Add support for FREX flawless frames, failed attempt to fix intel igpus

This commit is contained in:
mcrcortex
2025-05-22 19:50:25 +10:00
parent a314c26b89
commit cdfa15c1f6
8 changed files with 77 additions and 34 deletions

View File

@@ -43,12 +43,20 @@ void main() {
barrier();
#ifdef IS_INTEL
uint val = subgroupExclusiveAdd(warpPrefixSum[gl_SubgroupInvocationID]);
barrier();
if (gl_SubgroupID == 0) {
warpPrefixSum[gl_SubgroupInvocationID] = val;
}
#else
if (gl_SubgroupID == 0) {
uint val = warpPrefixSum[gl_SubgroupInvocationID];
subgroupBarrier();
//Use warp to do entire add in 1 reduction
warpPrefixSum[gl_SubgroupInvocationID] = subgroupExclusiveAdd(val);
}
#endif
barrier();
//Add the computed sum across all threads and warps