[Mlir-commits] [mlir] [mlir][Vector] Improve support for vector.extract(broadcast) (PR #116234)

[Andrzej Warzyński]

================
@@ -652,24 +652,44 @@ func.func @fold_extract_transpose(
 
 // -----
 
-// CHECK-LABEL: fold_extract_broadcast
+// CHECK-LABEL: fold_extract_broadcast_same_type
 //  CHECK-SAME:   %[[A:.*]]: f32
 //       CHECK:   return %[[A]] : f32
-func.func @fold_extract_broadcast(%a : f32) -> f32 {
+func.func @fold_extract_broadcast_same_type(%a : f32, 
+                                            %idx0 : index, 
+                                            %idx1 : index) -> f32 {
   %b = vector.broadcast %a : f32 to vector<1x2x4xf32>
-  %r = vector.extract %b[0, 1, 2] : f32 from vector<1x2x4xf32>
+  // The indices don't batter for this folder, so we use mixed indices.
+  %r = vector.extract %b[%idx0, %idx1, 2] : f32 from vector<1x2x4xf32>
   return %r : f32
 }
 
 // -----
 
-// CHECK-LABEL: fold_extract_broadcast_0dvec
+// CHECK-LABEL: fold_extract_broadcast_same_type_vec
+//  CHECK-SAME:   %[[A:.*]]: vector<4xf32>
+//       CHECK:   return %[[A]] : vector<4xf32>
+func.func @fold_extract_broadcast_same_type_vec(%a : vector<4xf32>, 
+                                                %idx0 : index) 
+                                                -> vector<4xf32> {
+  %b = vector.broadcast %a : vector<4xf32> to vector<1x2x4xf32>
+  // The indices don't batter for this folder, so we use mixed indices.
+  %r = vector.extract %b[0, %idx0] : vector<4xf32> from vector<1x2x4xf32>
+  return %r : vector<4xf32>
+}
+
+// -----
+
+// CHECK-LABEL: fold_extract_broadcast_0dvec_and_scalar
 //  CHECK-SAME:   %[[A:.*]]: vector<f32>
 //       CHECK:   %[[B:.+]] = vector.extractelement %[[A]][] : vector<f32>
 //       CHECK:   return %[[B]] : f32
-func.func @fold_extract_broadcast_0dvec(%a : vector<f32>) -> f32 {
+func.func @fold_extract_broadcast_0dvec_and_scalar(%a : vector<f32>, 
----------------
banach-space wrote:

[nit] Since you are updating the function names, how about `brodcast_0dvec` -> `broadcast_0d_extract_scalar`? That would be more descriptive and consistent with what Diego has been adding here: https://github.com/llvm/llvm-project/pull/115824/files

https://github.com/llvm/llvm-project/pull/116234