User reported 1.14 regressions #16638
Milestone
Comments
|
Toca Race Driver 2 is crashing using opengl backend only. |
|
I discover that Toca Race Driver series is also affected by #16015. |
ghost
referenced
this issue
Dec 22, 2022
…6015 See #16015 Partly derived from master...unknownbrackets:ppsspp:adreno-deadlock But this one surgically activates it only for the known affected games, to avoid any additional performance regressions in the release. Entirely untested.
hrydgard
added a commit
that referenced
this issue
Dec 23, 2022
…pat workaround. See #16638, reported by Gamemulatorer
|
@Gamemulatorer Added. |
|
I think some people have said GLES is slower than before, but I've started to suspect that maybe these people were using a speedhack that was either disabled by default again or removed. Interestingly, at least one person said Vulkan was a bit faster than before, but was less stable FPS wise than GLES. -[Unknown] |
|
Just a heads up, this change (200c055) switching on "Retain changed textures" in compat.ini for Final Fantasy Tactics breaks some character stat drawing in the party screen. I have to override it manually in PPSSPP/system/compat.ini or the stat display gets stuck and doesn't update when switching between characters using L and R, it's much worse when switching very quickly. I'm on Android, but people on the FFT subreddit are seeing the same on Windows. v1.13.2 is unaffected up through git build v1.13.2-1801. If it's helpful to see what's happening I have a couple of quick screen capture clips of the menu pre and post change. |
|
Good discovery. I'm gonna back that change out, I don't know if it had enough performance benefit to motivate itself anyway. |
hrydgard
added a commit
that referenced
this issue
Dec 27, 2022
|
Looked at Dante's Inferno, as mentioned in #16327 , it does an effect that's very inefficient to emulate (texturing from the same image it's rendering to to perform a bloom effect). In 1.14, it's turned even more inefficient, since previously we were somehow able to (incorrectly in my opinion!) not do a new copy after every single draw call, instead it only happend after each little bunch of them drawing to the same location. So we fixed a logical error but it got slower. To get back the old speed correctly we need to track modified rectangles of the current texture, only "re-copying" it to a source texture when we texture from a region the previous draw call modified. We have some limited tracking similar to that but not enough, I think. |
|
Using 1 or more post processing shaders can decrease performance in Tekken6. |
|
@Gamemulatorer Well, post-shaders aren't free. Is it somehow worse in 1.14 than in 1.13? |
Using 1.14.3 version this issue of performance using shaders is ok now :) |
|
The Dante and Toca issues are fixed, and the others are a bit nebulous and hard to reproduce. Closing. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Been getting some emails and Discord messages reporting reduced performance or graphical glitches in some games in 1.14, collecting them here to maybe later split out into issues. I've added partial names so I can search my email for them without being too specific.
Naruto Ultimate Ninja Impact
For example, Sebastian reports that on a Huawei y5 2018 , he's got new graphical problems in Naruto Ultimate ninja impact. PowerVR GE8100.
FIFA "22"
Esethu reports that FIFA 22 (assuming one of those roster hacks) and Dragon Ball Z (unclear which game) has suddenly started crashing a lot on a "Samsung Android (latest version)".
Dante's Inferno
Performance regression, seems fillrate related (user can only detect performance loss at high resolutions) so likely some new passes are happening that didn't before (framebuffer reinterpret or similar). (yes, see below) See #17030
Toca Race Driver
The range culling issues are back on Mali. (Fixed)
Gran Turismo
Apparently it ran better on Galaxy A5 (android 7, adreno) in older versions.
Fate Unlimited Codes
This happens on D3D11 for some reason.
There are additional even less specific ones.
More as they arrive, maybe there are patterns.
The text was updated successfully, but these errors were encountered: