PR #22593: [XLA:GPU] Fix triton dot op on sm120 (RTX50xx)
Imported from GitHub PR https://github.com/openxla/xla/pull/22593 Triton doesn't currently support sm120 GPUs - adding a patch to fix that, the upstream support should be available soon. Converting "12.0" arch into "10.0" is not correct, as they're not compatible - removing this. Copybara import of the project: -- 5b5752dc3b3ae3611ffeca0f55a87f130ff1e8bb by Sergey Kozub <skozub@nvidia.com>: [XLA:GPU] Fix triton dot op on sm120 (RTX50xx) Merging this change closes #22593 FUTURE_COPYBARA_INTEGRATE_REVIEW=https://github.com/openxla/xla/pull/22593 from openxla:skozub/sm120_dot 5b5752dc3b3ae3611ffeca0f55a87f130ff1e8bb PiperOrigin-RevId: 725968626
Showing
- third_party/triton/temporary/series.bzl 1 addition, 0 deletionsthird_party/triton/temporary/series.bzl
- third_party/triton/temporary/sm120.patch 13 additions, 0 deletionsthird_party/triton/temporary/sm120.patch
- third_party/xla/third_party/triton/temporary/series.bzl 1 addition, 0 deletionsthird_party/xla/third_party/triton/temporary/series.bzl
- third_party/xla/third_party/triton/temporary/sm120.patch 13 additions, 0 deletionsthird_party/xla/third_party/triton/temporary/sm120.patch
- third_party/xla/xla/backends/gpu/codegen/triton/fusion_emitter.cc 0 additions, 5 deletions...rty/xla/xla/backends/gpu/codegen/triton/fusion_emitter.cc
Please register or sign in to comment