Improve SmackDown 2 runtime stability, audio, and Vulkan performance - #16
Closed
shaneomac1337 wants to merge 7 commits into
Closed
Improve SmackDown 2 runtime stability, audio, and Vulkan performance#16shaneomac1337 wants to merge 7 commits into
shaneomac1337 wants to merge 7 commits into
Conversation
- psx_cyc.h / psx_cycles.c: _BitScanForward/_BitScanReverse shims for __builtin_ctz/__builtin_clz under _MSC_VER - full_function_emitter.cpp / main_psx.cpp: emit a portable .CRT$XCU init-pointer instead of GCC-only __attribute__((constructor)) in generated dispatch/game code - main.cpp: declare cross-language globals extern "C" so MSVC links them against their C definitions (no-op for GCC/Clang) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Two deterministic boot hangs (WWF SmackDown! 2, SLUS-01234): 1. overlay_loader_dispatch()/is_candidate()/call_native() ran during early BIOS kernel init (0xBFC0DB10 memcard polling loop) before overlay_loader_init() built the range-index structures, walking uninitialized lists until a PC=0 trap (froze at cycle 208195 every boot). Guard all three entry points on s_active. 2. Code at 0x800D1F28 (boot-text populated by bulk host transfer that bypassed the RAM write-marking hooks) was neither compiled, dirty, nor overlay -> unknown-dispatch fail at frame 1340. Add a fallback: an address above overlay_region_floor whose word decodes as valid MIPS is marked executable and handed to the dirty-RAM interpreter. Validated 5/5 reproducible boots, 2000+ frame sustained runs. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- Replace nearest-sample voice pitch resampling with the PS1 hardware 4-tap Gaussian (No$ table, new spu_gauss.h; formula (g[0xFF-i]*s[-3] + g[0x1FF-i]*s[-2] + g[0x100+i]*s[-1] + g[i]*s[0])>>15 with i=(phase>>4)&0xFF). Nearest-sample put a flat aliasing shelf at ~-31 dB across 8-22 kHz (measured); Gaussian drops it 18 dB and restores natural spectral decay. Adds prev[3] cross-block history. - Beetle-model reverb engine on the SPU mix. - Gate the float-shadow substitution off while any voice has a reverb send (the shadow models only the dry mix). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The PS1 GPU does not render a polygon or line when the distance between any two vertices exceeds 1023 horizontally or 511 vertically (No$ docs; Beetle/DuckStation enforce it per rendered triangle). Games rely on the cull during zoom transitions and close camera cuts: their 11-bit-wrapped vertices span the whole coordinate space and, without the reject, get rasterized as giant flat-color triangles covering the frame (SmackDown 2 rendered its menu star wipe and in-match crowd cuts as a full black cover in both the software and GL backends). Applied per-triangle in all polygon execs (quads reject each half independently; textured variants still latch the texpage word first, matching hardware), plus lines and both polyline continuations. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Fixes found chasing block corruption on AMD RDNA3 (RX 7900 XT) that the NVIDIA driver this backend was developed on tolerates: - flush_cpu_upload: keep every packed rect's VkBufferImageCopy bufferOffset 4-byte aligned (an odd w*h rect misaligned every following rect's R16 offset to 2 mod 4) - begin_geo_pass + wide passes: color-aspect self-barrier mirroring the existing stencil self-barrier (img_to emits no barrier when the layout is unchanged, and the render pass declares no EXTERNAL dependency, so back-to-back passes had no memory dependency on the color attachment) - submit_present: wait the acquire semaphore at TRANSFER (all swapchain writes here are transfer ops); waiting at COLOR_ATTACHMENT_OUTPUT left them unordered against presentation-engine release Still NOT fixed on AMD: black block confetti on BIOS/FMV screens. VRAM-state dumps prove the hr image content is bit-perfect at present time, so remaining corruption is in the present chain. Env-gated diagnostics (tagged [DEBUG-vk01]) included for the ongoing hunt: PSX_VK_VERIFY / PSX_VK_DUMP / PSX_VK_ONEPASS / PSX_VK_EAGER_UPLOAD / PSX_VK_NO_MERGE. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This was referenced Jul 14, 2026
Owner
|
Hello!. I have pulled the MSVC generated-initializer fix into #1 and the overlay pre-initialization guardinto #23 The broader executable discovery, GPU/SPU changes, Vulkan work, and shaders are not yet pulled in. I need to give those more of my time before I can merge them out. But thank you for the work so far! Eager to find time to dig into the rest |
Owner
|
Tested the extracted Vulkan changes from this PR in Tomba 2 on an RTX 3080 Ti. It booted through SCEA and the Whoopee Camp FMV into gameplay, but it does not feel merge-ready yet:
Keeping the Vulkan work parked for further isolation. |
This was referenced Jul 15, 2026
mstan
added a commit
that referenced
this pull request
Jul 15, 2026
Port the reusable Gaussian interpolation slice from PR #16 while keeping its reverb engine on a separate branch. Co-authored-by: Martin Penkava <mpenkava1337@gmail.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Owner
|
Hello! Wanted to follow up on this draft. The SPU reverb implementation closely seems to model Beetle/Medafen's, which are GPLv2. Want to make sure before I include anything that I understand attribute. Similar question for the CRT shader. Suggest it's inspired by CRT-Royale. I'm not sure what that is, and would like to understand its licensing as well before taking it into consideration. Take care! |
Owner
|
Following up. The Vulkan work was ultimately a net positive. Very performant, but still suffers from some visual fidelity issues. In Tomba 1, a lot of sprites had some garble. In Tomba 2, the HUD item selector was showing hidden element that's normally toggle in. There was also some items poking through terrain. I'll go ahead and accept the work, but will not be making Vulkan an exposed option yet until these types of bugs are ironed out. With regards to the remaining unmerged items; license concerns remain outstanding. Given that this PR is somewhat stale and the artifacts are independent items anyhow, I'll go ahead and close this draft PR. Please submit any follow ups as individual items, please! 
