4 ms·

Nintendo 64 Vulkan Low-Level emulator – paraLLel – pre-alpha release

19 points by libretro 10y ago

Jasper_ 10y ago
Should probably make an effortpost on this to explain what "low-level emulation" is. That said, I don't support the libretro effort. The Nintendo 64 is an architecture composed of a 64-bit MIPS CPU, 4MB of memory, and, two chips for the GPU: the Reality Signal Processor (RSP), and the Reality Display Processor (RDP). The codename for the N64 was "Reality", and the system was co-developed by Nintendo and a small team at SGI. The RSP is a programmable unit, while the RDP is a fixed-function unit that does rasterization, perspective texture mapping, and basic glTexEnv-style multitexturing. When N64 emulators talk about "high-level" vs. "low-level", they're pretty exclusively talking about how the RSP is implemented. The RSP is programmable in that you can upload your own microcodes and provide your own instruction set in return. When an RSP emulator interprets these microcodes directly, it's "low-level". When an emulator author studies these microcodes, and then reimplements them in C/C++, this is "high-level". There are around 15 microcodes in existence, the most common one being F3DEX2, which was the standard Nintendo-provided microcode which appeared a year after the system was released. You can see the instruction set of F3DEX2 here: http://wiki.cloudmodding.com/oot/F3DZEX http://wiki.cloudmodding.com/oot/F3DZEX Very few developers wrote their own microcodes -- pretty much only Rareware and Factor 5 did their own thing. This is one reason why Rareware and Factor 5 games are harder to emulate. Most emulators only implement low-level mode for compatibility and testing purposes, since it tends to be pretty slow. paraLLel uses Vulkan and Compute Shaders to run the RSP microcode on the GPU. I'm not convinced that's a good approach, but good on them for trying.
- Fej 10y ago
  It's, at the very least, a cool, novel use for Vulkan.
- anonbanker 10y ago
  why don't you support it?
  - Jasper_ 10y ago
    The technical reason is that the microcode isn't designed to be run in parallel -- it does DMA transfers between main memory and texture memory, and such. I don't understand why running it on the GPU would help anybody. For personal / drama reasons, the main author of libretro (squarepusher / twinaphex) has a bad track record in the emulation community, but this isn't the time or place to talk about it. Just know that not a lot of people in the scene hold a very high opinion of him.

19 points by libretro 10y ago

Should probably make an effortpost on this to explain what "low-level emulation" is. That said, I don't support the libretro effort. The Nintendo 64 is an architecture composed of a 64-bit MIPS CPU, 4MB of memory, and, two chips for the GPU: the Reality Signal Processor (RSP), and the Reality Display Processor (RDP). The codename for the N64 was "Reality", and the system was co-developed by Nintendo and a small team at SGI. The RSP is a programmable unit, while the RDP is a fixed-function unit that does rasterization, perspective texture mapping, and basic glTexEnv-style multitexturing. When N64 emulators talk about "high-level" vs. "low-level", they're pretty exclusively talking about how the RSP is implemented. The RSP is programmable in that you can upload your own microcodes and provide your own instruction set in return. When an RSP emulator interprets these microcodes directly, it's "low-level". When an emulator author studies these microcodes, and then reimplements them in C/C++, this is "high-level". There are around 15 microcodes in existence, the most common one being F3DEX2, which was the standard Nintendo-provided microcode which appeared a year after the system was released. You can see the instruction set of F3DEX2 here: http://wiki.cloudmodding.com/oot/F3DZEX http://wiki.cloudmodding.com/oot/F3DZEX Very few developers wrote their own microcodes -- pretty much only Rareware and Factor 5 did their own thing. This is one reason why Rareware and Factor 5 games are harder to emulate. Most emulators only implement low-level mode for compatibility and testing purposes, since it tends to be pretty slow. paraLLel uses Vulkan and Compute Shaders to run the RSP microcode on the GPU. I'm not convinced that's a good approach, but good on them for trying.

Fej 10y ago
It's, at the very least, a cool, novel use for Vulkan.
anonbanker 10y ago
why don't you support it?
- Jasper_ 10y ago
  The technical reason is that the microcode isn't designed to be run in parallel -- it does DMA transfers between main memory and texture memory, and such. I don't understand why running it on the GPU would help anybody. For personal / drama reasons, the main author of libretro (squarepusher / twinaphex) has a bad track record in the emulation community, but this isn't the time or place to talk about it. Just know that not a lot of people in the scene hold a very high opinion of him.