Skip to content

Upstream sync 6: shared AGC driver fixes onto Metal (slice A, first chunk) - #442

Merged
RINNECODER merged 52 commits into
mainfrom
upstream/agc-sync6
Oct 13, 2026
Merged

Upstream sync 6: shared AGC driver fixes onto Metal (slice A, first chunk)#442
RINNECODER merged 52 commits into
mainfrom
upstream/agc-sync6

Conversation

@RINNECODER

@RINNECODER RINNECODER commented Oct 13, 2026

Copy link
Copy Markdown
Owner

Part of #415. Refs #418 (first chunk; not all shared-logic commits are in, see "Remaining").

Ports upstream libSceAgcDriver fixes onto the fork's Metal path with git cherry-pick -x (each commit carries its upstream sha). Range: non-merge commits in 2ccbd376f..upstream/main touching core/libs/prx/libSceAgcDriver (118), plus the shared-file prerequisites from a11c8f7f7..2ccbd376f that later commits build on, plus two small open upstream PRs.

Note on range. 2ccbd376f is not merged into the fork: the fork's last upstream base is still a11c8f7f7, and none of the 104 commits in a11c8f7f7..2ccbd376f (the range #418's body describes) had been ported. This PR takes the 8 of those that touch Metal-compiled shared files and apply on the fork. The upstream MSAA rework in that range is not taken (see "Remaining").

What the Metal path compiles

The Metal and Rosetta builds compile only part of the driver (cmake/RosettaGraphics.cmake, core/shader/recompiler/MetalReplay/CMakeLists.txt): Graphics/src/State.cpp, ShaderInputState, ColorTargetLayout, SurfaceMetadata, DccClearEncoding, MeshDraw, Execution/src/Pm4.cpp, EntryPoints.cpp, Submit/*, Eq/*, State/*, Resource/*, Unimplemented.cpp. Graphics/src/Texture.cpp, DepthSurface.cpp, Draw.cpp, GuestMemory.cpp, GuestBufferMemory.cpp and ShaderRegistry.cpp are the Vulkan driver's and are not built on macOS. Commits for those files are taken where they apply cleanly, so later syncs do not conflict, but they are compile-checked only by upstream.

Ported

Shared logic the Metal path compiles:

  • 61a9f5dc1 render 1_5_5_5 UNORM color targets (State.cpp), with deaa2e3ef (test)
  • c914bcd72, 485c0f334 color mip targets with parent layout (ColorTargetLayout), with c58c9488e, 6c94d31d7, 4dfe0526b; Draw.cpp side 89c128782, 9394c7430
  • f27bdf4d7 64 KiB display XOR color targets (ColorTargetLayout), 4326c7c21 validate display XOR bits before resident lookup (State.cpp), b6f1a4a69 (test)
  • d9fbaf767 accept a CB metadata pass beside depth state without depth or stencil planes (State.cpp; conflict, hand-resolved to the fork's line)
  • fdfbeba8a accept MSAA_ENABLE on a single-sample target sampled at the pixel center (State.cpp). Adapted to the fork's MSAA model (rasterSamples from PA_SC_AA_CONFIG, AttachmentSampleCount): the fork's scanModeMask is replaced by upstream's check, and a single-sample scan with MSAA_ENABLE that the attachments raise to a multisampled pass stays refused. Upstream's sample-location tables are not taken.
  • 2a7b38ab7 KytyPS5 workload packets, workload stream registry (libSceAgc Workload.cpp, Resource/Registration.cpp); fork Commands.cpp tests merged with upstream's testWorkloads
  • 661f8e1e7 sceAgcAcbWaitUntilSafeForRendering (libSceAgc, Pm4.cpp)
  • bdb0be019 export sceAgcDriverGetHsOffchipParam (Unimplemented.cpp; the validation stubs upstream still had there stay in the fork's State/src/Status.cpp)
  • prerequisites from a11c8f7f7..2ccbd376f: a553e2263 WRITE_DATA DST_SEL 3, 5550ca63d wait for shutdown at shader ABI entrypoints (EntryPoints.cpp conflict: the fork's Metal no-op branches kept, upstream's ProcessShutdown handling wraps the Vulkan branch), 79c8d3690 split SWAP_ALT blends, b63996f1f merged program vertex tables, eba67d413 2D array color slice, 2f140cb8b zero for unexported parameters, c1287d73c depth surface left out without a tested plane, 82ac3bd93 read-only depth resummarize draws

