Repository navigation
NHL 26 (PPSA26785) #7304
Replies: 6 comments
|
Logs for the NHL 26 (PPSA26785 01.060.000) runs on stderr of run 2 (warm), trimmedstdout of run 1: the title's own breadcrumb dump after the abortAI-assisted: yes (Claude Code). |
|
Update from the RTX 4090 / Windows 11 box: NHL 26 (PPSA26785, 01.060.000) on
Open PRs this title still needs on this box: #7346, then #7048. Earlier: #6180 and #5405 are in AI-assisted: yes (Claude Code). |
|
Following up on the report above regarding I have implemented a fix on branch Opening a PR shortly. |
|
Update from the RTX 4090 / Windows 11 box,
Open pull requests this title still needs on this box: #7048 (#7346 is in AI-assisted: yes (Claude Code). |
|
NHL 26 (PPSA26785, 01.060.000) on Input came from the project's controller command file (
What the logs say about the slow phase after the press (run 2):
Timeline of the warm runs for anyone scripting this: first frames 4:45 to 5:13, the EA logo screen after that (6:01 in run 1), the title screen at 6:23 to 6:50 (frame ~700); the Cross press has to wait for the title screen, a press on the black window is lost (run 1). Next: ACCEPT on the legal agreements page, in the note that follows. Open pull requests this title still needs on this box: #7048. AI-assisted: yes (Claude Code). |
|
NHL 26 (PPSA26785, 01.060.000) on
The a1 crash, from the dump: the Heartbeat thread runs a loop at Host memory, now sampled from launch: Timeline of a warm run for scripting: first frames about 5:00, the EA logo screen right after, the title screen at 6:20 to 6:50 (frame about 700), the legal agreements page about 25 s after the Cross, the quick settings page about 15 s after ACCEPT. A press on the black window is lost; gate on the frame counter. Next: ADVANCE on the quick settings page (Cross) and whatever follows; the libSceJson2 AI-assisted: yes (Claude Code). |
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Existing thread
Title
NHL 26 (PPSA26785)
Content version tested
01.060.000
AnyPS5 commit
main376c040 (2026-10-18) with PR #7048 on top; the first report below wasmain922dc89 (2026-10-17), and the 2026-10-18 runs with #7346 (since merged) are in the comments.System tested
Windows 11 (26200), GeForce RTX 4090 (NVIDIA 616.92, Vulkan 1.4.341, subgroup size 32), AMD Ryzen 7 5800X, 32 GB RAM, MinGW-w64 GCC 15.2.0 r7 (UCRT), Release
libs.Furthest point
Menu
What is shown
With
main376c040 + #7048 and scripted controller input (APS5_PAD_INPUT_FILE): the EA logo screen, the title screen with its "START" prompt at 11 to 12 FPS, a Cross takes it to the "NHL 26 LEGAL AGREEMENTS" page (0.2 FPS for 3 min on its first visit while 167 shaders compile, then 3 to 7 FPS), ACCEPT takes it to the "QUICK SETTINGS / WELCOME TO NHL 26" page (experience, controls, difficulty), which runs at about 4 FPS until the cap. Picture only; audio not checked. On stockmain922dc89 (first report): a black window, then the abort below.Current blocker
Nothing stops it on the quick settings page yet; ADVANCE there is the next step. Stops seen on the way, all intermittent: #7700 (
FATAL: unhandled exception 0xc0000005 on thread 'Heartbeat', the title walking a freed thread stack, once in three visits to the legal agreements page) and the storage-image write-back refusal that #7048 is for (storing storage image results pending over memory outside the write watch without the guest bytes saved before them is not implemented), which still fires on about 1 run in 7 with #7048 applied (reported on the PR). Thesce::Json::Parser::parse: null textabort from libSceJson2 was seen once with #7346 and not since (no issue). Stockmain922dc89 stopped oncompute shader 0x1150101000: AGC graphics: 32-bit integer read of the depth-layout texture 0x11c10f0000, which is no depth surface drawn with, is not implemented(#5398, closed, fixed by #7346, merged). Open pull requests the title still needs: #7048.How it was run
eboot.binand the four modules unwrapped withtools/unwrap_self.py, relinked withrelinker --windows --registry --module-dir .;app.exefrom the desktop session withANYPS5_REQUIRE_VULKAN=1, no controller; button presses throughAPS5_PAD_INPUT_FILEwithtools/pad_input.py(Cross on the title screen, Cross on ACCEPT), 600 to 1200 s cap.(Original report of 2026-10-17 follows unchanged.)
