paraLLEl N64 – Low-level RDP upscaling is finally here!

June 16, 2020August 17, 2021 Libretro Team

ParaLLEl RDP this year has singlehandedly caused a breakthrough in N64 emulation. For the first time, the very CPU-intensive accurate Angrylion renderer was lifted from CPU to GPU thanks to the powerful low-level graphics API Vulkan. This combined with a dynarec-powered RSP plugin has made low-level N64 emulation finally possible for the masses at great speeds on modest hardware configurations.

ParaLLEl RDP Upscaling

Jet Force Gemini running with 2x internal upscale

It quickly became apparent after launching ParaLLEl RDP that users have grown accustomed to seeing upscaled N64 graphics over the past 20 years. So something rendering at native resolution, while obviously accurate, bit-exact and all, was seen as unpalatable to them. Many users indicated over the past few weeks that upscaling was desired.

Well, now it’s here. ParaLLEl RDP is the world’s first Low-Level RDP renderer capable of upscaling. The graphics output you get is unlike any HLE renderer you’ve ever seen before for the past twenty years, since unlike them, there is full VI emulation (including dithering, divot filtering, and basic edge anti-aliasing). You can upscale in integer steps of the base resolution. When you set resolution upscaling to 2x, you are multiplying the input resolution by 2x. So 256×224 would become 512×448, 4x would be 1024×896, and 8x would be 2048×1792.

Now, here comes the good stuff with LLE RDP emulation. As said before, unlike so many HLE renderers, ParaLLEl RDP fully emulates the RCP’s VI Interface. As part of this interface’s postprocessing routines, it automatically applies an approximation of 8x MSAA (Multi-Sampled Anti-Aliasing) to the image. This means that even though our internal resolution might be 1024×896, this will then be further smoothed out by this aggressive AA postprocessing step.

Super Mario 64 running on ParaLLEl RDP with 2x internal upscale

This results in even games that run at just 2x native resolution looking significantly better than the same resolution running on an HLE RDP renderer. Look for instance at this Mario 64 screenshot here with the game running at 2x internal upscale (512×448).

How to install and set it up

RDP upscaling is available right now on Windows, Linux, and Android. We make no guarantees as to what kind of performance you can expect across these platforms, this is all contingent on your GPU’s Vulkan drivers and its compute power.

Anyway, here is how you can get it.

In RetroArch, go to Online Updater.
(If you have paraLLEl N64 already installed) – Select ‘Update Installed Cores’. This will update all the cores that you already installed.
(If you don’t have paraLLEl N64 installed already) – go to ‘Core Updater’ (older versions of RA) or ‘Core Downloader’ (newer version of RA), and select ‘Nintendo – Nintendo 64 (paraLLEl N64)’.
Now start up a game with this core.
Go to the Quick Menu and go to ‘Options’. Scroll down the list until you reach ‘GFX Plugin’. Set this to ‘parallel’. Set ‘RSP plugin’ to ‘parallel’ as well.
For the changes to take effect, we now need to restart the core. You can either close the game or quit RetroArch and start the game up again.

In order to upscale, you need to first set the Upscaling factor. By default, it is set to 1x (native resolution). Setting it to 2x/4x/8x then restarting the core makes the upscaling take effect.

Explanation of core options

A few new core option features have been added. We’ll briefly explain what they do and how you can go about using them.

(ParaLLEl-RDP) Upscaling factor (Restart)

Available options: 1x, 2x, 4x, 8x

The upscaling factor for the internal resolution. 1x is default and is the native resolution. 2x, 4x, and 8x are all possible. NOTE: It bears noting that 8x requires at least 5GB/6GB VRAM on your GPU. System requirements are steep for 8x and we generally don’t recommend anything less than a 1080 Ti or better for this. Your mileage may vary, just be forewarned. 2x and 4x by comparison are much lighter. Even when upscaling, the rendering is still rendering at full accuracy, and it is still all software rendered on the GPU. 4x upscale means 16x times the work that 1x Angrylion would churn through.

