ob64 0.1.0 → 0.5.0

data/vendor/libbase64/lib/arch/ssse3/dec_loop.c

@@ -0,0 +1,173 @@
+// 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
+//
+// We will use lookup tables for character validation and offset computation.
+// Remember that 0x2X and 0x0X are the same index for _mm_shuffle_epi8, this
+// allows to mask with 0x2F instead of 0x0F and thus save one constant
+// declaration (register and/or memory access).
+//
+// For offsets:
+// Perfect hash for lut = ((src >> 4) & 0x2F) + ((src == 0x2F) ? 0xFF : 0x00)
+// 0000 = garbage
+// 0001 = /
+// 0010 = +
+// 0011 = 0-9
+// 0100 = A-Z
+// 0101 = A-Z
+// 0110 = a-z
+// 0111 = a-z
+// 1000 >= garbage
+//
+// For validation, here's the table.
+// A character is valid if and only if the AND of the 2 lookups equals 0:
+//
+// hi \ lo              0000 0001 0010 0011 0100 0101 0110 0111 1000 1001 1010 1011 1100 1101 1110 1111
+//      LUT             0x15 0x11 0x11 0x11 0x11 0x11 0x11 0x11 0x11 0x11 0x13 0x1A 0x1B 0x1B 0x1B 0x1A
+//
+// 0000 0x10 char        NUL  SOH  STX  ETX  EOT  ENQ  ACK  BEL   BS   HT   LF   VT   FF   CR   SO   SI
+//           andlut     0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10
+//
+// 0001 0x10 char        DLE  DC1  DC2  DC3  DC4  NAK  SYN  ETB  CAN   EM  SUB  ESC   FS   GS   RS   US
+//           andlut     0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10
+//
+// 0010 0x01 char               !    "    #    $    %    &    '    (    )    *    +    ,    -    .    /
+//           andlut     0x01 0x01 0x01 0x01 0x01 0x01 0x01 0x01 0x01 0x01 0x01 0x00 0x01 0x01 0x01 0x00
+//
+// 0011 0x02 char          0    1    2    3    4    5    6    7    8    9    :    ;    <    =    >    ?
+//           andlut     0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x02 0x02 0x02 0x02 0x02 0x02
+//
+// 0100 0x04 char          @    A    B    C    D    E    F    G    H    I    J    K    L    M    N    O
+//           andlut     0x04 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00
+//
+// 0101 0x08 char          P    Q    R    S    T    U    V    W    X    Y    Z    [    \    ]    ^    _
+//           andlut     0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x08 0x08 0x08 0x08 0x08
+//
+// 0110 0x04 char          `    a    b    c    d    e    f    g    h    i    j    k    l    m    n    o
+//           andlut     0x04 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00
+// 0111 0x08 char          p    q    r    s    t    u    v    w    x    y    z    {    |    }    ~
+//           andlut     0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x08 0x08 0x08 0x08 0x08
+//
+// 1000 0x10 andlut     0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10
+// 1001 0x10 andlut     0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10
+// 1010 0x10 andlut     0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10
+// 1011 0x10 andlut     0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10
+// 1100 0x10 andlut     0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10
+// 1101 0x10 andlut     0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10
+// 1110 0x10 andlut     0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10
