diff options
author | yazevnul <yazevnul@yandex-team.ru> | 2022-02-10 16:46:46 +0300 |
---|---|---|
committer | Daniil Cherednik <dcherednik@yandex-team.ru> | 2022-02-10 16:46:46 +0300 |
commit | 8cbc307de0221f84c80c42dcbe07d40727537e2c (patch) | |
tree | 625d5a673015d1df891e051033e9fcde5c7be4e5 /contrib/libs/base64/neon64 | |
parent | 30d1ef3941e0dc835be7609de5ebee66958f215a (diff) | |
download | ydb-8cbc307de0221f84c80c42dcbe07d40727537e2c.tar.gz |
Restoring authorship annotation for <yazevnul@yandex-team.ru>. Commit 1 of 2.
Diffstat (limited to 'contrib/libs/base64/neon64')
-rw-r--r-- | contrib/libs/base64/neon64/codec_neon64.c | 196 | ||||
-rw-r--r-- | contrib/libs/base64/neon64/codecs.h | 70 | ||||
-rw-r--r-- | contrib/libs/base64/neon64/dec_head.c | 58 | ||||
-rw-r--r-- | contrib/libs/base64/neon64/dec_neon.c | 154 | ||||
-rw-r--r-- | contrib/libs/base64/neon64/dec_tail.c | 130 | ||||
-rw-r--r-- | contrib/libs/base64/neon64/dec_uint64.c | 136 | ||||
-rw-r--r-- | contrib/libs/base64/neon64/enc_head.c | 46 | ||||
-rw-r--r-- | contrib/libs/base64/neon64/enc_neon.c | 74 | ||||
-rw-r--r-- | contrib/libs/base64/neon64/enc_tail.c | 56 | ||||
-rw-r--r-- | contrib/libs/base64/neon64/enc_uint64.c | 56 | ||||
-rw-r--r-- | contrib/libs/base64/neon64/lib.c | 244 | ||||
-rw-r--r-- | contrib/libs/base64/neon64/libbase64.h | 178 | ||||
-rw-r--r-- | contrib/libs/base64/neon64/ya.make | 36 |
13 files changed, 717 insertions, 717 deletions
diff --git a/contrib/libs/base64/neon64/codec_neon64.c b/contrib/libs/base64/neon64/codec_neon64.c index 7a352c3adf..0a954a6712 100644 --- a/contrib/libs/base64/neon64/codec_neon64.c +++ b/contrib/libs/base64/neon64/codec_neon64.c @@ -1,98 +1,98 @@ -#if (defined(__ARM_NEON) && !defined(__ARM_NEON__)) -#define __ARM_NEON__ -#endif - -#include <stdint.h> -#include <stddef.h> -#include <stdlib.h> -#ifdef __ARM_NEON__ -#include <arm_neon.h> -#endif - -#include "libbase64.h" -#include "codecs.h" - -#if (defined(__aarch64__) && defined(__ARM_NEON__)) - -#define CMPGT(s,n) vcgtq_u8((s), vdupq_n_u8(n)) -#define CMPEQ(s,n) vceqq_u8((s), vdupq_n_u8(n)) -#define REPLACE(s,n) vandq_u8((s), vdupq_n_u8(n)) -#define RANGE(s,a,b) vandq_u8(vcgeq_u8((s), vdupq_n_u8(a)), vcleq_u8((s), vdupq_n_u8(b))) - -// With this transposed encoding table, we can use -// a 64-byte lookup to do the encoding. -// Read the table top to bottom, left to right. -static const char *neon64_base64_table_enc_transposed = -{ - "AQgw" - "BRhx" - "CSiy" - "DTjz" - "EUk0" - "FVl1" - "GWm2" - "HXn3" - "IYo4" - "JZp5" - "Kaq6" - "Lbr7" - "Mcs8" - "Ndt9" - "Oeu+" - "Pfv/" -}; -#endif - -// Stride size is so large on these NEON 64-bit functions -// (48 bytes encode, 64 bytes decode) that we inline the -// uint64 codec to stay performant on smaller inputs. - -void -neon64_base64_stream_encode - ( struct neon64_base64_state *state - , const char *src - , size_t srclen - , char *out - , size_t *outlen - ) -{ -#if (defined(__aarch64__) && defined(__ARM_NEON__)) - uint8x16x4_t tbl_enc = vld4q_u8((uint8_t const*)neon64_base64_table_enc_transposed); - - #include "enc_head.c" - #include "enc_neon.c" - #include "enc_uint64.c" - #include "enc_tail.c" -#else - (void)state; - (void)src; - (void)srclen; - (void)out; - (void)outlen; - abort(); -#endif -} - -int -neon64_base64_stream_decode - ( struct neon64_base64_state *state - , const char *src - , size_t srclen - , char *out - , size_t *outlen - ) -{ -#if (defined(__aarch64__) && defined(__ARM_NEON__)) - #include "dec_head.c" - #include "dec_neon.c" - #include "dec_uint64.c" - #include "dec_tail.c" -#else - (void)state; - (void)src; - (void)srclen; - (void)out; - (void)outlen; - abort(); -#endif -} +#if (defined(__ARM_NEON) && !defined(__ARM_NEON__)) +#define __ARM_NEON__ +#endif + +#include <stdint.h> +#include <stddef.h> +#include <stdlib.h> +#ifdef __ARM_NEON__ +#include <arm_neon.h> +#endif + +#include "libbase64.h" +#include "codecs.h" + +#if (defined(__aarch64__) && defined(__ARM_NEON__)) + +#define CMPGT(s,n) vcgtq_u8((s), vdupq_n_u8(n)) +#define CMPEQ(s,n) vceqq_u8((s), vdupq_n_u8(n)) +#define REPLACE(s,n) vandq_u8((s), vdupq_n_u8(n)) +#define RANGE(s,a,b) vandq_u8(vcgeq_u8((s), vdupq_n_u8(a)), vcleq_u8((s), vdupq_n_u8(b))) + +// With this transposed encoding table, we can use +// a 64-byte lookup to do the encoding. +// Read the table top to bottom, left to right. +static const char *neon64_base64_table_enc_transposed = +{ + "AQgw" + "BRhx" + "CSiy" + "DTjz" + "EUk0" + "FVl1" + "GWm2" + "HXn3" + "IYo4" + "JZp5" + "Kaq6" + "Lbr7" + "Mcs8" + "Ndt9" + "Oeu+" + "Pfv/" +}; +#endif + +// Stride size is so large on these NEON 64-bit functions +// (48 bytes encode, 64 bytes decode) that we inline the +// uint64 codec to stay performant on smaller inputs. + +void +neon64_base64_stream_encode + ( struct neon64_base64_state *state + , const char *src + , size_t srclen + , char *out + , size_t *outlen + ) +{ +#if (defined(__aarch64__) && defined(__ARM_NEON__)) + uint8x16x4_t tbl_enc = vld4q_u8((uint8_t const*)neon64_base64_table_enc_transposed); + + #include "enc_head.c" + #include "enc_neon.c" + #include "enc_uint64.c" + #include "enc_tail.c" +#else + (void)state; + (void)src; + (void)srclen; + (void)out; + (void)outlen; + abort(); +#endif +} + +int +neon64_base64_stream_decode + ( struct neon64_base64_state *state + , const char *src + , size_t srclen + , char *out + , size_t *outlen + ) +{ +#if (defined(__aarch64__) && defined(__ARM_NEON__)) + #include "dec_head.c" + #include "dec_neon.c" + #include "dec_uint64.c" + #include "dec_tail.c" +#else + (void)state; + (void)src; + (void)srclen; + (void)out; + (void)outlen; + abort(); +#endif +} diff --git a/contrib/libs/base64/neon64/codecs.h b/contrib/libs/base64/neon64/codecs.h index d97269a4a5..0b8138c5bb 100644 --- a/contrib/libs/base64/neon64/codecs.h +++ b/contrib/libs/base64/neon64/codecs.h @@ -1,35 +1,35 @@ -#pragma once - -// Define machine endianness. This is for GCC: -#if (__BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) - #define BASE64_NEON64_LITTLE_ENDIAN 1 -#else - #define BASE64_NEON64_LITTLE_ENDIAN 0 -#endif - -// This is for Clang: -#ifdef __LITTLE_ENDIAN__ - #define BASE64_NEON64_LITTLE_ENDIAN 1 -#endif - -#ifdef __BIG_ENDIAN__ - #define BASE64_NEON64_LITTLE_ENDIAN 0 -#endif - -// Endian conversion functions -#if BASE64_NEON64_LITTLE_ENDIAN - #define cpu_to_be32(x) __builtin_bswap32(x) - #define cpu_to_be64(x) __builtin_bswap64(x) - #define be32_to_cpu(x) __builtin_bswap32(x) - #define be64_to_cpu(x) __builtin_bswap64(x) -#else - #define cpu_to_be32(x) (x) - #define cpu_to_be64(x) (x) - #define be32_to_cpu(x) (x) - #define be64_to_cpu(x) (x) -#endif - -// These tables are used by all codecs -// for fallback plain encoding/decoding: -extern const uint8_t neon64_base64_table_enc[]; -extern const uint8_t neon64_base64_table_dec[]; +#pragma once + +// Define machine endianness. This is for GCC: +#if (__BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) + #define BASE64_NEON64_LITTLE_ENDIAN 1 +#else + #define BASE64_NEON64_LITTLE_ENDIAN 0 +#endif + +// This is for Clang: +#ifdef __LITTLE_ENDIAN__ + #define BASE64_NEON64_LITTLE_ENDIAN 1 +#endif + +#ifdef __BIG_ENDIAN__ + #define BASE64_NEON64_LITTLE_ENDIAN 0 +#endif + +// Endian conversion functions +#if BASE64_NEON64_LITTLE_ENDIAN + #define cpu_to_be32(x) __builtin_bswap32(x) + #define cpu_to_be64(x) __builtin_bswap64(x) + #define be32_to_cpu(x) __builtin_bswap32(x) + #define be64_to_cpu(x) __builtin_bswap64(x) +#else + #define cpu_to_be32(x) (x) + #define cpu_to_be64(x) (x) + #define be32_to_cpu(x) (x) + #define be64_to_cpu(x) (x) +#endif + +// These tables are used by all codecs +// for fallback plain encoding/decoding: +extern const uint8_t neon64_base64_table_enc[]; +extern const uint8_t neon64_base64_table_dec[]; diff --git a/contrib/libs/base64/neon64/dec_head.c b/contrib/libs/base64/neon64/dec_head.c index 10f0def5be..89decb1be7 100644 --- a/contrib/libs/base64/neon64/dec_head.c +++ b/contrib/libs/base64/neon64/dec_head.c @@ -1,29 +1,29 @@ -int ret = 0; -const uint8_t *c = (const uint8_t *)src; -uint8_t *o = (uint8_t *)out; -uint8_t q; - -// Use local temporaries to avoid cache thrashing: -size_t outl = 0; -struct neon64_base64_state st; -st.eof = state->eof; -st.bytes = state->bytes; -st.carry = state->carry; - -// If we previously saw an EOF or an invalid character, bail out: -if (st.eof) { - *outlen = 0; - return 0; -} - -// Turn four 6-bit numbers into three bytes: -// out[0] = 11111122 -// out[1] = 22223333 -// out[2] = 33444444 - -// Duff's device again: -switch (st.bytes) -{ - for (;;) - { - case 0: +int ret = 0; +const uint8_t *c = (const uint8_t *)src; +uint8_t *o = (uint8_t *)out; +uint8_t q; + +// Use local temporaries to avoid cache thrashing: +size_t outl = 0; +struct neon64_base64_state st; +st.eof = state->eof; +st.bytes = state->bytes; +st.carry = state->carry; + +// If we previously saw an EOF or an invalid character, bail out: +if (st.eof) { + *outlen = 0; + return 0; +} + +// Turn four 6-bit numbers into three bytes: +// out[0] = 11111122 +// out[1] = 22223333 +// out[2] = 33444444 + +// Duff's device again: +switch (st.bytes) +{ + for (;;) + { + case 0: diff --git a/contrib/libs/base64/neon64/dec_neon.c b/contrib/libs/base64/neon64/dec_neon.c index 713d8ca9a4..30d846a916 100644 --- a/contrib/libs/base64/neon64/dec_neon.c +++ b/contrib/libs/base64/neon64/dec_neon.c @@ -1,77 +1,77 @@ -// If we have NEON support, pick off 64 bytes at a time for as long as we can. -// Unlike the SSE codecs, we don't write trailing zero bytes to output, so we -// don't need to check if we have enough remaining input to cover them: -while (srclen >= 64) -{ - uint8x16x4_t set1, set2, set3, set4, set5, set6, set7, delta; - uint8x16x3_t dec; - - // Load 64 bytes and deinterleave: - uint8x16x4_t str = vld4q_u8((uint8_t *)c); - - // The input consists of six character sets in the Base64 alphabet, - // which we need to map back to the 6-bit values they represent. - // There are three ranges, two singles, and then there's the rest. - // - // # From To Add Characters - // 1 [43] [62] +19 + - // 2 [47] [63] +16 / - // 3 [48..57] [52..61] +4 0..9 - // 4 [65..90] [0..25] -65 A..Z - // 5 [97..122] [26..51] -71 a..z - // (6) Everything else => invalid input - - // Benchmarking on the Raspberry Pi 2B and Clang shows that looping - // generates slightly faster code than explicit unrolling: - for (int i = 0; i < 4; i++) { - set1.val[i] = CMPEQ(str.val[i], '+'); - set2.val[i] = CMPEQ(str.val[i], '/'); - set3.val[i] = RANGE(str.val[i], '0', '9'); - set4.val[i] = RANGE(str.val[i], 'A', 'Z'); - set5.val[i] = RANGE(str.val[i], 'a', 'z'); - set6.val[i] = CMPEQ(str.val[i], '-'); - set7.val[i] = CMPEQ(str.val[i], '_'); - - delta.val[i] = REPLACE(set1.val[i], 19); - delta.val[i] = vorrq_u8(delta.val[i], REPLACE(set2.val[i], 16)); - delta.val[i] = vorrq_u8(delta.val[i], REPLACE(set3.val[i], 