[mlir] Add forall canonicalization to replace constant induction vars

[Max191]

Adds a canonicalization pattern for scf.forall that replaces constant induction variables with a constant index. There is a similar canonicalization that completely removes constant induction variables from the loop, but that pattern does not apply on foralls with mappings, so this one is necessary for those cases.

[llvmbot]

@llvm/pr-subscribers-mlir-scf

@llvm/pr-subscribers-mlir

Author: None (Max191)

Changes

Adds a canonicalization pattern for scf.forall that replaces constant induction variables with a constant index. There is a similar canonicalization that completely removes constant induction variables from the loop, but that pattern does not apply on foralls with mappings, so this one is necessary for those cases.

---

Full diff: https://github.com/llvm/llvm-project/pull/112764.diff

2 Files Affected:

• (modified) mlir/lib/Dialect/SCF/IR/SCF.cpp (+28-1)
• (modified) mlir/test/Dialect/SCF/canonicalize.mlir (+2)

diff --git a/mlir/lib/Dialect/SCF/IR/SCF.cpp b/mlir/lib/Dialect/SCF/IR/SCF.cpp
index 2582d4e0df1920..7789f21af00780 100644
--- a/mlir/lib/Dialect/SCF/IR/SCF.cpp
+++ b/mlir/lib/Dialect/SCF/IR/SCF.cpp
@@ -1767,6 +1767,32 @@ struct ForallOpSingleOrZeroIterationDimsFolder
   }
 };
 
+struct ForallOpReplaceConstantInductionVar : public OpRewritePattern<ForallOp> {
+  using OpRewritePattern<ForallOp>::OpRewritePattern;
+
+  LogicalResult matchAndRewrite(ForallOp op,
+                                PatternRewriter &rewriter) const override {
+    // Replace all induction vars with a single trip count with their lower
+    // bound.
+    Location loc = op.getLoc();
+    bool replacedIv = false;
+    for (auto [lb, ub, step, iv] :
+         llvm::zip(op.getMixedLowerBound(), op.getMixedUpperBound(),
+                   op.getMixedStep(), op.getInductionVars())) {
+      if (iv.getUses().begin() == iv.getUses().end())
+        continue;
+      auto numIterations = constantTripCount(lb, ub, step);
+      if (!numIterations.has_value() || numIterations.value() != 1) {
+        continue;
+      }
+      rewriter.replaceAllUsesWith(
+          iv, getValueOrCreateConstantIndexOp(rewriter, loc, lb));
+      return success();
+    }
+    return failure();
+  }
+};
+
 struct FoldTensorCastOfOutputIntoForallOp
     : public OpRewritePattern<scf::ForallOp> {
   using OpRewritePattern<scf::ForallOp>::OpRewritePattern;
@@ -1851,7 +1877,8 @@ void ForallOp::getCanonicalizationPatterns(RewritePatternSet &results,
                                            MLIRContext *context) {
   results.add<DimOfForallOp, FoldTensorCastOfOutputIntoForallOp,
               ForallOpControlOperandsFolder, ForallOpIterArgsFolder,
-              ForallOpSingleOrZeroIterationDimsFolder>(context);
+              ForallOpSingleOrZeroIterationDimsFolder,
+              ForallOpReplaceConstantInductionVar>(context);
 }
 
 /// Given the region at `index`, or the parent operation if `index` is None,
diff --git a/mlir/test/Dialect/SCF/canonicalize.mlir b/mlir/test/Dialect/SCF/canonicalize.mlir
index c68369a8e4fce7..6f4703c04dc768 100644
--- a/mlir/test/Dialect/SCF/canonicalize.mlir
+++ b/mlir/test/Dialect/SCF/canonicalize.mlir
@@ -1632,6 +1632,8 @@ func.func @do_not_inline_distributed_forall_loop(
 }
 // CHECK-LABEL: @do_not_inline_distributed_forall_loop
 // CHECK: scf.forall
+// CHECK:   tensor.extract_slice %{{.*}}[0, 0] [2, 3] [1, 1]
+// CHECK:   tensor.parallel_insert_slice %{{.*}}[0, 0] [2, 3] [1, 1]
 
 // -----
 

[hanhanW]

LGTM, just one suggestion about the comment.