[Bug target/125189] [missed optimization] VPERM instruction not used to vectorize gather (lookup) in AVX512

[vincenzo.innocente at cern dot ch via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125189

--- Comment #2 from vincenzo Innocente <vincenzo.innocente at cern dot ch> ---
godbold emits
        vcvttps2dq      zmm0, zmm2
        vpsrad  zmm1, zmm0, 31
        vpsrld  zmm1, zmm1, 28
        vpaddd  zmm0, zmm0, zmm1
        vpandd  zmm0, zmm0, zmm3
        vpsubd  zmm0, zmm0, zmm1
        vmovups zmm1, ZMMWORD PTR "y"[rip]

for
"for (int k=0; k<16; ++k) j[k] = int(a[i+k])%16;"

so I did not bother to write it myself...
(and it is not a realistic use case: usually one builds the index with the most
significand bits)