Fix linux CI and AVX512-128/256

.github/workflows/linux.yml

@@ -119,7 +119,7 @@ jobs:
           CMAKE_EXTRA_ARGS="$CMAKE_EXTRA_ARGS -DTARGET_ARCH=knm"
         fi
         if [[ '${{ matrix.sys.flags }}' == 'i386' ]]; then
-          CXX_FLAGS="$CXX_FLAGS -m32"
+          CXXFLAGS="$CXXFLAGS -m32"
         fi
         if [[ '${{ matrix.sys.flags }}' == 'force_no_instr_set' ]]; then
           :
@@ -128,17 +128,19 @@ jobs:
         fi
 
         # Cheap way of spotting uninitialized read
-        CXX_FLAGS="$CXX_FLAGS -ftrivial-auto-var-init=pattern"
+        if echo 'int main(){}' | $CXX -x c++ -ftrivial-auto-var-init=pattern -fsyntax-only - 2>/dev/null; then
+          CXXFLAGS="$CXXFLAGS -ftrivial-auto-var-init=pattern"
+        fi
 
         cmake -B _build \
               -DBUILD_TESTS=ON \
-              -DBUILD_BENCHMARK=ON \
               -DBUILD_EXAMPLES=ON \
+              -DBUILD_ARCHITECTURE_TESTS=OFF \
               -DCMAKE_BUILD_TYPE=Release \
               -DCMAKE_C_COMPILER=$CC \
               -DCMAKE_CXX_COMPILER=$CXX \
               $CMAKE_EXTRA_ARGS \
-              -DCMAKE_CXX_FLAGS='$CXX_FLAGS' \
+              -DCMAKE_CXX_FLAGS="$CXXFLAGS" \
               -G Ninja
     - name: Build
       run: cmake --build _build
@@ -161,3 +163,33 @@ jobs:
           export XSIMD_TEST_CPU_ASSUME_MANUFACTURER="intel,amd"
           ./test_xsimd
         fi
+
+  build-extra:
+    runs-on: ubuntu-latest
+    name: 'clang 18 - architectures and benchmark'
+    steps:
+    - name: Setup compiler
+      run: |
+        sudo apt-get update || exit 1
+        sudo apt-get --no-install-suggests --no-install-recommends install clang-18 || exit 1
+        echo "CC=clang-18" >> $GITHUB_ENV
+        echo "CXX=clang++-18" >> $GITHUB_ENV
+    - name: Checkout xsimd
+      uses: actions/checkout@v6
+    - name: Install mamba
+      uses: mamba-org/setup-micromamba@v2
+      with:
+        environment-file: environment.yml
+    - name: Configure build
+      run: |
+        cmake -B _build \
+              -DBUILD_TESTS=ON \
+              -DBUILD_ARCHITECTURE_TESTS=ON \
+              -DBUILD_BENCHMARK=ON \
+              -DCMAKE_BUILD_TYPE=Release \
+              -DCMAKE_C_COMPILER=$CC \
+              -DCMAKE_CXX_COMPILER=$CXX \
+              -DXSIMD_ENABLE_WERROR=ON \
+              -G Ninja
+    - name: Build
+      run: cmake --build _build

include/xsimd/arch/xsimd_avx.hpp

@@ -1726,12 +1726,12 @@ namespace xsimd
                                           { return ssub(batch<T, sse4_2>(s), batch<T, sse4_2>(o)); },
                                           self, other);
             }
-            else if (std::is_signed_v<T>)
+            else if constexpr (std::is_signed_v<T>)
             {
-                auto mask = (other >> (8 * sizeof(T) - 1));
+                auto other_is_negative = other < batch<T, A>(T(0));
                 auto self_overflow_branch = min(std::numeric_limits<T>::max() + other, self);
                 auto self_underflow_branch = max(std::numeric_limits<T>::min() + other, self);
-                return select(batch_bool<T, A>(mask.data), self_overflow_branch, self_underflow_branch) - other;
+                return select(other_is_negative, self_overflow_branch, self_underflow_branch) - other;
             }
             else
             {

include/xsimd/arch/xsimd_avx512f.hpp

@@ -1561,7 +1561,7 @@ namespace xsimd
 
                     const uint32_t magic = (0x80 + 0x4000 + 0x200000 + 0x10000000);
 
-                    unsigned char res = ((data * magic) >> 24) & 0xFF;
+                    unsigned char res = (data * magic) >> 28;
                     return res;
                 }
                 else if constexpr (N == 2)
@@ -1571,7 +1571,7 @@ namespace xsimd
 
                     const uint16_t magic = (0x80 + 0x4000);
 
-                    unsigned char res = ((data * magic) >> 8) & 0xFF;
+                    unsigned char res = (data * magic) >> 14;
                     return res;
                 }
             }

include/xsimd/arch/xsimd_common_fwd.hpp

@@ -60,6 +60,10 @@ namespace xsimd
         XSIMD_INLINE batch<T, A> bitwise_rshift(batch<T, A> const& self, batch<T, A> const& other, requires_arch<common>) noexcept;
         template <size_t shift, class A, class T, class = std::enable_if_t<std::is_integral_v<T>>>
         XSIMD_INLINE batch<T, A> bitwise_rshift(batch<T, A> const& self, requires_arch<common>) noexcept;
+        template <class A, class T, class Mask>
+        XSIMD_INLINE batch<T, A> decr_if(batch<T, A> const& self, Mask const& mask, requires_arch<common>) noexcept;
+        template <class A, class T, class Mask>
+        XSIMD_INLINE batch<T, A> incr_if(batch<T, A> const& self, Mask const& mask, requires_arch<common>) noexcept;
         template <class A, class T>
         XSIMD_INLINE batch_bool<T, A> gt(batch<T, A> const& self, batch<T, A> const& other, requires_arch<common>) noexcept;
         template <class A, class T, class = std::enable_if_t<std::is_integral_v<T>>>

test/CMakeLists.txt

@@ -218,7 +218,10 @@ if (XSIMD_ENABLE_WERROR)
 endif()
 
 add_subdirectory(doc)
-add_subdirectory(architectures)
+OPTION(BUILD_ARCHITECTURE_TESTS "xsimd compilation tests for various -march targets" ON)
+if(BUILD_ARCHITECTURE_TESTS)
+    add_subdirectory(architectures)
+endif()
 
 if(EMULATED_COMPILE_FLAGS)
     message(STATUS ${EMULATED_COMPILE_FLAGS})