[Bug target/92492] New: [AVX512F] Icc generate much better code for loop vectorization

[crazylht at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92492

            Bug ID: 92492
           Summary: [AVX512F] Icc generate much better code for loop
                    vectorization
           Product: gcc
           Version: 10.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: crazylht at gmail dot com
                CC: hjl.tools at gmail dot com
  Target Milestone: ---
            Target: i386, x86-64

Created attachment 47231
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=47231&action=edit
testcase cut from 525.x264_r

for simple testcase
cat test.c:

typedef unsigned char uint8_t;

static inline uint8_t x264_clip_uint8( int x )
{
  return x&(~63) ? (-x)>>7 : x;
}


void mc_weight( uint8_t *dst, uint8_t *src, int i_width, int i_height )
{
        for( int x = 0; x < i_width; x++ )
            dst[x] = x264_clip_uint8(src[x]);
}

Refer to

https://godbolt.org/z/TnACA-

Icc generate much better code by using vptestmd and maskmov

gcc loop:
------------
        vmovdqu8        (%rsi,%rax), %ymm6
        vpmovzxbw       %xmm6, %ymm3
        vpmovzxwd       %xmm3, %ymm0
        vextracti128    $0x1, %ymm3, %xmm3
        vpand   %ymm5, %ymm0, %ymm9
        vpmovzxwd       %xmm3, %ymm3
        vextracti128    $0x1, %ymm6, %xmm2
        vpcmpd  $0, %ymm4, %ymm9, %k0
        vpmovzxbw       %xmm2, %ymm2
        vpand   %ymm5, %ymm3, %ymm9
        vpcmpd  $0, %ymm4, %ymm9, %k1
        vpmovzxwd       %xmm2, %ymm1
        vextracti128    $0x1, %ymm2, %xmm2
        vpand   %ymm5, %ymm1, %ymm9
        vpmovzxwd       %xmm2, %ymm2
        vpcmpd  $0, %ymm4, %ymm9, %k2
        vpand   %ymm5, %ymm2, %ymm9
        kunpckbw        %k0, %k1, %k0
        vpcmpd  $0, %ymm4, %ymm9, %k1
        vpsubd  %ymm0, %ymm4, %ymm0
        vpsubd  %ymm3, %ymm4, %ymm3
        vpsubd  %ymm1, %ymm4, %ymm1
        vpsubd  %ymm2, %ymm4, %ymm2
        kunpckbw        %k2, %k1, %k1
        kunpckwd        %k0, %k1, %k1
        vpermt2w        %ymm3, %ymm8, %ymm0
        vpermt2w        %ymm2, %ymm8, %ymm1
        vpsraw  $7, %ymm0, %ymm0
        vpsraw  $7, %ymm1, %ymm1
        vpand   %ymm0, %ymm7, %ymm0
        vpand   %ymm1, %ymm7, %ymm1
        vpackuswb       %ymm1, %ymm0, %ymm0
        vpermq  $216, %ymm0, %ymm0
        vmovdqu8        %ymm6, %ymm0{%k1}
        vmovdqu8        %ymm0, (%rdi,%rax)
        addq    $32, %rax
        cmpq    %rcx, %rax
------------

icc loop:
----------
        vpmovzxbd (%rsi,%r8), %ymm3                             #12.31
        vptestmd  %ymm1, %ymm3, %k1                             #5.12
        vpsubd    %ymm3, %ymm0, %ymm2                           #12.31
        vpsrad    $7, %ymm2, %ymm3{%k1}                         #12.31
        vpmovdb   %ymm3, (%r8,%rdi)                             #12.6
        addq      $8, %r8                                       #11.2
        cmpq      %rcx, %r8                                     #11.2
        jb        ..B1.7        # Prob 82%                      #11.2
----------


origin case cut from SPEC2017 525.x264_r, refer to attachment