fix(shader): serve ds_ordered_count in wave launch order by ZSWolt · Pull Request #12944 · boykopovar/AnyPS5

ZSWolt · 2026-10-29T15:42:29Z

What

Background. On AMD GCN/RDNA GPUs, GDS (global data share) is a small on-chip memory shared by all waves of a dispatch. ds_ordered_count is a GDS instruction that adds to, or swaps, a counter in GDS and returns the old value. Unlike an ordinary atomic, the hardware serves the waves in wavefront launch order: the first-launched wave gets the first value, the next wave the next, and so on. Shaders use it to give each wave a contiguous, ordered slice of an output buffer (stream compaction, particle lists, culling output). Because the order is guaranteed, the result is deterministic and the slices never overlap.

The bug. ds_ordered_count (ordered append on GDS) ran as a plain device atomic, so the waves of a dispatch added in whatever order the host ran them. The hardware hands the counter out in wavefront launch order. radeonsi's si_compute_prim_discard.c (mesa-20.3.0) describes it as "an atomic counter that is incremented in the wavefront launch order", with ORDER_MODE 0 launching in order. A title that turns the returned values into output offsets therefore got overlapping or shifted ranges, intermittently.

In compute shaders the ordered add and swap now run in launch order:

Launch index: each wave computes its launch index: the flattened workgroup ID times the waves per workgroup, plus the wave index in the workgroup.
Waiting for its turn: one lane per guest wave waits until a turn dword kept one dword past the 64 KiB GDS equals that index. The GDS binding grows by two dwords (RuntimeAbi::GdsMetadataDwords), the turn and a timeout count (Pm4::GdsBindingBytes). Ordinary GDS accesses stay bounded at 64 KiB; only the ordering code reaches the two dwords. The wait, the add and the release run inside one loop, so guest waves that share a host subgroup (wave32 code in a 64-lane subgroup) do not wait on each other.
Passing the turn on: it performs the add or swap, then, when offset1 bit 1 (wave done) is set, compare-exchanges the turn from its own index to the next (0 after the dispatch's last wave). Only the wave that holds the turn moves it.
Reset per dispatch: the compiled program records whether it orders (orderedAppend, disk cache format 57). For those programs only, the driver zeroes the turn in the buffer the dispatch binds, before the dispatch and before each chunk of a split one. It uses a COMPUTE_SHADER → TRANSFER barrier, a 4-byte vkCmdFillBuffer, then TRANSFER → COMPUTE_SHADER. A GDS copy the CPU made is created fillable, so it is reset the same way. Each chunk counts its own waves.
Bounded wait: a wave goes ahead after 2^20 reads of the turn (about 130 ms on an RX 9060 XT). It adds unordered (as on main), marks the turn abandoned so the rest of the dispatch runs unordered at once, and adds 1 to the timeout count, which the test checks. With one wave that never takes its turn, a 65536-group dispatch takes about 350 ms; without the mark it lost the device after 2 s.
Graphics stages keep the unordered add.

The wait relies on waves being launched in workgroup order and on earlier waves running while later ones spin. AMD hardware does this, but Vulkan does not guarantee it. This is recorded in docs/dev/debt/shader.md, which now describes the new behaviour in place of "not modeled".

Tested

Ubuntu 26.04.1 (Linux 7.0.0-38), GCC 15.2.0, Release, RX 9060 XT (RADV GFX1200, Mesa 26.0.8-1ubuntu0.3). Based on main 6d56eb8 (merged into the branch on Oct 30).

New cases in agc_driver_gds_append_consume, each checking every lane's value, the counter and a timeout count of 0:
- 64 single-wave workgroups each add 1, and group g must get start + g. Runs 8 dispatches back to back.
- The same with group 0 delayed by a scalar loop, so later waves reach the instruction first.
- 16 groups of 4 waves, group 0 delayed.
- 4 groups where group 1 skips ds_ordered_count: group 0 gets the start value, groups 2–3 time out and get the remaining values, and exactly 2 timeouts are counted.
- A normal dispatch right after that one, with the turn left at 1: it is ordered only because of the per-dispatch reset.
- On main bb04a4f (with only this test file added) it fails: "64 single-wave groups, dispatch 1, ds_ordered_count: group 0 lane 0 returned 1025, expected 1000 (the counter in launch order)".
- With this change it passes.
- The existing single-workgroup ordered add and swap cases still pass.
Split-dispatch case (1 x 65537 single-wave groups, row 1 skips ds_ordered_count), in agc_driver_gds_append_consume and _copied. The RX 9060 XT on RADV reports maxComputeWorkGroupCount[1] = 65535, so the case takes the split path here: 2 chunks, row 0 gets the start value, 65533 waves of the first chunk run unordered, and the second chunk and the dispatch after it are ordered. It passes 3 of 3 in both registrations, with 0 VUID- messages.
- On AMD's Windows driver (RX 9070 XT, reported by BunnyFl) row 0 got a value about 65,300 places late, which means it started after most of the dispatch. Vulkan does not guarantee that workgroups start in WorkgroupId order. For a row 0 served that late, the case skips only the row 0 value and the unordered count, and names the skip in its success line; the skipped row, the second chunk's order after its reset and the final counter are still checked. A row 0 served early but not first still fails. The debt entry names the assumption.
ctest -R '^agc_|^shader_|recompiler' -j4 on this branch: 1130 of 1132 pass. The 2 failures, agc_idle_publish_wait and agc_driver_native_multisample_depth, also fail on main on this machine (the second because a host import gets vkAllocateMemory -13 here).
Validation-layer compare: ANYPS5_REQUIRE_VULKAN=1 ANYPS5_NO_SHADER_CACHE=1 VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation ctest -V -R gds_append on the same machine and GPU, on main bb04a4f and on this branch.
- Neither log has the "was not found but was requested" line, and both have 0 VUID- messages.
- On main the test fails only on the new case, as intended.
God of War Ragnarök (PPSA08329), title-screen fireplace scene. Plain main stops at startup in this title (sceHttp2SetResolveRetry not implemented), so this ran on my local integration build (main plus local branches) with and without this PR.
- Before the change, a particle compute shader that uses ordered append (0x1038fb8200) intermittently produced overlapping particle records, drawn as long streaks. That was 1–9 corrupted frames per 8-minute run; with the change, 0 in 3 runs (Oct 23, first version of this PR).
- A/B on the current head, tools/trace_digest.py, APS5_TRACE_SECONDS=200:60, order without, with, with, without:
  - presents 203 / 217 / 214 / 198;
  - frame p50 287.0 / 275.9 / 277.9 / 292.2 ms;
  - gpu_busy_ms_per_present 148.5 / 128.9 / 128.6 / 151.9. This metric spans each batch from its first to its last command, idle gaps included, so it is not shader time. I have not attributed the change to a phase.
- Earlier versions of this description gave presents "851–857 against about 790" on main. That was wrong: those were runs of the integration build against a baseline from another day.
python3 tools/check_conventions.py --base origin/main: all conventions checks passed.
tools/pr_overlap.py --branch fix/recompiler-ordered-append: no open PR changes ds_ordered_count. The overlaps are on shared files only: Graphics/src/ShaderResources.cpp, SpirvMemoryInstructions.cpp, SpirvEmitterHelpers.cpp, Execution/src/Pm4.cpp and docs/dev/debt/shader.md (e.g. feat(agc): serve lane-keyed, pointer and runtime image-table columns #10193, feat(agc): test the depth bounds in the pixel program when the device lacks the depthBounds feature #11551, fix(agc): span table-reading wave64 programs by default on Intel drivers that share handles across fused threads #10846).

Progress

PPSA08329 on this machine (RX 9060 XT, RADV), on main with this change and other local, unrelated fixes; this PR depends on none of them. Bring-up thread: #12128.

Streaks: the particle streaks from earlier runs are gone. Particle draws read records that compute shader 0x1038fb8200 writes through ds_ordered_count slots, and serving those slots in launch order removed every streak.
Gameplay reached: from a new game, the opening cutscene and the sled chase run through the first Freya fight, with the forest, rocks, snow ground and characters drawn. A 10-minute run from that save on Oct 29 showed the fight lit, with the forest and snow present.
Frame time, title-screen fireplace scene (Oct 29), A/B on the local build with and without this PR:
- presents in 60 s: 203 / 198 without, 217 / 214 with;
- gpu_busy_ms_per_present: 148.5 / 151.9 without, 128.9 / 128.6 with.
Still open, not addressed here: the CPU-bound packet thread (about 2 FPS in the chase).

Checklist

🤖 Generated with Claude Code

The hardware hands out an ordered-append counter in wavefront launch order. Compute waves now wait for a turn dword kept past the 64 KiB GDS to reach their launch index (flattened workgroup ID times waves per workgroup plus the wave index), perform the add or swap, and pass the turn on when the wave-done bit is set; the last wave of a dispatch resets it. The wait is bounded so a wave that never issues a done operation cannot hang the device. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

oneandonlydean · 2026-10-29T16:18:49Z

@ZSWolt thanks, this is a good approach (the radeonsi-style turn past the GDS range, per-chunk reset for split dispatches, graphics left unordered, and a bounded wait so nothing can hang). One problem in the timeout path needs fixing before it can land:

A wave that gives up waiting still moves the turn. The wait loop leaves on ready || exhausted (SpirvMemoryInstructions.cpp around l.1333), and the wave-done release then stores launch + 1 (or 0 after the last wave) unconditionally with OpAtomicStore. So a wave that timed out also writes the turn. With two waves and the turn at 0: if wave 1 runs first and exhausts its wait, it adds, and as the last wave stores 0; wave 0 then adds and stores 1. The results come out reversed, and the dispatch ends with the turn at 1, so every wave of the next dispatch waits the full 2^20 reads and then runs unordered as well. One timeout turns into a slowdown and wrong order for everything after it, which is worse than main's plain unordered behaviour.

Please:

Only the wave that holds the turn passes it on. Release with a compare-exchange from launch to launch + 1 (or 0), so a wave that timed out leaves the turn alone. Running that one wave unordered after a timeout is acceptable (it matches main), as long as it can't affect the waves after it.
Make the reset belong to the dispatch, so the next dispatch starts at 0 even if the last wave finished early, timed out, or never reached the instruction (for example, reset it in the PM4 path before each dispatch that uses ordered count, rather than relying on the last wave's store).
Make timeouts visible: count them (a debug counter or log line is enough), so a passing test can't hide that the fallback ran.
Tests: add a case where a later wave reaches the instruction first (delay the first wave), multi-wave work-groups, and a second dispatch after one with a timeout, checking per-wave results and that the second dispatch is ordered. The current test (64 single-wave groups, 8 runs) only covers the in-order path.

The debt entry already says that in-order launch and progress are assumed and that Vulkan doesn't guarantee them, which is right; please add what happens on a timeout after the fix.

We'll run it on an RTX 4090 when it's updated (the agc_driver_gds_append_consume test plus a frame-time check on titles we have).

github-actions · 2026-10-29T16:38:27Z

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 6146	`core/libs/prx/libSceAgcDriver/Graphics/include/GuestBufferMemory.hpp` L607
PR 6858	`core/shader/recompiler/ShaderDiskCache.hpp` L16
PR 7395	`core/shader/recompiler/ShaderDiskCache.hpp` L16
PR 8939	`core/shader/recompiler/ShaderDiskCache.hpp` L16
PR 9457	`core/shader/recompiler/ShaderDiskCache.hpp` L16
PR 9508	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp` L2607-2608, L5861-5862 `core/libs/prx/libSceAgcDriver/Graphics/include/GuestBufferMemory.hpp` L381 `core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp` L2315 `core/shader/recompiler/RuntimeAbi.hpp` L20 `core/shader/recompiler/ShaderDiskCache.hpp` L16
PR 9633	`core/shader/recompiler/ShaderDiskCache.hpp` L16
PR 9749	`core/shader/recompiler/ShaderDiskCache.hpp` L16
PR 9790	`core/shader/recompiler/ShaderDiskCache.hpp` L16
PR 9969	`core/shader/recompiler/ShaderDiskCache.hpp` L16
PR 10049	`core/shader/recompiler/ShaderDiskCache.hpp` L16
PR 12840	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp` L5904 `core/libs/prx/libSceAgcDriver/Graphics/include/ShaderResources.hpp` L265
PR 14032	`core/shader/recompiler/ShaderDiskCache.hpp` L16
PR 14636	`core/shader/recompiler/ShaderDiskCache.hpp` L16
PR 14853	`core/shader/recompiler/ShaderDiskCache.hpp` L16
PR 14871	`core/shader/recompiler/ShaderDiskCache.hpp` L16
PR 15555	`core/shader/recompiler/ShaderDiskCache.hpp` L16
PR 16472	`core/shader/recompiler/ShaderDiskCache.hpp` L16

