Dolphin, the GameCube and Wii emulator - Forums

Full Version: Pixel Processing Problems: On the Road to Pixel Perfection
You're currently viewing a stripped down version of our content. View the full version with proper formatting.
Pages: 1 2 3 4 5 6 7 8
Has anyone checked OGL performance on Linux with Nvidia drivers? I'm swamped with my own development stuff atm. Nvidia was always rumored to spend more effort on their Linux drivers when it came to OGL compared to Windows (key part is "rumored" though).
(03-16-2014, 02:24 AM)Dolp Wrote: [ -> ]I'm curious, how does this affect Nvidia Maxwell, the newest GPU architecture from Nvidia.

Unfortunately, this does affect Maxwell like the Kepler cards before it when it comes to OpenGL. I ran a quick test, using the intro scene for LoZ: TP on rev 4.0-681 (before tev_fixes_new) and 4.0-1206 (after merge). Using OpenGL, the performance loss is about 50%. Using D3D post tev_fixes_new showed no noticeable performance loss on my GTX 750 Ti vs OpenGL or D3D pre-merge. Hopefully Nvidia can figure out why their integer performance is so bad in OpenGL.

This is *not* whining about the performance though. The inaccuracy and bugs had to go, and the fixes are especially important to me since I pretty much use Dolphin to play LoZ games. Great work neobrain!
I still hope Nvidia will improve their shader compiler so we will be as fast as before. Instructions like those should never be emitted:
AND.S R0.y, {255, 0, 0, 0}.x, {31, 0, 0, 0}.x

For reference, the complete shader is here: http://pastie.org/8866041
It sounds like they are aware of the issue, correct? Are they generally responsive to these kinds of issue requests?

Also, forgive me, but what does that shader instruction do? I'm not pixel shader literate Wink
It will calculate the bitwise_and between 31 and 255 (which is 31) and stores it in the register R0.y. This isn't a big deal, but as both input values are const, there is no need to calculate this for _every_ pixel...
http://www.geeks3d.com/20140310/nvidia-r335-23-whql-released-for-titanfall-new-gl_nv-extensions/

I wonder if any of these new extensions will improve performance for Nvidia hardware.

Quote:Compared to previous versions, R335.23 adds 3 extensions:

GL_EXT_shader_integer_mix
GL_NV_shader_thread_group
GL_NV_shader_thread_shuffle

R335.23 exposes two additional extensions for Maxwell GPUs:

GL_EXT_shader_image_load_formatted
GL_NV_blend_equation_advanced_coherent
All I want to say is thank you so much to the amazing developers! I really appreciate the article, too, giving us insight into the methodologies of emulation-improvement. I've always believed that emulation is providing a critical preservation of history, so major kudos to all involved with Dolphin. I eagerly await the next article!
I wonder if PCSX2 is plagued by the same issue as I remember Jak & Daxter having similar texture defects. Does anyone know if PCSX2 uses floating-point math?
Which part? PCSX2 is a big application.
(03-17-2014, 10:30 AM)NaturalViolence Wrote: [ -> ]Which part? PCSX2 is a big application.

I don't know, I'm not at all adept at code let alone code in emulation. Essentially what I'm asking is could the pcsx2 team solve problems I assume having to do with pixel processing, such as the anomalies in Jak & Daxter the same way this Dolphin merge fixed issues in WW. Just curiosity.
Pages: 1 2 3 4 5 6 7 8