Initial support for Altivec

Only supports altivec with vxs extension, tested on both little and big
endian architectures.

Several operations could be improved, but this should pass CI.

Notable changes:

- document the fact that slide_left and slide_right are unreliable on
  big endian architecture

- gcc chokes on force_inline for altivec, so dismiss it

- some common code were making assumption of conversion between
  batch_bool registers and batch_registers that proved to be difficult
  for altivec type system, make the conversion explicit.

Author: serge-sans-paille

.github/toolchains/gcc-powerpc64-linux-gnu.cmake

@@ -0,0 +1,5 @@
+set(CMAKE_SYSTEM_PROCESSOR powerpc64)
+set(triple powerpc64-linux-gnu)
+
+include(${CMAKE_CURRENT_LIST_DIR}/gcc.cmake)
+

.github/toolchains/gcc-powerpc64le-linux-gnu.cmake

@@ -0,0 +1,5 @@
+set(CMAKE_SYSTEM_PROCESSOR powerpc64le)
+set(triple powerpc64le-linux-gnu)
+
+include(${CMAKE_CURRENT_LIST_DIR}/gcc.cmake)
+

.github/workflows/cross.yml renamed to .github/workflows/cross-arm.yml

@@ -0,0 +1,44 @@
+name: PowerPC cross-compilation build
+on: [push, pull_request]
+concurrency:
+  group: ${{ github.workflow }}-${{ github.job }}-${{ github.ref }}
+  cancel-in-progress: true
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    name: '${{ matrix.target.arch }}, ${{ matrix.sys.compiler }} ${{ matrix.sys.version }}'
+    strategy:
+      matrix:
+        target:
+          - { platform: 'ppc64le',     dir: 'powerpc64le-linux-gnu',   flags: '-maltivec -mvsx -mcpu=power10', full: 'OFF' }
+          - { platform: 'ppc64',     dir: 'powerpc64-linux-gnu',   flags: '-maltivec -mvsx -mcpu=power10', full: 'OFF' }
+        sys:
+          - { compiler: 'gcc',   version: '12' }
+    steps:
+    - name: Setup compiler
+      if: ${{ matrix.sys.compiler == 'gcc' }}
+      run: |
+        sudo apt-get update || exit 1
+        sudo apt-get --no-install-suggests --no-install-recommends install g++-${{ matrix.sys.version }}-${{ matrix.target.dir }} g++-${{ matrix.sys.version }}-multilib || exit 1
+        sudo update-alternatives --remove-all ${{ matrix.target.dir }}-gcc || true
+        sudo update-alternatives --remove-all ${{ matrix.target.dir }}-g++ || true
+        sudo update-alternatives --install /usr/bin/${{ matrix.target.dir }}-gcc ${{ matrix.target.dir }}-gcc /usr/bin/${{ matrix.target.dir }}-gcc-${{ matrix.sys.version }} 20
+        sudo update-alternatives --install /usr/bin/${{ matrix.target.dir }}-g++ ${{ matrix.target.dir }}-g++ /usr/bin/${{ matrix.target.dir }}-g++-${{ matrix.sys.version }} 20
+    - name: Setup QEMU
+      run: |
+        sudo apt-get --no-install-suggests --no-install-recommends install qemu-user
+    - name: Setup Ninja
+      run: |
+        sudo apt-get install ninja-build
+    - name: Checkout xsimd
+      uses: actions/checkout@v3
+    - name: Setup
+      run: |
+        mkdir _build
+        cd _build && cmake .. -DBUILD_TESTS=ON -DDOWNLOAD_DOCTEST=ON -DBUILD_BENCHMARK=${{ 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
+    - name: Testing xsimd
+      run: |
+        qemu-${{ matrix.target.platform }} -cpu power10 -L /usr/${{ matrix.target.dir}}/ ./test/test_xsimd
+      working-directory: ${{ github.workspace }}/_build

