Add NVIDIA and AMD targets in CI builds (#13)

.github/workflows/ci.yml

@@ -102,7 +102,9 @@ jobs:
           -DCMAKE_BUILD_TYPE=${{env.BUILD_TYPE}}
           -DCMAKE_CXX_COMPILER=clang++
           -DENABLE_GRAPHICS=ON
-          -DENABLE_SPIR=ON -DENABLE_CUDA=OFF -DENABLE_HIP=OFF
+          -DENABLE_SPIR=ON
+          -DENABLE_CUDA=ON -DCUDA_COMPUTE_CAPABILITY=80
+          -DENABLE_HIP=ON -DHIP_GFX_ARCH=gfx90a
           -DCMAKE_CXX_FLAGS='-Wall -Wextra -Wpedantic -Werror'
           -G Ninja
 

ci/Dockerfile

@@ -5,14 +5,27 @@ RUN apt update && apt -y install wget gpg git cmake ninja-build \
     gcc-12 libstdc++-12-dev libsdl2-dev \
     && apt clean
 
-# Install Intel oneAPI repositories
+# Install nvcc (dependency for compiling for a CUDA target)
+ARG CUDA_VERSION=12-4
+RUN wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.0-1_all.deb \
+    && dpkg -i cuda-keyring_1.0-1_all.deb && rm cuda-keyring_1.0-1_all.deb \
+    && apt update && apt -y install cuda-nvcc-${CUDA_VERSION} && apt clean
+
+# Install ROCm device libs (dependency for compiling for a HIP target)
+ARG ROCM_VERSION=5.4.3
+RUN wget https://repo.radeon.com/rocm/rocm.gpg.key -O - \
+    | gpg --dearmor | tee /etc/apt/keyrings/rocm.gpg > /dev/null \
+    && echo "deb [arch=amd64 signed-by=/etc/apt/keyrings/rocm.gpg] https://repo.radeon.com/rocm/apt/${ROCM_VERSION} jammy main" \
+    | tee /etc/apt/sources.list.d/rocm.list \
+    && apt update && apt -y install rocm-device-libs && apt clean
+
+# Install DPC++ and remove parts we don't need to reduce the container size
+ARG ONEAPI_VERSION=2024.1
 RUN wget -O- https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB \
     | gpg --dearmor | tee /usr/share/keyrings/oneapi-archive-keyring.gpg > /dev/null \
     && echo "deb [signed-by=/usr/share/keyrings/oneapi-archive-keyring.gpg] https://apt.repos.intel.com/oneapi all main" \
-    | tee /etc/apt/sources.list.d/oneAPI.list
-
-# Install DPC++ and remove parts we don't need to reduce the container size
-RUN apt update && apt -y install intel-oneapi-compiler-dpcpp-cpp-2024.0 \
+    | tee /etc/apt/sources.list.d/oneAPI.list \
+    && apt update && apt -y install intel-oneapi-compiler-dpcpp-cpp-${ONEAPI_VERSION} \
     && apt clean \
     && cd /opt/intel/oneapi \
     && rm -rf conda_channel debugger dev-utilities dpl compiler/latest/linux/lib/oclfpga
@@ -24,6 +37,7 @@ ENV PATH=${CMPLR_ROOT}/bin:${CMPLR_ROOT}/bin/compiler:${PATH}
 ENV CPATH=${CMPLR_ROOT}/include:${CPATH}
 ENV LIBRARY_PATH=${CMPLR_ROOT}/lib:${LIBRARY_PATH}
 ENV LD_LIBRARY_PATH=${CMPLR_ROOT}/lib:${LD_LIBRARY_PATH}
+ENV HIP_DEVICE_LIB_PATH=/usr/lib/x86_64-linux-gnu/amdgcn/bitcode
 
