yuzu/yuzu-emu.github.io
1
0
Fork 0
mirror of https://github.com/yuzu-emu/yuzu-emu.github.io.git synced 2026-08-07 18:23:33 +00:00
Code Issues Packages Projects Releases Wiki Activity

Flam suggestions

Browse source 
Co-authored-by: Schplee <24275329+Schplee@users.noreply.github.com>
This commit is contained in:
Matías Locatti 2022-12-08 07:32:25 -03:00 committed by GitHub
parent 6bd32ce9a4
commit bcb015af99
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 79 additions and 81 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

160
site/content/entry/yuzu-progress-report-nov-2022/index.md
View file

@ -6,48 +6,48 @@ forum = 0
+++ +++
As a dear friend likes to say, meowdy yuwu-zers! As a dear friend likes to say, meowdy yuwu-zers!
This month is special, a report of developer tears caused mostly by the release of new titles in the goliath franchise that is Pokémon, the adventures on new architecture lands, and the list of fixes and improvements along the way. Fasten your seat-belts, here we go! This month is special: a report detailing the uphill battle for the release of new titles in the goliath franchise that is Pokémon, adventures on *new architecture* lands, and a list of fixes and improvements that were made along the way. Fasten your seat-belts, here we go!
<!--more--> <!--more-->
## PSA for NVIDIA users: Almost a flawless victory ## PSA for NVIDIA users (almost a flawless victory)
Before we move to what everyone wants to learn about, just a new Pokémon game, we have very good news to announce. Before we move on to what everyone wants to hear about (just a new Pokémon game, nothing special 😉), we have very exciting news to announce!
NVIDIA driver version `527.37` finally solves all known driver related regressions on Pascal and newer hardware! NVIDIA driver version `527.37` finally solves all known driver related regressions on Pascal and newer hardware!
And to make it better, it affects all cards on active support by NVIDIA, from the GT 1010 to the RTX 4090, there are no more driver related graphical issues (it's only our fault from now on), exactly like the cards used to perform with version 472.12 all the way back in September 2021. And to make it better, it affects all cards actively supported by NVIDIA. From the GT 1010 to the RTX 4090, there are no more driver related graphical issues (it's only our fault for now... until NVIDIA strikes again), exactly how the cards used to perform with driver version `472.12` all the way back in September 2021.
Sadly, Maxwell users (900 series and older) will have to remain in version 512.95 to avoid very frequent device losses when using Vulkan, but for those with GPUs newer than 8 years, there’s no longer a need to stick to old driver versions, feel free to update to the latest release! Sadly, Maxwell users (900 series and older) will have to remain on version `512.95` to avoid frequent device loss errors when using Vulkan, but for those with GPUs newer than 8 years, there’s no longer a need to stick to old driver versions. Feel free to update to the latest release!
Here are links for [desktop](https://www.nvidia.com/download/driverResults.aspx/197300/en-us/) and [laptop](https://www.nvidia.com/download/driverResults.aspx/197301/en-us/) GPUs respectively. Here are links for the [desktop](https://www.nvidia.com/download/driverResults.aspx/197300/en-us/) and [laptop](https://www.nvidia.com/download/driverResults.aspx/197301/en-us/) GPUs respectively.
## Guess who’s back ## Guess who’s back
It’s finally here. It’s finally here.
~~For better or worse.~~ ~~For better or worse.~~
Taking some of the new gameplay elements introduced with `Pokémon Legends: Arceus`, is now with us `Pokémon Scarlet/Violet`! Adopting some of the new gameplay elements introduced with `Pokémon Legends: Arceus`, `Pokémon Scarlet/Violet` has been unleashed upon the world!
This time with ultra-realistic tree textures, and 5 FPS walking NPCs. This time with ultra-realistic tree textures, and NPCs walking at 5 FPS.
Fun new gameplay aside, this release has been quite the challenge for yuzu, exposing several issues and their corresponding fixes helping other games in unexpected ways. Fun new gameplay aside, this release has been quite the challenge for yuzu. It doesn’t help that the game has problems running natively on the Switch to begin with.
It doesn’t help that the game has problems running under native conditions already.
Low performance, storage related stuttering, and memory leaks being some of them. Low performance, storage related stuttering, and memory leaks being some of them.
However, while exposing several issues in yuzu, their corresponding fixes have helped other games in unexpected ways!
When we first got our hands on some copies, the game didn’t render anything, at all, which is not good, something needed to be done. When the game first released, it didn’t render anything in yuzu at all, which is not good. Something needed to be done.
[vonchenplus](https://github.com/vonchenplus) was the first to answer, and identified that this was due the game's use of `gl_Layer`, and {{< gh-hovercard "9253" "added support for it in the shader compiler." >}} [vonchenplus](https://github.com/vonchenplus) was the first to answer, and identified that this was due the game's use of `gl_Layer`, and {{< gh-hovercard "9253" "added support for it in the shader compiler." >}}
You will hear more about this gl_Layer character later. You will hear more about this `gl_Layer` "character" later.
{{< imgs {{< imgs
"./poke.png| One thing is for sure, graphics improved (Pokémon Scarlet)" "./poke.png| One thing is for sure, the graphics have definitely improved (Pokémon Scarlet)"
>}} >}}
With this change, the game starts to render and the team can focus on finding, documenting and fixing any other bugs that show up. With this change, the game starts to render and the team can focus on finding, documenting, and fixing any other bugs that show up.
