Shmupmame 5 – Side effects be gone

[nimitz]

Shmupmame 5.4 (sh3, psx, stv, gradius IV, Cave 68000 and more)

Version 5.0 marks a somewhat new approach to reduce latency, focusing on latency lowering modifications that have no visual side effects or glitches. This means no more background layer de-sync on games with left/right panning when moving sideways. I included in this build most shmups that are playable at relatively low latency (usually means ~33ms delay or less).

Following the release of Shmupmame sh3 I was made aware of some issues with cv1k slowdown in the first few seconds of play of Deathsmiles in the build that wasn't supposed to be present. I tried fixing by modifying the memory timings, but turns out this caused more inaccuracies than it fixed, so the values are back to what they were (in 5.1) and we'll have to live with that minor inconvenience with Deathsmiles for now.

After that I decided to add some drivers to the sh3 build just so it could play most Cave games and noticed that the Cave 68000 games still had relatively high latency, I tried the "cleanest" method of reducing latency (input injection) and it worked very well. I also tried to find games which had early 3D graphics devices which could be de-lagged using a method similar to STV and PS1 hardware, and the Gradius IV hornet driver uses a voodoo1 to which I was able to apply the fix easily. As far as I know this would make it the best way to play Gradius IV as of now.

Finally, earlier builds of shmupmame had a way of enabling a loop2 start of Dodonpachi, but it was through either mame cheats or savestates which was far from ideal, so I took the time to actually make it a proper romset in mame and did the same for some games which I thought could benefit from it, here are the new romsets that enable a loop2 start using the "loop2 shmupmame" clones:
• Batsugun special (Starts at loop 2 with max power, but "level 1")
• Donpachi (jp) (Starts at loop 2 with max power)
• Dodonpachi (Starts at loop 2 at max power)
• Dai-ou-Jou Black Label (also includes WL) (Starts at loop 2 with max power for both shot and laser)
• Ketsui Ura Loop (Starts at loop 2 with max power)
• Ketsui Omote Loop (Starts at loop 2 with max power)
• DaiFukkatsu 1.5 Omote Loop (Starts at loop 2 with max power)


Downloads (source code is in the /src folder):
Shmupmame 5.4 Windows 64
Shmupmame 5.4 Linux x86_64
Shmupmame 5.4 Linux Appimage x86_64

The non Appimage Linux version should "just work" on most linux systems, the AppImage is included as a "last resort" for those who can't get it to run otherwise.
See post below for changes in v5.1

[nimitz]

New in version 5.4:

Code: Select all

 -Added Mars Matrix fixes from trap0xf (https://bitbucket.org/trap15/mmatrix_unfuck/src/master/)
   to all mars matrix roms, further reducing lag, 3-button controls, better sound balance and better reset behavior
	Thanks to trap15 for the permission to distribute his patch!
 -Fixed a bug with the windows version that complained about the loop2 hacks being clones of clones despite working
 -Added the loop2 hacks to the favorites folder since mame search functionality can be pretty bad
 -Fixed a small bug in the cv1k slowdown code (the fix will also be applied to bio_endio's implementation)
    And included the latest improvements from bio_endio

New in version 5.3.1:

Code: Select all

Added bio_endio's accurate slowdown emulationfor cv1k (Big thanks!)

New in version 5.11:

Code: Select all

Corrected issue where Batsugun special and Daifukkatsu Loop 2 modes could be incorrectly initialized on normal sets.

New in version 5.1:

Code: Select all

	-sh3 memory timings back to CoffeePope's build values since most games behave better. Only the first ~30 seconds of Deathsmiles behave oddly. 
		Which seems to be a GFX upload issue (on the cpu side, not blitter side) that only shows up in that specific instance.
	-Stopped injecting Game Tengoku coin/credit switch to prevent weirdness (too many coins)
	-DDP loop 2 start now starts the player at max power as you would if playing normally.
	-DDP loop 2 start now work for both p1 and p2 start.

[nesrulz]

Thx!

[emphatic]

What a gift to the community, kudos.

[copy-paster]

No one expected of all things, Shmupmame 5 is real.

[maximo310]

Wow, this caught me off guard with all of the new drivers/QoL fixes & drivers from Shmupmame 4.2, thank you so much!

Out of curiosity, I remember reading a while back that Stagger I runs on NMK hardware, so I'm wondering if it'd be possible to add it to the game/driver list in the future. I know it's not anywhere as good as other NMK stuff, but I do have a soft spot for putting in credits every once in a while.

[birb]

Seriously, good work on this. Having an emulator that can run shmups at their best is a blessing. While I do dislike the new MAME GUI. I don't mind it as much as I use command line

[BareKnuckleRoo]

We're not worthy, we're not worthy!

