https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99411
--- Comment #8 from Jan Hubicka <hubicka at gcc dot gnu.org> --- Compared to aocc we also do worse on zen4: jh@alberti:~/tsvc/bin> ~/trunk-install/bin/gcc -Ofast -march=native s311.c jh@alberti:~/tsvc/bin> time ./a.out real 0m3.207s user 0m3.206s sys 0m0.000s jh@alberti:~/tsvc/bin> ~/aocc-compiler-4.0.0/bin/clang -Ofast -march=native s311.c jh@alberti:~/tsvc/bin> time ./a.out real 0m1.221s user 0m1.221s sys 0m0.000s aocc code seems similar to clangs from two years ago except for additional use of avx512. main: # @main .cfi_startproc # %bb.0: # %entry xorl %eax, %eax .p2align 4, 0x90 .LBB0_1: # %vector.ph # =>This Loop Header: Depth=1 # Child Loop BB0_2 Depth 2 vxorps %xmm0, %xmm0, %xmm0 movq $-128000, %rcx # imm = 0xFFFE0C00 vxorps %xmm1, %xmm1, %xmm1 vxorps %xmm2, %xmm2, %xmm2 vxorps %xmm3, %xmm3, %xmm3 .p2align 4, 0x90 .LBB0_2: # %vector.body # Parent Loop BB0_1 Depth=1 # => This Inner Loop Header: Depth=2 vaddps a+128000(%rcx), %zmm0, %zmm0 vaddps a+128064(%rcx), %zmm1, %zmm1 vaddps a+128128(%rcx), %zmm2, %zmm2 vaddps a+128192(%rcx), %zmm3, %zmm3 addq $256, %rcx # imm = 0x100 jne .LBB0_2 # %bb.3: # %middle.block # in Loop: Header=BB0_1 Depth=1 incl %eax cmpl $1000000, %eax # imm = 0xF4240 jne .LBB0_1 # %bb.4: # %for.cond.cleanup vaddps %zmm0, %zmm1, %zmm0 xorl %eax, %eax vaddps %zmm0, %zmm2, %zmm0 vaddps %zmm0, %zmm3, %zmm0 vextractf64x4 $1, %zmm0, %ymm1 vaddps %zmm1, %zmm0, %zmm0 vextractf128 $1, %ymm0, %xmm1 vaddps %xmm1, %xmm0, %xmm0 vpermilpd $1, %xmm0, %xmm1 # xmm1 = xmm0[1,0] vaddps %xmm1, %xmm0, %xmm0 vmovshdup %xmm0, %xmm1 # xmm1 = xmm0[1,1,3,3] vaddss %xmm1, %xmm0, %xmm0 vucomiss .LCPI0_0(%rip), %xmm0 seta %al vzeroupper retq