Vulkan-side Graphics/Execution (applied cleanly; not built on macOS):

  • 92b46a2b7 clear integer storage images with DCC code 0000, 09aa34d06, b3c8be588 (tests; CMake conflict resolved to this commit's test only), 0b54425e3
  • ca99658ce memoize untracked texture comparisons
  • 61180981c, eb03e0e15, 85929389b, b35117c28 host-memory fallback when device memory runs out
  • 512ea3ec0, dd4504ee4 recorded fast clears synchronized with attachments
  • 2441a4b0d, cd067d156 host import retry and parked imports (CMake conflict: test list merged)
  • 471d9eced, fc7a9650f write tracker lifetime at teardown
  • 6b2432875 linear 2D texture with default-pitch DEPTH
  • 768929bb7, e71e3a4b3, 56df89366 shader preparation failures name address and stage (ShaderRegistry; applied cleanly onto the fork's shortened RegisterShader, no deleted path restored)
  • a3cba2d1f index move past vertex fetch records (VertexInput.hpp is also used by MetalDraw.mm)
  • 4309bffdb dword pattern-fill kernel only for 32-bit output
  • a55472cf9 test registration overlap

Open upstream PRs (shared State.cpp logic):

Fork commits:

  • link Resource/src/Registration.cpp into anyps5_metal_native_execution, so the Metal libSceAgc host fixture resolves AgcDriverWorkloadStreamRegistered (the Rosetta libs already link libSceAgc to libSceAgcDriver)
  • anyps5_metal_render_state gains SharedStateDecode: decodes a guest draw with BLEND_BYPASS and with MSAA_ENABLE on a single-sample scan and renders both through the Metal backend (blend 128,128,0,191 vs bypass 255,0,0,128; MSAA_ENABLE at the pixel center renders the same pixels as without; a sample off the center is refused). It fails on main ("blend bypass conflicts with enabled blending"). MSAA_ENABLE is read only when PA_SC_MODE_CNTL_0 is written, as before.

Docs (docs/dev/TechnicalDebt.md, docs/dev/debt/*.md) conflicts: the fork's text is kept and only the entry each commit adds is appended.

Skipped

Remaining (next chunk)

Tests

  • macOS guest replay config (-DANYPS5_METAL_UTILITIES_ONLY=ON -DANYPS5_METAL_SHADER_BRIDGE=ON -DANYPS5_BUILD_METAL_SHADER_BRIDGE_TESTS=ON -DANYPS5_METAL_GUEST_REPLAY=ON): full build, ctest --output-on-failure --timeout 300 36/36 passed.
  • macOS shader bridge config: full build, ctest 4/4 passed.
  • anyps5_metal_render_state with main's State.cpp: fails with "blend bypass conflicts with enabled blending"; with this branch: passes.
  • macOS Rosetta libs (x86_64, -DANYPS5_ROSETTA_LIBS=ON): targets rosetta_libs rosetta_engine rosetta_launcher_tests direct_memory_darwin_tests metal_utility_tests rosetta_direct_graphics_tests rosetta_cpuid_tests and rosetta_runtime_tests built; ctest -R 'direct_memory_darwin|metal_utility_tests|rosetta_direct_graphics|rosetta_lib_binds|rosetta_runtime|rosetta_launcher|rosetta_cpuid' 9/9 passed.

Game run (one, reference title)

tools/rosetta/run_title.py --wall 280 with this branch's Rosetta build, under with-lock eboot, on a private APFS clone of the data image mounted read-only (detached and deleted afterwards), private empty shader cache.

