[Mlir-commits] [mlir] [MLIR] Allowing unsupported conv2d op to fail gracefully from vectorization (PR #130181)

[Zhuoran Yin]

================
@@ -3125,58 +3225,32 @@ bool isSupportedPoolKind(vector::CombiningKind kind) {
 /// kw is unrolled, w is unrolled iff dilationW > 1.
 struct Conv1DGenerator
     : public StructuredGenerator<LinalgOp, utils::IteratorType> {
-  Conv1DGenerator(RewriterBase &rewriter, LinalgOp linalgOp, int strideW,
-                  int dilationW)
-      : StructuredGenerator<LinalgOp, utils::IteratorType>(rewriter, linalgOp),
-        strideW(strideW), dilationW(dilationW) {
-    // Determine whether `linalgOp` can be generated with this generator
-    if (linalgOp.getNumDpsInputs() != 2 || linalgOp.getNumDpsInits() != 1)
-      return;
+  Conv1DGenerator(RewriterBase &rewriter, LinalgOp linalgOp)
+      : StructuredGenerator<LinalgOp, utils::IteratorType>(rewriter, linalgOp) {
+
     lhsShaped = linalgOp.getDpsInputOperand(0)->get();
     rhsShaped = linalgOp.getDpsInputOperand(1)->get();
     resShaped = linalgOp.getDpsInitOperand(0)->get();
     lhsShapedType = dyn_cast<ShapedType>(lhsShaped.getType());
     rhsShapedType = dyn_cast<ShapedType>(rhsShaped.getType());
     resShapedType = dyn_cast<ShapedType>(resShaped.getType());
----------------
jerryyin wrote:

Those are all member variables, please refer to:

https://github.com/llvm/llvm-project/blob/2ef5555f82feb9f8dbaa1cac04cfc25013c0d16f/mlir/lib/Dialect/Linalg/Transforms/Vectorization.cpp#L3906-L3907

I'd be tempted to get rid of those too because those are a lot of state to track. I'm refraining myself from doing it to control the amount of refactor in this PR.

https://github.com/llvm/llvm-project/pull/130181