+// 1111 0x10 andlut     0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10 0x10
+
+static inline int
+dec_loop_ssse3_inner (const uint8_t **s, uint8_t **o, size_t *rounds)
+{
+	const __m128i lut_lo = _mm_setr_epi8(
+		0x15, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11,
+		0x11, 0x11, 0x13, 0x1A, 0x1B, 0x1B, 0x1B, 0x1A);
+
+	const __m128i lut_hi = _mm_setr_epi8(
+		0x10, 0x10, 0x01, 0x02, 0x04, 0x08, 0x04, 0x08,
+		0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10);
+
+	const __m128i lut_roll = _mm_setr_epi8(
+		0,  16,  19,   4, -65, -65, -71, -71,
+		0,   0,   0,   0,   0,   0,   0,   0);
+
+	const __m128i mask_2F = _mm_set1_epi8(0x2F);
+
+	// Load input:
+	__m128i str = _mm_loadu_si128((__m128i *) *s);
+
+	// Table lookups:
+	const __m128i hi_nibbles = _mm_and_si128(_mm_srli_epi32(str, 4), mask_2F);
+	const __m128i lo_nibbles = _mm_and_si128(str, mask_2F);
+	const __m128i hi         = _mm_shuffle_epi8(lut_hi, hi_nibbles);
+	const __m128i lo         = _mm_shuffle_epi8(lut_lo, lo_nibbles);
+
+	// Check for invalid input: if any "and" values from lo and hi are not
+	// zero, fall back on bytewise code to do error checking and reporting:
+	if (_mm_movemask_epi8(_mm_cmpgt_epi8(_mm_and_si128(lo, hi), _mm_setzero_si128())) != 0) {
+		return 0;
+	}
+
+	const __m128i eq_2F = _mm_cmpeq_epi8(str, mask_2F);
+	const __m128i roll  = _mm_shuffle_epi8(lut_roll, _mm_add_epi8(eq_2F, hi_nibbles));
+
+	// Now simply add the delta values to the input:
+	str = _mm_add_epi8(str, roll);
+
+	// Reshuffle the input to packed 12-byte output format:
+	str = dec_reshuffle(str);
+
+	// Store the output:
+	_mm_storeu_si128((__m128i *) *o, str);
+
+	*s += 16;
+	*o += 12;
+	*rounds -= 1;
+
+	return 1;
+}
+
+static inline void
+dec_loop_ssse3 (const uint8_t **s, size_t *slen, uint8_t **o, size_t *olen)
+{
+	if (*slen < 24) {
+		return;
+	}
+
+	// Process blocks of 16 bytes per round. Because 4 extra zero bytes are
+	// written after the output, ensure that there will be at least 8 bytes
+	// of input data left to cover the gap. (6 data bytes and up to two
+	// end-of-string markers.)
+	size_t rounds = (*slen - 8) / 16;
+
+	*slen -= rounds * 16;	// 16 bytes consumed per round
+	*olen += rounds * 12;	// 12 bytes produced per round
+
+	do {
+		if (rounds >= 8) {
+			if (dec_loop_ssse3_inner(s, o, &rounds) &&
+			    dec_loop_ssse3_inner(s, o, &rounds) &&
+			    dec_loop_ssse3_inner(s, o, &rounds) &&
+			    dec_loop_ssse3_inner(s, o, &rounds) &&
+			    dec_loop_ssse3_inner(s, o, &rounds) &&
+			    dec_loop_ssse3_inner(s, o, &rounds) &&
+			    dec_loop_ssse3_inner(s, o, &rounds) &&
+			    dec_loop_ssse3_inner(s, o, &rounds)) {
+				continue;
+			}
+			break;
+		}
+		if (rounds >= 4) {
+			if (dec_loop_ssse3_inner(s, o, &rounds) &&
+			    dec_loop_ssse3_inner(s, o, &rounds) &&
+			    dec_loop_ssse3_inner(s, o, &rounds) &&
+			    dec_loop_ssse3_inner(s, o, &rounds)) {
+				continue;
+			}
+			break;
+		}
+		if (rounds >= 2) {
+			if (dec_loop_ssse3_inner(s, o, &rounds) &&
+			    dec_loop_ssse3_inner(s, o, &rounds)) {
+				continue;
+			}
+			break;
+		}
+		dec_loop_ssse3_inner(s, o, &rounds);
+		break;
+
+	} while (rounds > 0);
+
+	// Adjust for any rounds that were skipped:
+	*slen += rounds * 16;
+	*olen -= rounds * 12;
+}

