memset

Introduction

From The Open Group Base Specifications issue 7, 2018:

#include <string.h>

void *memset(void *s, int c, size_t n);

• The memset() function shall copy c (converted to an unsigned char) into each of the first n bytes of the object pointed to by s.
• The memset() function shall return s; no return value is reversed to indicate an error.
• No error are defined.

Implementation

This implementation will not work on older and/or specialized hardware.

• This memset() sets up to 8 bytes per iteration.
• An alignment loop is performed when necessary, setting the first byte one by one (up to 7 bytes).
• If the remaining bytes are less than 8, the last bytes are set one by one (up to 7 bytes).

Speed tests

• Results are a percentage of the performance against a glibc memset [ldd (Ubuntu GLIBC 2.27-3ubuntu1) 2.27].
• Compiler is gcc [(Ubuntu 7.3.0-27ubuntu1~18.04) 7.3.0].
• for each size the test runtime is 5 seconds.

Basic implementation

Size\flag	-O3 -fno-builtin	-O2	-Os	(none)
1	99%	104%	101%	100%
10	100%	101%	100%	101%
100	100%	97%	96%	87%
1K	105%	83%	81%	41%
10K	100%	35%	35%	8%
100K	104%	12%	13%	2%
1M	79%	10%	10%	2%

Current implementation

Size\flag	-O3 -fno-builtin	-O2	-Os	(none)
1	102%	100%	100%	98%
10	101%	102%	99%	99%
100	98%	101%	101%	98%
1K	101%	95%	95%	84%
10K	98%	88%	78%	41%
100K	99%	74%	57%	17%
1M	80%	68%	53%	13%

Other implementations

• Some example implementations of memset
• MUSL libc memset
• GNU libc memset

Further reading

• Dead Store Elimination (Still) Considered Harmful: Zhaomo Yang, Brian Johannesmeyer, Anders Trier Olesen, Sorin Lerner, Kirill Levchenko. 26th USENIX Security Symposium. August 2017.