third_party/utf8_range: support arm neon (#18126)
Protobuf uses utf8_range library for utf8 string validation.
Currently, only SSE implementation is integrated.
This patch adapts utf8_range Neon implementation to protobuf.
Closes #18126
COPYBARA_INTEGRATE_REVIEW=https://github.com/protocolbuffers/protobuf/pull/18126 from cyb70289:utf8-neon 5edbcc2692
PiperOrigin-RevId: 680711032
pull/18514/head
Yibo Cai
2 months ago
committed by
Copybara-Service
parent
cd6df36de5
commit
d83ad15cd1
6 changed files with 410 additions and 275 deletions
@ -0,0 +1,117 @@ |
|||||||
|
#include <arm_neon.h> |
||||||
|
|
||||||
|
/* This code is almost the same as SSE implementation, please reference |
||||||
|
* utf8-range-sse.inc for detailed explanation. |
||||||
|
* The only difference is the range adjustment step. NEON code is more |
||||||
|
* straightforward. |
||||||
|
*/ |
||||||
|
|
||||||
|
static FORCE_INLINE_ATTR inline size_t utf8_range_ValidateUTF8Simd( |
||||||
|
const char* data, const char* end, int return_position) { |
||||||
|
const uint8x16_t first_len_tbl = { |
||||||
|
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 2, 3, |
||||||
|
}; |
||||||
|
const uint8x16_t first_range_tbl = { |
||||||
|
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 8, 8, 8, 8, |
||||||
|
}; |
||||||
|
const uint8x16_t range_min_tbl = { |
||||||
|
0x00, 0x80, 0x80, 0x80, 0xA0, 0x80, 0x90, 0x80, |
||||||
|
0xC2, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, |
||||||
|
}; |
||||||
|
const uint8x16_t range_max_tbl = { |
||||||
|
0x7F, 0xBF, 0xBF, 0xBF, 0xBF, 0x9F, 0xBF, 0x8F, |
||||||
|
0xF4, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, |
||||||
|
}; |
||||||
|
/* Range adjustment in NEON uint8x16x2 table. Note that lanes are interleaved |
||||||
|
* in register. The table below is plotted vertically to ease understanding. |
||||||
|
* The 1st column is for E0~EF, 2nd column for F0~FF. |
||||||
|
*/ |
||||||
|
// clang-format off |
||||||
|
const uint8_t range_adjust_tbl_data[] = { |
||||||
|
/* index -> 0~15 16~31 <- index */ |
||||||
|
/* E0 -> */ 2, 3, /* <- F0 */ |
||||||
|
0, 0, |
||||||
|
0, 0, |
||||||
|
0, 0, |
||||||
|
0, 4, /* <- F4 */ |
||||||
|
0, 0, |
||||||
|
0, 0, |
||||||
|
0, 0, |
||||||
|
0, 0, |
||||||
|
0, 0, |
||||||
|
0, 0, |
||||||
|
0, 0, |
||||||
|
0, 0, |
||||||
|
/* ED -> */ 3, 0, |
||||||
|
0, 0, |
||||||
|
0, 0, |
||||||
|
}; |
||||||
|
// clang-format on |
||||||
|
const uint8x16x2_t range_adjust_tbl = vld2q_u8(range_adjust_tbl_data); |
||||||
|
|
||||||
|
const uint8x16_t const_1 = vdupq_n_u8(1); |
||||||
|
const uint8x16_t const_2 = vdupq_n_u8(2); |
||||||
|
const uint8x16_t const_e0 = vdupq_n_u8(0xE0); |
||||||
|
|
||||||
|
uint8x16_t prev_input = vdupq_n_u8(0); |
||||||
|
uint8x16_t prev_first_len = vdupq_n_u8(0); |
||||||
|
uint8x16_t error = vdupq_n_u8(0); |
||||||
|
|
||||||
|
const char* const data_original = data; |
||||||
|
while (end - data >= 16) { |
||||||
|
const uint8x16_t input = vld1q_u8((const uint8_t*)data); |
||||||
|
|
||||||
|
const uint8x16_t high_nibbles = vshrq_n_u8(input, 4); |