  • Guest entry after 0.6 s; ran the full 280 s until the wall limit's SIGTERM. 0 unresolved imports, 1 unimplemented call (an Ajm Opus mode, audio).
  • Furthest point: the title was loading its UI shell assets (boot-shell and DLC shell textures) at the end, past the early black/first-logo screen that Metal: cut per-draw shader conversions and copies (gpu-pipeline) #440's two 180 s cold-cache runs on main stopped at. No frame captures were taken (ANYPS5_FRAME_CAPTURE_DIR was not set), so there is no fps number from this run; Metal: cut per-draw shader conversions and copies (gpu-pipeline) #440's main figure is 28–31 fps on the early screen.
  • No AGC graphics: refusals in the log (none of the decode paths changed here refused). The Metal backend logged skipped draw ... Subgroup builtins are not available in this type of function (log-scaled counter reached x100000). That comes from the shader bridge's MSL conversion, which this PR does not touch; not compared against a main run.
  • Exit: after the wall-limit SIGTERM, a guest thread in sceAgcSuspendPoint got DriverStopped from MetalDriver::SuspendPoint and aborted (exit 134). That code is unchanged here; it is a teardown path on SIGTERM, worth its own issue.

🤖 Generated with Claude Code

songanane and others added 30 commits October 12, 2026 20:24
@songanane @claude @RINNECODER
feat(agc): write memory for WRITE_DATA with DST_SEL 3
35bfcde 
Killzone: Liberation (PPSA11142) writes its frame fence with
`sceAgcDcbWriteData(dst 6, cache policy 3, write confirm)`, which the
SPRX-confirmed encoding turns into WRITE_DATA with DST_SEL 3 and a label
address in the title's memory that RELEASE_MEM and WAIT_REG_MEM then
use. The validator rejected DST_SEL 3 as the GDS; it now writes guest
memory like DST_SEL 5, resolves as a label, and the title runs past its
first frame. TechnicalDebt records where the semantics come from.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
(cherry picked from commit 9c53b7f)
(cherry picked from commit a553e22)
@MarcZX10 @RINNECODER
fix(agc): wait for shutdown at shader ABI entrypoints
2513b04 
(cherry picked from commit 5550ca6)
@XairRrR @RINNECODER
fix(agc): reject constant color factors and memory-writing stages in …
204d186 
…split SWAP_ALT blends

Review follow-up to the 8_8 SWAP_ALT blending. The second channel of
such a target holds the guest's alpha. Vulkan applies the green blend
constant to it for CONSTANT_COLOR and ONE_MINUS_CONSTANT_COLOR, where the
guest's alpha equation uses the alpha constant. Decoding now throws for
these factors on a SWAP_ALT 8_8 target, unless the equation is MIN or
MAX, which ignore factors.

The second pass of a split draw runs every stage again, not only the
pixel shader, so a vertex or geometry stage that writes memory would
write twice. The memory-write check (now ShadersWriteMemory) covers all
the draw's stages.

(cherry picked from commit 79c8d36)
@songanane @claude @RINNECODER
fix(agc): read a merged program's vertex tables past its hidden user …
05f051d 
…words

A merged (NGG geometry path) program's user data starts with eight
hidden words, so the vertex buffer and attribute table offsets of its
AGC header, which count from the first user SGPR, were read eight words
too early and found zeros. Killzone: Liberation (PPSA11142) stopped on
`AGC graphics: null vertex buffer table address` at the first frame of
its PSP game. The decoder now adds the hidden word count to the table
reads; the registers reported to the recompiler are unchanged, as it
already maps them from the first user SGPR. `agc_driver_graphics` gains
a merged and a plain layout case.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
(cherry picked from commit b63996f)
@XairRrR @RINNECODER
fix(agc): render a slice of a 2D array color target into that layer o…
3001d3d 
…f the array

A color view of slice N of a 2D array (CB_COLOR_VIEW SLICE_START, with
CB_COLOR_ATTRIB3 MIP0_DEPTH as the last slice) was rendered as a 2D
surface at the slice's address. The swizzle equations of the XOR tile
modes take the slice index (the detiler's slice bits), so the texels of
slice N of an array are not where a 2D surface at that address puts
them: a title that renders into array slices and then samples the array
read them scrambled.

The decoded target now keeps the array's address, its last slice and the
view's slice. A resident target that is a slice of an array attaches
that layer of the array's storage image, so the texels follow the
array's layout; storage images of 2D arrays are color attachments for
this, and an attachment view selects a layer. A slice whose array has
DCC, CMASK or an attachment proxy has no resident array image, and a
non-resident slice of an XOR-tiled array now throws instead of being
written in the 2D layout.

