topic Re: Re:Bug in zgetrs() under Windows 10 2004 in Intel® oneAPI Math Kernel Library

Bug in zgetrs() under Windows 10 2004

vssalbert — Mon, 24 Aug 2020 12:26:29 GMT

When running under Windows 10 2004, if the CRT function fmod(x, y) is called with 0 for the x argument prior to calling zgetrs(), zgetrs() returns -nan(ind) for many of the values of the solved for x. The call to fmod() can occur far away and long before.

This is with Visual Studio Professional 2019 versions 16.7.2 and 16.2.3. Attached is code illustrating the problem, when run under Release the output for the vector pB contains several -nan(ind) values.

zgetrs() returns the expected results if any of the following is true:

- the x argument to fmod() is not 0.

- Debug build is run.

- It is run under Windows 10 1909 or 1903.

- The environment variable MKL_CBWR=COMPATIBLE or MKL_CBWR=SSE2 is set.

This was tested using:

- MKL versions 2019 build 4, 2019 build 5, and 2020 build 2.

- Intel(R) Core(TM) i7-4770 CPU @ 3.40GHz 3.40 GHz, 16 GB RAM

- Intel(R) Core(TM) i7-7820HQ CPU @ 2.90GHz 2.90GHz, 32 GB RAM

- Windows 10 Pro 2004

Our production product is experiencing this problem under Windows 10 Pro 2004 with MKL version 2019 build 5. The call to fmod() is occurring much earlier than the call to zgetrs(), and is known to occur in the same thread (it may also happen in separate threads, I have not checked that scenario).

Thanks,

Albert Santos.

Re:Bug in zgetrs() under Windows 10 2004

Gennady_F_Intel — Tue, 25 Aug 2020 07:42:03 GMT

I don't see the problem on my side: win10, command line, MKL version 2020.2

Intel(R) Core(TM) i5-7300U CPU @ 2.60GHz

....MKL_CBWR=COMPATIBLE...

MKL 2020.0.2 Intel(R) 64 architecture

Processor: Intel(R) Streaming SIMD Extensions 2 (Intel(R) SSE2) enabled processors

Info: 0

14.8677 -6.71688

2.89337 -7.12151

2.52554 -5.46729

...MKL_CBWR=SSE2...

MKL 2020.0.2 Intel(R) 64 architecture

Processor: Intel(R) Streaming SIMD Extensions 2 (Intel(R) SSE2) enabled processors

Info: 0

14.8677 -6.71688

2.89337 -7.12151

2.52554 -5.46729

...MKL_CBWR=SSE42...

MKL 2020.0.2 Intel(R) 64 architecture

Processor: Intel(R) Streaming SIMD Extensions 4.2 (Intel(R) SSE4.2) enabled processors

Info: 0

14.8677 -6.71688

2.89337 -7.12151

2.52554 -5.46729

...MKL_CBWR=AVX...

MKL 2020.0.2 Intel(R) 64 architecture

Processor: Intel(R) Advanced Vector Extensions (Intel(R) AVX) enabled processors

Info: 0

14.8677 -6.71688

2.89337 -7.12151

2.52554 -5.46729

...MKL_CBWR=AVX2...

MKL 2020.0.2 Intel(R) 64 architecture

Processor: Intel(R) Advanced Vector Extensions 2 (Intel(R) AVX2) enabled processors

Info: 0

14.8677 -6.71688

2.89337 -7.12151

2.52554 -5.46729

Re: Re:Bug in zgetrs() under Windows 10 2004

vssalbert — Tue, 25 Aug 2020 11:27:28 GMT

Are you running Windows 10 version 2004? It only started occurring with version 2004. I've also only tried it using Windows 10 Pro, I don't know if the edition affects it.

Re:Bug in zgetrs() under Windows 10 2004

Gennady_F_Intel — Wed, 26 Aug 2020 03:36:27 GMT

my testing system: Windows 10 Enterprise, version 1809

Re:Bug in zgetrs() under Windows 10 2004

Gennady_F_Intel — Wed, 26 Aug 2020 04:17:31 GMT

the same results were obtained at the Win Server 2012 R2 OS.

I think maybe someone from the community has such OS available and checks the problem.

Re: Re:Bug in zgetrs() under Windows 10 2004

vss_albert — Thu, 27 Aug 2020 11:56:02 GMT

I get the same results you do in Windows 10 Enterprise, version 1809.

