Fix bad vcplmulq detection and update simd feature tests.

awslabs · Feb 5, 2024 · 8cac3d7 · 8cac3d7
1 parent ddb7d02
commit 8cac3d7
Show file tree

Hide file tree

Showing 4 changed files with 10 additions and 5 deletions.
diff --git a/cmake/AwsFeatureTests.cmake b/cmake/AwsFeatureTests.cmake
@@ -17,6 +17,8 @@ if(MINGW)
     set(USE_CPU_EXTENSIONS OFF)
 endif()
 
+set(AWS_USE_CPU_EXTENSIONS USE_CPU_EXTENSIONS)
+
 if(NOT CMAKE_CROSSCOMPILING)
     check_c_source_runs("
     #include <stdbool.h>

diff --git a/cmake/AwsSIMD.cmake b/cmake/AwsSIMD.cmake
@@ -23,18 +23,20 @@ else()
 endif()
 
 if (USE_CPU_EXTENSIONS)
+    set(AVX_CFLAGS ${AWS_SSE4_2_FLAG})
+
     check_c_compiler_flag(${AWS_AVX2_FLAG} HAVE_M_AVX2_FLAG)
     if (HAVE_M_AVX2_FLAG)
-        set(AVX_CFLAGS ${AWS_AVX2_FLAG})
+        set(AVX_CFLAGS "${AWS_AVX2_FLAG} ${AVX_CFLAGS}")
     endif()
 
     check_c_compiler_flag("${AWS_AVX512_FLAG} ${AWS_CLMUL_FLAG}" HAVE_M_AVX512_FLAG)
     if (HAVE_M_AVX512_FLAG)
-        set(AVX_CFLAGS "${AVX_CFLAGS} ${AWS_AVX512_FLAG} ${AWS_CLMUL_FLAG} ${AWS_SSE4_2_FLAG}")
+        set(AVX_CFLAGS "${AWS_AVX512_FLAG} ${AWS_CLMUL_FLAG} ${AVX_CFLAGS}")
     endif()
 
     set(old_flags "${CMAKE_REQUIRED_FLAGS}")
-    set(CMAKE_REQUIRED_FLAGS "${CMAKE_REQUIRED_FLAGS} ${AVX_CFLAGS}")
+    set(CMAKE_REQUIRED_FLAGS "${CMAKE_REQUIRED_FLAGS} ${AVX_CFLAGS} ${WERROR_FLAG}")
 
     check_c_source_compiles("
         #include <immintrin.h>

diff --git a/include/aws/common/config.h.in b/include/aws/common/config.h.in
@@ -28,5 +28,6 @@
 #cmakedefine AWS_ARCH_ARM64
 #cmakedefine AWS_ARCH_INTEL
 #cmakedefine AWS_ARCH_INTEL_X64
+#cmakedefine AWS_USE_CPU_EXTENSIONS
 
 #endif
diff --git a/source/arch/intel/cpuid.c b/source/arch/intel/cpuid.c
@@ -116,8 +116,8 @@ static bool s_has_bmi2(void) {
 static bool s_has_vpclmulqdq(void) {
     uint32_t abcd[4];
     /* Check VPCLMULQDQ:
-     * CPUID.(EAX=07H, ECX=0H):ECX.VPCLMULQDQ[bit 20]==1 */
-    uint32_t vpclmulqdq_mask = (1 << 20);
+     * CPUID.(EAX=07H, ECX=0H):ECX.VPCLMULQDQ[bit 10]==1 */
+    uint32_t vpclmulqdq_mask = (1 << 10);
     aws_run_cpuid(7, 0, abcd);
     if ((abcd[2] & vpclmulqdq_mask) != vpclmulqdq_mask) {
         return false;