(cherry picked from commit eba67d4)
@XairRrR @RINNECODER
fix(agc): read zero for pixel inputs of parameters the vertex stage d…
fe98b2c 
…oes not export

A pixel input whose SPI_PS_INPUT_CNTL offset names a parameter the
previous stage does not export failed validation ("graphics interfaces
disagree"), so the draw threw. No export writes that parameter slot for
the draw. radeonsi programs a pixel input without a matching vertex
output as OFFSET 0x20 with DEFAULT_VAL 0, so that it reads (0, 0, 0, 0)
(SI_PS_INPUT_CNTL_UNUSED in si_shader.h, applied in si_shader.c), and
rect-list draws already pass zero for such inputs.

Before the pixel stage of a draw is prepared, inputs that read a
parameter the previous stage's program does not export now read zero:
OFFSET bit 5 is set, DEFAULT_VAL and FLAT_SHADE are cleared, and an FP16
input also reads zero in its high half (USE_DEFAULT_ATTR1 set,
DEFAULT_VAL_ATTR1 cleared). Inputs that already read a default value
are unchanged.

Fixes boykopovar#2975.

(cherry picked from commit 2f140cb)
@oneandonlydean @RINNECODER
fix(agc): leave the depth surface out when no enabled test has its plane
2164cc1 
DecodeState bound the depth surface whenever DB_DEPTH_CONTROL enabled a
depth, stencil or depth bounds test and either plane was valid. A stencil
test beside a depth-only surface (DB_STENCIL_INFO format invalid, depth
test off) tests nothing, but the surface was still attached and the
render extent was clamped to it. A draw into a 1920x1080 target with a
960x540 depth surface left bound then covered only the top-left quarter.

A test whose plane is present but invalid no longer binds the surface.
Such a draw renders without a depth attachment, as it already does when
both planes are invalid. DrawRejection follows the same rule.

(cherry picked from commit c1287d7)
@Mouhand-Kaddo @RINNECODER
fix(agc): skip read-only depth resummarize draws
f74e27a 
(cherry picked from commit 82ac3bd)
@JAB-dev @claude @RINNECODER
feat(agc): render 1_5_5_5 UNORM color targets
06b3a2f 
CB_COLOR_INFO format 17 with UNORM and the standard component swap renders
into VK_FORMAT_A1R5G5B5_UNORM_PACK16, whose attachment and blend support
Vulkan requires. AMD keeps X in the low bits and Vulkan keeps R in the
high bits, so the exports are mapped with X and Z exchanged and the stored
bits follow the guest layout. PPSA06699 renders into this format.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
(cherry picked from commit 61a9f5d)
@devavijitsingh @RINNECODER
fix(agc): keep write tracker alive through driver teardown
42387f3 
Initialize the guest write tracker during Driver construction so its function-local static is destroyed after Driver on process exit.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
(cherry picked from commit 471d9ec)
@xXPBSXx @claude @RINNECODER
fix(libSceAgc): write the KytyPS5 workload packets and return null fo…
7ceb80c 
…r unregistered streams

sceAgcDcbSetWorkloadsActive writes an 18-dword IT_NOP packet (stream id, 64-bit workload mask) and sceAgcDcbSetWorkloadComplete a 12-dword one (stream id, workload id, 64-bit clear mask), following KytyPS5. Both return null for a null buffer, an unregistered or out-of-range stream, a workload id above 63, and for active a duplicate id or a count of 0 or above 63. sceAgcDriverRegisterWorkloadStream and sceAgcDriverUnregisterWorkloadStream now record the stream ids the DCB calls check.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
(cherry picked from commit 2a7b38a)
@devavijitsingh @RINNECODER
style(agc): remove unnecessary shutdown explanation
c677758 
Keep the tracker initialization fix without an explanatory code comment or technical-debt entry, as requested in review.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
(cherry picked from commit fc7a965)
@reyani-dev @claude @RINNECODER
fix(agc): accept a linear 2D texture whose DEPTH holds its default pitch
a0b4fc7 
On GFX10.3 the DEPTH field of a 1D or 2D non-array linear T# carries the
row pitch minus one (Mesa's ac_descriptors.c sets it for custom pitches,
which must be multiples of 256 bytes). A single-level, uncompressed 2D
linear texture whose DEPTH equals its default 256-byte-aligned pitch now
decodes as that 2D texture; any other pitch still throws.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
(cherry picked from commit 6b24328)
@danzaio @RINNECODER
test(agc): preparation failures name the shader address and stage
7ef506a 
FailureCapture now requires the code address and stage in the diagnosed failure, with and without the request dump. Fails before the wrap, passes after. AI-assisted: yes.