4)); - delta.val[i] = vorrq_u8(delta.val[i], REPLACE(set4.val[i], -65)); - delta.val[i] = vorrq_u8(delta.val[i], REPLACE(set5.val[i], -71)); - delta.val[i] = vorrq_u8(delta.val[i], REPLACE(set6.val[i], 17)); - delta.val[i] = vorrq_u8(delta.val[i], REPLACE(set7.val[i], -32)); - } - - // Check for invalid input: if any of the delta values are zero, - // fall back on bytewise code to do error checking and reporting: - uint8x16_t classified = CMPEQ(delta.val[0], 0); - classified = vorrq_u8(classified, CMPEQ(delta.val[1], 0)); - classified = vorrq_u8(classified, CMPEQ(delta.val[2], 0)); - classified = vorrq_u8(classified, CMPEQ(delta.val[3], 0)); - - // Extract both 32-bit halves; check that all bits are zero: - if (vgetq_lane_u32((uint32x4_t)classified, 0) != 0 - || vgetq_lane_u32((uint32x4_t)classified, 1) != 0 - || vgetq_lane_u32((uint32x4_t)classified, 2) != 0 - || vgetq_lane_u32((uint32x4_t)classified, 3) != 0) { - break; - } - - // Now simply add the delta values to the input: - str.val[0] = vaddq_u8(str.val[0], delta.val[0]); - str.val[1] = vaddq_u8(str.val[1], delta.val[1]); - str.val[2] = vaddq_u8(str.val[2], delta.val[2]); - str.val[3] = vaddq_u8(str.val[3], delta.val[3]); - - // Compress four bytes into three: - dec.val[0] = vshlq_n_u8(str.val[0], 2) | vshrq_n_u8(str.val[1], 4); - dec.val[1] = vshlq_n_u8(str.val[1], 4) | vshrq_n_u8(str.val[2], 2); - dec.val[2] = vshlq_n_u8(str.val[2], 6) | str.val[3]; - - // Interleave and store decoded result: - vst3q_u8((uint8_t *)o, dec); - - c += 64; - o += 48; - outl += 48; - srclen -= 64; -} +// If we have NEON support, pick off 64 bytes at a time for as long as we can. +// Unlike the SSE codecs, we don't write trailing zero bytes to output, so we +// don't need to check if we have enough remaining input to cover them: +while (srclen >= 64) +{ + uint8x16x4_t set1, set2, set3, set4, set5, set6, set7, delta; + uint8x16x3_t dec; + + // Load 64 bytes and deinterleave: + uint8x16x4_t str = vld4q_u8((uint8_t *)c); + + // The input consists of six character sets in the Base64 alphabet, + // which we need to map back to the 6-bit values they represent. + // There are three ranges, two singles, and then there's the rest. + // + // # From To Add Characters + // 1 [43] [62] +19 + + // 2 [47] [63] +16 / + // 3 [48..57] [52..61] +4 0..9 + // 4 [65..90] [0..25] -65 A..Z + // 5 [97..122] [26..51] -71 a..z + // (6) Everything else => invalid input + + // Benchmarking on the Raspberry Pi 2B and Clang shows that looping + // generates slightly faster code than explicit unrolling: + for (int i = 0; i < 4; i++) { + set1.val[i] = CMPEQ(str.val[i], '+'); + set2.val[i] = CMPEQ(str.val[i], '/'); + set3.val[i] = RANGE(str.val[i], '0', '9'); + set4.val[i] = RANGE(str.val[i], 'A', 'Z'); + set5.val[i] = RANGE(str.val[i], 'a', 'z'); + set6.val[i] = CMPEQ(str.val[i], '-'); + set7.val[i] = CMPEQ(str.val[i], '_'); + + delta.val[i] = REPLACE(set1.val[i], 19); + delta.val[i] = vorrq_u8(delta.val[i], REPLACE(set2.val[i], 16)); + delta.val[i] = vorrq_u8(delta.val[i], REPLACE(set3.val[i], 4)); + delta.val[i] = vorrq_u8(delta.val[i], REPLACE(set4.val[i], -65)); + delta.val[i] = vorrq_u8(delta.val[i], REPLACE(set5.val[i], -71)); + delta.val[i] = vorrq_u8(delta.val[i], REPLACE(set6.val[i], 17)); + delta.val[i] = vorrq_u8(delta.val[i], REPLACE(set7.val[i], -32)); + } + + // Check for invalid input: if any of the delta values are zero, + // fall back on bytewise code to do error checking and reporting: + uint8x16_t classified = CMPEQ(delta.val[0], 0); + classified = vorrq_u8(classified, CMPEQ(delta.val[1], 0)); + classified = vorrq_u8(classified, CMPEQ(delta.val[2], 0)); + classified = vorrq_u8(classified, CMPEQ(delta.val[3], 0)); + + // Extract both 32-bit halves; check that all bits are zero: + if (vgetq_lane_u32((uint32x4_t)classified, 0) != 0 + || vgetq_lane_u32((uint32x4_t)classified, 1) != 0 + || vgetq_lane_u32((uint32x4_t)classified, 2) != 0 + || vgetq_lane_u32((uint32x4_t)classified, 3) != 0) { + break; + } + + // Now simply add the delta values to the input: + str.val[0] = vaddq_u8(str.val[0], delta.val[0]); + str.val[1] = vaddq_u8(str.val[1], delta.val[1]); + str.val[2] = vaddq_u8(str.val[2], delta.val[2]); + str.val[3] = vaddq_u8(str.val[3], delta.val[3]); + + // Compress four bytes into three: + dec.val[0] = vshlq_n_u8(str.val[0], 2) | vshrq_n_u8(str.val[1], 4); + dec.val[1] = vshlq_n_u8(str.val[1], 4) | vshrq_n_u8(str.val[2], 2); + dec.val[2] = vshlq_n_u8(str.val[2], 6) | str.val[3]; + + // Interleave and store decoded result: + vst3q_u8((uint8_t *)o, dec); + + c += 64; + o += 48; + outl += 48; + srclen -= 64; +} diff --git a/contrib/libs/base64/neon64/dec_tail.c b/contrib/libs/base64/neon64/dec_tail.c index 90a3eec2d3..b530aa62cc 100644 --- a/contrib/libs/base64/neon64/dec_tail.c +++ b/contrib/libs/base64/neon64/dec_tail.c @@ -1,65 +1,65 @@ - if (srclen-- == 0) { - ret = 1; - break; - } - if ((q = neon64_base64_table_dec[*c++]) >= 254) { - st.eof = 1; - // Treat character '=' as invalid for byte 0: - break; - } - st.carry = q << 2; - st.bytes++; - - case 1: if (srclen-- == 0) { - ret = 1; - break; - } - if ((q = neon64_base64_table_dec[*c++]) >= 254) { - st.eof = 1; - // Treat character '=' as invalid for byte 1: - break; - } - *o++ = st.carry | (q >> 4); - st.carry = q << 4; - st.bytes++; - outl++; - - case 2: if (srclen-- == 0) { - ret = 1; - break; - } - if ((q = neon64_base64_table_dec[*c++]) >= 254) { - st.eof = 1; - // When q == 254, the input char is '='. Return 1 and EOF. - // Technically, should check if next byte is also '=', but never mind. - // When q == 255, the input char is invalid. Return 0 and EOF. - ret = (q == 254) ? 