||||||
|
|
||||||
|
const uint8x16_t first_len = vqtbl1q_u8(first_len_tbl, high_nibbles); |
||||||
|
|
||||||
|
uint8x16_t range = vqtbl1q_u8(first_range_tbl, high_nibbles); |
||||||
|
|
||||||
|
range = vorrq_u8(range, vextq_u8(prev_first_len, first_len, 15)); |
||||||
|
|
||||||
|
uint8x16_t shift2 = vextq_u8(prev_first_len, first_len, 14); |
||||||
|
shift2 = vqsubq_u8(shift2, const_1); |
||||||
|
range = vorrq_u8(range, shift2); |
||||||
|
|
||||||
|
uint8x16_t shift3 = vextq_u8(prev_first_len, first_len, 13); |
||||||
|
shift3 = vqsubq_u8(shift3, const_2); |
||||||
|
range = vorrq_u8(range, shift3); |
||||||
|
|
||||||
|
uint8x16_t shift1 = vextq_u8(prev_input, input, 15); |
||||||
|
shift1 = vsubq_u8(shift1, const_e0); |
||||||
|
range = vaddq_u8(range, vqtbl2q_u8(range_adjust_tbl, shift1)); |
||||||
|
|
||||||
|
const uint8x16_t min_range = vqtbl1q_u8(range_min_tbl, range); |
||||||
|
const uint8x16_t max_range = vqtbl1q_u8(range_max_tbl, range); |
||||||
|
|
||||||
|
if (return_position) { |
||||||
|
error = vcltq_u8(input, min_range); |
||||||
|
error = vorrq_u8(error, vcgtq_u8(input, max_range)); |
||||||
|
if (vmaxvq_u32(vreinterpretq_u32_u8(error))) { |
||||||
|
break; |
||||||
|
} |
||||||
|
} else { |
||||||
|
error = vorrq_u8(error, vcltq_u8(input, min_range)); |
||||||
|
error = vorrq_u8(error, vcgtq_u8(input, max_range)); |
||||||
|
} |
||||||
|
|
||||||
|
prev_input = input; |
||||||
|
prev_first_len = first_len; |
||||||
|
|
||||||
|
data += 16; |
||||||
|
} |
||||||
|
|
||||||
|
if (return_position && data == data_original) { |
||||||
|
return utf8_range_ValidateUTF8Naive(data, end, return_position); |
||||||
|
} |
||||||
|
const int32_t prev = vgetq_lane_s32(vreinterpretq_s32_u8(prev_input), 3); |
||||||
|
data -= utf8_range_CodepointSkipBackwards(prev); |
||||||
|
if (return_position) { |
||||||
|
return (data - data_original) + |
||||||
|
utf8_range_ValidateUTF8Naive(data, end, return_position); |
||||||
|
} |
||||||
|
if (vmaxvq_u32(vreinterpretq_u32_u8(error))) { |
||||||
|
return 0; |
||||||
|
} |
||||||
|
return utf8_range_ValidateUTF8Naive(data, end, return_position); |
||||||
|
} |
||||||
@ -0,0 +1,273 @@ |
|||||||
|
#include <emmintrin.h> |
||||||
|
#include <smmintrin.h> |
||||||
|
#include <tmmintrin.h> |
||||||
|
|
||||||
|
static FORCE_INLINE_ATTR inline size_t utf8_range_ValidateUTF8Simd( |
||||||
|
const char* data, const char* end, int return_position) { |
||||||
|
/* This code checks that utf-8 ranges are structurally valid 16 bytes at once |
||||||
|
* using superscalar instructions. |
||||||
|
* The mapping between ranges of codepoint and their corresponding utf-8 |
||||||
|
* sequences is below. |
||||||
|
*/ |
||||||
|
|
||||||
|
/* |
||||||
|
* U+0000...U+007F 00...7F |
||||||
|
* U+0080...U+07FF C2...DF 80...BF |
||||||
|
* U+0800...U+0FFF E0 A0...BF 80...BF |
||||||
|
* U+1000...U+CFFF E1...EC 80...BF 80...BF |
||||||
|
* U+D000...U+D7FF ED 80...9F 80...BF |
||||||
|
* U+E000...U+FFFF EE...EF 80...BF 80...BF |
||||||
|
* U+10000...U+3FFFF F0 90...BF 80...BF 80...BF |
||||||
|
