Bump CI to TheRock 7.14

CHANGELOG.md

@@ -8,6 +8,7 @@ Full documentation for MIGraphX is available at
 ### Added
 
 * Added `ArrayFeatureExtractor` ONNX operator support (#4742).
+* Added support for building against ROCm 7.13 and newer using TheRock (#4952)
 * Added YOLO26 object detection example notebook.
 * Added `auto_pad` attribute support for the ONNX `ConvTranspose` operator, supporting `SAME_UPPER`, `SAME_LOWER`, and `VALID` padding modes for static shapes (#4638).
 * Added a dedicated logger for MIGraphX.

CMakeLists.txt

@@ -195,6 +195,7 @@ endif()
 rocm_enable_clang_tidy(
     CHECKS
         bugprone-*
+        -bugprone-std-namespace-modification
         cert-*
         clang-analyzer-*
         clang-diagnostic-*
@@ -212,11 +213,16 @@ rocm_enable_clang_tidy(
         -bugprone-crtp-constructor-accessibility
         -bugprone-derived-method-shadowing-base-method
         -bugprone-easily-swappable-parameters
+        -bugprone-exception-escape
+        -bugprone-fold-init-type
         -bugprone-implicit-widening-of-multiplication-result
         -bugprone-macro-parentheses
         -bugprone-nondeterministic-pointer-iteration-order
+        -bugprone-random-generator-seed
         -bugprone-return-const-ref-from-parameter
+        -bugprone-signed-bitwise
         -bugprone-signed-char-misuse
+        -bugprone-std-exception-baseclass
         -bugprone-unchecked-optional-access
         -bugprone-unintended-char-ostream-output
         # Disable the aliased reserved identifiers
@@ -271,21 +277,30 @@ rocm_enable_clang_tidy(
         -misc-const-correctness
         -misc-include-cleaner
         -misc-macro-parentheses
+        -misc-multiple-inheritance
         -misc-no-recursion
         -misc-override-with-different-visibility
+        -misc-use-internal-linkage
+        -modernize-avoid-c-style-cast
         -modernize-concat-nested-namespaces
         -modernize-pass-by-value
         -modernize-type-traits
         -modernize-use-default-member-init
         -modernize-use-nodiscard
         -modernize-use-override
         -modernize-use-scoped-lock
+        -modernize-use-string-view
+        -modernize-use-structured-binding
         -modernize-use-trailing-return-type
         -modernize-use-transparent-functors
         -performance-avoid-endl
+        -performance-faster-string-find
+        -performance-prefer-single-char-overloads
         -performance-type-promotion-in-math-fn
         -performance-enum-size
+        -performance-use-std-move
         -readability-braces-around-statements
+        -readability-inconsistent-ifelse-braces
         -readability-container-size-empty
         -readability-avoid-nested-conditional-operator
         -readability-convert-member-functions-to-static
@@ -294,10 +309,13 @@ rocm_enable_clang_tidy(
         -readability-identifier-length
         -readability-math-missing-parentheses
         -readability-named-parameter
+        -readability-redundant-lambda-parameter-list
         -readability-redundant-member-init
         -readability-redundant-parentheses
+        -readability-redundant-qualified-alias
         -readability-redundant-string-init
         -readability-suspicious-call-argument
+        -readability-trailing-comma
         -readability-uppercase-literal-suffix
         -readability-use-concise-preprocessor-directives
         -*-avoid-c-arrays

Dockerfile

@@ -1,27 +1,57 @@
-FROM ubuntu:22.04
+# MIGraphX build environment with TheRock (amdrocm-*) deb packages.
+#
+# Build args:
+#   ROCM_VERSION      ROCm release version for versioned package names (e.g. 7.13)
+#   GPU_ARCH          GPU architecture family (e.g. gfx120x, gfx94x)
+#   USE_WHL           Set to a non-empty value to install ROCm from Python
+#                     wheels (pip) instead of system packages
+#   INDEX_URL         pip --index-url used when installing ROCm from wheels
+#                     (only effective together with USE_WHL)
+#
+# Build:
+#   docker build --build-arg GPU_ARCH=<gpu_arch> \
+#       --build-arg ROCM_VERSION=<rocm_version> \
+#       -t migraphx-therock .
+#
+# Run:
+#   docker run -it --device=/dev/kfd --device=/dev/dri --group-add video \
+#       -v $(pwd):/code/AMDMIGraphX migraphx-therock
+#
+# Build MIGraphX inside the container:
+#   cd /code/AMDMIGraphX
+#   rbuild build -d depend -B build
+#
+
+FROM ubuntu:24.04
 