1 : 0; - break; - } - *o++ = st.carry | (q >> 2); - st.carry = q << 6; - st.bytes++; - outl++; - - case 3: if (srclen-- == 0) { - ret = 1; - break; - } - if ((q = neon64_base64_table_dec[*c++]) >= 254) { - st.eof = 1; - // When q == 254, the input char is '='. Return 1 and EOF. - // When q == 255, the input char is invalid. Return 0 and EOF. - ret = (q == 254) ? 1 : 0; - break; - } - *o++ = st.carry | q; - st.carry = 0; - st.bytes = 0; - outl++; - } -} -state->eof = st.eof; -state->bytes = st.bytes; -state->carry = st.carry; -*outlen = outl; -return ret; + if (srclen-- == 0) { + ret = 1; + break; + } + if ((q = neon64_base64_table_dec[*c++]) >= 254) { + st.eof = 1; + // Treat character '=' as invalid for byte 0: + break; + } + st.carry = q << 2; + st.bytes++; + + case 1: if (srclen-- == 0) { + ret = 1; + break; + } + if ((q = neon64_base64_table_dec[*c++]) >= 254) { + st.eof = 1; + // Treat character '=' as invalid for byte 1: + break; + } + *o++ = st.carry | (q >> 4); + st.carry = q << 4; + st.bytes++; + outl++; + + case 2: if (srclen-- == 0) { + ret = 1; + break; + } + if ((q = neon64_base64_table_dec[*c++]) >= 254) { + st.eof = 1; + // When q == 254, the input char is '='. Return 1 and EOF. + // Technically, should check if next byte is also '=', but never mind. + // When q == 255, the input char is invalid. Return 0 and EOF. + ret = (q == 254) ? 1 : 0; + break; + } + *o++ = st.carry | (q >> 2); + st.carry = q << 6; + st.bytes++; + outl++; + + case 3: if (srclen-- == 0) { + ret = 1; + break; + } + if ((q = neon64_base64_table_dec[*c++]) >= 254) { + st.eof = 1; + // When q == 254, the input char is '='. Return 1 and EOF. + // When q == 255, the input char is invalid. Return 0 and EOF. + ret = (q == 254) ? 1 : 0; + break; + } + *o++ = st.carry | q; + st.carry = 0; + st.bytes = 0; + outl++; + } +} +state->eof = st.eof; +state->bytes = st.bytes; +state->carry = st.carry; +*outlen = outl; +return ret; diff --git a/contrib/libs/base64/neon64/dec_uint64.c b/contrib/libs/base64/neon64/dec_uint64.c index 0ea70a3424..90d134af38 100644 --- a/contrib/libs/base64/neon64/dec_uint64.c +++ b/contrib/libs/base64/neon64/dec_uint64.c @@ -1,68 +1,68 @@ -// If we have native uint64's, pick off 8 bytes at a time for as long as we -// can, but make sure that we quit before seeing any == markers at the end of -// the string. Also, because we write two zeroes at the end of the output, -// ensure that there are at least 3 valid bytes of input data remaining to -// close the gap. 8 + 2 + 3 = 13 bytes: -while (srclen >= 13) -{ - uint64_t str, res, dec; - - // Load string: - str = *(uint64_t *)c; - - // Shuffle bytes to 64-bit bigendian: - str = cpu_to_be64(str); - - // Lookup each byte in the decoding table; if we encounter any - // "invalid" values, fall back on the bytewise code: - if ((dec = neon64_base64_table_dec[str >> 56]) > 63) { - break; - } - res = dec << 58; - - if ((dec = neon64_base64_table_dec[(str >> 48) & 0xFF]) > 63) { - break; - } - res |= dec << 52; - - if ((dec = neon64_base64_table_dec[(str >> 40) & 0xFF]) > 63) { - break; - } - res |= dec << 46; - - if ((dec = neon64_base64_table_dec[(str >> 32) & 0xFF]) > 63) { - break; - } - res |= dec << 40; - - if ((dec = neon64_base64_table_dec[(str >> 24) & 0xFF]) > 63) { - break; - } - res |= dec << 34; - - if ((dec = neon64_base64_table_dec[(str >> 16) & 0xFF]) > 63) { - break; - } - res |= dec << 28; - - if ((dec = neon64_base64_table_dec[(str >> 8) & 0xFF]) > 63) { - break; - } - res |= dec << 22; - - if ((dec = neon64_base64_table_dec[str & 0xFF]) > 63) { - break; - } - res |= dec << 16; - - // Reshuffle and repack into 6-byte output format: - res = be64_to_cpu(res); - - // Store back: - *(uint64_t *)o = res; - - c += 8; - o += 6; - outl += 6; - srclen -= 8; -} +// If we have native uint64's, pick off 8 bytes at a time for as long as we +// can, but make sure that we quit before seeing any == markers at the end of +// the string. Also, because we write two zeroes at the end of the output, +// ensure that there are at least 3 valid bytes of input data remaining to +// close the gap. 8 + 2 + 3 = 13 bytes: +while (srclen >= 13) +{ + uint64_t str, res, dec; + + // Load string: + str = *(uint64_t *)c; + + // Shuffle bytes to 64-bit bigendian: + str = cpu_to_be64(str); + + // Lookup each byte in the decoding table; if we encounter any + // "invalid" values, fall back on the bytewise code: + if ((dec = neon64_base64_table_dec[str >> 56]) > 63) { + break; + } + res = dec << 58; + + if ((dec = neon64_base64_table_dec[(str >> 48) & 0xFF]) > 63) { + break; + } + res |= dec << 52; + + if ((dec = neon64_base64_table_dec[(str >> 40) & 0xFF]) > 63) { + break; + } + res |= dec << 46; + + if ((dec = neon64_base64_table_dec[(str >> 32) & 0xFF]) > 63) { + break; + } + res |= dec << 40; + + if ((dec = neon64_base64_table_dec[(str >> 24) & 0xFF]) > 63) { + break; + } + res |= dec << 34; + + if ((dec = neon64_base64_table_dec[(str >> 16) & 0xFF]) > 63) { + break; + } + res |= dec << 28; + + if ((dec = neon64_base64_table_dec[(str >> 8) & 0xFF]) > 63) { + break; + } + res |= dec << 22; + + if ((dec = neon64_base64_table_dec[str & 0xFF]) > 63) { + break; + } + res |= dec << 16; + + // Reshuffle and repack into 6-byte output format: + res = be64_to_cpu(res); + + // Store back: + *(uint64_t *)o = res; + + c += 8; + o += 6; + outl += 6; + srclen -= 8; +} diff --git a/contrib/libs/base64/neon64/enc_head.c