[ANONYMOUSZED-beep]

What

Part of #6191: the Guardians of the Galaxy chain (constant buffer → pointer → pointer → T#). The Ratchet & Clank chain (record → pointer → V# → T# at a second readfirstlane index) needs a second key and is not covered here.

#4475 plans an image-table record that holds a pointer to the T#. When the pointer itself is loaded through another pointer, the column was refused: TryMakeIndirectColumn wanted both pointer dwords to be scalar buffer reads, so the stop is GetImageResource dword 0 is not a valid runtime value with the chain LoadAddressU32(GetAddressResource(LoadAddressU32(GetAddressResource(ReadConstBuffer …)))).

• Tracker. Walks back from the T#'s GetAddressResource while both of its dwords are consecutive LoadAddressU32 reads (offset n and n + 4, immediate offset, same or equivalent pointer). Each step is one PointerHop {offset, immediate}. The record at the end of the chain is matched as before. Any depth is accepted; a chain that does not end in a keyed record read is refused as before.
• Materializer. After it reads a record's pointer, it follows column.pointerHops in order. A hop that is unaligned, wraps, or reaches an unmapped page makes the record Unmapped poison, like an unreadable T#. Each link's 8 bytes are added to the table's ranges, so a changed link rebuilds the table.
• Planning-only loads. A hop load is planning-only when every use of its pointer is a planning-only load. This runs to a fixed point, so the whole chain stops reading through BDA.
• TableColumn gains pointerHops; the plan encoder and decoder carry it, and the ShaderDiskCache layout asserts are updated. A column with no hops is planned and encoded as before.

Tested

Amazon Linux 2023 (x86-64, GCC 14.2.1, Release), main d92a8dad3. I have no GPU and no game, so I could not run Guardians or Ratchet.

• agc_shader_bindless: new verifyPointerChainImageTable. A compute shader reads a record pointer with s_buffer_load_dwordx2 at key * 16 + 8, loads a second pointer with s_load_dwordx2 at +0x10, and the T# with s_load_dwordx8 at +0x20. It checks the column (indirect, stride 16, offset 8, pointerHops == {{0, 0x10}}), that every scalar address load is planning-only and the shader does not use BDA, that two chains to the same T# share one element and a different T# gets its own, that records whose chain hits a null pointer are Unmapped poison, that the links are in the captured regions and snapshot ranges, and that moving one record's second pointer changes its T#.
  • On the unchanged tracker it fails with GetImageResource dword 0 is not a valid runtime value; chain: LoadAddressU32@pc=0x4c(GetAddressResource(LoadAddressU32@pc=0x44(GetAddressResource(ReadConstBuffer@pc=0x3c(…, the same shape as the Guardians report.
  • With the fixed-point marking turned off it fails with "a load along the pointer chain still reads through BDA".
• agc_shader_disk_cache. The hand-built plan now has two pointer hops. It passes, and it fails ("the decoded hand-built plan differs…") when the hop encoding is removed.
• agc_driver_image_table: new RunPointerChainRecordTests (PointerChainRecordCode, which is IndirectRecordCode plus one s_load_dwordx2). It checks texels per record, one element per distinct T#, and that moving record 1's second pointer changes what it samples. It passes on lavapipe (Mesa 24.2.6, CPU only, no hardware GPU). On the unchanged tracker it fails with the same GetImageResource dword 0 chain.
  • On lavapipe the rest of agc_driver_image_table behaves the same with and without this change. RunAnisoOverrideTests fails ("a direct pair changed its texels"), and RunGraphicsTests did not finish within 15 minutes. Each group was run on its own: the other groups pass, including RunPointerRecordTests, RunSplitRecordTests and RunPointerStorageWriteTests. I could not run the full suite on real hardware.
• agc_shader_memory, agc_shader_bindless, agc_shader_srt_read_patching, agc_shader_frontend_pair(_off), agc_shader_quad_mask_no_demote, agc_shader_layout_pair(_off) and agc_shader_disk_cache pass.
• python3 tools/check_conventions.py --base origin/main reports no findings for the code.

