try compiling support for more Arm CPUs

Author: jameslamb

.ci/ci-images/manylinux_2_28_aarch64/Dockerfile

@@ -43,15 +43,27 @@ git clone \
    --branch v7.1 \
    https://github.com/pocl/pocl.git
 
+# compile support for a bunch of different aarch64 CPUs
+#
+# Copied from https://github.com/conda-forge/pocl-feedstock/blob/64c2c02d96960160e20cabf1b1392a1752146f29/recipe/build.sh#L58-L62
+#
+# This support is mostly added for CI...testing device="gpu" codepaths on host CPUs, which tend to be cheaper and more
+# available than CI runners with GPUs. But using the host CPU as an OpenCL device can be useful for research as well.
+#
+AARCH64_CPUS="generic;cortex-a35;cortex-a53;cortex-a55;cortex-a57;cortex-a65;cortex-a72;cortex-a73;cortex-a75;cortex-a76"
+AARCH64_CPUS="${AARCH64_CPUS};cyclone;exynos-m3;exynos-m4;exynos-m5;falkor;kryo;neoverse-e1;neoverse-n1;saphira"
+AARCH64_CPUS="${AARCH64_CPUS};thunderx;thunderx2t99;thunderxt81;thunderxt83;thunderxt88;tsv110"
+
 # explanations for some flags:
 #
 #   * -DCMAKE_{C,CXX}_COMPILER: DEVTOOLSET_ROOTPATH is where manylinux puts the gcc toolset
 #   * -DENABLE_ICD=ON: select appropriate OpenCL platform at runtime (https://github.com/pocl/pocl/blob/013d2f19f4e8f2e0fd9aedcb70117d6dcc737aa9/doc/sphinx/source/using.rst#installable-client-driver-icd)
-#   * -DLLC_HOST_CPU="cortex-a53": passed to clang's -march/-mcpu flag, which is like "support a bunch of Arm CPUs".
+#   * -DLLC_HOST_CPU="cortex-a35": passed to clang's -march/-mcpu flag, which is like "support a bunch of Arm CPUs".
 #
 cmake \
    -B pocl/build \
    -S pocl \
+   -DCLANG_MARCH_FLAG="-mcpu=" \
    -DCMAKE_BUILD_TYPE=release \
    -DCMAKE_C_COMPILER="clang" \
    -DCMAKE_CXX_COMPILER="clang++" \
@@ -66,7 +78,8 @@ cmake \
    -DENABLE_TESTS=OFF \
    -DENABLE_VALGRIND=OFF \
    -DINSTALL_OPENCL_HEADERS=OFF \
-   -DLLC_HOST_CPU=cortex-a53 \
+   -DKERNELLIB_HOST_CPU_VARIANTS="${AARCH64_CPUS}" \
+   -DLLC_HOST_CPU=cortex-a35 \
    -DPOCL_DEBUG_MESSAGES=OFF \
    -DPOCL_INSTALL_ICD_VENDORDIR=/etc/OpenCL/vendors
 

.ci/install-opencl-intel-driver.sh

@@ -7,7 +7,7 @@ set -e -u -o pipefail
 # https://www.intel.com/content/www/us/en/docs/oneapi/installation-guide-linux/2023-0/yum-dnf-zypper.html#GUID-B5018FF2-B9F3-4ADC-9EB6-F99F6BFC7948
 # https://www.intel.com/content/www/us/en/developer/articles/technical/intel-cpu-runtime-for-opencl-applications-with-sycl-support.html
 
-sudo cat > /etc/yum.repos.d/oneAPI.repo <<EOF
+cat > /etc/yum.repos.d/oneAPI.repo <<EOF
 [oneAPI]
 name=Intel® oneAPI repository
 baseurl=https://yum.repos.intel.com/oneapi

.ci/setup.sh

@@ -137,13 +137,13 @@ else  # Linux
             # if [[ "${ARCH}" == "x86_64" ]]; then
             #     .ci/install-opencl-intel-driver.sh
             # fi
+            echo "--- clinfo: ---"
+            clinfo || true
+            # echo "--- llc -mcpu=help ---"
+            # # this shows the possible values for -DLLC_HOST_CPU
+            # # see https://github.com/pocl/pocl/blob/013d2f19f4e8f2e0fd9aedcb70117d6dcc737aa9/doc/sphinx/source/install.rst#L171
+            # llc -mcpu=help
         fi
-        echo "--- clinfo: ---"
-        clinfo || true
-        echo "--- llc -mcpu=help ---"
-        # this shows the possible values for -DLLC_HOST_CPU
-        # see https://github.com/pocl/pocl/blob/013d2f19f4e8f2e0fd9aedcb70117d6dcc737aa9/doc/sphinx/source/install.rst#L171
-        llc -mcpu=help
     fi
     if [[ $TASK == "cuda" ]]; then
         echo 'debconf debconf/frontend select Noninteractive' | debconf-set-selections