 ARG PREFIX=/usr/local
+# ROCm release version (used in versioned package names, e.g. amdrocm-developer-tools7.13)
+ARG ROCM_VERSION="7.14"
+# GPU architecture family (e.g. gfx942, gfx120x); leave empty for arch-independent packages
+ARG GPU_ARCH=""
+# Install location for the prebuilt MIGraphX dependencies.
+ARG PREFIX=/usr/local
+# Install the MIGraphX build prerequisites (system packages + ROCm components).
+# Set USE_WHL to any non-empty value to install ROCm from Python wheels instead
+# of system packages (passes --whl to the prereqs script).
+ARG USE_WHL=""
+# pip index URL for the wheel-based ROCm install (only used when USE_WHL is set).
+ARG INDEX_URL="https://repo.amd.com/rocm/whl-multi-arch/"
 
 # Support multiarch
 RUN dpkg --add-architecture i386
 
 # Install rocm key
 RUN apt-get update && apt-get install -y software-properties-common gnupg2 --no-install-recommends curl && \
     mkdir -p /etc/apt/keyrings && \
-    curl -fsSL https://repo.radeon.com/rocm/rocm.gpg.key | gpg --dearmor -o /etc/apt/keyrings/rocm.gpg
+    curl -fsSL https://repo.amd.com/rocm/packages/gpg/rocm.gpg | gpg --dearmor -o /etc/apt/keyrings/amdrocm.gpg
 
 # Add rocm repository
-RUN sh -c 'echo deb [arch=amd64 signed-by=/etc/apt/keyrings/rocm.gpg] https://repo.radeon.com/rocm/apt/7.2.3/ jammy main > /etc/apt/sources.list.d/rocm.list'
-
-# From docs.amd.com for installing rocm. Needed to install properly
-RUN sh -c "echo 'Package: *\nPin: release o=repo.radeon.com\nPin-priority: 600' > /etc/apt/preferences.d/rocm-pin-600"
-
-# rocgdb doesn't work on 22.04, workaround by installing the older python packages that are in 20.04
-RUN add-apt-repository -y ppa:deadsnakes/ppa
+RUN sh -c 'echo deb [arch=amd64 signed-by=/etc/apt/keyrings/amdrocm.gpg] https://repo.amd.com/rocm/packages-multi-arch/ubuntu2404 stable main > /etc/apt/sources.list.d/rocm.list'
 
 # Add LLVM repository for Clang 17 (ROCm 7.x ships with Clang 20 which has ODR false positives in ASAN)
 RUN curl -sL https://apt.llvm.org/llvm-snapshot.gpg.key | apt-key add - && \
-    add-apt-repository -y "deb http://apt.llvm.org/jammy/ llvm-toolchain-jammy-17 main"
+    add-apt-repository -y "deb http://apt.llvm.org/noble/ llvm-toolchain-noble-17 main"
 
 # Install dependencies
 RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y \
@@ -44,84 +74,69 @@ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y \
     python3-full \
     libpython3.8 \
     wget \
-    rocm-device-libs \
-    hip-dev \
     libnuma-dev \
-    miopen-hip \
     libomp-17-dev \
-    rocblas \
-    hipfft \
-    hipsolver \
-    rocthrust \
-    rocrand \
-    rocprofiler-sdk \
-    hipsparse \
-    rccl \
-    rocm-smi-lib \
-    rocminfo \
-    roctracer-dev \
-    hipcub  \
-    hipblas  \
-    hipify-clang \
-    hiprand-dev \
-    hipsparselt \
-    hsa-amd-aqlprofile \
-    half \
     libssl-dev \
     zlib1g-dev && \
     apt-get clean && \
     rm -rf /var/lib/apt/lists/*
 