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 (none added)
• 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 (none)
• Generic behaviour, not specific to one title
• Depends on: none (feat(agc): image-table records that hold a pointer to the T# or split it over two loads #4475 is merged)
• Unblocks: none confirmed (meant for PPSA01748 in Recompiler: a T# reached through two loads after a dynamically indexed read (pointer → pointer → T#, pointer → V# → buffer → T#) stops Ratchet & Clank (PPSA01474) and Guardians of the Galaxy (PPSA01748) #6191, not run on the title)
• GPUs tested: lavapipe only (Mesa 24.2.6 llvmpipe, Amazon Linux 2023, Intel Xeon CPU); no hardware GPU, so a test on AMD/NVIDIA/Intel is still needed
• AI-assisted: yes

[github-actions]

Other open pull requests touch the same code. Merging one will break the other, or both implement the same thing. If this one builds on them, list them in Depends on.

Pull request	Conflicting files (lines on main)	Same exports	Same new files
PR 6440	core/libs/prx/libSceAgcDriver/tests/ShaderMemory.cpp L2517, L6166, L6219
PR 7395	core/libs/prx/libSceAgcDriver/tests/execution/ImageTable.cpp L281-282, L1218 core/shader/recompiler/IntermediateRepresentation/include/IntermediateRepresentation/IrMetadata/ControlFlowInfo.hpp L51 core/shader/recompiler/Optimization/src/ResourceMaterializer.cpp L821 core/shader/recompiler/Optimization/src/ResourceTracker.cpp L690-692, L699, L734-735 core/shader/recompiler/ShaderDiskCache.cpp L152

Some comparisons could not be completed because a pull request conflicts with the base branch. Review these shared files manually before concluding that there is no overlap.

