Merge branch 'master' of https://github.com/intel/ARM_NEON_2_x86_SSE
diff --git a/ReadMe.md b/ReadMe.md index ef5b90f..dbaedd6 100644 --- a/ReadMe.md +++ b/ReadMe.md
@@ -6,8 +6,10 @@ To take advantage of this file just include it in your project that uses ARM NEON intinsics instead of "arm_neon.h", compile it as usual and enjoy the result. -Considering 64-128 bit width of the NEON istructions AVX usage in x86 implementaion is litited, SSE is used in more than 90% of functions. -For significant performance improvement in some cases you might need to define USE_SSE4 or USE_AVX2 in your project settings. Otherwise by default SIMD up to SSSE3 to be used. +Considering 64-128 bit width of the NEON istructions AVX vector instructions usage in x86 implementaion is litited, SSE is used in more than 90% of functions. +However for some functions significant performance improvement it is recommended to compile your x86 code with the highest level of vector instructions supported by the target CPU -say SSE4 or AVX2. +Please notice that AVX2 enabling for this project compilation includes FMA instructions as well. +Just use the corresponding compiler flags for the purpose. In some cases you might need to define USE_SSE4 or USE_AVX2 in your project settings manually. Otherwise by default SIMD up to SSSE3 to be used. If NEON2SSE_DISABLE_PERFORMANCE_WARNING macro is defined, then the performance warnings for serial functions implementaions are disabled.