(paraLLEl-RDP) Downsampling

Available options: Disabled, 1/2, 1/4, 1/8

Also known as SSAA, this works pretty similar to the SSAA downscaling feature in Beetle PSX HW’s Vulkan renderer. The idea is that you internally upscale at a higher resolution, then set this option from ‘Disabled’ to any of the other values. What happens from there is that this internal higher resolution image is then downscaled to either half its size, one quarter of its size, or one eight of its size. This gives you a very smoothed out anti-aliased picture that for all intents and purposes still outputs at 240p/240i. From there, you can apply some frontend shaders on top to create a very nice and compelling look that still looks better than native resolution but is also still very faithful to it.

So, if you would want 4x resolution upscaling with 4x SSAA, you’d set ‘Downsample’ to ‘1/2’. With 4x upscale, and 1/4 downsample, you get 240p output with 16x SSAA, which looks great with CRT shaders.

(paraLLEl-RDP) Use native texture LOD when upscaling

This option is disabled by default.

We have so far only found one game that absolutely required this to be turned on for gameplay purposes. If you don’t have this enabled, the Princess-to-Bowser painting transition in Mario 64 is not there and instead you just see Bowser in the portrait from a far distance. There might be further improvements later to attempt to automatically detect these cases.

Most N64 games didn’t use mipmapping, but the ones that do on average benefit from this setting being off – you get higher quality LOD textures instead of a lower-quality LOD texture eventually making way for a more detailed one as you look closer. However, turning this option on could also be desirable depending on whether you favor accurate looking graphics or a facsimile of how things used to look.

(paraLLEl-RDP) Use native resolution for TEX_RECT

This option is on by default.

2D elements such as sprites are usually rendered with TEX_RECT commands, and trying to upscale them inevitably leads to ugly “seams” in the picture. This option forces native resolution rendering for such sprites.

Managing expectations

It’s important that people understand what the focus of this renderer is. There is no intent to have yet another enhancement-focused renderer here. This is the closest there has ever been to date of a full software rendered reimplementation of Angrylion on the GPU with additional niceties like upscaling. The renderer guarantees bit-exactness, what you see is what you would get on a real N64, no exceptions.

With a HLE renderer, the scene is rendered using either OpenGL or Vulkan rasterization rules. Here, neither is done – the exact rasterization steps of the RDP are followed instead, there are no API calls to GL to draw triangles here or there. So how is this done? Through compute shaders. It’s been established that you cannot correctly emulate the RDP’s rasterization rules by just simply mapping it to OpenGL. This is why previous attempts like z64gl fell flat after an initial promising start.

So the value proposition here for upscaling with ParaLLEl RDP is quite compelling – you get upscaling with the most accurate renderer this side of Angrylion. It runs well thanks to Vulkan, you can upscale all the way to 8x (which is an insane workload for a GPU done this way). And purists get the added satisfaction of seeing for the first time upscaled N64 graphics using the N64’s entire postprocessing pipeline finally in action courtesy of the VI Interface. You get nice dither filtering that smooths out really well at higher resolutions and can really fake the illusion of higher bit depth. HLE renderers have a lot of trouble with the kind of depth cuing and dithering being applied on the geometry, but ParaLLEl RDP does this effortlessly. This causes the upscaled graphics to look less sterile, whereas with traditional GL/Vulkan rasterization, you’d just see the same repeated textures everywhere with the same basic opacity everywhere. Here, we get dithering and divot filtering creating additional noise to the image leading to an overall richer picture.

So basically, the aim here is actually emulating the RDP and RSP. The focus is not on getting the majority of commercial games to just run and simulating the output they would generate through higher level API calls.

Won’t be done – where HLE wins

Therefore, the following requests will not be pursued at least in the near future:

* Widescreen rendering – Can be done through game patches (ASM patches applied directly to the ROM, or bps/ups patches or something similar). Has to be done on a per-game basis, with HLE there is some way to modify the view frustum and viewport dimensions to do this but it almost never works right due to the way the game occludes geometry and objects based on your view distance, so game patches implementing widescreen and DOF/draw distance enhancements would always be preferable.

