https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121970
--- Comment #5 from Hongtao Liu <liuhongt at gcc dot gnu.org> --- (In reply to Andrew Pinski from comment #4) > (In reply to Hongtao Liu from comment #1) > > Although the option is x86 specific, but I think the issue is middle-end, > > it's related how MOVE_MAX and STORE_MAX_PIECES is used. > > #define MOVE_MAX \ > ((TARGET_AVX512F \ > && (ix86_move_max == PVW_AVX512 \ > || ix86_store_max == PVW_AVX512)) \ > ? 64 \ > : ((TARGET_AVX \ > && (ix86_move_max >= PVW_AVX256 \ > || ix86_store_max >= PVW_AVX256)) \ > ? 32 \ > : ((TARGET_SSE2 \ > && TARGET_SSE_UNALIGNED_LOAD_OPTIMAL \ > && TARGET_SSE_UNALIGNED_STORE_OPTIMAL) \ > ? 16 : UNITS_PER_WORD))) > > > x86 defines MOVE_MAX based on if either option is set to 512 here. > > While STORE_MAX_PIECES is defined by store max only: > > #define STORE_MAX_PIECES \ > (TARGET_INTER_UNIT_MOVES_TO_VEC \ > ? ((TARGET_AVX512F && ix86_store_max == PVW_AVX512) \ > ? 64 \ > : ((TARGET_AVX \ > && ix86_store_max >= PVW_AVX256) \ > ? 32 \ > : ((TARGET_SSE2 \ > && TARGET_SSE_UNALIGNED_STORE_OPTIMAL) \ > ? 16 : UNITS_PER_WORD))) \ > : UNITS_PER_WORD) > > > So this seems like a target issue. Ah, I see, thanks. H.J, I think we should remove ix86_store_max from MOVE_MAX.