* U+40000...U+FFFFF F1...F3 80...BF 80...BF 80...BF |
||||||
|
* U+100000...U+10FFFF F4 80...8F 80...BF 80...BF |
||||||
|
*/ |
||||||
|
|
||||||
|
/* First we compute the type for each byte, as given by the table below. |
||||||
|
* This type will be used as an index later on. |
||||||
|
*/ |
||||||
|
|
||||||
|
/* |
||||||
|
* Index Min Max Byte Type |
||||||
|
* 0 00 7F Single byte sequence |
||||||
|
* 1,2,3 80 BF Second, third and fourth byte for many of the sequences. |
||||||
|
* 4 A0 BF Second byte after E0 |
||||||
|
* 5 80 9F Second byte after ED |
||||||
|
* 6 90 BF Second byte after F0 |
||||||
|
* 7 80 8F Second byte after F4 |
||||||
|
* 8 C2 F4 First non ASCII byte |
||||||
|
* 9..15 7F 80 Invalid byte |
||||||
|
*/ |
||||||
|
|
||||||
|
/* After the first step we compute the index for all bytes, then we permute |
||||||
|
the bytes according to their indices to check the ranges from the range |
||||||
|
table. |
||||||
|
* The range for a given type can be found in the range_min_table and |
||||||
|
range_max_table, the range for type/index X is in range_min_table[X] ... |
||||||
|
range_max_table[X]. |
||||||
|
*/ |
||||||
|
|
||||||
|
/* Algorithm: |
||||||
|
* Put index zero to all bytes. |
||||||
|
* Find all non ASCII characters, give them index 8. |
||||||
|
* For each tail byte in a codepoint sequence, give it an index corresponding |
||||||
|
to the 1 based index from the end. |
||||||
|
* If the first byte of the codepoint is in the [C0...DF] range, we write |
||||||
|
index 1 in the following byte. |
||||||
|
* If the first byte of the codepoint is in the range [E0...EF], we write |
||||||
|
indices 2 and 1 in the next two bytes. |
||||||
|
* If the first byte of the codepoint is in the range [F0...FF] we write |
||||||
|
indices 3,2,1 into the next three bytes. |
||||||
|
* For finding the number of bytes we need to look at high nibbles (4 bits) |
||||||
|
and do the lookup from the table, it can be done with shift by 4 + shuffle |
||||||
|
instructions. We call it `first_len`. |
||||||
|
* Then we shift first_len by 8 bits to get the indices of the 2nd bytes. |
||||||
|
* Saturating sub 1 and shift by 8 bits to get the indices of the 3rd bytes. |
||||||
|
* Again to get the indices of the 4th bytes. |
||||||
|
* Take OR of all that 4 values and check within range. |
||||||
|
*/ |
||||||
|
/* For example: |
||||||
|
* input C3 80 68 E2 80 20 A6 F0 A0 80 AC 20 F0 93 80 80 |
||||||
|
* first_len 1 0 0 2 0 0 0 3 0 0 0 0 3 0 0 0 |
||||||
|
* 1st byte 8 0 0 8 0 0 0 8 0 0 0 0 8 0 0 0 |
||||||
|
* 2nd byte 0 1 0 0 2 0 0 0 3 0 0 0 0 3 0 0 // Shift + sub |
||||||
|
* 3rd byte 0 0 0 0 0 1 0 0 0 2 0 0 0 0 2 0 // Shift + sub |
||||||
|
* 4th byte 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 // Shift + sub |
||||||
|
* Index 8 1 0 8 2 1 0 8 3 2 1 0 8 3 2 1 // OR of results |
||||||
|
*/ |
||||||
|
|
||||||
|
/* Checking for errors: |
||||||
|
* Error checking is done by looking up the high nibble (4 bits) of each byte |
||||||
|
against an error checking table. |
||||||
|
* Because the lookup value for the second byte depends of the value of the |
