[MLIR][MemRef] Add alloca support for erase_dead_alloc_and_stores

[Lallapallooza]

Previously, erase_dead_alloc_and_stores didn't support memref.alloca. This patch introduces support for it.

[github-actions]

Thank you for submitting a Pull Request (PR) to the LLVM Project!

This PR will be automatically labeled and the relevant teams will be notified.

If you wish to, you can add reviewers by using the "Reviewers" section on this page.

If this is not working for you, it is probably because you do not have write permissions for the repository. In which case you can instead tag reviewers by name in a comment by using @ followed by their GitHub username.

If you have received no comments on your PR for a week, you can request a review by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate is once a week. Please remember that you are asking for valuable time from other developers.

If you have further questions, they may be answered by the LLVM GitHub User Guide.

You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums.

[llvmbot]

@llvm/pr-subscribers-mlir

@llvm/pr-subscribers-mlir-memref

Author: Vitalii Shutov (Lallapallooza)

Changes

Previously, erase_dead_alloc_and_stores didn't support memref.alloca. This patch introduces support for it.

---

Full diff: https://github.com/llvm/llvm-project/pull/142131.diff

3 Files Affected:

• (modified) mlir/include/mlir/Dialect/MemRef/TransformOps/MemRefTransformOps.td (+1-1)
• (modified) mlir/lib/Dialect/MemRef/Utils/MemRefUtils.cpp (+7-3)
• (modified) mlir/test/Dialect/MemRef/transform-ops.mlir (+24)

diff --git a/mlir/include/mlir/Dialect/MemRef/TransformOps/MemRefTransformOps.td b/mlir/include/mlir/Dialect/MemRef/TransformOps/MemRefTransformOps.td
index 2d060f3c2da64..f4694a30a8a12 100644
--- a/mlir/include/mlir/Dialect/MemRef/TransformOps/MemRefTransformOps.td
+++ b/mlir/include/mlir/Dialect/MemRef/TransformOps/MemRefTransformOps.td
@@ -245,7 +245,7 @@ def MemRefEraseDeadAllocAndStoresOp
     ]> {
   let description = [{
     This applies memory optimization on memref. In particular it does store to
-    load forwarding, dead store elimination and dead alloc elimination.
+    load forwarding, dead store elimination and dead alloc/alloca elimination.
 
     #### Return modes
 
diff --git a/mlir/lib/Dialect/MemRef/Utils/MemRefUtils.cpp b/mlir/lib/Dialect/MemRef/Utils/MemRefUtils.cpp
index 3f9fb071e0ba8..bd30276339812 100644
--- a/mlir/lib/Dialect/MemRef/Utils/MemRefUtils.cpp
+++ b/mlir/lib/Dialect/MemRef/Utils/MemRefUtils.cpp
@@ -156,13 +156,17 @@ static bool resultIsNotRead(Operation *op, std::vector<Operation *> &uses) {
 
 void eraseDeadAllocAndStores(RewriterBase &rewriter, Operation *parentOp) {
   std::vector<Operation *> opToErase;
-  parentOp->walk([&](memref::AllocOp op) {
+  auto collectOpsToErase = [&](Operation *op) {
     std::vector<Operation *> candidates;
     if (resultIsNotRead(op, candidates)) {
       llvm::append_range(opToErase, candidates);
-      opToErase.push_back(op.getOperation());
+      opToErase.push_back(op);
     }
-  });
+  };
+
+  parentOp->walk([&](memref::AllocOp op) { collectOpsToErase(op); });
+  parentOp->walk([&](memref::AllocaOp op) { collectOpsToErase(op); });
+
   for (Operation *op : opToErase)
     rewriter.eraseOp(op);
 }
diff --git a/mlir/test/Dialect/MemRef/transform-ops.mlir b/mlir/test/Dialect/MemRef/transform-ops.mlir
index acab37e482cfe..3b37c62fcb28e 100644
--- a/mlir/test/Dialect/MemRef/transform-ops.mlir
+++ b/mlir/test/Dialect/MemRef/transform-ops.mlir
@@ -327,6 +327,30 @@ module attributes {transform.with_named_sequence} {
   }
 }
 
+// -----
+
+// CHECK-LABEL: func.func @dead_alloca
+func.func @dead_alloca() {
+  // CHECK-NOT: %{{.+}} = memref.alloca
+  %0 = memref.alloca() : memref<8x64xf32, 3>
+  %1 = memref.subview %0[0, 0] [8, 4] [1, 1] : memref<8x64xf32, 3> to
+    memref<8x4xf32, affine_map<(d0, d1) -> (d0 * 64 + d1)>, 3>
+  %c0 = arith.constant 0 : index
+  %cst_0 = arith.constant dense<0.000000e+00> : vector<1x4xf32>
+  vector.transfer_write %cst_0, %1[%c0, %c0] {in_bounds = [true, true]} :
+    vector<1x4xf32>, memref<8x4xf32, affine_map<(d0, d1) -> (d0 * 64 + d1)>, 3>
+  return
+}
+
+module attributes {transform.with_named_sequence} {
+  transform.named_sequence @__transform_main(%arg1: !transform.any_op {transform.readonly}) {
+    %0 = transform.structured.match ops{["func.func"]} in %arg1 : (!transform.any_op) -> !transform.any_op
+    transform.memref.erase_dead_alloc_and_stores %0 : (!transform.any_op) -> ()
+    transform.yield
+  }
+}
+
+
 // -----
 
 // CHECK-LABEL: @store_to_load

[MaheshRavishankar]

We dont need two separate walks. We can do it in a single walk and check for AllocOp or AllocaOp.

[Lallapallooza]

Done

[Lallapallooza]

Ping.

[Lallapallooza]

Ping

[amirBish]

LGTM.

[github-actions]

@Lallapallooza Congratulations on having your first Pull Request (PR) merged into the LLVM Project!

Your changes will be combined with recent changes from other authors, then tested by our build bots. If there is a problem with a build, you may receive a report in an email or a comment on this PR.

Please check whether problems have been caused by your change specifically, as the builds can include changes from many authors. It is not uncommon for your change to be included in a build that fails due to someone else's changes, or infrastructure issues.

How to do this, and the rest of the post-merge process, is covered in detail here.

If your change does cause a problem, it may be reverted, or you can revert it yourself. This is a normal part of LLVM development. You can fix your changes and open a new PR to merge them again.

If you don't get any reports, no action is required from you. Your changes are working as expected, well done!