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 1395	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp`
PR 4040	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp`
PR 4472	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp`
PR 5021	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp` `core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp` `core/shader/recompiler/Recompiler.cpp` `core/shader/recompiler/SpirvBackend/include/SpirvBackend/SpirvEmitterState.hpp` `core/shader/recompiler/SpirvBackend/src/SpirvEmitter.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvMemory/SpirvMemoryInstructions.cpp` `docs/dev/debt/shader.md`
PR 5371	`core/shader/recompiler/RuntimeAbi.hpp` `docs/dev/debt/shader.md`
PR 5609	`core/shader/recompiler/Recompiler.cpp` `core/shader/recompiler/Recompiler.hpp` `core/shader/recompiler/ShaderDiskCache.cpp`
PR 5925	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp` `core/shader/recompiler/Recompiler.hpp` `core/shader/recompiler/ShaderDiskCache.cpp` `core/shader/recompiler/ShaderDiskCache.hpp`
PR 6186	`core/libs/prx/libSceAgcDriver/Graphics/include/GuestBufferMemory.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp`
PR 6188	`core/libs/prx/libSceAgcDriver/Execution/include/Pm4.hpp` `core/libs/prx/libSceAgcDriver/Execution/src/Pm4.cpp` `core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp` `core/shader/recompiler/Recompiler.hpp`
PR 6192	`core/libs/prx/libSceAgcDriver/Graphics/include/ShaderResources.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp`
PR 6194	`core/libs/prx/libSceAgcDriver/Graphics/include/ShaderResources.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp`
PR 6197	`core/libs/prx/libSceAgcDriver/Graphics/include/GuestBufferMemory.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp`
PR 6223	`core/libs/prx/libSceAgcDriver/Graphics/include/ShaderResources.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp`
PR 6235	`core/shader/recompiler/Recompiler.hpp` `core/shader/recompiler/SpirvBackend/src/SpirvMemory/SpirvEmitterHelpers.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvModuleEmitter.cpp` `docs/dev/debt/shader.md`
PR 6280	`core/shader/recompiler/RuntimeAbi.hpp` `core/shader/recompiler/ShaderDiskCache.cpp` `core/shader/recompiler/ShaderDiskCache.hpp` `docs/dev/debt/shader.md`
PR 6340	`core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp`
PR 6412	`core/libs/prx/libSceAgcDriver/Graphics/include/ShaderResources.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp` `core/shader/recompiler/Recompiler.cpp` `core/shader/recompiler/Recompiler.hpp` `core/shader/recompiler/ShaderDiskCache.cpp` `core/shader/recompiler/ShaderDiskCache.hpp` `docs/dev/debt/shader.md`
PR 6416	`core/shader/recompiler/ShaderDiskCache.hpp` `core/shader/recompiler/SpirvBackend/include/SpirvBackend/SpirvEmitterState.hpp` `core/shader/recompiler/SpirvBackend/src/SpirvEmitter.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvMemory/SpirvMemoryInstructions.cpp`
PR 6481	`core/libs/prx/libSceAgcDriver/Graphics/include/GuestBufferMemory.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp` `core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp`
PR 6488	`core/shader/recompiler/Recompiler.cpp` `core/shader/recompiler/ShaderDiskCache.cpp` `core/shader/recompiler/ShaderDiskCache.hpp` `core/shader/recompiler/SpirvBackend/src/SpirvEmitter.cpp` `docs/dev/debt/shader.md`
PR 6549	`core/libs/prx/libSceAgcDriver/Execution/include/Pm4.hpp` `core/libs/prx/libSceAgcDriver/Execution/src/Pm4.cpp`
PR 6804	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp` `core/libs/prx/libSceAgcDriver/Graphics/include/ShaderResources.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp`
PR 6918	`core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp`
PR 6924	`core/shader/recompiler/Recompiler.cpp` `core/shader/recompiler/ShaderDiskCache.cpp` `core/shader/recompiler/ShaderDiskCache.hpp` `docs/dev/debt/shader.md`
PR 7085	`core/shader/recompiler/RuntimeAbi.hpp` `core/shader/recompiler/ShaderDiskCache.cpp` `core/shader/recompiler/ShaderDiskCache.hpp` `docs/dev/debt/shader.md`
PR 7275	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp`
PR 7410	`core/shader/recompiler/IntermediateRepresentation/src/IrOpcode.cpp` `core/shader/recompiler/Recompiler.cpp` `core/shader/recompiler/Recompiler.hpp` `core/shader/recompiler/SpirvBackend/src/SpirvAnalysis.cpp` `docs/dev/debt/shader.md`
PR 7449	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp`
PR 7457	`core/libs/prx/libSceAgcDriver/Graphics/include/GuestBufferMemory.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp`
PR 7514	`core/libs/prx/libSceAgcDriver/Graphics/include/GuestBufferMemory.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp`
PR 7642	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp` `core/libs/prx/libSceAgcDriver/Graphics/include/GuestBufferMemory.hpp` `core/libs/prx/libSceAgcDriver/Graphics/include/ShaderResources.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp` `core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp` `core/shader/recompiler/Recompiler.cpp` `core/shader/recompiler/Recompiler.hpp` `core/shader/recompiler/RuntimeAbi.hpp` `core/shader/recompiler/ShaderDiskCache.cpp` `core/shader/recompiler/ShaderDiskCache.hpp` `core/shader/recompiler/SpirvBackend/include/SpirvBackend/SpirvEmitter.hpp` `core/shader/recompiler/SpirvBackend/include/SpirvBackend/SpirvEmitterState.hpp` `core/shader/recompiler/SpirvBackend/src/SpirvEmitter.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvMemory/SpirvBufferAccess.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvMemory/SpirvEmitterHelpers.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvMemory/SpirvMemoryInstructions.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvModuleEmitter.cpp`
PR 7643	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp`
PR 7684	`core/shader/recompiler/ShaderDiskCache.cpp` `core/shader/recompiler/ShaderDiskCache.hpp` `docs/dev/debt/shader.md`
PR 7689	`core/libs/prx/libSceAgcDriver/Graphics/include/ShaderResources.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp` `core/shader/recompiler/Recompiler.hpp` `core/shader/recompiler/ShaderDiskCache.cpp`
PR 8004	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp` `core/shader/recompiler/Recompiler.cpp` `core/shader/recompiler/Recompiler.hpp` `core/shader/recompiler/SpirvBackend/include/SpirvBackend/SpirvEmitter.hpp` `core/shader/recompiler/SpirvBackend/include/SpirvBackend/SpirvEmitterState.hpp` `core/shader/recompiler/SpirvBackend/src/SpirvEmitter.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvMemory/SpirvMemoryInstructions.cpp` `docs/dev/debt/shader.md`
PR 9012	`core/libs/prx/libSceAgcDriver/Graphics/include/GuestBufferMemory.hpp` `core/libs/prx/libSceAgcDriver/Graphics/include/ShaderResources.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp` `core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp`
PR 9289	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp` `core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp`
PR 9356	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp` `core/shader/recompiler/Recompiler.cpp` `core/shader/recompiler/Recompiler.hpp` `core/shader/recompiler/ShaderDiskCache.cpp` `core/shader/recompiler/ShaderDiskCache.hpp` `core/shader/recompiler/SpirvBackend/include/SpirvBackend/SpirvEmitter.hpp` `core/shader/recompiler/SpirvBackend/include/SpirvBackend/SpirvEmitterState.hpp` `core/shader/recompiler/SpirvBackend/src/SpirvEmitter.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvMemory/SpirvEmitterHelpers.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvModuleEmitter.cpp`
PR 9387	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp` `core/shader/recompiler/Recompiler.cpp` `core/shader/recompiler/Recompiler.hpp` `core/shader/recompiler/ShaderDiskCache.cpp` `core/shader/recompiler/ShaderDiskCache.hpp` `core/shader/recompiler/SpirvBackend/include/SpirvBackend/SpirvEmitter.hpp` `core/shader/recompiler/SpirvBackend/include/SpirvBackend/SpirvEmitterState.hpp` `core/shader/recompiler/SpirvBackend/src/SpirvEmitter.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvMemory/SpirvEmitterHelpers.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvModuleEmitter.cpp` `docs/dev/debt/shader.md`
PR 9592	`core/libs/prx/libSceAgcDriver/Graphics/include/ShaderResources.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp`
PR 9748	`core/shader/recompiler/Recompiler.cpp` `core/shader/recompiler/Recompiler.hpp` `core/shader/recompiler/ShaderDiskCache.hpp` `core/shader/recompiler/SpirvBackend/src/SpirvModuleEmitter.cpp` `docs/dev/debt/shader.md`
PR 9852	`core/libs/prx/libSceAgcDriver/Graphics/include/GuestBufferMemory.hpp` `core/libs/prx/libSceAgcDriver/Graphics/include/ShaderResources.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp` `core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp` `core/shader/recompiler/Recompiler.hpp` `core/shader/recompiler/ShaderDiskCache.cpp` `core/shader/recompiler/ShaderDiskCache.hpp`
PR 9884	`core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp`
PR 9893	`core/libs/prx/libSceAgcDriver/Execution/src/Pm4.cpp`
PR 10193	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp` `core/libs/prx/libSceAgcDriver/Graphics/include/GuestBufferMemory.hpp` `core/libs/prx/libSceAgcDriver/Graphics/include/ShaderResources.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp` `core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp` `core/shader/recompiler/Recompiler.cpp` `core/shader/recompiler/Recompiler.hpp` `core/shader/recompiler/RuntimeAbi.hpp` `core/shader/recompiler/ShaderDiskCache.cpp` `core/shader/recompiler/ShaderDiskCache.hpp` `core/shader/recompiler/SpirvBackend/include/SpirvBackend/SpirvEmitter.hpp` `core/shader/recompiler/SpirvBackend/include/SpirvBackend/SpirvEmitterState.hpp` `core/shader/recompiler/SpirvBackend/src/SpirvEmitter.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvMemory/SpirvBufferAccess.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvMemory/SpirvEmitterHelpers.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvMemory/SpirvMemoryInstructions.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvModuleEmitter.cpp`
PR 10233	`docs/dev/debt/shader.md`
PR 10249	`core/libs/prx/libSceAgcDriver/Execution/src/Pm4.cpp`
PR 10359	`core/shader/recompiler/Recompiler.cpp` `core/shader/recompiler/Recompiler.hpp`
PR 10442	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp` `core/libs/prx/libSceAgcDriver/Graphics/include/ShaderResources.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp` `core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp`
PR 10652	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp` `core/shader/recompiler/Recompiler.cpp` `core/shader/recompiler/Recompiler.hpp` `core/shader/recompiler/SpirvBackend/include/SpirvBackend/SpirvEmitter.hpp` `core/shader/recompiler/SpirvBackend/include/SpirvBackend/SpirvEmitterState.hpp` `core/shader/recompiler/SpirvBackend/src/SpirvEmitter.cpp` `docs/dev/debt/shader.md`
PR 10678	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp`
PR 10757	`core/shader/recompiler/Recompiler.cpp` `core/shader/recompiler/Recompiler.hpp`
PR 10894	`core/libs/prx/libSceAgcDriver/Graphics/include/ShaderResources.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp`
PR 11435	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp`
PR 11436	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp`
PR 11444	`core/libs/prx/libSceAgcDriver/Graphics/include/ShaderResources.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp`
PR 11483	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp` `core/shader/recompiler/Recompiler.cpp` `core/shader/recompiler/Recompiler.hpp` `core/shader/recompiler/SpirvBackend/include/SpirvBackend/SpirvEmitter.hpp` `core/shader/recompiler/SpirvBackend/include/SpirvBackend/SpirvEmitterState.hpp` `core/shader/recompiler/SpirvBackend/src/SpirvEmitter.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvModuleEmitter.cpp` `docs/dev/debt/shader.md`
PR 12157	`core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp`
PR 12324	`core/shader/recompiler/Recompiler.cpp`
PR 12619	`core/shader/recompiler/SpirvBackend/src/SpirvMemory/SpirvEmitterHelpers.cpp`
PR 12828	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp` `core/shader/recompiler/Recompiler.cpp` `core/shader/recompiler/Recompiler.hpp`
PR 12904	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp`
PR 13148	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp` `core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp`
PR 13486	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp` `core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp` `core/shader/recompiler/Recompiler.cpp` `core/shader/recompiler/ShaderDiskCache.cpp` `core/shader/recompiler/ShaderDiskCache.hpp` `core/shader/recompiler/SpirvBackend/src/SpirvModuleEmitter.cpp` `docs/dev/debt/shader.md`
PR 13796	`core/libs/prx/libSceAgcDriver/Graphics/include/ShaderResources.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp`
PR 13809	`core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp`
PR 13847	`docs/dev/debt/shader.md`
PR 13863	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp` `core/libs/prx/libSceAgcDriver/Graphics/include/ShaderResources.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp`
PR 13920	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp` `core/libs/prx/libSceAgcDriver/Graphics/include/ShaderResources.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp`
PR 13928	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp` `core/libs/prx/libSceAgcDriver/Graphics/include/GuestBufferMemory.hpp` `core/libs/prx/libSceAgcDriver/Graphics/include/ShaderResources.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp` `core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp` `core/shader/recompiler/Recompiler.cpp` `core/shader/recompiler/Recompiler.hpp` `core/shader/recompiler/RuntimeAbi.hpp` `core/shader/recompiler/ShaderDiskCache.cpp` `core/shader/recompiler/ShaderDiskCache.hpp` `core/shader/recompiler/SpirvBackend/include/SpirvBackend/SpirvEmitter.hpp` `core/shader/recompiler/SpirvBackend/include/SpirvBackend/SpirvEmitterState.hpp` `core/shader/recompiler/SpirvBackend/src/SpirvEmitter.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvMemory/SpirvBufferAccess.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvMemory/SpirvEmitterHelpers.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvMemory/SpirvMemoryInstructions.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvModuleEmitter.cpp`
PR 13929	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp` `core/libs/prx/libSceAgcDriver/Graphics/include/GuestBufferMemory.hpp` `core/libs/prx/libSceAgcDriver/Graphics/include/ShaderResources.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp` `core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp` `core/shader/recompiler/Recompiler.cpp` `core/shader/recompiler/Recompiler.hpp` `core/shader/recompiler/RuntimeAbi.hpp` `core/shader/recompiler/ShaderDiskCache.cpp` `core/shader/recompiler/ShaderDiskCache.hpp` `core/shader/recompiler/SpirvBackend/include/SpirvBackend/SpirvEmitter.hpp` `core/shader/recompiler/SpirvBackend/include/SpirvBackend/SpirvEmitterState.hpp` `core/shader/recompiler/SpirvBackend/src/SpirvEmitter.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvMemory/SpirvBufferAccess.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvMemory/SpirvEmitterHelpers.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvMemory/SpirvMemoryInstructions.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvModuleEmitter.cpp`
PR 14212	`docs/dev/debt/shader.md`
PR 14218	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp` `core/shader/recompiler/Recompiler.cpp` `core/shader/recompiler/Recompiler.hpp` `core/shader/recompiler/SpirvBackend/include/SpirvBackend/SpirvEmitter.hpp` `core/shader/recompiler/SpirvBackend/include/SpirvBackend/SpirvEmitterState.hpp` `core/shader/recompiler/SpirvBackend/src/SpirvEmitter.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvMemory/SpirvMemoryInstructions.cpp` `docs/dev/debt/shader.md`
PR 14344	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp`
PR 14377	`docs/dev/debt/shader.md`
PR 14944	`core/libs/prx/libSceAgcDriver/Graphics/include/ShaderResources.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp` `core/shader/recompiler/Recompiler.cpp` `core/shader/recompiler/Recompiler.hpp` `core/shader/recompiler/ShaderDiskCache.cpp` `core/shader/recompiler/ShaderDiskCache.hpp`
PR 15059	`core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp`
PR 15136	`core/libs/prx/libSceAgcDriver/Execution/include/Pm4.hpp` `core/libs/prx/libSceAgcDriver/Execution/src/Pm4.cpp` `core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp`
PR 15169	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp`
PR 15296	`core/libs/prx/libSceAgcDriver/Graphics/include/GuestBufferMemory.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp`
PR 15302	`core/libs/prx/libSceAgcDriver/Graphics/include/GuestBufferMemory.hpp` `core/libs/prx/libSceAgcDriver/Graphics/include/ShaderResources.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp` `core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp`
PR 15529	`core/libs/prx/libSceAgcDriver/Graphics/include/ShaderResources.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp`
PR 15548	`core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp` `core/shader/recompiler/Recompiler.cpp` `core/shader/recompiler/Recompiler.hpp` `core/shader/recompiler/ShaderDiskCache.cpp` `core/shader/recompiler/ShaderDiskCache.hpp` `core/shader/recompiler/SpirvBackend/include/SpirvBackend/SpirvEmitter.hpp` `core/shader/recompiler/SpirvBackend/include/SpirvBackend/SpirvEmitterState.hpp` `core/shader/recompiler/SpirvBackend/src/SpirvEmitter.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvMemory/SpirvEmitterHelpers.cpp` `core/shader/recompiler/SpirvBackend/src/SpirvModuleEmitter.cpp`
PR 15570	`core/libs/prx/libSceAgcDriver/Graphics/include/ShaderResources.hpp` `core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp` `core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp`

A wave now passes the turn on with a compare-exchange from its launch index, so a wave that gave up waiting adds unordered and leaves the turn alone. The driver zeroes the turn in the buffer each dispatch binds, before the dispatch and each chunk of a split one, instead of relying on the last wave. Waves that give up are counted in a dword after the turn. The wait, the add and the release now run inside the loop, one lane per guest wave, so guest waves sharing a host subgroup no longer wait on each other until the timeout. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

ZSWolt · 2026-10-29T16:53:53Z

@oneandonlydean thanks, all five points are addressed in the new commit on top (no force-push).

Only the wave that holds the turn passes it on. The wave-done release is now OpAtomicCompareExchange from launch to launch + 1 (or 0 after the last wave). A wave that timed out adds unordered, as on main, and leaves the turn alone.
The reset belongs to the dispatch. recordDispatch zeroes the turn in the buffer the dispatch binds, before every dispatch that binds the GDS and before each chunk of a split one. It uses vkCmdFillBuffer between barriers, or a CPU store when that buffer is a CPU-made copy. The last-wave reset is gone; the release from the last wave only wraps to 0.
Timeouts are counted. A wave that gives up adds 1 to a dword after the turn (Pm4::GdsOrderedTimeoutsOffset, binding now GdsBytes + 8). The test checks the count is 0 in every ordered case.
Tests (GdsAppendConsume.cpp), each checking per-wave results, the counter and the timeout count:
- 64 single-wave groups with group 0 delayed by a scalar loop, so later waves reach the instruction first.
- 16 groups of 4 waves, group 0 delayed.
- 4 groups where group 1 skips ds_ordered_count. Group 0 gets the start value, groups 2–3 each time out and get one of the rest, and exactly 2 timeouts are counted. Before the fix, a timed-out wave's store let the next one through.
- A normal 64-group dispatch right after that one. The turn was left at 1, so it is ordered only because of the per-dispatch reset; without it, group 0 got the last value.
Debt entry updated with the reset, the compare-exchange, and what a timeout does.

A second bug the multi-wave test found. When the host subgroup is wider than the guest wave (wave32 code in a 64-lane subgroup on RADV), two guest waves shared one wait loop. The first could not leave the loop to release until the second stopped waiting, so the second always timed out (29 timeouts in that test). The wait, the add and the release now run inside the loop, done by one lane per guest wave, so guest waves in one host subgroup no longer wait on each other.

Checks (RX 9060 XT, RADV):

agc_driver_gds_append_consume passes.
ctest -R '^agc_|^shader_|recompiler': only the two tests that also fail on main fail.
Validation layer: 0 VUIDs.
check_conventions passes.

A run on the RTX 4090 would be welcome. The delayed and multi-wave cases are the ones that depend on the host's subgroup layout.

🤖 Addressed by Claude Code

CnRJay

The ordering fix is sound, and oneandonlydean's timeout points are addressed. One cost in recordDispatch is not covered yet and applies to every dispatch that binds the GDS, not just ordered ones.

The reset is gated on bindsGds, not on orderedAppend. bindsGds is true for any shader with a DescriptorRole::Gds binding, which includes plain append/consume and any other GDS user. Those shaders never read the turn dword, but they now get:

vkCmdPipelineBarrier ALL_COMMANDS → TRANSFER
vkCmdFillBuffer of 4 bytes
vkCmdPipelineBarrier TRANSFER → COMPUTE_SHADER

ALL_COMMANDS as the source stage drains everything recorded before it: preceding draws, other dispatches and copies. Each such dispatch also becomes a hard sync point, so it can't overlap the work before it. The same is added before each chunk of a split dispatch. A title that issues many small GDS dispatches per frame (particle and culling passes are the typical case) pays a GPU drain for each one.