(cherry picked from commit 768929b)
@danzaio @RINNECODER
fix(agc): name the shader address and stage in preparation failures
607973c 
PrepareShaderWithDiagnostics dumped the request but rethrew the raw error, so a failure never said which shader it came from. Wrap it with the shader address and stage, keeping the dump. AI-assisted: yes.

(cherry picked from commit e71e3a4)
@danzaio @RINNECODER
fix(agc): preserve dumps for non-standard shader failures
010c105 
Capture the request once before rethrowing from the shared handler. Standard exceptions keep the shader address, stage, optional dump filename and full message; non-standard exceptions remain unchanged after capture.

AI-assisted: yes.
(cherry picked from commit 56df893)
@TibixDev @RINNECODER
fix(agc): retry failed host imports after range replacement
4064218 
Retain weak identities and the full failed span so a replacement of either owner or a shorter remaining run retries the import. Keep unchanged failures cached across unrelated mapping changes without pinning their allocations. Existing upstream import lifetime, span windows and Windows fallback remain in place. AI-assisted implementation.

(cherry picked from commit 2441a4b)
@diversao @claude @RINNECODER
fix(agc): release parked host imports a pinned guest range waits for
032089e 
A device replaced at AttachWindow keeps its host imports in
HostImports::parked until it is destroyed, and each parked import keeps
its guest ranges pinned. A guest munmap, mprotect or free of such a range
spun in the pin waiter until APS5_PIN_WAIT_MS and then failed, because the
waiter only looked at the live imports. The replaced device was already
waited idle by PrepareForReplacement, so the waiter now drops the parked
imports over the range.

agc_parked_import_unmap imports a range on one device, again on a second
one, and unmaps it with a 3 s pin wait.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
(cherry picked from commit cd067d1)
@utkuhalis @RINNECODER
feat(libSceAgc): implement sceAgcAcbWaitUntilSafeForRendering
007593f 
(cherry picked from commit 661f8e1)
@JAB-dev @claude @RINNECODER
test(agc): draw into a 1_5_5_5 target and read the 16 bits back
da0cfcd 
Draws 32_ABGR exports into a tiled and a linear format 17 target and compares every texel with
the layout X in bits 0-4, Y in 5-9, Z in 10-14, W in bit 15, including the X/Z-swapped pairs, rounding
edges and texels outside the scissor. Reverting the X/Z exchange fails it.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
(cherry picked from commit deaa2e3)
@devavijitsingh @RINNECODER
feat(agc): describe color mip targets with parent layout
6526e5d 
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
(cherry picked from commit c914bcd)
@devavijitsingh @RINNECODER
test(agc): cover packed color mip detiling
6c731d3 
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
(cherry picked from commit c58c948)
@devavijitsingh @RINNECODER
fix(agc): render nonresident packed mip targets
9fbcef2 
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
(cherry picked from commit 89c1287)
@devavijitsingh @RINNECODER
style(agc): keep mip layout API comment-free
fe53d6c 
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
(cherry picked from commit 485c0f3)
@pipe1os @RINNECODER
fix(agc): keep untailed color targets on their own detiler layout
0d140eb 
(cherry picked from commit 9394c74)
@pipe1os @RINNECODER
test(agc): render into packed mip tails through the guest-memory fall…
2cc97ff 
…back

(cherry picked from commit 6c94d31)
@pipe1os @RINNECODER
fix(agc): link the texture mip layout into the BDA device tests
4cbcc8c 
(cherry picked from commit 4dfe052)
@krishalivaibhav @RINNECODER
fix(agc): synchronize recorded fast clears with attachments
e1452f1 
(cherry picked from commit 512ea3e)
@krishalivaibhav @RINNECODER
test(agc): follow repository comment conventions
271c413 
(cherry picked from commit dd4504e)
@gravelas @claude @RINNECODER
fix(agc-driver): move indices far past every vertex fetch's records t…
a3c8505 
…o the first zero vertex

