jayfoad wrote:
> This logic would need updating again for GFX12. It seems like it's
> duplicating a lot of knowledge which is already implemented in
> SIInsertWaitcnts.
Just to demonstrate, you could implement this feature in SIInsertWaitcnts for
**all** supported architectures with something like:
```diff
diff --git a/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp
b/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp
index 6ecb1c8bf6e1..910cd094f8f2 100644
--- a/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp
+++ b/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp
@@ -2299,6 +2299,12 @@ bool
SIInsertWaitcnts::insertWaitcntInBlock(MachineFunction &MF,
updateEventWaitcntAfter(Inst, &ScoreBrackets);
+ AMDGPU::Waitcnt Wait =
+ AMDGPU::Waitcnt::allZeroExceptVsCnt(ST->hasExtendedWaitCounts());
+ ScoreBrackets.simplifyWaitcnt(Wait);
+ Modified |= generateWaitcnt(Wait, std::next(Inst.getIterator()), Block,
+ ScoreBrackets, /*OldWaitcntInstr=*/nullptr);
+
#if 0 // TODO: implement resource type check controlled by options with ub =
LB.
// If this instruction generates a S_SETVSKIP because it is an
// indexed resource, and we are on Tahiti, then it will also force
```
Handling VSCNT/STORECNT correctly is a little more complicated but not much.
https://github.com/llvm/llvm-project/pull/79236
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
