Hi,
I got around to running your tests on Windows. On my Radeon HD 5770 the texldp results match what I get on your Wine implementation. The DF16 and DF24 results differ in the green and blue channels - it seems that this particular GPU uses "X001" for both of them. See df24_off_evergreen.png, texldp_evergreen.png and evergreen.txt The attached screenshots show the output with your filtering settings (mag = linear, min = point, mip = linear).
Testing on the r500 GPU reveals that the weird 0.5 pixel offset is gone. This is a result of the linear mag filter, the min or mip filter appear to have no effect. I recommend to set the filters in the way AMD mandates and continue to go for the weird 0.5 texel offset.
I don't think the difference on the r500 card is a result of clamping at the border. The entire image is shifted, including the pixels that sample from the center texels. See r500_ffp_point.png and r500_ffp_lin.png.
Switching to linear mag filtering obviously causes test failures in the tests that have fetch4 off. I'll let you handle the fallout of that and then re-run the tests on my two machines.
Maybe I'll get around to installing Windows 10 on my new Mac with an AMD Polaris GPU. That should give some clues how the behavior developed since then.
Stefan
Am 24.01.19 um 02:47 schrieb DarkZeros:
If you test it, test this new set, some results are now corrected.
El mié., 23 ene. 2019 a las 1:29, DarkZeros (<mailszeros@gmail.com mailto:mailszeros@gmail.com>) escribió:
Hi, Please find attached (compressed format) the set of patches, Hopefully the testbot will not compile and run them. I would say the implementation on Wine is almost finished, projection and 0.5 texel offset are working. 3D textures are not, but don't work on AMD anyway. I just need the values for texldp on AMD, DF16/24 in order to adjust the test results expected (and maybe correct some minor deviation) BR, Daniel El mar., 22 ene. 2019 a las 20:33, Stefan Dösinger (<stefandoesinger@gmail.com <mailto:stefandoesinger@gmail.com>>) escribió: Hi,Am 22.01.2019 um 01:25 schrieb DarkZeros <mailszeros@gmail.com <mailto:mailszeros@gmail.com>>: Hi Stefan, Instead of keeping the discussion private I prefer to send it to the same thread.Good idea :-)I got a new file for you to test. https://www.dropbox.com/s/9a3t1j2nsy3nlm0/d3d9_crosstest.exe?dl=0 I made some big mistakes on the first one (like wine and windows flags were inverted). This new one will hopefully only fail on the middle of the projection (due to rounding i will fix tomorrow), and DFXX.Can you send me a diff for the tests? I have no problem compiling the crosstests myself, but a compiled .exe severely limits my abilities to experiment with it.