||||||
|
first byte in codepoint, we use saturated operations to adjust the index. |
||||||
|
* Specifically we need to add 2 for E0, 3 for ED, 3 for F0 and 4 for F4 to |
||||||
|
match the correct index. |
||||||
|
* If we subtract from all bytes EF then EO -> 241, ED -> 254, F0 -> 1, |
||||||
|
F4 -> 5 |
||||||
|
* Do saturating sub 240, then E0 -> 1, ED -> 14 and we can do lookup to |
||||||
|
match the adjustment |
||||||
|
* Add saturating 112, then F0 -> 113, F4 -> 117, all that were > 16 will |
||||||
|
be more 128 and lookup in ef_fe_table will return 0 but for F0 |
||||||
|
and F4 it will be 4 and 5 accordingly |
||||||
|
*/ |
||||||
|
/* |
||||||
|
* Then just check the appropriate ranges with greater/smaller equal |
||||||
|
instructions. Check tail with a naive algorithm. |
||||||
|
* To save from previous 16 byte checks we just align previous_first_len to |
||||||
|
get correct continuations of the codepoints. |
||||||
|
*/ |
||||||
|
|
||||||
|
/* |
||||||
|
* Map high nibble of "First Byte" to legal character length minus 1 |
||||||
|
* 0x00 ~ 0xBF --> 0 |
||||||
|
* 0xC0 ~ 0xDF --> 1 |
||||||
|
* 0xE0 ~ 0xEF --> 2 |
||||||
|
* 0xF0 ~ 0xFF --> 3 |
||||||
|
*/ |
||||||
|
const __m128i first_len_table = |
||||||
|
_mm_setr_epi8(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 2, 3); |
||||||
|
|
||||||
|
/* Map "First Byte" to 8-th item of range table (0xC2 ~ 0xF4) */ |
||||||
|
const __m128i first_range_table = |
||||||
|
_mm_setr_epi8(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 8, 8, 8, 8); |
||||||
|
|
||||||
|
/* |
||||||
|
* Range table, map range index to min and max values |
||||||
|
*/ |
||||||
|
const __m128i range_min_table = |
||||||
|
_mm_setr_epi8(0x00, 0x80, 0x80, 0x80, 0xA0, 0x80, 0x90, 0x80, 0xC2, 0x7F, |
||||||
|
0x7F, 0x7F, 0x7F, 0x7F, 0x7F, 0x7F); |
||||||
|
|
||||||
|
const __m128i range_max_table = |
||||||
|
_mm_setr_epi8(0x7F, 0xBF, 0xBF, 0xBF, 0xBF, 0x9F, 0xBF, 0x8F, 0xF4, 0x80, |
||||||
|
0x80, 0x80, 0x80, 0x80, 0x80, 0x80); |
||||||
|
|
||||||
|
/* |
||||||
|
* Tables for fast handling of four special First Bytes(E0,ED,F0,F4), after |
||||||
|
* which the Second Byte are not 80~BF. It contains "range index adjustment". |
||||||
|
* +------------+---------------+------------------+----------------+ |
||||||
|
* | First Byte | original range| range adjustment | adjusted range | |
||||||
|
* +------------+---------------+------------------+----------------+ |
||||||
|
* | E0 | 2 | 2 | 4 | |
||||||
|
* +------------+---------------+------------------+----------------+ |
||||||
|
* | ED | 2 | 3 | 5 | |
||||||
|
* +------------+---------------+------------------+----------------+ |
||||||
|
* | F0 | 3 | 3 | 6 | |
||||||
|
* +------------+---------------+------------------+----------------+ |
||||||
|
* | F4 | 4 | 4 | 8 | |
||||||
|
* +------------+---------------+------------------+----------------+ |
||||||
|
*/ |
||||||
|
|
||||||
|
/* df_ee_table[1] -> E0, df_ee_table[14] -> ED as ED - E0 = 13 */ |
||||||
|
// The values represent the adjustment in the Range Index table for a correct |
