[fma-rc]

What

1348 Ex Voto (PPSA24249) stops on current main 2559e96 when a compute shader calls through libSceAgc's FS table:

[videoout] presentation failed: compute shader 0x16800e3000: shader 0x16800e3000 stage 0: ShaderRecompiler::Recompile: computed/data-dependent s_swappc_b64 call target at program counter 0x00000130 is not statically resolvable

The shader builds the constant address 0xF_E0040000 with two s_mov_b32, loads slot 0 there as a V#, reads a call target at +0x20 and skips the s_swappc_b64 when it is zero. Issue #6117 reports the same shape in MXGP 26 and asks what a console keeps at that address.

On the PS5 12.00 modules, libSceAgc's init gets a block from libSceAgcDriver, aborts unless the table lands at 0xF_E0040000, zeroes six 16-byte slots, and makes slot 0 a V# (stride 16, eight records) over a buffer of eight {routine address, 0} records. This PR:

• maps 2 MiB at 0xF_E0000000 once per process when the driver is created (MAP_FIXED_NOREPLACE on Linux, a commit inside the guest arena on Windows), fills the table that way and registers the block with GuestAllocations;
• points every record at one routine that only returns (s_setpc_b64 s[14:15]); Sony's routines are not modeled (TechnicalDebt);
• extends the table-callee path from feat(shader): link callees of calls whose target is loaded from a table #4706 to a V# pointer set by two s_mov_b32 of constants, and lets the writer search cross a conditional branch that jumps past the call (the null check).

Refs #6117 (same call; its unaligned atomic stays open).

Tested

Linux, current main 2559e96 plus this PR:

Run	main	this PR
Ex Voto, warm shader cache, 300 s limit	stops at 57.1 s (frame 30) with the line above	no stop, frame 566 at 300 s
Ex Voto, cold shader cache, 300 s limit	frame 32 at 300 s	no stop, frame 472 at 300 s

Rebased onto main 4524944 (2026-10-16): agc_driver_compute_fs_table_calls passes on RTX 5080 (NVIDIA 615.78.08), Radeon 610M (RADV) and lavapipe. Ex Voto, warm cache, 300 s: main stops at 53.1 s with the same s_swappc_b64 line; with this PR no stop, frame 851 at 300 s, on the title screen.

With the PR and a warm cache, pressing Cross at the title screen reaches the main menu (Play / Options / Credits) with no stop in 390 s.

MXGP 26 (PPSA30839), tested by milescoviello on Linux (RTX 4070 Ti, main d92a8da merged with d1a68c4, comment): main stops at 135 s with the issue's s_swappc_b64 line; with this PR 2 of 2 runs reach the 180 s limit without it.

Silent Hill: Townfall (PPSA33286), tested by tv7 on Windows (RTX 3070, main ff23657 merged with d1a68c4, comment): main stops on the s_swappc_b64 line in 3 of 3 runs after about 2.5 min; with this PR it appears in none of 5 runs and 4 of 5 reach the 900 s limit (the fifth hits a render-thread timeout tv7 has seen before without this PR).

Hot Wheels Infinite Rush (PPSA22408), tested by mrfelek on Linux (RTX 3080 Ti, main 5d9dcc4 merged with d1a68c4, comment): main aborts on the same s_swappc_b64 line at PC 0x134 after the boot screens; with this PR none of 3 runs shows it (one reaches the loading screen, one stops on a storage-image write-watch error also seen once without it, one runs out of device memory while storage images grow).

agc_driver_compute_fs_table_calls (the Ex Voto shape: the marker store runs only after the call returns) and scalar_call fail with GraphBuilder.cpp and Lifecycle.cpp from main (FS table at 0xfe0040000 is not mapped, scalar call regression; NVIDIA) and pass with the PR on NVIDIA RTX 5080 (615.78.08), RADV (Ryzen 7 7800X3D iGPU) and lavapipe. Windows (WinLibs GCC 15.2, VM): the full suite passes except guest_kernel_stack_protect and guest_kqueue, which fail the same way on main there.

