[diegoparrilla]

On an STE or a Mega STE the sound now goes through the DMA sound chip: 8-bit samples at 12,517 Hz, with no interrupt per sample. A plain ST or Mega ST plays through the YM as before. Both run from the same firmware.

How it works

• The chip plays from ST RAM, from a 2 KB ring just below screen page A ($6F800). Every VBL, right after the wake, userfw:
  • reads where the chip plays;
  • reports it at $FB8Dxx;
  • copies the cart buffer (the ring's mirror) from where its last copy ended, up to 768 bytes (3 VBLs) ahead of the chip.
  • The copy costs about 0.35 ms at 8 MHz. Timer-B stays off on this path, which frees its 1.6 ms per VBL.
• At boot, userfw sets the Microwire as TOS does. It stops the chip on the way out to GEM and before a cold reset.
• Both sides pick the output by the same rule: DMA when the hello's machine family is the STE's and the RP has not asked for the YM (slot 6 of the shared variables, audio_prefer_ym()).
• The RP writes ahead of the chip. audio.c's writer keeps the mirror written 928 bytes ahead of where the chip plays now: the last report, plus the samples played since, since a missed VBL sends no report. It never writes where the ST copies.
• Sources: apps give YM pairs as before, or signed 8-bit PCM at any rate (audio_play_pcm_loop, audio_set_pcm_callback).
  • Either plays on both outputs, converted at runtime through the 64-entry table both ways, with linear resampling.
  • A YM source on the YM path goes through byte for byte as before.
• The audio buffer grows to 2 KB (the mirror), so APP_FREE moves to $FA4980 and has about 14.4 KB.
• Tools:
  • wav_to_ym4.py --mode pcm writes PCM headers.
  • Two debug commands: audio_out 0|1, which chooses the output from the ST's next boot, and tone HZ.
  • The audio output counter in swd.py counters.

Tests

• test_audio_dma.c runs the output against a simulated STE:
  • steady play;
  • main-loop stalls within and beyond the 80 ms tolerance;
  • a missed VBL;
  • an ST reset;
  • YM pairs on the DMA chip, PCM on the YM, and the output switch.
• Every sample played must be the source in order, the held last sample, or silence before the start.
• Breaking the writer on purpose (no byte swap, no time estimate, no margin, stale data on underrun, no inverse table) fails it every time.
• The YM path's 315 existing checks are unchanged.
• test_wav_to_ym4.py checks that audio.c's tables match the converter's.
• The cartridge code is 8,928 bytes.

Hardware

Test	Mega STE (TOS 2.06), DMA path	Mega STE, forced onto the YM	Mega ST (TOS 1.04), release build
DEMO.YMS, converted to PCM	clean, better than the YM	—	clean
440 Hz tone through the PCM path	clean	clean	—
Mouse moved flat out: frames per second	49.5–50.5	30–38	—
Late / underruns / resyncs	0 / 0 / 0	0 / 0 / 0	0 / 0 / 0
Exit to GEM	chip silent, key click plays	—	—

• Mega ST: the release build stays on the YM and never touches the DMA registers.
• Not tested here:
  • STE TOS 1.06 and 1.62;
  • a 30-minute soak.
• Not included:
  • PCM files streamed from SD (.YMS files play on both outputs);
  • a PCM version of the jingle, whose source recording is not in the repository.
• Latency on the DMA path is up to about 175 ms from the source to the speaker (120 ms on the YM), because of the 3-VBL lead. AUDIO_FIFO_SLICES lowers it.

Docs

CLAUDE.md (the audio pipeline for two outputs, the Atari RAM it uses, the layout), README.md (audio), the framebuffer-app skill, tools/dev/README.md and CHANGELOG.md.