================
@@ -1333,6 +1308,56 @@ static void addPreloadKernArgHint(Function &F,
TargetMachine &TM) {
}
}
+/// The final check and update of the attribute 'amdgpu-waves-per-eu' based on
+/// the determined 'amdgpu-flat-work-group-size' attribute. We can't do this
+/// during attributor run because the two attributes grow in opposite
direction,
+/// we should not use any intermediate value to calculate waves per eu until we
+/// have a determined flat workgroup size.
+static void updateWavesPerEU(Module &M, TargetMachine &TM) {
+ for (Function &F : M) {
+ const GCNSubtarget &ST = TM.getSubtarget<GCNSubtarget>(F);
+
+ auto FlatWgrpSizeAttr =
+ AMDGPU::getIntegerPairAttribute(F, "amdgpu-flat-work-group-size");
+
+ unsigned MinWavesPerEU = ST.getMinWavesPerEU();
+ unsigned MaxWavesPerEU = ST.getMaxWavesPerEU();
+
+ unsigned MinFlatWgrpSize = ST.getMinFlatWorkGroupSize();
+ unsigned MaxFlatWgrpSize = ST.getMaxFlatWorkGroupSize();
+ if (FlatWgrpSizeAttr.has_value()) {
+ MinFlatWgrpSize = FlatWgrpSizeAttr->first;
+ MaxFlatWgrpSize = *(FlatWgrpSizeAttr->second);
+ }
+
+ // Start with the max range.
+ unsigned Min = MinWavesPerEU;
+ unsigned Max = MinWavesPerEU;
+
+ // Compute the range from flat workgroup size. `getWavesPerEU` will also
+ // account for the 'amdgpu-waves-er-eu' attribute.
+ auto [MinFromFlatWgrpSize, MaxFromFlatWgrpSize] =
+ ST.getWavesPerEU(F, std::make_pair(MinFlatWgrpSize, MaxFlatWgrpSize));
----------------
arsenm wrote:
```suggestion
ST.getWavesPerEU(F, {MinFlatWgrpSize, MaxFlatWgrpSize});
```
https://github.com/llvm/llvm-project/pull/123995
_______________________________________________
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