data/vendor/libbase64/lib/arch/ssse3/dec_reshuffle.c

@@ -0,0 +1,33 @@
+static inline __m128i
+dec_reshuffle (const __m128i in)
+{
+	// in, bits, upper case are most significant bits, lower case are least significant bits
+	// 00llllll 00kkkkLL 00jjKKKK 00JJJJJJ
+	// 00iiiiii 00hhhhII 00ggHHHH 00GGGGGG
+	// 00ffffff 00eeeeFF 00ddEEEE 00DDDDDD
+	// 00cccccc 00bbbbCC 00aaBBBB 00AAAAAA
+
+	const __m128i merge_ab_and_bc = _mm_maddubs_epi16(in, _mm_set1_epi32(0x01400140));
+	// 0000kkkk LLllllll 0000JJJJ JJjjKKKK
+	// 0000hhhh IIiiiiii 0000GGGG GGggHHHH
+	// 0000eeee FFffffff 0000DDDD DDddEEEE
+	// 0000bbbb CCcccccc 0000AAAA AAaaBBBB
+
+	const __m128i out = _mm_madd_epi16(merge_ab_and_bc, _mm_set1_epi32(0x00011000));
+	// 00000000 JJJJJJjj KKKKkkkk LLllllll
+	// 00000000 GGGGGGgg HHHHhhhh IIiiiiii
+	// 00000000 DDDDDDdd EEEEeeee FFffffff
+	// 00000000 AAAAAAaa BBBBbbbb CCcccccc
+
+	// Pack bytes together:
+	return  _mm_shuffle_epi8(out, _mm_setr_epi8(
+		 2,  1,  0,
+		 6,  5,  4,
+		10,  9,  8,
+		14, 13, 12,
+		-1, -1, -1, -1));
+	// 00000000 00000000 00000000 00000000
+	// LLllllll KKKKkkkk JJJJJJjj IIiiiiii
+	// HHHHhhhh GGGGGGgg FFffffff EEEEeeee
+	// DDDDDDdd CCcccccc BBBBbbbb AAAAAAaa
+}

data/vendor/libbase64/lib/arch/ssse3/enc_loop.c

@@ -0,0 +1,67 @@
+static inline void
+enc_loop_ssse3_inner (const uint8_t **s, uint8_t **o)
+{
+	// Load input:
+	__m128i str = _mm_loadu_si128((__m128i *) *s);
+
+	// Reshuffle:
+	str = enc_reshuffle(str);
+
+	// Translate reshuffled bytes to the Base64 alphabet:
+	str = enc_translate(str);
+
+	// Store:
+	_mm_storeu_si128((__m128i *) *o, str);
+
+	*s += 12;
+	*o += 16;
+}
+
+static inline void
+enc_loop_ssse3 (const uint8_t **s, size_t *slen, uint8_t **o, size_t *olen)
+{
+	if (*slen < 16) {
+		return;
+	}
+
+	// Process blocks of 12 bytes at a time. Because blocks are loaded 16
+	// bytes at a time, ensure that there will be at least 4 remaining
+	// bytes after the last round, so that the final read will not pass
+	// beyond the bounds of the input buffer:
+	size_t rounds = (*slen - 4) / 12;
+
+	*slen -= rounds * 12;	// 12 bytes consumed per round
+	*olen += rounds * 16;	// 16 bytes produced per round
+
+	do {
+		if (rounds >= 8) {
+			enc_loop_ssse3_inner(s, o);
+			enc_loop_ssse3_inner(s, o);
+			enc_loop_ssse3_inner(s, o);
+			enc_loop_ssse3_inner(s, o);
+			enc_loop_ssse3_inner(s, o);
+			enc_loop_ssse3_inner(s, o);
+			enc_loop_ssse3_inner(s, o);
+			enc_loop_ssse3_inner(s, o);
+			rounds -= 8;
+			continue;
+		}
+		if (rounds >= 4) {
+			enc_loop_ssse3_inner(s, o);
+			enc_loop_ssse3_inner(s, o);
+			enc_loop_ssse3_inner(s, o);
+			enc_loop_ssse3_inner(s, o);
+			rounds -= 4;
+			continue;
+		}
+		if (rounds >= 2) {
+			enc_loop_ssse3_inner(s, o);
+			enc_loop_ssse3_inner(s, o);
+			rounds -= 2;
+			continue;
+		}
+		enc_loop_ssse3_inner(s, o);
+		break;
+
+	} while (rounds > 0);
+}

