Thank you, Alex. There are also a couple of redundant bitcasts in LHLO memref cast ops -> llvm conversion. That reminds me to clean them up as well.

Address the comments.

Address the comments and fix mlir-cpu-runner tests in cmake.

lint

In D85192#2196724, @ftynse wrote:

It looks like it would require using TestDialect and writing a test pass from test dialect to llvm. Could there be a simpler way of testing this?

How did you stumble upon this? I think this function is called when an std.call is converted, so it should be sufficient to test that an std.call to a function that takes a mix of memrefs and scalars as arguments.

In D85192#2195280, @mehdi_amini wrote:

Can you add a test please?

Lint

Test

Address the comments.

In D84918#2185567, @mehdi_amini wrote:

If you are just experimenting with it, you can use one of the PRIVATE_EXPERIMENTAL one, otherwise this is intended to be "cheap" to add.

Address the comments.

Fix linter.

Address the comments.

Split getSizes() into getMemRefDescriptorsizes and getCumulativeSizeInBytes.

Addressed the comment

x index -> x i32

Address the comments in https://llvm.discourse.group/t/standardops-add-reshapememrefcastop-to-standard/1350/2

Remove verification for unranked -> unranked memrefs.

Addressed the comments.

Thank you!

In D83068#2129809, @mehdi_amini wrote:

Can you make sure there is a thread on discourse before adding ops to the standard dialect?