Skip to content

[SYCL][UR] move UR headers to dedicated unified-runtime include directory#21399

Open
ldorau wants to merge 1 commit intointel:syclfrom
ldorau:move-ur-headers-to-dir
Open

[SYCL][UR] move UR headers to dedicated unified-runtime include directory#21399
ldorau wants to merge 1 commit intointel:syclfrom
ldorau:move-ur-headers-to-dir

Conversation

@ldorau
Copy link
Contributor

@ldorau ldorau commented Feb 27, 2026

This aligns how headers are installed by other components.
While UR was never meant to be used directly,
this change may break existing software that includes <ur_api.h>.

Ref: #21421

@ldorau
Copy link
Contributor Author

ldorau commented Feb 27, 2026

Please review @pbalcer

@ldorau
Copy link
Contributor Author

ldorau commented Mar 2, 2026

Please review @intel/bindless-images-reviewers @intel/dpcpp-nativecpu-reviewers @intel/llvm-reviewers-cuda @intel/llvm-reviewers-runtime @intel/sycl-graphs-reviewers @intel/unified-runtime-reviewers @intel/unified-runtime-reviewers-level-zero @intel/unified-runtime-reviewers-opencl

@ldorau ldorau force-pushed the move-ur-headers-to-dir branch from e6fb3a0 to 8ddbcdb Compare March 2, 2026 13:47
@mmichel11
Copy link
Contributor

The nightly benchmarks from intel/compute-benchmarks rely on including ur_api.h for direct UR benchmarking. Could we couple this PR with a change to the compute benchmarks and a benchmark commit id update, so we don't break the nightly run?

Probably need to ensure compatibility with older versions with a __has_include check.

@pbalcer
Copy link
Contributor

pbalcer commented Mar 2, 2026

The nightly benchmarks from intel/compute-benchmarks rely on including ur_api.h for direct UR benchmarking. Could we couple this PR with a change to the compute benchmarks and a benchmark commit id update, so we don't break the nightly run?

Probably need to ensure compatibility with older versions with a __has_include check.

That's a good shout, thanks! Yes, we should do that.

ldorau added a commit to ldorau/compute-benchmarks that referenced this pull request Mar 2, 2026
First look for <ur/ur_api.h>, then <ur_api.h>
for compatibility with older versions.

Ref: intel/llvm#21399

Signed-off-by: Lukasz Dorau <lukasz.dorau@intel.com>
@ldorau
Copy link
Contributor Author

ldorau commented Mar 2, 2026

The nightly benchmarks from intel/compute-benchmarks rely on including ur_api.h for direct UR benchmarking. Could we couple this PR with a change to the compute benchmarks and a benchmark commit id update, so we don't break the nightly run?
Probably need to ensure compatibility with older versions with a __has_include check.

That's a good shout, thanks! Yes, we should do that.

@mmichel11 Done: intel/compute-benchmarks#42

@ldorau ldorau requested a review from pbalcer March 2, 2026 16:02
ldorau added a commit to ldorau/compute-benchmarks that referenced this pull request Mar 2, 2026
First look for <ur/ur_api.h>, then <ur_api.h>
for compatibility with older versions.

Ref: intel/llvm#21399

Signed-off-by: Lukasz Dorau <lukasz.dorau@intel.com>
ldorau added a commit to ldorau/compute-benchmarks that referenced this pull request Mar 3, 2026
First look for <ur/ur_api.h>, then <ur_api.h>
for compatibility with older versions.

Ref: intel/llvm#21399

Signed-off-by: Lukasz Dorau <lukasz.dorau@intel.com>
@ldorau
Copy link
Contributor Author

ldorau commented Mar 3, 2026

The UR CUDA precommit conformance test exp_usm_context_memcpy/urUSMContextMemcpyExp fails because of #19688

@ldorau
Copy link
Contributor Author

ldorau commented Mar 3, 2026

Please review @intel/bindless-images-reviewers @intel/dpcpp-nativecpu-reviewers @intel/llvm-reviewers-cuda @intel/llvm-reviewers-runtime @intel/sycl-graphs-reviewers @intel/unified-runtime-reviewers @intel/unified-runtime-reviewers-level-zero @intel/unified-runtime-reviewers-opencl

@lukaszstolarczuk
Copy link
Contributor

The nightly benchmarks from intel/compute-benchmarks rely on including ur_api.h for direct UR benchmarking. Could we couple this PR with a change to the compute benchmarks and a benchmark commit id update, so we don't break the nightly run?
Probably need to ensure compatibility with older versions with a __has_include check.

