Intel® Integrated Performance Primitives
Deliberate problems developing high-performance vision, signal, security, and storage applications.
Announcements
FPGA community forums and blogs have moved to the Altera Community. Existing Intel Community members can sign in with their current credentials.

IPP 9.0 - ippiWarpAffineLinear - Aligned Vs non-Aligned inputs

Anchit_D_
Beginner
1,076 Views

Hello,

I have an implementation for Image Warp using the ippiWarpAffineLinear and related set up functions. The implementation works great and gives the right results.
But, I get a difference in results between when-

  1.  input is aligned to 64.
  2.  unaligned input.

This is *not* seen with the ippiWarpAffineNearest symbol.
I have attached the warped image and the diff'ed warped image between aligned and unaligned input.

Here are the questions that I have-

  1. Does the ippiWarpAffineLinear expect memory aligned inputs?
  2. If not, is this mismatch in results a known issue?
  3. If it is a known issue, is there a workaround (other than making a aligned copy of the input image)?

A few more specific of the implementation-
borderType = ippBorderConst
WarpDirection = ippWarpForward
FillValues = 0
dataType = uint8
 

Let me know if I can provide more information.

Thanks.
-Anchit

0 Kudos
1 Reply
Zhen_Z_Intel
Employee
1,075 Views

Hi Anchit,

Normally while you using ippsMalloc/ ippiMalloc, the memory will be aligned to 64 bytes. But we do not required must to align memory, you could choose to use malloc without alignment. It might be some a difference just with some byte, normally you could find the difference hardly. But seems in your test case, the whole image has been changed. It is unusual. I tried with your description, but could not find any difference between aligned input & non-aligned input. could you please provide a test case & tell me your ipp version & system environment. Thank you.

Best regards,
Fiona

0 Kudos
Reply