#jsmess 2022-01-03,Mon

↑back Search

Time Nickname Message
00:20 🔗 DFJustin has joined #jsmess
04:46 🔗 SketchCo2 I've got the Arcade Repair set down to 223 and dropping.
10:10 🔗 Sgeo has quit IRC (Read error: Connection reset by peer)
16:03 🔗 dboard420 has quit IRC (Read error: Operation timed out)
16:03 🔗 dboard420 has joined #jsmess
18:48 🔗 Sgeo has joined #jsmess
20:01 🔗 SketchCo2 Sharing for DFJustin and others:
20:01 🔗 SketchCo2 https://docs.google.com/spreadsheets/d/1v5PgyIEDkXbK8jtBNSKTW-nmPTwqdNUQ8_ElKWF1jsk/edit?usp=sharing
20:07 🔗 SketchCo2 Uncaught in Promise are the straight-up blow-ups.
20:07 🔗 SketchCo2 If I fix them, they leave this list.
20:42 🔗 DFJustin so there is some discussion about enabling threading in the mame emscripten build, that would require some server side changes to send the correct cross origin headers as per https://emscripten.org/docs/porting/pthreads.html
20:43 🔗 DFJustin could provide some speed benefits
22:18 🔗 SketchCo2 We can always do 39in1 as a test, and then talk to the devs when it works out to see what's being blocked
22:26 🔗 SketchCo2 I'm continuing to find working CHDs and the amount of brokes in that google docs continues.
22:26 🔗 SketchCo2 Hopefully you will see trends in what's causing the promises
22:26 🔗 SketchCo2 Happy to recompile via whatever tricks you want
22:27 🔗 DFJustin I'd have to make some changes to the codebase first
22:33 🔗 balrog DFJustin, SketchCo2: gtfore06 (or anything that uses the mips3 drc) will not work without -nodrc, and with -nodrc performance will be terrible
22:33 🔗 balrog I've been trying to track down the bugs that cause the C DRC interpreter to fail, but it's ... difficult
22:33 🔗 balrog even with the C DRC interpreter, performance will likely still be just as bad as with -nodrc, though
22:38 🔗 SketchCo2 So, feel free to look over the list of what I've put up, and add a note in the D column
22:39 🔗 SketchCo2 I mean, first, there are SLOW machines, which is one issue, and then there are CRASHING machines.
22:39 🔗 SketchCo2 For the record, the most "important" to me are the blitz machines, as I'm close friends with the lead dev, and he would love to see this emulated.
22:42 🔗 balrog yeah, and those use mips3. :(
22:43 🔗 SketchCo2 so, you're saying they need the -nodrc line or they'll crash
22:43 🔗 balrog almost certainly. I'm chasing this bug and am not getting very far (so far)
22:43 🔗 balrog this CPU core relies on recompiling instructions to x86 for performance; I'll do a quick performance test on my M1 Mac (where there is no DRC) to see what speeds I get
22:44 🔗 balrog CPU emulation core**
22:44 🔗 SketchCo2 I mean, tha'ts fine
22:44 🔗 balrog yeah, can't really do that in emscripten
22:44 🔗 SketchCo2 I'm fine with all of it. I'm trying to just limit DFJustin chasing pain
22:44 🔗 balrog (the DRC part)
22:44 🔗 balrog I'll at least give y'all a performance baseline
22:44 🔗 SketchCo2 If you could go on that list on the google doc and add to Column D "add -nodrc"
22:45 🔗 balrog I just did
22:45 🔗 balrog for the two that I know off the top of my head
22:46 🔗 SketchCo2 Doing a blitz test
22:56 🔗 balrog blitz, using an optimized release build: without -nodrc I get a hard lock, with it I get performance around 50-60%
22:57 🔗 SketchCo2 So is there something in the code I can find that tells me which of these locks I have can use NODRC
22:57 🔗 DFJustin you can probably just add -nodrc always, it has no effect if the CPU has no DRC like a Z80
22:58 🔗 balrog it depends — drc c interpreter may be faster than -nodrc for some systems
22:58 🔗 balrog but ... it's completely broken for mips3
22:58 🔗 DFJustin it's been slower in my experience but I suppose there could be one out there
22:58 🔗 SketchCo2 Well, again, I'm trying to reduce your time wasted
22:58 🔗 balrog if the driver has a #include "cpu/mips/mips3.h" — yeah, you'll need -nodrc
22:59 🔗 SketchCo2 So if after I'm done, I have a pile of nodrc to add, or even looking for mips in the driver, that works.