Amazing contribution to the community, thank you. <3

[MFC]

i believe sdoj expert with A ship is now missing slowdown, here are two examples of little to no slowdown where the PCB footage shows slowdown (comparing with https://www.youtube.com/watch?v=Z9skBAPlC2U)

stage 3 boss final - https://files.catbox.moe/s7krsv.mp4 (compare to 12:56 in youtube video)
stage 5 first midboss - https://files.catbox.moe/ci204j.mp4 (compare to 20:22 in youtube video)

i have also noticed minor audio stutters that were not in the previous version

[nimitz]

Thanks for kind words guys, I'm happy to know that people will get some enjoyment out of this project!

i believe sdoj expert with A ship is now missing slowdown, here are two examples of little to no slowdown where the PCB footage shows slowdown (comparing with https://www.youtube.com/watch?v=Z9skBAPlC2U)

stage 3 boss final - https://files.catbox.moe/s7krsv.mp4 (compare to 12:56 in youtube video)
stage 5 first midboss - https://files.catbox.moe/ci204j.mp4 (compare to 20:22 in youtube video)

i have also noticed minor audio stutters that were not in the previous version

Thank you! This is exactly the kind of info needed to make the slowdown accurate. I'll check on that "soon" and hopefully have a more general solution for the next version/revision. Please keep it coming if you (or anyone else) finds anything else that doesn't match pcb behavior.

[RZM]

Awesome work, I'll definitely try it!

Just a note about the DDP loop2 rom. I don't know what other ddp lovers think about it but starting at minimum power isn't ideal, since it's significantly harder and not much fun to play stages 1 and 2 in this configuration, which is the worst possible for chaining & survival (equivalent to emulating a loop 1 with a first death on the last boss).

[Jaimers]

Doing God's work man, absolutely incredible advancements.
Return of the king.

Thank you! This is exactly the kind of info needed to make the slowdown accurate. I'll check on that "soon" and hopefully have a more general solution for the next version/revision. Please keep it coming if you (or anyone else) finds anything else that doesn't match pcb behavior.

Yeah it's worth noting that this wasn't an issue with the previous 'Shmupmame sh3' version.
For SDOJ at least so far you still want to stick to that previous version, as the slowdown was pretty decently accurate on that.

[EmperorIng]

I don't know if it's just me but a weird thing with Game Tengoku is pressing the insert credit button autofire spams coins (a light tap can give you 6 or 7, a normal press 15 or more, ha). I dunno if the frame reduction is interfacing with the credit input, since it just seems to be that game (I should look into other Jaleco board roms as well). Not a huge deal but I figured I'd bring it up!

Though overall I am really quite happy and pleased with the release - if I have other things I find I hope you don't mind me chiming in.

I've focused mostly on the "other" stuff, like the PS1, STV, etc. stuff so far, since that's where my interests lie. But what a treat!

[nimitz]

Thanks for the feedback guys, I uploaded new version (5.1) with these fixes:
-sh3 memory timings back to coffeePope's build values since only the first ~30 seconds of Deathsmiles behave oddly. (Seems to be a GFX upload issue that only shows up in the first few seconds of play on deathsmiles)
-Stopped injecting Game Tengoku credit switch to prevent weirdness (too many coins)
-DDP loop 2 start now starts the player at max power as you would if playing normally and now work for both p1 and p2 start.

Out of curiosity, I remember reading a while back that Stagger I runs on NMK hardware, so I'm wondering if it'd be possible to add it to the game/driver list in the future. I know it's not anywhere as good as other NMK stuff, but I do have a soft spot for putting in credits every once in a while.

Stagger I is included in the build (since 5.0)

[RadiantAnsel]

We started this year really strong with this

[joeh]

This is awesome. Appreciate the releases!

[Feedbacker]

Thank you nimitz for your hard work.

For other hardwares than the ones you listed, does this build still have advantages over official MAME with enabled low latency?
Also, may I ask you what "input injection" is?

[KMH]

Something is likely buggy about the way you implemented the batsugun loop2 start hack.

When I start my normal demo recording/playback command line shortcuts for both Batsugun and Special Version, it often boots straight into the loop2 start version. Even in normal Batsugun which doesn't have a second loop, it will start at full power some of the time. Happens in both 5.0 and 5.1.

I thought it was related to me using -nonvram_save on most of my shortcuts, but removing it doesn't fix the issue (and it doesn't seem like batsugun saves nvram anyway). So i can't really find a consistent pattern, it just seems random. Sometimes you can record a demo fine, and when you play it back it's the loop2 start version, and vice versa. I did try deleting batsugspl2.zip from the roms folder also, thinking that may have something to do with it, but it didn't change anything.

This type of issue also hasn't happened yet for me in donpachi, dodonpachi, doj, or ketsui. They all play normally from command line shortcuts so far (but i haven't tested them that much).

Edit: Just realized i might have overcomplicated things. The demo recording/playback part is likely not relevant, because even just a normal shortcut to the shmupmame exe with "batsugunsp" in it and nothing else, launches straight into the loop2 start.

[nimitz]

For other hardwares than the ones you listed, does this build still have advantages over official MAME with enabled low latency?
Also, may I ask you what "input injection" is?

A good example would be the seibu drivers, those are fine with standard mame with the "lowlatency" mode/flag enabled and they are the exact same as the ones included in this build. I included them for convenience and as a curated list of drivers/shmups that are playable (with low enough latency).

