https://bugs.llvm.org/show_bug.cgi?id=48893
Bug ID: 48893
Summary: The performance of vector::assign is 100 times worse
than that of libstdc++.
Product: libc++
Version: 8.0
Hardware: PC
OS: Windows NT
Status: NEW
Severity: release blocker
Priority: P
Component: All Bugs
Assignee: unassignedclangbugs at nondot.org
Reporter: 2077213809 at qq.com
CC: llvm-bugs at lists.llvm.org, mclow.lists at gmail.com
Created attachment 24423
--> https://bugs.llvm.org/attachment.cgi?id=24423&action=edit
the testcase
For details about test cases, see the attachment.
./gccO0.out
copy 1610957372.518706
copy1 1610957372.526591
7885
./clangO0.out
copy 1610957377.298682
copy1 1610957378.455563
1156881
./gccO2.out
copy 1610957396.558745
copy1 1610957396.566606
7861
./clangO2.out
copy 1610957407.81578
copy1 1610957407.179067
97489
the test function
int main(int argc, char* argv[])
{
std::vector<uint8_t> data;
data.resize(1024UL*1024*24);
std::vector<uint8_t> data1;
struct timeval tv,tv1;
gettimeofday(&tv, NULL);
data1.assign(data.begin(), data.end());
gettimeofday(&tv1, NULL);
std::cout << "copy " << tv.tv_sec <<"." << tv.tv_usec << std::endl;
std::cout << "copy1 " << tv1.tv_sec <<"." << tv1.tv_usec << std::endl;
std::cout << tv1.tv_sec * 1000000 + tv1.tv_usec - (tv.tv_sec*1000000 +
tv.tv_usec) << std::endl;
return 0;
}
The main function consumed is __construct_range_forward. libcxx use the
__construct_range_forward
Each element is constructed using the construct method, which is constructed
using the placement new method. GCC uses std::uninitialized_copy to copy the
memory. The performance is significantly better than that of libcxx.
--
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20210127/bd6badc5/attachment-0001.html>