#[cfg(target_arch = "x86")] use std::arch::x86::{ __m128i, _mm_and_si128, _mm_cmpeq_epi8, _mm_extract_epi32, _mm_loadu_si128, _mm_sad_epu8, _mm_set1_epi8, _mm_setzero_si128, _mm_sub_epi8, _mm_xor_si128, }; #[cfg(target_arch = "x86_64")] use std::arch::x86_64::{ __m128i, _mm_and_si128, _mm_cmpeq_epi8, _mm_extract_epi32, _mm_loadu_si128, _mm_sad_epu8, _mm_set1_epi8, _mm_setzero_si128, _mm_sub_epi8, _mm_xor_si128, }; #[target_feature(enable = "sse2")] pub unsafe fn _mm_set1_epu8(a: u8) -> __m128i { _mm_set1_epi8(a as i8) } #[target_feature(enable = "sse2")] pub unsafe fn mm_cmpneq_epi8(a: __m128i, b: __m128i) -> __m128i { _mm_xor_si128(_mm_cmpeq_epi8(a, b), _mm_set1_epi8(-1)) } const MASK: [u8; 32] = [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, ]; #[target_feature(enable = "sse2")] unsafe fn mm_from_offset(slice: &[u8], offset: usize) -> __m128i { _mm_loadu_si128(slice.as_ptr().offset(offset as isize) as *const _) } #[target_feature(enable = "sse2")] unsafe fn sum(u8s: &__m128i) -> usize { let sums = _mm_sad_epu8(*u8s, _mm_setzero_si128()); (_mm_extract_epi32(sums, 0) + _mm_extract_epi32(sums, 2)) as usize } #[target_feature(enable = "sse2")] pub unsafe fn chunk_count(haystack: &[u8], needle: u8) -> usize { assert!(haystack.len() >= 16); let mut offset = 0; let mut count = 0; let needles = _mm_set1_epu8(needle); // 4080 while haystack.len() >= offset + 16 * 255 { let mut counts = _mm_setzero_si128(); for _ in 0..255 { counts = _mm_sub_epi8( counts, _mm_cmpeq_epi8(mm_from_offset(haystack, offset), needles) ); offset += 16; } count += sum(&counts); } // 2048 if haystack.len() >= offset + 16 * 128 { let mut counts = _mm_setzero_si128(); for _ in 0..128 { counts = _mm_sub_epi8( counts, _mm_cmpeq_epi8(mm_from_offset(haystack, offset), needles) ); offset += 16; } count += sum(&counts); } // 16 let mut counts = _mm_setzero_si128(); for i in 0..(haystack.len() - offset) / 16 { counts = _mm_sub_epi8( counts, _mm_cmpeq_epi8(mm_from_offset(haystack, offset + i * 16), needles) ); } if haystack.len() % 16 != 0 { counts = _mm_sub_epi8( counts, _mm_and_si128( _mm_cmpeq_epi8(mm_from_offset(haystack, haystack.len() - 16), needles), mm_from_offset(&MASK, haystack.len() % 16) ) ); } count += sum(&counts); count } #[target_feature(enable = "sse2")] unsafe fn is_leading_utf8_byte(u8s: __m128i) -> __m128i { mm_cmpneq_epi8(_mm_and_si128(u8s, _mm_set1_epu8(0b1100_0000)), _mm_set1_epu8(0b1000_0000)) } #[target_feature(enable = "sse2")] pub unsafe fn chunk_num_chars(utf8_chars: &[u8]) -> usize { assert!(utf8_chars.len() >= 16); let mut offset = 0; let mut count = 0; // 4080 while utf8_chars.len() >= offset + 16 * 255 { let mut counts = _mm_setzero_si128(); for _ in 0..255 { counts = _mm_sub_epi8( counts, is_leading_utf8_byte(mm_from_offset(utf8_chars, offset)) ); offset += 16; } count += sum(&counts); } // 2048 if utf8_chars.len() >= offset + 16 * 128 { let mut counts = _mm_setzero_si128(); for _ in 0..128 { counts = _mm_sub_epi8( counts, is_leading_utf8_byte(mm_from_offset(utf8_chars, offset)) ); offset += 16; } count += sum(&counts); } // 16 let mut counts = _mm_setzero_si128(); for i in 0..(utf8_chars.len() - offset) / 16 { counts = _mm_sub_epi8( counts, is_leading_utf8_byte(mm_from_offset(utf8_chars, offset + i * 16)) ); } if utf8_chars.len() % 16 != 0 { counts = _mm_sub_epi8( counts, _mm_and_si128( is_leading_utf8_byte(mm_from_offset(utf8_chars, utf8_chars.len() - 16)), mm_from_offset(&MASK, utf8_chars.len() % 16) ) ); } count += sum(&counts); count }