Goat Simulator 3 draws a triangle list whose indices are all 0xffffffff
with primitive restart off. Each fetch past a descriptor's records reads
zeros (out-of-bounds select 0 or 1), but the zero-padded vertex copies
were sized to the highest index: 160 GiB, which vkAllocateMemory refused.
When the highest index is at least 64Ki vertices past the largest
per-vertex record count, the draw now uses a copy of its indices with
every index past that count moved to it. Each attribute reads zeros
there as at the original index; only the vertex index the shader sees
changes, so indices closer to the records are left as they are.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
(cherry picked from commit a3cba2d)
GustavoGamarra95 and others added 17 commits October 12, 2026 20:25
@GustavoGamarra95 @RINNECODER
fix(agc): choose host memory from the heap budget instead of retrying
8bd6137 
Read the device-local heap's VK_EXT_memory_budget figures before a texture
or storage image allocation and pick a host memory type when the
allocation does not fit, rather than allocating again after the driver
refused the first attempt. The allocation itself is made once and still
throws when it fails.

(cherry picked from commit b35117c)
@ThatsLeo @RINNECODER
perf(agc): memoize untracked texture comparisons by the stores that r…
feeb82c 
…each the surface

The memo stamp held TrackerGeneration, which every collect and driver store
advances, so it almost never matched. Driver stores into uncovered ranges
record their generation per 64 KiB block, and NewestStamp of the surface
replaces TrackerGeneration in the stamp.

(cherry picked from commit ca99658)
@AlxcBTW @claude @RINNECODER
fix(agc): take the dword pattern-fill kernel only through a 32-bit ou…
f482724 
…tput format

The kernel boykopovar#2081 matches stores each pattern dword with
buffer_store_format_x, so what lands in memory depends on the output
descriptor's FORMAT (bits 12-18 of user[3]). Only 32_UINT, 32_SINT and
32_FLOAT (20-22) store the dword unchanged. Other formats convert it,
and an 8- or 16-bit one writes 1 or 2 bytes of every 4-byte element, so
a buffer fill would also overwrite the bytes in between. Such a dispatch
now goes to the recompiler.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
(cherry picked from commit 4309bff)
@hansjm10 @RINNECODER
feat(libSceAgcDriver): export sceAgcDriverGetHsOffchipParam
f434eaf 
(cherry picked from commit bdb0be0)
@Amoenus @RINNECODER
feat(agc): support 64 KiB display XOR color targets
09a6f9e 
Map SW_64KB_D_X color surfaces through the existing source-backed texture equations and mip geometry. Keep display and render-target equation caches separate. Reject unresolved raw column bits for the new mode instead of accepting them.

Cover the captured PPSA99902 descriptor, independent source offsets, pipe/bank tags, transfers, array/mip placement and resident/non-resident GPU draws. Record unmeasured console behavior in TechnicalDebt.

CPU validation on main e013f8c: captured-descriptor failure 5/5; changed passes 5/5; pinned AMD addrlib reference 536940 addresses per run, five runs. Full build and GPU validation are tracked externally. AI-assisted implementation and validation: Codex.

(cherry picked from commit f27bdf4)
@Amoenus @RINNECODER
fix(agc): validate display XOR bits before resident lookup
cdd57cc 
Apply the source-configured D_X XOR refusal during color decoding as well as direct layout construction. Regression tests reject each raw column-bit combination before storage or rendering is selected. Existing resident-array and packed-mip refusal paths remain in force.

AI-assisted implementation and validation: Codex.

(cherry picked from commit 4326c7c)
@Amoenus @RINNECODER
test(agc): verify display XOR array attachment transport
5fd202f 
Run the existing array-layer draw, recompiled shader load, untouched-layer and guest-publication checks with D_X as well as R_X. This exercises the resident array path that carries slice-dependent address XOR; a direct per-slice copy is not an equivalent model.

AI-assisted implementation and validation: Codex.