So, in short, yes, you can do this with ParaLLEl RDP too, just with per-game specific patches. Don’t expect a core option that you can just toggle on or off.
* Rendering framebuffer effects at higher resolution – not really possible with LLE, don’t see much payoff to it either. Super-sampled framebuffer effects might be possible in theory.
* Texture resolution packs – Again, no. The nature of an LLE renderer is right there in the name, Low-Level. While the RDP is processing streams of data (fed to it by the RSP), there is barely any notion whatsoever of a ‘texture’ – it only sees TMEM uploads and tile descriptors which point to raw bytes. With High Level emulation, you have a higher abstraction level where you can ‘hook’ into the parts where you think a texture upload might be going on so you can replace it on the fly. Anyway, those looking for something like that are really at the wrong address with ParaLLEl RDP anyway. ParaLLEl RDP is about making authentic N64 rendering look as good as possible without resorting to replacing original assets or anything bootleg like that.
* Z-fighting/subpixel precision: In some games, there is some slight Z-fighting in the distance that you might see which HLE renderers typically don’t have. Again, this is because this is accurate RDP emulation. Z-fighting is a thing. The RDP only has 18-bit UNORM of depth precision with 10 bits of fractional precision during interpolation, and compression on top of that to squeeze it down to 14 bits. A HLE emulator can render at 24+ bits depth. Compounding this, because the RSP is Low-level, it’s sending 16-bit fixed point vertex coordinates to the RDP for rendering. A typical HLE renderer and HLE RSP would just determine that we are about to draw some 3D geometry and then just turn it into float values so that there is a higher level of precision when it comes to vertex positioning. If you recall, the PlayStation1’s GTE also did not deal with vertex coordinates in floats but in fixed point. There, we had to go to the effort of doing PGXP in order to convert it to float. I really doubt there is any interest to contemplate this at this point. Best to let sleeping dogs lie.
* Raw performance. HLE uses the hardware rasterization and texture units of the GPU which is far more efficient than software, but of course, it is far less accurate than software rendering.

Where LLE wins

Conversely, there are parts where LLE wins over HLE, and where HLE can’t really go –

* HLE tends to struggle with decals, depth bias doesn’t really emulate the RDP’s depth bias scheme at all since RDP depth bias is a double sided test. Depth bias is also notorious for behaving differently on different GPUs.
* Correct dithering. A HLE renderer still has to work with a fixed function blending pipeline. A software rendered rasterizer like ParaLLEl RDP does not have to work with any pre-existing graphics API setup, it implements its own rasterizer and outputs that to the screen through compute shading. Correct dither means applying dither after blending, among other things, which is not something you can generally do [with HLE]. It generally looks somewhat tacky to do dithering in OpenGL. You need 8-bit input and do blending in 8-bit but dither + quantization at the end, which you can’t do in fixed function blending.
* The entire VI postprocessing pipeline. Again, it bears repeating that not only is the RDP graphics being upscaled, so is the VI filtering. VI filtering got a bad rep on the N64 because this overaggressive quasi-8x MSAA would tend to make the already low-resolution images look even blurrier. But at higher resolutions as you can see here, it can really shine. You need programmable blending to emulate the VI’s coverage, and this just is not practical with OpenGL and/or the current HLE renderers out there. The VI has a quite ingenious filter that distributes the dither noise where it reconstructs more color depth. So not only are we getting post-processing AA courtesy of the RDP, we’re also getting more color depth.
* What You See Is What You Get. This renderer is the hypothetical what-if scenario of how an N64 Pro unit would look like that could pump out insane resolutions while still having the very same hardware. Notice that the RDP/VI implementations in ParaLLEl RDP have NOT been enhanced in any way. The only real change was modifying the rasterizer to test fractional pixel coordinates as well.
* Full accuracy with CPU readbacks. CPU can freely read and write on top of RDP rendered data, and we can easily deal with it without extra hacks.

Known issues

The deinterlacing process for interlaced video modes is still rather poor (just like Angrylion), basic bob and weave setup. There are plans to come up with a completely new system.
Mario Tennis glitches out a bit with upscaling for some reason, there might be subtle bugs in the implementation that only manifest on that game. This seems to not happen on Nvidia Windows drivers though.

Screenshots

The screenshots below here show ParaLLEl RDP running at its maximum internal input resolution, 8x the original native image. This means that when your game is running at say 256×224, it would be running at 2048×1792. But if your game is running at say 640×480 (some interlaced games actually set the resolution that high, Indiana Jones IIRC), then we’d be looking at 5120×3840. That’s bigger than 4K! Then bear in mind that on top of that you’re going to get the VI’s 8x MSAA on top of that, and you can probably begin to imagine just how demanding this is on your GPU given that it’s trying to run a custom software rasterizer on hardware. Suffice it to say, the demands for 2x and 4x will probably not be too steep, but if you’re thinking of using 8x, you better bring some serious GPU horsepower. You’ll need at least 5-6GB of VRAM for 8x internal resolution for starters.