Re:Bug in zgetrs() under Windows 10 2004

Gennady_F_Intel — Thu, 27 Aug 2020 15:03:23 GMT

Could you share the output you obtained?

Re: Re:Bug in zgetrs() under Windows 10 2004

vss_albert — Thu, 27 Aug 2020 16:23:03 GMT

The following are all under Windows 10 Pro version 2004, on an Intel(R) Core(TM) i7-7820HQ CPU @ 2.90GHz 2.90GHz
32.0 GB RAM:

----

MKL Version String: Intel(R) Math Kernel Library Version 2019.0.5 Product Build 20190808 for Intel(R) 64 architecture applications

Info: 0
-nan(ind) -nan(ind)
-nan(ind) -nan(ind)
-nan(ind) -5.46729

----
MKL Version String: Intel(R) Math Kernel Library Version 2020.0.2 Product Build 20200624 for Intel(R) 64 architecture applications

Info: 0
-nan(ind) -nan(ind)
-nan(ind) -nan(ind)
-nan(ind) -5.46729

--- If the x = 0.0; line is commented out so fmod(x, 1.) is fmod(10., 1.):
MKL Version String: Intel(R) Math Kernel Library Version 2020.0.2 Product Build 20200624 for Intel(R) 64 architecture applications

Info: 0
14.8677 -6.71688
2.89337 -7.12151
2.52554 -5.46729

--- If MKL_CBWR=SSE2 environment variable is set:
MKL Version String: Intel(R) Math Kernel Library Version 2020.0.2 Product Build 20200624 for Intel(R) 64 architecture applications

Info: 0
14.8677 -6.71688
2.89337 -7.12151
2.52554 -5.46729

Re: Re:Bug in zgetrs() under Windows 10 2004

Gennady_F_Intel — Fri, 28 Aug 2020 08:27:28 GMT

please try to run the exe file from zip file I attached to this reply. the password intel_123_

Re: Re:Bug in zgetrs() under Windows 10 2004

vss_albert — Mon, 31 Aug 2020 11:38:55 GMT

I get the following:

MKL 2020.0.2 Intel(R) 64 architecture
Processor: Intel(R) Advanced Vector Extensions 2 (Intel(R) AVX2) enabled processors
Info: 0
14.8677 -6.71688
2.89337 -7.12151
2.52554 -5.46729

Re: Re:Bug in zgetrs() under Windows 10 2004

vss_albert — Tue, 01 Sep 2020 16:56:57 GMT

Note that when I run your test_zgetrs.cpp file as-is I get:

MKL 2020.0.2Intel(R) 64 architecture
Processor: Intel(R) Advanced Vector Extensions 2 (Intel(R) AVX2) enabled processors
Info: 0
14.8677 -6.71688
2.89337 -7.12151
2.52554 -5.46729

But if I remove the comment on line 85, the std::cout << "Hello World " << x << std::endl; I get:

MKL 2020.0.2Intel(R) 64 architecture
Processor: Intel(R) Advanced Vector Extensions 2 (Intel(R) AVX2) enabled processors
Info: 0
-nan(ind) -nan(ind)
-nan(ind) -nan(ind)
-nan(ind) -5.46729

Hello World! 0

I believe what's happening there is the call to fmod() is being optimized away since the result is not needed.

Re: Re:Bug in zgetrs() under Windows 10 2004

pwc — Tue, 15 Sep 2020 01:51:44 GMT

Is there any progress on this bug report? It's been nearly two weeks with no response. We are seeing more customers with this issue when they upgrade to Windows 10 2004.

Re:Bug in zgetrs() under Windows 10 2004

Gennady_F_Intel — Tue, 15 Sep 2020 03:54:16 GMT

we don't reproduce the problem

with or without comments on line 85, we see the same outputs:

MKL 2020.0.2 Intel(R) 64 architecture

Processor: Intel(R) Advanced Vector Extensions 2 (Intel(R) AVX2) enabled processors

Info: 0

14.8677 -6.71688

2.89337 -7.12151

2.52554 -5.46729

Hello World! -0

Win 10, 64 bit. a sequential version of MKL

Win10, Version 1809 Build 17763.1397

Re: Re:Bug in zgetrs() under Windows 10 2004

vss_albert — Tue, 15 Sep 2020 10:55:09 GMT

You need to be running Windows 10 Build 2004 (which came out in May, 2020).

