raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-20 02:47:02 +02:00
Code Issues Packages Projects Releases Wiki Activity

gpuTiler: fix performance (dispatch a 2D 8×8 workgroup)

Browse Source
This commit is contained in:
nmzik
2026-07-22 23:36:00 +02:00
parent f60007e923
commit 85876388cf
2 changed files with 16 additions and 38 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/host_gpu/shaders/gpu_tiler_common.inc
+3 -13
View File
@@ -1,4 +1,4 @@
layout(local_size_x = 64) in;
layout(local_size_x = 8, local_size_y = 8) in;
layout(set = 0, binding = 0, std430) readonly buffer Input { uint data[]; } input_buffer;
layout(set = 0, binding = 1, std430) buffer Output { uint data[]; } output_buffer;
@@ -17,8 +17,6 @@ layout(push_constant) uniform Push {
uint tail_x;
uint tail_y;
uint tail;
uint first;
uint count;
} params;
void copy_element(uint src, uint dst) {
@@ -34,19 +32,11 @@ void copy_element(uint src, uint dst) {
}
void main() {
uint index = gl_GlobalInvocationID.x;
if (index >= params.count) {
uvec3 p = gl_GlobalInvocationID;
if (p.x >= params.width || p.y >= params.height || p.z >= params.depth) {
return;
}
uint element = params.first + index;
uint plane = params.width * params.height;
uvec3 p;
p.z = element / plane;
element -= p.z * plane;
p.y = element / params.width;
p.x = element - p.y * params.width;
uvec3 extent = block_extent();
uvec3 swizzle = p;
uvec3 block = uvec3(0);