A limitation was quickly found by the community, gl_Layer depends on having driver support for the `GL_ARB_shader_viewport_layer_array` extension in OpenGL, or the Vulkan equivalent, `VK_EXT_shader_viewport_index_layer`. A limitation was quickly found by the community, `gl_Layer` depends on having driver support for the `GL_ARB_shader_viewport_layer_array` extension in OpenGL, or the Vulkan equivalent, `VK_EXT_shader_viewport_index_layer`.
While most GPU drivers do indeed support these extensions, the NVIDIA drivers for Maxwell V1 and older cards don't, meaning users trying to run `Pokémon Scarlet/Violet` on GPUs older than the 900 series wouldn’t get the game to display anything on screen, in either graphics API and any shader backend. While most GPU drivers do indeed support these extensions, the NVIDIA drivers don't for Maxwell V1 and older cards. This means users trying to run `Pokémon Scarlet/Violet` on GPUs older than the 900 series wouldn’t get the game to display anything on screen, in either graphics API and any shader backend.
Must be part of the promised package that included Vulkan support for Fermi, pure smoke. Must be part of the promised package that included Vulkan support for Fermi, pure smoke.
The way byte[] solved this limitation is by {{< gh-hovercard "9636" "translating the gl_Layer assignments with geometry shaders," >}} basically a brute-forced hardware accelerated way to replicate the missing extensions on older hardware. The way [byte[]](https://github.com/liamwhite) solved this limitation is by {{< gh-hovercard "9363" "translating the gl_Layer assignments with geometry shaders," >}} basically a brute-forced hardware accelerated way to replicate the missing extensions on older hardware.
This results in proper rendering for the decade-old Geforce Fermi, Kepler, and Maxwell V1 series GPUs, in either OpenGL and Vulkan. This results in proper rendering for the decade-old Geforce Fermi, Kepler, and Maxwell V1 series GPUs, in either OpenGL and Vulkan.
Now an extra 6% of users can play the game. Now an extra 6% of users can play the game.
@ -69,31 +69,31 @@ What do you do when all issues seem totally unrelated and random?
You blame the CPU precision, of course! You blame the CPU precision, of course!
More, ahem, precisely, in what optimizations are enabled by default. More, ahem, precisely, in what optimizations are enabled by default.
The responsible one turned out to be `Inaccurate NaN handling`, an optimization made to improve the performance of some rooms in `Luigi’s Mansion 3`. The offending option turned out to be `Inaccurate NaN handling`, an optimization made to improve the performance of some rooms in `Luigi’s Mansion 3`.
[Blinkhawk](https://github.com/FernandoS27) handled the change in the Auto setting for CPU accuracy to skip this optimization. [Blinkhawk](https://github.com/FernandoS27) updated the `Auto` setting for CPU accuracy to skip this optimization.
For anyone wishing to restore the performance for `Luigi’s Mansion 3`, you can use per-game settings, just right click the game, go to Properties, CPU tab, set CPU accuracy to unsafe and make sure Inaccurate NaN handling is enabled, like in the following example: For anyone wishing to restore the performance for `Luigi’s Mansion 3`, we recommend using the per-game settings feature. Right click the game, go to `Properties`, `CPU` tab, set CPU accuracy to `Unsafe`, and make sure `Inaccurate NaN handling` is enabled, as shown in the example below:
{{< imgs {{< imgs
"./lm3.png| The per-game settings window is a pathway to knowledge many consider... Unnatural" "./lm3.png| The per-game settings window is a pathway to knowledge many consider... Unnatural"
>}} >}}
We strongly recommend using a global setting of Auto CPU accuracy, and only implement custom/unsafe settings on a per game basis, never globally. We strongly recommend using a global setting of `Auto` CPU accuracy, and only implement custom/unsafe settings on a per game basis, never globally.
Games WILL break with unsafe settings. Games WILL break with unsafe settings.
Back on GPU related issues, AMD has its own share of issues, it always does. Now going back to GPU related issues, AMD has its own share of problems, as it always does.
Users were quick to point out that the game looked like the usual cheap method USA films resort to when showing anything happening outside their borders, with a yellow filter applied. Users were quick to point out that the game looked as though the director of Breaking Bad produced it, seemingly rendering like a yellow/orange filter was applied.
This issue is for the history books, as both the official AMD drivers (the Windows driver, amdvlk, and AMDGPU-PRO) *and* mesa (RADV) share the same behaviour. This issue is for the history books, as both the official AMD drivers (the Windows driver, amdvlk, and AMDGPU-PRO) *and* mesa (RADV) share the same behaviour.
Let’s provide some context. Let’s provide some context.
The default shader backend of Vulkan, [SPIR-V](https://en.wikipedia.org/wiki/Standard_Portable_Intermediate_Representation), describes the special properties variable decorations have. The default shader backend of Vulkan, [SPIR-V](https://en.wikipedia.org/wiki/Standard_Portable_Intermediate_Representation), describes the special properties variable decorations have.
Usually, a fragment shader’s input variables are set as `smooth`, interpolating the values between each vertex. Usually, a fragment shader’s input variables are set as `smooth`, interpolating the values between each vertex.