 # Set up entry point
 ENTRYPOINT []

cmake/ConfigureSYCL.cmake

@@ -42,12 +42,18 @@ set(SYCL_TARGETS "")
 if(${ENABLE_CUDA})
     string(JOIN "," SYCL_TARGETS "${SYCL_TARGETS}" "nvptx64-nvidia-cuda")
     set(DEFAULT_CUDA_COMPUTE_CAPABILITY "50")
-    execute_process(
-        COMMAND bash -c "which nvidia-smi >/dev/null && nvidia-smi --query-gpu=compute_cap --format=csv,noheader | head -n 1 | tr -d '.'"
-        OUTPUT_VARIABLE CUDA_COMPUTE_CAPABILITY
-        OUTPUT_STRIP_TRAILING_WHITESPACE)
+    set(CUDA_COMPUTE_CAPABILITY "" CACHE BOOL
+        "CUDA architecture (compute capability), e.g. sm_80. Default value is auto-configured using nvidia-smi.")
+    # Auto-configure if not specified by user
     if ("${CUDA_COMPUTE_CAPABILITY}" STREQUAL "")
-        message(WARNING "Failed to autoconfigure CUDA Compute Capability using nvidia-smi. Will default to sm_${DEFAULT_CUDA_COMPUTE_CAPABILITY}")
+        execute_process(
+            COMMAND bash -c "which nvidia-smi >/dev/null && nvidia-smi --query-gpu=compute_cap --format=csv,noheader | head -n 1 | tr -d '.'"
+            OUTPUT_VARIABLE CUDA_COMPUTE_CAPABILITY
+            OUTPUT_STRIP_TRAILING_WHITESPACE)
+    endif()
+    # Warn if not specified and failed to auto-configure
+    if ("${CUDA_COMPUTE_CAPABILITY}" STREQUAL "")
+        message(WARNING "Failed to autoconfigure CUDA_COMPUTE_CAPABILITY using nvidia-smi. Will default to sm_${DEFAULT_CUDA_COMPUTE_CAPABILITY}")
         set(CUDA_COMPUTE_CAPABILITY ${DEFAULT_CUDA_COMPUTE_CAPABILITY} CACHE STRING "CUDA Compute Capability")
     else()
         message(STATUS "Enabled SYCL target CUDA with Compute Capability sm_${CUDA_COMPUTE_CAPABILITY}")
@@ -60,12 +66,18 @@ endif()
 if(${ENABLE_HIP})
     string(JOIN "," SYCL_TARGETS "${SYCL_TARGETS}" "amdgcn-amd-amdhsa")
     set(DEFAULT_HIP_GFX_ARCH "gfx906")
-    execute_process(
-        COMMAND bash -c "which rocminfo >/dev/null && rocminfo | grep -o 'gfx[0-9]*' | head -n 1"
-        OUTPUT_VARIABLE HIP_GFX_ARCH
-        OUTPUT_STRIP_TRAILING_WHITESPACE)
+    set(HIP_GFX_ARCH "" CACHE BOOL
+        "HIP architecture tag, e.g. gfx90a. Default value is auto-configured using rocminfo.")
+    # Auto-configure if not specified by user
+    if ("${CUDA_COMPUTE_CAPABILITY}" STREQUAL "")
+        execute_process(
+            COMMAND bash -c "which rocminfo >/dev/null && rocminfo | grep -o 'gfx[0-9]*' | head -n 1"
+            OUTPUT_VARIABLE HIP_GFX_ARCH
+            OUTPUT_STRIP_TRAILING_WHITESPACE)
+    endif()
+    # Warn if not specified and failed to auto-configure
     if ("${HIP_GFX_ARCH}" STREQUAL "")
-        message(WARNING "Failed to autoconfigure HIP gfx arch using rocminfo. Will default to ${DEFAULT_HIP_GFX_ARCH}")
+        message(WARNING "Failed to autoconfigure HIP_GFX_ARCH using rocminfo. Will default to ${DEFAULT_HIP_GFX_ARCH}")
         set(HIP_GFX_ARCH ${DEFAULT_HIP_GFX_ARCH} CACHE STRING "HIP gfx arch")
     else()
         message(STATUS "Enabled SYCL target HIP with gfx arch ${HIP_GFX_ARCH}")