PR #21396: [ROCm] Fix build break due to XNNPACK update and add NCCL_…

…MAX_NCHANNELS to multi gpu tests Imported from GitHub PR openxla/xla#21396 `NCCL_MAX_NCHANNELS=1` is necessary for collective ops tests to pass in CI. As for XNNPACK problem, similar fix has already been done for single-gpu tests -> openxla/xla#20975 Copybara import of the project: -- 631fa6b7fc859c083e0735d2ce47167cbf57c174 by Milica Makevic <[email protected]>: Fix build break due to XNNPACK update -- d226a07701ddd88d45e7b27406d2915d032832a4 by Milica Makevic <[email protected]>: Add NCCL_MAX_NCHANNELS env variable to multi gpu tests -- b826eee7ab0e2a1f4d062871465ab4dac48ead37 by Milica Makevic <[email protected]>: Split bazel command arguments in multiple lines Merging this change closes #21396 PiperOrigin-RevId: 715779335
google · Jan 15, 2025 · 32732fa · 32732fa
1 parent 61d03c8
commit 32732fa
Showing 1 changed file with 2 additions and 0 deletions.
diff --git a/.bazelrc b/.bazelrc
@@ -294,6 +294,8 @@ build:tpu --define=enable_mlir_bridge=true
 build:rocm --crosstool_top=@local_config_rocm//crosstool:toolchain
 build:rocm --define=using_rocm_hipcc=true
 build:rocm --define=tensorflow_mkldnn_contraction_kernel=0
+build:rocm --define=xnn_enable_avxvnniint8=false
+build:rocm --define=xnn_enable_avx512fp16=false
 build:rocm --repo_env TF_NEED_ROCM=1
 
 build:rocm_clang_official --config=rocm