(cherry picked from commit b6f1a4a)
@CnRJay @RINNECODER
perf(agc): clear integer-format storage images with DCC code 0000 on …
e7a60f6 
…the GPU

ClearColorFor refused every integer format, so a fast-cleared R32_UINT/R16_UINT/R8_UINT surface was uploaded through the CPU path: a guest read that waits for the GPU writers, then FillDccClear's zero texels. Code 0000 is zero bits in any format, so it maps to a zero integer clear. PPSA28416: ~1.4 s per 10 s of CPU uploads gone, 4.9 -> 5.3 fps.

(cherry picked from commit 92b46a2)
@adriadam10 @RINNECODER
test(agc): pin the integer storage clear to DCC code 0000
938de81 
StorageClearAvailable takes code 0000 on R8/R16/R32_UINT and refuses every
other code; drop the header comment that said integer formats were refused.

(cherry picked from commit 09aa34d)
@adriadam10 @RINNECODER
test(agc): run the integer storage clear check from ctest
4b4f889 
(cherry picked from commit b3c8be5)
@krishalivaibhav @RINNECODER
test(agc): remove non-discriminating recorded clear fixture
1940215 
(cherry picked from commit 0b54425)
@Nesyth @RINNECODER
fix(agc): accept a CB metadata pass beside depth state without depth …
d49cddd 
…or stencil planes

(cherry picked from commit d9fbaf7)
@oneandonlydean @claude @RINNECODER
fix(agc): accept MSAA_ENABLE on a single-sample target sampled at the…
40c2771 
… pixel center

(cherry picked from commit fdfbeba)

Fork adaptation: applied to the fork MSAA model (rasterSamples from PA_SC_AA_CONFIG); a single-sample scan with MSAA_ENABLE that the attachments raise to a multisampled pass stays refused. Upstream sample-location tables are not taken here.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@Tamalero @claude @RINNECODER
fix(agc): leave blending off for a color target with CB_COLOR_INFO.BL…
f57452e 
…END_BYPASS

DecodeState threw "blend bypass conflicts with enabled blending" when
CB_BLENDn_CONTROL enabled blending for a target whose CB_COLORn_INFO has
BLEND_BYPASS (bit 16, Mesa src/amd/registers/gfx10.json) set. The bit makes
the color block skip the blender for that target: Mesa sets it "according
to docs" for UINT/SINT and 8_24/24_8/X24_8_32_FLOAT targets, which the
blender cannot process (src/amd/common/ac_descriptors.c, shared by radeonsi
and RADV). Marvel's Guardians of the Galaxy (PPSA01748) issues such draws
right after its loading screen. Blending is now enabled only when it is
requested and not bypassed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
(cherry picked from commit baaad78)
@Tamalero @claude @RINNECODER
fix(agc): accept a CB metadata pass whose depth test passes every fra…
15fa612 
…gment without writing

DecodeColorMetadataPass refuses an eliminate-fast-clear or DCC-decompress
pass when DB_DEPTH_CONTROL has any of its low four bits set and depth or
stencil planes are bound. Marvel's Guardians of the Galaxy (PPSA01748)
issues one with the Z test enabled, ZFUNC ALWAYS and no Z write
(DB_DEPTH_CONTROL 0x72) while a Z32 and an S8 target are bound. That test
passes every fragment and leaves depth unchanged, so the pass does no
depth work. Draws already treat this state as a pass-through
(depthPassThrough); the metadata pass now uses the same rule. Depth
bounds, depth writes, stencil tests and DB_RENDER_CONTROL clears or copies
are still refused, and the error names the four registers involved.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
(cherry picked from commit b90ec1e)
@RINNECODER @claude
build(metal): link the workload stream registry into the native AGC e…
750e395 
…xecution

