EA SPORTS FC 27 (PPSA34015)

[klosixgaming-arch]

Existing thread

• I searched this category for the title ID and no thread exists for this title on any platform.

Title

EA SPORTS FC 27 (PPSA34015)

Content version tested

01.000.004

AnyPS5 commit

main beb723a with #5352 #6034 #6049 #6056 #6058 #6062 #6072 #6074 #6079 #6084 #6085 #6086 #6087 #6088 #6089 #6090 #6092 #6343 merged on top

System tested

Windows 11, WinLibs GCC 15.2.0 (x86_64-posix-seh, ucrt, r7), RTX 5090 (NVIDIA 610.47), AMD Ryzen 9 9950X3D

Furthest point

Logos or loading screen

What is shown

The EA SPORTS FC 27 splash screen, drawn correctly. The frame rate starts at 0.3 fps and rises to about 1.5 fps while shaders are recompiled on the first run (about 50 to 100 shaders a second). Audio and input were not checked yet.

Current blocker

The frame rate during first-run shader compilation. The title logs this once per frame, and its breadcrumb report says all markers completed:

Failure in SubmissionManagerPs5::waitForFrame with error code 0x8002003C.  See the log for additional details (e.g. potential GPU breadcrumbs).

Open pull requests this run needs:

• notaxeltv's PPSA34015 PRs, applied together: feat(libSceGameLiveStreaming): accept GetProgramInfo with no live stream #5352 feat(libSceLoginService): load Initialize/Terminate so PPSA34015 can start #6034 feat(libSceNpPartner001): answer sceNpHasEAAccessSubscription without an EA Access server #6049 feat(libSceNpManager): export two unknown NIDs so PPSA34015 can load #6056 feat(libScePlayerSelectionDialog): run common-dialog state machine without UI #6058 feat(libSceVdecsw): block on SyncDecodeInput/Output like TrySync #6062 feat(libc): raise logic_error::_Doraise and _Xruntime_error #6072 feat(libc): allocate through operator new with align_val_t #6074 feat(libc): throw bad_optional_access like bad_weak_ptr #6079 feat(libc): implement C11 timespec_get (PPSA34015) #6084 feat(libc): export _FDsign/_Dsign like MSVC CRT (PPSA34015) #6085 feat(libc): export std::thread::hardware_concurrency (PPSA34015) #6086 feat(libc): export locale::classic (PPSA34015) #6087 feat(libc): _Cnd_init_with_default_name_override (PPSA34015) #6088 feat(libSceSystemService): accept OpenTournamentOccurrence without system UI #6089 feat(libSceVoice): accept WriteToIPort on input ports (PPSA34015) #6090 feat(libSceWebBrowserDialog): finish PredeterminedContent without UI (PPSA34015) #6092. Without them the title does not load on main: libSceLoginService.prx is missing and libraries such as libScePlayerSelectionDialog and libSceVdecsw lack imports it needs.
• feat(libkernel): accept a kevent read event on descriptor 0 #6343 kevent on descriptor 0. Without it, the title stops about 40 seconds after start with kevent: EVFILT_READ on a non-socket descriptor not implemented, because it registers its standard input in a kqueue.

Also logged, without stopping the title: sceSysmoduleIsLoaded unknown id: 306.

How it was run

The title's eboot.bin and its bundled modules (sce_module/libc.prx, sce_module/libSceNpCppWebApi.prx, Engine.MovieVLC.LibVLC.prx and Engine.Render.Core2.PlatformPs5.prx) were unwrapped to ELF and relinked with relinker --windows --sce-module-path <elf dir> --module-dir . eboot.elf app.exe. app0 holds the dump's Data, Patch, Scripts, sce_sys and library as junctions, plus the original eboot.bin, package.mft and Runtime_Ps5_retail.BuildSettings. Run from a console with ANYPS5_GPU=5090.

[pushkarreddyy]

Claiming the start-up sysmodule notice sceSysmoduleIsLoaded unknown id: 306 (�x00000132). Adding the �x00000132 mapping to libSceProprietaryVoiceChatHelperPrx in core/libs/prx/libSceSysmodule/ModuleTable.hpp (matching the ID verified in #5017) with unit test coverage in guest_sysmodule_tests; branch and PR in progress.