Checklist

• Based on current main; no other open PR implements the same functions
• One topic per PR; follow-ups go in a new PR
• No comments in code except technical debt
• Unimplemented paths throw (NotImplemented_nid_no_patch); silent stubs are listed under Silent stubs in their library's technical debt file
• No notes, investigation .md files or images added to the repository (attach them to this PR)
• New third-party code is a submodule built from source
• Generic behaviour, not specific to one title
• Depends on: none
• Unblocks: PPSA24249 (1348 Ex Voto), PPSA30839 (MXGP 26, with fix(shader): act on the containing dword for unaligned 32-bit buffer atomics #6094 for its atomics), PPSA33286 (Silent Hill: Townfall), PPSA22408 (Hot Wheels Infinite Rush)
• GPUs tested: NVIDIA RTX 5080 (615.78.08, Arch Linux), AMD Ryzen 7 7800X3D iGPU (RADV, Arch Linux), lavapipe
• AI-assisted: yes

[milescoviello]

MXGP 26 (PPSA30839), the title in #6117, with this PR on Linux.

Setup: Ubuntu 24.04, GCC 13.3, Release, RTX 4070 Ti (NVIDIA 595.99), main d92a8da merged with this head d1a68c4ae (clean merge), relinked with --to-intel, 180 s runs with pad input (a button every 1.5 s from 8 s).

• main alone: stops at 135 s with computed/data-dependent s_swappc_b64 call target at program counter 0x00000134 is not statically resolvable, the AGC: MXGP 26 (PPSA30839) aborts on a compute shader that calls a function pointer read through a V# at the fixed address 0xF_E0040000 #6117 shader.
• With this PR: 2 of 2 runs reach the 180 s limit and that line does not appear. The unaligned buffer_atomic_or faults (BDA access failed: ... bytes=4 stage=4 reason=7) remain, 36 and 70 per run; fix(shader): act on the containing dword for unaligned 32-bit buffer atomics #6094 is for those.
• With this PR and fix(shader): act on the containing dword for unaligned 32-bit buffer atomics #6094 (6c188638a): 2 of 2 runs reach the limit with no BDA fault. A third run with screenshots goes through the publisher splash and the title screen to the "Welcome to MXGP26" account prompt; the FPS counter reads 0.2–0.5 there.

Not checked on Windows.

AI-assisted: yes

[ps54all]

Tested on God of War Ragnarök (PPSA08330, 06.006.000): the computed s_swappc_b64 stop is gone.

• Setup: Windows 11, RTX 3080 10 GB (NVIDIA 610.74), i9-10850K, WinLibs GCC 15.2, Release. Our test branch (main 82f0c9aec + perf(agc): draw finisher pipeline (APS5_DRAW_PIPELINE) #6946 + fix(agc): skip a periodic pipeline cache save while a pipeline creation holds the cache #6958 + fix(recompiler): give a dynamic-mip storage image a slot for every level #7238 + feat(shader): resolve nested image tables through runtime pointers #7223 + fix(recompiler): read lane spills back through relative-addressing window registers #6277 + fix(shader): plan image tables through invariant descriptor phis #6417 + fix(ampr): return EBUSY when a command does not fit and EPERM before a buffer is bound #7300 and a few open picks), with and without this PR; cold shader cache in both.
• Without: dies ~400 s, computed/data-dependent s_swappc_b64 call target at program counter 0x00004044 is not statically resolvable (compute shader 0x10527b8200).
• With: that line is gone; the same shader then stops at 438 s on ds_ordered_count at pc 16504: ordered swap is not modeled, a separate gap (we have a local change for it and will open a PR). With both, alive for the full 720 s run. Four other shaders still log could not be prepared at registration (its s_swappc_b64 at program counter 0x74 calls a computed target); preparing it when it is used, and none of them stops the run.
• The window stays black at ~0.1 FPS, so this is only about the stop. One run per build.

Tested and written with Claude Code (AI assistant) on a contributor's machine.

---

Generated by Claude Code