I've tested XXH3 using xxhash's built-in benchmark tool with clang-7.0.1 and gcc-8.2.1 on an Intel i9-9900K. The processor was otherwise idle, and was running at 5 GHz. The command I tested is `xxhsum -b5i10`.

    SSE2: CFLAGS=-O3
    AVX2: CFLAGS="-O3 -mavx2"
    ARCH: CFLAGS="-O3 -march=native"

    Compiler  Mode  Speed
    gcc-8     SSE2  32.8 GB/s
    clang-7   SSE2  36.5 GB/s 
    gcc-8     AVX2  44.1 GB/s
    clang-7   AVX2  68.3 GB/s
    gcc-8     ARCH  60.9 GB/s
    clang-7   ARCH  69.7 GB/s

gcc nearly catches up with clang when compiled with -march=native, but with only -mavx2 it isn't performing as well.