5 ms·

is this allowed for speedruns

danaris 5y ago
Allowed? Maybe, maybe not. Helpful? Extremely not. I just ran this on my high-end M1 Max, and it rendered 2 frames in roughly 3 minutes. Hardware-accurate emulation is, generally speaking, not fast emulation.
- rl3 5y ago
  Hardware-accurate emulation in software is typically not fast. FPGAs however can achieve low-level emulation with a high degree of hardware accuracy: https://github.com/MiSTer-devel/Main_MiSTer/wiki https://github.com/MiSTer-devel/Main_MiSTer/wiki
  - monocasa 5y ago
    The 6502 unfortunately relies on some effects that can't be described in FPGA synthesizable logic. You might get a lot of speed up over the extremely simplified spice style simulation you have to do that still might get you to real time, but it's not the slam dunk you might think.
    - russdill 5y ago
      Sounds interesting. Any citations where I can read more about those effects?
      - monocasa 5y ago
        You can sort of divine them from how the undocumented instructions work. https://www.pagetable.com/?p=39 https://www.pagetable.com/?p=39 It more or less falls under true classic highz buses with multiple drivers. You have three options I can see 1) treat it as a spice simulation that aggressively optmizes to synthesizable logic most places other than the problem parts. 2) hand replace the problem parts with synthesizable logic sorta like how Nvidia replaces graphics shaders with had written replacements in their drivers 3) just throw the whole thing into a spice simulation
      - tenebrisalietum 5y ago
        I think the decrement relies on open bus behavior I think - the 6502 uses it to cheat and set a value to FF in the same cycle which is then sent through the same path as increment or add, effectively adding -1 to a value.
    - comex 5y ago
      Out of curiosity, what effects are you talking about?
      - monocasa 5y ago
        The internal buses are basically true highz buses with multiple drivers and latches hanging off of them.
        russdill 5y ago
        That shouldn't be a problem to implement on an FPGA with a tiny bit of extra logic.
        monocasa 5y ago
        Totally agreed, but you can't just throw the netlist at a fpga and expect the synthesizer to be happy is all I'm saying.
  - stormbrew 5y ago
    I mean, aside from the question of whether fpgas can truly accurately simulate everything at the transistor level, the truth is that most fpga emulation cores for this sort of thing don't. They are, afaik, still mostly 'just' cycle-level accurate emulations built off the same reverse engineering that accurate software emulators are. They can do it more efficiently for sure, but that doesn't mean it's worth it for them to go much farther. Like, for an extreme example there's no world in which the in-development PSX mister core, running on a de10 nano, is particularly more accurate than a software emulator. The chips involved have too many transistors to completely simulate. But this will be true to some degree for everything above some complexity threshold. Maybe the Atari 2600 core is perfectly accurate. I wouldn't be surprised if the most accurate software emulators are more accurate than the most accurate fpga cores, especially for anything newer than the NES, just because that's where most of the original work involved happens. This idea that FPGAs are inherently more accurate seems to be rooted in marketing FUD from a company called Analogue.
    - philistine 5y ago
      I’ll 100% give you the marketing spin, but FPGAs provide an element of accuracy with frame delay. If you plug in a controller and use a CRT, there no more lost frames than the original hardware. That’s the accuracy that’s above what an emulator on our Babel’s tower of a stack on Windows.
      - stormbrew 5y ago
        That's not because of the fpga though. You can also plug a CRT into a PC. You can even chase the beam if you're willing to go to a low enough level. People generally don't but this is a limitation born advantage. There's no OS (sort of) to get in your way and why would you put one there if it's not helping you. (I say sort of because there usually is a system that could be described as an OS running somewhere on the fpga board, to load the fpga bitstreams, if nothing else, but also sometimes to supplement the fpga's abilities)
        philistine 5y ago
  - taneq 5y ago
    Is it still really emulation at that stage? Or is it more like a reimplementation?
teaearlgraycold 5y ago
I think it'd be considered TAS if you report your time by frame count. Slowing down an emulator is a TAS.
lanewinfield 5y ago
yes, they're called slowruns

