unir/MastersThesis
0
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

ci update
Some checks failed
build_docker / build_cpu (linux/arm64) (pull_request) Has been cancelled
Details
build_docker / build_gpu (linux/amd64) (pull_request) Has been cancelled
Details
build_docker / build_gpu (linux/arm64) (pull_request) Has been cancelled
Details
build_docker / build_cpu (linux/amd64) (pull_request) Has been cancelled
Details
build_docker / essential (push) Successful in 0s
Details
build_docker / build_gpu (linux/amd64) (push) Has been cancelled
Details
build_docker / build_gpu (linux/arm64) (push) Has been cancelled
Details
build_docker / build_cpu (linux/amd64) (push) Has been cancelled
Details
build_docker / build_cpu (linux/arm64) (push) Has been cancelled
Details
build_docker / essential (pull_request) Successful in 1s
Details

Browse Source
This commit is contained in:
Sergio Jimenez Jimenez
2026-01-17 16:15:53 +01:00
parent 7ac0971153
commit a89ddd2d13
9 changed files with 525 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

63
src/paddle_ocr/Dockerfile.build-paddle
View File

@@ -91,6 +91,43 @@ RUN sed -i 's/-m64//g' cmake/flags.cmake && \
find . -name "*.cmake" -exec sed -i 's/-m64//g' {} \; 2>/dev/null || true && \
echo "Patched -m64 flag for ARM64 compatibility"
# Patch for ARM64: Install sse2neon to translate x86 SSE intrinsics to ARM NEON
# sse2neon provides drop-in replacements for x86 SIMD headers
RUN git clone --depth 1 https://github.com/DLTcollab/sse2neon.git /tmp/sse2neon && \
mkdir -p /usr/local/include/sse2neon && \
cp /tmp/sse2neon/sse2neon.h /usr/local/include/sse2neon/ && \
rm -rf /tmp/sse2neon && \
echo "Installed sse2neon for x86->ARM NEON translation"
# Create wrapper headers that use sse2neon for ARM64
RUN mkdir -p /usr/local/include/x86_stubs && \
echo "#ifndef __x86_64__" > /usr/local/include/x86_stubs/immintrin.h && \
echo "#include <sse2neon/sse2neon.h>" >> /usr/local/include/x86_stubs/immintrin.h && \
echo "#else" >> /usr/local/include/x86_stubs/immintrin.h && \
echo "#include_next <immintrin.h>" >> /usr/local/include/x86_stubs/immintrin.h && \
echo "#endif" >> /usr/local/include/x86_stubs/immintrin.h && \
echo "#ifndef __x86_64__" > /usr/local/include/x86_stubs/xmmintrin.h && \
echo "#include <sse2neon/sse2neon.h>" >> /usr/local/include/x86_stubs/xmmintrin.h && \
echo "#else" >> /usr/local/include/x86_stubs/xmmintrin.h && \
echo "#include_next <xmmintrin.h>" >> /usr/local/include/x86_stubs/xmmintrin.h && \
echo "#endif" >> /usr/local/include/x86_stubs/xmmintrin.h && \
echo "#ifndef __x86_64__" > /usr/local/include/x86_stubs/emmintrin.h && \
echo "#include <sse2neon/sse2neon.h>" >> /usr/local/include/x86_stubs/emmintrin.h && \
echo "#else" >> /usr/local/include/x86_stubs/emmintrin.h && \
echo "#include_next <emmintrin.h>" >> /usr/local/include/x86_stubs/emmintrin.h && \
echo "#endif" >> /usr/local/include/x86_stubs/emmintrin.h && \
echo "#ifndef __x86_64__" > /usr/local/include/x86_stubs/pmmintrin.h && \
echo "#include <sse2neon/sse2neon.h>" >> /usr/local/include/x86_stubs/pmmintrin.h && \
echo "#else" >> /usr/local/include/x86_stubs/pmmintrin.h && \
echo "#include_next <pmmintrin.h>" >> /usr/local/include/x86_stubs/pmmintrin.h && \
echo "#endif" >> /usr/local/include/x86_stubs/pmmintrin.h && \
echo "#ifndef __x86_64__" > /usr/local/include/x86_stubs/smmintrin.h && \
echo "#include <sse2neon/sse2neon.h>" >> /usr/local/include/x86_stubs/smmintrin.h && \
echo "#else" >> /usr/local/include/x86_stubs/smmintrin.h && \
echo "#include_next <smmintrin.h>" >> /usr/local/include/x86_stubs/smmintrin.h && \
echo "#endif" >> /usr/local/include/x86_stubs/smmintrin.h && \
echo "Created x86 intrinsic wrapper headers for ARM64 using sse2neon"
# Install additional Python requirements for building
RUN pip install -r python/requirements.txt || true
@@ -99,6 +136,7 @@ RUN mkdir -p build
WORKDIR /build/Paddle/build
# Configure CMake for ARM64 + CUDA build
# Note: -Wno-class-memaccess fixes Eigen NEON warning on ARM64
RUN echo "Building for CUDA architecture: sm_${CUDA_ARCH}" && \
cmake .. \
-GNinja \
@@ -118,6 +156,7 @@ RUN echo "Building for CUDA architecture: sm_${CUDA_ARCH}" && \
-DCMAKE_CUDA_ARCHITECTURES="${CUDA_ARCH}" \
-DCMAKE_CXX_COMPILER_LAUNCHER=ccache \
-DCMAKE_C_COMPILER_LAUNCHER=ccache \
-DCMAKE_CXX_FLAGS="-Wno-class-memaccess -Wno-error=class-memaccess -I/usr/local/include/x86_stubs" \
-DCMAKE_EXPORT_COMPILE_COMMANDS=ON
# Build external dependencies first (cacheable layer)
@@ -142,14 +181,28 @@ RUN ninja paddle_python || true
# Create output directory
RUN mkdir -p /output
# Build wheel package
# Build wheel package - try multiple methods since PaddlePaddle build structure varies
WORKDIR /build/Paddle
RUN cd python && python setup.py bdist_wheel || pip wheel . -w dist/
RUN echo "=== Looking for wheel build method ===" && \
ls -la python/ 2>/dev/null && \
ls -la build/python/ 2>/dev/null && \
if [ -f build/python/setup.py ]; then \
echo "Using build/python/setup.py" && \
cd build/python && python setup.py bdist_wheel; \
elif [ -f python/setup.py ]; then \
echo "Using python/setup.py" && \
cd python && python setup.py bdist_wheel; \
else \
echo "Looking for existing wheel..." && \
find /build -name "paddlepaddle*.whl" -type f 2>/dev/null; \
fi
# Copy wheel to output
RUN cp python/dist/*.whl /output/ 2>/dev/null || \
cp build/python/dist/*.whl /output/ 2>/dev/null || \
find /build -name "paddlepaddle*.whl" -exec cp {} /output/ \;
RUN find /build -name "paddlepaddle*.whl" -type f -exec cp {} /output/ \; && \
ls -la /output/ && \
if [ ! "$(ls -A /output/*.whl 2>/dev/null)" ]; then \
echo "ERROR: No wheel found!" && exit 1; \
fi
# List what was built
RUN ls -la /output/ && \