Anyway, without much further ado, here are some glorious screenshots. GoldenEye 007 now looks dangerously close to the upscaled bullshot images on the back of the boxart!

GoldenEye 007 running with ParaLLEl RDP at 8x internal upscale

Super Mario 64 running on ParaLLEl RDP with 8x internal upscale

Star Fox 64 running on ParaLLEl RDP with 8x internal upscale

Perfect Dark running on ParaLLEl RDP with 8x internal upscale in high-res mode

World Driver Championship running on ParaLLEl RDP with 8x internal upscale

Videos

Body Harvest

Perfect Dark

Legend of Zelda: Ocarina of Time

Super Mario 64

Coming to Mupen64Plus Next soon

ParaLLEl RDP will also be making its way into the upcoming new version of Mupen64Plus Next as well. Expect increased compatibility over ParaLLEl N64 (especially on Android) and potentially better performance in many games.

Future blog posts

There might eventually be some future blog post by Themaister going into more technical detail on the inner workings of ParaLLEl RDP. I will also probably release a performance test-focused blog post later testing a variety of different GPUs and how far we can take them as far as upscaling is concerned.

I can already tell you to neuter your expectations with regards to Android/mobile GPUs. I tested ParaLLEl RDP with 2x upscaling on a Samsung Galaxy S10+ and performance was about 36fps, this is with vsync off. With 1x native resolution I manage to get on average 64 to 70fps with the same games. So obviously mobile GPUs still have a lot of catching up to do with their discrete big brothers on the desktop.

At least it will make for a nice GPU benchmark for mobile hardware until we eventually crack fullspeed with 2x native!

Reviving and rewriting paraLLEl-RDP – Fast and accurate low-level N64 RDP emulation

May 13, 2020August 17, 2021 Maister

Over the last few months after completing the paraLLEl-RSP rewrite to a Lightrec based recompiler, I’ve been plugging away on a project which I had been putting off for years, to implement the N64 RDP with Vulkan compute shaders in a low-level fashion. Every design of the old implementation has been scrapped, and a new implementation has arisen from the ashes. I’ve learned a lot of advanced compute techniques, and I’m able to use far better methods than I was ever able to use back in the early days. This time, I wanted to do it right. Writing a good, accurate software renderer on a massively parallel architecture is not easy and you need to rethink everything. Serial C code will get you nowhere on a GPU, but it’s a fun puzzle, and quite rewarding when stuff works.

The new implementation is a standalone repository that could be integrated into any emulator given the effort: https://github.com/Themaister/parallel-rdp. For this first release, I integrated it into parallel-n64. It is licensed as MIT, so feel free to integrate it in other emulators as well.

Why?

I wanted to prove to myself that I could, and it’s … a little fun? I won’t claim this is more than it is. 🙂

Chasing bit-exactness

The new implementation is implemented in a test-driven way. The Angrylion renderer is used as a reference, and the goal is to generate the exact same output in the new renderer. I started writing an RDP conformance suite. Here, we generate RDP commands in C++, run the commands across different implementations, and compare results in RDRAM (and hidden RDRAM of course, 9-bit RAM is no joke). To pass, we must get an exact match. This is all fixed-point arithmetic, no room for error! I’ve basically just been studying Angrylion to understand what on earth is supposed to happen, and trying to make sense of what the higher level goal of everything is. In LLE, there’s a lot of weird magic that just happens to work out.

