topic ippsConv_32f / ippStaticInit failure in Intel® Integrated Performance Primitives

ippsConv_32f / ippStaticInit failure

astrasel — Tue, 13 Jul 2010 20:48:19 GMT

I have an interesting situation that is manifested in a call to ippsConv_32f. This function takes two source vectors and produces the convoluted result. In my case, the all values of the input vectors are >= 0. This means that all values of the output vector should be >= 0.

The problem is that I have an example where the call to ippsConv_32f produces some resulting values that are < 0.

What is interesting is that I have tracked this down to whether or not I call ippStaticInit (I amusing static dispatching) and which platform I am building for.

Specifically, here are the results I am seeing (where 'works' means that the destination does NOT have any values < 0)

Target Platform: Win32 x64
Call ippStaticInit works fails
No ippStaticInit fails works

I have this captured in a stand alone application and can submit it to technical support, but this seems to clearly be a bug and thought I'd submit it here first.

Here is some IPP meta data:

IPP Version Structure:

Version Numbers: 6.1.137.844

Target CPU: mx

Name: ippsmxl.lib

Version: 6.1 build 137.53

Build date: Jan 23 2010

ippsConv_32f / ippStaticInit failure

Gennady_F_Intel — Thu, 15 Jul 2010 06:58:45 GMT

this is an unknown problem - can you give us the test case?

--Gennady

ippsConv_32f / ippStaticInit failure

Gennady_F_Intel — Fri, 16 Jul 2010 10:14:36 GMT

I mised that you sent the sources to the premier #595725.

Nevertheless,I tried to check the problem on my side with IPP 6.1.5 and the probelm has not been reproduced:

see the results below:

test64bit.exe

IPP Version Structure:

Version Numbers: 6.1.137.847

Target CPU: mx

Name: ippsmxl.lib

Version: 6.1 build 137.56

Build date: Mar 17 2010

Static IPP: Intel

Input: src1 - is all positive

Input: src2 - is all positive

Input: dst - NOT all positive

and the same for ia32:

test32bit.exe

IPP Version Structure:

Version Numbers: 6.1.137.847

Target CPU: px

Name: ippspxl.lib

Version: 6.1 build 137.56

Build date: Mar 17 2010

Static IPP: Intel

Input: src1 - is all positive

Input: src2 - is all positive

Input: dst - is all positive

- I used makefile and ran the code whitout MVSC IDE.

- because you use only one function from ipps domain, therefore I linked only

lib32:

cl main.cpp /Fetest32bit.exe -I$(IPP_INCL) \

$(IPPROOT)\ippsemerged.lib $(IPPROOT)\ippsmerged.lib \

$(IPPROOT)\ippiemerged.lib $(IPPROOT)\ippimerged.lib \

$(IPPROOT)\ippcorel.lib

em64t:

cl main.cpp /Fetest64bit.exe -I$(IPP_INCL) \

$(IPP64ROOT)\ippsemergedem64t.lib $(IPP64ROOT)\ippsmergedem64t.lib \

$(IPP64ROOT)\ippiemergedem64t.lib $(IPP64ROOT)\ippimergedem64t.lib \

$(IPP64ROOT)\ippcoreem64tl.lib

--Gennady