topic IPP filter problems in Intel® Integrated Performance Primitives

ipp filter problems

steffenroeber — Wed, 08 Aug 2012 07:55:37 GMT

Hi,
I have two problems with the ipp filter functions.
1.
IppiSize dstROI = {718, 404}
IppiSize kernelSize = {3, 3}
IppiPoint anchorIpp = {-1, -1};
status = ippiFilter32f_8u_C1R(u8srcdata, 718, u8dstdata, 718,
dstROI, &kernelMatrix(0, 0), kernelSize, anchorIpp);

return always -14 (step error), altough the step is not less than dstROI.height.

2. If i use
status = ippiFilterLowpass_8u_AC4R(i_src->getPixel(pixelOffset, pixelOffset), yPitch,
dstImg->getPixel(pixelOffset, pixelOffset), yPitch, dstROI, mask);

for example, then the alpha channel of the destination image is always 0.
Cananybody help me?

Regards
Steffen

ipp filter problems

pvonkaenel — Wed, 08 Aug 2012 12:33:45 GMT

Hi Steffen,

I think the step size needs to be in bytes instead of samples. Not sure about the alpha channel problem.

Peter

IPP filter problems

SergeyKostrov — Thu, 09 Aug 2012 00:32:12 GMT

Hi Steffen,

Quoting steffenroeber

I have two problems with the ipp filter functions...
1.
IppiSize dstROI = {718, 404}
IppiSize kernelSize = {3, 3}
IppiPoint anchorIpp = {-1, -1};
status = ippiFilter32f_8u_C1R(u8srcdata, 718, u8dstdata, 718,
dstROI, &kernelMatrix(0, 0), kernelSize, anchorIpp);

return always -14 (step error), altough the step is not less than dstROI.height.
...

These two values don't look right. This is what a header says forippiFilter32f_8u_C1R:

...
// Parameters:
// pSrc Pointer to the source buffer
// srcStep Step in bytes through the source image buffer
// pDst Pointer to the destination buffer
// dstStep Step in bytes through the destination image buffer
// dstRoiSize Size of the source and destination ROI in pixels
// pKernel Pointer to the kernel values ( 32f kernel )
// kernelSize Size of the rectangular kernel in pixels.
// anchor Anchor cell specifying the rectangular kernel alignment
// with respect to the position of the input pixel
...

and for ippiMalloc set of functions:

...
// Parameter:
// widthPixels width of image in pixels
// heightPixels height of image in pixels
// pStepBytes the pointer to the image step, it is an out parameter calculated
// by the function
...

Best regards,
Sergey

IPP filter problems

SergeyKostrov — Thu, 09 Aug 2012 04:07:02 GMT

Quoting steffenroeber

...
status = ippiFilter32f_8u_C1R(u8srcdata, 718, u8dstdata, 718,
dstROI, &kernelMatrix(0, 0), kernelSize, anchorIpp);

return always -14 (step error), altough the step is not less than dstROI.height
...

It is recommended to use a value forthe'Step In Bytes' parameters returned from IPP memory allocation functions,
like ippiMalloc_xxx orippsMalloc_xxx.

Best regards,
Sergey

ipp filter problems

steffenroeber — Thu, 09 Aug 2012 10:54:28 GMT

But for 8 bit images the ypitch is the same like the width of the roi, isn't it?

ipp filter problems

pvonkaenel — Thu, 09 Aug 2012 10:58:13 GMT

No necessarily. If you used the IPP malloc routine it may allocate more per line to try and get the start of each line to align with a certain byte alignment to allow the use of vector instructions. If you're managing the memory your self, then most likely yes. Note that the filter function you're using is on 32-bit floating point data, so the step size will be at least 4 times the width.

Peter

IPP filter problems

SergeyKostrov — Thu, 09 Aug 2012 13:09:51 GMT

Quoting pvonkaenel

... Note that the filter function you're using is on 32-bit floating point data, so the step size will be at least 4 times the width...

Here is a declaration of the function:

IPPAPI( IppStatus, ippiFilter32f_8u_C1R, ( const Ipp8u* pSrc, int srcStep,
Ipp8u* pDst, int dstStep, IppiSize dstRoiSize, const Ipp32f* pKernel,
IppiSize kernelSize, IppiPoint anchor ))

As you can see it uses a floating point type only for the kernel.