I’m quite happy with where I’ve ended up with testing and seeing output like this gives me a small dopamine shot before committing:

…

122/163 Test #122: rdp-test-interpolation-color-texture-ci4-tlut-ia16 ………………………….. Passed 2.50 sec
Start 123: rdp-test-interpolation-color-texture-ci8-tlut-ia16
123/163 Test #123: rdp-test-interpolation-color-texture-ci8-tlut-ia16 ………………………….. Passed 2.40 sec
Start 124: rdp-test-interpolation-color-texture-ci16-tlut-ia16
124/163 Test #124: rdp-test-interpolation-color-texture-ci16-tlut-ia16 …………………………. Passed 2.37 sec
Start 125: rdp-test-interpolation-color-texture-ci32-tlut-ia16
125/163 Test #125: rdp-test-interpolation-color-texture-ci32-tlut-ia16 …………………………. Passed 2.45 sec
Start 126: rdp-test-interpolation-color-texture-2cycle-lod-frac
126/163 Test #126: rdp-test-interpolation-color-texture-2cycle-lod-frac ………………………… Passed 2.51 sec
Start 127: rdp-test-interpolation-color-texture-perspective
127/163 Test #127: rdp-test-interpolation-color-texture-perspective ……………………………. Passed 2.50 sec
Start 128: rdp-test-interpolation-color-texture-perspective-2cycle-lod-frac
128/163 Test #128: rdp-test-interpolation-color-texture-perspective-2cycle-lod-frac ……………… Passed 3.29 sec
Start 129: rdp-test-interpolation-color-texture-perspective-2cycle-lod-frac-sharpen
129/163 Test #129: rdp-test-interpolation-color-texture-perspective-2cycle-lod-frac-sharpen ………. Passed 3.26 sec
Start 130: rdp-test-interpolation-color-texture-perspective-2cycle-lod-frac-detail
130/163 Test #130: rdp-test-interpolation-color-texture-perspective-2cycle-lod-frac-detail ……….. Passed 3.48 sec
Start 131: rdp-test-interpolation-color-texture-perspective-2cycle-lod-frac-sharpen-detail
131/163 Test #131: rdp-test-interpolation-color-texture-perspective-2cycle-lod-frac-sharpen-detail … Passed 3.26 sec
Start 132: rdp-test-texture-load-tile-16-yuv

…

151/163 Test #151: vi-test-aa-none …………………………………………………………. Passed 21.19 sec
Start 152: vi-test-aa-extra-dither-filter
152/163 Test #152: vi-test-aa-extra-dither-filter ……………………………………………. Passed 48.77 sec
Start 153: vi-test-aa-extra-divot
153/163 Test #153: vi-test-aa-extra-divot …………………………………………………… Passed 64.29 sec
Start 154: vi-test-aa-extra-dither-filter-divot
154/163 Test #154: vi-test-aa-extra-dither-filter-divot ………………………………………. Passed 65.90 sec
Start 155: vi-test-aa-extra-gamma
155/163 Test #155: vi-test-aa-extra-gamma …………………………………………………… Passed 48.28 sec
Start 156: vi-test-aa-extra-gamma-dither
156/163 Test #156: vi-test-aa-extra-gamma-dither …………………………………………….. Passed 48.18 sec
Start 157: vi-test-aa-extra-nogamma-dither
157/163 Test #157: vi-test-aa-extra-nogamma-dither …………………………………………… Passed 47.56 sec

…

100% tests passed, 0 tests failed out of 163 #feelsgoodman

Ideally, if someone is clever enough to hook up a serial connection to the N64, it might be possible to run these tests through a real N64, that would be interesting.

I also fully implemented the VI this time around. It passes bit-exact output with Angrylion in my tests and there is a VI conformance suite to validate this as well. I implemented almost the entire thing without even running actual content. Once I got to test real content and sort out the last weird bugs, we get to the next important part of a test-driven development workflow …

The importance of dumping formats

A critical aspect of verifying behavior is being able to dump RDP commands from the emulator and replay them.

On the left I have Angrylion and on the right paraLLEl-RDP running side by side from a dump where I can step draw by draw, and drill down any pesky bugs quite effectively. This humble tool has been invaluable. The Angrylion backend in parallel-n64 can be configured to generate dumps which are then used to drill down rendering bugs offline.

Compatibility

The compatibility is much improved and should be quite high, I won’t claim its perfect, but I’m quite happy with it so far. We went through essentially all relevant titles during testing (just the first few minutes), and found and fixed the few issues which popped up. Many games which were completely broken in the old implementation now work just fine. I’m fairly confident that those bugs are solvable this time around though if/when they show up.

Implementation techniques

With Vulkan in 2020 I have some more tools in my belt than was available back in the day. Vulkan is a quite capable compute API now.