But sometimes you end up with a value that is constant across all vertices in a triangle. There’s no need to waste computational power smoothing anything in a case like this, so you can set the variable as `flat` instead. But sometimes you end up with a value that is constant across all vertices in a triangle. There’s no need to waste computational power smoothing anything in a case like this, so you can set the variable as `flat` instead.
Both the official AMD drivers and RADV think that the previously mentioned `gl_Layer` input in the fragment shader needs to be decorated as flat, or they treat it as a constant zero across the triangle, instead of the value set in the vertex shader. Both the official AMD and RADV drivers think that the previously mentioned `gl_Layer` input in the fragment shader needs to be decorated as flat, or they treat it as a constant zero across the triangle, instead of the value set in the vertex shader.
Jury is out on this, we reported the issue to AMD. Jury is out on this, we reported the issue to AMD.
A normal Vulkan program would never find itself in this situation because the GLSL compiler always decorates gl_Layer as flat. A normal Vulkan program would never find itself in this situation because the GLSL compiler always decorates `gl_Layer` as flat.
The NVIDIA driver doesn’t make this assumption and doesn’t treat gl_Layer as a constant zero, regardless of the presence of the decoration. The NVIDIA driver doesn’t make this assumption and doesn’t treat `gl_Layer` as a constant zero, regardless of the presence of the decoration.
{{< single-title-imgs-compare {{< single-title-imgs-compare
"Wasn't this game based in Spain? (Pokémon Scarlet)" "Wasn't this game based in Spain? (Pokémon Scarlet)"
@ -101,27 +101,26 @@ The NVIDIA driver doesn’t make this assumption and doesn’t treat gl_Layer as
"./amdfix.png" "./amdfix.png"
>}} >}}
{{< gh-hovercard "9260" "By adding the missing flat decoration," >}} [byte[]](https://github.com/liamwhite) gives us Radeon users proper blue skies, in both Windows and Linux. {{< gh-hovercard "9260" "By adding the missing flat decoration," >}} byte[] allows Radeon users to view the proper beautiful blue skies, in both Windows and Linux.
Let’s do a tennis match, the last bug was GPU related, time for a CPU one. Ping-ponging back to CPU bugs:
Even with all the work so far, `Pokémon Scarlet/Violet` could periodically crash.
Even with all the work so far, the game could periodically crash.
This gave us a sense of dread we haven’t felt since the release of `Pokémon Sword/Shield`. This gave us a sense of dread we haven’t felt since the release of `Pokémon Sword/Shield`.
Thankfully our fear was unfounded, as only Windows users reported back. Thankfully, our fear was unfounded, as only Windows users reported this issue.
Linux users, for example Steam Deck users, didn’t crash at all. Linux users, for example Steam Deck users, didn’t crash at all.
This gave us the needed hook to reel the line of the cause, which ended up being a problem in dynarmic, {{< gh-hovercard "9271" "a stack misalignment in one of the memory accessors." >}} This gave us the needed hook to reel the line of the cause, which ended up being a problem in dynarmic, {{< gh-hovercard "9271" "a stack misalignment in one of the memory accessors." >}}
A quick call to [Merry](https://github.com/merryhime) over the red phone, and crashing is no more. A quick call to [Merry](https://github.com/merryhime) over the red phone, and the crashing is no more.
Back to GPU, and this one was really annoying. Back to the GPU, and this one was really annoying.
Your writer had to do an all-nighter running tests with [bunnei](https://github.com/bunnei) to figure it out. Your writer had to do an all-nighter running tests with [bunnei](https://github.com/bunnei) to figure it out.
NPCs and characters could vertex explode at random, and while using high GPU accuracy would mitigate the problem to a certain extent, it was still very common in cities, particularly in the academy. NPCs and characters could encounter vertex explosions at random, and while using high GPU accuracy would mitigate the problem to a certain extent, it was still very common in cities, particularly in the academy.
To make it even uglier, those vertex explosions were permanent at specific camera angles. To make it even uglier, those vertex explosions were permanent at specific camera angles.
After a few dozen rounds of regression testing, the root of the cause was found in one of the changes introduced by `Project Y.F.C.`, so Blinkhawk took over and implemented the proper fixes, {{< gh-hovercard "9312" "fixing some buffer cache and engine upload issues." >}} After a few dozen rounds of regression testing, the root cause was found in one of the changes introduced by `Project Y.F.C.`, so Blinkhawk took over and implemented the proper changes, {{< gh-hovercard "9312" "fixing some buffer cache and engine upload issues." >}}
{{< single-title-imgs {{< single-title-imgs
"It's like our personal curse, happens with every release (Pokémon Scarlet & Violet)" "It's like our personal curse, seems to happen with every release (Pokémon Scarlet & Violet)"
"./ver1.png" "./ver1.png"
"./ver2.png" "./ver2.png"
"./ver3.png" "./ver3.png"
@ -130,21 +129,21 @@ After a few dozen rounds of regression testing, the root of the cause was found
While the NPC vertex explosions are fixed, some geometry pop-up issues remain. Homework for later. While the NPC vertex explosions are fixed, some geometry pop-up issues remain. Homework for later.
Now any vertex related glitch only lasts one frame and should go away once the relevant shader gets cached. Now any vertex related glitch only lasts one frame and should go away once the relevant shader gets cached.
