[Mlir-commits] [mlir] [mlir][vector] add tensor.concat, bitcast, expand_shape, collapse_shape vectorization support (PR #97297)

[Oleksandr Alex Zinenko]

================
@@ -1718,6 +1718,209 @@ vectorizeAsTensorPadOp(RewriterBase &rewriter, tensor::PadOp padOp,
   return success();
 }
 
+/// Vectorize a `tensor::expandshape` to these 3 Ops:
+///   Vector::TransferReadOp - Reads a vector from the source tensor
+///   ShapeCastOp - Reshape the data based on the target.
+///   vector::TransferWriteOp. - Write the result vector back to the destination
+///   tensor
+static LogicalResult lowerTensorReshape(RewriterBase &rewriter,
+                                        Operation *inputOp,
+                                        ArrayRef<int64_t> inputVectorSizes,
+                                        SmallVectorImpl<Value> &newResults) {
+  OpBuilder::InsertionGuard g(rewriter);
+  rewriter.setInsertionPoint(inputOp);
+  auto src = inputOp->getOperand(0);
+  auto srcType = mlir::dyn_cast<ShapedType>(src.getType());
+  auto result = inputOp->getResults()[0];
+  auto resultType = mlir::dyn_cast<ShapedType>(result.getType());
+  ArrayRef<int64_t> resultShape = resultType.getShape();
+  ArrayRef<int64_t> srcShape = srcType.getShape();
----------------
ftynse wrote:

This most likely needs at least an assertion that shapes are static. The logic below will become wild if given a `kDynamic`.

https://github.com/llvm/llvm-project/pull/97297