22:59 🔗 balrog DFJustin: hmmmm, good to know
23:00 🔗 SketchCo2 aleck64.cpp:#include "cpu/mips/mips3.h"
23:00 🔗 SketchCo2 atlantis.cpp:#include "cpu/mips/mips3.h"
23:00 🔗 SketchCo2 caprcyc.cpp:#include "cpu/mips/mips3.h"
23:00 🔗 SketchCo2 crimson.cpp:#include "cpu/mips/mips3.h"
23:00 🔗 SketchCo2 decstation.cpp:#include "cpu/mips/mips3.h"
23:00 🔗 SketchCo2 hng64.cpp:#include "cpu/mips/mips3.h"
23:00 🔗 SketchCo2 iteagle.cpp:#include "cpu/mips/mips3.h"
23:00 🔗 SketchCo2 kinst.cpp:#include "cpu/mips/mips3.h"
23:00 🔗 SketchCo2 kpython.cpp:#include "cpu/mips/mips3.h"
23:00 🔗 SketchCo2 magictg.cpp:#include "cpu/mips/mips3.h"
23:00 🔗 SketchCo2 n64.cpp:#include "cpu/mips/mips3.h"
23:00 🔗 SketchCo2 namcops2.cpp:#include "cpu/mips/mips3.h"
23:00 🔗 SketchCo2 namcos23.cpp:#include "cpu/mips/mips3.h"
23:00 🔗 SketchCo2 ncdmips.cpp:#include "cpu/mips/mips3.h"
23:00 🔗 SketchCo2 o2.cpp:#include "cpu/mips/mips3.h"
23:00 🔗 SketchCo2 octane.cpp:#include "cpu/mips/mips3.h"
23:00 🔗 SketchCo2 ps2sony.cpp:#include "cpu/mips/mips3.h"
23:00 🔗 SketchCo2 seattle.cpp:#include "cpu/mips/mips3.h"
23:01 🔗 SketchCo2 vegas.cpp:#include "cpu/mips/mips3.h"
23:01 🔗 SketchCo2 vp101.cpp:#include "cpu/mips/mips3.h"
23:01 🔗 SketchCo2 And from that
23:01 🔗 SketchCo2 biofreak.json: "wasm_filename": "mameseattle.wasm.gz",
23:01 🔗 SketchCo2 blitz2k.json: "wasm_filename": "mameseattle.wasm.gz",
23:01 🔗 SketchCo2 blitz99.json: "wasm_filename": "mameseattle.wasm.gz",
23:01 🔗 SketchCo2 blitz.json: "wasm_filename": "mameseattle.wasm.gz",
23:01 🔗 SketchCo2 calspeed.json: "wasm_filename": "mameseattle.wasm.gz",
23:01 🔗 SketchCo2 carnevil.json: "wasm_filename": "mameseattle.wasm.gz",
23:01 🔗 SketchCo2 hyprdriv.json: "wasm_filename": "mameseattle.wasm.gz",
23:01 🔗 SketchCo2 mace.json: "wasm_filename": "mameseattle.wasm.gz",
23:01 🔗 SketchCo2 sfrush.json: "wasm_filename": "mameseattle.wasm.gz",
23:01 🔗 SketchCo2 sfrushrk.json: "wasm_filename": "mameseattle.wasm.gz",
23:02 🔗 SketchCo2 vaportrx.json: "wasm_filename": "mameseattle.wasm.gz",
23:02 🔗 SketchCo2 wg3dh.json: "wasm_filename": "mameseattle.wasm.gz",
23:02 🔗 SketchCo2 So we start to solve the mysteries. That's all I want to do.
23:04 🔗 balrog chances are these machines work at a playable rate... on native x86_64.
23:30 🔗 SketchCo2 Are you sure it's "-nodrc"
23:32 🔗 SketchCo2 {
23:32 🔗 SketchCo2 "name": "wg3dh",
23:32 🔗 SketchCo2 "arcade": "1",
23:32 🔗 SketchCo2 "bios_filenames": [
23:32 🔗 SketchCo2 ""
23:32 🔗 SketchCo2 ],
23:32 🔗 SketchCo2 "peripherals": [
23:32 🔗 SketchCo2 ""
23:32 🔗 SketchCo2 ],
23:32 🔗 SketchCo2 "native_resolution": [
23:32 🔗 SketchCo2 640,
23:32 🔗 SketchCo2 480
23:32 🔗 SketchCo2 ],
23:32 🔗 SketchCo2 "extra_args=": [
23:32 🔗 SketchCo2 ""
23:32 🔗 SketchCo2 ],
23:32 🔗 SketchCo2 "driver": "wg3dh",
23:32 🔗 SketchCo2 "wasmjs_filename": "mameseattle.js.gz",
23:32 🔗 SketchCo2 "wasm_filename": "mameseattle.wasm.gz",
23:33 🔗 SketchCo2 "file_locations": {
23:33 🔗 SketchCo2 "mamesegag80r.wasm": "mameseattle.wasm.gz"
23:33 🔗 SketchCo2 }
23:33 🔗 SketchCo2 }
23:33 🔗 SketchCo2 So, here, I was putting -nodrc in the extra_args
23:33 🔗 SketchCo2 https://archive.org/details/arcade_wg3dh is clearly the best test, with a mere 87mb of CHD
23:35 🔗 SketchCo2 "extra_args=": [
23:35 🔗 SketchCo2 "-nodrc"
23:35 🔗 SketchCo2 ],
23:35 🔗 SketchCo2 I used this, and it is still blowing up
23:35 🔗 SketchCo2 The new blowup is definitely different.
23:36 🔗 DFJustin extra_args shouldn't have an =
23:36 🔗 DFJustin I feel like we've been through this a few times
23:36 🔗 SketchCo2 Well
23:37 🔗 SketchCo2 The main reason here is that these are all the CHD-ified machines that I just literally dropped the tools and tooling on to walk over to where effort was really needed.
23:37 🔗 SketchCo2 Like, some of these were just the CHD uploaded and then walked away when we couldn't get CHDs to load
23:37 🔗 SketchCo2 So I'm already forwarding 10-20 steps before you see them
23:37 🔗 SketchCo2 (adding ROMs, getting other configs right, etc.)
23:37 🔗 SketchCo2 But when they hit the promise, I have to stop
23:38 🔗 SketchCo2 That's mostly why. Luckily, we're down to, like, 200 items and even these are falling
23:45 🔗 SketchCo2 emularity_engine_v1 is definitely in need of a cleanup.
irclogger-viewer