-
-
Notifications
You must be signed in to change notification settings - Fork 1.9k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
rsx: Fixes #3664
rsx: Fixes #3664
Conversation
kd-11
commented
Oct 27, 2017
- Texture cache fixes
- Tag framebuffer memory when strict mode is enabled to check if content has been written to. This way, framebuffer contents can be ignored if they are known to be incorrect.
- Rework memory protection and ignore flush requests if the data has already been written to the CPU. Speedup when using WCB
- Stability fixes (See Regressions with #3638 #3649)
- Vulkan: Flush command queue before attempting to perform texture writeback to guarantee draw order
- Vulkan: Tag primary command buffers with a flush_only access hint to ensure they are always reopened if submitted outside flush_command_queue such as when dealing with access violations
|
Army of Two, Ni No Kuni and Young justice no longer crash RPCS3 with a nvglv64.dll error when using Vulkan+WCB. Young Justice does freeze without any specific error in log at the same point (loading screen) though. Which is a regression as it used to go ingame before #3638 (although with completely broken graphics). |
|
Freezing is expected and will be fixed in a later PR. As with everything else its about balancing performance and guaranteed accuracy. |
|
The unreleased texture issue seems to have come back. Ni No Kuni - PR: 400-750 unreleased Master: 0-10 Graphics no longer flicker at least (regression from #3638 ). |
EDIT: Its possible that there is double counting of sections causing the reported value to grow excessively while the real number is quite small. Its also possible that there is excessive invalidation of sections going on. I'll look into what could be causing that to happen |
Edit: LBP Karting doesn't crash with d9671bf as well. |
|
@Asinin3 Demon's Souls uses cpu readback to only fetch a 16x16 luminance buffer, not actual color data. Thats why the threshold slider is there. At default settings, 16x16 is below the threshold and is not actually scaled. The same goes for any game that uses the readback for anything that is not actual color buffer data. |
|
427df64 Fixed Performance loss with Gattling Gears VRAM usage is also normal now too 👍 I'll try a few other games but everything seems good so far. The scaling issue is still there but it really doesn't matter as it seems to be an edge case. Edit: Ni No Kuni performance back to normal as well. |
|
Imo, wcb should not be used with scaling if color data is being written back to the cpu. It will downsample the image to fit in the cpu space and then the image will be stretched again to fit the screen. This usually gives very blurry result due to the upscale->downscale->upscale cycle where information is lost every time. I will however set up a mechanism to scale the output if conditions are right. |
- vk: Always reopen primary command buffers. They should only be closed in flush_command_queue - If uploading a texture and there are collisions with protected buffers, do not rebuild the cache - Perform writes via flush before reprotecting pages that were not trampled - Only flush no pages once
…ing violations - This allows for better handling of deferred flushes. -- There's still no guarantee that cache contents will have changed between the set acquisition and following flush operation -- Hopefully this is rare enough that it doesnt cause serious issues for now
…hing to a minimum
|
half fps (12->6) in midnight club LA after this merge WCB / GPU scal. doesnt matter |
|
Please pay attention to this: #3681 |