data/vendor/libbase64/lib/arch/ssse3/enc_reshuffle.c

@@ -0,0 +1,48 @@
+static inline __m128i
+enc_reshuffle (__m128i in)
+{
+	// Input, bytes MSB to LSB:
+	// 0 0 0 0 l k j i h g f e d c b a
+
+	in = _mm_shuffle_epi8(in, _mm_set_epi8(
+		10, 11,  9, 10,
+		 7,  8,  6,  7,
+		 4,  5,  3,  4,
+		 1,  2,  0,  1));
+	// in, bytes MSB to LSB:
+	// k l j k
+	// h i g h
+	// e f d e
+	// b c a b
+
+	const __m128i t0 = _mm_and_si128(in, _mm_set1_epi32(0x0FC0FC00));
+	// bits, upper case are most significant bits, lower case are least significant bits
+	// 0000kkkk LL000000 JJJJJJ00 00000000
+	// 0000hhhh II000000 GGGGGG00 00000000
+	// 0000eeee FF000000 DDDDDD00 00000000
+	// 0000bbbb CC000000 AAAAAA00 00000000
+
+	const __m128i t1 = _mm_mulhi_epu16(t0, _mm_set1_epi32(0x04000040));
+	// 00000000 00kkkkLL 00000000 00JJJJJJ
+	// 00000000 00hhhhII 00000000 00GGGGGG
+	// 00000000 00eeeeFF 00000000 00DDDDDD
+	// 00000000 00bbbbCC 00000000 00AAAAAA
+
+	const __m128i t2 = _mm_and_si128(in, _mm_set1_epi32(0x003F03F0));
+	// 00000000 00llllll 000000jj KKKK0000
+	// 00000000 00iiiiii 000000gg HHHH0000
+	// 00000000 00ffffff 000000dd EEEE0000
+	// 00000000 00cccccc 000000aa BBBB0000
+
+	const __m128i t3 = _mm_mullo_epi16(t2, _mm_set1_epi32(0x01000010));
+	// 00llllll 00000000 00jjKKKK 00000000
+	// 00iiiiii 00000000 00ggHHHH 00000000
+	// 00ffffff 00000000 00ddEEEE 00000000
+	// 00cccccc 00000000 00aaBBBB 00000000
+
+	return _mm_or_si128(t1, t3);
+	// 00llllll 00kkkkLL 00jjKKKK 00JJJJJJ
+	// 00iiiiii 00hhhhII 00ggHHHH 00GGGGGG
+	// 00ffffff 00eeeeFF 00ddEEEE 00DDDDDD
+	// 00cccccc 00bbbbCC 00aaBBBB 00AAAAAA
+}

data/vendor/libbase64/lib/arch/ssse3/enc_translate.c

@@ -0,0 +1,33 @@
+static inline __m128i
+enc_translate (const __m128i in)
+{
+	// A lookup table containing the absolute offsets for all ranges:
+	const __m128i lut = _mm_setr_epi8(
+		 65,  71, -4, -4,
+		 -4,  -4, -4, -4,
+		 -4,  -4, -4, -4,
+		-19, -16,  0,  0
+	);
+
+	// Translate values 0..63 to the Base64 alphabet. There are five sets:
+	// #  From      To         Abs    Index  Characters
+	// 0  [0..25]   [65..90]   +65        0  ABCDEFGHIJKLMNOPQRSTUVWXYZ
+	// 1  [26..51]  [97..122]  +71        1  abcdefghijklmnopqrstuvwxyz
+	// 2  [52..61]  [48..57]    -4  [2..11]  0123456789
+	// 3  [62]      [43]       -19       12  +
+	// 4  [63]      [47]       -16       13  /
+
+	// Create LUT indices from the input. The index for range #0 is right,
+	// others are 1 less than expected:
+	__m128i indices = _mm_subs_epu8(in, _mm_set1_epi8(51));
+
+	// mask is 0xFF (-1) for range #[1..4] and 0x00 for range #0:
+	__m128i mask = _mm_cmpgt_epi8(in, _mm_set1_epi8(25));
+
+	// Subtract -1, so add 1 to indices for range #[1..4]. All indices are
+	// now correct:
+	indices = _mm_sub_epi8(indices, mask);
+
+	// Add offsets to input values:
+	return _mm_add_epi8(in, _mm_shuffle_epi8(lut, indices));
+}