.github/workflows/cross-ppc.yml

@@ -49,6 +49,7 @@ ${XSIMD_INCLUDE_DIR}/xsimd/arch/xsimd_sse3.hpp
 ${XSIMD_INCLUDE_DIR}/xsimd/arch/xsimd_sse4_1.hpp
 ${XSIMD_INCLUDE_DIR}/xsimd/arch/xsimd_sse4_2.hpp
 ${XSIMD_INCLUDE_DIR}/xsimd/arch/xsimd_ssse3.hpp
+${XSIMD_INCLUDE_DIR}/xsimd/arch/xsimd_altivec.hpp
 ${XSIMD_INCLUDE_DIR}/xsimd/arch/xsimd_sve.hpp
 ${XSIMD_INCLUDE_DIR}/xsimd/config/xsimd_arch.hpp
 ${XSIMD_INCLUDE_DIR}/xsimd/config/xsimd_config.hpp
@@ -70,6 +71,7 @@ ${XSIMD_INCLUDE_DIR}/xsimd/types/xsimd_fma3_sse_register.hpp
 ${XSIMD_INCLUDE_DIR}/xsimd/types/xsimd_fma4_register.hpp
 ${XSIMD_INCLUDE_DIR}/xsimd/types/xsimd_common_arch.hpp
 ${XSIMD_INCLUDE_DIR}/xsimd/types/xsimd_register.hpp
+${XSIMD_INCLUDE_DIR}/xsimd/types/xsimd_altivec_register.hpp
 ${XSIMD_INCLUDE_DIR}/xsimd/types/xsimd_rvv_register.hpp
 ${XSIMD_INCLUDE_DIR}/xsimd/types/xsimd_sse2_register.hpp
 ${XSIMD_INCLUDE_DIR}/xsimd/types/xsimd_sse3_register.hpp

CMakeLists.txt

@@ -54,6 +54,7 @@ x86          | AVX512BW, AVX512CD, AVX512DQ, AVX512F (gcc7 and higher)
 x86 AMD      | FMA4
 ARM          | NEON, NEON64, SVE128/256/512 (fixed vector size)
 WebAssembly  | WASM
+powerpc64    | Altivec+VSX
 RISC-V       | RISC-V128/256/512 (fixed vector size)
 
 ## Installation

README.md

@@ -9,6 +9,7 @@ INPUT             = ../include/xsimd/types/xsimd_api.hpp \
                     ../include/xsimd/memory/xsimd_aligned_allocator.hpp \
                     ../include/xsimd/types/xsimd_common_arch.hpp \
                     ../include/xsimd/types/xsimd_traits.hpp \
+                    ../include/xsimd/types/xsimd_altivec_register.hpp \
                     ../include/xsimd/types/xsimd_avx2_register.hpp \
                     ../include/xsimd/types/xsimd_avx512bw_register.hpp \
                     ../include/xsimd/types/xsimd_avx512cd_register.hpp \

docs/Doxyfile

@@ -203,10 +203,9 @@ namespace xsimd
         {
             if (std::is_signed<T>::value)
             {
-                auto mask = (other >> (8 * sizeof(T) - 1));
                 auto self_pos_branch = min(std::numeric_limits<T>::max() - other, self);
                 auto self_neg_branch = max(std::numeric_limits<T>::min() - other, self);
-                return other + select(batch_bool<T, A>(mask.data), self_neg_branch, self_pos_branch);
+                return other + select(other >= 0, self_pos_branch, self_neg_branch);
             }
             else
             {

include/xsimd/arch/common/xsimd_common_arithmetic.hpp

@@ -1087,7 +1087,7 @@ namespace xsimd
         template <class A, class T>
         XSIMD_INLINE batch<T, A> from_bool(batch_bool<T, A> const& self, requires_arch<common>) noexcept
         {
-            return batch<T, A>(self.data) & batch<T, A>(1);
+            return batch<T, A>((typename batch<T, A>::register_type)self.data) & batch<T, A>(1);
         }
 
         // horner