Also, as a bonus, now `Pokémon Legends: Arceus` can be played at Normal GPU accuracy without having its previously experiened vertex explosions, like we promised to fix [back in January!](https://yuzu-emu.org/entry/yuzu-progress-report-jan-2022/#a-new-legend) Also, as a bonus, `Pokémon Legends: Arceus` can now be played at `Normal` GPU accuracy without encountering vertex explosions, making good on our [promise to fix]((https://yuzu-emu.org/entry/yuzu-progress-report-jan-2022/#a-new-legend))!
This can greatly increase performance on the superior older Pokémon title. One to scratch from the list. This can greatly increase performance on the older Pokémon title. One more thing to scratch off the list.
That buffer cache rewrite is still needed though. That buffer cache rewrite is still needed though.
CPU's turn, or, well, memory in this case. CPU's turn again, or rather memory in this case.
byte[] fixed {{< gh-hovercard "9279" "a silly mistake in the cheat engine" >}} that caused it to crash when using speedhack cheats. byte[] fixed {{< gh-hovercard "9279" "a silly mistake in the cheat engine" >}} that caused it to crash when using speedhack cheats.
GPU now, or more precisely, an user interface aspect related to it. Quick, back to the GPU! Or more precisely, a change to the user defaults.
As you may know from previous reports, Vulkan has a tendency to break when the user installs outdated screen recorders, overlays, or bloatware that messes with the Vulkan layers. As you may know from previous reports, Vulkan has a tendency to break when the user installs outdated screen recorders, overlays, or bloatware that messes with the Vulkan layers.
This leads people to still use the OpenGL API, which is not only considerably slower for `Pokémon Scarlet/Violet` (over three times slower in some cases), but also run the `GLASM` shader backend, an NVIDIA exclusive feature which is not particularly good with recent game releases, Scarlet/Violet not being an exception. This leads people to still use the OpenGL API, which is not only considerably slower for `Pokémon Scarlet/Violet` (over three times slower in some cases), but also run the `GLASM` shader backend, an NVIDIA exclusive feature which is not particularly good with recent game releases, `Scarlet/Violet` not being an exception.
yuzu used to default to `GLASM` so NVIDIA users could enjoy lower shader compilation related stuttering, as OpenGL’s default shader backend, `GLSL`, is irritatingly slow in this aspect. yuzu used to default to `GLASM` so NVIDIA users could enjoy lower shader compilation related stuttering, as OpenGL’s default shader backend, `GLSL`, is irritatingly slow in this aspect.
The problem with this approach is that the `GLASM` backend (assembly shaders) was developed as an experiment, and its two primary maintainers have moved on from the project, with one literally turning green with envy. The problem with this approach is that the `GLASM` backend (assembly shaders) was developed as an experiment, and its two primary maintainers have moved on from the project, with one literally "turning green with envy."
No one remaining on the team is taking the time to maintain a backend that has decades old and difficult-to-parse documentation, no debug tools, only partially helps a single GPU vendor, and has its advantages negated by a superior alternative, Vulkan. No one on the team is taking the time to maintain a backend that has decades old and difficult-to-parse documentation, no debug tools, only partially helps a single GPU vendor, and has its advantages negated by a superior alternative, Vulkan.
{{< single-title-imgs-compare {{< single-title-imgs-compare
"Lavender Town? (Pokémon Violet)" "Lavender Town? (Pokémon Violet)"
@ -152,22 +151,22 @@ No one remaining on the team is taking the time to maintain a backend that has d
"./vulkan.png" "./vulkan.png"
>}} >}}
Assembly shaders was an useful alternative while Vulkan was in early development, now it’s just dead weight taking precious development time that could be used to improve Vulkan instead. Assembly shaders were a useful alternative while Vulkan was in early development, but now it’s just dead weight taking precious development time that could be used to improve Vulkan instead.
This problem is far more common than [your writer](https://github.com/goldenx86) would like, so for those NVIDIA users with the *special superpower* to always break Vulkan, {{< gh-hovercard "9318" "OpenGL will default to GLSL now." >}} This problem is far more common than [your writer](https://github.com/goldenx86) would like, so for those NVIDIA users with the *special superpower* to always break Vulkan, {{< gh-hovercard "9318" "OpenGL will default to GLSL now." >}}
The option to use GLASM will remain available, as Fermi users love it due to their chronic lack of Vulkan support. The option to use GLASM will remain available, as Fermi users love it due to their chronic lack of Vulkan support.
To finish the changes implemented this month regarding this flawed best-seller, bunnei fixed an {{< gh-hovercard "9320" "assert spam in the audio suspend process." >}} To finish the changes implemented this month regarding this flawed best-seller, bunnei fixed an {{< gh-hovercard "9320" "assert spam in the audio suspend process." >}}
This change not only cleans up logs, it has the potential to improve performance a bit. This change not only cleans up logs, it has the potential to improve performance to a degree.
That’s all for this November’s list of Pokémon fixes and improvements. That’s all for this November’s list of Pokémon fixes and improvements.
More work is in development, as there is still stuff to fix, so while we wait let’s end this section with some recommendations we found to get the best experience while playing: More work is in development, as there is still stuff to fix, so while we wait let’s end this section with some recommendations we found to get the best experience while playing:
- NVIDIA, Intel, and AMD users [must run the latest driver versions](https://community.citra-emu.org/t/recommended-settings/319349). - NVIDIA, Intel, and AMD users [must run the latest driver versions](https://community.citra-emu.org/t/recommended-settings/319349).
