768k Skylake Problem/Bug - Page 23

megabit8 · 2016-01-02, 02:15

I have attached 2 pictures of this trivial application showing the iterations per thread before "Pause" is set in VS2015, and the "after1.png" file which shows that 2 errors appeared instantly as soon as Debug Pause is pressed then continue.
On my Skylake the error appears instantly.

I am going to leave this application running overnight to see if any errors appear on their own.

chalsall · 2016-01-02, 02:20

Quote:

Originally Posted by megabit8

I am going to leave this application running overnight to see if any errors appear on their own.

Dumber than bricks....

megabit8 · 2016-01-02, 03:06

Mr. Chris, are you characterizing yourself ?

Do you know what's the difference between genius and "brick stupid" people ?

For those willing to make this test in Visual Studio 2015, I have attached the small project.
You need to download the 2.79 source from:
http://www.mersenne.org/ftp_root/gim...279.source.zip
and copy the p95v279.source.zip\gwnum\gwnum64.lib into the project "AVXCore" folder.
You can skip the project upgrade. That's it. Then compile.

The test itself should be almost as good as the torture test, even better because it is sensitive on every processor to every bit changed.

chalsall · 2016-01-02, 03:16

Quote:

Originally Posted by megabit8

Mr. Chris, are you characterizing yourself ?

Possibly...

But then, maybe not....

megabit8 · 2016-01-02, 03:23

Quote:

Originally Posted by chalsall

Possibly...

But then, maybe not....

Maybe so or maybe no.

Ingenious people have limits and bounds whereas stupidity is unlimited... (Einstein)
What's the only thing that's infinite ?

chalsall · 2016-01-02, 03:31

Quote:

Originally Posted by megabit8

What's the only thing that's infinite ?

Thought?

Let's be honest here... We might have found a bug on Intel's latest and greatest chip.

Let's figure this out. It's not really that big of a deal.

megabit8 · 2016-01-02, 03:32

Now someone with a Haswell processor could validate if the "Pause" issue happens on the 5'th gen Intel Processor. I was thinking of: "kracker" since he validated first the cosine rounding problem.

megabit8 · 2016-01-02, 03:34

Quote:

Originally Posted by chalsall

Thought?

Let's be honest here... We might have found a bug on Intel's latest and greatest chip.

Let's figure this out. It's not really that big of a deal.

The only thing that's infinite is the human stupidity. (Einstein)

Dubslow · 2016-01-02, 03:40

At the very least, you should format your forum posts properly. Avoid unnecessary bold (and I would classify all such usage so far thusly) and use the [code] tags which were designed for code, rather than this bold nonsense.

And then: George?

Prime95 · 2016-01-02, 05:06

Quote:

Originally Posted by megabit8

I had an idea to use the Prime 95 v27.9 code to perform a 768K FFT per thread.

also need to know if I use the functions: addr and gwnum_datasize correctly so that this code does not touch memory zones outside the transform.

I have linked gwnum64.lib (non-debug library) to this project and included all the *.h files from the prime95 v27.9 "gwnum" folder.

Kudos to someone willing to delve deep into the source code and try to potentially learn something new about the problem at hand.

I did not see you using the addr function. This is fortunate. Since you are reading the raw doubles straight out of gwnum memory (looping from 0 to gwnum_datasize/8-1). The addr function is used when looping from 0 to fftlen-1). The gwnum array contains a number of unused cache lines to avoid power-of-2 strides which is why gwnum_datasize is not 768K * 8 bytes.

To make sure you are using AVX FFTs you need to call guessCpuType and then turn off
the CPU_FMA3 flag in CPU_FLAGS.

I don't know what to make of your program working standalone, but not after pausing in the debugger. Could well be a completely hardware different problem or a software problem with MSVC.

Madpoo · 2016-01-02, 05:59

Quote:

Originally Posted by Prime95

Kudos to someone willing to delve deep into the source code and try to potentially learn something new about the problem at hand.

I wanted to chime in and say ditto. It would be nice to have an isolated bit of code that could simply be run and get the error (or not).

It would make it easier for folks to troubleshoot and walk through what's going on. I know that Intel would want to take possible problems seriously and would look into it, but getting easily replicated code would certainly let them focus on the issue itself.

2016-01-02, 02:15	#243
megabit8 Dec 2015 2³×3 Posts	I have attached 2 pictures of this trivial application showing the iterations per thread before "Pause" is set in VS2015, and the "after1.png" file which shows that 2 errors appeared instantly as soon as Debug Pause is pressed then continue. On my Skylake the error appears instantly. I am going to leave this application running overnight to see if any errors appear on their own. Attached Thumbnails

2016-01-02, 03:32	#249
megabit8 Dec 2015 2³·3 Posts	It will be interesting to see if the Pause aka. Break-All issue happens on Haswell Now someone with a Haswell processor could validate if the "Pause" issue happens on the 5'th gen Intel Processor. I was thinking of: "kracker" since he validated first the cosine rounding problem.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Skylake vs Kabylake	ET_	Hardware	17	2017-05-24 16:19
Skylake and RAM scaling	mackerel	Hardware	34	2016-03-03 19:14
So does skylake-nonXeon actually get us anything?	fivemack	Hardware	36	2015-09-08 01:42
Skylake processor	tha	Hardware	7	2015-03-05 23:49
Skylake AVX-512	clarke	Software	15	2015-03-04 21:48

2016-01-02, 03:40	#251
Dubslow Basketry That Evening! "Bunslow the Bold" Jun 2011 40<A<43 -89<O<-88 16065₈ Posts	At the very least, you should format your forum posts properly. Avoid unnecessary bold (and I would classify all such usage so far thusly) and use the [code] tags which were designed for code, rather than this bold nonsense. And then: George?