Allowed? Maybe, maybe not. Helpful? Extremely not. I just ran this on my high-end M1 Max, and it rendered 2 frames in roughly 3 minutes. Hardware-accurate emulation is, generally speaking, not fast emulation.

rl3 5y ago
Hardware-accurate emulation in software is typically not fast. FPGAs however can achieve low-level emulation with a high degree of hardware accuracy: https://github.com/MiSTer-devel/Main_MiSTer/wiki https://github.com/MiSTer-devel/Main_MiSTer/wiki
- monocasa 5y ago
  The 6502 unfortunately relies on some effects that can't be described in FPGA synthesizable logic. You might get a lot of speed up over the extremely simplified spice style simulation you have to do that still might get you to real time, but it's not the slam dunk you might think.
  - russdill 5y ago
    Sounds interesting. Any citations where I can read more about those effects?
    - monocasa 5y ago
      You can sort of divine them from how the undocumented instructions work. https://www.pagetable.com/?p=39 https://www.pagetable.com/?p=39 It more or less falls under true classic highz buses with multiple drivers. You have three options I can see 1) treat it as a spice simulation that aggressively optmizes to synthesizable logic most places other than the problem parts. 2) hand replace the problem parts with synthesizable logic sorta like how Nvidia replaces graphics shaders with had written replacements in their drivers 3) just throw the whole thing into a spice simulation
    - tenebrisalietum 5y ago
      I think the decrement relies on open bus behavior I think - the 6502 uses it to cheat and set a value to FF in the same cycle which is then sent through the same path as increment or add, effectively adding -1 to a value.
  - comex 5y ago
    Out of curiosity, what effects are you talking about?
    - monocasa 5y ago
      The internal buses are basically true highz buses with multiple drivers and latches hanging off of them.
      - russdill 5y ago
        That shouldn't be a problem to implement on an FPGA with a tiny bit of extra logic.
        monocasa 5y ago
        Totally agreed, but you can't just throw the netlist at a fpga and expect the synthesizer to be happy is all I'm saying.
- stormbrew 5y ago
  I mean, aside from the question of whether fpgas can truly accurately simulate everything at the transistor level, the truth is that most fpga emulation cores for this sort of thing don't. They are, afaik, still mostly 'just' cycle-level accurate emulations built off the same reverse engineering that accurate software emulators are. They can do it more efficiently for sure, but that doesn't mean it's worth it for them to go much farther. Like, for an extreme example there's no world in which the in-development PSX mister core, running on a de10 nano, is particularly more accurate than a software emulator. The chips involved have too many transistors to completely simulate. But this will be true to some degree for everything above some complexity threshold. Maybe the Atari 2600 core is perfectly accurate. I wouldn't be surprised if the most accurate software emulators are more accurate than the most accurate fpga cores, especially for anything newer than the NES, just because that's where most of the original work involved happens. This idea that FPGAs are inherently more accurate seems to be rooted in marketing FUD from a company called Analogue.
  - philistine 5y ago
    I’ll 100% give you the marketing spin, but FPGAs provide an element of accuracy with frame delay. If you plug in a controller and use a CRT, there no more lost frames than the original hardware. That’s the accuracy that’s above what an emulator on our Babel’s tower of a stack on Windows.
    - stormbrew 5y ago
      That's not because of the fpga though. You can also plug a CRT into a PC. You can even chase the beam if you're willing to go to a low enough level. People generally don't but this is a limitation born advantage. There's no OS (sort of) to get in your way and why would you put one there if it's not helping you. (I say sort of because there usually is a system that could be described as an OS running somewhere on the fpga board, to load the fpga bitstreams, if nothing else, but also sometimes to supplement the fpga's abilities)
      - philistine 5y ago
- taneq 5y ago
  Is it still really emulation at that stage? Or is it more like a reimplementation?

teaearlgraycold 5y ago

I think it'd be considered TAS if you report your time by frame count. Slowing down an emulator is a TAS.

lanewinfield 5y ago

yes, they're called slowruns