- High GPU accuracy ensures proper rendering of vegetation and buildings. You see missing graphics? Switch to High. - `High` GPU accuracy ensures proper rendering of vegetation and buildings. Missing graphics? Switch to High.
- Handheld mode improves performance considerably over Docked. - Handheld mode improves performance considerably over Docked.
- Gym trials may require switching to Handheld mode for stability, this could be a game issue as it is known to have memory leaks on console. Later game updates seem to solve this. - Gym trials may require switching to Handheld mode for stability, this could be a game issue as it is known to have memory leaks on console. Later game updates seem to solve this.
- 4 cores/8 threads users seem to improve performance by a good margin by disabling SMT/HT. - 4 core/8 thread users seem to improve performance by a good margin by disabling SMT/HT.
- Installing the latest game update improves performance and stability over long gameplay sessions. - Installing the latest game update improves performance and stability over long gameplay sessions.
## Graphical fixes ## Graphical fixes
@ -176,7 +175,7 @@ Back to the usual agenda, games in general.
vonchenplus found an {{< gh-hovercard "9167" "issue in how yuzu handles tessellation shaders" >}} which causes black backgrounds in `The Legend of Heroes: Trails from Zero`. vonchenplus found an {{< gh-hovercard "9167" "issue in how yuzu handles tessellation shaders" >}} which causes black backgrounds in `The Legend of Heroes: Trails from Zero`.
Why would you want to use tessellation shaders for 2D content on an RPG? Why would you want to use tessellation shaders for 2D content in an RPG?
{{< imgs {{< imgs
"./tfz.png| A tessellated 2D background, in all its glory (The Legend of Heroes: Trails from Zero)" "./tfz.png| A tessellated 2D background, in all its glory (The Legend of Heroes: Trails from Zero)"
@ -184,13 +183,13 @@ Why would you want to use tessellation shaders for 2D content on an RPG?
Immediately after implementing the change, we noticed that games didn’t render correctly, and it wasn’t the fix’s fault. Immediately after implementing the change, we noticed that games didn’t render correctly, and it wasn’t the fix’s fault.
In cases like this, a cache invalidation is not only recommended, but needed. In cases like this, a cache invalidation is not only recommended, but needed.
[gidoly](https://github.com/gidoly) {{< gh-hovercard "9175" "updated the cache version number," >}} invalidating all previous ones, and causing a huge disturbance in the Net, as if thousands of Smash players suddenly cried out in terror, and were suddenly silenced. [gidoly](https://github.com/gidoly) {{< gh-hovercard "9175" "updated the cache version number," >}} invalidating all previous caches, and causing a large disturbance in the Net, as if thousands of Smash players suddenly cried out in terror and were suddenly silenced.
One of the important changes that didn’t make it in time for the first part of `Project Y.F.C.`, Blinkhawk series of GPU related changes and fixes, is {{< gh-hovercard "9194" " improving the handling of ASTC texture mipmap uploads." >}} One of the important changes that didn’t make it in time for the first part of `Project Y.F.C.`, Blinkhawk's series of GPU related changes and fixes, is {{< gh-hovercard "9194" " improving the handling of ASTC texture mipmap uploads." >}}
The original implementation rendered the ASTC mipmaps directly. The original implementation rendered the ASTC mipmaps directly.
What actually happens on Switch is that the NVIDIA driver uploads to its 2D engine first. What actually happens on the Switch is that the NVIDIA driver uploads to its 2D engine first.
The incorrect implementation caused severe texture corruption on many many cases, most commonly on games running Unreal Engine 4. The incorrect implementation caused severe texture corruption in many cases, most commonly in games running Unreal Engine 4.
{{< single-title-imgs-compare {{< single-title-imgs-compare
"Fluffy (Yoshi's Crafted World)" "Fluffy (Yoshi's Crafted World)"
@ -198,7 +197,7 @@ The incorrect implementation caused severe texture corruption on many many cases
"./ue4fix.png" "./ue4fix.png"
>}} >}}
The list is long, so here’s a selection of games that improved with this change: The list is long, so here’s a selection of games that were improved with this change:
- `Blue Fire` - `Blue Fire`
- `A Hat in Time` - `A Hat in Time`
@ -210,21 +209,21 @@ The list is long, so here’s a selection of games that improved with this chang
- `Yoshi's Crafted World` - `Yoshi's Crafted World`
- `F.I.S.T.: Forged In Shadow Torch` - `F.I.S.T.: Forged In Shadow Torch`
- `Digimon Story Cyber Sleuth: Complete Edition` - `Digimon Story Cyber Sleuth: Complete Edition`
- `Grand Theft Auto` trilogy - `Grand Theft Auto: The Trilogy – The Definitive Edition`
Working on improving OpenGL stability, vonchenplus {{< gh-hovercard "9216" "reimplemented the inline index buffer binding," >}} making the OpenGL implementation match the Vulkan one. Working on improving OpenGL stability, vonchenplus {{< gh-hovercard "9216" "reimplemented the inline index buffer binding," >}} making the OpenGL implementation match the Vulkan one.
This solves certain cases where OpenGL would crash when using non-unified memory mode. This solves certain cases where OpenGL would crash when using non-unified memory mode.
Here’s an interesting one. Here’s an interesting one.
If a game had bad performance on a specific system and produced too low framerates at boot, it could hang due to a lost wakeup. If a game had bad performance on a specific system and produced too low of framerates at boot, it could hang due to a lost wakeup.
