_mm_cmpestri instruction alternative on NEON

220 views Asked by At

I am trying to run PicoHTTPParser on ARM platform. There is a function that uses SSE instruction _mm_cmpestri for fast char comparisson. Is there an alternative in NEON? It seems I'll have to use something like VCGT.U8, but it does not look very efficient.

static const char *findchar_fast(const char *buf, const char *buf_end, const char *ranges, size_t ranges_size, int *found)
{
    *found = 0;
    #if __SSE4_2__
    if (likely(buf_end - buf >= 16)) {
        __m128i ranges16 = _mm_loadu_si128((const __m128i *)ranges);

        size_t left = (buf_end - buf) & ~15;
        do {
            __m128i b16 = _mm_loadu_si128((const __m128i*)buf);
            int r = _mm_cmpestri(ranges16, ranges_size, b16, 16, _SIDD_LEAST_SIGNIFICANT | _SIDD_CMP_RANGES | _SIDD_UBYTE_OPS);
            if (unlikely(r != 16)) {
                buf += r;
                *found = 1;
                break;
            }
            buf += 16;
            left -= 16;
        } while (likely(left != 0));
    }
    #else
    /* suppress unused parameter warning */
    (void)buf_end;
    (void)ranges;
    (void)ranges_size;
    #endif
    return buf;
}
0

There are 0 answers