That's a good shout, thanks! Yes, we should do that.

@mmichel11 Done: intel/compute-benchmarks#42

I will update the benchmarks' scripts in #21421 once your commit is available.

VPG-SWE-Github pushed a commit to intel/compute-benchmarks that referenced this pull request Mar 3, 2026
First look for <ur/ur_api.h>, then <ur_api.h>
for compatibility with older versions.

Ref: intel/llvm#21399

Signed-off-by: Lukasz Dorau <lukasz.dorau@intel.com>
@ldorau
Copy link
Contributor Author

ldorau commented Mar 3, 2026

The nightly benchmarks from intel/compute-benchmarks rely on including ur_api.h for direct UR benchmarking. Could we couple this PR with a change to the compute benchmarks and a benchmark commit id update, so we don't break the nightly run?
Probably need to ensure compatibility with older versions with a __has_include check.

That's a good shout, thanks! Yes, we should do that.

@mmichel11 Done: intel/compute-benchmarks#42

I will update the benchmarks' scripts in #21421 once your commit is available.

There already is: intel/compute-benchmarks@48504d2

@ldorau
Copy link
Contributor Author

ldorau commented Mar 3, 2026

The nightly benchmarks from intel/compute-benchmarks rely on including ur_api.h for direct UR benchmarking. Could we couple this PR with a change to the compute benchmarks and a benchmark commit id update, so we don't break the nightly run?
Probably need to ensure compatibility with older versions with a __has_include check.

That's a good shout, thanks! Yes, we should do that.

@mmichel11 Done: intel/compute-benchmarks#42

Regarding "a benchmark commit id update":
@mmichel11 Done: #21421

@ldorau
Copy link
Contributor Author

ldorau commented Mar 4, 2026

Please review @intel/bindless-images-reviewers @intel/dpcpp-nativecpu-reviewers @intel/llvm-reviewers-cuda @intel/llvm-reviewers-runtime @intel/sycl-graphs-reviewers @intel/unified-runtime-reviewers @intel/unified-runtime-reviewers-level-zero @intel/unified-runtime-reviewers-opencl

@bratpiorka
Copy link
Contributor

CUDA/HIP adapters LGTM

Copy link
Member

@adamfidel adamfidel left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Graph changes LGTM.

@ldorau ldorau force-pushed the move-ur-headers-to-dir branch from 8ddbcdb to b3e091f Compare March 4, 2026 17:06
@ldorau
Copy link
Contributor Author

ldorau commented Mar 4, 2026

Rebased because of conflicts.

@ldorau
Copy link
Contributor Author

ldorau commented Mar 4, 2026

Please review @intel/bindless-images-reviewers, @intel/llvm-reviewers-runtime and @intel/unified-runtime-reviewers-level-zero

@ldorau ldorau changed the title [SYCL][UR] move UR headers to dedicated ur include directory [SYCL][UR] move UR headers to dedicated unified-runtime include directory Mar 5, 2026
@ldorau ldorau force-pushed the move-ur-headers-to-dir branch from b3e091f to 81973e8 Compare March 5, 2026 13:50
@ldorau
Copy link
Contributor Author

ldorau commented Mar 5, 2026

Renamed include/ur to include/unified-runtime.

@ldorau
Copy link
Contributor Author

ldorau commented Mar 5, 2026

Please review @intel/bindless-images-reviewers, @intel/llvm-reviewers-runtime and @intel/unified-runtime-reviewers-level-zero

@ldorau ldorau force-pushed the move-ur-headers-to-dir branch from 81973e8 to 08a647c Compare March 5, 2026 14:32
…tory

This aligns how headers are installed by other components. While UR
was never meant to be used directly, this change may break existing
software that includes <ur_api.h>.

Signed-off-by: Lukasz Dorau <lukasz.dorau@intel.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

7 participants