Since the "secret project" byte[] was working on was a platform slower than a normal gaming PC (more on this in the next section), byte[] blasted this limitation to the Dark World, {{< gh-hovercard "9244" "allowing everyone to be able to boot their game dumps." >}} Since the "secret project" byte[] was working on was a platform slower than a normal gaming PC (more on this in the next section), byte[] blasted this limitation to the Dark World, {{< gh-hovercard "9244" "allowing everyone to be able to boot their game dumps." >}}
A problem we reported a couple of times already, and AMD confirmed to be working on, was crashes happening in `Xenoblade Chronicles 3` with Vulkan, thanks to improving the precision of our MacroJIT code. A problem we reported a couple of times already, and AMD confirmed to be working on, was crashes happening in `Xenoblade Chronicles 3` with Vulkan, thanks to improving the precision of our MacroJIT code.
Since it became clear that more time was needed to solve this issue on the driver side, byte[] managed to implement a workaround, {{< gh-hovercard "9252" "an HLE multi-layer clear mechanism" >}} which bypasses this driver-level limitation. Since it became clear that more time was needed to solve this issue on the driver side, byte[] managed to implement a workaround, {{< gh-hovercard "9252" "an HLE multi-layer clear mechanism" >}} which bypasses this driver-level limitation.
This way the game remains playable in Vulkan for AMD Windows, where the good performance is at. This way, the game remains playable in Vulkan for AMD on Windows.
{{< imgs {{< imgs
"./xc3.png| Always with the good title screen music (Xenoblade Chronicles 3)" "./xc3.png| Always has great title screen music (Xenoblade Chronicles 3)"
>}} >}}
Here’s an interesting find from one of our *old* testers. Here’s an interesting find from one of our *old* testers.
@ -266,7 +265,7 @@ Thank you [toastUnlimited](https://github.com/lat9nq) for helping me with the pe
"./pgfsr.png| Right-click > Properties > Graphics" "./pgfsr.png| Right-click > Properties > Graphics"
>}} >}}
Our own POYO enjoyer [Morph](https://github.com/Morph1984), with help from byte[], worked on {{< gh-hovercard "9307" "improving the usage bits" >}} Vulkan demands of the device’s supported formats, while at it also adding a new format to the list, `R16_SINT`. Our own Poyo enjoyer, [Morph](https://github.com/Morph1984), with help from byte[], worked on {{< gh-hovercard "9307" "improving the usage bits" >}} Vulkan demands of the device’s supported formats, while at it also adding a new format to the list, `R16_SINT`.
One of the “improved” games from this change is `Xenoblade Chronicles: Definitive Edition`. One of the “improved” games from this change is `Xenoblade Chronicles: Definitive Edition`.
While the game complained about missing a texture format, graphics don’t seem to have changed. While the game complained about missing a texture format, graphics don’t seem to have changed.
@ -275,11 +274,11 @@ While the game complained about missing a texture format, graphics don’t seem
Let’s address the elephant in the room, which byte[] has been feeding tons of peanuts lately. Let’s address the elephant in the room, which byte[] has been feeding tons of peanuts lately.
yuzu now supports being {{< gh-hovercard "9198" "built and run on ARM64 devices!" >}} yuzu now supports being {{< gh-hovercard "9198" "built and run on ARM64 devices!" >}}
Of course, hardware requirements remain the same, at least 8GB of RAM and a OpenGL 4.6 compatibility profile/Vulkan 1.2 plus specific extensions GPU driver is required. Of course, hardware requirements remain the same: at least 8GB of RAM and an OpenGL 4.6 compatibility profile/Vulkan 1.2 plus specific extensions GPU driver is required.
This means that yuzu can run on, for example, Asahi Linux on an M1 Apple MacBook Air, but by using Lavapipe, Mesa’s CPU rendering Vulkan driver, which of course results in very slow performance. This means that yuzu can run on, for example, Asahi Linux on an M1 Apple MacBook Air, but by using Lavapipe, Mesa’s CPU rendering Vulkan driver, which of course results in very slow performance.
{{< single-title-imgs {{< single-title-imgs
"This is what's possible right now (Mario Kart 8 Deluxe & Super Mario Galaxy, loaded directly)" "This is what's possible right now (Mario Kart 8 Deluxe & Super Mario Galaxy)"
"./asahi.png" "./asahi.png"
"./asahi2.png" "./asahi2.png"
>}} >}}
@ -301,12 +300,11 @@ Further improvements by byte[] includes {{< gh-hovercard "9215" "corrections on
"./macos.png| If only Apple wasn't a reptilian company and offered Vulkan support" "./macos.png| If only Apple wasn't a reptilian company and offered Vulkan support"
>}} >}}
We have a very long road ahead to get full macOS support ongoing. We have a very long road ahead to get full macOS support, but a journey of a thousand miles begins with a single step, says the old man.
But a journey of a thousand miles begins with a single step, says the old man.
Outside of venturing into new architecture seas, byte also contributed a few general changes too. Outside of venturing into new architecture seas, byte[] also contributed a few general changes too.
Switch games can be weird, for example `MONSTER HUNTER RISE` has some fundamental love for opening services. Switch games can be weird. For example, `MONSTER HUNTER RISE` has some fundamental love for opening services.
