Re: Vectorization, SIMD

[Gerd Möllmann]

Gerd Möllmann <gerd.moellmann@gmail.com> writes:

> Hi Stef,
>
> I've played with this a bit and made notes, which I'd like to share.
> Please find attached.
>
> My summary so far: Can be done and it's not that difficult. There is
> even some degree of "portability" achievable when writing SIMD code by
> hand. What's will be really no fun, from my POV, is configuration stuff,
> making it work on N platforms with M compilers and so on.
>
> Don't know, maybe it's best to wait for compilers to get better at
> auto-vectorization. Or maybe GCC is better than LLVM in this regard. I
> haven't checked that because I can't really use GCC here on macOS.

FWIW, this is what it could look like in C using ARM Neon.

  // clang -S neon.c
  #include <arm_neon.h>

  int count_char_heads(const char *s, int len) {
      // Create vectors of constants 16x 0xC0 and 0x80.
      uint8x16_t v0xC0 = vmovq_n_u8(0xC0);
      uint8x16_t v0x80 = vmovq_n_u8(0x80);
      int count = 0;

      const uint8x16_t *p = (const uint8x16_t *) s;
      const uint8x16_t *end = (const uint8x16_t *) (s + len);
      for (; p < end; ++p) {
          // And each byte with 0xC0
          uint8x16_t a_and_c0 = vandq_u8(*p, v0xC0);

          // Compare each byte with 0x80
          uint8x16_t eq = vceqq_u8(a_and_c0, v0x80);

          // Reduce vector. Add up all non-zero matches to obtain the
          // number of CHAR_HEAD_P and use that. This could be further
          // improved by using an accumulator vector and reduce when that
          // would overflow (i < 127).
          count += vaddlvq_s8(eq);
      }

    return count;
  }

In Emacs, one would of course functions around such function, that take
the gap into account and so on.

Not difficult, but I don't think I'll do that, since we don't have a
problem :-).