Enforcing RDRAM coherency

A major pain point of any N64 emulator is the fact that RDRAM is shared for the CPU and RDP, and games sure know how to take advantage of this. This creates a huge burden on GPU-accelerated implementations as we now have to ensure full coherency to make it accurate. Most HLE emulators simply don’t care or employ complicated heuristics and workarounds, and that’s fine, but it’s not good enough for LLE.

In the previous implementation, it would try to do “framebuffer manager” techniques similar to HLE emulators, but this was the wrong approach and lead to a design which was impossible to fix. What if … we just import RDRAM as buffer straight into the Vulkan driver and render to that, wouldn’t that be awesome? Yes … yes, it would be, and that’s what I did. We have an obscure, but amazing extension in Vulkan called VK_EXT_external_memory_host which lets me import RDRAM from the emulator straight into Vulkan and render to it over the PCI-e bus. That way, all framebuffer management woes simply disappear, I render straight into RDRAM, and the only thing left to do is to handle synchronization. If you’re worried about rendering over the PCI-e bus, then don’t be. The bandwidth required to write out a 320×240 framebuffer is absolutely trivial especially considering that we’re doing …

Tile-based rendering

The last implementation was tile-based as well, but the design is much improved. This time around all tile binning is done entirely on the GPU in parallel, using techniques I implemented in https://github.com/Themaister/RetroWarp, which was the precursor project for this new paraLLEl-RDP. Using tile-based rendering, it does not really matter that we’re effectively rendering over the PCI-e bus as tile-based rendering is extremely good at minimizing external memory bandwidth. Of course, for iGPU, there is no (?) external PCI-e bus to fight with to begin with, so that’s nice!

Ubershaders with asynchronous pipeline optimization

The entire renderer is split into a very small selection of Vulkan GLSL shaders which are precompiled into SPIR-V. This time, I take full advantage of Vulkan specialization constants which allow me to fine-tune the shader for specific RDP state. This turned out to be an absolute massive win for performance. To avoid the dreaded shader compilation stutter, I can always fallback to a generic ubershader while pipeline is being compiled which is slow, but works for any combination of state. This is a very similar idea to what Dolphin pioneered for emulation a few years ago.

8/16-bit integer support

Memory accesses in the RDP are often 8 or 16 bits, and thus it is absolutely critical that we make use of 8/16-bit storage features to interact directly with RDRAM, and if the GPU supports it, we can make use of 8 and 16-bit arithmetic as well for good measure.

Async compute

Async compute is critical as well, since we can make the async compute queue high priority and ensure that RDP shading work happens with very low latency, while VI filtering and frontend shaders can happily chug along in the fragment/graphics queue. Both AMD and NVIDIA now have competent implementations here.

GPU-driven TMEM management

A big mistake I made previously was doing TMEM management in CPU timeline, this all came crashing down once we needed framebuffer effects. To avoid this, all TMEM uploads are now driven by the GPU. This is probably the hairiest part of paraLLEl-RDP by far, but I have quite a lot of gnarly tests to test all the relevant corner cases. There are some true insane edge cases that I cannot handle yet, but the results created would be completely meaningless to any actual content.

Performance

To talk about FPS figures it’s important to consider the three major performance hogs in a low-level N64 emulator, the VR4300 CPU, the RSP and finally the RDP. Emulating the RSP in an LLE fashion is still somewhat taxing, even with a dynarec (paraLLEl-RSP) and even if I make the RDP infinitely fast, there is an upper bound to how fast we can make the emulator run as the CPU and RSP are still completely single threaded affairs. Do keep that in mind. Still, even with multithreaded Angrylion, the RDP represents a quite healthy chunk of overhead that we can almost entirely remove with a GPU implementation.

GPU bound performance

It’s useful to look at what performance we’re getting if emulation was no constraint at all. By adding PARALLEL_RDP_BENCH=1 to environment variables, I can look at how much time is spent on GPU rendering.

