Skip to content
This repository has been archived by the owner on Jun 28, 2024. It is now read-only.
/ Turbo-Base64 Public archive
forked from powturbo/Turbo-Base64

Turbo Base64 - Fastest Base64 SIMD/Neon/Altivec

License

Notifications You must be signed in to change notification settings

unum-cloud/Turbo-Base64

 
 

Repository files navigation

Turbo Base64:Fastest Base64 Scalar+SIMD

Fastest Base64 SIMD Encoding library
  • 🆕 (2022.02) the fastest now more faster
  • 🆕 (2022.02) improved speed for short strings
  • 100% C (C++ headers), as simple as memcpy
  • No other base64 library encode or decode faster
  • Scalar can be faster than other SSE or ARM Neon based base64 libraries
  • SSE faster than other SSE/AVX/AVX2! base64 library
  • Fastest AVX2 implementation
  • TurboBase64 AVX2 decoding up to ~2x faster than other AVX2 libs.
  • TurboBase64 is 3-4 times faster than other libs for short strings
  • Fastest ARM Neon base64
  • 👍 Dynamic CPU detection and JIT scalar/sse/avx/avx2 switching
  • Base64 robust error checking, optimized for long+short strings
  • Portable library, 32/64 bits, SSE/AVX/AVX2, ARM Neon, Power9 Altivec
  • OS:Linux amd64, arm64, Power9, MacOs+Apple M1, s390x. Windows: Mingw, visual c++
  • Big endian + Little endian
  • Ready and simple to use library, no armada of files, no hassles dependencies
  • LICENSE GPL 3


Benchmark incl. the best SIMD Base64 libs:

  • Single thread
  • Including base64 error checking
  • Small file + realistic and practical (no PURE cache) benchmark
  • Unlike other benchmarks, the best of the best scalar+simd libraries are included
  • all libraries with the latest version

Benchmark Intel CPU: i7-9700k 3.6GHz gcc 11.2

E Size ratio% E MB/s D MB/s Name 50,000 bytes - 2022.02
66668 133.3 32794 37837 tb64v256 Turbo Base64 avx2
66668 133.3 27789 22264 b64avx2 aklomp Base64 avx2
66668 133.3 25305 21980 fb64avx2 lemire Fastbase64 avx2
66668 133.3 17348 20686 tb64v128a Turbo Base64 avx
66668 133.3 16035 18865 tb64v128 Turbo Base64 sse
66668 133.3 15820 13078 b64avx aklomp Base64 avx
66668 133.3 15322 11302 b64sse aklomp Base64 sse41
50000 100.0 47593 47623 memcpy
E Size ratio% E MB/s D MB/s Name 1 MB - 2022.02
1333336 133.3 29086 29748 tb64v256 Turbo Base64 avx2
1333336 133.3 26153 22515 b64avx2 Base64 avx2
1333336 133.3 23686 21231 fb64avx2 Fastbase64 avx2
1333336 133.3 16897 20215 tb64v128a Turbo Base64 avx
1333336 133.3 15932 18749 tb64v128 Turbo Base64 sse
1333336 133.3 15537 12959 b64avx Base64 avx
1333336 133.3 15135 11304 b64sse Base64 sse41
1333336 133.3 6546 5473 TB64x Turbo Base64 scalar
1333336 133.3 6495 4454 b64plain Base64 plain
1333336 133.3 1908 2752 TB64s Turbo Base64 scalar
1333336 133.3 2541 4289 chrome Google Chrome base64
1333336 133.3 2670 2299 fb64plain FastBase64 plain
1333334 135.4 1754 219 linux Linux base64
1000000 100.0 28688 28656 memcpy

TurboBase64 vs. Base64 for short strings (incl. checking)

String length E MB/s D MB/s Name 50,000 bytes - short strings 2022.02
4 - 16 2330 2161 TB64avx2 Turbo Base64 avx2
891 734 b64avx2 Base64 avx2
8 - 32 3963 3570 TB64avx2 Turbo Base64 avx2
1348 943 b64avx2 Base64 avx2
16 - 64 6881 5937 TB64avx2 Turbo Base64 avx2
2509 1488 b64avx2 Base64 avx2
32 - 128 10946 8880 TB64avx2 Turbo Base64 avx2
4902 2777 b64avx2 Base64 avx2
Benchmark ARM Neon: ARMv8 A73-ODROID-N2 1.8GHz (clang 6.0)
E Size ratio% E MB/s D MB/s Name 30MB binary 2019.12
40000000 133.3 2026 1650 TB64neon Turbo Base64 Neon
40000000 133.3 1795 1285 b64neon64 Base64 Neon
40000000 133.3 1270 1095 TB64x Turbo Base64 scalar
40000000 133.3 695 965 TB64s Turbo Base64 scalar
40000000 133.3 512 782 fb64neon Fastbase64 SIMD Neon
40000000 133.3 565 460 Chrome Google Chrome base64
40000000 133.3 642 614 b64plain Base64 plain
40000000 133.3 506 548 fb64plain Fastbase64 plain
40500000 135.4 314 91 Linux Linux base64
30000000 100.0 3820 3834 memcpy
  • (bold = pareto in category) MB=1.000.000
  • (E/D) : Encode/Decode
  • Timmings are respectively relative to the base64 output/input in encode/decode.

Compile: (Download or clone Turbo Base64 SIMD)

    git clone git://github.com/powturbo/TurboBase64.git
    make

Usage: (Benchmark App)

    ./tb64app file 
    or  
    ./tb64app

Function usage:

static inline unsigned turbob64len(unsigned n)
Base64 output length after encoding

unsigned tb64enc(const unsigned char *in, unsigned inlen, unsigned char *out)
Encode binary input 'in' buffer into base64 string 'out'
with automatic cpu detection for simd and switch (sse/avx2/scalar
in : Input buffer to encode
inlen : Length in bytes of input buffer
out : Output buffer
return value: Length of output buffer
Remark : byte 'zero' is not written to end of output stream
Caller must add 0 (out[outlen] = 0) for a null terminated string

unsigned tb64dec(const unsigned char *in, unsigned inlen, unsigned char *out)
Decode base64 input 'in' buffer into binary buffer 'out'
in : input buffer to decode
inlen : length in bytes of input buffer
out : output buffer
return value: >0 output buffer length
0 Error (invalid base64 input or input length = 0)

Environment:

OS/Compiler (32 + 64 bits):
  • Windows: Visual C++ (2017)
  • Windows: MinGW-w64 makefile
  • Linux amd/intel: GNU GCC (>=4.6)
  • Linux amd/intel: Clang (>=3.2)
  • Linux arm: aarch64 ARMv8 Neon: gcc (>=6.3)
  • Linux arm: aarch64 ARMv8 Neon: clang (>=6.0)
  • MaxOS: XCode (>=9), apple M1
  • PowerPC ppc64le: gcc (>=8.0) incl. SIMD Altivec
References:
* SIMD Base64 publications:

Last update: 21 FEB 2022

About

Turbo Base64 - Fastest Base64 SIMD/Neon/Altivec

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • C 96.8%
  • CMake 3.2%