vdSqr function seems not to return correct answer on Xeon Gold 6132

Kurokawa__Hiroshi — Mon, 26 Feb 2018 06:39:14 GMT

(Sorry for my poor English, I'm not a native.)

Hello,

vdSqrt function of MKL returns incorrect answer if I used this function under my developement environment.

Here is my environment.

OS:Windows 10 x64 Pro

CPU: Xeon Gold 6132 * 2

Mem:288GB

DevTools: Visual Studio 2017 (using MS C/C++ compiler , not Intel's C)+ MKL 2018 Upd1(AcademicResearch free License)

Here is short reproduction code,

#include "stdafx.h"
#include <vector>
#include <mkl.h>

int main()
{
std::vector<double> Source, Normal, MKL;
size_t Size = 10000;

Source.resize(Size);
Normal.resize(Size);
MKL.resize(Size);

// create calc. target source 1, 2, 3, .....
for(auto i = 0; i < Size; i++)
{
Source = i + 1;
}

// Exec std::sqrt to all target
for(auto i = 0; i < Size; i++)
{
Normal = std::sqrt(Source);
}

// Exec vdSqr to all target
vdSqr((int)Size, Source.data(), MKL.data());

for(auto i = 0; i < Size; i++)
{
  if(std::abs(Normal - MKL) > 0.0001)
  {
   printf("No = %d, Val:%f Normal:%f MKL:%f\n", i, Source, Normal, MKL);
   throw std::runtime_error("mismatch!");
  }
}
return 0;
}

normal std::sqrt function returns 1, 1.4142, 1.732, .....

but MKL vdSqrt function returns 1, 4, 9, 16, .......

Sincerely,

pls try to use vdSqrt( n, a,

Gennady_F_Intel — Mon, 26 Feb 2018 15:26:02 GMT

pls try to use vdSqrt( n, a, y );

I'm sorry for taking your

Kurokawa__Hiroshi — Tue, 27 Feb 2018 01:27:04 GMT

I'm sorry for taking your time for my easy careless mistake.

I confirmed the correct result in the way you taught.

Thank you for your help.

topic pls try to use vdSqrt( n, a, in Intel® oneAPI Math Kernel Library

vdSqr function seems not to return correct answer on Xeon Gold 6132

pls try to use vdSqrt( n, a,

I'm sorry for taking your