Update atari800 core from 3.1.0 to 5.2.0

[WizzardSK]

Summary

• Replace atari800/src/ with standalone atari800 5.2.0 source (from 3.1.0)
• Re-apply all libretro-specific patches (state save, sound, input, UI, config)
• Fix crash when TV mode changes before sound initialization (guard POKEYSND_Init against Sound_out.freq==0)

Changes

• New files: cartridge_info.c/h, file_export.c/h, altirra_5200_charset.c, codecs/, libatari800/, pokeyrec.c/h
• Removed: emuos.c/h (replaced by EMUOS_ALTIRRA), sndsave.c/h
• Config: PACKAGE_VERSION 3.1.0 → 5.2.0, added EMUOS_ALTIRRA=1

[warmenhoven]

https://git.libretro.com/libretro/libretro-atari800/-/pipelines/64699

Looks like most failures are related to termios.h

[WizzardSK]

Thanks for the pointer! Fixed in 1d7a65f — libretro/config.h was defining HAVE_TERMIOS_H unconditionally. Now guarded against Windows and devkitPro targets (PSP, Vita, 3DS, GC/Wii, Wii U, Switch) where termios.h is either missing or transitively pulls in a nonexistent sys/termios.h.

[KamiLordus]

I don't think .a52 files for the Atari 5200 system work. They don't boot. At least on Win 11.

[warmenhoven]

Only one remaining build failure for WiiU missing nanosleep

https://git.libretro.com/libretro/libretro-atari800/-/jobs/1295954

[WizzardSK]

Pushed 69e3f38 — #undef HAVE_NANOSLEEP for WiiU so Util_sleep() falls back to usleep() (same approach as VITA/PSP/PS3). devkitPPC declares nanosleep in <time.h> but doesn't provide the symbol at link time.

[KamiLordus]

@WizzardSK Have you tested the Atari 5200 ROMs?
After testing new build on Windows 11, the .a52 and .bin files have a black screen.
Could you check if it works for you?

[WizzardSK]

Yes, I have tested Atari 5200 games yesterday on Linux and I also having black screen.
I only tested Atari 800 games before, so we need further testing.

edit: I think I found the bug. Tomorrow I will try to do the fix.

[WizzardSK]

Fix for Atari 5200 black screen regression introduced by this PR (commit bc35963).

Bug

Replacing atari800/src/ wholesale dropped the libretro-specific patch in cartridge.c from 2017 (commit 1ac0cbc, "autodetect a52 cart type"), which used atari5200_hash.h CRCs to force the right CARTRIDGE_5200_* type. Without it, raw .a52/.bin images of common sizes (16K, 32K, 64K, 128K, 256K, 512K) match multiple entries in CARTRIDGES[] — so CARTRIDGE_ReadImage() returns CARTRIDGE_UNKNOWN, Atari800_Initialise() falls into UI_SelectCartType() (atari.c:904), and the first-boot libretro UI can't answer → black screen. Matches the long-standing upstream report #79.

Fix (no change to atari800/src/)

• libretro/carts_hash.{c,h} — new get_5200_cart_atari800_type(crc, size) maps internal a5200_* categories (+ size-based fallback for unlisted .a52) to CARTRIDGE_5200_* enums.
• libretro/libretro-core.c — compute CRC for .a52 too, store the resolved type in autorun5200CartType.
• libretro/cmdline.c — for autorunCartridge == A5200_CART, invoke atari800 with -5200 -cart-type <N> -cart <file> instead of the positional prg <file>. CARTRIDGE_Initialise() then preserves the commandline-specified type (see InitInsert() at cartridge.c:1856) even when CARTRIDGE_ReadImage() returns UNKNOWN.

Test

5200 Menu (USA) (Proto).a52 (8 KB → CARTRIDGE_5200_8) boots cleanly and shows the Atari 5200 splash screen in RetroArch (180-frame --max-frames-ss screenshot confirms). Before: black screen.

[WizzardSK]

Follow-up fix: Bounty Bob Strikes Back! (5200) was crashing with CIM at $0007 (commit 3a6b4c5).

