[Mlir-commits] [mlir] [mlir][linalg] Enable masked vectorisation for depthwise convolutions (PR #81625)

[Diego Caballero]

================
@@ -1608,7 +1610,31 @@ static LogicalResult reductionPreconditions(LinalgOp op) {
   return success();
 }
 
+static LogicalResult vectorizeDynamicConvOpPrecondition(linalg::LinalgOp conv) {
+  if (!isa<linalg::DepthwiseConv1DNwcWcOp>(conv)) {
+    LDBG("Not a depth-wise 1D conv, dynamic shapes are not supported\n");
+    return failure();
+  }
+
+  // Support dynamic shapes in 1D depthwise convolution, but only in the
+  // _channel_ dimension. That's exclusively to support scalable
+  // vectorisation.
----------------
dcaballe wrote:

Is this still the case?

https://github.com/llvm/llvm-project/pull/81625