scott/rocm-faster-whisper
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Fix all ROCm 6.2 hipBLAS API incompatibilities

Browse Source 
ROCm 6.2 hipblasGemmEx uses hipblasDatatype_t for the compute type arg,
not hipblasComputeType_t (that came in ROCm 7). Map compute type and
HIPBLAS_COMPUTE_* constants down to the old hipblasDatatype_t/HIPBLAS_R_*
equivalents. Also fix HIPBLAS_R_16B (not 16BF) and add rocrand-dev.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
scott
2026-04-02 21:46:56 -04:00
parent f0504b0ff2
commit f5a19920d7
1 changed files with 8 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
Dockerfile
View File

@@ -12,7 +12,7 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
git cmake cmake-curses-gui \ git cmake cmake-curses-gui \
libopenblas-dev wget \ libopenblas-dev wget \
python3-dev python3-pip \ python3-dev python3-pip \
hipblas-dev miopen-hip-dev rocthrust-dev hiprand-dev hipcub-dev \ hipblas-dev miopen-hip-dev rocthrust-dev hiprand-dev hipcub-dev rocrand-dev \
&& rm -rf /var/lib/apt/lists/* && rm -rf /var/lib/apt/lists/*
# Build oneDNN as a static library (CPU fallback / conv ops) # Build oneDNN as a static library (CPU fallback / conv ops)
@@ -42,11 +42,17 @@ RUN git clone --branch v4.0.0 --depth 1 \
git apply /tmp/ct2_4.0.0_rocm.patch && \ git apply /tmp/ct2_4.0.0_rocm.patch && \
sed -i \ sed -i \
-e 's/hipDataType/hipblasDatatype_t/g' \ -e 's/hipDataType/hipblasDatatype_t/g' \
-e 's/hipblasComputeType_t/hipblasDatatype_t/g' \
-e 's/HIP_R_16F/HIPBLAS_R_16F/g' \ -e 's/HIP_R_16F/HIPBLAS_R_16F/g' \
-e 's/HIP_R_32F/HIPBLAS_R_32F/g' \ -e 's/HIP_R_32F/HIPBLAS_R_32F/g' \
-e 's/HIP_R_16BF/HIPBLAS_R_16BF/g' \ -e 's/HIP_R_16BF/HIPBLAS_R_16B/g' \
-e 's/HIP_R_32I/HIPBLAS_R_32I/g' \ -e 's/HIP_R_32I/HIPBLAS_R_32I/g' \
-e 's/HIP_R_8I/HIPBLAS_R_8I/g' \ -e 's/HIP_R_8I/HIPBLAS_R_8I/g' \
-e 's/HIPBLAS_COMPUTE_16F/HIPBLAS_R_16F/g' \
-e 's/HIPBLAS_COMPUTE_32F_FAST_16BF/HIPBLAS_R_32F/g' \
-e 's/HIPBLAS_COMPUTE_32F_FAST_16F/HIPBLAS_R_32F/g' \
-e 's/HIPBLAS_COMPUTE_32F/HIPBLAS_R_32F/g' \
-e 's/HIPBLAS_COMPUTE_32I/HIPBLAS_R_32I/g' \
src/cuda2hip_macros.hpp && \ src/cuda2hip_macros.hpp && \
mkdir build && cd build && \ mkdir build && cd build && \
cmake -DCMAKE_INSTALL_PREFIX=${CTRANSLATE2_ROOT} \ cmake -DCMAKE_INSTALL_PREFIX=${CTRANSLATE2_ROOT} \