大雑把改行カウント

lf_count.c

size_t simple_count(const char buf[], size_t n)
{
	size_t count = 0;
	for(size_t i = 0; i < n; i++)
	{
		if(buf[i] == '\n')
		{
			++count;
		}
	}
	return count;
}

// 手抜きで buf は 32 バイトアライン限定
size_t simd_count(const char buf[], size_t n)
{
	const __m256i lfs = _mm256_set1_epi8('\n');
	const __m256i ones = _mm256_set1_epi8(1);

	size_t count = 0;
	size_t i;
	for(i = 0; i + 32 <= n; i += 32)
	{
		__m256i x = _mm256_load_si256((const __m256i *)(buf + i));
		x = _mm256_cmpeq_epi8(x, lfs) & ones;
		x = _mm256_hadd_epi16(x, _mm256_permute2x128_si256(x, x, 1));
		x = _mm256_hadd_epi16(x, x);
		x = _mm256_hadd_epi16(x, x);
		x = _mm256_hadd_epi16(x, x);
		size_t c = (x[0] & 0xffff);
		count += (c >> 8) + (c & 0xff);
	}

	count += simple_count(buf + i, n - i);

	return count;
}