Skip to content

Possible alternative to our latin1_to_utf8_avx512 routine. #346

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
lemire wants to merge 1 commit into
base: master
Choose a base branch
from
Draft

Possible alternative to our latin1_to_utf8_avx512 routine. #346

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
40 changes: 34 additions & 6 deletions src/icelake/icelake_convert_latin1_to_utf8.inl.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -70,15 +70,43 @@ We adjust for the bytes that have their two most significant bits. This takes ca
}
return output_size;
}

static inline size_t latin1_to_utf8_avx512_branch(__m512i input, char *utf8_output) {


// We take 32 zero-extended Latin1 characters and we write between 32 and 64 UTF-8 bytes.
// Returns the number of bytes written.
static simdutf_really_inline size_t latin1_to_utf8_avx512_vec2(__m512i in, char *utf8_output, __mmask32 is2byte) {
const __m512i twobytes = _mm512_ternarylogic_epi32(
_mm512_slli_epi16(in, 8), _mm512_srli_epi16(in, 6),
_mm512_set1_epi16(0x3f3f), 0xa8); // (A|B)&C
in = _mm512_mask_add_epi16(in, is2byte, twobytes,
_mm512_set1_epi16(int16_t(0x80c0)));
const __m512i cmpmask = _mm512_set1_epi16(0x0800);
// in >= cmpmask, always true for low bytes in 2-byte word.
// , always false for high bytes in ASCII 2-byte word.
// , always true for high bytes in UTF-8 2-byte word.
const __mmask64 smoosh = _mm512_cmp_epu8_mask(in, cmpmask, _MM_CMPINT_NLT);
const __m512i out = _mm512_maskz_compress_epi8(smoosh, in);
_mm512_storeu_epi8(utf8_output, out);
return 32 + _mm_popcnt_u32(_cvtmask32_u32(is2byte));
}

static simdutf_really_inline size_t latin1_to_utf8_avx512_branch( __m512i input, char *utf8_output) {
__mmask64 nonascii = _mm512_movepi8_mask(input);
size_t nonascii_count = (size_t)count_ones(nonascii);
if(nonascii_count > 0){
return latin1_to_utf8_avx512_vec(input, 64, utf8_output, 0);
if(nonascii) {
__mmask32 nonascii0 = (__mmask32)nonascii;
__mmask32 nonascii1 = (__mmask32)_kshiftri_mask64(nonascii,32);
const __m256i h0 = _mm512_castsi512_si256(input);
const __m256i h1 = _mm512_extracti64x4_epi64(input, 1);
const __m512i input0 = _mm512_cvtepu8_epi16(h0);
const __m512i input1 = _mm512_cvtepu8_epi16(h1);
size_t written0 = latin1_to_utf8_avx512_vec2(input0, utf8_output, nonascii0);
utf8_output += written0;
size_t written1 = latin1_to_utf8_avx512_vec2(input1, utf8_output, nonascii1);
return written0 + written1;
} else {
_mm512_storeu_si512(utf8_output, input);
return 64 + nonascii_count;}
return 64;
}
}

size_t latin1_to_utf8_avx512_start(const char *buf, size_t len, char *utf8_output) {
Expand Down