https://bugs.llvm.org/show_bug.cgi?id=43888
Bug ID: 43888
Summary: Inline memcpy/memmove/memset with unknown size but
bounded by derefenceable info
Product: libraries
Version: trunk
Hardware: PC
OS: Linux
Status: NEW
Severity: enhancement
Priority: P
Component: Scalar Optimizations
Assignee: unassignedb...@nondot.org
Reporter: david.bolvan...@gmail.com
CC: llvm-bugs@lists.llvm.org
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#define C 8
int d[C];
int s[C];
__attribute__((noinline))
void old(int *d, int *s, int n) {
memcpy(d, s, n*sizeof(int));
}
__attribute__((noinline))
void new(int *d, int *s, int n) {
switch(n) {
case 8: memcpy(d, s, n*sizeof(int)); break;
case 7: memcpy(d, s, n*sizeof(int)); break;
case 6: memcpy(d, s, n*sizeof(int)); break;
case 5: memcpy(d, s, n*sizeof(int)); break;
case 4: memcpy(d, s, n*sizeof(int)); break;
case 3: memcpy(d, s, n*sizeof(int)); break;
case 2: memcpy(d, s, n*sizeof(int)); break;
case 1: memcpy(d, s, n*sizeof(int)); break;
case 0: break;
}
}
int main(void) {
for (int i = 0; i < 1000000000; ++i) {
new(d, s, C);
}
}
Idea:
Analyze memcpy's dst an src buffers - for example: infer that sizeof(dst) is 8
ints, sizeof(dst) is 8 ints. Replace small memcpy with jump table with inlined
memcpys (see "new" function).
Performance improvements:
For code above:
time ./old.out
real 0m2,535s
user 0m2,534s
sys 0m0,000s
time ./new.out
real 0m1,981s
user 0m1,981s
sys 0m0,000s
Or C = 8 ints and copy 6 ints.
time ./old.out
real 0m3,123s
user 0m3,123s
sys 0m0,000s
time ./new.out
real 0m1,979s
user 0m1,973s
sys 0m0,004s
--
You are receiving this mail because:
You are on the CC list for the bug._______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
