Skip to content
This repository was archived by the owner on Mar 28, 2023. It is now read-only.

[SYCL] Add test for reductions using the resource pool #873

Closed

Conversation

steffenlarsen
Copy link

The current implementation of SYCL reductions may require some additional device-memory resources. To reduce the overhead of allocating memory every time a reduction runs, a resource pool is employed. These changes add a test to ensure that the resource pool prevents repeated allocations for reductions needing these resources.

The current implementation of SYCL reductions may require some
additional device-memory resources. To reduce the overhead of allocating
memory every time a reduction runs, a resource pool is employed. These
changes add a test to ensure that the resource pool prevents repeated
allocations for reductions needing these resources.

Signed-off-by: Steffen Larsen <[email protected]>
@steffenlarsen steffenlarsen requested a review from a team as a code owner February 25, 2022 16:46
@steffenlarsen
Copy link
Author

Signed-off-by: Steffen Larsen <[email protected]>
@vladimirlaz
Copy link

/verify with intel/llvm#5662

@@ -0,0 +1,35 @@
// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_60

Choose a reason for hiding this comment

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

the test is built for CUDA BE
Is it expected?

Suggested change
// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_60
// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out

Copy link
Author

Choose a reason for hiding this comment

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

These options only have an effect if the sycl target is the CUDA backend and are ignored otherwise. All it does is instruct the generation of NVPTX to use SM60, which I believe is required here to enable an extended set of atomic operations used by reductions.

myler pushed a commit to myler/llvm-test-suite that referenced this pull request Jun 17, 2022
Fix CMPLRTST-16222: disable hier_par_wgscope_O0 on discrete GPUs because of a hang issue
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants