LGTM.

LGTM. This approach is used in TensorFlow XLA for ROCm, where XLA build may not have lld configured, and thus must rely on the external lld binary in a ROCm installation.

LGTM.

How about add a unit test to check the attribute can be overriden?

LGTM! I'd like to hear from @t-tye or @kzhuravl , whom to my knowledge are contributors to this documentation.

In D98447#2631114, @herhut wrote:

Thanks for cleaning this up! Looks good in general but I also cannot test this. It would be nice if we had a builder for this. @whchung do you have a setup that could test this continuously or at least periodically to make sure it does not break?

The patch has been verified on ROCm 4.0.

@csigg I would need your help advising me the proper way to configure it so I could test the patch. With the cmake command I use downstream I run into link errors aforementioned.

@csigg May I understand how you configure the build?

@csigg May I understand the goal for this changeset is to merge everything under mlir-cpu-runner?

Add MIR-level tests.

@arsenm Could you help me understand if my understanding is correct to fulfill your requirement of 'test globalisel'?

The patterns were found in some MLIR-based applications.

@aartbik The patch is no longer necessary.

Remove an empty line.

Update commit message.

Address review comments by improve logic.

Address code review comments.

LGTM. @aartbik / @nicolasvasilache would you mind give this patch another round of review?

This patch was landed in commit 603b974cf7103766a0e5e4a0320fedb7c4b570f9.

Fix typos in the commit log.

Revise HIP version detection logic to be more descriptive.

Add logic to detech AMD ISA version.

Fix build error due to incorrect lint.

Remove NOLINT.

Apply NOLINT.

@jerryyin I've revised the commit based on your comments. Could you give this patch another around of review? Thanks.

Address code review comments.

Address code review comments.

Revise the fix and added positive and negative tests.

Rebase.

std::string -> StringRef.

Fix build errors in a shared lib build.

@ftynse Thanks for the patch. Wondering would it be a better idea to amend the text at: https://mlir.llvm.org/getting_started/ so -DBUILD_SHARED_LIBS=ON is always turned on?

Thanks for the patch. On the application I'm working on, the longest vector type would be vector<?x4>, plus I use scalarize-masked-mem-intrin in my pipeline so the issue addressed in this patch was not encountered on my targets before.

Add CMake checks to ensure lld is enabled building mlir-rocm-runner.

Improve ROCm HIP library search logic.

Tame clang-tidy.

This patch depends on D80142.

Remove unused headers.

Remove obsolete comment.

Revise the patch addressing code review comments.