b/contrib/libs/base64/neon64/enc_head.c index 92d5ed50b4..9f8147d608 100644 --- a/contrib/libs/base64/neon64/enc_head.c +++ b/contrib/libs/base64/neon64/enc_head.c @@ -1,23 +1,23 @@ -// Assume that *out is large enough to contain the output. -// Theoretically it should be 4/3 the length of src. -const uint8_t *c = (const uint8_t *)src; -uint8_t *o = (uint8_t *)out; - -// Use local temporaries to avoid cache thrashing: -size_t outl = 0; -struct neon64_base64_state st; -st.bytes = state->bytes; -st.carry = state->carry; - -// Turn three bytes into four 6-bit numbers: -// in[0] = 00111111 -// in[1] = 00112222 -// in[2] = 00222233 -// in[3] = 00333333 - -// Duff's device, a for() loop inside a switch() statement. Legal! -switch (st.bytes) -{ - for (;;) - { - case 0: +// Assume that *out is large enough to contain the output. +// Theoretically it should be 4/3 the length of src. +const uint8_t *c = (const uint8_t *)src; +uint8_t *o = (uint8_t *)out; + +// Use local temporaries to avoid cache thrashing: +size_t outl = 0; +struct neon64_base64_state st; +st.bytes = state->bytes; +st.carry = state->carry; + +// Turn three bytes into four 6-bit numbers: +// in[0] = 00111111 +// in[1] = 00112222 +// in[2] = 00222233 +// in[3] = 00333333 + +// Duff's device, a for() loop inside a switch() statement. Legal! +switch (st.bytes) +{ + for (;;) + { + case 0: diff --git a/contrib/libs/base64/neon64/enc_neon.c b/contrib/libs/base64/neon64/enc_neon.c index 9cf28a11f8..2ba5a561e9 100644 --- a/contrib/libs/base64/neon64/enc_neon.c +++ b/contrib/libs/base64/neon64/enc_neon.c @@ -1,37 +1,37 @@ -// If we have ARM NEON support, pick off 48 bytes at a time: -while (srclen >= 48) -{ - uint8x16x3_t str; - uint8x16x4_t res; - - // Load 48 bytes and deinterleave: - str = vld3q_u8((uint8_t *)c); - - // Divide bits of three input bytes over four output bytes: - res.val[0] = vshrq_n_u8(str.val[0], 2); - res.val[1] = vshrq_n_u8(str.val[1], 4) | vshlq_n_u8(str.val[0], 4); - res.val[2] = vshrq_n_u8(str.val[2], 6) | vshlq_n_u8(str.val[1], 2); - res.val[3] = str.val[2]; - - // Clear top two bits: - res.val[0] &= vdupq_n_u8(0x3F); - res.val[1] &= vdupq_n_u8(0x3F); - res.val[2] &= vdupq_n_u8(0x3F); - res.val[3] &= vdupq_n_u8(0x3F); - - // The bits have now been shifted to the right locations; - // translate their values 0..63 to the Base64 alphabet. - // Use a 64-byte table lookup: - res.val[0] = vqtbl4q_u8(tbl_enc, res.val[0]); - res.val[1] = vqtbl4q_u8(tbl_enc, res.val[1]); - res.val[2] = vqtbl4q_u8(tbl_enc, res.val[2]); - res.val[3] = vqtbl4q_u8(tbl_enc, res.val[3]); - - // Interleave and store result: - vst4q_u8((uint8_t *)o, res); - - c += 48; // 3 * 16 bytes of input - o += 64; // 4 * 16 bytes of output - outl += 64; - srclen -= 48; -} +// If we have ARM NEON support, pick off 48 bytes at a time: +while (srclen >= 48) +{ + uint8x16x3_t str; + uint8x16x4_t res; + + // Load 48 bytes and deinterleave: + str = vld3q_u8((uint8_t *)c); + + // Divide bits of three input bytes over four output bytes: + res.val[0] = vshrq_n_u8(str.val[0], 2); + res.val[1] = vshrq_n_u8(str.val[1], 4) | vshlq_n_u8(str.val[0], 4); + res.val[2] = vshrq_n_u8(str.val[2], 6) | vshlq_n_u8(str.val[1], 2); + res.val[3] = str.val[2]; + + // Clear top two bits: + res.val[0] &= vdupq_n_u8(0x3F); + res.val[1] &= vdupq_n_u8(0x3F); + res.val[2] &= vdupq_n_u8(0x3F); + res.val[3] &= vdupq_n_u8(0x3F); + + // The bits have now been shifted to the right locations; + // translate their values 0..63 to the Base64 alphabet. + // Use a 64-byte table lookup: + res.val[0] = vqtbl4q_u8(tbl_enc, res.val[0]); + res.val[1] = vqtbl4q_u8(tbl_enc, res.val[1]); + res.val[2] = vqtbl4q_u8(tbl_enc, res.val[2]); + res.val[3] = vqtbl4q_u8(tbl_enc, res.val[3]); + + // Interleave and store result: + vst4q_u8((uint8_t *)o, res); + + c += 48; // 3 * 16 bytes of input + o += 64; // 4 * 16 bytes of output + outl += 64; + srclen -= 48; +} diff --git a/contrib/libs/base64/neon64/enc_tail.c b/contrib/libs/base64/neon64/enc_tail.c index 3c97141902..86b49fa6af 100644 --- a/contrib/libs/base64/neon64/enc_tail.c +++ b/contrib/libs/base64/neon64/enc_tail.c @@ -1,28 +1,28 @@ - if (srclen-- == 0) { - break; - } - *o++ = neon64_base64_table_enc[*c >> 2]; - st.carry = (*c++ << 4) & 0x30; - st.bytes++; - outl += 1; - - case 1: if (srclen-- == 0) { - break; - } - *o++ = neon64_base64_table_enc[st.carry | (*c >> 4)]; - st.carry = (*c++ << 2) & 0x3C; - st.bytes++; - outl += 1; - - case 2: if (srclen-- == 0) { - break; - } - *o++ = neon64_base64_table_enc[st.carry | (*c >> 6)]; - *o++ = neon64_base64_table_enc[*c++ & 0x3F]; - st.bytes = 0; - outl += 2; - } -} -state->bytes = st.bytes; -state->carry = st.carry; -*outlen = outl; + if (srclen-- == 0) { + break; + } + *o++ = neon64_base64_table_enc[*c >> 2]; + st.carry = (*c++ << 4) & 0x30; + st.bytes++; + outl += 1; + + case 1: if (srclen-- == 0) { + break; + } + *o++ = neon64_base64_table_enc[st.carry | (*c >> 4)]; + st.carry = (*c++ << 2) & 0x3C; + st.bytes++; + outl += 1; + + case 2: if (srclen-- == 0) { + break; + } + *o++ = neon64_base64_table_enc[st.carry | (*c >> 6)]; + *o++ = neon64_base64_table_enc[*c++ & 0x3F]; + st.bytes = 0; + outl += 2; + } +} +state->bytes = st.bytes; +state->carry = st.carry; +*outlen = outl; diff --git a/contrib/libs/base64/neon64/enc_uint64.c b/contrib/libs/base64/neon64/enc_uint64.c index 2db3c0be97..15e4436fe3 100644 --- a/contrib/libs/base64/neon64/enc_uint64.c +++ b/contrib/libs/base64/neon64/enc_uint64.c @@ -1,28 +1,28 @@ -// If we