diff options
Diffstat (limited to 'vendor/bytecount/src/integer_simd.rs')
-rw-r--r-- | vendor/bytecount/src/integer_simd.rs | 111 |
1 files changed, 111 insertions, 0 deletions
diff --git a/vendor/bytecount/src/integer_simd.rs b/vendor/bytecount/src/integer_simd.rs new file mode 100644 index 000000000..48f2ee8d9 --- /dev/null +++ b/vendor/bytecount/src/integer_simd.rs @@ -0,0 +1,111 @@ +#[cfg(not(feature = "runtime-dispatch-simd"))] +use core::{mem, ptr, usize}; +#[cfg(feature = "runtime-dispatch-simd")] +use std::{mem, ptr, usize}; + +fn splat(byte: u8) -> usize { + let lo = usize::MAX / 0xFF; + lo * byte as usize +} + +unsafe fn usize_load_unchecked(bytes: &[u8], offset: usize) -> usize { + let mut output = 0; + ptr::copy_nonoverlapping( + bytes.as_ptr().add(offset), + &mut output as *mut usize as *mut u8, + mem::size_of::<usize>() + ); + output +} + +fn bytewise_equal(lhs: usize, rhs: usize) -> usize { + let lo = usize::MAX / 0xFF; + let hi = lo << 7; + + let x = lhs ^ rhs; + !((((x & !hi) + !hi) | x) >> 7) & lo +} + +fn sum_usize(values: usize) -> usize { + let every_other_byte_lo = usize::MAX / 0xFFFF; + let every_other_byte = every_other_byte_lo * 0xFF; + + // Pairwise reduction to avoid overflow on next step. + let pair_sum: usize = (values & every_other_byte) + ((values >> 8) & every_other_byte); + + // Multiplication results in top two bytes holding sum. + pair_sum.wrapping_mul(every_other_byte_lo) >> ((mem::size_of::<usize>() - 2) * 8) +} + +fn is_leading_utf8_byte(values: usize) -> usize { + // a leading UTF-8 byte is one which does not start with the bits 10. + ((!values >> 7) | (values >> 6)) & splat(1) +} + +pub fn chunk_count(haystack: &[u8], needle: u8) -> usize { + let chunksize = mem::size_of::<usize>(); + assert!(haystack.len() >= chunksize); + + unsafe { + let mut offset = 0; + let mut count = 0; + + let needles = splat(needle); + + // 2040 + while haystack.len() >= offset + chunksize * 255 { + let mut counts = 0; + for _ in 0..255 { + counts += bytewise_equal(usize_load_unchecked(haystack, offset), needles); + offset += chunksize; + } + count += sum_usize(counts); + } + + // 8 + let mut counts = 0; + for i in 0..(haystack.len() - offset) / chunksize { + counts += bytewise_equal(usize_load_unchecked(haystack, offset + i * chunksize), needles); + } + if haystack.len() % 8 != 0 { + let mask = usize::from_le(!(!0 >> ((haystack.len() % chunksize) * 8))); + counts += bytewise_equal(usize_load_unchecked(haystack, haystack.len() - chunksize), needles) & mask; + } + count += sum_usize(counts); + + count + } +} + +pub fn chunk_num_chars(utf8_chars: &[u8]) -> usize { + let chunksize = mem::size_of::<usize>(); + assert!(utf8_chars.len() >= chunksize); + + unsafe { + let mut offset = 0; + let mut count = 0; + + // 2040 + while utf8_chars.len() >= offset + chunksize * 255 { + let mut counts = 0; + for _ in 0..255 { + counts += is_leading_utf8_byte(usize_load_unchecked(utf8_chars, offset)); + offset += chunksize; + } + count += sum_usize(counts); + } + + // 8 + let mut counts = 0; + for i in 0..(utf8_chars.len() - offset) / chunksize { + counts += is_leading_utf8_byte(usize_load_unchecked(utf8_chars, offset + i * chunksize)); + } + if utf8_chars.len() % 8 != 0 { + let mask = usize::from_le(!(!0 >> ((utf8_chars.len() % chunksize) * 8))); + counts += is_leading_utf8_byte(usize_load_unchecked(utf8_chars, utf8_chars.len() - chunksize)) & mask; + } + count += sum_usize(counts); + + count + } +} |