As for input injection, it is basically sending the most "fresh" controller input data to the game at the right time. This is in a way part of what the mame lowlatency flag is doing (waiting before processing/emulating a frame) but some games are coded in such a way that fresh inputs are never used in the same frame, sometimes due to obvious coding errors that cause multiple frame delays (e.g guardian force) sometimes just a bad ordering of events that adds a single frame. So in short, in only helps for drivers that have input processing "bugs" or that are too computationally heavy for frame delay to be a big help.

KMH:
Thanks for the report, it was an omission on my part, the same could also happen with dfk. I'll release a fixed build soon.

[KV344]

Really appreciate your great work! The glsl shader included also looks great

Well, this might be a dumb question... Is there any extra fiddling needed to utilize the input lag improvements you incorporated, or do I just turn on "low latency mode" like in regular MAME and be done with it?
Thanks in advance!

[blackrabite]

Haven't logged in here for quite some time, just wanted to say "thank you very much" for your work on this. I really appreciate your focus on putting together a package that works well out the gates and for documenting your implementations/changes for the various drivers. Motivated by this release (and a modest local STG scene that has been picking up), I've been playing more lately and got my first 1-All for DOJ BL on release 5.1 just a few nights ago! Cheers.

[nimitz]

There's been quite a few new developments in the past few weeks. A pretty cool guy named bio_endio has took upon himself to take what coffeepope had started about a year ago and actually finish it.

The results are pretty amazing and as far as most can tell fully accurate slowdown for cv1k, I included this code in the new version (shmupmame 5.3) which should make a huge difference in terms of accurate slowdown for those games!

[emphatic]

There's been quite a few new developments in the past few weeks. A pretty cool guy named bio_endio has took upon himself to take what coffeepope hard started about a year ago and actually finish it.

The result is pretty amazing and as far as most can tell fully accurate slowdown for cv1k, I included this code in the new version (shmupmame 5.3) which should make a huge difference in terms of accurate slowdown for those games!

Wow, very cool.

[Aquas]

Wow, didn't know G IV was playable now. What a time to be alive. This is shifting the focus away from 360 ports - for MMP and PS for me. MAME revenge!

[blackrabite]

Thank you so much, this rocks! I can't wait to test it out this evening. Also, thanks for providing the binaries - my s.o. plays on Linux so we really appreciate the ease of access to your releases. Have a great weekend.

[nimitz]

Quick update, version 5.3.1 released, which fixed a slowdown related issue specific to shmupmame 5.3

[Bladeforce]

Hi, THanks for this but i'm finding the appimage is extremely laggy. WHen i load it the cursor is very slow and keyboard responsiveness is very slow too

[KV344]

Again, thanks for your efforts benefitting the community!

I want to report some minor imperfections on Linux... While Windows version runs perfectly smooth with Ryzen 7 5800H, in Linux (BazziteOS) on a GPD Win Mini 8840U audio will run very jittery during heavy "cinematic" slowdowns in SDOJ (like explosions after defeating a boss), which was not present in 5.1.1 build. I can't comment on gameplay/bullet count induced slowdown accuracy, but at least in those parts audio behaves normally.
Both standard Linux executable build and Appimage seem to function exactly the same. I hadn't run into issues regarding keyboard/cursor response problems like another commenter mentioned eariler

Not too big of a deal, but us Linux users will greatly appreciate it if this could be improved....

[nimitz]

Quick update, version 5.4 is released which fixes a small bug with the windows version when playing the loop2 hacks. (an error popup appeared in the background but the games ran/worked anyway)

Also added the Mars matrix Improvement patch from trap15 to the mars matrix roms, so that loading any version of mars matrix applies the patch. The patch improves the latency, sound balance, adds 3 button control (was already un shmupmame but I'm using his patch as-is) and better reset behavior.

And also the best and latest cv1k slowdown implementation from bio_endio is included. Please report any slowdown discrepancies, but other than sdoj that is not quite 100% perfect, The slowdown behavior for the other games should be extremely close to the real thing!

Hi, THanks for this but i'm finding the appimage is extremely laggy. WHen i load it the cursor is very slow and keyboard responsiveness is very slow too

I'd venture to guess this is a combination of a high display resolution and a "slow" gpu. Try standard mame Appimage and report if the same behavior is observed.

I want to report some minor imperfections on Linux... While Windows version runs perfectly smooth with Ryzen 7 5800H, in Linux (BazziteOS) on a GPD Win Mini 8840U audio will run very jittery during heavy "cinematic" slowdowns in SDOJ (like explosions after defeating a boss), which was not present in 5.1.1 build. I can't comment on gameplay/bullet count induced slowdown accuracy, but at least in those parts audio behaves normally.
Both standard Linux executable build and Appimage seem to function exactly the same. I hadn't run into issues regarding keyboard/cursor response problems like another commenter mentioned eariler

Not too big of a deal, but us Linux users will greatly appreciate it if this could be improved....

The newer versions of the cv1k slowdown are a bit heavier on the cpu, part of the "cost" of making things accurate. Maybe try running the windows version on that linux box with wine and report if there's a difference in performance.

[joeh]

Just tried Futari on the Windows 5.4 release. Game ran very slow. Almost around half speed? Went back to 5.31 and didn't have any issues.