IPP filter problems

steffenroeber — Thu, 09 Aug 2012 13:25:50 GMT

I don't understand. The docu says:
ippStsStepErr:
Indicates an error condition if one the step value is less than roiSize.width*sizeof(pSrc)*numberOfChannels.

My roiSize.width=718, sizeof(pSrc) = 1, numberOfChannels =1
-> 718
and my step value is also 718. Do, what does this has to du with the ippimalloc?

Maybe the questions: Is the step val related to the original image or to the dstRoiSize?

IPP filter problems

SergeyKostrov — Fri, 10 Aug 2012 05:40:40 GMT

Quoting steffenroeber

I don't understand. The docu says:
ippStsStepErr:
Indicates an error condition if one the step value is less than roiSize.width*sizeof(pSrc)*numberOfChannels.

My roiSize.width=718, sizeof(pSrc) = 1, numberOfChannels =1
-> 718
and my step value is also 718. Do, what does this has to du with the ippimalloc?...

[SergeyK] It returns a 'Step Bytes' value. For example:
...
...ippiMalloc_8u_C1( iSrcWidth, iSrcHeight, &iSrcStepBytes );
...

Steffen,

How did you allocate the memory for yoursource and destinationimages? Did you use someippiMalloc_xxx function?

In order to help you here is a very small test-case and please take a look / investigate:
[cpp] ... // Sub-Test 06 - Tests for 'ippiMalloc_8u_C1' and 'CrtMalloc' functions { CrtPrintf( RTU("Sub-Test 06n") ); // ippiMalloc_8u_C1: Max Image size = 34207 x 34207 // CrtMalloc : Max Image size = 34208 x 34208 Ipp8u *pSrcImage = RTnull; Ipp8u *pDstImage = RTnull; while( RTtrue ) { RTint iSrcWidth; RTint iSrcHeight; iSrcWidth = 34207; // For ippiMalloc_8u_C1 function ( ~1.08975GB ) iSrcHeight = 34207; // iSrcWidth = 34208; // For CrtMalloc function ( ~1.08982GB ) // iSrcHeight = 34208; CrtPrintf( RTU("Source Image size: %5ld x %5ldn"), iSrcWidth, iSrcHeight ); RTint iSrcStepBytes = -1; iSrcStepBytes = iSrcWidth; pSrcImage = ( Ipp8u * )::ippiMalloc_8u_C1( iSrcWidth, iSrcHeight, &iSrcStepBytes ); // pSrcImage = ( Ipp8u * )::CrtMalloc( iSrcWidth * iSrcHeight * sizeof( RTuchar ) ); if( pSrcImage == RTnull ) { CrtPrintf( RTU("Memory for Source Image is NOT allocatedn") ); break; } CrtPrintf( RTU("Memory for Source Image is allocatedn") ); IppiSize srcSize = { 0 }; srcSize.width = ( RTint )( iSrcWidth * iSrcHeight ); srcSize.height = 1; // st = ::ippiSet_8u_C1R( 1, pSrcImage, 1, srcSize ); // if( st != ippStsNoErr ) // { // CrtPrintf( RTU("[ ippiSet_8u_C1R ] Failedn") ); // break; // } // CrtPrintf( RTU("[ ippiSet_8u_C1R ] Completedn") ); break; } SysSleep( 2000 ); if( pSrcImage != RTnull ) { ::ippiFree( pSrcImage ); // For ippiMalloc_8u_C1 function // CrtFree( pSrcImage ); // For CrtMalloc function pSrcImage = RTnull; CrtPrintf( RTU("Memory for Source Image deallocatedn") ); } } ...[/cpp]Best regards,
Sergey

IPP filter problems

steffenroeber — Fri, 10 Aug 2012 06:39:49 GMT

Hi,
I used my own memorly allocation. And I fixed the issue meanwhile. It seems that I have to use the step value of the whole image instead of the ROI. At least this does not create this error code.

IPP filter problems

SergeyKostrov — Fri, 10 Aug 2012 13:11:24 GMT

Quoting steffenroeber

...I used my own memorly allocation...

[SergeyK] This is OK and I also use 'malloc' instead of 'ippiMalloc_xxx' in some cases. But, I always verify a 'Step Byte' value with
a right'ippiMalloc_xxx' function and then use it.

And I fixed the issue meanwhile. It seems that I have to use the step value of the whole image instead of the ROI. At least this does not create this error code.