[mlir][vector] Prevent folding non memref-type gather into maskedload (llvm#135371)

This patch fixes an issue in the FoldContiguousGather pattern which was
incorrectly folding vector.gather operations with contiguous indices
into vector.maskedload operations regardless of the base operand type.

While vector.gather operations can work on both tensor and memref types,
vector.maskedload operations are only valid for memref types. The
pattern was incorrectly lowering a tensor-based gather into a
masked-load, which is invalid.

This fix adds a type check to ensure the pattern only applies to
memref-based gather operations.

Co-authored-by: Sagar Kulkarni <[email protected]>

Author: sagarkulkarni19

mlir/lib/Dialect/Vector/IR/VectorOps.cpp

@@ -5348,6 +5348,9 @@ class FoldContiguousGather final : public OpRewritePattern<GatherOp> {
   using OpRewritePattern::OpRewritePattern;
   LogicalResult matchAndRewrite(GatherOp op,
                                 PatternRewriter &rewriter) const override {
+    if (!op.getBase().getType().isa<MemRefType>())
+      return rewriter.notifyMatchFailure(op, "base must be of memref type");
+
     if (failed(isZeroBasedContiguousSeq(op.getIndexVec())))
       return failure();
 

mlir/test/Dialect/Vector/canonicalize.mlir

@@ -3198,6 +3198,19 @@ func.func @contiguous_gather_step(%base: memref<?xf32>,
 
 // -----
 
+// CHECK-LABEL: @no_fold_contiguous_gather_tensor
+func.func @no_fold_contiguous_gather_tensor(%base: tensor<8xf32>, %mask: vector<4xi1>, %pass_thru: vector<4xf32>) -> vector<4xf32> {
+  %c0 = arith.constant 0 : index
+  %indices = arith.constant dense<[0, 1, 2, 3]> : vector<4xindex>
+  // CHECK: vector.gather
+  // CHECK-NOT: vector.maskedload
+  %0 = vector.gather %base[%c0][%indices], %mask, %pass_thru :
+    tensor<8xf32>, vector<4xindex>, vector<4xi1>, vector<4xf32> into vector<4xf32>
+  return %0 : vector<4xf32>
+}
+
+// -----
+
 // CHECK-LABEL: @gather_broadcast(
 // TODO: Broadcast is not supported yet
 //       CHECK:   %[[R:.*]] = vector.gather