topic Re: Accelerating random access to memory in Intel® Integrated Performance Primitives

Accelerating random access to memory

nizanh — Wed, 02 Feb 2005 20:01:56 GMT

hi,

I am looking for a way to get faster memory access to, say, some random pixels in an image. I couldn't find any IPP function that dose so (or close to). Presently I am using IPPS function to compute the offset of the memory address, but IPPS doesn't support pointer arithmatic, so I still have to do the last step in a loop.

Any suggestion?

thanks,

Nizan.

Message Edited by nizanh on 02-02-2005 04:14 AM

Re: Accelerating random access to memory

Vladimir_Dudnik — Wed, 09 Feb 2005 19:57:48 GMT

Hi Nizan

It is a bit unclear what you are trying to do. To get access to arbitrary pixel in image you can use

val = pSrc[y*srcStep) + x]

where x and y is pixel coordinates

Is it what you are asking about?

Regards,

Vladimir

Re: Accelerating random access to memory

nizanh — Mon, 14 Feb 2005 17:13:46 GMT

thanks for the response.

I do access the memory with

pSrc[y*srcStep) + x].

but suppose that y & x are vectors. Access the memory through loop

for (i=1;i= pSrc[y*srcStep + x ];}

is quite slow. I can speed it up if I calculate

offset = y*srcStep + x

as vectors, using ipps functions, but I still need a loop

for (i=1;i= pSrc[offset];}

I want to speed it up further. If mmx/sse (and better, IPP) supportsome memory access for multiple address it would probablyboost the speed further.

thanks,

Nizan.