Intel® Integrated Performance Primitives

Interleaved data re-packing

Hello,

Ihave a short (signed 16-bit integers) interleaved data set that I would like to re-pack into different format used for further processing. The data is arranged M x N, where M is number of different data channel information, and N is the number of data points. I would like to extract only a single channel of information 1 x N, but I do not know which IPP function to use so the re-packing is fast and efficient.

Anyone successfully converted a Firewire camera image?

I'm trying to convert an image in the format YUV422 given by a firewire camera to an RGB image, but unsuccessfull. No matter what I do I always get a very weird image.

I'm not sure which step to use, but I think that for an image width 320x240 the YUV422 image step would be 640, right?

This is what I do:

imgSize.width = 320;
imgSize.height = 240;

rgbImg = ippiMalloc_8u_C3( 320, 240, &step );
ippiYUV422ToRGB_8u_C2C3R( img, 640, rgbImg,
step, imgSize );

Thanks in advanced for any help.

V4.0 -> V4.1 performace degradation

My application links to static Ipp libraries ( merged.lib + emerged.lib + IppStaticInitBest()).When I upgraded libraryfrom V4.0 to V4.1 I noticed a huge performance degradation.Ippmpeg4 codecruns about 2x slower.I didn't change anything but simplychanged ipp*.lib and *h files to newer versions.Everything is ok whenI revert to the previous version.
What could be a reason ?
Processortype detected correctly as "Pentium4HT2 2800MHz". mpeg codec is put in separate static library, butIppStaticInitBest() is called from my main exe.

Thanks in advance,
Kamil

Addition with overflow?

Hi all -

I'm writing a function that allows the user to shift the hue of an image. I convert the image into an HSV format, then use the AddC function. However, because the function automatically saturates pixel values in the case of overflow, the hue shift does not work correctly. Ideally, I need a function that does not saturate the pixel values, but wraps around in the case of addition overflow.

Is there some function that will not saturate overflow values? Or is there some other method to achieve the same result? I apologize if this is an elementary question.

interleaved image to planar image

I am doing image processing on a P4 processor.

The image are interleaved 8 bit integers i.e (rgbrgbrgb...) when I get the images but it is much easier to perform fast processing using sse when the image is planar (rrrrrggggggbbbb...) I would like to do this conversion. Does anyone know if the source code for a conversion is available anywhere? The code will be quite short but I would like to do other things than optimizing this if it is possible.
Any suggestions how this should be done as fast as possible? is SSE fastest?

/Johan

Subscribe to Intel® Integrated Performance Primitives