data/vendor/libbase64/lib/codec_choose.c

@@ -0,0 +1,281 @@
+#include <stdbool.h>
+#include <stdint.h>
+#include <stddef.h>
+#include <stdint.h>
+
+#include "../include/libbase64.h"
+#include "codecs.h"
+#include "config.h"
+#include "env.h"
+
+#if (__x86_64__ || __i386__ || _M_X86 || _M_X64)
+  #define BASE64_X86
+  #if (HAVE_SSSE3 || HAVE_SSE41 || HAVE_SSE42 || HAVE_AVX || HAVE_AVX2)
+    #define BASE64_X86_SIMD
+  #endif
+#endif
+
+#ifdef BASE64_X86
+#ifdef _MSC_VER
+	#include <intrin.h>
+	#define __cpuid_count(__level, __count, __eax, __ebx, __ecx, __edx) \
+	{						\
+		int info[4];				\
+		__cpuidex(info, __level, __count);	\
+		__eax = info[0];			\
+		__ebx = info[1];			\
+		__ecx = info[2];			\
+		__edx = info[3];			\
+	}
+	#define __cpuid(__level, __eax, __ebx, __ecx, __edx) \
+		__cpuid_count(__level, 0, __eax, __ebx, __ecx, __edx)
+#else
+	#include <cpuid.h>
+	#if HAVE_AVX2 || HAVE_AVX
+		#if ((__GNUC__ > 4 || __GNUC__ == 4 && __GNUC_MINOR__ >= 2) || (__clang_major__ >= 3))
+			static inline uint64_t _xgetbv (uint32_t index)
+			{
+				uint32_t eax, edx;
+				__asm__ __volatile__("xgetbv" : "=a"(eax), "=d"(edx) : "c"(index));
+				return ((uint64_t)edx << 32) | eax;
+			}
+		#else
+			#error "Platform not supported"
+		#endif
+	#endif
+#endif
+
+#ifndef bit_AVX2
+#define bit_AVX2 (1 << 5)
+#endif
+#ifndef bit_SSSE3
+#define bit_SSSE3 (1 << 9)
+#endif
+#ifndef bit_SSE41
+#define bit_SSE41 (1 << 19)
+#endif
+#ifndef bit_SSE42
+#define bit_SSE42 (1 << 20)
+#endif
+#ifndef bit_AVX
+#define bit_AVX (1 << 28)
+#endif
+
+#define bit_XSAVE_XRSTORE (1 << 27)
+
+#ifndef _XCR_XFEATURE_ENABLED_MASK
+#define _XCR_XFEATURE_ENABLED_MASK 0
+#endif
+
+#define _XCR_XMM_AND_YMM_STATE_ENABLED_BY_OS 0x6
+#endif
+
+// Function declarations:
+#define BASE64_CODEC_FUNCS(arch)	\
+	BASE64_ENC_FUNCTION(arch);	\
+	BASE64_DEC_FUNCTION(arch);	\
+
+BASE64_CODEC_FUNCS(avx2)
+BASE64_CODEC_FUNCS(neon32)
+BASE64_CODEC_FUNCS(neon64)
+BASE64_CODEC_FUNCS(plain)
+BASE64_CODEC_FUNCS(ssse3)
+BASE64_CODEC_FUNCS(sse41)
+BASE64_CODEC_FUNCS(sse42)
+BASE64_CODEC_FUNCS(avx)
+
+static bool
+codec_choose_forced (struct codec *codec, int flags)
+{
+	// If the user wants to use a certain codec,
+	// always allow it, even if the codec is a no-op.
+	// For testing purposes.
+
+	if (!(flags & 0xFF)) {
+		return false;
+	}
+	if (flags & BASE64_FORCE_AVX2) {
+		codec->enc = base64_stream_encode_avx2;
+		codec->dec = base64_stream_decode_avx2;
+		return true;
+	}
+	if (flags & BASE64_FORCE_NEON32) {
+		codec->enc = base64_stream_encode_neon32;
+		codec->dec = base64_stream_decode_neon32;
+		return true;
+	}
+	if (flags & BASE64_FORCE_NEON64) {
+		codec->enc = base64_stream_encode_neon64;
+		codec->dec = base64_stream_decode_neon64;
+		return true;
+	}
+	if (flags & BASE64_FORCE_PLAIN) {
+		codec->enc = base64_stream_encode_plain;
+		codec->dec = base64_stream_decode_plain;
+		return true;
+	}
+	if (flags & BASE64_FORCE_SSSE3) {
+		codec->enc = base64_stream_encode_ssse3;
+		codec->dec = base64_stream_decode_ssse3;
+		return true;
+	}
+	if (flags & BASE64_FORCE_SSE41) {
+		codec->enc = base64_stream_encode_sse41;
+		codec->dec = base64_stream_decode_sse41;
+		return true;
+	}
+	if (flags & BASE64_FORCE_SSE42) {
+		codec->enc = base64_stream_encode_sse42;
+		codec->dec = base64_stream_decode_sse42;
+		return true;
+	}
+	if (flags & BASE64_FORCE_AVX) {
+		codec->enc = base64_stream_encode_avx;
+		codec->dec = base64_stream_decode_avx;
