]
tile = Tile_read(read_tile_alloc(element_ref_ix, mem_ok), TileRef(sh_tile_base[element_ref_ix]
+ (sh_tile_stride[element_ref_ix] * tile_y + tile_x) * Tile_size));
if (tile.tile.offset == 0 && tile.backdrop == 0) {
clip_zero_depth = clip_depth + 1;