libSceAgc's workload packets (upstream 2a7b38a) ask libSceAgcDriver whether a stream
is registered. The Rosetta libs link the two libraries; the Metal utilities build
links libSceAgc's host objects against anyps5_metal_native_execution, which now
carries Resource/src/Registration.cpp.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@RINNECODER @claude
test(metal): cover ported State decode fixes on the Metal path
cf3e75f 
anyps5_metal_render_state now decodes a guest draw with CB_COLOR_INFO BLEND_BYPASS
(upstream PR 4862) and with MSAA_ENABLE on a single-sample scan (upstream
fdfbeba) and renders both through the Metal backend. MSAA_ENABLE is read only
when PA_SC_MODE_CNTL_0 is written, as the fork's decode accepted before.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@RINNECODER RINNECODER added the upstream-sync Upstream boykopovar/AnyPS5 intake label Oct 13, 2026
@chatgpt-codex-connector

chatgpt-codex-connector Bot commented Oct 13, 2026 •
edited
Loading

Copy link
Copy Markdown

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review Status Commit Review trigger
📝 Code Review ✅ Completed 2026-10-13T01:47:29.797036Z cf3e75f PR opened
ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

@coderabbitai

coderabbitai Bot commented Oct 13, 2026 •
edited
Loading

Copy link
Copy Markdown

Important

  • 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration
  • Configuration used: defaults
  • Review profile: CHILL
  • Plan: Advanced
  • Run ID: 1796fa82-fd86-4737-a350-9e631064d162
  • Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

Comment @coderabbitai help to get the list of available commands.

@RINNECODER

RINNECODER commented Oct 13, 2026

Copy link
Copy Markdown
Owner Author

Merged main (#441) into the branch; one docs conflict in docs/dev/TechnicalDebt.md (kept main's removal, kept this branch's new entries). Re-ran after the merge: Rosetta libs build + regex ctest 9/9, guest-replay build + ctest 36/36.

This was referenced Oct 13, 2026
Closed
Closed
@RINNECODER
RINNECODER merged commit 87d1e9b into main Oct 13, 2026
5 checks passed
@RINNECODER
RINNECODER deleted the upstream/agc-sync6 branch October 13, 2026 01:46
chatgpt-codex-connector[bot]
chatgpt-codex-connector Bot reviewed Oct 13, 2026
View reviewed changes

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: cf3e75f357

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

Comment thread core/libs/prx/libSceAgcDriver/Graphics/include/VertexInput.hpp
if (VertexBufferOutOfBoundsSelect(attribute) > 1u) return std::nullopt;
zeroFrom = std::max(zeroFrom, attribute.resource.fields[2]);
}
return zeroFrom;

@chatgpt-codex-connector chatgpt-codex-connector Bot Oct 13, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Badge Skip index clamping when no vertex fetch qualifies

ZeroFetchVertex returns an engaged optional containing zero when the shader has no attributes, or when every attribute is null, zero-stride, instance-indexed, or otherwise skipped. Consequently, prepareDrawInputs rewrites 32-bit indices at least 65,536 past zero to index 0 even though there is no vertex-indexed buffer fetch to protect. Procedural shaders that derive positions from the vertex ID, or shaders with only instance attributes, therefore receive incorrect vertex IDs and render different geometry; return nullopt unless at least one qualifying vertex fetch contributed a record limit.

Useful? React with 👍 / 👎.

Comment thread core/libs/prx/libSceAgcDriver/Execution/src/Driver/Draw/DrawPrecheck.cpp
localDevice->ColorMetadataPass(*pass);
return DrawVerdict::Drawn;
}
if (Graphics::DepthResummarizeDrawsNothing(queue)) return DrawVerdict::Nothing;

@chatgpt-codex-connector chatgpt-codex-connector Bot Oct 13, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Badge Preserve shader writes from resummarize draws

When this predicate matches a read-only resummarize draw, the precheck returns before decoding or executing any shader. PixelProgramSkipped only proves that the pixel stage is absent; a vertex, tessellation, or geometry stage can still perform supported guest-buffer or image writes (as accounted for by ShadersWriteMemory in Pipeline.cpp). Such a draw now silently loses those side effects instead of being rejected, so this fast path should only classify the draw as empty after ruling out shader writes, or retain the previous rejection for that case.

Useful? React with 👍 / 👎.

This was referenced Oct 13, 2026
Merged
Merged
Merged
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@chatgpt-codex-connector chatgpt-codex-connector[bot] chatgpt-codex-connector[bot] left review comments

Assignees

No one assigned

Labels

upstream-sync Upstream boykopovar/AnyPS5 intake

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.