The TOSEC dump CRC32 `0x7873c6dd` has an alternative ROM layout: the fixed 8K block is at image offset 0 (mapped to $8000–$BFFF), with the two banked 16K regions shifted by +0x2000. Running it through the standard `CARTRIDGE_5200_40` (which expects the fixed 8K at image offset 0x8000) produces garbage at the reset vector → CIM → Atari800 crash monitor.

This was already handled in the 3.1.0 fork by PR #106 via a dedicated `CARTRIDGE_5200_40_ALT` cart type (slot 73 back then) with its own `MapActiveCart()` case and shifted offsets in the Bounty Bob bank-switch handlers. That custom cart type was lost when the core was replaced wholesale with upstream 5.2.0 in this PR.

Re-added at slot 104 (first unused slot in upstream 5.2.0's enum):

• `atari800/src/cartridge_info.h` — `CARTRIDGE_5200_40_ALT = 104`
• `atari800/src/cartridge_info.c` — `CARTRIDGES[104]` entry (40 KB)
• `atari800/src/cartridge.c` — ALT case in `CartIsFor5200()`, cold mapping in `MapActiveCart()` (fixed 8K from image+0, banks from image+0x2000 and image+0x6000), runtime bank-switch offset handling in `access_BountyBob1()/access_BountyBob2()`
• `libretro/carts_hash.c` — `a5200_40_ALT` now maps to `CARTRIDGE_5200_40_ALT` instead of `CARTRIDGE_5200_40`

Verified: `Bounty Bob Strikes Back! (USA).a52` boots to the title screen (Big Five Software copyright 1985) in RetroArch. Before: black screen on boot, then Atari computer crash monitor.

[KamiLordus]

@WizzardSK Unfortunately, all versions of this game still don't work properly:
Atari 5200 version (.a52, .bin files) - freezes in the menu
Atari 800 cartridge version (.car file) - freezes in the menu
Atari 800 tape version (.cas file) - no sound in the game

This is a problem that also occurred in 3.1. But in the standalone 5.2 version, the games work fine.

[KamiLordus]

After some research, I found information that the game uses some form of protection. This likely applies to the Atari 800 and Atari 5200 cartridge versions. I don't think the tape version has protection. The sound issue is likely caused by something else. I don't know if this will be useful (information from the Atariage forum):

" bounty bob uses custom banking scheme, developed for protecting atari 2600 games
repeated pia adresses - resulting in what you can observe here repeated pokey adresses - resulting in lazy birds and game unplayable at all if there is something more than single pokey on the bus if there was a vast amount of games that would use that banking scheme, or coding "standards" this would be noticed, and fixed, but there are none the other example i know of was road race, and i took care to make it playable when releasing u1mb board according to all available literature pokey address range is 0xd200-0xd20f, PIA is 0xd300-0xd303, remaining space is simply the result of address decoder being cheap, and there is no advantage of using remaining space for addressing these chips other than obscuring the code".

https://forums.atariage.com/topic/314050-bounty-bob-strikes-back-problem/#comments

[WizzardSK]

@KamiLordus Identified the root cause and pushed two follow-up commits.

Root cause

The 3.1.0 fork's cartridge.c had an explicit POKEYSND_stereo_enabled = FALSE for Bounty Bob with the comment "game locks" (PR #106). When this PR replaced atari800/src/ wholesale with upstream 5.2.0 (32cc3b2), that protection was lost.

POKEY masks register addresses with 0x0F in mono and 0x1F in stereo (pokey.c:188). Bounty Bob writes to mirrored POKEY addresses (e.g. $E810 on 5200, $D210 on 800) expecting them to alias to $E800/$D200 on a single-POKEY system. With stereo enabled, bit 4 is decoded as chip-select and those writes vanish into a nonexistent second chip — game locks in the title screen. This matches exactly the AtariAge forum description you linked.