Carry the orderedAppend requirement into the compiled program, or the binding info, and reset only for programs that contain DataOrderedCount or DataOrderedSwap in a compute stage. Shaders that don't order then get exactly what main gives them.

The first barrier is wider than it needs to be. The only prior writer of the turn dword is an earlier ordered dispatch. A COMPUTE_SHADER → TRANSFER barrier with SHADER_WRITE|SHADER_READ → TRANSFER_WRITE covers that. ALL_COMMANDS / TRANSFER_WRITE also orders graphics work against the fill, which nothing needs.

Alternative that avoids the fill and both barriers. Have the shader own the reset. The wave holding the last launch index already compare-exchanges the turn to 0. With a launch-epoch dword that the host bumps once per dispatch through push constants, the compare-exchange targets epoch * stride + launch and no dispatch ever needs a zeroed turn. A timed-out dispatch then can't poison the next one either, and the fill and barriers go away.

CPU-copy path (zeroGuestBytes returning nullopt). The memset runs at record time, while the GPU may still be executing a previously submitted dispatch that uses the same CPU-made GDS copy. Two dispatches recorded back to back into one command buffer also both zero the turn before either runs. Please confirm that case can't happen, or say what guarantees it can't.

Wait loop. Each waiting wave does up to 2^20 device-scope acquire loads with no backoff. If one wave never issues the done operation (a divergent path where a whole guest wave skips it), every later wave in that dispatch spends the full budget. The test counts the timeouts (2 here), but nothing records how long a timeout costs on hardware that doesn't schedule groups strictly in order. On a 4090, include the time of the skipped-group case in the numbers; it matters for Windows TDR.

Measurement. The title numbers are the corrupted-frame counts and the present counts (851–857 against about 790). Please add tools/trace_digest.py summaries of main and of this change from the same machine and scene, with the frame time and the gpu_busy_ms_per_present of the GoW fireplace scene. Presents going up while a drain is added before every GDS dispatch is surprising, and the digest will show which phase moved.

The compiled program now records whether it orders GDS appends, and the driver resets the turn only for those, behind a compute-to-transfer barrier instead of draining all prior work. A GDS copy the CPU made is no longer zeroed at record time. The first wave that gives up waiting marks the turn abandoned, so the rest of the dispatch runs unordered at once instead of each later wave waiting a full budget. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

TechLevelZero · 2026-10-29T19:22:45Z

✅ NVIDIA GPU test: passed

PR head af3c0c1 merged onto main bb04a4f · tested 2026-10-29 19:22 UTC · 26 min


GPU	NVIDIA RTX 2000 Ada Generation, driver 595.99.02, Vulkan 1.4.329
Host	Ubuntu 26.04.1 LTS, kernel 7.0.0-38-generic, container ubuntu:24.04
Logs	build and test logs for this run
Build	✅ passed (Release, `-DANYPS5_ENABLE_SPIRV_TOOLS=ON`)
Tests	✅ 2,234 passed · 6 failed (0 new, 5 also fail on main, 1 unstable on main) · 40 skipped (`ANYPS5_REQUIRE_VULKAN=1`)
Test setup	CPU-only tests without a GPU; GPU tests (and any CPU-only test that failed or skipped without one) once on the GPU with `VK_LAYER_KHRONOS_validation` and `ANYPS5_NO_SHADER_CACHE=1`; main ran the same way
Test selection	2,280 of 2,918 tests run: the ones this change can reach, chosen by `tools/select_tests.py` from main's build graph
Vulkan validation	✅ no new validation IDs vs main (857 tests run on the GPU)
Slow tests (>30 s, GPU tests timed with validation layers)	⚠️ `agc_driver_bda_device` 65.2 s

Unstable on main (1, these also failed in recent main runs on this machine; not counted against this PR)

agc_driver_submit_progress: failed in 4 of the last 6 main runs

Failing on main too (5, not counted against this PR)

agc_compute_pipeline_inspection
agc_submit_progress
agc_submit_progress_no_timeline
guest_mmap_executable
guest_socket_options_multicast

Validation IDs already present on main (4 IDs in 397 tests)

VUID-VkDeviceCreateInfo-pNext-pNext (397 tests): agc_compute_pipeline_replay, agc_driver, agc_driver_address_passes_draw_transitions, agc_driver_address_passes_targets_split, agc_driver_alias_copy_label, agc_driver_array_layer_target, agc_driver_array_layer_target_cpu, agc_driver_bda_anchor_reads, and 389 more
VUID-vkCmdCopyImage-srcImage-01548 (11 tests): agc_driver_depth_clear_hle, agc_driver_depth_plane_buffer_memory, agc_driver_depth_plane_cpu_write_after_memo, agc_driver_depth_plane_cpu_write_after_page_unwatch, agc_driver_depth_plane_cpu_write_after_partial_unwatch, agc_driver_depth_plane_partial_store, agc_driver_depth_plane_partial_store_unimported, agc_driver_depth_storage_dispatch, and 3 more
VUID-vkCmdCopyImage-srcImage-01551 (11 tests): agc_driver_depth_clear_hle, agc_driver_depth_plane_buffer_memory, agc_driver_depth_plane_cpu_write_after_memo, agc_driver_depth_plane_cpu_write_after_page_unwatch, agc_driver_depth_plane_cpu_write_after_partial_unwatch, agc_driver_depth_plane_partial_store, agc_driver_depth_plane_partial_store_unimported, agc_driver_depth_storage_dispatch, and 3 more
VUID-VkPipelineColorBlendStateCreateInfo-pAttachments-00605 (2 tests): agc_driver_prepared_exports, agc_driver_srgb8_color_target

Skipped (40)

agc_driver_bda_reads_shadowed_results: skipped, the device reports watched host memory written when it imports or reads it
agc_driver_bda_reads_shadowed_results_per_build: skipped, the device reports watched host memory written when it imports or reads it
agc_driver_msaa_dcc_compared_import_2: skipped, the device reports host imports that keep write protection, so the write-watch probe compares none of them
agc_driver_msaa_dcc_compared_import_4: skipped, the device reports host imports that keep write protection, so the write-watch probe compares none of them
agc_driver_msaa_dcc_compared_import_8: skipped, the device reports host imports that keep write protection, so the write-watch probe compares none of them
agc_driver_sample_order_grid_reference: skipped, the device does not keep depth drawn at custom sample locations at those locations, even after a transition that names them: the composite of uniform renders holds the guest's depth in 0 of 3005 sloped samples, so no path can be compared with it bit for bit
agc_driver_sparse_dma_buf_import: skipped, host imports are unavailable: /dev/udmabuf is not accessible
agc_driver_stencil_ref_export: skipped, the device lacks VK_EXT_shader_stencil_export
agc_host_import_sparse: skipped, host imports are unavailable: /dev/udmabuf is not accessible
agc_import_settle_initial_dcc: skipped, the device has no deferred settle of host imports: imports under 1 MiB wait for one only on Windows
agc_unit_shadow_completion_store: skipped, the device reports watched host memory written when it imports or reads it
guest_http_uri_locale: 
guest_net_ctl: skipped, the host has no IPv4 address
guest_video_out_color_space: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_hdr: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_high_frame_rate: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_output_mode: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_resolution_status: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_wqhd_detection: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
video_out_flip_controls: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_cpuflip: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_cpuflip_in_order: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_eop: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_eop_aftergpu: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_eop_main: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_fresh: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_labels: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not

Reproduce

git fetch origin af3c0c1e51c4bf63eaa69c5ce4437a9fedca0fd3 bb04a4f1ea0175e9e4f149f0359a764d62ff3296
git checkout bb04a4f1ea0175e9e4f149f0359a764d62ff3296 && git merge --no-ff af3c0c1e51c4bf63eaa69c5ce4437a9fedca0fd3
git submodule update --init --recursive
cmake -S . -B build -G Ninja -DCMAKE_BUILD_TYPE=Release -DBUILD_TESTING=ON -DANYPS5_ENABLE_SPIRV_TOOLS=ON
cmake --build build && cmake --build build --target libs
VK_DRIVER_FILES=/nonexistent/icd.json ctest --test-dir build -E '^(agc|shader|recompiler|videoout|video_out|guest_video_out|vulkan)|gpu' --timeout 120 --parallel 32 --output-on-failure
ANYPS5_REQUIRE_VULKAN=1 ANYPS5_NO_SHADER_CACHE=1 VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation \
  ctest --test-dir build -R '^(agc|shader|recompiler|videoout|video_out|guest_video_out|vulkan)|gpu' --timeout 120 --parallel 4 --output-on-failure
# failures are then re-run alone: --timeout 360 --parallel 2; new validation IDs: the test alone, twice, --parallel 1

_{Automated build and test on a contributor's NVIDIA machine; it does not replace upstream CI. A new comment is posted for each new push. Full logs.}

ZSWolt · 2026-10-29T20:03:56Z

@CnRJay thanks. All points are addressed in a new commit on top (no force-push).

Reset only for ordering programs. The compiled program now carries orderedAppend: it goes from the SPIR-V emitter's requirements into CompiledShaderArtifact, the disk cache (format 50) and RecompileResult. recordDispatch resets the turn only when it is set. Plain append/consume and every other GDS user get exactly what main records.

Narrower barrier. The reset is now:

COMPUTE_SHADER → TRANSFER (SHADER_READ|SHADER_WRITE → TRANSFER_WRITE);
the 4-byte fill;
TRANSFER → COMPUTE_SHADER.

Epoch alternative. I kept the reset. An epoch needs a push word in every ordering program, bumped per dispatch, including dispatches replayed from recipes. A 32-bit turn of epoch * stride + launch would also wrap for large dispatches. With the gate, the cost is two narrow barriers and a 4-byte fill, and only for ordering dispatches. I can switch if you prefer the epoch.

CPU-copy path. You're right, the record-time store was not safe to rely on, and it is removed. The reset is recorded only when the turn lives in an import or a GPU copy (GuestBufferMemory::FillTarget). A GDS copy the CPU made is not reset: its dispatch starts from the turn the previous one left, which is 0 after a dispatch whose waves all took their turns. The debt entry says so. In the test the reset takes effect, so here the GDS is bound as an import or a GPU copy.

Wait loop cost. On an RX 9060 XT, 2^20 reads take about 130 ms. Your concern was right. I timed a dispatch where group 1 never takes its turn (a local timing build):

Groups	Before this commit	With this commit
4	~130 ms	~140 ms
4096	1357 ms	~350 ms
65536	device lost after ~2 s (`VK_ERROR_DEVICE_LOST`)	~350 ms

Before this commit, waves that launched after earlier ones had given up each waited a full budget again. Now the first wave that gives up marks the turn abandoned (0xffffffff). Every later wave of that dispatch sees the mark and runs unordered at once, as on main, and is counted. The next ordering dispatch's reset clears the mark. The holder's compare-exchange cannot overwrite the mark, because it compares against its own launch index.

Measurement. A correction first: the earlier numbers in the description (851–857 presents against about 790, "on main 0cd07dc") were wrong. They came from Oct 23, on my local integration build, against a baseline from another run; they were not a same-build A/B on main. I've corrected the description.

Plain main cannot reach this scene: GoW Ragnarök stops at startup on sceHttp2SetResolveRetry not implemented. So the A/B below runs on my local integration build (main plus local branches, d3b93c0ed8), with and without this PR:

built in one worktree, one variant after the other;
each run uses its own libs/ through --rpath;
the build is confirmed per run by the shader cache format (49 without, 50 with);
title-screen fireplace scene, warm caches, 3840x2160, otherwise default settings;
APS5_TRACE_SECONDS=200:60, order without, with, with, without.

tools/trace_digest.py summary lines:

	presents	frame p50 / p90 / p99 ms	`gpu_busy_ms_per_present`
without, run 1	203	287.0 / 343.4 / 416.7	148.5
with, run 1	217	275.9 / 295.0 / 338.0	128.9
with, run 2	214	277.9 / 300.1 / 374.6	128.6
without, run 2	198	292.2 / 360.3 / 493.2	151.9

Where the time moved, per frame (top_phases_ms_per_frame):

Driver.Packet.draw is 173–179 ms without and 158–161 ms with.
Driver.Packet.memory_wait is 204–209 ms without and 184–186 ms with.
Driver.Packet.dispatch is 166–169 ms in all runs, so the reset adds nothing visible.

GPU busy per present drops by about 20 ms. My guess, not verified, is that the drop comes from the corrupted particle records: without the fix they are drawn as long streaks across the screen, which is far more raster work than correct particles. The traces dropped events after about 10 s on the busiest worker (trace_dropped: true); the summary values above come from the per-present totals.

Checks on this commit:

agc_driver_gds_append_consume passes.
ctest -R '^agc_|^shader_|recompiler': only the two tests that also fail on main fail.
Validation layer: 0 VUIDs.
check_conventions passes.

TechLevelZero's NVIDIA run (RTX 2000 Ada) passed on the previous head, af3c0c1.

🤖 Addressed by Claude Code

TechLevelZero · 2026-10-29T21:20:42Z

✅ NVIDIA GPU test: passed

PR head 3cea7d4 merged onto main bb04a4f · tested 2026-10-30 01:10 UTC · 29 min


GPU	NVIDIA RTX 2000 Ada Generation, driver 595.99.02, Vulkan 1.4.329
Host	Ubuntu 26.04.1 LTS, kernel 7.0.0-38-generic, container ubuntu:24.04
Logs	build and test logs for this run
Build	✅ passed (Release, `-DANYPS5_ENABLE_SPIRV_TOOLS=ON`)
Tests	✅ 2,242 passed · 6 failed (0 new, 5 also fail on main, 1 unstable on main) · 40 skipped (`ANYPS5_REQUIRE_VULKAN=1`)
Test setup	CPU-only tests without a GPU; GPU tests (and any CPU-only test that failed or skipped without one) once on the GPU with `VK_LAYER_KHRONOS_validation` and `ANYPS5_NO_SHADER_CACHE=1`; main ran the same way
Test selection	2,288 of 2,918 tests run: the ones this change can reach, chosen by `tools/select_tests.py` from main's build graph
Vulkan validation	✅ no new validation IDs vs main (861 tests run on the GPU)
Slow tests (>30 s, GPU tests timed with validation layers)	⚠️ `agc_driver_bda_device` 68.7 s

Unstable on main (1, these also failed in recent main runs on this machine; not counted against this PR)

agc_driver_depth_clear_hle: failed in 1 of the last 7 main runs

Failing on main too (5, not counted against this PR)

agc_compute_pipeline_inspection
agc_submit_progress
agc_submit_progress_no_timeline
guest_mmap_executable
guest_socket_options_multicast

Validation IDs already present on main (5 IDs in 400 tests)

VUID-VkDeviceCreateInfo-pNext-pNext (400 tests): agc_compute_pipeline_replay, agc_driver, agc_driver_address_passes_draw_transitions, agc_driver_address_passes_targets_split, agc_driver_alias_copy_label, agc_driver_array_layer_target, agc_driver_array_layer_target_cpu, agc_driver_bda_anchor_reads, and 392 more
VUID-vkCmdCopyImage-srcImage-01548 (11 tests): agc_driver_depth_clear_hle, agc_driver_depth_plane_buffer_memory, agc_driver_depth_plane_cpu_write_after_memo, agc_driver_depth_plane_cpu_write_after_page_unwatch, agc_driver_depth_plane_cpu_write_after_partial_unwatch, agc_driver_depth_plane_partial_store, agc_driver_depth_plane_partial_store_unimported, agc_driver_depth_storage_dispatch, and 3 more
VUID-vkCmdCopyImage-srcImage-01551 (11 tests): agc_driver_depth_clear_hle, agc_driver_depth_plane_buffer_memory, agc_driver_depth_plane_cpu_write_after_memo, agc_driver_depth_plane_cpu_write_after_page_unwatch, agc_driver_depth_plane_cpu_write_after_partial_unwatch, agc_driver_depth_plane_partial_store, agc_driver_depth_plane_partial_store_unimported, agc_driver_depth_storage_dispatch, and 3 more
VUID-VkPipelineColorBlendStateCreateInfo-pAttachments-00605 (2 tests): agc_driver_prepared_exports, agc_driver_srgb8_color_target
VUID-vkCmdCopyBuffer-commonparent (1 test): agc_driver_depth_clear_hle

