[mlir][vector] Make TransposeOpLowering configurable (#73915)

Following the discussion here:

  * #72105

this patch makes the `TransposeOpLowering` configurable so that one can select
whether to favour `vector.shape_cast` over `vector.transpose`.

As per the discussion in #72105, using `vector.shape_cast` is very beneficial
and desirable when targeting `LLVM IR` (CPU lowering), but won't work when
targeting `SPIR-V` today (GPU lowering). Hence the need for a mechanism to be
able to disable/enable the pattern introduced in #72105. This patch proposes one
such mechanism.

While this should solve the problem that we are facing today, it's understood to
be a temporary workaround. It should be removed once support for lowering
`vector.shape_cast` to SPIR-V is added. Also, (once implemented) the following
proposal might make this workaround redundant:

  * https://discourse.llvm.org/t/improving-handling-of-unit-dimensions-in-the-vector-dialect/

Author: banach-space

mlir/include/mlir/Dialect/Vector/Transforms/VectorTransforms.h

@@ -59,6 +59,16 @@ struct VectorTransformsOptions {
     vectorTransferSplit = opt;
     return *this;
   }
+
+  /// Option to control if vector.transpose can lower to a vector.shape_cast.
+  /// TODO: ATM it's not possible to lower `vector.shape_cast` to SPIR-V
+  /// and hence the need for this opt-out. Once the missing support has been
+  /// added, this option can be removed.
+  bool useShapeCast = true;
+  VectorTransformsOptions &setUseShapeCast(bool opt = true) {
+    useShapeCast = opt;
+    return *this;
+  }
 };
 
 //===----------------------------------------------------------------------===//

mlir/lib/Dialect/Vector/Transforms/LowerVectorTranspose.cpp

@@ -334,22 +334,24 @@ class TransposeOpLowering : public OpRewritePattern<vector::TransposeOp> {
       return rewriter.notifyMatchFailure(
           op, "Options specifies lowering to shuffle");
 
-    // Replace:
-    //   vector.transpose %0, [1, 0] : vector<nx1x<eltty>> to
-    //                                 vector<1xnxelty>
-    // with:
-    //   vector.shape_cast %0 : vector<nx1x<eltty>> to vector<1xnxelty>
-    //
-    // Source with leading unit dim (inverse) is also replaced. Unit dim must
-    // be fixed. Non-unit can be scalable.
-    if (resType.getRank() == 2 &&
-        ((resType.getShape().front() == 1 &&
-          !resType.getScalableDims().front()) ||
-         (resType.getShape().back() == 1 &&
-          !resType.getScalableDims().back())) &&
-        transp == ArrayRef<int64_t>({1, 0})) {
-      rewriter.replaceOpWithNewOp<vector::ShapeCastOp>(op, resType, input);
-      return success();
+    if (vectorTransformOptions.useShapeCast) {
+      // Replace:
+      //   vector.transpose %0, [1, 0] : vector<nx1x<eltty>> to
+      //                                 vector<1xnxelty>
+      // with:
+      //   vector.shape_cast %0 : vector<nx1x<eltty>> to vector<1xnxelty>
+      //
+      // Source with leading unit dim (inverse) is also replaced. Unit dim must
+      // be fixed. Non-unit can be scalable.
+      if (resType.getRank() == 2 &&
+          ((resType.getShape().front() == 1 &&
+            !resType.getScalableDims().front()) ||
+           (resType.getShape().back() == 1 &&
+            !resType.getScalableDims().back())) &&
+          transp == ArrayRef<int64_t>({1, 0})) {
+        rewriter.replaceOpWithNewOp<vector::ShapeCastOp>(op, resType, input);
+        return success();
+      }
     }
 
     if (inputType.isScalable())