After a recent refactor, our kernel emulation incorrectly used a process per session, and this could lead to running out of slab heap, resulting in a hard crash. After a recent refactor, our kernel emulation incorrectly used a process per session, and this could lead to running out of slab heap, resulting in a hard crash.
As a workaround while byte[] refactors services to have a process tree accurate to the Switch, {{< gh-hovercard "9224" "the extraneous processes have been removed," >}} keeping stability in check. As a workaround while byte[] refactors services to have a process tree accurate to the Switch, {{< gh-hovercard "9224" "the extraneous processes have been removed," >}} keeping stability in check.
@ -315,7 +313,7 @@ This brings memory management improvements (must resist making a boomerang joke)
While no game seems to specifically improve with this change, future games targeting the latest firmware versions will certainly benefit. While no game seems to specifically improve with this change, future games targeting the latest firmware versions will certainly benefit.
Nothing better than avoiding future headaches! Nothing better than avoiding future headaches!
## Services, input and audio changes ## Services, input, and audio changes
On the last day of the month, when we cut the list of pull requests that will make it into this article, Morph contacts your writer asking for one last addition. On the last day of the month, when we cut the list of pull requests that will make it into this article, Morph contacts your writer asking for one last addition.
And it was one important addition. And it was one important addition.
@ -327,7 +325,7 @@ Morph {{< gh-hovercard "9348" "improved the stubs" >}} for the `Submit`, `GetReq
"./sp32.png" "./sp32.png"
>}} >}}
A lot more work is required to make this game fully playable, but taking out boot related issues is always a great start. A lot more work is required to make this game fully playable, but fixing boot related issues is always a great start.
[jbeich](https://github.com/jbeich) is back in action, {{< gh-hovercard "9181" "ensuring that yuzu is compatible with BSD systems!" >}} [jbeich](https://github.com/jbeich) is back in action, {{< gh-hovercard "9181" "ensuring that yuzu is compatible with BSD systems!" >}}
This change adds support for previous additions and fixes that were only considering Linux at the time. This change adds support for previous additions and fixes that were only considering Linux at the time.
@ -337,7 +335,7 @@ As part of byte[]’s efforts to get yuzu running on macOS, he and toastUnlimite
First, byte[] {{< gh-hovercard "9304" "assigned the menuRole property for actions to improve Cocoa support," >}} and then toastUnlimited {{< gh-hovercard "9308" "changed the Vulkan check on other platforms to behave closer to how the check for Windows works," >}} improving the compatibility with the macOS file manager. First, byte[] {{< gh-hovercard "9304" "assigned the menuRole property for actions to improve Cocoa support," >}} and then toastUnlimited {{< gh-hovercard "9308" "changed the Vulkan check on other platforms to behave closer to how the check for Windows works," >}} improving the compatibility with the macOS file manager.
[german77](https://github.com/german77) joins the fun, {{< gh-hovercard "9322" "making changes to ensure reading SDL events doesn't crash on macOS." >}} [german77](https://github.com/german77) joins the fun, {{< gh-hovercard "9322" "making changes to ensure reading SDL events doesn't crash on macOS." >}}
Regarding input in general, german77 went to infinity and beyond with yuzu’s multitouch detection, raising the 16 fingers limit to, well, infinite. Regarding input in general, german77 went to infinity and beyond with yuzu’s multitouch detection, raising the 16-finger limit to, well, infinite!
As part of this change, he also {{< gh-hovercard "9191" "tuned the response speed of touches," >}} which were a little bit too fast, resulting in missing input. As part of this change, he also {{< gh-hovercard "9191" "tuned the response speed of touches," >}} which were a little bit too fast, resulting in missing input.
As a result `Mini Motorways` is far more playable now! As a result `Mini Motorways` is far more playable now!
@ -355,8 +353,8 @@ The first example is {{< gh-hovercard "9219" "implementing the NFC" >}} `IUser`
>}} >}}
Next, german77 {{< gh-hovercard "9238" "implemented the" >}} `cabinet applet` (Amiibo manager window). Next, german77 {{< gh-hovercard "9238" "implemented the" >}} `cabinet applet` (Amiibo manager window).
This new settings window allows the user to register, format and delete game data of their Amiibos, while also displaying additional data available. This new settings window allows the user to register, format, and delete game data from their Amiibos, while also displaying any additional data available.
This first implementation will only show up when a game request it, an UI option will be added in the future This first implementation will only show up when a game requests it, a manual UI option will be added in the future.
{{< imgs {{< imgs
"./cabinet.png| Images optional for now (Shovel Knight)" "./cabinet.png| Images optional for now (Shovel Knight)"
@ -370,22 +368,22 @@ yuzu operated under the assumption that games would use buffers 0 to 5 for outpu
This particular game uses something similar to 8/9/C/D/A/B for its 6 buffer outputs, as you can see we’re going hexadecimal here. This particular game uses something similar to 8/9/C/D/A/B for its 6 buffer outputs, as you can see we’re going hexadecimal here.
The previous implementation took the first buffer as a starting point, buffer 0, and added +1 to fill the remaining 5 slots. The previous implementation took the first buffer as a starting point, buffer 0, and added +1 to fill the remaining 5 slots.
That’s obviously not valid for Skyward Sword, the moment you want to access a buffer that is not part of those 6 first consecutive buffers, you get a crash for getting out-of-bounds, and this happens even with the first actual output buffer, at 8! That’s definitely not valid for Skyward Sword, the moment you want to access a buffer that is not part of those 6 first consecutive buffers, you get a crash for getting out-of-bounds, and this happens even with the first actual output buffer, at 8!