Skipped (40)

agc_driver_bda_reads_shadowed_results: skipped, the device reports watched host memory written when it imports or reads it
agc_driver_bda_reads_shadowed_results_per_build: skipped, the device reports watched host memory written when it imports or reads it
agc_driver_msaa_dcc_compared_import_2: skipped, the device reports host imports that keep write protection, so the write-watch probe compares none of them
agc_driver_msaa_dcc_compared_import_4: skipped, the device reports host imports that keep write protection, so the write-watch probe compares none of them
agc_driver_msaa_dcc_compared_import_8: skipped, the device reports host imports that keep write protection, so the write-watch probe compares none of them
agc_driver_sample_order_grid_reference: skipped, the device does not keep depth drawn at custom sample locations at those locations, even after a transition that names them: the composite of uniform renders holds the guest's depth in 0 of 3005 sloped samples, so no path can be compared with it bit for bit
agc_driver_sparse_dma_buf_import: skipped, host imports are unavailable: /dev/udmabuf is not accessible
agc_driver_stencil_ref_export: skipped, the device lacks VK_EXT_shader_stencil_export
agc_host_import_sparse: skipped, host imports are unavailable: /dev/udmabuf is not accessible
agc_import_settle_initial_dcc: skipped, the device has no deferred settle of host imports: imports under 1 MiB wait for one only on Windows
agc_unit_shadow_completion_store: skipped, the device reports watched host memory written when it imports or reads it
guest_http_uri_locale: 
guest_net_ctl: skipped, the host has no IPv4 address
guest_video_out_color_space: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_hdr: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_high_frame_rate: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_output_mode: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_resolution_status: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_wqhd_detection: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
video_out_flip_controls: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_cpuflip: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_cpuflip_in_order: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_eop: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_eop_aftergpu: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_eop_main: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_fresh: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_labels: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not

Reproduce

git fetch origin 3cea7d40118adb552c0df9666555fb738b88ce04 bb04a4f1ea0175e9e4f149f0359a764d62ff3296
git checkout bb04a4f1ea0175e9e4f149f0359a764d62ff3296 && git merge --no-ff 3cea7d40118adb552c0df9666555fb738b88ce04
git submodule update --init --recursive
cmake -S . -B build -G Ninja -DCMAKE_BUILD_TYPE=Release -DBUILD_TESTING=ON -DANYPS5_ENABLE_SPIRV_TOOLS=ON
cmake --build build && cmake --build build --target libs
VK_DRIVER_FILES=/nonexistent/icd.json ctest --test-dir build -E '^(agc|shader|recompiler|videoout|video_out|guest_video_out|vulkan)|gpu' --timeout 120 --parallel 32 --output-on-failure
ANYPS5_REQUIRE_VULKAN=1 ANYPS5_NO_SHADER_CACHE=1 VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation \
  ctest --test-dir build -R '^(agc|shader|recompiler|videoout|video_out|guest_video_out|vulkan)|gpu' --timeout 120 --parallel 4 --output-on-failure
# failures are then re-run alone: --timeout 360 --parallel 2; new validation IDs: the test alone, twice, --parallel 1

_{Automated build and test on a contributor's NVIDIA machine; it does not replace upstream CI. A new comment is posted for each new push. Full logs.}

…d-append Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

bunbeans2-arch · 2026-10-30T10:42:00Z

Windows data point to go with the RADV RX 9060 XT run: AMD Radeon RX 9070 XT (RDNA 4) on AMD's Windows driver 32.0.32015.7029, Windows 11 Pro 22631, MSYS2 UCRT64 GCC 16.1, Release, BUILD_TESTING=ON, with the Khronos validation layer loaded.

Tree: main 409fe64199 with this PR merged on top (it merges cleanly). Every test executable was rebuilt for the tree before running.

	`main`	with this PR
`agc_driver_gds_append_consume`	passes 3 of 3	passes 3 of 3 (the PR's version of the test)
`agc_driver_graphics`, `agc_driver_gcn_detile`	pass 3 of 3	pass 3 of 3
`VUID-` lines	0	0

I could not run the PR's test against main's driver code: it uses Pm4::GdsOrderedTimeoutsOffset, which only this PR adds. So this shows the PR's test passes on AMD's Windows driver (it did not skip), not that it catches the old behaviour on this card.

AI-assisted: yes (Claude Code)

TechLevelZero · 2026-10-30T10:46:53Z

✅ NVIDIA GPU test: passed

PR head 7bdeb14 merged onto main 73f4f7e · tested 2026-10-31 18:27 UTC · 20 min


GPU	NVIDIA RTX 2000 Ada Generation, driver 595.99.02, Vulkan 1.4.329
Host	Ubuntu 26.04.1 LTS, kernel 7.0.0-38-generic, container ubuntu:24.04
Logs	build and test logs for this run
Build	✅ passed (Release, `-DANYPS5_ENABLE_SPIRV_TOOLS=ON`)
Tests	✅ 2,347 passed · 5 failed (0 new, 4 also fail on main, 1 unstable on main) · 48 skipped (`ANYPS5_REQUIRE_VULKAN=1`)
Test setup	CPU-only tests without a GPU; GPU tests (and any CPU-only test that failed or skipped without one) once on the GPU with `VK_LAYER_KHRONOS_validation` and `ANYPS5_NO_SHADER_CACHE=1`; main ran the same way
Test selection	2,400 of 3,072 tests run: the ones this change can reach, chosen by `tools/select_tests.py` from main's build graph
Vulkan validation	✅ no new validation IDs vs main (884 tests run on the GPU)
Slow tests (>30 s, GPU tests timed with validation layers)	⚠️ `agc_driver_bda_device` 39.8 s, ⚠️ `agc_driver_graphics_maps_fallback` 34.7 s, ⚠️ `agc_driver_graphics` 31.0 s

Unstable on main (1, these also failed in recent main runs on this machine; not counted against this PR)

agc_driver_submit_progress: failed in 7 of the last 10 main runs

Failing on main too (4, not counted against this PR)

agc_submit_progress
agc_submit_progress_no_timeline
guest_mmap_executable
guest_socket_options_multicast

Passed on retry (1, failed in the parallel run, passed when re-run with a 360 s timeout; not counted against this PR)

agc_driver_job_manager_compute_capacity: Failed in the parallel run; also passed only on retry in 192 other runs in the last 7 days

Validation IDs already present on main (3 IDs in 390 tests)

VUID-VkDeviceCreateInfo-pNext-pNext (390 tests): agc_compute_pipeline_replay, agc_driver, agc_driver_address_passes, agc_driver_address_passes_draw_transitions, agc_driver_address_passes_targets_split, agc_driver_alias_copy_label, agc_driver_alpha_to_coverage, agc_driver_array_layer_target, and 382 more
VUID-vkCmdCopyImage-srcImage-01548 (12 tests): agc_driver_depth_clear_hle, agc_driver_depth_plane_buffer_memory, agc_driver_depth_plane_cpu_write_after_memo, agc_driver_depth_plane_cpu_write_after_page_unwatch, agc_driver_depth_plane_cpu_write_after_partial_unwatch, agc_driver_depth_plane_partial_store, agc_driver_depth_plane_partial_store_unimported, agc_driver_depth_plane_sampling, and 4 more
VUID-vkCmdCopyImage-srcImage-01551 (12 tests): agc_driver_depth_clear_hle, agc_driver_depth_plane_buffer_memory, agc_driver_depth_plane_cpu_write_after_memo, agc_driver_depth_plane_cpu_write_after_page_unwatch, agc_driver_depth_plane_cpu_write_after_partial_unwatch, agc_driver_depth_plane_partial_store, agc_driver_depth_plane_partial_store_unimported, agc_driver_depth_plane_sampling, and 4 more

Skipped (48)

agc_driver_bda_reads_shadowed_results: skipped, the device reports watched host memory written when it imports or reads it
agc_driver_bda_reads_shadowed_results_per_build: skipped, the device reports watched host memory written when it imports or reads it
agc_driver_msaa_dcc_compared_import_2: skipped, the device reports host imports that keep write protection, so the write-watch probe compares none of them
agc_driver_msaa_dcc_compared_import_4: skipped, the device reports host imports that keep write protection, so the write-watch probe compares none of them
agc_driver_msaa_dcc_compared_import_8: skipped, the device reports host imports that keep write protection, so the write-watch probe compares none of them
agc_driver_sample_order_grid_reference: skipped, the device does not keep depth drawn at custom sample locations at those locations, even after a transition that names them: the composite of uniform renders holds the guest's depth in 0 of 3005 sloped samples, so no path can be compared with it bit for bit
agc_driver_vertex_elected_atomic_max: skipped, the device runs vertex programs in subgroups, so they keep their cross-lane reductions
agc_host_import_sparse: skipped, host imports are unavailable: /dev/udmabuf is not accessible
agc_import_settle_initial_dcc: skipped, the device has no deferred settle of host imports: imports under 1 MiB wait for one only on Windows
agc_unit_shadow_completion_store: skipped, the device reports watched host memory written when it imports or reads it
guest_http_uri_locale: 
guest_net_ctl: skipped, the host has no IPv4 address
guest_video_out_color_space: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_hdr: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_high_frame_rate: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_output_mode: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_resolution_status: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_wqhd_detection: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
video_out_flip_aftergpu: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_backtoback: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_controls: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_cpuflip: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_cpuflip_in_order: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_dcc: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_deferred: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_eop: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or plat

Reproduce

git fetch origin 7bdeb149a8ca09d481971915c14b06633bb27956 73f4f7e06e4f210a8c6570e335fd654dc0a86134
git checkout 73f4f7e06e4f210a8c6570e335fd654dc0a86134 && git merge --no-ff 7bdeb149a8ca09d481971915c14b06633bb27956
git submodule update --init --recursive
cmake -S . -B build -G Ninja -DCMAKE_BUILD_TYPE=Release -DBUILD_TESTING=ON -DANYPS5_ENABLE_SPIRV_TOOLS=ON
cmake --build build && cmake --build build --target libs
VK_DRIVER_FILES=/nonexistent/icd.json ctest --test-dir build -E '^(agc|shader|recompiler|videoout|video_out|guest_video_out|vulkan)|gpu' --timeout 120 --parallel 32 --output-on-failure
ANYPS5_REQUIRE_VULKAN=1 ANYPS5_NO_SHADER_CACHE=1 VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation \
  ctest --test-dir build -R '^(agc|shader|recompiler|videoout|video_out|guest_video_out|vulkan)|gpu' --timeout 120 --parallel 4 --output-on-failure
# failures are then re-run alone: --timeout 360 --parallel 2; new validation IDs: the test alone, twice, --parallel 1

_{Automated build and test on a contributor's NVIDIA machine; it does not replace upstream CI. A new comment is posted for each new push. Full logs.}

JJenkx · 2026-10-30T15:53:18Z

Windows GPU run of head 7bdeb149a8, merged onto main 1c6b316cf7 (clean merge, local only).

Windows 11 Enterprise 10.0.22621, Ryzen 9 5950X, Radeon RX 6700 XT with AMD proprietary driver 26.10.1 (Vulkan 1.4.315), WinLibs GCC 15.2.0 UCRT r7, Ninja, Release, BUILD_TESTING=ON. Khronos validation layer 1.4.309 requested for every test (VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation).

agc_driver_gds_append_consume, built from the test files this PR changes: pass, 3 of 3 runs each (--repeat until-fail:3), no failures, no skips.
They create Vulkan instances on the RX 6700 XT: the validation layer was inserted 3 times across the runs and reported no VUIDs.
No negative control: this PR's test files do not build against main's sources, because they use what the PR adds.

AI-assisted: yes (Claude Code).

jacknieland · 2026-10-31T14:53:48Z

NVIDIA on Windows, for head 7bdeb149a8 merged into main 73f4f7e06e (clean merge). The NVIDIA runs so far are RTX 2000 Ada on Linux, and the Windows runs are AMD.

GPU: NVIDIA GeForce GTX 1650 (Turing, 0x1f91), driver 566.41, Vulkan 1.3.289, subgroup size 32. The driver logs Selected GPU name=NVIDIA GeForce GTX 1650. Host: Windows 11 Pro 26200, i7-9750H, WinLibs GCC 15.2.0 (x86_64-ucrt-posix-seh, r7), Release, BUILD_TESTING=ON. Vulkan SDK 1.4.363.0 validation layer.

ctest -R agc_driver_gds_append_consume -V, on main and on the merge, with ANYPS5_GPU=NVIDIA, ANYPS5_REQUIRE_VULKAN=1, ANYPS5_NO_SHADER_CACHE=1 and VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation:

main: passes, with 0 VUID- lines.
This PR: passes, and passes 3/3 with --repeat until-fail:3, with 0 VUID- lines. That includes the in-order group case, which requires that no wave gave up waiting for its turn, and the timeout case.
Neither log contains was not found but was requested by env var VK_INSTANCE_LAYERS.
No red/green: this PR's GdsAppendConsume.cpp uses Pm4::GdsOrderedTimeoutsOffset, so it does not build against main's driver.

Only this test target was built, not the full suite, and no title was run. AI-assisted.

CnRJay

Thanks for the gate, the narrower barrier, the timing table and the correction of the present counts. One consequence of the CPU-copy change is not covered in your reply.

An abandoned mark is never cleared when the GDS binding is a CPU-made copy. The first wave that gives up stores 0xffffffff in the turn, and "the next ordering dispatch's reset clears the mark". For a CPU-made copy you no longer record a reset (FillTarget returns nothing), and you rely on the previous dispatch leaving 0. After a timeout it leaves 0xffffffff. The holder's compare-exchange cannot replace the mark, so every later ordering dispatch on that copy sees the mark at its first wait and runs unordered, with every wave counted as a timeout. That is main's behaviour, but it is permanent, and nothing in the log distinguishes it from a first timeout.

Please say which binding the GDS gets by default on the three platforms (import, GPU copy or CPU copy), since that decides how often this path is the one titles take. Then either:

clear the mark from the CPU side when a CPU copy is bound and the timeout count has moved, before the next dispatch is recorded and with the previous submission retired; or
refuse ordering programs on a CPU-made copy with a throw, so the gap is visible rather than a silent fallback.

A test that times out once and then runs a normal dispatch on a CPU-made binding would pin it. The existing "dispatch after a timeout" case runs on an import or GPU copy, where the reset covers it.

The frame-time table is what I asked for; gpu_busy_ms_per_present going from about 150 to about 129 ms with the change is a large move for a GDS-only fix, so it is worth a line saying which phase it comes from.

main moved the shader disk cache to format 56; this branch's ordered-append field moves it to 57. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

TechLevelZero · 2026-11-04T07:55:18Z

✅ NVIDIA GPU test: passed

PR head 22c1254 merged onto main 5b1d381 · tested 2026-11-04 07:55 UTC · 35 min


GPU	NVIDIA RTX 2000 Ada Generation, driver 595.99.02, Vulkan 1.4.329
Host	Ubuntu 26.04.1 LTS, kernel 7.0.0-38-generic, container ubuntu:24.04
Logs	build and test logs for this run
Build	✅ passed (Release, `-DANYPS5_ENABLE_SPIRV_TOOLS=ON`)
Tests	✅ 2,836 passed · 6 failed (0 new, 3 also fail on main, 3 unstable on main) · 49 skipped (`ANYPS5_REQUIRE_VULKAN=1`)
Test setup	CPU-only tests without a GPU; GPU tests (and any CPU-only test that failed or skipped without one) once on the GPU with `VK_LAYER_KHRONOS_validation` and `ANYPS5_NO_SHADER_CACHE=1`; main ran the same way
Test selection	2,891 of 3,883 tests run: the ones this change can reach, chosen by `tools/select_tests.py` from main's build graph
Vulkan validation	✅ no new validation IDs vs main (979 tests run on the GPU)
Slow tests (>30 s, GPU tests timed with validation layers)	⚠️ `agc_driver_pipeline_journal` 120.3 s

Unstable on main (3, these also failed in recent main runs on this machine; not counted against this PR)

agc_driver_pipeline_journal: failed in 3 of the last 10 main runs
agc_submit_progress: failed in 4 of the last 10 main runs
agc_submit_progress_no_timeline: failed in 3 of the last 10 main runs

Failing on main too (3, not counted against this PR)

agc_driver_image_table
guest_mmap_executable
guest_socket_options_multicast

Validation IDs already present on main (13 IDs in 543 tests)