[klosixgaming-arch]

Second run with a warm shader cache, same build (local/fc27-stack: main beb723a18a + the 17 PPSA34015 PRs + #6343):

• The [shader-disk-cache] lines now report hits instead of misses: 26,341 hits against 350 misses (the 350 were written to the cache during this run), and the Vulkan pipeline cache loaded from disk.
• Visually the title is still on the splash screen after 30+ minutes, but it has progressed further in the background than before: it opens /download0/EA SPORTS FC 27/mainmenuconfig.json, creates menu textures, and opens its EA DP SDK persistence files (environment.dat gets a device UUID written; EadpErrorsData.log.st.dat stays empty).
• The per-frame Failure in SubmissionManagerPs5::waitForFrame with error code 0x8002003C is gone (zero occurrences): with the cache warm, the GPU no longer misses its frame deadline while compiling pipelines, so that error was a side effect of first-run shader compilation.
• Still logged without stopping: sceSysmoduleIsLoaded unknown id: 306. That module id (0x132) is covered by notaxeltv's open PR feat(libSceSysmodule): register libSceProprietaryVoiceChatHelperPrx module id 0x132 #6370.

Currently investigating what holds it on the splash screen (thread sampling); I will report back here with the next stop and any PRs it needs.

[klosixgaming-arch]

Update, 15 Oct (same PC: Windows 11, RTX 5090, NVIDIA 610.47). This run applied main beb723a with notaxeltv's PRs, #6343, #6370, #6342 and #6487.

Furthest point: past the splash screen, into main-menu loading. The title opens /download0/EA SPORTS FC 27/mainmenuconfig.json, filesystemcache/ and /app0/Scripts/ini/osdkdebugmanager.ini, about 55 s after start with a warm shader cache.

• feat(libSceSysmodule): register libSceProprietaryVoiceChatHelperPrx module id 0x132 #6370 removes sceSysmoduleIsLoaded unknown id: 306.
• With a warm cache the waitForFrame 0x8002003C lines are gone; they came from first-run shader compilation.

Current blocker (every run on the 5090):

what():  shader 0x11a0709c00 stage 5: ShaderRecompiler::Recompile: SPIR-V emission failed: hash=0x75706178560 stage=2 opcode=ReadLane reason=v_readlane_b32 of lane 63 is outside the 32-lane host subgroup that runs this wave64 program at one lane per invocation

It is a wave64 pixel shader, so #5405 (compute only) does not cover it. The block is a UMin half-wave reduction:

s_orn2_saveexec_b64 vcc, exec
v_cndmask_b32_e32 v2, -1, v0, vcc_lo
v_min_u32_dpp v2, v2, v2 row_shr:1 ... row_shr:8
v_permlanex16_b32 v1, v2, -1, -1 op_sel:[0,1]
v_min_u32_e32 v2, v2, v1
s_mov_b64 exec, vcc
v_readlane_b32 s4, v2, 31
v_readlane_b32 s5, v2, 63
s_min_u32 s28, s4, s5

The failure reproduces offline with agc_shader_replay on the logged RecompileRequest. MatchHalfWaveScan recognizes the scan, the row steps and the permlane. Only the key check fails: EntryLaneWalk::ZeroBit on the v_cndmask condition. That condition is the loop's exec (a Phi of the helper test and exec & ~ballot(v_cmp)) ANDed with Ballot(true), and NotHelper does not accept Ballot(true) in a pixel shader. I can attach the request blob if someone working on the half-wave reduction wants it.

Tried for comparison:

• With feat(recompiler): run single-wave wave64 compute programs at one lane per invocation on 32-lane hosts #5405 applied, the title recompiles every shader and then shows only a black screen at 0 fps, with waitForFrame 0x8002003C every frame.
• On the Ryzen iGPU (AMD Radeon(TM) Graphics, device 0x13c0, subgroup 64) it gets past compilation but never finishes a frame (waitForFrame timeouts); the iGPU is probably just too slow.

Also seen once: Joining guest thread: Access is denied (abort) from the title's thread wrapper. #6487 follows FreeBSD's joiner rules for it.

[klosixgaming-arch]

Same stop with NVIDIA driver 617.x (updated from 610.47): the v_readlane_b32 of lane 63 failure above, about 55 s in, during main-menu loading. The recompiler fails before the driver is involved, so this was expected.

Question for whoever owns the half-wave reduction matcher (SpirvAluEmitterMath.cpp): how should this pixel-shader case be handled?

In this shader the v_cndmask key condition is the loop exec (a Phi of the helper test and exec & ~ballot(v_cmp)) ANDed with Ballot(true). EntryLaneWalk::NotHelper rejects Ballot(true) in a pixel shader, because helper lanes are active there, so ZeroBit fails.

The options I see:

1. Prove the keys still exclude helpers through the exec Phi: its entry value is the helper test, and its back edge only clears bits. That means teaching Word that x & ballot(true) is no wider than x.
2. Something else you prefer.

I can turn either into a PR with a ShaderMemory test built from this shader's block. I can attach the RecompileRequest blob, which agc_shader_replay reproduces offline. I'd rather not guess at these semantics without a maintainer's view.