Thank you for the contributions! |
fabioap-cpu
pushed a commit
to fabioap-cpu/Tekken1Recomp
that referenced
this pull request
Jul 30, 2026
Adapt only the fail-closed overlay lifecycle checks from mstan/psxrecomp#16 commit e6809ccf7d778a4c2f32d9e27c0ec31a44cbd2ba. Add a focused structural regression test and record the excluded dirty-interpreter and game/runtime work. Co-authored-by: Martin Penkava <mpenkava1337@gmail.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Owner
|
A follow up on this PR. PR #104 introduced a clean room psx-spx/nocash register instead. One thing that was not addressed in that was the hardware's 22.05->44.1 kHz reverb reconstruction filter. Given Beetle's GPL licensing, I wasn't able to borrow it from there and would need to derive a clean room implementation. The functionality is isolated and a future PR that modifies it is always welcome; assuming it follows proper licensing. |
ClaudioVarandas
pushed a commit
to ClaudioVarandas/psxrecomp
that referenced
this pull request
Aug 18, 2026
…inational Closes the four gaps named in issue mstan#103 and lands the separable, verified parts of PR mstan#102. See docs/internal/SPU_FIDELITY_103.md for the per-piece disposition and the oracle-verification queue. SPU DSP (issue mstan#103), clean-room from hardware documentation: - Reverb: the documented 22050 Hz engine — vLIN/vRIN input, same-side and crossed different-side IIR reflections, 4-tap comb early echo, both all-pass filters, vLOUT/vROUT output, and the work-area address advance. Gated on SPUCNT bit 7 for WRITES only; reads, IRQ checks, address advance and output always run, as on hardware. Addressing is confined to [mBASE, 0x80000). Per-voice sends from EON, CD send from SPUCNT bits 0+2. - Noise: 16-bit LFSR with the documented parity feedback, clocked at the SPUCNT bits 8-13 rate; NON voices substitute it for their ADPCM sample. - Volume sweeps: all 24 voice L/R plus main L/R. Bit 15 selects sweep vs direct; sweep is a live envelope on the same rate machinery as ADSR, and a guest read of a sweeping register returns the live level. - Capture buffers: CD L/R and voice 1/3 output written to SPU RAM, so a game that parks the IRQ address there still gets its interrupt. - SPU IRQ (I_STAT bit 9): checked at every SPU RAM access class — FIFO, DMA both directions, ADPCM block fetch, capture writes, reverb work-area accesses — gated on SPUCNT bit 6, mirrored in SPUSTAT bit 6, acked by clearing SPUCNT bit 6. The CD-only fast path in spu_render was removed rather than extended: the SPU now has per-frame work that must run in every path, and an FMV (XA audio with zero active voices) is exactly the CD-with-reverb case. This is a clean-room implementation from the psx-spx/nocash register map and documented algorithm. Beetle's source was deliberately not consulted: it is GPL-2.0-or-later and this project is PolyForm Noncommercial. PR mstan#16 was parked for exactly that reason and PR mstan#13's reverb was separately wrong (it gated on bit 15 of dAPF1, an address offset, instead of SPUCNT bit 7). The one place this knowingly deviates from hardware is the 22.05 -> 44.1 kHz reconstruction filter, which the documentation does not specify; it is isolated in a single function and flagged for oracle comparison. CAUSE.IP2 (from PR mstan#102): IP2 is combinational on R3000A — it mirrors the interrupt line and falls the instant the guest acks I_STAT or masks the source. It was being OR'd in at delivery and never cleared, in TWO places (the compiled path and psx_interpreter.c, the latter missed by PR mstan#102), leaving a phantom pending interrupt that can spin a kernel dispatcher looping on CAUSE.IP & SR.IM. psx_irq_refresh_cause_ip2() is now the sole writer of bit 10, refreshed at raise, I_STAT ack, I_MASK write, HLE context restore and power-on. Verified against the Beetle oracle's irq.cpp/cpu.cpp rather than asserted, and covered by test_cause_ip2_combinational. Mid-dispatch audio pump (from PR mstan#102), with a fix: routed through a gate that mirrors sdl_audio_update's mute/sink decision. Pumping unconditionally from the VBlank edge, as submitted, would have pushed real audio during every turbo-load hard mute and defeated the freeze-in-place mute model. SPU RAM DMA readback (from PR mstan#102): DMA4 in the SPU->RAM direction wrote literal zeros; SPU RAM is readable memory and titles carry state through it across an Exec boundary. Also fixes an unrelated build race found on the way: tools/embed_spirv.py wrote its intermediate SPIR-V to <source>.spv, so the two runtime targets that embed the same shaders raced on one path and one deleted the file the other was reading. Co-Authored-By: Alexandros Mandravillis <Alexbeav@live.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What changed
Why
SmackDown 2 previously encountered boot freezes, missing reverb, and Vulkan frame-time spikes caused by repeated command submissions, staging allocations, and competing presentation pacing.
Impact
The game boots reliably, retains higher-quality PSX audio, and renders through Vulkan at a much steadier cadence while preserving configurable screen simulation.
Validation
SmackDown2Recomp.exesuccessfully with Vulkan enabledThe generic BIOS-only framework link still reports the existing unresolved
psx_game_address_in_textsymbol; the game target itself builds successfully.