Hi All,
ippiPyramidLayerDown consistently fails to compute the last pyramid level (Exit status -6 (ippStsSizeErr: Incorrect value for data size)). All one has to do is to feed it some non-default resolutions. Reproducer is attached, PyramidLayerDown.c. All I changed was the resoution, the rest of the code is identical with the original sample code.
The underlying reason seems to be an issue in IppiPyramidDownState. If I create a separate IppiPyramidDownState for each downsampling step from level i-1 to level i, then the error does not show. But IppiPyramidDownState should work for all pyramid levels, not just some.
Below you find the output of GetLibVersion for the IPP installation on my machine (Windows 8.1, 64 Bit).
Thanks,
Thomas
GetLibVersion output:
version: 2019.0.4 (r62443), [2019.0.62443.4] name: ippCore date: Mar 28 2019
version: 2019.0.4 (r62443), [2019.0.62443.4] name: ippIP AVX2 (l9) date: Mar 28 2019
version: 2019.0.4 (r62443), [2019.0.62443.4] name: ippSP AVX2 (l9) date: Mar 28 2019
version: 2019.0.4 (r62443), [2019.0.62443.4] name: ippCC AVX2 (l9) date: Mar 29 2019
version: 2019.0.4 (r62443), [2019.0.62443.4] name: ippCH AVX2 (l9) date: Mar 29 2019
version: 2019.0.4 (r62443), [2019.0.62443.4] name: ippVM AVX2 (l9) date: Mar 28 2019
version: 2019.0.4 (r62443), [2019.0.62443.4] name: ippCV AVX2 (l9) date: Mar 29 2019
version: 2019.0.4 (r62443), [2019.0.62443.4] name: ippDC AVX2 (l9) date: Mar 29 2019
Link Copied
Thanks for the report, Thomas. The problem happens when the destination size is {1,1}. We know about such an issue and we are planning to fix this issue into the nearest future. We will update this thread as soon as the fix of the problem will be available.
Hi All,
It's even worse than thought. Now I omit the generation of the last pyramid level (1x1), and I still get failures, e.g. with 64x6. I am running pyramid generation for all the combinations of the following image widths and heights:
widths: [1-32 64 128 256 512 1024 1280];
heights: [1-32 64 128 256 512 1024];
Moreover it occurs for the _32f and _8u variants.
Regards,
Thomas
Thank you, Thomas,
We are investigating it.
Pavel
Hi Thomas.
I confirm your issue. It will be fixed in the next IPP release. As a temporal workaround could you please avoid a situation when two levels of pyramid have equal height, i.e {3x1, 2x1, 1x1}?
Thanks for your feedback.
For more complete information about compiler optimizations, see our Optimization Notice.