Fixes (commits 01d80d3, d66c6a2)

1. 5200 always mono — real Atari 5200 has only one POKEY, stereo is non-authentic for any 5200 cart
2. .car BBSB_40 always mono — peek the 4-byte big-endian type field in the 16-byte CART header before Sound_Initialise()
3. New atari800_pokey_stereo core option (default: disabled) — stock 800/XL/XE has a single POKEY, so default mono is most authentic and most compatible (BBSB, Road Race, etc.). Users who want to emulate the Atari Stereo Sound Mod toggle it on and restart. Matches standalone atari800 behavior where stereo is a -stereo/-nostereo cmdline / config / UI choice.

Verified

• Bounty Bob Strikes Back! (USA).a52 (CRC 0x7873c6dd) — attract-mode demo runs after the fix; before, the title screen was static even after 2000 frames.
• Synthesized BBSB .car triggers the header-peek path (BBSB_40 .car detected, forcing mono POKEY in libretro log).

Not addressed: .cas

I tested both .cas dumps you mentioned ((1985)(U.S. Gold)(PAL)(GB)[64K] CRC 30a923e4, (198x)(Aackosoft)(PAL)(NL)(en)[64K] CRC 874ab34c) headless and via Xvfb+xdotool: both crash with CIM $52 @ $6764 immediately on cold-boot, well before tape loading can begin. The crash reproduces with cassboot enabled/disabled, internalbasic enabled/disabled, on 800XL and 400/800 OS B, and is not affected by the POKEY stereo state. It's a separate regression in the cassette/SIO path of the 5.2.0 import, unrelated to POKEY mirroring — out of scope for this PR. Worth a separate issue if you can reproduce on Windows with the mono default.

[KamiLordus]

First of all, thanks a lot for your work on this.
Regarding tapes, you need to set them in the Atari800 core:
Video > Video Standard > PAL (if you leave NTSC - it will crash)
SIO Acceleration > OFF
Boot from casette > ON

With these settings, the tapes work (no crashes) in version 3a6b4c5 – except there's no sound. Although I don't rule out that the Linux version might be different.

For testing purposes, I'll include my versions of the tapes. If you can, please test them and then remove them. If that doesn't help, let's leave it as is.

[WizzardSK]

@KamiLordus Did some more digging on the .cas audio question with both your tapes (PAL versions you uploaded). Tested on Linux with PAL + cassboot=ON, both with SIO acceleration ON (instant load) and OFF (real ~14 min cassette timing). Here's what I found:

Both .cas files load and run gameplay correctly, with full POKEY audio in-game (verified by recording the libretro audio output to a .wav and measuring RMS — clearly active output during gameplay, not silence).

Title screen comparison:

Variant	Title-screen audio	In-game audio
Bounty Bob Strikes Back! (USA).a52 (cart)	full title music	✅
Bounty Bob Strikes Back! (1985)(U.S. Gold)(PAL)(GB).cas	typewriter SFX as letters scroll in, no melody	✅
Bounty Bob Strikes Back! (198x)(Aackosoft)(PAL)(NL).cas	typewriter SFX only	✅

So the cart has proper title music and the European budget tape conversions don't — they reuse the simple typewriter "tick" SFX during the intro instead of a melodic theme. That's a property of those specific tape dumps (the conversions stripped the music to fit cassette memory layout), not a regression in this PR. I also cross-checked one of the tapes against standalone atari800 5.0.0 with -pal -boottape — same behavior, no title music there either.

If you're getting genuine silence in-game on Win11 (not just the missing title music), that's almost certainly a RetroArch audio driver / buffer issue rather than a core problem — the POKEY mono fix in d66c6a2 already addresses the underlying mirror-write issue that was breaking BBSB gameplay sound. Worth trying a different audio driver in RetroArch settings (WASAPI ↔ XAudio ↔ DirectSound) once a Win build with this PR's changes is available.

[KamiLordus]

@WizzardSK Very interesting. I usually use sio acceleration off, which often causes crashes for various games on tapes, but I always use the RA fast forward option. I wonder if that's the problem with the lack of sound.
I asked @warmenhoven to merge your latest commits into the WizzardSK repository.

Thanks a lot for checking the tapes and providing the information.

[KamiLordus]

Tested and hurray, all versions work without any problem. Great work WizzardSK!!!