||||||
|
// index. |
||||||
|
const __m128i df_ee_table = |
||||||
|
_mm_setr_epi8(0, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 3, 0); |
||||||
|
|
||||||
|
/* ef_fe_table[1] -> F0, ef_fe_table[5] -> F4, F4 - F0 = 4 */ |
||||||
|
// The values represent the adjustment in the Range Index table for a correct |
||||||
|
// index. |
||||||
|
const __m128i ef_fe_table = |
||||||
|
_mm_setr_epi8(0, 3, 0, 0, 0, 4, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0); |
||||||
|
|
||||||
|
__m128i prev_input = _mm_set1_epi8(0); |
||||||
|
__m128i prev_first_len = _mm_set1_epi8(0); |
||||||
|
__m128i error = _mm_set1_epi8(0); |
||||||
|
|
||||||
|
// Save buffer start address for later use |
||||||
|
const char* const data_original = data; |
||||||
|
while (end - data >= 16) { |
||||||
|
const __m128i input = _mm_loadu_si128((const __m128i*)(data)); |
||||||
|
|
||||||
|
/* high_nibbles = input >> 4 */ |
||||||
|
const __m128i high_nibbles = |
||||||
|
_mm_and_si128(_mm_srli_epi16(input, 4), _mm_set1_epi8(0x0F)); |
||||||
|
|
||||||
|
/* first_len = legal character length minus 1 */ |
||||||
|
/* 0 for 00~7F, 1 for C0~DF, 2 for E0~EF, 3 for F0~FF */ |
||||||
|
/* first_len = first_len_table[high_nibbles] */ |
||||||
|
__m128i first_len = _mm_shuffle_epi8(first_len_table, high_nibbles); |
||||||
|
|
||||||
|
/* First Byte: set range index to 8 for bytes within 0xC0 ~ 0xFF */ |
||||||
|
/* range = first_range_table[high_nibbles] */ |
||||||
|
__m128i range = _mm_shuffle_epi8(first_range_table, high_nibbles); |
||||||
|
|
||||||
|
/* Second Byte: set range index to first_len */ |
||||||
|
/* 0 for 00~7F, 1 for C0~DF, 2 for E0~EF, 3 for F0~FF */ |
||||||
|
/* range |= (first_len, prev_first_len) << 1 byte */ |
||||||
|
range = _mm_or_si128(range, _mm_alignr_epi8(first_len, prev_first_len, 15)); |
||||||
|
|
||||||
|
/* Third Byte: set range index to saturate_sub(first_len, 1) */ |
||||||
|
/* 0 for 00~7F, 0 for C0~DF, 1 for E0~EF, 2 for F0~FF */ |
||||||
|
__m128i tmp1; |
||||||
|
__m128i tmp2; |
||||||
|
/* tmp1 = saturate_sub(first_len, 1) */ |
||||||
|
tmp1 = _mm_subs_epu8(first_len, _mm_set1_epi8(1)); |
||||||
|
/* tmp2 = saturate_sub(prev_first_len, 1) */ |
||||||
|
tmp2 = _mm_subs_epu8(prev_first_len, _mm_set1_epi8(1)); |
||||||
|
/* range |= (tmp1, tmp2) << 2 bytes */ |
||||||
|
range = _mm_or_si128(range, _mm_alignr_epi8(tmp1, tmp2, 14)); |
||||||
|
|
||||||
|
/* Fourth Byte: set range index to saturate_sub(first_len, 2) */ |
||||||
|
/* 0 for 00~7F, 0 for C0~DF, 0 for E0~EF, 1 for F0~FF */ |
||||||
|
/* tmp1 = saturate_sub(first_len, 2) */ |
||||||
|
tmp1 = _mm_subs_epu8(first_len, _mm_set1_epi8(2)); |
||||||
|
/* tmp2 = saturate_sub(prev_first_len, 2) */ |
||||||
|
tmp2 = _mm_subs_epu8(prev_first_len, _mm_set1_epi8(2)); |
||||||
|
/* range |= (tmp1, tmp2) << 3 bytes */ |
||||||
|
range = _mm_or_si128(range, _mm_alignr_epi8(tmp1, tmp2, 13)); |
||||||
|
|
||||||
|
/* |
||||||
|
* Now we have below range indices calculated |
||||||
|
* Correct cases: |
||||||
|
* - 8 for C0~FF |
||||||
|