have 64-bit ints, pick off 6 bytes at a time for as long as we can, -// but ensure that there are at least 8 bytes available to avoid segfaulting: -while (srclen >= 8) -{ - // Load string: - uint64_t str = *(uint64_t *)c; - - // Reorder to 64-bit big-endian, if not already in that format. The - // workset must be in big-endian, otherwise the shifted bits do not - // carry over properly among adjacent bytes: - str = cpu_to_be64(str); - - // Shift input by 6 bytes each round and mask in only the lower 6 bits; - // look up the character in the Base64 encoding table and write it to - // the output location: - *o++ = neon64_base64_table_enc[(str >> 58) & 0x3F]; - *o++ = neon64_base64_table_enc[(str >> 52) & 0x3F]; - *o++ = neon64_base64_table_enc[(str >> 46) & 0x3F]; - *o++ = neon64_base64_table_enc[(str >> 40) & 0x3F]; - *o++ = neon64_base64_table_enc[(str >> 34) & 0x3F]; - *o++ = neon64_base64_table_enc[(str >> 28) & 0x3F]; - *o++ = neon64_base64_table_enc[(str >> 22) & 0x3F]; - *o++ = neon64_base64_table_enc[(str >> 16) & 0x3F]; - - c += 6; // 6 bytes of input - outl += 8; // 8 bytes of output - srclen -= 6; -} +// If we have 64-bit ints, pick off 6 bytes at a time for as long as we can, +// but ensure that there are at least 8 bytes available to avoid segfaulting: +while (srclen >= 8) +{ + // Load string: + uint64_t str = *(uint64_t *)c; + + // Reorder to 64-bit big-endian, if not already in that format. The + // workset must be in big-endian, otherwise the shifted bits do not + // carry over properly among adjacent bytes: + str = cpu_to_be64(str); + + // Shift input by 6 bytes each round and mask in only the lower 6 bits; + // look up the character in the Base64 encoding table and write it to + // the output location: + *o++ = neon64_base64_table_enc[(str >> 58) & 0x3F]; + *o++ = neon64_base64_table_enc[(str >> 52) & 0x3F]; + *o++ = neon64_base64_table_enc[(str >> 46) & 0x3F]; + *o++ = neon64_base64_table_enc[(str >> 40) & 0x3F]; + *o++ = neon64_base64_table_enc[(str >> 34) & 0x3F]; + *o++ = neon64_base64_table_enc[(str >> 28) & 0x3F]; + *o++ = neon64_base64_table_enc[(str >> 22) & 0x3F]; + *o++ = neon64_base64_table_enc[(str >> 16) & 0x3F]; + + c += 6; // 6 bytes of input + outl += 8; // 8 bytes of output + srclen -= 6; +} diff --git a/contrib/libs/base64/neon64/lib.c b/contrib/libs/base64/neon64/lib.c index de5d49e711..63d514476e 100644 --- a/contrib/libs/base64/neon64/lib.c +++ b/contrib/libs/base64/neon64/lib.c @@ -1,122 +1,122 @@ -#include <stdint.h> -#include <stddef.h> -#include <stdlib.h> - -#include "libbase64.h" -#include "codecs.h" - -const uint8_t -neon64_base64_table_enc[] = - "ABCDEFGHIJKLMNOPQRSTUVWXYZ" - "abcdefghijklmnopqrstuvwxyz" - "0123456789+/"; - -// In the lookup table below, note that the value for '=' (character 61) is -// 254, not 255. This character is used for in-band signaling of the end of -// the datastream, and we will use that later. The characters A-Z, a-z, 0-9 -// and + / are mapped to their "decoded" values. The other bytes all map to -// the value 255, which flags them as "invalid input". - -const uint8_t -neon64_base64_table_dec[] = -{ - 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, // 0..15 - 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, // 16..31 - 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 62, 254, 62, 255, 63, // 32..47 - 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 255, 255, 255, 254, 255, 255, // 48..63 - 255, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, // 64..79 - 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 255, 255, 255, 255, 63, // 80..95 - 255, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, // 96..111 - 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 255, 255, 255, 255, 255, // 112..127 - 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, // 128..143 - 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, - 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, - 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, - 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, - 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, - 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, - 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, -}; - -void -neon64_base64_stream_encode_init (struct neon64_base64_state *state) -{ - state->eof = 0; - state->bytes = 0; - state->carry = 0; -} - -void -neon64_base64_stream_encode_final - ( struct neon64_base64_state *state - , char *out - , size_t *outlen - ) -{ - uint8_t *o = (uint8_t *)out; - - if (state->bytes == 1) { - *o++ = neon64_base64_table_enc[state->carry]; - *o++ = '='; - *o++ = '='; - *outlen = 3; - return; - } - if (state->bytes == 2) { - *o++ = neon64_base64_table_enc[state->carry]; - *o++ = '='; - *outlen = 2; - return; - } - *outlen = 0; -} - -void -neon64_base64_stream_decode_init (struct neon64_base64_state *state) -{ - state->eof = 0; - state->bytes = 0; - state->carry = 0; -} - -void -neon64_base64_encode - ( const char *src - , size_t srclen - , char *out - , size_t *outlen - ) -{ - size_t s; - size_t t; - struct neon64_base64_state state; - - // Init the stream reader: - neon64_base64_stream_encode_init(&state); - - // Feed the whole string to the stream reader: - neon64_base64_stream_encode(&state, src, srclen, out, &s); - - // Finalize the stream by writing trailer if any: - neon64_base64_stream_encode_final(&state, out + s, &t); - - // Final output length is stream length plus tail: - *outlen = s + t; -} - -int -neon64_base64_decode - ( const char *src - , size_t srclen - , char *out - , size_t *outlen - ) -{ - struct neon64_base64_state state; - - // Init the stream reader: - neon64_base64_stream_decode_init(&state); - - // Feed the whole string to the stream reader: - return neon64_base64_stream_decode(&state, src, srclen, out, outlen); -} +#include <stdint.h> +#include <stddef.h> +#include <stdlib.h> + +#include "libbase64.h" +#include "codecs.h" + +const uint8_t +neon64_base64_table_enc[] = + "ABCDEFGHIJKLMNOPQRSTUVWXYZ" + "abcdefghijklmnopqrstuvwxyz" + "0123456789+/"; + +// In the lookup table below, note that the value for '=' (character 61) is +// 254, not 255. This character is used for in-band signaling of the end of +// the datastream, and we will use that later. The characters A-Z, a-z, 0-9 +// and + / are mapped to their "decoded" values. The other bytes all map to +// the value 255, which flags them as "invalid input". + +const uint8_t +neon64_base64_table_dec[] = +{ + 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, // 0..15 + 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, // 16..31 + 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 62, 254, 62, 255, 63, // 32..47 + 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 255, 255, 255, 254, 255, 255, // 48..63 + 255, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, // 64..79 + 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 255, 255, 255, 255, 63, // 80..95 + 255, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, // 96..111 + 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 255, 255, 255, 255, 255, // 112..127 + 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, // 128..143 + 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, + 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, + 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, + 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, + 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, + 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, + 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, +}; + +void +neon64_base64_stream_encode_init (struct neon64_base64_state *state) +{ + state->eof = 0; + state->bytes = 0; + state->carry = 0; +} + +void +neon64_base64_stream_encode_final + ( struct neon64_base64_state *state + , char *out + , size_t *outlen + ) +{ + uint8_t *o = (uint8_t *)out; + + if (state->bytes == 1) { + *o++ = neon64_base64_table_enc[state->carry]; + *o++ = '='; + *o++ = '='; + *outlen = 3; + return; + } + if (state->bytes == 2) { + *o++ = neon64_base64_table_enc[state->carry]; + *o++ = '='; + *outlen = 2; + return; + } + *outlen = 0; +} + +void +neon64_base64_stream_decode_init (struct neon64_base64_state *state) +{ + state->eof = 0; + state->bytes = 0; + state->carry = 0; +} + +void +neon64_base64_encode + ( const char *src + , size_t srclen + , char *out + , size_t *outlen + ) +{ + size_t s; + size_t t; + struct neon64_base64_state state; + + // Init the stream reader: + neon64_base64_stream_encode_init(&state); + + // Feed the whole string to the stream reader: + neon64_base64_stream_encode(&state, src, srclen, out, &s); + + // Finalize the stream by writing trailer if any: + neon64_base64_stream_encode_final(&state, out + s, &t); + + // Final output length is stream length plus tail: + *outlen = s + t; +} + +int +neon64_base64_decode + ( const char *src + , size_t srclen + , char *out + , size_t *outlen + ) +{ + struct neon64_base64_state state; + + // Init the stream reader: + neon64_base64_stream_decode_init(&state); + + // Feed the whole string to the stream reader: + return neon64_base64_stream_decode(&state, src, srclen, out, outlen); +} diff --git a/contrib/libs/base64/neon64/libbase64.h b/contrib/libs/base64/neon64/libbase64.h index a7224965d7..7566eb7545 100644 --- a/contrib/libs/base64/neon64/libbase64.h +++ b/contrib/libs/base64/neon64/libbase64.h @@ -1,89 +1,89 @@ -#pragma once - -#ifdef __cplusplus -extern "C" { -#endif - -struct neon64_base64_state { - int eof; - int bytes; - unsigned char carry; -}; - -/* Wrapper function to encode a plain string of given length. Output is written - * to *out without trailing zero. Output length in bytes is written to *outlen. - * The buffer in `out` has been allocated by the caller and is at least 4/3 the - * size of the input. See above for `flags`; set to 0 for default operation: */ -void neon64_base64_encode - ( const char *src - , size_t srclen - , char *out - , size_t *outlen - ) ; - -/* Call this before calling base64_stream_encode() to init the state. See above - * for `flags`; set to 0 for default operation: */ -void neon64_base64_stream_encode_init - ( struct neon64_base64_state *state - ) ; - -/* Encodes the block of data of given length at `src`, into the buffer at - * `out`. Caller is responsible for allocating a large enough out-buffer; it - * must be at least 4/3 the size of the in-buffer, but take some margin. Places - * the number of new bytes written into `outlen` (which is set to zero when the - * function starts). Does not zero-terminate or finalize the output. */ -void neon64_base64_stream_encode - ( struct neon64_base64_state *state - , const char *src - , size_t srclen - , char *out - , size_t *outlen - ) ; - -/* Finalizes the output begun by previous calls to `base64_stream_encode()`. - * Adds the required end-of-stream markers if appropriate. `outlen` is modified - * and will contain the number of new bytes written at `out` (which will