Playing on an GTX 166o Ti outside the castle in Mario 64:

[INFO]: Timestamp tag report: render-pass
[INFO]: 0.196 ms / frame context
[INFO]: 0.500 iterations / frame context

We’re talking ~0.2ms on GPU to render one frame on average, hello theoretical 5000 VI/s … Somewhat smaller frame times can be observed on my Radeon 5700 XT, but we’re getting frame rates so ridiciously high they become meaningless here. We’ve tested it on quite old cards as well and the difference in FPS on even something ancient like an R9 290x card and a 2080 Ti is minimal since the time now spent in RDP rendering is completely irrelevant compared to CPU + RSP workloads. We seem to be getting about a 50-100% uplift in FPS, which represents the shaved away overhead that the CPU renderer had. Hello 300+ VI/s!

Unfortunately, Intel iGPU does not fare as well, with an overhead high enough that it does not generally beat multithreaded Angrylion running on CPU. I was somewhat disappointed by this, but I have not gone into any real shader optimization work. My early analysis suggests extremely poor occupancy and a ton of register spilling. I want to create a benchmark tool at some point to help drill down these issues down the line.

It would be interesting to test on the AMD APUs, but none of us have the hardware handy sadly 🙁

Synchronous vs Asynchronous RDP

There are two modes for the RDP. In async mode, the emulation thread does not wait for the GPU to complete rendering. This improves performance, at the cost of accuracy. Many games unfortunately really rely on the unified memory architecture of the N64. The default option is sync, and should be used unless you have a real need for speed, or the game in question does not need sync.

Here we see an example of broken blob shadows caused by async RDP in Jet Force Gemini. This happens because the CPU is actually reading the shadowmap rendered by the RDP, and blurring it on the CPU timeline (why on earth the game would do that is another question), then reuploading it to the RDP. These kinds of effects require very tight sync between CPU and GPU and comes up in many games. N64 is particularly notorious for these kinds of rendering challenges.

Of course, given how fast the GPU implementation is on discrete GPUs, sync mode does not really pose an issue. Do note that since we’re using async compute queues here, we are not stalling on frontend shading or anything like that. The typical stall times on the CPU is in the order of 1 ms per frame, which is very acceptable. That includes the render thread doing its thing, submitting that to GPU, getting it executed and coming back to CPU, which has some extra overhead.

Road-map for future improvement

I believe this is solid enough for a first release, but there are further avenues for improvement.

Figure out poor performance on Intel iGPU

There is something going on here that we should be able to improve.

Implement a workaround for implementations without VK_EXT_external_memory_host (EDIT: Now implemented as of 2020-05-18)

Unfortunately there is one particular driver on desktop which doesn’t support this, and that’s NVIDIA on Linux (Windows has been supported since 2018 …). Hopefully this gets implemented soon, but we will need a fallback. This will get ugly since we’ll need to start shuffling memory back and forth between RDRAM and a GPU buffer. Hopefully the async transfer queue can help make this less painful. It might also open up some opportunities for mobile, which also don’t implement this extension as we speak. There might also be incentives to rewrite some fundamental assumptions in the N64 emulator plugin specifications (can we please get rid of this crap …). If we can let the GPU backend allocate memory, we don’t need any fancy extension, but that means uprooting 20 years of assumptions and poking into the abyss … Perhaps a new implementation can break new ground here (hi @ares_emu!).

EDIT: This is now done! Takes a 5-10% performance hit in sync mode, but the workaround works quite well. A fine blend of masked SIMD moves, a writemask buffer, and atomics …

Internal upscaling?

It is rather counter-intuitive to do upscaling in an LLE emulator, but it might yield some very interesting results. Given how obscenely fast the discrete GPUs are at this task, we should be able to do a 2x or maybe even 4x upscale at way-faster-than-realtime speeds. It would be interesting to explore if this lets us avoid the worst artifacts commonly associated with upscaling in HLE.

Fancier deinterlacer?

Some N64 content runs at 480i, and we can probably spare some GPU cycles running a fancier deinterlacer 😉

Esoteric use cases?

PS1 wobbly polygon rendering has seen some kind of resurgence in the last years in the indie scene, perhaps we’ll see the same for the fuzzy N64 look eventually. With paraLLEl-RDP, it should be possible to build a rendering engine around a N64-lookalike game. That would be cool to see.

Conclusion

This is a somewhat esoteric implementation, but I hope I’ve inspired more implementations like this. Compute-based accurate renderers will hopefully spread to more systems that have difficulties with accurate rendering. I think it’s a very interesting topic, and it’s a fun take on emulation that is not well explored in general.