NHL 26 (PPSA26785, 01.060.000, retail dump, owned) on Windows / NVIDIA, stock
mainof today. Short version: the wave64 pixel stop from #5398 is gone on currentmainhere too, the title compiles ~27,000 shaders, presents a black window for a few hundred frames and then aborts on the same depth-layout integer read that @milchgeldkassierer hit second on Linux, with the same shader hash and the same texture address.Setup. Windows 11 (26200), GeForce RTX 4090 (NVIDIA 616.92, Vulkan 1.4.341, subgroup size 32), AMD Ryzen 7 5800X, 32 GB RAM, MinGW-w64 GCC 15.2.0 r7 (UCRT), Release
libs. Tree:main922dc89 (2026-10-17, includes #6180 and #5405), no local changes.eboot.binand the four modules (sce_module/libc.prx,sce_module/libSceNpCppWebApi.prx, and the two at the dump root,Engine.MovieVLC.LibVLC.prxandEngine.Render.Core2.PlatformPs5.prx) unwrapped withtools/unwrap_self.py, relinked withrelinker --windows --registry --module-dir .(no--to-intel, the host is AMD). With--module-dir .the relinker writesapp0/Engine.Render.Core2.PlatformPs5.prx.guest.prxbeside the original, so thesceKernelLoadStartModule("/app0/Engine.Render.Core2.PlatformPs5.prx")load needs no hand-made link. Import audit (tools/import_audit.pyover the five registries): 1,332 unique imports, 0 absent system imports (the 246 it lists as absent are exports of the title's own modules: 162libSceNpCppWebApi, 78Engine.MovieVLC.LibVLC, 6libc), 27 stubs. Launch:app.exefrom the desktop session withANYPS5_REQUIRE_VULKAN=1only, no language variable, no controller, nobody at the keyboard, 900 s cap, three runs (one cold shader cache, two warm).Result. 3/3 identical. No
SPIR-V emission failed/v_readlane_b32 of lane 63anywhere (the #5398 stop is gone on a plainmainbuild, as the #6180 replay predicted). The title creates the device, registers its shaders (cold run: 27,297 compiled, 0 cache hits, ~11 min; warm runs: 26,758 / 27,232 hits, ~4 min), then opens its 1920x1080 window and flips black frames with the emulator's FPS counter running (run 1 reached frame 504, run 3 frame 272), and aborts:Every capture is black (the title never gets a frame with content out before the abort). Device-local memory at the stop: 242 to 375 MiB of the 23,370 MiB budget, so none of the 8 GB-card problems from #5398 apply here, and none of them is reached anyway.
Where it is, for @MarcZX100's question on #5398. The failing dispatch itself is not in the kept part of the ring dump (the last 16 graphics submissions are cut at 512 dwords each; the two compute-queue submissions it keeps dispatch neighbouring programs such as 0x1150100100, not 0x1150101000). What the log does give is the title's own breadcrumb dump, printed when its
SubmissionManagerPs5::waitForFramefails (0x8002003C) after the abort, for the graphics queue of that frame:preIntegratedFG,planarShadowDL,planarShadowsBlur(x2) andmainForegroundGBuffer(x2) complete, first incompleteVelocityVectorFromDepth(last begun render passVelocityVectorFromDepth, last bound pipelineShaderProgram_GTAODenoisePass, last opcodeDispatch 60 x 45 x 1), thenGpuProbes_ExtractProbeBit,RenderDirectionalShadow,drawShadowBufferSlicenot started. The log does not say what 0x11c10f0000 was drawn as before: the only 1920x1080 registrations it prints are the storage image at 0x11c0000000 (format 36) and the texture at 0x11c0870000 (format 56). This failure does not dump aRecompileRequest; if one of shader 0x1150101000 or the descriptor words would help, tell me what to print and I will run it again.Other lines seen before the stop, none of them a stop: 104 shaders deferred at registration (
s_swappc_b64calls a computed target), one rect-list draw prepared at draw,layer/viewport index vertex exports are ignored (PA_CL_VS_OUT_CNTL=0x01240000),sceSysmoduleIsLoaded unknown id: 306twice per run. Nothing fromlibSceVdecsworsce::Jsonis hit (theSync*functions andsetNullAccessCallBackare onmainnow).I found no open PR for this read. The local
cachedTextureswitch that @milchgeldkassierer used to get past it is the next thing to try here; I can run a branch for it on this card.Trimmed stderr of a warm run and the title's breadcrumb dump: #7304 (comment)
AI-assisted: yes (Claude Code).
All reactions