julian/amd-strix-halo-toolboxes
1
0
Fork 0
Code Issues Pull Requests 2 Actions Packages Projects Releases Wiki Activity

fix: Add temporary workaround for ROCm 7 performance regression by setting HIP_LLVM_FLAGS.

Browse Source
This commit is contained in:
Donato Capitella
2026-02-04 14:50:32 +00:00
parent 51aab9665d
commit 7ffa22d8de
2 changed files with 6 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
toolboxes/Dockerfile.rocm-7.2
+3
View File
@@ -36,6 +36,9 @@ ENV ROCM_PATH=/opt/rocm \
WORKDIR /opt/llama.cpp WORKDIR /opt/llama.cpp
RUN git clone --recursive https://github.com/ggerganov/llama.cpp.git . RUN git clone --recursive https://github.com/ggerganov/llama.cpp.git .
# Temporary workaround for ROCm 7 performance regression (https://github.com/ROCm/rocm-systems/issues/2865#issuecomment-3846679806)
ENV HIP_LLVM_FLAGS="-mllvm --amdgpu-unroll-threshold-local=600"
# build # build
RUN git clean -xdf \ RUN git clean -xdf \
&& git submodule update --recursive \ && git submodule update --recursive \
toolboxes/Dockerfile.rocm7-nightlies
+3
View File
@@ -57,6 +57,9 @@ RUN git clone --recursive https://github.com/ggerganov/llama.cpp.git . \
&& git clean -xdf \ && git clean -xdf \
&& git submodule update --recursive && git submodule update --recursive
# Temporary workaround for ROCm 7 performance regression (https://github.com/ROCm/rocm-systems/issues/2865#issuecomment-3846679806)
ENV HIP_LLVM_FLAGS="-mllvm --amdgpu-unroll-threshold-local=600"
RUN cmake -S . -B build \ RUN cmake -S . -B build \
-DGGML_HIP=ON \ -DGGML_HIP=ON \
-DAMDGPU_TARGETS=gfx1151 \ -DAMDGPU_TARGETS=gfx1151 \