mirror of
https://github.com/RetroPortingToolKit/psxrecomp
synced 2026-10-26 10:19:47 -04:00
Alex Vanderveen
39bcdf0d56
Slicing the GPU linked-list DMA across guest cycles (8af48ae9, PR #277) also
changed what a walk costs: 8 clocks for every node header plus 5 setup clocks
per non-empty node, where the pre-slicing path charged one clock per word via
schedule_delayed_complete(2, actual_words, DMA_GPU_CYCLES_PER_WORD).
An ordering table is mostly EMPTY nodes -- Final Fantasy VII's runs to
ot_max 4094 -- and each empty slot still paid the full 8-cycle header for a
packet that does not exist. Measured on FF7 (SCUS-94163):
4,807 nodes / 12,757 words 52,384 cycles vs 17,564 (3.0x)
44,010 nodes / 117,507 words 499,187 cycles vs 161,517 (3.1x)
An NTSC frame is 564,480 cycles, so a single walk reached 88% of a frame.
FF7's libgpu polls CHCR bit 24, finds the channel still busy after 790-890
reads, and force-clears the start bit at 0x800484F0 -- the documented
stuck-channel path. cancel_async_transfer() then abandons the walk partway
through, and the rest of the frame's geometry is never sent to GP0. In a
battle 12.6% of ordering tables were aborted this way; models rendered with
missing or shattered limbs.
The guest is behaving correctly: it inspected the busy bit before clearing it.
The slicing is also correct and worth keeping -- a guest really can change a
packet after starting DMA and before DMA reaches it, and only a sliced walker
models that. Only the cost was wrong.
Header is one clock, setup is free, so a walk costs exactly (nodes + words).
When setup is free the payload boundary is the same service as the header, so
the header phase falls through instead of returning; cycles_to_event() never
reports zero, and returning would spend an unaccounted scheduler cycle per
non-empty node. The packet is still read at its own boundary, so #277's
mid-walk mutation behaviour is unchanged and its test still passes.
(nodes + words) is not a new assertion -- it is the cost this runtime shipped
and was validated against. A non-zero per-node overhead may well be real on
hardware, but it needs a measurement behind it, not a substituted constant.
Observability, because none of the above was visible:
dma_state gains a gpu_ot block -- starts, completes, cancels, starts_dropped,
nodes/words/cycles of the last walk and the max, plus guest reads of CHCR(2)
and a ring of the last 8 aborts carrying the guest PC, CHCR value, how far
the walk got, and how many times the guest polled before giving up. The
silent early return in start_async_gpu_linked_list() is now counted as
starts_dropped rather than losing an ordering table with nothing said.
PSX_GPU_LL_SYNC=1 drains a walk at start. Diagnostic A/B lever only: it
completes immediately as well, so it is not a faithful restoration of the
pre-slicing timing and must not be used as a fix.
The timing test now asserts the cost model rather than describing it: a table
of empty nodes must cost exactly one cycle per node. That is the check that
would have caught this.
Verified on FF7 SCUS-94163 from a save state: cycles_max 58,933 -> 16,972,
a 144-node/490-word walk costs exactly 490 cycles, and battle geometry renders
correctly with the walk still sliced.
Not fixed here: the same commit also broke gpu_frame_dump's func attribution.
GP0 writes now issue from dma_advance() on the cycle scheduler, so every
primitive is stamped with whatever the scheduler interrupted -- all 1,803
primitives in a captured battle frame attributed to BIOS 0x1FC02B50 instead of
the issuing game function. That disables gpu_frame_diff's "a function stopped
drawing" verdict and gpu_frame_layers entirely. Stamping the DMA initiator was
tried and does not work: g_debug_current_func_addr reads 0 at kick time. The
ra column still carries real game addresses as a fallback. Left as a separate
issue rather than shipped as a fix that does not fix it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01L9v7e6AVZe3eRP8VXey9Pz