https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91875
Bug ID: 91875
Summary: Performance drop with mt19937 with -O2/-O3/-Ofast
compared to -O1
Product: gcc
Version: 7.1.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c++
Assignee: unassigned at gcc dot gnu.org
Reporter: hehaochen at hotmail dot com
Target Milestone: ---
Compile the following code and run:
> g++ -O0/-O1/-O2/-O3/-Ofast test.cpp
> ./a.out
---------------------------------------------------
#include <iostream>
#include <vector>
#include <chrono>
#include <random>
template< typename T_rng, typename T_dist>
double timer(T_rng& rng, T_dist& dist, int n){
std::vector< typename T_dist::result_type > vec(n, 0);
auto t1 = std::chrono::high_resolution_clock::now();
for (int i = 0; i < n; ++i)
vec[i] = dist(rng);
auto t2 = std::chrono::high_resolution_clock::now();
auto runtime =
std::chrono::duration_cast<std::chrono::microseconds>(t2-t1).count()/1000.0;
return runtime;
}
int main(){
const int n = 10000000;
std::default_random_engine rng_default(1);
std::mt19937 rng_mt (1);
std::mt19937_64 rng_mt_64 (1);
std::uniform_int_distribution<int> dist_int(0,1000);
std::uniform_real_distribution<float> dist_float(0.0, 1.0);
std::cout << "float_default: " << timer(rng_default, dist_float, n) <<
std::endl;
std::cout << "float_mt: " << timer(rng_mt, dist_float, n) << std::endl;
std::cout << "float_mt_64: " << timer(rng_mt_64, dist_float, n) <<
std::endl;
}
---------------------------------------------------
We get the following result:
=========+===============+===========+===========+==========================
Compile | | gcc 6.5.0 | gcc 7.1.0 | Comments
With | | | |
=========+===============+===========+===========+==========================
| float_default | 1680.99 | 470.338 |
+---------------+-----------+-----------+ Good to see the
-O0 | float_mt | 1777.31 | 559.226 | performance improvement
+---------------+-----------+-----------+
| float_mt_64 | 1886.36 | 649.42 |
---------+---------------+-----------+-----------+-------------------------
| float_default | 67.532 | 69.127 |
+---------------+-----------+-----------+
-O1 | float_mt | 60.976 | 60.864 | Always good and stable
+---------------+-----------+-----------+
| float_mt_64 | 126.056 | 128.175 |
---------+---------------+-----------+-----------+-------------------------
| float_default | 69.31 | 67.628 |
+---------------+-----------+-----------+
-O2 | float_mt | 54.214 | 101.032 |
+---------------+-----------+-----------+
| float_mt_64 | 124.479 | 167.792 | float_mt drops by ~85%
---------+---------------+-----------+-----------+
| float_default | 67.513 | 58.234 |float_mt_64 drops by ~35%
+---------------+-----------+-----------+
-O3 | float_mt | 57.296 | 100.051 | *********************
+---------------+-----------+-----------+ *** Even slower ***
| float_mt_64 | 125.185 | 168.787 | *** than '-O1'! ***
---------+---------------+-----------+-----------+ *********************
| float_default | 58.442 | 60.228 |
+---------------+-----------+-----------+
-Ofast | float_mt | 47.885 | 92.448 |
+---------------+-----------+-----------+
| float_mt_64 | 126.016 | 170.85 |
=========+===============+===========+===========+=========================
gcc with former version than 7 has almost same result as gcc 6.5.0.
I think the performance with -O2/-O3/-Ofast shouldn't worse than that with -O1,
that is counter-intuitive.
Thank you.