* - 3 for 1st byte after F0~FF |
||||||
|
* - 2 for 1st byte after E0~EF or 2nd byte after F0~FF |
||||||
|
* - 1 for 1st byte after C0~DF or 2nd byte after E0~EF or |
||||||
|
* 3rd byte after F0~FF |
||||||
|
* - 0 for others |
||||||
|
* Error cases: |
||||||
|
* >9 for non ascii First Byte overlapping |
||||||
|
* E.g., F1 80 C2 90 --> 8 3 10 2, where 10 indicates error |
||||||
|
*/ |
||||||
|
|
||||||
|
/* Adjust Second Byte range for special First Bytes(E0,ED,F0,F4) */ |
||||||
|
/* Overlaps lead to index 9~15, which are illegal in range table */ |
||||||
|
__m128i shift1; |
||||||
|
__m128i pos; |
||||||
|
__m128i range2; |
||||||
|
/* shift1 = (input, prev_input) << 1 byte */ |
||||||
|
shift1 = _mm_alignr_epi8(input, prev_input, 15); |
||||||
|
pos = _mm_sub_epi8(shift1, _mm_set1_epi8(0xEF)); |
||||||
|
/* |
||||||
|
* shift1: | EF F0 ... FE | FF 00 ... ... DE | DF E0 ... EE | |
||||||
|
* pos: | 0 1 15 | 16 17 239| 240 241 255| |
||||||
|
* pos-240: | 0 0 0 | 0 0 0 | 0 1 15 | |
||||||
|
* pos+112: | 112 113 127| >= 128 | >= 128 | |
||||||
|
*/ |
||||||
|
tmp1 = _mm_subs_epu8(pos, _mm_set1_epi8(-16)); |
||||||
|
range2 = _mm_shuffle_epi8(df_ee_table, tmp1); |
||||||
|
tmp2 = _mm_adds_epu8(pos, _mm_set1_epi8(112)); |
||||||
|
range2 = _mm_add_epi8(range2, _mm_shuffle_epi8(ef_fe_table, tmp2)); |
||||||
|
|
||||||
|
range = _mm_add_epi8(range, range2); |
||||||
|
|
||||||
|
/* Load min and max values per calculated range index */ |
||||||
|
__m128i min_range = _mm_shuffle_epi8(range_min_table, range); |
||||||
|
__m128i max_range = _mm_shuffle_epi8(range_max_table, range); |
||||||
|
|
||||||
|
/* Check value range */ |
||||||
|
if (return_position) { |
||||||
|
error = _mm_cmplt_epi8(input, min_range); |
||||||
|
error = _mm_or_si128(error, _mm_cmpgt_epi8(input, max_range)); |
||||||
|
/* 5% performance drop from this conditional branch */ |
||||||
|
if (!_mm_testz_si128(error, error)) { |
||||||
|
break; |
||||||
|
} |
||||||
|
} else { |
||||||
|
error = _mm_or_si128(error, _mm_cmplt_epi8(input, min_range)); |
||||||
|
error = _mm_or_si128(error, _mm_cmpgt_epi8(input, max_range)); |
||||||
|
} |
||||||
|
|
||||||
|
prev_input = input; |
||||||
|
prev_first_len = first_len; |
||||||
|
|
||||||
|
data += 16; |
||||||
|
} |
||||||
|
/* If we got to the end, we don't need to skip any bytes backwards */ |
||||||
|
if (return_position && data == data_original) { |
||||||
|
return utf8_range_ValidateUTF8Naive(data, end, return_position); |
||||||
|
} |
||||||
|
/* Find previous codepoint (not 80~BF) */ |
||||||
|
data -= utf8_range_CodepointSkipBackwards(_mm_extract_epi32(prev_input, 3)); |
||||||
|
if (return_position) { |
||||||
|
return (data - data_original) + |
||||||
|
utf8_range_ValidateUTF8Naive(data, end, return_position); |
||||||
|
} |
||||||
|
/* Test if there was any error */ |
||||||
|
if (!_mm_testz_si128(error, error)) { |
||||||
|
return 0; |
||||||
|
} |
||||||
|
/* Check the tail */ |
||||||
|
return utf8_range_ValidateUTF8Naive(data, end, return_position); |
||||||
|
} |
||||||
Loading…
Reference in new issue