Pull request	Shared files not compared
PR 907	core/shader/recompiler/Optimization/src/ResourceMaterializer.cpp
PR 1341	core/libs/prx/libSceAgcDriver/tests/ShaderMemory.cpp core/shader/recompiler/Optimization/src/ResourceMaterializer.cpp
PR 3247	core/libs/prx/libSceAgcDriver/tests/ShaderDiskCache.cpp core/shader/recompiler/ShaderDiskCache.cpp
PR 5371	core/libs/prx/libSceAgcDriver/tests/ShaderMemory.cpp core/shader/recompiler/Optimization/src/ResourceMaterializer.cpp
PR 5500	core/libs/prx/libSceAgcDriver/tests/ShaderMemory.cpp core/shader/recompiler/ShaderDiskCache.cpp
PR 5609	core/libs/prx/libSceAgcDriver/tests/ShaderDiskCache.cpp core/shader/recompiler/ShaderDiskCache.cpp
PR 5925	core/shader/recompiler/ShaderDiskCache.cpp
PR 6177	core/libs/prx/libSceAgcDriver/tests/ShaderDiskCache.cpp core/libs/prx/libSceAgcDriver/tests/ShaderMemory.cpp
PR 6235	core/libs/prx/libSceAgcDriver/tests/ShaderDiskCache.cpp core/libs/prx/libSceAgcDriver/tests/ShaderMemory.cpp
PR 6280	core/libs/prx/libSceAgcDriver/tests/ShaderDiskCache.cpp core/shader/recompiler/Optimization/src/ResourceMaterializer.cpp core/shader/recompiler/Optimization/src/ResourceTracker.cpp core/shader/recompiler/ShaderDiskCache.cpp
PR 6312	core/shader/recompiler/Optimization/src/ResourceMaterializer.cpp
PR 6412	core/libs/prx/libSceAgcDriver/tests/ShaderDiskCache.cpp core/libs/prx/libSceAgcDriver/tests/ShaderMemory.cpp core/shader/recompiler/Optimization/src/ResourceMaterializer.cpp core/shader/recompiler/ShaderDiskCache.cpp
PR 6481	core/libs/prx/libSceAgcDriver/tests/execution/ImageTable.cpp
PR 6486	core/libs/prx/libSceAgcDriver/tests/ShaderDiskCache.cpp core/shader/recompiler/ShaderDiskCache.cpp
PR 6488	core/libs/prx/libSceAgcDriver/tests/ShaderDiskCache.cpp core/libs/prx/libSceAgcDriver/tests/ShaderMemory.cpp core/shader/recompiler/IntermediateRepresentation/include/IntermediateRepresentation/IrMetadata/ControlFlowInfo.hpp core/shader/recompiler/Optimization/src/ResourceMaterializer.cpp core/shader/recompiler/Optimization/src/ResourceTracker.cpp core/shader/recompiler/ShaderDiskCache.cpp
PR 6503	core/libs/prx/libSceAgcDriver/tests/ShaderMemory.cpp
PR 6539	core/libs/prx/libSceAgcDriver/tests/ShaderDiskCache.cpp core/shader/recompiler/ShaderDiskCache.cpp
PR 6645	core/libs/prx/libSceAgcDriver/tests/ShaderMemory.cpp
PR 6683	core/libs/prx/libSceAgcDriver/tests/ShaderMemory.cpp
PR 6879	core/shader/recompiler/ShaderDiskCache.cpp
PR 6941	core/libs/prx/libSceAgcDriver/tests/ShaderDiskCache.cpp core/libs/prx/libSceAgcDriver/tests/ShaderMemory.cpp core/shader/recompiler/ShaderDiskCache.cpp
PR 6970	core/libs/prx/libSceAgcDriver/tests/ShaderDiskCache.cpp core/shader/recompiler/ShaderDiskCache.cpp
PR 6977	core/shader/recompiler/Optimization/src/ResourceMaterializer.cpp core/shader/recompiler/ShaderDiskCache.cpp
PR 7223	core/libs/prx/libSceAgcDriver/tests/ShaderDiskCache.cpp core/libs/prx/libSceAgcDriver/tests/execution/ImageTable.cpp core/shader/recompiler/IntermediateRepresentation/include/IntermediateRepresentation/IrMetadata/ControlFlowInfo.hpp core/shader/recompiler/Optimization/src/ResourceMaterializer.cpp core/shader/recompiler/Optimization/src/ResourceTracker.cpp core/shader/recompiler/ShaderDiskCache.cpp
PR 7332	core/libs/prx/libSceAgcDriver/tests/ShaderMemory.cpp
PR 7381	core/libs/prx/libSceAgcDriver/tests/ShaderDiskCache.cpp core/libs/prx/libSceAgcDriver/tests/ShaderMemory.cpp core/shader/recompiler/Optimization/src/ResourceMaterializer.cpp core/shader/recompiler/Optimization/src/ResourceTracker.cpp core/shader/recompiler/ShaderDiskCache.cpp
PR 7388	core/libs/prx/libSceAgcDriver/tests/ShaderMemory.cpp core/shader/recompiler/Optimization/src/ResourceTracker.cpp
PR 7390	core/shader/recompiler/ShaderDiskCache.cpp
PR 7642	core/libs/prx/libSceAgcDriver/tests/ShaderDiskCache.cpp core/libs/prx/libSceAgcDriver/tests/ShaderMemory.cpp core/shader/recompiler/Optimization/src/ResourceMaterializer.cpp core/shader/recompiler/Optimization/src/ResourceTracker.cpp core/shader/recompiler/ShaderDiskCache.cpp

[Neithennn]

Windows NVIDIA run, since the description lists lavapipe only.

