[Mlir-commits] [mlir] [mlir][vector] Promote in-bounds sub-vector transfer reads with Mem2Reg (PR #236320)

[llvmlistbot at llvm.org]

================
@@ -690,3 +689,495 @@ func.func @negative_masked_write(%v: vector<8xf32>, %w: vector<8xf32>, %m: vecto
   %r = vector.transfer_read %a[%c0], %pad {in_bounds = [true]} : memref<8xf32>, vector<8xf32>
   return %r : vector<8xf32>
 }
+
+// -----
+
+// CHECK-LABEL: func.func @partial_read_origin
+// CHECK-SAME: (%[[V:.*]]: vector<8xf32>,
+// CHECK-NOT: memref.alloca
+// CHECK-NOT: vector.transfer
+// CHECK: %[[R:.*]] = vector.extract_strided_slice %[[V]] offsets = [0], sizes = [4], strides = [1]
+// CHECK: return %[[R]] : vector<4xf32>
+func.func @partial_read_origin(%v: vector<8xf32>, %pad: f32) -> vector<4xf32> {
----------------
meinie0826 wrote:

ditto

https://github.com/llvm/llvm-project/pull/236320