count-newlines.cpp

// Count newlines, 32 characters at a time
__forceinline void countNewlines( __m256i ascii, __m256i& counters )
{
	// Find '\n' characters, 32 bytes at a time
	__m256i newlines = _mm256_cmpeq_epi8( ascii, _mm256_set1_epi8( '\n' ) );
	// Convert 0xFF to 1
	newlines = _mm256_and_si256( newlines, _mm256_set1_epi8( 1 ) );
	// Accumulate the counters.
	// The key part is _mm256_sad_epu8 instruction, here’s copy-paste from the docs:
	// Compute the absolute differences of packed unsigned 8-bit integers in "a" and "b",
	// then horizontally sum each consecutive 8 differences to produce four unsigned 16-bit integers,
	// and pack these unsigned 16-bit integers in the low 16 bits of 64-bit elements in "dst".
	counters = _mm256_add_epi64( counters, _mm256_sad_epu8( newlines, _mm256_setzero_si256() ) );
}

// Horizontal sum of 64-bit integer lanes.
// You should only call this once, after the complete file is processed.
int64_t horizontalAdd( __m256i counters )
{
	// Add 4 lanes into 2
	__m128i tmp = _mm_add_epi64( _mm256_castsi256_si128( counters ), _mm256_extracti128_si256( counters, 1 ) );
	// Add 2 lanes into 1
	tmp = _mm_add_epi64( tmp, _mm_unpackhi_epi64( tmp, tmp ) );
	// Return the result
	return _mm_cvtsi128_si64( tmp );
}