-
-# Install pytorch
-RUN pip3 install https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2.3/torch-2.8.0%2Brocm7.2.3.lw.git2742f6d1-cp310-cp310-linux_x86_64.whl \
-                 https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2.3/torchvision-0.24.0%2Brocm7.2.3.gitb919bd0c-cp310-cp310-linux_x86_64.whl \
-                 https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2.3/triton-3.4.0%2Brocm7.2.3.git0cace8d2-cp310-cp310-linux_x86_64.whl
-
-# add this for roctracer dependencies
-RUN pip3 install CppHeaderParser
-
-# Workaround broken rocm packages
-RUN ln -s /opt/rocm-* /opt/rocm
-RUN echo "/opt/rocm/lib" > /etc/ld.so.conf.d/rocm.conf
-RUN echo "/opt/rocm/llvm/lib" > /etc/ld.so.conf.d/rocm-llvm.conf
-RUN ldconfig
-
-# ATT library
-RUN wget -O /opt/rocm/lib/librocprof-trace-decoder.so https://github.com/ROCm/rocprof-trace-decoder/raw/7e58204a955e5787b9b38087f3ad502f07ff78ef/releases/linux_glibc_2_28_x86_64/librocprof-trace-decoder.so
-
-# Workaround broken miopen cmake files
-RUN sed -i 's,;/usr/lib/x86_64-linux-gnu/librt.so,,g' /opt/rocm/lib/cmake/miopen/miopen-targets.cmake
-
-# Workaround for distributions running cmake < 3.25
-RUN sed -i -e 's/^block/if(COMMAND block)\nblock/g' -e 's/^endblock/endblock\(\)\nendif/g' /opt/rocm/lib/cmake/hipblaslt/hipblaslt-config.cmake
-
 RUN locale-gen en_US.UTF-8
 RUN update-locale LANG=en_US.UTF-8
 
 ENV LC_ALL=C.UTF-8
 ENV LANG=C.UTF-8
 
+ENV PIP_BREAK_SYSTEM_PACKAGES=1
+
 # Install dependencies
 ADD dev-requirements.txt /dev-requirements.txt
 ADD requirements.txt /requirements.txt
 ADD rbuild.ini /rbuild.ini
 
-# Location where onnx unit tests models are cached
-ENV ONNX_HOME=/.onnx
-RUN mkdir -p $ONNX_HOME/models && chmod 777 $ONNX_HOME/models
-
 COPY ./tools/install_prereqs.sh /
 COPY ./tools/requirements-py.txt /requirements-py.txt
-RUN /install_prereqs.sh /usr/local / && rm /install_prereqs.sh && rm /requirements-py.txt
+RUN ./install_prereqs.sh \
+        --rocm-version ${ROCM_VERSION} \
+        ${GPU_ARCH:+--gpu ${GPU_ARCH}} \
+        --index-url ${INDEX_URL} \
+        ${USE_WHL:+--whl}
+RUN rm /install_prereqs.sh && rm /*.txt
 RUN test -f /usr/local/hash || exit 1
 
+# Workaround broken rocm packages
+RUN echo "/opt/rocm/lib" > /etc/ld.so.conf.d/rocm.conf
+RUN echo "/opt/rocm/llvm/lib" > /etc/ld.so.conf.d/rocm-llvm.conf
+RUN ldconfig
+
+# Manually remove rocm-cmake, since it shouldnt be installed in the first place
+RUN rm -rf /opt/rocm/share/rocmcmakebuildtools
+
+# Install pytorch
+RUN pip3 install --index-url "${INDEX_URL}" \
+        "torch==2.11.0+rocm${ROCM_VERSION}.0" \
+        "torchvision==0.26.0+rocm${ROCM_VERSION}.0" \
+        "torchaudio==2.11.0+rocm${ROCM_VERSION}.0"
+
+# Location where onnx unit tests models are cached
+ENV ONNX_HOME=/.onnx
+RUN mkdir -p $ONNX_HOME/models && chmod 777 $ONNX_HOME/models
+
 # Install yapf
-RUN pip3 install yapf==0.28.0
+RUN pipx install --global yapf==0.28.0
+
+# Install clang format
+RUN pipx install --global clang-format==22.1.5
 
 # Install doc requirements
 ADD docs/sphinx/requirements.txt /doc-requirements.txt
-RUN pip3 install -r /doc-requirements.txt
+# pip rejects extras in a constraints file (the pip-compile output pins
+# pyjwt[crypto]), and extras carry no meaning in a constraint, so strip them to
+# pin the sphinx install.
+RUN sed 's/\[[^][]*\]//' /doc-requirements.txt > /doc-constraints.txt && \
+    pipx install --global sphinx --pip-args="-c /doc-constraints.txt" && \
+    rm /doc-constraints.txt
+RUN pipx inject --global sphinx -r /doc-requirements.txt
 
 # Install latest ccache version
 RUN cget -p $PREFIX install facebook/zstd@v1.4.5 -X subdir -DCMAKE_DIR=build/cmake
@@ -139,3 +154,4 @@ ENV UBSAN_OPTIONS=print_stacktrace=1
 # See: https://github.com/google/sanitizers/issues/1017
 ENV ASAN_OPTIONS=detect_stack_use_after_return=1:check_initialization_order=1:strict_init_order=1
 RUN ln -s /opt/rocm/llvm/bin/llvm-symbolizer /usr/bin/llvm-symbolizer
+