Attachment 762 Details for Bug 1365 – Test C++ code and assembly obtained using the 2 versions of Eigen

This bugzilla service is closed. All entries have been migrated to https://gitlab.com/libeigen/eigen

Test C++ code and assembly obtained using the 2 versions of Eigen

file_1365.txt (text/plain), 2.80 KB, created by Luca Vandelli on 2016-12-19 18:07:38 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: Luca Vandelli

Created: 2016-12-19 18:07:38 UTC

Size: 2.80 KB

patch

obsolete

>#include <Eigen\Dense>
>#include <vector>
>
>#include <Windows.h>
>
>#include <iostream>
>#include <time.h>
>#include <stdio.h>
>
>using namespace Eigen;
>
>static void computeDistances(Ref<MatrixXf> out, const std::vector<Eigen::Vector4f> & points) {
>
>	auto np = points.size();
>	for (int i = 0; i < np; ++i) {
>		for (int j = 0; j < np; ++j) {
>			out(i, j) = (points[j] - points[i]).squaredNorm();
>		}
>	}
>}
>
>static void performTest(int np, int nTrials) {
>
>	// Create a sample set of points
>	std::vector<Eigen::Vector4f> points;
>	points.reserve(np);	
>	for (int i = 0; i < np; ++i) {
>		float iFloat = (float)i;
>		Vector4f point(0.0 + iFloat, 1.0 + iFloat, 2.0 + iFloat, 3.0 + iFloat);
>		points.push_back(point);
>	}
>
>	MatrixXf result(np, np);
>
>	clock_t begin = clock();
>
>	for (int i = 0; i < nTrials; ++i) {
>	     // Compute the (squared) distances between all the couples of points.
>	     computeDistances(result, points);
>	}
>
>	clock_t end = clock();
>
>	double elapsed_secs = double(end - begin) / CLOCKS_PER_SEC;
>
>	std::cout << "nTrials\t" << nTrials << "\tnPoints\t" << np << "\ttime\t" << elapsed_secs << "\r\n";
>}
>
>int main(int ac, char* av[])
>{
>	performTest(1000, 200);
>	getchar();
>}
>
>Assembly obtained with Eigen 3.2.10:
>00007FF6D64F12F0  add         rcx,rcx  
>00007FF6D64F12F3  vmovups     xmm1,xmmword ptr [rbx+rcx*8]  
>00007FF6D64F12F8  vsubps      xmm1,xmm1,xmmword ptr [rbx+rax*8]  
>00007FF6D64F12FD  vmulps      xmm3,xmm1,xmm1  
>00007FF6D64F1301  vmovhlps    xmm1,xmm3,xmm3  
>00007FF6D64F1305  vaddps      xmm3,xmm1,xmm3  
>00007FF6D64F1309  vshufps     xmm0,xmm3,xmm3,1  
>00007FF6D64F130E  vaddss      xmm2,xmm3,xmm0  
>00007FF6D64F1312  vmovss      dword ptr [r10],xmm2  
>00007FF6D64F1317  inc         r9d  
>00007FF6D64F131A  lea         r10,[r10+0FA0h]  
>00007FF6D64F1321  movsxd      rcx,r9d  
>00007FF6D64F1324  cmp         rcx,r11  
>00007FF6D64F1327  jb          performTest+170h (07FF6D64F12F0h) 
>
>Assembly obtained with Eigen 3.3.1:
>00007FF6D13812E0  shl         rcx,4  
>00007FF6D13812E4  add         rcx,rbx  
>00007FF6D13812E7  mov         qword ptr [rbp-79h],rcx  
>00007FF6D13812EB  mov         qword ptr [rbp-71h],rax  
>00007FF6D13812EF  vmovups     ymm0,ymmword ptr [rsp+28h]  
>00007FF6D13812F5  vmovups     ymmword ptr [rbp-59h],ymm0  
>00007FF6D13812FA  vmovups     xmm0,xmmword ptr [rcx]  
>00007FF6D13812FE  vsubps      xmm1,xmm0,xmmword ptr [rax]  
>00007FF6D1381302  vmulps      xmm3,xmm1,xmm1  
>00007FF6D1381306  vhaddps     xmm1,xmm3,xmm3  
>00007FF6D138130A  vhaddps     xmm3,xmm1,xmm1  
>00007FF6D138130E  vmovss      dword ptr [r9],xmm3  
>00007FF6D1381313  inc         r8d  
>00007FF6D1381316  lea         r9,[r9+0FA0h]  
>00007FF6D138131D  movsxd      rcx,r8d  
>00007FF6D1381320  cmp         rcx,r11  
>00007FF6D1381323  jb          performTest+150h (07FF6D13812E0h)

Actions: View

Attachments on bug 1365: 762