On Mon, Nov 25, 2024 at 12:13 PM Li, Pan2 <pan2...@intel.com> wrote: > > > This doesn't look right, you need scan-tree-dump-times with tree dumps. > > Oh, yes, interesting the make check doesn't complain anything about it. > > Summaries valid vectorize cases as blow, is there anything missing? > > X86 uint8_t uint16_t > X86_32 uint8_t, uint16_t, uint32_t > X86_64 uint8_t, uint16_t, uint32_t, uint64_t
We only have x86_32 and x86_64. Both can vectorize uint8_t and uint16_t when -msse2 is added to the compile flags. For scalar code, only x86_64 supports uint64_t. > BTW, how can I tell the x86, x86_32, x86_64 from the test source code? #ifdef __i386__ for x86_32 and #ifdef __x86_64__ for x86_64, but you won't need that because SSE vectorization is supported for both with -msse2 compile flag. Uros.
