[Mlir-commits] [mlir] [MLIR][SCF] Add canonicalization pattern to fold away iter args of scf.forall (PR #90189)
Matthias Springer
llvmlistbot at llvm.org
Mon May 6 04:32:50 PDT 2024
================
@@ -1415,6 +1415,19 @@ InParallelOp ForallOp::getTerminator() {
return cast<InParallelOp>(getBody()->getTerminator());
}
+SmallVector<Operation *> ForallOp::getStoreOpUser(BlockArgument bbArg) {
+ SmallVector<Operation *> storeOps;
+ InParallelOp inParallelOp = getTerminator();
+ for (Operation &yieldOp : inParallelOp.getYieldingOps()) {
----------------
matthias-springer wrote:
I was hoping that `ParallelCombiningOpInterface` is implemented by the ops in the "combining" region (e.g., `parallel_insert_slice`), but its actually implemented by the op that has the region (`in_parallel`). Then we wouldn't have to hard-code `parallel_insert_slice` here. But nvm, we can fix that later...
https://github.com/llvm/llvm-project/pull/90189
More information about the Mlir-commits
mailing list