[Mlir-commits] [mlir] [MLIR][Linalg] pack, unpack to take memref inputs (PR #129036)

[Han-Chung Wang]

================
@@ -706,3 +706,21 @@ func.func @conv2d_channel_first_q_promote(%img: tensor<100x3x224x224xi8>, %filt:
 // CHECK-LABEL: func @conv2d_channel_first_q_promote(
 // CHECK:   %[[arg0:[a-zA-z0-9]*]]: tensor<100x3x224x224xi8>, %[[arg1:[a-zA-z0-9]*]]: tensor<64x3x5x5xi8>, %[[arg2:[a-zA-z0-9]*]]: i8, %[[arg3:[a-zA-z0-9]*]]: i8)
 // CHECK:         linalg.conv_2d_nchw_fchw_q {dilations = dense<1> : tensor<2xi64>, strides = dense<1> : tensor<2xi64>} ins(%[[arg0]], %[[arg1]], %[[arg2]], %[[arg3]] : tensor<100x3x224x224xi8>, tensor<64x3x5x5xi8>, i8, i8) outs(%{{.*}} : tensor<100x64x220x220xi32>) -> tensor<100x64x220x220xi32>
+
+// -----
+// Test that we can lower all the way to LLVM without crashing, don't check results here.
----------------
hanhanW wrote:

If the comment is copied from other places, maybe they are obsolete. The comment was added with lowering linalg to llvm. I think the modern convention is that `roundtrips/ops` tests are verifying printers and parsers. The lowering should be moved to other lit test files. Thus we don't need such comment.

https://github.com/llvm/llvm-project/blob/b3f01a6aa45b00240cec1c64286b85d7ba87e2af/mlir/test/Dialect/Linalg/roundtrip.mlir#L6-L25

btw, we don't need any return in the memref tests.

https://github.com/llvm/llvm-project/pull/129036