[Mlir-commits] [mlir] [mlir][amdgpu] Fuse adjacent `MemoryCounterWaitOp` (PR #171148)

[Jakub Kuderski]

================
@@ -596,6 +596,49 @@ LogicalResult PermlaneSwapOp::verify() {
   return success();
 }
 
+//===----------------------------------------------------------------------===//
+// MemoryCounterWaitOp
+//===----------------------------------------------------------------------===//
+
+namespace {
+/// Fuse adjacent memory counter wait ops, taking the minimum value of the
+/// counters.
+struct FuseMemoryCounterWaitOp final : OpRewritePattern<MemoryCounterWaitOp> {
+  using OpRewritePattern::OpRewritePattern;
+
+  LogicalResult matchAndRewrite(MemoryCounterWaitOp op,
+                                PatternRewriter &rewriter) const override {
+    auto next = dyn_cast<MemoryCounterWaitOp>(op->getNextNode());
+    if (!next)
+      return failure();
+
+    auto setters = {&MemoryCounterWaitOp::setLoad,
+                    &MemoryCounterWaitOp::setStore, &MemoryCounterWaitOp::setDs,
+                    &MemoryCounterWaitOp::setExp};
+    auto lhsVals = {op.getLoad(), op.getStore(), op.getDs(), op.getExp()};
+    auto rhsVals = {next.getLoad(), next.getStore(), next.getDs(),
+                    next.getExp()};
+    for (const auto &[setter, lhs, rhs] :
+         llvm::zip(setters, lhsVals, rhsVals)) {
----------------
kuhar wrote:

can we zip_equals?

https://github.com/llvm/llvm-project/pull/171148