mirror of
https://github.com/ROCm/ROCm.git
synced 2026-04-05 03:01:17 -04:00
fix ifu gemm perf regression (#348)
This commit is contained in:
@@ -419,6 +419,13 @@ SmallVector<Value> LayoutPropagation::propagateToUsers(Value value,
|
||||
user->hasTrait<mlir::OpTrait::Elementwise>() ||
|
||||
isa<triton::ReduceOp, triton::ExpandDimsOp,
|
||||
triton::gpu::ConvertLayoutOp>(user)) {
|
||||
#ifdef USE_ROCM
|
||||
if (auto convertOp = dyn_cast<triton::gpu::ConvertLayoutOp>(user)) {
|
||||
if (triton::gpu::isSharedEncoding(convertOp.getResult()) ||
|
||||
triton::gpu::isSharedEncoding(convertOp.getOperand()))
|
||||
continue;
|
||||
}
|
||||
#endif
|
||||
setEncoding(user->getResults(), info, changed, user);
|
||||
continue;
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user