Dynamic dispatch VSX

.github/workflows/cross-ppc.yml

@@ -35,15 +35,23 @@ jobs:
       uses: actions/checkout@v6
     - name: Setup
       run: |
-        cmake -B _build \
+        cmake -B build/ \
               -DBUILD_TESTS=ON -DDOWNLOAD_DOCTEST=ON \
               -DBUILD_BENCHMARK=${{ matrix.target.full }} -DBUILD_EXAMPLES=${{ matrix.target.full }} \
               -DCMAKE_BUILD_TYPE=Release \
               -DCMAKE_C_FLAGS="${{ matrix.target.flags }}" \
               -DCMAKE_CXX_FLAGS="${{ matrix.target.flags }}" \
               -DCMAKE_TOOLCHAIN_FILE=${{ github.workspace }}/.github/toolchains/${{ matrix.sys.compiler }}-${{ matrix.target.dir }}.cmake
     - name: Build
-      run: cmake --build _build --verbose -j1
+      run: cmake --build build/ --verbose -j1
+    - name: Set CPU feature test expectations
+      run: |
     - name: Testing xsimd
-      run: qemu-${{ matrix.target.platform }} -cpu power10 -L /usr/${{ matrix.target.dir}}/ ./test/test_xsimd
-      working-directory: ${{ github.workspace }}/_build
+      run: |
+        # Set CPU feature test expectations, 0 is explicit absence of the feature
+        export XSIMD_TEST_CPU_ASSUME_SSE4_2="0"
+        export XSIMD_TEST_CPU_ASSUME_NEON64="0"
+        export XSIMD_TEST_CPU_ASSUME_RVV="0"
+        export XSIMD_TEST_CPU_ASSUME_VSX="1"
+
+        qemu-${{ matrix.target.platform }} -cpu power10 -L /usr/${{ matrix.target.dir}}/ ./build/test/test_xsimd

.github/workflows/linux.yml

@@ -123,6 +123,7 @@ jobs:
         # Set CPU feature test expectations, 0 is explicit absence of the feature
         export XSIMD_TEST_CPU_ASSUME_NEON64="0"
         export XSIMD_TEST_CPU_ASSUME_RVV="0"
+        export XSIMD_TEST_CPU_ASSUME_VSX="0"
         cd _build/test
         if echo '${{ matrix.sys.flags }}' | grep -q 'avx512' ; then
           # Running with emulation, must have AVX512, lower tier are checked by implications in tests

include/xsimd/config/xsimd_config.hpp

@@ -499,6 +499,17 @@
 #define XSIMD_WITH_WASM 0
 #endif
 
+/**
+ * @ingroup xsimd_config_macro
+ *
+ * Set to 1 if the target is in the PowerPC architecture family, to 0 otherwise
+ */
+#if defined(__powerpc__) || defined(__powerpc64__) || defined(_ARCH_PPC) || defined(_ARCH_PPC64)
+#define XSIMD_TARGET_PPC 1
+#else
+#define XSIMD_TARGET_PPC 0
+#endif
+
 /**
  * @ingroup xsimd_config_macro
  *

include/xsimd/config/xsimd_cpu_features_ppc.hpp

@@ -0,0 +1,61 @@
+/***************************************************************************
+ * Copyright (c) Johan Mabille, Sylvain Corlay, Wolf Vollprecht and         *
+ * Martin Renou                                                             *
+ * Copyright (c) QuantStack                                                 *
+ * Copyright (c) Serge Guelton                                              *
+ *                                                                          *
+ * Distributed under the terms of the BSD 3-Clause License.                 *
+ *                                                                          *
+ * The full license is in the file LICENSE, distributed with this software. *
+ ***************************************************************************/
+
+#ifndef XSIMD_CPU_FEATURES_PPC_HPP
+#define XSIMD_CPU_FEATURES_PPC_HPP
+
+#include "./xsimd_config.hpp"
+#include "./xsimd_getauxval.hpp"
+
+#if XSIMD_TARGET_PPC && XSIMD_HAVE_LINUX_GETAUXVAL
+// HWCAP_XXX masks to use on getauxval results.
+// Header does not exists on all architectures and masks are architecture
+// specific.
+#include <asm/hwcap.h>
+#endif // XSIMD_TARGET_PPC && XSIMD_HAVE_LINUX_GETAUXVAL
+
+namespace xsimd
+{
+    /**
+     * An opinionated CPU feature detection utility for PowerPC.
+     *
+     * On Linux, runtime detection uses getauxval to query the auxiliary vector.
+     * On other platforms, only compile-time information is used.
+     *
+     * This is well defined on all architectures.
+     * It will always return false on non-PowerPC architectures.
+     */
+    class ppc_cpu_features : private linux_hwcap_backend_default
+    {
+    public:
+        inline bool vsx() const noexcept;
+    };
+
+    /********************
+     *  Implementation  *
+     ********************/
+
+    inline bool ppc_cpu_features::vsx() const noexcept
+    {
+#if XSIMD_TARGET_PPC && XSIMD_HAVE_LINUX_GETAUXVAL
+#ifdef PPC_FEATURE_HAS_VSX
+        return hwcap().has_feature(PPC_FEATURE_HAS_VSX);
+#else
+        // Possibly missing on older Linux distributions
+        return hwcap().has_feature(0x00000080);
+#endif
+#else
+        return XSIMD_WITH_VSX;
+#endif
+    }
+}
+
+#endif

include/xsimd/config/xsimd_cpuid.hpp

@@ -15,6 +15,7 @@
 #include "../types/xsimd_all_registers.hpp"
 #include "./xsimd_cpu_features_arm.hpp"
 #include "./xsimd_cpu_features_riscv.hpp"
+#include "./xsimd_cpu_features_ppc.hpp"
 #include "./xsimd_cpu_features_x86.hpp"
 #include "./xsimd_inline.hpp"
 
@@ -80,9 +81,10 @@ namespace xsimd
                 wasm = 1;
 #endif
 
-#if XSIMD_WITH_VSX
-                vsx = 1;
-#endif
+                // Safe on all platforms, it will be false if non PowerPC.
+                const auto ppc_cpu = xsimd::ppc_cpu_features();
+
+                vsx = ppc_cpu.vsx();
 
                 // Safe on all platforms, it will be all false if non risc-v.
                 const auto riscv_cpu = xsimd::riscv_cpu_features();

test/test_cpu_features.cpp

@@ -169,3 +169,10 @@ TEST_CASE("[cpu_features] risc-v features from environment")
 
     CHECK_ENV_FEATURE("XSIMD_TEST_CPU_ASSUME_RVV", cpu.rvv());
 }
+
+TEST_CASE("[cpu_features] ppc features from environment")
+{
+    xsimd::ppc_cpu_features cpu;
+
+    CHECK_ENV_FEATURE("XSIMD_TEST_CPU_ASSUME_VSX", cpu.vsx());
+}