http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56538
Bug #: 56538
Summary: No opiton to disable slow 'lock' instr. one does not
need
Classification: Unclassified
Product: gcc
Version: 4.8.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
AssignedTo: unassig...@gcc.gnu.org
ReportedBy: jan.kratoch...@redhat.com
Target: x86_64-unknown-linux-gnu
When no threads / shared mem is in use 'lock' has no benefits.
But it still has heavy performance hit.
#include <memory>
#include <unistd.h>
#include <sys/wait.h>
int main() {
const int jobs=8;
pid_t pid[jobs];
std::shared_ptr<int> p(new int());
for (int job=0;job<jobs;job++)
if ((pid[job]=fork())==0) {
for (int i=0;i<1000000000;i++)
std::shared_ptr<int> q(p);
_exit(0);
}
for (int job=0;job<jobs;job++)
waitpid(pid[job],NULL,0);
}
g++ -o bench1 bench1.C -Wall -g -std=gnu++11 -O3
7.816s
g++ bench1.C -Wall -g -std=gnu++11 -O3 -S -o -|g++ -Dlock= -o bench1 -x
assembler-with-cpp - -lstdc++
7.150s
= 9.3% improvement
fork() has to be there to see an improvement - Intel i7-920 probably ignores
'lock' when only single core runs.
gcc-4.8.0-0.14.fc19.x86_64