Windows 11 Pro 26200, NVIDIA GeForce RTX 4080 (driver 617.42), AMD Ryzen 7 5800X, WinLibs GCC 15.2.0 r7, Ninja, Release, bundled dependencies, BUILD_TESTING=ON, ANYPS5_NO_SHADER_CACHE=1. Head 6b662dd8b merged onto main ff23657f3; full build succeeds.

• Targeted, with ANYPS5_REQUIRE_VULKAN=1: agc_shader_bindless, agc_shader_memory, agc_shader_disk_cache, agc_driver_bindless_images, agc_driver_image_table, agc_driver_image_table_cache, agc_driver_image_table_cache_bounds, agc_driver_image_table_reuse: 8/8 pass. verifyPointerChainImageTable runs in both agc_shader_bindless and agc_shader_memory.
• ctest -R "agc_|shader|recompiler|spirv|video_out" -j 8: 622/623, same as unchanged main on this machine; the one failure is agc_driver_store_over_untracked (Windows: StoreOverUntracked fails at "the first result was not stored" #6347/AGC: StoreOverUntracked's NeighbourCopied case intermittently misses its pending result on Windows / RTX 4080 #6383), also failing on main.

As far as I can tell from the source, verifyPointerChainImageTable compiles and materializes on the host and does not dispatch on the GPU, so this mainly shows the change does not regress the GPU-dispatching image-table tests on NVIDIA/Windows. No title run.

AI-assisted: yes (runs done with Claude Code).

[Tamalero]

Title run on Guardians of the Galaxy (PPSA01748), the title this PR targets from #6191. Head 6b662dd8b merged onto main 03c4cb5e7, with a control on that main alone. Both use the same relinked executable and no local patches.

Build	Level-1 load (shield screen)
main 03c4cb5e7	music starts, then stops on the #6191 chain: shader 0x3b94a7500 stage 5, GetImageResource dword 0 is not a valid runtime value, LoadAddressU32@pc=0x84(GetAddressResource(LoadAddressU32@pc=0x70(GetAddressResource(ReadConstBuffer@pc=0x30(…ReadFirstLane(GetInterpolationParameter…
this PR	no descriptor stop in 3 runs; the load gets further, then stops on something unrelated (below)

So the constant buffer → pointer → pointer → T# half of #6191 is covered on the title. Menu frame rate with a warm shader cache is the same on both builds, about 11–14 FPS.

The next stop is in the draw path, not in this PR's code. A DRAW_INDEX_INDIRECT_MULTI (opcode 0x38, 144 records, stride 20) is issued after INDEX_BUFFER_SIZE is set to 0x10000000 with 16-bit indices. For indirect indexed draws the driver binds that whole declared range (Pm4.cpp, "the declared range is the hardware's view of the index buffer"), so it asks for 512 MiB at 0x383240000. Only about 54 MiB of it is mapped:

AGC driver: guest memory is not readable at 0x386590000 (range 0x383240000+0x20000000)

The stack, from a local print added for this run only: Driver::draw → Pm4::ResolveDraw (indirect path) → GuestMemory::CheckRange. I'll open a separate issue for it.

Details

• Linux (CachyOS), RTX 4090, i9-14900K. The executable was relinked with --to-intel by the relinker at 03c4cb5e7.
• Runs on this PR: three normal runs and one with a local stack print. All stop at the same range; the first unreadable page is 0x3865a0000 in one run and 0x386590000 in the others. With a warm cache the stop comes before the shield screen is drawn, though music has already started.
• Still unplanned on both builds, as listed in Recompiler: a T# reached through two loads after a dynamically indexed read (pointer → pointer → T#, pointer → V# → buffer → T#) stops Ratchet & Clank (PPSA01474) and Guardians of the Galaxy (PPSA01748) #6191: the three 0x3b91c3e00-family shaders, which use the pc 0x4378/0x4380 key read through ImageRead. They are deferred at registration and don't stop the load.
• Not run: Ratchet & Clank, which the description says this PR doesn't cover.

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