quite - * often be zero). */ -void neon64_base64_stream_encode_final - ( struct neon64_base64_state *state - , char *out - , size_t *outlen - ) ; - -/* Wrapper function to decode a plain string of given length. Output is written - * to *out without trailing zero. Output length in bytes is written to *outlen. - * The buffer in `out` has been allocated by the caller and is at least 3/4 the - * size of the input. See above for `flags`, set to 0 for default operation: */ -int neon64_base64_decode - ( const char *src - , size_t srclen - , char *out - , size_t *outlen - ) ; - -/* Call this before calling base64_stream_decode() to init the state. See above - * for `flags`; set to 0 for default operation: */ -void neon64_base64_stream_decode_init - ( struct neon64_base64_state *state - ) ; - -/* Decodes the block of data of given length at `src`, into the buffer at - * `out`. Caller is responsible for allocating a large enough out-buffer; it - * must be at least 3/4 the size of the in-buffer, but take some margin. Places - * the number of new bytes written into `outlen` (which is set to zero when the - * function starts). Does not zero-terminate the output. Returns 1 if all is - * well, and 0 if a decoding error was found, such as an invalid character. - * Returns -1 if the chosen codec is not included in the current build. Used by - * the test harness to check whether a codec is available for testing. */ -int neon64_base64_stream_decode - ( struct neon64_base64_state *state - , const char *src - , size_t srclen - , char *out - , size_t *outlen - ) ; - -#ifdef __cplusplus -} -#endif - +#pragma once + +#ifdef __cplusplus +extern "C" { +#endif + +struct neon64_base64_state { + int eof; + int bytes; + unsigned char carry; +}; + +/* Wrapper function to encode a plain string of given length. Output is written + * to *out without trailing zero. Output length in bytes is written to *outlen. + * The buffer in `out` has been allocated by the caller and is at least 4/3 the + * size of the input. See above for `flags`; set to 0 for default operation: */ +void neon64_base64_encode + ( const char *src + , size_t srclen + , char *out + , size_t *outlen + ) ; + +/* Call this before calling base64_stream_encode() to init the state. See above + * for `flags`; set to 0 for default operation: */ +void neon64_base64_stream_encode_init + ( struct neon64_base64_state *state + ) ; + +/* Encodes the block of data of given length at `src`, into the buffer at + * `out`. Caller is responsible for allocating a large enough out-buffer; it + * must be at least 4/3 the size of the in-buffer, but take some margin. Places + * the number of new bytes written into `outlen` (which is set to zero when the + * function starts). Does not zero-terminate or finalize the output. */ +void neon64_base64_stream_encode + ( struct neon64_base64_state *state + , const char *src + , size_t srclen + , char *out + , size_t *outlen + ) ; + +/* Finalizes the output begun by previous calls to `base64_stream_encode()`. + * Adds the required end-of-stream markers if appropriate. `outlen` is modified + * and will contain the number of new bytes written at `out` (which will quite + * often be zero). */ +void neon64_base64_stream_encode_final + ( struct neon64_base64_state *state + , char *out + , size_t *outlen + ) ; + +/* Wrapper function to decode a plain string of given length. Output is written + * to *out without trailing zero. Output length in bytes is written to *outlen. + * The buffer in `out` has been allocated by the caller and is at least 3/4 the + * size of the input. See above for `flags`, set to 0 for default operation: */ +int neon64_base64_decode + ( const char *src + , size_t srclen + , char *out + , size_t *outlen + ) ; + +/* Call this before calling base64_stream_decode() to init the state. See above + * for `flags`; set to 0 for default operation: */ +void neon64_base64_stream_decode_init + ( struct neon64_base64_state *state + ) ; + +/* Decodes the block of data of given length at `src`, into the buffer at + * `out`. Caller is responsible for allocating a large enough out-buffer; it + * must be at least 3/4 the size of the in-buffer, but take some margin. Places + * the number of new bytes written into `outlen` (which is set to zero when the + * function starts). Does not zero-terminate the output. Returns 1 if all is + * well, and 0 if a decoding error was found, such as an invalid character. + * Returns -1 if the chosen codec is not included in the current build. Used by + * the test harness to check whether a codec is available for testing. */ +int neon64_base64_stream_decode + ( struct neon64_base64_state *state + , const char *src + , size_t srclen + , char *out + , size_t *outlen + ) ; + +#ifdef __cplusplus +} +#endif + diff --git a/contrib/libs/base64/neon64/ya.make b/contrib/libs/base64/neon64/ya.make index 82d821b4ee..2b171abf95 100644 --- a/contrib/libs/base64/neon64/ya.make +++ b/contrib/libs/base64/neon64/ya.make @@ -1,11 +1,11 @@ -OWNER( - yazevnul +OWNER( + yazevnul g:contrib g:cpp-contrib -) - -LIBRARY() - +) + +LIBRARY() + LICENSE( BSD-2-Clause AND MIT @@ -13,20 +13,20 @@ LICENSE( LICENSE_TEXTS(.yandex_meta/licenses.list.txt) -NO_UTIL() - -SRCS( - codec_neon64.c - lib.c -) - -IF (ARCH_AARCH64 OR ARCH_ARM64) +NO_UTIL() + +SRCS( + codec_neon64.c + lib.c +) + +IF (ARCH_AARCH64 OR ARCH_ARM64) IF (OS_LINUX OR OS_DARWIN OR OS_ANDROID) CONLYFLAGS( -march=armv8-a -std=c11 ) - ENDIF() -ENDIF() - -END() + ENDIF() +ENDIF() + +END() |