saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-07-29 20:14:01 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

HSA_NO_SCRATCH_RECLAIM and RCCL_ALLTOALL_KERNEL_DISABLE are not needed for ROCm 4.1 (#7224)

Browse source 
Co-authored-by: Weixing Zhang <wezhan@microsoft.com>
This commit is contained in:
Weixing Zhang 2021-04-02 18:19:11 -07:00 committed by GitHub
parent ba5f056b09
commit 59b57d8322
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 4 additions and 4 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

8
orttraining/tools/amdgpu/Dockerfile.rocm4.1.pytorch
View file

@ -108,7 +108,7 @@ RUN mv $OPENMPI_DIR/bin/mpirun $OPENMPI_DIR/bin/mpirun.real && \
# install mpi4py (be sure to link existing /opt/openmpi-xxx)
RUN CC=mpicc MPICC=mpicc pip install mpi4py --no-binary mpi4py
ARG CACHE_DATA=2020-12-06
ARG CACHE_DATA=2021-04-02
# ONNX Runtime
WORKDIR $GITHUB_DIR
@ -186,12 +186,12 @@ RUN locale-gen en_US.UTF-8
# Workaround an issue in AMD compiler which generates poor GPU ISA
# when the type of kernel parameter is a structure and “pass-by-value” is used
ENV HSA_NO_SCRATCH_RECLAIM=1
# ENV HSA_NO_SCRATCH_RECLAIM=1
# Distributed training related environment variables
ENV HSA_FORCE_FINE_GRAIN_PCIE=1
ENV NCCL_DEBUG=INFO
ENV RCCL_ALLTOALL_KERNEL_DISABLE=1
# ENV NCCL_DEBUG=INFO
# ENV RCCL_ALLTOALL_KERNEL_DISABLE=1
# ENV NCCL_DEBUG_SUBSYS=INIT,COLL
WORKDIR ${WORKSPACE_DIR}/script