[Mlir-commits] [mlir] [mlir][linalg] Split GenericPadOpVectorizationPattern into two patterns (PR #111349)

[Andrzej Warzyński]

================
@@ -2604,6 +2495,177 @@ struct PadOpVectorizationWithTransferWritePattern
   }
 };
 
+/// Given an ArrayRef of OpFoldResults, return a vector of Values.
+/// IntegerAttrs are converted to ConstantIndexOps. Other attribute types are
+/// not supported.
+static SmallVector<Value> ofrToIndexValues(RewriterBase &rewriter, Location loc,
+                                           ArrayRef<OpFoldResult> ofrs) {
+  SmallVector<Value> result;
+  for (auto o : ofrs) {
+    if (auto val = llvm::dyn_cast_if_present<Value>(o)) {
+      result.push_back(val);
+    } else {
+      result.push_back(rewriter.create<arith::ConstantIndexOp>(
+          loc, cast<IntegerAttr>(cast<Attribute>(o)).getInt()));
+    }
+  }
+  return result;
+}
+
+/// Returns the effective Pad value for the input op, provided it's a scalar.
+///
+/// Many Ops exhibit pad-like behaviour, but this isn't always explicit. If
+/// this Op performs padding, retrieve the padding value provided that it's
+/// a scalar and static/fixed for all the padded values. Returns an empty value
+/// otherwise.
+static Value getStaticPadVl(Operation *op) {
+  if (!op)
+    return {};
+
+  // 1. vector.broadcast - return the value that's being broadcast,
+  // provided that it's a scalar.
+  if (auto bcast = llvm::dyn_cast<vector::BroadcastOp>(op)) {
+    auto source = bcast.getSource();
+    if (llvm::dyn_cast<VectorType>(source.getType()))
+      return {};
----------------
banach-space wrote:

Added a small note to clarify.

https://github.com/llvm/llvm-project/pull/111349