Problem with running test filter

[Dmitry Plakhin]

I'm writing simple filters "add", "sub", "mul", "div". Each of them should work like this:
input: 0000.tif, 0001.tif, output: 0000.tif (0000.tif + 0001.tif)
input: 0001.tif, 0002.tif, output: 0001.tif (0001.tif + 0002.tif)
...
But every time I running this filter I have different results with the same input. Sometimes I have two or more identical images, sometimes parts of image are mixed from other images.

The code looks correct. I use Eclipse C++ to write filters. Also I use
Intel(R) OpenCL 1.5 SDK package with OpenCL 1.1 support, version 1.5-15293 and new version of UFO Framework (downloaded at 27 of April).

To run filter I use next lines:
plakhin@linux-blqp:~> sudo cp ~/Desktop/add.cl /usr/lib64/ufo/test.cl
plakhin@linux-blqp:~> sudo cp ~/workspace/tmp/Debug/libtmp.so /usr/lib64/ufo/libufofiltertest.so
plakhin@linux-blqp:~> python ~/Desktop/run_filter.py

So I don't know what to do with this problem...

[Dmitry Plakhin]

All needed files

[Matthias Vogelgesang]

Hi Dmitry, I proposed to ask either Suren or Tomy to give you access to one of the servers that are equipped with a real GPU, so that you can test your filters there.

Nevertheless, I will also setup a virtual machine with the exact same setup as you and see what the problems are.

[Matthias Vogelgesang]

I am into this. Can you please re-compile the latest ufo-core and ufo-filters and tell me if this still occurs?

What I suspect is, that NVIDIA automatically blocks on data transfer from host to device, whereas Intel respects the block flag that was set to FALSE when "transfering" the data.

[Dmitry Plakhin]

Replying to Vogelgesang:

I am into this. Can you please re-compile the latest ufo-core and ufo-filters and tell me if this still occurs?

What I suspect is, that NVIDIA automatically blocks on data transfer from host to device, whereas Intel

respects the block flag that was set to FALSE when "transfering" the data.

Hello, Matthias!

I have upgraded ufo-core and ufo-filters. It works! Thank you very much) I don't know if all results are right, because in some cases I wait another result. But every time I run some filter I have the same result with previous run. Moreover, inversion and subtract filter working right in all tests.

In 'addition' filter I have only one image that I can't understand why it looks like it looks. But maybe it's right, I don't know. In other two filters I also can't understand some results, but I think if it's not right, it is my mistake =)