[Bug c/83202] Try joining operations on consecutive array elements during tree vectorization

[bugzilla@poradnik-webmastera.com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83202

--- Comment #1 from Daniel Fruzynski <bugzilla@poradnik-webmastera.com> ---
This was compiled with -O3 -mavx -ftree-vectorize
After sending this I noticed that I wrote inner loop incorrectly, I meant one
below. Anyway, it it also not optimized:
for (int j = 0; j < i; j+=4)

I also checked code which could be optimized using operations on YMM registers:

void test(double data[8][8])
{
  for (int i = 0; i < 8; i++)
  {
    for (int j = 0; j < i; j+=4)
    {
      data[i][j] *= data[i][j];
      data[i][j+1] *= data[i][j+1];
      data[i][j+2] *= data[i][j+2];
      data[i][j+3] *= data[i][j+3];
    }
  }
}

gcc output is, hmm, interesting:

test(double (*) [8]):
  vmovupd xmm0, XMMWORD PTR [rdi+64]
  vinsertf128 ymm0, ymm0, XMMWORD PTR [rdi+80], 0x1
  vmulpd ymm0, ymm0, ymm0
  vmovups XMMWORD PTR [rdi+64], xmm0
  vextractf128 XMMWORD PTR [rdi+80], ymm0, 0x1
  vmovupd xmm0, XMMWORD PTR [rdi+128]
  vinsertf128 ymm0, ymm0, XMMWORD PTR [rdi+144], 0x1
  vmulpd ymm0, ymm0, ymm0
  vmovups XMMWORD PTR [rdi+128], xmm0
  vextractf128 XMMWORD PTR [rdi+144], ymm0, 0x1
  vmovupd xmm0, XMMWORD PTR [rdi+192]
  vinsertf128 ymm0, ymm0, XMMWORD PTR [rdi+208], 0x1
  vmulpd ymm0, ymm0, ymm0
  vmovups XMMWORD PTR [rdi+192], xmm0
  vextractf128 XMMWORD PTR [rdi+208], ymm0, 0x1
  vmovupd xmm0, XMMWORD PTR [rdi+256]
  vinsertf128 ymm0, ymm0, XMMWORD PTR [rdi+272], 0x1
  vmulpd ymm0, ymm0, ymm0
  vmovups XMMWORD PTR [rdi+256], xmm0
  vextractf128 XMMWORD PTR [rdi+272], ymm0, 0x1
  vmovupd xmm0, XMMWORD PTR [rdi+320]
  vinsertf128 ymm0, ymm0, XMMWORD PTR [rdi+336], 0x1
  vmulpd ymm0, ymm0, ymm0
  vmovups XMMWORD PTR [rdi+320], xmm0
  vextractf128 XMMWORD PTR [rdi+336], ymm0, 0x1
  vmovupd xmm0, XMMWORD PTR [rdi+352]
  vinsertf128 ymm0, ymm0, XMMWORD PTR [rdi+368], 0x1
  vmulpd ymm0, ymm0, ymm0
  vmovups XMMWORD PTR [rdi+352], xmm0
  vextractf128 XMMWORD PTR [rdi+368], ymm0, 0x1
  vmovupd xmm0, XMMWORD PTR [rdi+384]
  vinsertf128 ymm0, ymm0, XMMWORD PTR [rdi+400], 0x1
  vmulpd ymm0, ymm0, ymm0
  vmovups XMMWORD PTR [rdi+384], xmm0
  vextractf128 XMMWORD PTR [rdi+400], ymm0, 0x1
  vmovupd xmm0, XMMWORD PTR [rdi+416]
  vinsertf128 ymm0, ymm0, XMMWORD PTR [rdi+432], 0x1
  vmulpd ymm0, ymm0, ymm0
  vmovups XMMWORD PTR [rdi+416], xmm0
  vextractf128 XMMWORD PTR [rdi+432], ymm0, 0x1
  vmovupd xmm0, XMMWORD PTR [rdi+448]
  vinsertf128 ymm0, ymm0, XMMWORD PTR [rdi+464], 0x1
  vmulpd ymm0, ymm0, ymm0
  vmovups XMMWORD PTR [rdi+448], xmm0
  vextractf128 XMMWORD PTR [rdi+464], ymm0, 0x1
  vmovupd xmm0, XMMWORD PTR [rdi+480]
  vinsertf128 ymm0, ymm0, XMMWORD PTR [rdi+496], 0x1
  vmulpd ymm0, ymm0, ymm0
  vmovups XMMWORD PTR [rdi+480], xmm0
  vextractf128 XMMWORD PTR [rdi+496], ymm0, 0x1
  vzeroupper
  ret