Re:Bug in zgetrs() under Windows 10 2004

Gennady_F_Intel — Tue, 15 Sep 2020 11:30:35 GMT

Yes, it seems the issue happens on this specific OS. The problem is that we still don't use such OS available for testing and validation.

Re: Re:Bug in zgetrs() under Windows 10 2004

pwc — Tue, 15 Sep 2020 15:19:46 GMT

Can't you simply update one of your machines in your build/test lab or get access to a VM? We can't wait around for Intel to update their internal corporate standard for Win10. If you claim to support Windows, why don't you test on all released versions of Windows?

Re: Re:Bug in zgetrs() under Windows 10 2004

Bernard — Tue, 15 Sep 2020 18:21:17 GMT

If I remember correctly you may enable the the trapping of floating-point exceptions by calling __control87() function. The next step would be a registering your own FP exception (SIGFPE) handler with signal function.

The other option is to break with the help of windbg at the prologue of zgetrs() and observe the content of SIMD registers and in general whole function execution untill QNAN is visible in SIMD registers.

Re:Bug in zgetrs() under Windows 10 2004

Gennady_F_Intel — Wed, 23 Sep 2020 11:28:06 GMT

The problem has been checked on Windows 10 version 2004 (OS Build 19041.508)

CPU E5-2697 v2@2.7 GHz

set MKL_VERBOSE=1

//with the line x = 0.0 is commented out

MKL 2020.0.2Intel(R) 64 architecture

Processor: Intel(R) Advanced Vector Extensions (Intel(R) AVX) enabled processors

MKL_VERBOSE Intel(R) MKL 2020.0 Update 2 Product build 20200624 for Intel(R) 64 architecture Intel(R) Advanced Vector Extensions (Intel(R) AVX) enabled processors, Win 2.70GHz cdecl sequential

MKL_VERBOSE ZGETRS(N,3,1,000001E475CF70C0,3,000001E475D0D600,000001E475D0D500,3,0) 32.13us CNR:OFF Dyn:1 FastMM:1 TID:0 NThr:1

Info: 0

14.8677 -6.71688

2.89337 -7.12151

2.52554 -5.46729

//without the line x = 0.0

MKL 2020.0.2Intel(R) 64 architecture

Processor: Intel(R) Advanced Vector Extensions (Intel(R) AVX) enabled processors

MKL_VERBOSE Intel(R) MKL 2020.0 Update 2 Product build 20200624 for Intel(R) 64 architecture Intel(R) Advanced Vector Extensions (Intel(R) AVX) enabled processors, Win 2.70GHz cdecl sequential

MKL_VERBOSE ZGETRS(N,3,1,0000014FE28070C0,3,0000014FE281D600,0000014FE281D500,3,0) 19.75us CNR:OFF Dyn:1 FastMM:1 TID:0 NThr:1

Info: 0

14.8677 -6.71688

2.89337 -7.12151

2.52554 -5.46729

The results are the same.

Re:Bug in zgetrs() under Windows 10 2004

Gennady_F_Intel — Wed, 23 Sep 2020 11:31:20 GMT

The issue is closing and we will no longer respond to this thread. If you require additional assistance from Intel, please start a new thread or submit the ticket to the official Intel Online Service Center.

Any further interaction in this thread will be considered community only.

Re: Re:Bug in zgetrs() under Windows 10 2004

vss_albert — Wed, 23 Sep 2020 11:49:17 GMT

Well, his is what I get with MKL_VERBOSE=1:

MKL 2020.0.2Intel(R) 64 architecture
Processor: Intel(R) Advanced Vector Extensions 2 (Intel(R) AVX2) enabled processors
MKL_VERBOSE Intel(R) MKL 2020.0 Update 2 Product build 20200624 for Intel(R) 64 architecture Intel(R) Advanced Vector Extensions 2 (Intel(R) AVX2) enabled processors, Win 3.40GHz cdecl tbb_thread
MKL_VERBOSE ZGETRS(N,3,1,000001E875538F80,3,000001E875539080,000001E8755415C0,3,0) 82.19us CNR:OFF Dyn:1 FastMM:1 TID:0 NThr:4,BLAS:1,FFT:1,VML:1,PARDISO:1
Info: 0
-nan(ind) -nan(ind)
-nan(ind) -nan(ind)
-nan(ind) -5.46729

Hello World! 0