VUID-VkDeviceCreateInfo-pNext-pNext (543 tests): agc_driver, agc_driver_address_passes, agc_driver_alias_copy_label, agc_driver_alpha_to_coverage, agc_driver_array_layer_target, agc_driver_bc_block_copy, agc_driver_bda_anchor_reads, agc_driver_bda_span_reads, and 535 more
VUID-VkShaderModuleCreateInfo-pCode-08737 (8 tests): agc_driver_dpp16_operands, agc_driver_dpp8_pk_fmac, agc_driver_dx10_clamp, agc_driver_float16_fma_rounding, agc_driver_float16_unused_nan_payloads, agc_driver_packed_alu, agc_driver_packed_f16_denormal_modes, agc_driver_packed_float16_native
VUID-VkShaderModuleCreateInfo-pCode-08739 (8 tests): agc_driver_dpp16_operands, agc_driver_dpp8_pk_fmac, agc_driver_dx10_clamp, agc_driver_float16_fma_rounding, agc_driver_float16_unused_nan_payloads, agc_driver_packed_alu, agc_driver_packed_f16_denormal_modes, agc_driver_packed_float16_native
VUID-VkShaderModuleCreateInfo-pCode-08741 (8 tests): agc_driver_dpp16_operands, agc_driver_dpp8_pk_fmac, agc_driver_dx10_clamp, agc_driver_float16_fma_rounding, agc_driver_float16_unused_nan_payloads, agc_driver_packed_alu, agc_driver_packed_f16_denormal_modes, agc_driver_packed_float16_native
VUID-vkCmdCopyImage-srcImage-01548 (7 tests): agc_driver_depth_clear_hle, agc_driver_depth_plane_buffer_memory, agc_driver_depth_plane_partial_store, agc_driver_depth_plane_sampling, agc_driver_depth_storage_dispatch, agc_driver_depth_surface_from_storage, agc_driver_dispatch_move_depth
VUID-vkCmdCopyImage-srcImage-01551 (7 tests): agc_driver_depth_clear_hle, agc_driver_depth_plane_buffer_memory, agc_driver_depth_plane_partial_store, agc_driver_depth_plane_sampling, agc_driver_depth_storage_dispatch, agc_driver_depth_surface_from_storage, agc_driver_dispatch_move_depth
VUID-VkPipelineColorBlendStateCreateInfo-pAttachments-00605 (2 tests): agc_driver_prepared_exports, agc_driver_srgb8_color_target
VUID-VkPipelineShaderStageCreateInfo-module-parameter (1 test): agc_driver_float16_unused_nan_payloads
VUID-VkPipelineShaderStageCreateInfo-pSpecializationInfo-06849 (1 test): agc_driver_float16_unused_nan_payloads
VUID-VkSamplerCreateInfo-addressModeU-01079 (1 test): agc_driver_image_gather_volume
VUID-VkSamplerCreateInfo-addressModeU-parameter (1 test): agc_driver_image_gather_volume
VUID-VkSamplerCreateInfo-addressModeV-parameter (1 test): agc_driver_image_gather_volume
VUID-VkSamplerCreateInfo-addressModeW-parameter (1 test): agc_driver_image_gather_volume

Skipped (49)

agc_driver_bda_reads_shadowed_results: skipped, the device reports watched host memory written when it imports or reads it
agc_driver_bda_reads_shadowed_results_per_build: skipped, the device reports watched host memory written when it imports or reads it
agc_driver_gcn_display_present: skipped, no display: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
agc_driver_sample_order_grid_reference: skipped, the device does not keep depth drawn at custom sample locations at those locations, even after a transition that names them: the composite of uniform renders holds the guest's depth in 0 of 3005 sloped samples, so no path can be compared with it bit for bit
agc_driver_scattered_runs: skipped, host imports are unavailable: /dev/udmabuf is not accessible
agc_driver_sparse_dma_buf_import: skipped, host imports are unavailable: /dev/udmabuf is not accessible
agc_driver_stencil_ref_export: skipped, the device lacks VK_EXT_shader_stencil_export
agc_host_import_sparse: skipped, host imports are unavailable: /dev/udmabuf is not accessible
agc_import_settle_initial_dcc: skipped, the device has no deferred settle of host imports: imports under 1 MiB wait for one only on Windows
agc_unit_shadow_completion_store: skipped, the device reports watched host memory written when it imports or reads it
guest_http_uri_locale: 
guest_net_ctl: skipped, the host has no IPv4 address
guest_video_out_color_space: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_hdr: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_high_frame_rate: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_output_mode: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_resolution_status: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_wqhd_detection: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
video_out_flip_aftergpu: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_backtoback: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_controls: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_cpuflip: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_cpuflip_in_order: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_dcc: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_deferred: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_eop: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_eop_aftergpu: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not av

Reproduce

git fetch origin 22c12543e63b265393731e179cb7e9d9d29d0808 5b1d381d5ad5d9f923326b18479c3bd889cdd588
git checkout 5b1d381d5ad5d9f923326b18479c3bd889cdd588 && git merge --no-ff 22c12543e63b265393731e179cb7e9d9d29d0808
git submodule update --init --recursive
cmake -S . -B build -G Ninja -DCMAKE_BUILD_TYPE=Release -DBUILD_TESTING=ON -DANYPS5_ENABLE_SPIRV_TOOLS=ON
cmake --build build && cmake --build build --target libs
VK_DRIVER_FILES=/nonexistent/icd.json ctest --test-dir build -E '^(agc|shader|recompiler|videoout|video_out|guest_video_out|vulkan)|gpu' --timeout 120 --parallel 32 --output-on-failure
ANYPS5_REQUIRE_VULKAN=1 ANYPS5_NO_SHADER_CACHE=1 VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation \
  ctest --test-dir build -R '^(agc|shader|recompiler|videoout|video_out|guest_video_out|vulkan)|gpu' --timeout 120 --parallel 4 --output-on-failure
# failures are then re-run alone: --timeout 360 --parallel 2; new validation IDs: the test alone, twice, --parallel 1

_{Automated build and test on a contributor's NVIDIA machine; it does not replace upstream CI. A new comment is posted for each new push. Full logs.}

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

TechLevelZero · 2026-11-04T14:09:00Z

✅ NVIDIA GPU test: passed

PR head 13c089e merged onto main b11c7bf · tested 2026-11-04 14:08 UTC · 42 min


GPU	NVIDIA RTX 2000 Ada Generation, driver 595.99.02, Vulkan 1.4.329
Host	Ubuntu 26.04.1 LTS, kernel 7.0.0-38-generic, container ubuntu:24.04
Logs	build and test logs for this run
Build	✅ passed (Release, `-DANYPS5_ENABLE_SPIRV_TOOLS=ON`)
Tests	✅ 2,849 passed · 7 failed (0 new, 3 also fail on main, 4 unstable on main) · 49 skipped (`ANYPS5_REQUIRE_VULKAN=1`)
Test setup	CPU-only tests without a GPU; GPU tests (and any CPU-only test that failed or skipped without one) once on the GPU with `VK_LAYER_KHRONOS_validation` and `ANYPS5_NO_SHADER_CACHE=1`; main ran the same way
Test selection	2,905 of 3,894 tests run: the ones this change can reach, chosen by `tools/select_tests.py` from main's build graph
Vulkan validation	✅ no new validation IDs vs main (985 tests run on the GPU)
Slow tests (>30 s, GPU tests timed with validation layers)	⚠️ `agc_driver_pipeline_journal` 120.2 s

Unstable on main (4, these also failed in recent main runs on this machine; not counted against this PR)

agc_driver_pipeline_journal: failed in 3 of the last 10 main runs
agc_driver_submit_progress: failed in 3 of the last 10 main runs
agc_submit_progress: failed in 4 of the last 10 main runs
agc_submit_progress_no_timeline: failed in 3 of the last 10 main runs

Failing on main too (3, not counted against this PR)

agc_driver_image_table
guest_mmap_executable
guest_socket_options_multicast

Validation IDs already present on main (13 IDs in 549 tests)

VUID-VkDeviceCreateInfo-pNext-pNext (549 tests): agc_driver, agc_driver_address_draw_space_reuse, agc_driver_address_passes, agc_driver_alias_copy_label, agc_driver_alpha_to_coverage, agc_driver_array_layer_target, agc_driver_bc_block_copy, agc_driver_bda_anchor_reads, and 541 more
VUID-VkShaderModuleCreateInfo-pCode-08737 (8 tests): agc_driver_dpp16_operands, agc_driver_dpp8_pk_fmac, agc_driver_dx10_clamp, agc_driver_float16_fma_rounding, agc_driver_float16_unused_nan_payloads, agc_driver_packed_alu, agc_driver_packed_f16_denormal_modes, agc_driver_packed_float16_native
VUID-VkShaderModuleCreateInfo-pCode-08739 (8 tests): agc_driver_dpp16_operands, agc_driver_dpp8_pk_fmac, agc_driver_dx10_clamp, agc_driver_float16_fma_rounding, agc_driver_float16_unused_nan_payloads, agc_driver_packed_alu, agc_driver_packed_f16_denormal_modes, agc_driver_packed_float16_native
VUID-VkShaderModuleCreateInfo-pCode-08741 (8 tests): agc_driver_dpp16_operands, agc_driver_dpp8_pk_fmac, agc_driver_dx10_clamp, agc_driver_float16_fma_rounding, agc_driver_float16_unused_nan_payloads, agc_driver_packed_alu, agc_driver_packed_f16_denormal_modes, agc_driver_packed_float16_native
VUID-vkCmdCopyImage-srcImage-01548 (7 tests): agc_driver_depth_clear_hle, agc_driver_depth_plane_buffer_memory, agc_driver_depth_plane_partial_store, agc_driver_depth_plane_sampling, agc_driver_depth_storage_dispatch, agc_driver_depth_surface_from_storage, agc_driver_dispatch_move_depth
VUID-vkCmdCopyImage-srcImage-01551 (7 tests): agc_driver_depth_clear_hle, agc_driver_depth_plane_buffer_memory, agc_driver_depth_plane_partial_store, agc_driver_depth_plane_sampling, agc_driver_depth_storage_dispatch, agc_driver_depth_surface_from_storage, agc_driver_dispatch_move_depth
VUID-VkPipelineColorBlendStateCreateInfo-pAttachments-00605 (1 test): agc_driver_srgb8_color_target
VUID-VkPipelineShaderStageCreateInfo-module-parameter (1 test): agc_driver_float16_unused_nan_payloads
VUID-VkPipelineShaderStageCreateInfo-pSpecializationInfo-06849 (1 test): agc_driver_float16_unused_nan_payloads
VUID-VkSamplerCreateInfo-addressModeU-01079 (1 test): agc_driver_image_gather_volume
VUID-VkSamplerCreateInfo-addressModeU-parameter (1 test): agc_driver_image_gather_volume
VUID-VkSamplerCreateInfo-addressModeV-parameter (1 test): agc_driver_image_gather_volume
VUID-VkSamplerCreateInfo-addressModeW-parameter (1 test): agc_driver_image_gather_volume

Skipped (49)

agc_driver_bda_reads_shadowed_results: skipped, the device reports watched host memory written when it imports or reads it
agc_driver_bda_reads_shadowed_results_per_build: skipped, the device reports watched host memory written when it imports or reads it
agc_driver_gcn_display_present: skipped, no display: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
agc_driver_sample_order_grid_reference: skipped, the device does not keep depth drawn at custom sample locations at those locations, even after a transition that names them: the composite of uniform renders holds the guest's depth in 0 of 3005 sloped samples, so no path can be compared with it bit for bit
agc_driver_scattered_runs: skipped, host imports are unavailable: /dev/udmabuf is not accessible
agc_driver_sparse_dma_buf_import: skipped, host imports are unavailable: /dev/udmabuf is not accessible
agc_driver_stencil_ref_export: skipped, the device lacks VK_EXT_shader_stencil_export
agc_host_import_sparse: skipped, host imports are unavailable: /dev/udmabuf is not accessible
agc_import_settle_initial_dcc: skipped, the device has no deferred settle of host imports: imports under 1 MiB wait for one only on Windows
agc_unit_shadow_completion_store: skipped, the device reports watched host memory written when it imports or reads it
guest_http_uri_locale: 
guest_net_ctl: skipped, the host has no IPv4 address
guest_video_out_color_space: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_hdr: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_high_frame_rate: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_output_mode: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_resolution_status: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_wqhd_detection: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
video_out_flip_aftergpu: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_backtoback: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_controls: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_cpuflip: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_cpuflip_in_order: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_dcc: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_deferred: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_eop: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_eop_aftergpu: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not av

Reproduce

git fetch origin 13c089ea9a9ce970a5cd383310f40f4b799ed971 b11c7bf84f5d2f636ff60166cc3371bd071e24d5
git checkout b11c7bf84f5d2f636ff60166cc3371bd071e24d5 && git merge --no-ff 13c089ea9a9ce970a5cd383310f40f4b799ed971
git submodule update --init --recursive
cmake -S . -B build -G Ninja -DCMAKE_BUILD_TYPE=Release -DBUILD_TESTING=ON -DANYPS5_ENABLE_SPIRV_TOOLS=ON
cmake --build build && cmake --build build --target libs
VK_DRIVER_FILES=/nonexistent/icd.json ctest --test-dir build -E '^(agc|shader|recompiler|videoout|video_out|guest_video_out|vulkan)|gpu' --timeout 120 --parallel 32 --output-on-failure
ANYPS5_REQUIRE_VULKAN=1 ANYPS5_NO_SHADER_CACHE=1 VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation \
  ctest --test-dir build -R '^(agc|shader|recompiler|videoout|video_out|guest_video_out|vulkan)|gpu' --timeout 120 --parallel 4 --output-on-failure
# failures are then re-run alone: --timeout 360 --parallel 2; new validation IDs: the test alone, twice, --parallel 1

_{Automated build and test on a contributor's NVIDIA machine; it does not replace upstream CI. A new comment is posted for each new push. Full logs.}

TechLevelZero · 2026-11-04T15:50:01Z

✅ AMD GPU test: passed

PR head 13c089e merged onto main b11c7bf · tested 2026-11-04 15:49 UTC · 39 min


GPU	AMD Radeon RX 6300 (RADV NAVI24), driver Mesa 25.2.8-0ubuntu0.24.04.4, Vulkan 1.4.318
Host	Ubuntu 26.04.1 LTS, kernel 7.0.0-42-generic, container ubuntu:24.04
Logs	build and test logs for this run
Build	✅ passed (Release, `-DANYPS5_ENABLE_SPIRV_TOOLS=ON`)
Tests	✅ 2,836 passed · 9 failed (0 new, 8 also fail on main, 1 unstable on main) · 60 skipped (`ANYPS5_REQUIRE_VULKAN=1`)
Test setup	CPU-only tests without a GPU; GPU tests (and any CPU-only test that failed or skipped without one) once on the GPU with `VK_LAYER_KHRONOS_validation` and `ANYPS5_NO_SHADER_CACHE=1`; main ran the same way
Test selection	2,905 of 3,894 tests run: the ones this change can reach, chosen by `tools/select_tests.py` from main's build graph
Vulkan validation	✅ no new validation IDs vs main (985 tests run on the GPU)
Slow tests (>30 s, GPU tests timed with validation layers)	⚠️ `agc_driver_pipeline_journal` 120.4 s, ⚠️ `public_headers` 74.4 s, ⚠️ `agc_shader_sampler_limits` 35.1 s, ⚠️ `agc_driver_image_sample_unnormalized` 34.5 s, ⚠️ `agc_driver_lazy_resolved_abi` 32.6 s

Unstable on main (1, these also failed in recent main runs on this machine; not counted against this PR)

agc_driver_pipeline_journal: failed in 8 of the last 10 main runs

Failing on main too (8, not counted against this PR)

agc_driver_image_table
agc_driver_native_depth_bda_clear
agc_driver_native_multisample_depth
agc_driver_submit_progress
agc_submit_progress
agc_submit_progress_no_timeline
guest_mmap_executable
guest_socket_options_multicast

Validation IDs already present on main (8 IDs in 549 tests)

VUID-VkDeviceCreateInfo-pNext-pNext (549 tests): agc_driver, agc_driver_address_draw_space_reuse, agc_driver_address_passes, agc_driver_alias_copy_label, agc_driver_alpha_to_coverage, agc_driver_array_layer_target, agc_driver_bc_block_copy, agc_driver_bda_anchor_reads, and 541 more
VUID-vkCmdCopyImage-srcImage-01548 (4 tests): agc_driver_depth_clear_hle, agc_driver_depth_storage_dispatch, agc_driver_depth_surface_from_storage, agc_driver_dispatch_move_depth
VUID-vkCmdCopyImage-srcImage-01551 (4 tests): agc_driver_depth_clear_hle, agc_driver_depth_storage_dispatch, agc_driver_depth_surface_from_storage, agc_driver_dispatch_move_depth
VUID-VkPipelineColorBlendStateCreateInfo-pAttachments-00605 (1 test): agc_driver_srgb8_color_target
VUID-VkSamplerCreateInfo-addressModeU-01079 (1 test): agc_driver_image_gather_volume
VUID-VkSamplerCreateInfo-addressModeU-parameter (1 test): agc_driver_image_gather_volume
VUID-VkSamplerCreateInfo-addressModeV-parameter (1 test): agc_driver_image_gather_volume
VUID-VkSamplerCreateInfo-addressModeW-parameter (1 test): agc_driver_image_gather_volume

Skipped (60)

agc_driver_bda_reads_shadowed_results: skipped, the device reports watched host memory written when it imports or reads it
agc_driver_bda_reads_shadowed_results_per_build: skipped, the device reports watched host memory written when it imports or reads it
agc_driver_copied_writer_label_order: skipped, the device reports a minStorageBufferOffsetAlignment of 4, so the dispatch binds its record at +0x1004 in place and nothing is written back; the test needs an alignment above 4
agc_driver_float16_unused_nan_payloads: skipped, the device has no native f16 conversions with RTE rounding and kept f16 denormals
agc_driver_gcn_display_present: skipped, no display: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
agc_driver_image_table_vouched_lookups: skipped, the device reports 4 of 4 host-imported pages written after a GPU read, so a watched imported table cannot be vouched for
agc_driver_linear_target_direct_write_completion: skipped, the device does not keep host imports write-watched, so no completion stamp over the target is tracked
agc_driver_ngg_vertex_wave_reduction: skipped, s3 describes a 32-lane wave only on 32-lane subgroups, and the device has 64 lanes
agc_driver_scattered_runs: skipped, host imports are unavailable: /dev/udmabuf is not accessible
agc_driver_sparse_dma_buf_import: skipped, host imports are unavailable: /dev/udmabuf is not accessible
agc_driver_unit_shadow: skipped, the device does not import watched host memory for unit shadows
agc_driver_unit_shadow_budget: skipped, the device does not import watched host memory for unit shadows
agc_host_import_sparse: skipped, host imports are unavailable: /dev/udmabuf is not accessible
agc_idle_publish_wait: skipped, the device reports 4 of 4 host-imported pages written after a GPU read, so a forced watch of the import reads its submissions as CPU stores
agc_import_settle_initial_dcc: skipped, the device has no deferred settle of host imports: imports under 1 MiB wait for one only on Windows
agc_shadowed_key_stores: skipped, the device imports no watched host memory under a unit shadow
agc_small_tiled_shadow_label: skipped, host imports are unavailable: the test block is not imported and watched
agc_storage_clear_hand_over: skipped, the device imports no watched host memory for the clear-key hand-over
agc_storage_pending_key_refresh: skipped, refreshes over a pending key store are not testable here
agc_unit_shadow: skipped, the device does not import watched host memory for unit shadows
agc_unit_shadow_completion_store: skipped, the device reports watched host memory written when it imports or reads it
guest_http_uri_locale: 
guest_net_ctl: skipped, the host has no IPv4 address
guest_video_out_color_space: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_hdr: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_high_frame_rate: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_output_mode: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_resolution_status: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_wqhd_detection: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
video_out_flip_aftergpu: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_backtoback: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platf

Reproduce

git fetch origin 13c089ea9a9ce970a5cd383310f40f4b799ed971 b11c7bf84f5d2f636ff60166cc3371bd071e24d5
git checkout b11c7bf84f5d2f636ff60166cc3371bd071e24d5 && git merge --no-ff 13c089ea9a9ce970a5cd383310f40f4b799ed971
git submodule update --init --recursive
cmake -S . -B build -G Ninja -DCMAKE_BUILD_TYPE=Release -DBUILD_TESTING=ON -DANYPS5_ENABLE_SPIRV_TOOLS=ON
cmake --build build && cmake --build build --target libs
VK_DRIVER_FILES=/nonexistent/icd.json ctest --test-dir build -E '^(agc|shader|recompiler|videoout|video_out|guest_video_out|vulkan)|gpu' --timeout 120 --parallel 12 --output-on-failure
ANYPS5_REQUIRE_VULKAN=1 ANYPS5_NO_SHADER_CACHE=1 VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation \
  ctest --test-dir build -R '^(agc|shader|recompiler|videoout|video_out|guest_video_out|vulkan)|gpu' --timeout 120 --parallel 16 --output-on-failure
# failures are then re-run alone: --timeout 360 --parallel 2; new validation IDs: the test alone, twice, --parallel 1

_{Automated build and test on a contributor's AMD machine; it does not replace upstream CI. A new comment is posted for each new push. Full logs.}

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

TechLevelZero · 2026-11-05T09:59:07Z

✅ NVIDIA GPU test: passed

PR head 850614c merged onto main c5bd55a · tested 2026-11-05 09:59 UTC · 39 min


GPU	NVIDIA RTX 2000 Ada Generation, driver 595.99.02, Vulkan 1.4.329
Host	Ubuntu 26.04.1 LTS, kernel 7.0.0-38-generic, container ubuntu:24.04
Logs	build and test logs for this run
Build	✅ passed (Release, `-DANYPS5_ENABLE_SPIRV_TOOLS=ON`)
Tests	✅ 2,887 passed · 6 failed (0 new, 3 also fail on main, 3 unstable on main) · 55 skipped (`ANYPS5_REQUIRE_VULKAN=1`)
Test setup	CPU-only tests without a GPU; GPU tests (and any CPU-only test that failed or skipped without one) once on the GPU with `VK_LAYER_KHRONOS_validation` and `ANYPS5_NO_SHADER_CACHE=1`; main ran the same way
Test selection	2,948 of 3,972 tests run: the ones this change can reach, chosen by `tools/select_tests.py` from main's build graph
Vulkan validation	✅ no new validation IDs vs main (994 tests run on the GPU)
Slow tests (>30 s, GPU tests timed with validation layers)	⚠️ `agc_driver_pipeline_journal` 120.2 s, ⚠️ `public_headers` 31.3 s, ⚠️ `agc_driver_native_multisample_depth` 30.2 s

Unstable on main (3, these also failed in recent main runs on this machine; not counted against this PR)

agc_driver_pipeline_journal: failed in 3 of the last 10 main runs
agc_driver_submit_progress: failed in 1 of the last 10 main runs
agc_submit_progress: failed in 2 of the last 10 main runs

Failing on main too (3, not counted against this PR)

agc_driver_image_table
guest_mmap_executable
guest_socket_options_multicast

Passed on retry (1, failed in the parallel run, passed when re-run with a 360 s timeout; not counted against this PR)

agc_driver_native_multisample_depth: Timeout in the parallel run; also passed only on retry in 17 other runs in the last 7 days

Validation IDs already present on main (9 IDs in 551 tests)

VUID-VkDeviceCreateInfo-pNext-pNext (551 tests): agc_driver, agc_driver_address_draw_space_reuse, agc_driver_address_passes, agc_driver_alias_copy_label, agc_driver_alpha_to_coverage, agc_driver_array_layer_target, agc_driver_bc_block_copy, agc_driver_bda_anchor_reads, and 543 more
VUID-VkShaderModuleCreateInfo-pCode-08737 (14 tests): agc_driver_dpp16_operands, agc_driver_dpp8_pk_fmac, agc_driver_dx10_clamp, agc_driver_f16_fma_denormal_modes, agc_driver_f16_fma_invalid_nan, agc_driver_float16_fma_rounding, agc_driver_float16_modifiers, agc_driver_float16_native, and 6 more
VUID-VkShaderModuleCreateInfo-pCode-08739 (14 tests): agc_driver_dpp16_operands, agc_driver_dpp8_pk_fmac, agc_driver_dx10_clamp, agc_driver_f16_fma_denormal_modes, agc_driver_f16_fma_invalid_nan, agc_driver_float16_fma_rounding, agc_driver_float16_modifiers, agc_driver_float16_native, and 6 more
VUID-VkShaderModuleCreateInfo-pCode-08741 (14 tests): agc_driver_dpp16_operands, agc_driver_dpp8_pk_fmac, agc_driver_dx10_clamp, agc_driver_f16_fma_denormal_modes, agc_driver_f16_fma_invalid_nan, agc_driver_float16_fma_rounding, agc_driver_float16_modifiers, agc_driver_float16_native, and 6 more
VUID-vkCmdCopyImage-srcImage-01548 (7 tests): agc_driver_depth_clear_hle, agc_driver_depth_plane_buffer_memory, agc_driver_depth_plane_partial_store, agc_driver_depth_plane_sampling, agc_driver_depth_storage_dispatch, agc_driver_depth_surface_from_storage, agc_driver_dispatch_move_depth
VUID-vkCmdCopyImage-srcImage-01551 (7 tests): agc_driver_depth_clear_hle, agc_driver_depth_plane_buffer_memory, agc_driver_depth_plane_partial_store, agc_driver_depth_plane_sampling, agc_driver_depth_storage_dispatch, agc_driver_depth_surface_from_storage, agc_driver_dispatch_move_depth
VUID-VkPipelineColorBlendStateCreateInfo-pAttachments-00605 (2 tests): agc_driver_prepared_exports, agc_driver_srgb8_color_target
VUID-VkPipelineShaderStageCreateInfo-module-parameter (1 test): agc_driver_float16_unused_nan_payloads
VUID-VkPipelineShaderStageCreateInfo-pSpecializationInfo-06849 (1 test): agc_driver_float16_unused_nan_payloads

Skipped (55)

agc_driver_bda_reads_shadowed_results: skipped, the device reports watched host memory written when it imports or reads it
agc_driver_bda_reads_shadowed_results_per_build: skipped, the device reports watched host memory written when it imports or reads it
agc_driver_gcn_display_present: skipped, no display: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
agc_driver_sample_order_grid_reference: skipped, the device does not keep depth drawn at custom sample locations at those locations, even after a transition that names them: the composite of uniform renders holds the guest's depth in 0 of 3005 sloped samples, so no path can be compared with it bit for bit
agc_driver_scattered_runs: skipped, host imports are unavailable: /dev/udmabuf is not accessible
agc_driver_sparse_dma_buf_import: skipped, host imports are unavailable: /dev/udmabuf is not accessible
agc_driver_stencil_ref_export: skipped, the device lacks VK_EXT_shader_stencil_export
agc_host_import_sparse: skipped, host imports are unavailable: /dev/udmabuf is not accessible
agc_unit_shadow_completion_store: skipped, the device reports watched host memory written when it imports or reads it
guest_http_uri_locale: 
guest_net_ctl: skipped, the host has no IPv4 address
guest_net_ctl_disconnected: skipped, the host has no IPv4 address
guest_video_out_color_space: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_hdr: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_high_frame_rate: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_host_resolution_disabled: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_host_resolution_four_k: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_host_resolution_height_only: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_host_resolution_limit: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_host_resolution_malformed: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_host_resolution_width_only: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_output_mode: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_resolution_status: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_wqhd_detection: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
video_out_flip_aftergpu: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_backtoback: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_controls: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_cpuflip: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_cpuflip_in_order: skipped, no disp

Reproduce

git fetch origin 850614c0132b9d6d051cfa87a2c247d9fff979a3 c5bd55a89a07d39840867cc671f622f3d7c4869c
git checkout c5bd55a89a07d39840867cc671f622f3d7c4869c && git merge --no-ff 850614c0132b9d6d051cfa87a2c247d9fff979a3
git submodule update --init --recursive
cmake -S . -B build -G Ninja -DCMAKE_BUILD_TYPE=Release -DBUILD_TESTING=ON -DANYPS5_ENABLE_SPIRV_TOOLS=ON
cmake --build build && cmake --build build --target libs
VK_DRIVER_FILES=/nonexistent/icd.json ctest --test-dir build -E '^(agc|shader|recompiler|videoout|video_out|guest_video_out|vulkan)|gpu' --timeout 120 --parallel 32 --output-on-failure
ANYPS5_REQUIRE_VULKAN=1 ANYPS5_NO_SHADER_CACHE=1 VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation \
  ctest --test-dir build -R '^(agc|shader|recompiler|videoout|video_out|guest_video_out|vulkan)|gpu' --timeout 120 --parallel 4 --output-on-failure
# failures are then re-run alone: --timeout 360 --parallel 2; new validation IDs: the test alone, twice, --parallel 1

_{Automated build and test on a contributor's NVIDIA machine; it does not replace upstream CI. A new comment is posted for each new push. Full logs.}

ZSWolt · 2026-11-05T10:00:50Z

@XairRrR thanks for both re-checks and the approvals. One correction to my earlier reply: you're right that BunnyFl's failing run didn't log its driver, so I shouldn't have named …7029 as the driver it failed on. I'll leave the debt sentence as it is. It doesn't name a driver, and the rerun doesn't settle the late-start reading either way.

Since then I've merged main into the branch three more times, each time for a conflict GitHub reported:

22c12543e6 (5b1d381d5a): main moved the shader disk cache to format 56, so this branch's orderedAppend field now takes 57.
13c089ea9a (b11c7bf84f): ShaderResources.hpp added accessors on both sides; I kept both.
850614c013 (4d513cadcc): recordDispatch gained main's NoteGuestWork call right where this branch resets the ordered turn. I kept both: the reset, then NoteGuestWork.

For each merge, the build, the agc, shader and recompiler tests (1342 of 1342 for the latest), and check_conventions pass locally on RADV (RX 9060 XT). The PR's own commits are unchanged.

@TechLevelZero thanks for the NVIDIA and AMD runs on each new head. The NVIDIA run on 850614c passed too.

On CI: the macOS relinker failure on 13c089e was guest_none_relocations hitting its 120 s timeout. The Windows job on 850614c was canceled at the 45-minute job limit, inside "Build and run changed guest core/libs tests". This PR doesn't touch either of those areas.

🤖 Addressed by Claude Code

adam25136 · 2026-11-05T10:37:45Z

Windows and NVIDIA Blackwell run, for head 850614c013 merged into main b98c4af193 (git merge --no-ff, clean merge). The hosted windows check on this head was cancelled at its 45-minute limit (relinker (windows-latest, g++) passed), so this gives the head a Windows test result.


GPU	NVIDIA GeForce RTX 5080 (`10de:2c02`), driver 621.14 (32.0.16.2114), Vulkan 1.4.357
Host	Windows 10 Pro 22H2 (19045), i9-12900KF, WinLibs GCC 15.2.0 r7, Release, `BUILD_TESTING=ON`
Selection	2,408 tests, chosen by `tools/select_tests.py select` for the 27 changed files
Run	`select_tests.py ctest build --parallel 4 --timeout 120` with `ANYPS5_REQUIRE_VULKAN=1 ANYPS5_NO_SHADER_CACHE=1`; `main` `b98c4af193` ran its full suite (4,172 tests) the same way on this machine
Result	2,329 passed, 76 skipped, 3 failed, 0 new
Validation	The 1,361 `agc_` tests of the selection again with `VK_LAYER_KHRONOS_validation` 1.4.367 (the Khronos nightly built from `3baa2b8`), `ctest -V --parallel 4`, implicit layers disabled; `main` ran its 1,363 the same way. This change logs the same 151 messages in the same 23 tests as `main`, and no ID that `main` doesn't have for the same test*. The new `agc_driver_gds_append_consume_copied` logs none

agc_driver_gds_append_consume, the new agc_driver_gds_append_consume_copied and agc_driver_ds_append_consume pass on this subgroup-32 device (gds append consume tests passed, ds append consume tests passed).

The 3 failures fail on main b98c4af193 here too:

public_headers: five WKSecurityOrigin* declarations are missing on main since feat(libSceNKWebKit): implement WKHitTestResult property getters #15171 (see the comments there). Its 64 public_headers_compile_* tests are not run on either side, and they are 64 of the 76 skips.
agc_driver_direct_write_completion is AGC: agc_driver_direct_write_completion fails on Windows since #13159 #15564.
agc_driver_quad_pixel_mask is AGC: agc_driver_quad_pixel_mask samples the wrong mip level under a partial quad mask on Windows (RTX 5060 Ti and Radeon 860M) #14237; it fails on main with driver 621.14 and passed with 617.42.

The other 12 skips also skip on main here. The slowest selected test, agc_driver_pipeline_journal, took 22.3 s.

No title was run.

AI-assisted: yes. Claude Code ran the builds and tests and drafted this comment; the numbers are from the runs on this machine.

TechLevelZero · 2026-11-05T15:03:05Z

✅ AMD GPU test: passed

PR head 850614c merged onto main e772fb2 · tested 2026-11-05 15:03 UTC · 39 min


GPU	AMD Radeon RX 6300 (RADV NAVI24), driver Mesa 25.2.8-0ubuntu0.24.04.4, Vulkan 1.4.318
Host	Ubuntu 26.04.1 LTS, kernel 7.0.0-42-generic, container ubuntu:24.04
Logs	build and test logs for this run
Build	✅ passed (Release, `-DANYPS5_ENABLE_SPIRV_TOOLS=ON`)
Tests	✅ 2,948 passed · 10 failed (0 new, 6 also fail on main, 4 unstable on main) · 62 skipped (`ANYPS5_REQUIRE_VULKAN=1`)
Test setup	CPU-only tests without a GPU; GPU tests (and any CPU-only test that failed or skipped without one) once on the GPU with `VK_LAYER_KHRONOS_validation` and `ANYPS5_NO_SHADER_CACHE=1`; main ran the same way
Test selection	3,020 of 4,074 tests run: the ones this change can reach, chosen by `tools/select_tests.py` from main's build graph
Vulkan validation	✅ no new validation IDs vs main (985 tests run on the GPU)
Slow tests (>30 s, GPU tests timed with validation layers)	⚠️ `agc_driver_pipeline_journal` 120.1 s, ⚠️ `public_headers` 71.8 s, ⚠️ `agc_driver_image_sample_unnormalized` 43.2 s, ⚠️ `agc_driver_lazy_resolved_abi` 35.5 s, ⚠️ `agc_driver_depth_clear_hle` 30.7 s

Unstable on main (4, these also failed in recent main runs on this machine; not counted against this PR)

agc_driver_pipeline_journal: failed in 9 of the last 10 main runs
agc_submit_progress: failed in 9 of the last 10 main runs
agc_submit_progress_no_timeline: failed in 9 of the last 10 main runs
guest_pthread_start: failed in 2 of the last 10 main runs

Failing on main too (6, not counted against this PR)

agc_driver_image_table
agc_driver_native_depth_bda_clear
agc_driver_native_multisample_depth
agc_driver_submit_progress
guest_mmap_executable
guest_socket_options_multicast

Validated but not compared (2, not in the main baseline's validation pass)

guest_pthread_start
guest_sleep_wake

Validation IDs already present on main (4 IDs in 593 tests)

VUID-VkDeviceCreateInfo-pNext-pNext (593 tests): agc_compute_pipeline_replay, agc_driver, agc_driver_address_draw_space_reuse, agc_driver_address_passes, agc_driver_alias_copy_label, agc_driver_alpha_to_coverage, agc_driver_array_layer_target, agc_driver_bc_block_copy, and 585 more
VUID-vkCmdCopyImage-srcImage-01548 (4 tests): agc_driver_depth_clear_hle, agc_driver_depth_storage_dispatch, agc_driver_depth_surface_from_storage, agc_driver_dispatch_move_depth
VUID-vkCmdCopyImage-srcImage-01551 (4 tests): agc_driver_depth_clear_hle, agc_driver_depth_storage_dispatch, agc_driver_depth_surface_from_storage, agc_driver_dispatch_move_depth
VUID-VkPipelineColorBlendStateCreateInfo-pAttachments-00605 (1 test): agc_driver_prepared_exports

Skipped (62)

agc_driver_bda_reads_shadowed_results: skipped, the device reports watched host memory written when it imports or reads it
agc_driver_bda_reads_shadowed_results_per_build: skipped, the device reports watched host memory written when it imports or reads it
agc_driver_copied_writer_label_order: skipped, the device reports a minStorageBufferOffsetAlignment of 4, so the dispatch binds its record at +0x1004 in place and nothing is written back; the test needs an alignment above 4
agc_driver_float16_unused_nan_payloads: skipped, the device has no native f16 conversions with RTE rounding and kept f16 denormals
agc_driver_gcn_display_present: skipped, no display: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
agc_driver_image_table_vouched_lookups: skipped, the device reports 4 of 4 host-imported pages written after a GPU read, so a watched imported table cannot be vouched for
agc_driver_linear_target_direct_write_completion: skipped, the device does not keep host imports write-watched, so no completion stamp over the target is tracked
agc_driver_ngg_vertex_wave_reduction: skipped, s3 describes a 32-lane wave only on 32-lane subgroups, and the device has 64 lanes
agc_driver_recorded_draw_staging: skipped, the device does not keep host imports write-watched, so the target is never resident
agc_driver_vertex_elected_atomic_max: skipped, the device runs vertex programs in subgroups, so they keep their cross-lane reductions
agc_host_import_sparse: skipped, host imports are unavailable: /dev/udmabuf is not accessible
agc_idle_publish_wait: skipped, the device reports 4 of 4 host-imported pages written after a GPU read, so a forced watch of the import reads its submissions as CPU stores
agc_shadowed_key_stores: skipped, the device imports no watched host memory under a unit shadow
agc_small_tiled_shadow_label: skipped, host imports are unavailable: the test block is not imported and watched
agc_unit_shadow_completion_store: skipped, the device reports watched host memory written when it imports or reads it
guest_http_uri_locale: 
guest_net_ctl: skipped, the host has no IPv4 address
guest_net_ctl_disconnected: skipped, the host has no IPv4 address
guest_video_out_color_space: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_device_capability: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_hdr: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_high_frame_rate: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_host_resolution_disabled: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_host_resolution_four_k: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_host_resolution_height_only: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_host_resolution_limit: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_host_resolution_malformed: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_host_resolution_width_only: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_output_mode: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_resolution_status: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECON

Reproduce

git fetch origin 850614c0132b9d6d051cfa87a2c247d9fff979a3 e772fb25cc72ee5faf77847f06bbc69533016e82
git checkout e772fb25cc72ee5faf77847f06bbc69533016e82 && git merge --no-ff 850614c0132b9d6d051cfa87a2c247d9fff979a3
git submodule update --init --recursive
cmake -S . -B build -G Ninja -DCMAKE_BUILD_TYPE=Release -DBUILD_TESTING=ON -DANYPS5_ENABLE_SPIRV_TOOLS=ON
cmake --build build && cmake --build build --target libs
VK_DRIVER_FILES=/nonexistent/icd.json ctest --test-dir build -E '^(agc|shader|recompiler|videoout|video_out|guest_video_out|vulkan)|gpu' --timeout 120 --parallel 12 --output-on-failure
ANYPS5_REQUIRE_VULKAN=1 ANYPS5_NO_SHADER_CACHE=1 VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation \
  ctest --test-dir build -R '^(agc|shader|recompiler|videoout|video_out|guest_video_out|vulkan)|gpu' --timeout 120 --parallel 16 --output-on-failure
# failures are then re-run alone: --timeout 360 --parallel 2; new validation IDs: the test alone, twice, --parallel 1

_{Automated build and test on a contributor's AMD machine; it does not replace upstream CI. A new comment is posted for each new push. Full logs.}

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

BunnyFl · 2026-11-06T02:08:29Z

@ZSWolt you're right about the driver, and sorry for the slow reply: your offer to reword the shader.md sentence has been sitting unanswered on my side for two days.

On the driver. My failing run did not log its driver. Naming …7029 as the driver it failed on was never something my evidence supported. That is my fault, not yours: I should have logged the driver in the run itself. I now log vendor, model and driver in every GPU run I post.

On rewording the entry. I would not name …7029 in it, for the reason above. What the entry says at c95587e63b is accurate as it stands: it names the GPU and not a driver version. If you want to add anything, the part I can stand behind is that the late row 0 was seen once on an RX 9070 XT on Windows and has not been reproduced on that machine afterwards, with no driver version attached to either observation. I would also be fine leaving it exactly as it is.

Re-run at the current head, which is what I owed you. Head c95587e63b merged into main 01f804bb2f (merge 0aea09833d), Windows 11, WinLibs GCC, Release, AMD Radeon RX 9070 XT (device 0x7550), driver 32.0.32015.12012, Vulkan validation layer, ANYPS5_REQUIRE_VULKAN=1 ANYPS5_NO_SHADER_CACHE=1 VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation, ctest -V:

agc_atomic_gds_size, agc_driver_gds_append_consume, agc_driver_gds_append_consume_copied, agc_driver_gds_dma_fill, agc_driver_gds_dma_fill_private and agc_driver_pm4: 6 of 6 pass, none skipped, none left unbuilt.
No log line reports the validation layer missing, and no VUID- or SYNC-HAZARD id appears.
Both append-consume tests printed the plain gds append consume tests passed, with no note in parentheses. At this head that is the empty-note branch at GdsAppendConsume.cpp:533-534, so neither the device-limit skip at :324 nor the late-row-0 skip at :369 fired.

One note on my own tooling rather than on this PR: in my first attempt agc_driver_pm4 came back ***Not Run, with agc_driver_pm4_tests declared EXCLUDE_FROM_ALL at tests/cmake/agc_driver_graphics_tests.cmake:1 and my helper rebuilding only executables that ctest already lists a command for. Passing that target explicitly fixed it, and the numbers above are from the complete run.

What this does and does not settle. This machine is on the same driver as my 11-03 rerun, 32.0.32015.12012, and my one failing observation was on an unrecorded driver. This run separates nothing: it neither confirms nor refutes the late-start reading, exactly as before. It does say that the strict path still holds on this GPU at the current head, after the three main merges you describe.

TechLevelZero · 2026-11-06T05:55:31Z

✅ NVIDIA GPU test: passed

PR head c95587e merged onto main 82fcf46 · tested 2026-11-06 05:55 UTC · 20 min


GPU	NVIDIA RTX 2000 Ada Generation, driver 595.99.02, Vulkan 1.4.329
Host	Ubuntu 26.04.1 LTS, kernel 7.0.0-38-generic, container ubuntu:24.04
Logs	build and test logs for this run
Build	✅ passed (Release, `-DANYPS5_ENABLE_SPIRV_TOOLS=ON`)
Tests	✅ 2,988 passed · 3 failed (0 new, 3 also fail on main) · 56 skipped (`ANYPS5_REQUIRE_VULKAN=1`)
Test setup	CPU-only tests without a GPU; GPU tests (and any CPU-only test that failed or skipped without one) once on the GPU with `VK_LAYER_KHRONOS_validation` and `ANYPS5_NO_SHADER_CACHE=1`; main ran the same way
Test selection	3,047 of 4,158 tests run: the ones this change can reach, chosen by `tools/select_tests.py` from main's build graph
Vulkan validation	✅ no new validation IDs vs main (990 tests run on the GPU)
Slow tests (>30 s, GPU tests timed with validation layers)	⚠️ `agc_driver_pipeline_journal` 114.7 s, ⚠️ `agc_shader_sampler_limits` 40.5 s

Failing on main too (3, not counted against this PR)

agc_driver_image_table
guest_mmap_executable
guest_socket_options_multicast

Validation IDs already present on main (9 IDs in 595 tests)

VUID-VkDeviceCreateInfo-pNext-pNext (595 tests): agc_driver, agc_driver_address_draw_space_reuse, agc_driver_address_passes, agc_driver_alias_copy_label, agc_driver_alpha_to_coverage, agc_driver_array_layer_target, agc_driver_bc_block_copy, agc_driver_bda_anchor_reads, and 587 more
VUID-VkShaderModuleCreateInfo-pCode-08737 (14 tests): agc_driver_dpp16_operands, agc_driver_dpp8_pk_fmac, agc_driver_dx10_clamp, agc_driver_f16_fma_denormal_modes, agc_driver_f16_fma_invalid_nan, agc_driver_float16_fma_rounding, agc_driver_float16_modifiers, agc_driver_float16_native, and 6 more
VUID-VkShaderModuleCreateInfo-pCode-08739 (14 tests): agc_driver_dpp16_operands, agc_driver_dpp8_pk_fmac, agc_driver_dx10_clamp, agc_driver_f16_fma_denormal_modes, agc_driver_f16_fma_invalid_nan, agc_driver_float16_fma_rounding, agc_driver_float16_modifiers, agc_driver_float16_native, and 6 more
VUID-VkShaderModuleCreateInfo-pCode-08741 (14 tests): agc_driver_dpp16_operands, agc_driver_dpp8_pk_fmac, agc_driver_dx10_clamp, agc_driver_f16_fma_denormal_modes, agc_driver_f16_fma_invalid_nan, agc_driver_float16_fma_rounding, agc_driver_float16_modifiers, agc_driver_float16_native, and 6 more
VUID-vkCmdCopyImage-srcImage-01548 (7 tests): agc_driver_depth_clear_hle, agc_driver_depth_plane_buffer_memory, agc_driver_depth_plane_partial_store, agc_driver_depth_plane_sampling, agc_driver_depth_storage_dispatch, agc_driver_depth_surface_from_storage, agc_driver_dispatch_move_depth
VUID-vkCmdCopyImage-srcImage-01551 (7 tests): agc_driver_depth_clear_hle, agc_driver_depth_plane_buffer_memory, agc_driver_depth_plane_partial_store, agc_driver_depth_plane_sampling, agc_driver_depth_storage_dispatch, agc_driver_depth_surface_from_storage, agc_driver_dispatch_move_depth
VUID-VkPipelineColorBlendStateCreateInfo-pAttachments-00605 (2 tests): agc_driver_prepared_exports, agc_driver_srgb8_color_target
VUID-VkPipelineShaderStageCreateInfo-module-parameter (1 test): agc_driver_float16_unused_nan_payloads
VUID-VkPipelineShaderStageCreateInfo-pSpecializationInfo-06849 (1 test): agc_driver_float16_unused_nan_payloads

Skipped (56)

agc_driver_bda_reads_shadowed_results: skipped, the device reports watched host memory written when it imports or reads it
agc_driver_bda_reads_shadowed_results_per_build: skipped, the device reports watched host memory written when it imports or reads it
agc_driver_gcn_display_present: skipped, no display: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
agc_driver_sample_order_grid_reference: skipped, the device does not keep depth drawn at custom sample locations at those locations, even after a transition that names them: the composite of uniform renders holds the guest's depth in 0 of 3005 sloped samples, so no path can be compared with it bit for bit
agc_driver_sparse_dma_buf_import: skipped, host imports are unavailable: /dev/udmabuf is not accessible
agc_driver_stencil_ref_export: skipped, the device lacks VK_EXT_shader_stencil_export
agc_driver_vertex_elected_atomic_max: skipped, the device runs vertex programs in subgroups, so they keep their cross-lane reductions
agc_host_import_sparse: skipped, host imports are unavailable: /dev/udmabuf is not accessible
agc_unit_shadow_completion_store: skipped, the device reports watched host memory written when it imports or reads it
guest_http_uri_locale: 
guest_net_ctl: skipped, the host has no IPv4 address
guest_net_ctl_disconnected: skipped, the host has no IPv4 address
guest_video_out_color_space: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_device_capability: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_hdr: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_high_frame_rate: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_host_resolution_disabled: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_host_resolution_four_k: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_host_resolution_height_only: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_host_resolution_limit: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_host_resolution_malformed: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_host_resolution_width_only: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_output_mode: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_resolution_status: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_wqhd_detection: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
video_out_flip_aftergpu: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_backtoback: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_controls: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_cpuflip: skipped, no display or Vulkan de

Reproduce

git fetch origin c95587e63bcae6b609d71381b3057c3438e602d4 82fcf46d250e0ec2f436ceffd607dd2ff04c62d7
git checkout 82fcf46d250e0ec2f436ceffd607dd2ff04c62d7 && git merge --no-ff c95587e63bcae6b609d71381b3057c3438e602d4
git submodule update --init --recursive
cmake -S . -B build -G Ninja -DCMAKE_BUILD_TYPE=Release -DBUILD_TESTING=ON -DANYPS5_ENABLE_SPIRV_TOOLS=ON
cmake --build build && cmake --build build --target libs
VK_DRIVER_FILES=/nonexistent/icd.json ctest --test-dir build -E '^(agc|shader|recompiler|videoout|video_out|guest_video_out|vulkan)|gpu' --timeout 120 --parallel 32 --output-on-failure
ANYPS5_REQUIRE_VULKAN=1 ANYPS5_NO_SHADER_CACHE=1 VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation \
  ctest --test-dir build -R '^(agc|shader|recompiler|videoout|video_out|guest_video_out|vulkan)|gpu' --timeout 120 --parallel 4 --output-on-failure
# failures are then re-run alone: --timeout 360 --parallel 2; new validation IDs: the test alone, twice, --parallel 1

_{Automated build and test on a contributor's NVIDIA machine; it does not replace upstream CI. A new comment is posted for each new push. Full logs.}

BunnyFl

@ZSWolt approving at c95587e63b. My only formal review here is the 11-02 comment on 5332c91593 about the late row 0. The split-dispatch case now reports that case instead of failing, and my 11-06 rerun at this head passed. This adds the gated run with a negative control that I hadn't done.

What I ran. This head merged onto main 1ff0fe98aa (merge 98ea5d2326, clean), Windows 11 Pro, WinLibs GCC, Release, AMD Radeon RX 9070 XT (vulkaninfo: driverInfo 26.10.2 (LLPC), driverVersion 2.0.416), with ANYPS5_REQUIRE_VULKAN=1, ANYPS5_NO_SHADER_CACHE=1, VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation and ctest -V:

agc_atomic_gds_size, agc_driver_gds_append_consume, agc_driver_gds_append_consume_copied, agc_driver_gds_dma_fill, agc_driver_gds_dma_fill_private, agc_driver_pm4 and agc_guarded_memory_loads: 7 of 7 pass, none unbuilt or skipped. Both append-consume tests print the plain gds append consume tests passed, the empty-note branch at GdsAppendConsume.cpp:533-534; neither skip note fired.
No missing-layer line, no VUID- and no SYNC-HAZARD ids.
Negative control: the same merge with only core/shader/recompiler/SpirvBackend/src/SpirvMemory/SpirvMemoryInstructions.cpp taken from main fails both append-consume tests, with group 0 lane 0 returned 1032, expected 1000 (the counter in launch order) and returned 1012 for the copied registration. The other five still pass.

Overlap, checked with tools/pr_overlap.py --branch at this head:

core/shader/recompiler/ShaderDiskCache.hpp L16 (the cache format number) is shared with 13 open pull requests; whichever lands later takes the next number, as with your earlier merges.
#6146 changes GuestBufferMemory.hpp L607 on main, which this branch also changes. Both are in the review queue; one of you will need to merge after the other lands.
#12840 shares VulkanDevice.cpp L5904 and ShaderResources.hpp L265, and #14247 shares docs/dev/debt/shader.md L210-211.
94 more pull requests touch the same files but conflict with main and could not be compared.

Approving.

AI-assisted: yes (Claude Code)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

TechLevelZero · 2026-11-06T18:27:35Z

✅ NVIDIA GPU test: passed

PR head 3f6d916 merged onto main 8fa2f2f · tested 2026-11-06 18:27 UTC · 31 min


GPU	NVIDIA RTX 2000 Ada Generation, driver 595.99.02, Vulkan 1.4.329
Host	Ubuntu 26.04.1 LTS, kernel 7.0.0-38-generic, container ubuntu:24.04
Logs	build and test logs for this run
Build	✅ passed (Release, `-DANYPS5_ENABLE_SPIRV_TOOLS=ON`)
Tests	✅ 3,023 passed · 4 failed (0 new, 3 also fail on main, 1 unstable on main) · 55 skipped (`ANYPS5_REQUIRE_VULKAN=1`)
Test setup	CPU-only tests without a GPU; GPU tests (and any CPU-only test that failed or skipped without one) once on the GPU with `VK_LAYER_KHRONOS_validation` and `ANYPS5_NO_SHADER_CACHE=1`; main ran the same way
Test selection	3,082 of 4,235 tests run: the ones this change can reach, chosen by `tools/select_tests.py` from main's build graph
Vulkan validation	✅ no new validation IDs vs main (1001 tests run on the GPU)
Slow tests (>30 s, GPU tests timed with validation layers)	⚠️ `agc_driver_pipeline_journal` 120.1 s

Unstable on main (1, these also failed in recent main runs on this machine; not counted against this PR)

agc_driver_pipeline_journal: failed in 1 of the last 10 main runs

Failing on main too (3, not counted against this PR)

agc_driver_image_table
guest_mmap_executable
guest_socket_options_multicast

Validation IDs already present on main (9 IDs in 599 tests)

VUID-VkDeviceCreateInfo-pNext-pNext (599 tests): agc_driver, agc_driver_address_draw_space_reuse, agc_driver_address_draw_space_reuse_verify, agc_driver_address_passes, agc_driver_alias_copy_label, agc_driver_alpha_to_coverage, agc_driver_array_layer_target, agc_driver_bc_block_copy, and 591 more
VUID-VkShaderModuleCreateInfo-pCode-08737 (14 tests): agc_driver_dpp16_operands, agc_driver_dpp8_pk_fmac, agc_driver_dx10_clamp, agc_driver_f16_fma_denormal_modes, agc_driver_f16_fma_invalid_nan, agc_driver_float16_fma_rounding, agc_driver_float16_modifiers, agc_driver_float16_native, and 6 more
VUID-VkShaderModuleCreateInfo-pCode-08739 (14 tests): agc_driver_dpp16_operands, agc_driver_dpp8_pk_fmac, agc_driver_dx10_clamp, agc_driver_f16_fma_denormal_modes, agc_driver_f16_fma_invalid_nan, agc_driver_float16_fma_rounding, agc_driver_float16_modifiers, agc_driver_float16_native, and 6 more
VUID-VkShaderModuleCreateInfo-pCode-08741 (14 tests): agc_driver_dpp16_operands, agc_driver_dpp8_pk_fmac, agc_driver_dx10_clamp, agc_driver_f16_fma_denormal_modes, agc_driver_f16_fma_invalid_nan, agc_driver_float16_fma_rounding, agc_driver_float16_modifiers, agc_driver_float16_native, and 6 more
VUID-vkCmdCopyImage-srcImage-01548 (7 tests): agc_driver_depth_clear_hle, agc_driver_depth_plane_buffer_memory, agc_driver_depth_plane_partial_store, agc_driver_depth_plane_sampling, agc_driver_depth_storage_dispatch, agc_driver_depth_surface_from_storage, agc_driver_dispatch_move_depth
VUID-vkCmdCopyImage-srcImage-01551 (7 tests): agc_driver_depth_clear_hle, agc_driver_depth_plane_buffer_memory, agc_driver_depth_plane_partial_store, agc_driver_depth_plane_sampling, agc_driver_depth_storage_dispatch, agc_driver_depth_surface_from_storage, agc_driver_dispatch_move_depth
VUID-VkPipelineColorBlendStateCreateInfo-pAttachments-00605 (1 test): agc_driver_prepared_exports
VUID-VkPipelineShaderStageCreateInfo-module-parameter (1 test): agc_driver_float16_unused_nan_payloads
VUID-VkPipelineShaderStageCreateInfo-pSpecializationInfo-06849 (1 test): agc_driver_float16_unused_nan_payloads

Skipped (55)

agc_driver_bda_reads_shadowed_results: skipped, the device reports watched host memory written when it imports or reads it
agc_driver_bda_reads_shadowed_results_per_build: skipped, the device reports watched host memory written when it imports or reads it
agc_driver_gcn_display_present: skipped, no display: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
agc_driver_sample_order_grid_reference: skipped, the device does not keep depth drawn at custom sample locations at those locations, even after a transition that names them: the composite of uniform renders holds the guest's depth in 0 of 3005 sloped samples, so no path can be compared with it bit for bit
agc_driver_scattered_runs: skipped, host imports are unavailable: /dev/udmabuf is not accessible
agc_driver_vertex_elected_atomic_max: skipped, the device runs vertex programs in subgroups, so they keep their cross-lane reductions
agc_host_import_sparse: skipped, host imports are unavailable: /dev/udmabuf is not accessible
agc_unit_shadow_completion_store: skipped, the device reports watched host memory written when it imports or reads it
guest_http_uri_locale: 
guest_net_ctl: skipped, the host has no IPv4 address
guest_net_ctl_disconnected: skipped, the host has no IPv4 address
guest_video_out_color_space: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_device_capability: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_hdr: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_high_frame_rate: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_host_resolution_disabled: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_host_resolution_four_k: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_host_resolution_height_only: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_host_resolution_limit: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_host_resolution_malformed: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_host_resolution_width_only: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_output_mode: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_resolution_status: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
guest_video_out_wqhd_detection: skipped, no display or Vulkan device: SDL_InitSubSystem(VIDEO | GAMECONTROLLER) failed: wayland,x11 not available
video_out_flip_aftergpu: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_backtoback: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_controls: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in current SDL video driver (offscreen) or platform
video_out_flip_cpuflip: skipped, no display or Vulkan device: DisplayWindow: Vulkan support is either not configured in SDL or not available in curre

Reproduce

git fetch origin 3f6d916dd55a5e838400086d868b98918b0a4972 8fa2f2f8881b537b24062bd377badebae8d97abf
git checkout 8fa2f2f8881b537b24062bd377badebae8d97abf && git merge --no-ff 3f6d916dd55a5e838400086d868b98918b0a4972
git submodule update --init --recursive
cmake -S . -B build -G Ninja -DCMAKE_BUILD_TYPE=Release -DBUILD_TESTING=ON -DANYPS5_ENABLE_SPIRV_TOOLS=ON
cmake --build build && cmake --build build --target libs
VK_DRIVER_FILES=/nonexistent/icd.json ctest --test-dir build -E '^(agc|shader|recompiler|videoout|video_out|guest_video_out|vulkan)|gpu' --timeout 120 --parallel 32 --output-on-failure
ANYPS5_REQUIRE_VULKAN=1 ANYPS5_NO_SHADER_CACHE=1 VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation \
  ctest --test-dir build -R '^(agc|shader|recompiler|videoout|video_out|guest_video_out|vulkan)|gpu' --timeout 120 --parallel 4 --output-on-failure
# failures are then re-run alone: --timeout 360 --parallel 2; new validation IDs: the test alone, twice, --parallel 1

_{Automated build and test on a contributor's NVIDIA machine; it does not replace upstream CI. A new comment is posted for each new push. Full logs.}

ZSWolt · 2026-11-06T18:28:53Z

@BunnyFl thanks for the rerun, the approval and the negative control. Taking only SpirvMemoryInstructions.cpp from main makes both append-consume tests return 1032 and 1012 instead of 1000. That's the check I most wanted from hardware that isn't mine. I agree on the entry and will leave it as it is. It names the GPU and not a driver, and neither of your runs since then has reproduced the late row 0. Thanks also for the overlap list:

Cache format: whichever of the 13 lands later takes the next number, as you say.
fix(shader): translate ds_min/max/add_src2_f32 through dsSrc2 #14247: it has since merged into main. Its docs/dev/debt/shader.md change was this branch's last conflict, resolved in 3f6d916dd5 with main's DS src2 wording and this PR's two ds_ordered_count entries.
fix(agc): execute the DB's out-of-place depth expand #6146 and fix(agc): write back BDA stores into writable heap mirrors #12840: I'll merge whichever of us lands second.

@adam25136 thanks for the Windows run on the RTX 5080. It's the first result for a Windows head with a subgroup-32 NVIDIA driver, and it covers the 850614c head whose hosted windows job was cancelled. Thanks too for checking that the three failures also fail on main, and for the validation comparison: no ID that main lacks.

@TechLevelZero thanks for the NVIDIA and AMD runs on every head, including 3f6d916 (0 new failures).

On the hosted CI for 3f6d916: relinker (windows-latest, g++) failed with four check_compile_modes_* tests at their 120 s timeout, and the other 786 of 790 passed. This PR doesn't touch compile-mode tooling or the relinker.

🤖 Addressed by Claude Code

BunnyFl

@ZSWolt re-checked at 3f6d916dd5, and my approval stands.

3f6d916dd5 merges main (cefc546c86) onto c95587e63b, the head I approved. The PR's own +/- lines are identical at both heads in every file it changes.

What I ran

Windows 11 Pro 10.0.26300, WinLibs GCC 15.2.0, Release, this head merged onto current main (merge cefe4daece). The GPU is an AMD Radeon RX 9070 XT (0x1002:0x7550, Vulkan API 0x404167, as the test logs report it). Windows reports driver 32.0.32015.12012, read right after the runs; the logs don't print it. ANYPS5_REQUIRE_VULKAN=1 ANYPS5_NO_SHADER_CACHE=1 VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation, ctest -V, with the same seven tests as at my approval:

PR: agc_atomic_gds_size, agc_driver_gds_append_consume, agc_driver_gds_append_consume_copied, agc_driver_gds_dma_fill, agc_driver_gds_dma_fill_private, agc_driver_pm4 and agc_guarded_memory_loads: 7 of 7 pass.
Negative control: the same tree with main's core/shader/recompiler/SpirvBackend/src/SpirvMemory/SpirvMemoryInstructions.cpp. agc_driver_gds_append_consume and agc_driver_gds_append_consume_copied fail; 5 of 7 pass.
Neither log reports was not found but was requested by env var VK_INSTANCE_LAYERS. Both have no VUID- and no SYNC-HAZARD IDs.
tools/check_conventions.py --base upstream/main --head on this head: all checks pass.

Overlap

tools/pr_overlap.py --branch on this head:

#6146 still conflicts at Graphics/include/GuestBufferMemory.hpp L607, as at my approval.
New since then: #12840 conflicts at Execution/src/VulkanDevice.cpp L5904 and Graphics/include/ShaderResources.hpp L265.
14 PRs change the same line of core/shader/recompiler/ShaderDiskCache.hpp (L16), the shader cache format number.

AI-assisted: yes (Claude Code)

github-actions Bot added the unblocks title Names a title ID on the Unblocks line label Oct 29, 2026

oneandonlydean self-assigned this Oct 29, 2026

CnRJay reviewed Oct 29, 2026

View reviewed changes

github-actions Bot added the needs rebase Conflicts with main, rebase needed label Oct 30, 2026

Merge remote-tracking branch 'origin/main' into fix/recompiler-ordere…

7bdeb14

…d-append Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

github-actions Bot added shader core/shader docs agc libSceAgc* and removed needs rebase Conflicts with main, rebase needed labels Oct 30, 2026

Gonic0 mentioned this pull request Oct 30, 2026

perf(agc): back the GDS with anonymous shared memory on integrated GPUs and keep its DMA_DATA on the GPU #13159

Merged

12 tasks done

MarcZX10 mentioned this pull request Oct 31, 2026

feat(progress): count stubs no measured use needs as unobserved #13801

Closed

12 tasks done

CnRJay reviewed Nov 1, 2026

View reviewed changes

github-actions Bot added the needs rebase Conflicts with main, rebase needed label Nov 2, 2026

Merge main 5b1d381 into boykopovar#12944 (390b05a)

22c1254

main moved the shader disk cache to format 56; this branch's ordered-append field moves it to 57. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

github-actions Bot added review queue Next in the maintainer review queue: checks pass, no open change request, no conflict and removed needs rebase Conflicts with main, rebase needed labels Nov 4, 2026

github-actions Bot added the needs rebase Conflicts with main, rebase needed label Nov 4, 2026

Merge main b11c7bf into boykopovar#12944 (22c1254)

13c089e

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

github-actions Bot removed review queue Next in the maintainer review queue: checks pass, no open change request, no conflict needs rebase Conflicts with main, rebase needed labels Nov 4, 2026

TripyFox mentioned this pull request Nov 4, 2026

perf: PPSA21567 Crash Site, a pixel shader's GDS append serializes on NVIDIA and costs ~16 ms of GPU per frame (RTX 5090, Windows) #15560

Open

3 tasks done

BunnyFl mentioned this pull request Nov 5, 2026

fix(agc): execute the DB's out-of-place depth expand #6146

Open

13 tasks done

Merge main 4d513ca into boykopovar#12944 (13c089e)

850614c

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

AlexWhitehouse mentioned this pull request Nov 5, 2026

feat(agc): poll and write registers in PM4 COND_WRITE #16184

Open

12 tasks done

Merge main ea07ec8 into boykopovar#12944 (850614c)

c95587e

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

github-actions Bot added needs rebase Conflicts with main, rebase needed review queue Next in the maintainer review queue: checks pass, no open change request, no conflict and removed needs rebase Conflicts with main, rebase needed labels Nov 5, 2026

BunnyFl approved these changes Nov 6, 2026

View reviewed changes

Merge main cefc546 into boykopovar#12944 (c95587e)

3f6d916

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

github-actions Bot removed the review queue Next in the maintainer review queue: checks pass, no open change request, no conflict label Nov 6, 2026

BunnyFl approved these changes Nov 6, 2026

View reviewed changes

Conversation

ZSWolt commented Oct 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What

Tested

Progress

Checklist

Uh oh!

oneandonlydean commented Oct 29, 2026

Uh oh!

github-actions Bot commented Oct 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ZSWolt commented Oct 29, 2026

Uh oh!

CnRJay left a comment

Choose a reason for hiding this comment

Uh oh!

TechLevelZero commented Oct 29, 2026

✅ NVIDIA GPU test: passed

Uh oh!

ZSWolt commented Oct 29, 2026

Uh oh!

TechLevelZero commented Oct 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

✅ NVIDIA GPU test: passed

Uh oh!

bunbeans2-arch commented Oct 30, 2026

Uh oh!

TechLevelZero commented Oct 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

✅ NVIDIA GPU test: passed

Uh oh!

JJenkx commented Oct 30, 2026

Uh oh!

jacknieland commented Oct 31, 2026

Uh oh!

CnRJay left a comment

Choose a reason for hiding this comment

Uh oh!

TechLevelZero commented Nov 4, 2026

✅ NVIDIA GPU test: passed

Uh oh!

TechLevelZero commented Nov 4, 2026

✅ NVIDIA GPU test: passed

Uh oh!

TechLevelZero commented Nov 4, 2026

✅ AMD GPU test: passed

Uh oh!

TechLevelZero commented Nov 5, 2026

✅ NVIDIA GPU test: passed

Uh oh!

ZSWolt commented Nov 5, 2026

Uh oh!

adam25136 commented Nov 5, 2026

Uh oh!

TechLevelZero commented Nov 5, 2026

✅ AMD GPU test: passed

Uh oh!

BunnyFl commented Nov 6, 2026

Uh oh!

TechLevelZero commented Nov 6, 2026

✅ NVIDIA GPU test: passed

Uh oh!

BunnyFl left a comment

Choose a reason for hiding this comment

Uh oh!

TechLevelZero commented Nov 6, 2026

✅ NVIDIA GPU test: passed

Uh oh!

ZSWolt commented Nov 6, 2026

Uh oh!

BunnyFl left a comment

Choose a reason for hiding this comment

What I ran

Overlap

Uh oh!

This branch has not been deployed

Reviewers

Uh oh!

ZSWolt commented Oct 29, 2026 •

edited

Loading

github-actions Bot commented Oct 29, 2026 •

edited

Loading

TechLevelZero commented Oct 29, 2026 •

edited

Loading

TechLevelZero commented Oct 30, 2026 •

edited

Loading