Heavily optimised versions of the string functions from the C standard library, for x86 and x86_64 platforms, developed for the XEOS Operating System.
32 and 64 bits versions are available.
When applicable, and if the CPU supports it, the functions will take advantage of the SSE2 instruction set.
If not, less-optimised versions will be used.
Source code is written in assembly, with a test C file and a standard C implementation for each function, benchmarking it versus the system implementation.
Here are a few benchmark results, compared to the C library implementation of OS X 10.9 (which is also heavily optimised):
Test consists of 10000000 (10 millions) calls to strlen() with a 1000 characters string:
OS X strlen() / 64bits: 0.531318 seconds
XEOS strlen() / 64bits: 0.527532 seconds <-- Best result: XEOS
OS X strlen() / 32bits: 0.610977 seconds
XEOS strlen() / 32bits: 0.607162 seconds <-- Best result: XEOS
Test consists of 10000000 (10 millions) calls to memset() with a 4096 bytes buffer:
OS X memset() / 64bits: 0.906685 seconds
XEOS memset() / 64bits: 0.778556 seconds <-- Best result: XEOS
OS X memset() / 32bits: 0.869590 seconds
XEOS memset() / 32bits: 0.798635 seconds <-- Best result: XEOS
Test consists of 10000000 (10 millions) calls to memchr() with a 1000 bytes buffer, searching last character:
OS X memchr() / 64bits: 0.516240 seconds
XEOS memchr() / 64bits: 0.476380 seconds <-- Best result: XEOS
OS X memchr() / 32bits: 6.374186 seconds
XEOS memchr() / 32bits: 0.532130 seconds <-- Best result (!!!): XEOS
Test consists of 10000000 (10 millions) calls to memcpy() with a 4096 bytes source buffer:
OS X memcpy() / 64bits: 0.928405 seconds
XEOS memcpy() / 64bits: 0.858043 seconds <-- Best result: XEOS
OS X memcpy() / 64bits - misaligned: 1.218205 seconds <-- Best result: OS X
XEOS memcpy() / 64bits - misaligned: 1.408296 seconds
OS X memcpy() / 32bits: 0.975120 seconds
XEOS memcpy() / 32bits: 0.952001 seconds <-- Best result: XEOS
OS X memcpy() / 32bits - misaligned: 1.446217 seconds <-- Best result: OS X
XEOS memcpy() / 32bits - misaligned: 1.457572 seconds
Test consists of 10000000 (10 millions) calls to strcat() with 1000 characters source and destination strings:
OS X strcat() / 64bits: 1.314561 seconds
XEOS strcat() / 64bits: 1.269549 seconds <-- Best result: XEOS
OS X strcat() / 64bits - misaligned: 3.920867 seconds
XEOS strcat() / 64bits - misaligned: 1.423539 seconds <-- Best result (!!!): XEOS
OS X strcat() / 32bits: 1.653358 seconds
XEOS strcat() / 32bits: 1.349833 seconds <-- Best result: XEOS
OS X strcat() / 32bits - misaligned: 4.644904 seconds
XEOS strcat() / 32bits - misaligned: 1.515425 seconds <-- Best result (!!!): XEOS
Whole source code is released under the terms of the XEOS Software License.
Owner: Jean-David Gadina - XS-Labs
Web: www.xs-labs.com
Blog: www.noxeos.com
Twitter: @macmade
GitHub: github.com/macmade
LinkedIn: ch.linkedin.com/in/macmade/
StackOverflow: stackoverflow.com/users/182676/macmade