Fixing this behaviour by {{< gh-hovercard "9297" "checking the highest buffer and adding +1" >}} stops the game from crashing, and should also have the benefit of not outputting weird sounds, as the correct buffers will be selected. Fixing this behaviour by {{< gh-hovercard "9297" "checking the highest buffer and adding +1" >}} stops the game from crashing, and should also have the benefit of not outputting weird sounds, as the correct buffers will be selected.
As most of you know, our current file system emulation is far from ideal. As most of you know, our current file system emulation is far from ideal.
Morph’s `Project Gaia` should address most issues we have with it once his rewrite is finished. Morph’s `Project Gaia` should address most issues we have with it once his rewrite is finished.
In the meantime, there’s nothing against fixing existing problems with it, and [v1993](https://github.com/v1993) provides a huge hand by addressing a case that could be invisible to the user, and really annoying as a result. In the meantime, there’s nothing against fixing existing problems with it, and [v1993](https://github.com/v1993) provides a huge hand by addressing a case that could be invisible to the user, and really annoying as a result.
The emulator can auto-generate some of the keys required to decrypt and run the Swtich games. The emulator can auto-generate some of the keys required to decrypt and run Switch games.
The problem is that those auto-generated keys are sometimes worse than the ones the user provided during the dumping process, yet they still took precedence. The problem is that those auto-generated keys are sometimes worse than those the user provided during the dumping process, yet they still took precedence.
V1993 {{< gh-hovercard "9324" "flips this behaviour," >}} solving cases of games refusing to boot due to this conflict. v1993 {{< gh-hovercard "9324" "flips this behaviour," >}} solving cases of games refusing to boot due to this conflict.
## User interface improvements ## User interface improvements
Localization is an ongoing effort, and anyone can contribute [here](https://www.transifex.com/yuzu-emulator/yuzu). Localization is an ongoing effort, and anyone can contribute [here](https://www.transifex.com/yuzu-emulator/yuzu).
This time the community {{< gh-hovercard "9166" "added support for a new language, Ukrainian!" >}} This time the community {{< gh-hovercard "9166" "added support for another language, Ukrainian!" >}}
This addition is possible thanks to the work done by [Docteh](https://github.com/Docteh) and GillianMC. This addition is possible thanks to the work done by [Docteh](https://github.com/Docteh) and GillianMC.
Keep at it, guys, we're all with you. Keep at it guys, we're all with you.
Docteh also {{< gh-hovercard "9180" "fixed the translation of pop-up warnings" >}} that show up when trying to remove game content. Docteh also {{< gh-hovercard "9180" "fixed the translation of pop-up warnings" >}} that show up when trying to remove game content.
@ -395,15 +393,15 @@ Docteh also {{< gh-hovercard "9180" "fixed the translation of pop-up warnings" >
"./removefix.png" "./removefix.png"
>}} >}}
No one wants to lose their saves, especially not if it is the own user’s fault. No one wants to lose their saves, especially not if it's the user’s own fault.
As per the request of an user, toastUnlimited added {{< gh-hovercard "9247" "a warning pop-up when an user attempts to delete an user profile." >}} As per the request of one such user, toastUnlimited added {{< gh-hovercard "9247" "a warning pop-up when a user attempts to delete a user profile." >}}
Because, yes, deleting a profile includes deleting its saves. Because, yes, deleting a profile includes deleting its saves.
{{< imgs {{< imgs
"./delete.png| The UI must also protect the users from themselves" "./delete.png| The UI must also protect the users from themselves"
>}} >}}
epicboy surprised everyone by slamming down the front door, throwing {{< gh-hovercard "9273" "per-game input profile support" >}} into our faces, and leaving, refusing to elaborate further. epicboy surprised everyone by slamming down the front door, throwing {{< gh-hovercard "9273" "per-game input profile support" >}} in our faces, and leaving, refusing to elaborate further.
A real Chad. A real Chad.
{{< imgs {{< imgs
@ -414,10 +412,10 @@ This feature uses previously created input profiles, make sure to create some in
## Future projects ## Future projects
There’s a *lot* going on behind doors, but it’s too early to start teasing you. There’s a *lot* going on behind closed doors, but it’s too early to start teasing you.
We can confirm that both `Project Gaia` and `Project Y.F.C. Part 2` are progressing healthily, but those are not the only very important changes in the oven. We can confirm that both `Project Gaia` and `Project Y.F.C. Part 2` are progressing healthily, but those are not the only important changes in the oven.
Personally, I’m trying to implement SMAA, but what stated as just a rebase of [BreadFish](https://github.com/breadfish64)'s OpenGL implementation ended up being more than a week of extra work dealing with the Vulkan implementation, taking precious time from the already busy schedule byte[] has. Personally, I’m trying to implement SMAA, but what started as just a rebase of [BreadFish's](https://github.com/breadfish64) OpenGL implementation ended up being more than a week of extra work dealing with the Vulkan implementation, taking precious time from the already busy schedule byte[] has.
Sorry! But the result will be worth it. Sorry! But the result will be worth it.
That’s all folks! Thank you so much for giving us a bit of your time, and we hope to see you next time! That’s all folks! Thank you so much for giving us a bit of your time, and we hope to see you next time!