+		return true;
+	}
+	return false;
+}
+
+static bool
+codec_choose_arm (struct codec *codec)
+{
+#if (defined(__ARM_NEON__) || defined(__ARM_NEON)) && ((defined(__aarch64__) && HAVE_NEON64) || HAVE_NEON32)
+
+	// Unfortunately there is no portable way to check for NEON
+	// support at runtime from userland in the same way that x86
+	// has cpuid, so just stick to the compile-time configuration:
+
+	#if defined(__aarch64__) && HAVE_NEON64
+	codec->enc = base64_stream_encode_neon64;
+	codec->dec = base64_stream_decode_neon64;
+	#else
+	codec->enc = base64_stream_encode_neon32;
+	codec->dec = base64_stream_decode_neon32;
+	#endif
+
+	return true;
+
+#else
+	(void)codec;
+	return false;
+#endif
+}
+
+static bool
+codec_choose_x86 (struct codec *codec)
+{
+#ifdef BASE64_X86_SIMD
+
+	unsigned int eax, ebx = 0, ecx = 0, edx;
+	unsigned int max_level;
+
+	#ifdef _MSC_VER
+	int info[4];
+	__cpuidex(info, 0, 0);
+	max_level = info[0];
+	#else
+	max_level = __get_cpuid_max(0, NULL);
+	#endif
+
+	#if HAVE_AVX2 || HAVE_AVX
+	// Check for AVX/AVX2 support:
+	// Checking for AVX requires 3 things:
+	// 1) CPUID indicates that the OS uses XSAVE and XRSTORE instructions
+	//    (allowing saving YMM registers on context switch)
+	// 2) CPUID indicates support for AVX
+	// 3) XGETBV indicates the AVX registers will be saved and restored on
+	//    context switch
+	//
+	// Note that XGETBV is only available on 686 or later CPUs, so the
+	// instruction needs to be conditionally run.
+	if (max_level >= 1) {
+		__cpuid_count(1, 0, eax, ebx, ecx, edx);
+		if (ecx & bit_XSAVE_XRSTORE) {
+			uint64_t xcr_mask;
+			xcr_mask = _xgetbv(_XCR_XFEATURE_ENABLED_MASK);
+			if (xcr_mask & _XCR_XMM_AND_YMM_STATE_ENABLED_BY_OS) {
+				#if HAVE_AVX2
+				if (max_level >= 7) {
+					__cpuid_count(7, 0, eax, ebx, ecx, edx);
+					if (ebx & bit_AVX2) {
+						codec->enc = base64_stream_encode_avx2;
+						codec->dec = base64_stream_decode_avx2;
+						return true;
+					}
+				}
+				#endif
+				#if HAVE_AVX
+				__cpuid_count(1, 0, eax, ebx, ecx, edx);
+				if (ecx & bit_AVX) {
+					codec->enc = base64_stream_encode_avx;
+					codec->dec = base64_stream_decode_avx;
+					return true;
+				}
+				#endif
+			}
+		}
+	}
+	#endif
+
+	#if HAVE_SSE42
+	// Check for SSE42 support:
+	if (max_level >= 1) {
+		__cpuid(1, eax, ebx, ecx, edx);
+		if (ecx & bit_SSE42) {
+			codec->enc = base64_stream_encode_sse42;
+			codec->dec = base64_stream_decode_sse42;
+			return true;
+		}
+	}
+	#endif
+
+	#if HAVE_SSE41
+	// Check for SSE41 support:
+	if (max_level >= 1) {
+		__cpuid(1, eax, ebx, ecx, edx);
+		if (ecx & bit_SSE41) {
+			codec->enc = base64_stream_encode_sse41;
+			codec->dec = base64_stream_decode_sse41;
+			return true;
+		}
+	}
+	#endif
+
+	#if HAVE_SSSE3
+	// Check for SSSE3 support:
+	if (max_level >= 1) {
+		__cpuid(1, eax, ebx, ecx, edx);
+		if (ecx & bit_SSSE3) {
+			codec->enc = base64_stream_encode_ssse3;
+			codec->dec = base64_stream_decode_ssse3;
+			return true;
+		}
+	}
+	#endif
+
+#else
+	(void)codec;
+#endif
+
+	return false;
+}
+
+void
+codec_choose (struct codec *codec, int flags)
+{
+	// User forced a codec:
+	if (codec_choose_forced(codec, flags)) {
+		return;
+	}
+
+	// Runtime feature detection:
+	if (codec_choose_arm(codec)) {
+		return;
+	}
+	if (codec_choose_x86(codec)) {
+		return;
+	}
+	codec->enc = base64_stream_encode_plain;
+	codec->dec = base64_stream_decode_plain;
+}

data/vendor/libbase64/lib/codecs.h

@@ -0,0 +1,65 @@
+#include <stdint.h>
+#include <stddef.h>
+
+#include "../include/libbase64.h"
+#include "config.h"
+
+// Function parameters for encoding functions:
+#define BASE64_ENC_PARAMS			\
+	( struct base64_state	*state		\
+	, const char		*src		\
+	, size_t		 srclen		\
+	, char			*out		\
+	, size_t		*outlen		\
+	)
+
+// Function parameters for decoding functions:
+#define BASE64_DEC_PARAMS			\
+	( struct base64_state	*state		\
+	, const char		*src		\
+	, size_t		 srclen		\
+	, char			*out		\
+	, size_t		*outlen		\
+	)
+
+// Function signature for encoding functions:
+#define BASE64_ENC_FUNCTION(arch)		\
+	void					\
+	base64_stream_encode_ ## arch		\
+	BASE64_ENC_PARAMS
+
+// Function signature for decoding functions:
+#define BASE64_DEC_FUNCTION(arch)		\
+	int					\
+	base64_stream_decode_ ## arch		\
+	BASE64_DEC_PARAMS
+
+// Cast away unused variable, silence compiler:
+#define UNUSED(x)		((void)(x))
+
+// Stub function when encoder arch unsupported:
+#define BASE64_ENC_STUB				\
+	UNUSED(state);				\
+	UNUSED(src);				\
+	UNUSED(srclen);				\
+	UNUSED(out);				\
+						\
+	*outlen = 0;
+
+// Stub function when decoder arch unsupported:
+#define BASE64_DEC_STUB				\
+	UNUSED(state);				\
+	UNUSED(src);				\
+	UNUSED(srclen);				\
+	UNUSED(out);				\
+	UNUSED(outlen);				\
+						\
+	return -1;
+
+struct codec
+{
+	void (* enc) BASE64_ENC_PARAMS;
+	int  (* dec) BASE64_DEC_PARAMS;
+};
+
+extern void codec_choose (struct codec *, int flags);