[milescoviello]

What

tests/Commands.cpp gets a testDrawPackets case after testIndexBuffer. With modifier 0, sceAgcDcbDrawIndirect writes a five-word DRAW_INDIRECT with the argument offset, both patch offsets at 0x280 and the auto-index initiator 2. Modifier bit 8 adds 0x20 to the initiator unless bit 32 is set. sceAgcDcbDrawIndexOffset writes a five-word DRAW_INDEX_OFFSET_2 whose first word is the index count, or 1 for a count of 0. sceAgcDcbSetNumInstances writes the two-word NUM_INSTANCES. sceAgcDcbSetIndexSize writes the index size and cache policy (bits 6-7) into the second word of its three-word packet, and sceAgcDcbSetIndexSize_0600 adds the per-instance object id at bit 14. A misaligned argument offset, a modifier past 33 bits, an index size above 2, a cache policy above 3 and an object id above 1 are refused before anything is written, and the size getters report 20, 20, 8 and 12 bytes.

It is called by all eight titles run here for sceAgcDcbSetNumInstances; PPSA01494, PPSA03984, PPSA18194, PPSA30839, PPSA31162, PPSA34502 and PPSA36591 for sceAgcDcbDrawIndirect; PPSA01494, PPSA18194, PPSA21564, PPSA30839, PPSA31162, PPSA34502 and PPSA36591 for sceAgcDcbSetIndexSize_0600; PPSA01494, PPSA21564, PPSA30839, PPSA31162, PPSA34502 and PPSA36591 for sceAgcDcbDrawIndexOffset; PPSA03984 for sceAgcDcbSetIndexSize (call sites found through the import's PLT stub in each title's relinked eboot.bin), and until now no test called it.

Changed: core/libs/prx/libSceAgc/tests/Commands.cpp. Test only, no change to the library.

Tested

Linux (Ubuntu 26.04, GCC 15.2), Release: cmake --build build --target agc_command_tests and ctest -R '^agc_commands$' --repeat until-fail:3 pass on this branch, which is current main plus the test. tools/check_conventions.py --base upstream/main passes.

Windows 11 with an RTX 2070 SUPER (maskym's run, NVIDIA 621.01, Khronos validation layer, WinLibs GCC 15.2, Release): this PR's tests pass at head 60ee0115d0 merged into main f759e165d9, in a joint tree with other open PRs, and every other agc and video_out test gives the same result as on main.

Checklist

• Based on current main; no other open PR implements the same functions
• One topic per PR; follow-ups go in a new PR
• No comments in code except technical debt
• Unimplemented paths throw (NotImplemented_nid_no_patch); silent stubs are listed under Silent stubs in their library's technical debt file
• No notes, investigation .md files or images added to the repository (attach them to this PR)
• New third-party code is a submodule built from source
• Generic behaviour, not specific to one title
• Depends on: none
• Fixes: none
• Unblocks: none
• GPUs tested: n/a
• AI-assisted: yes

🤖 Generated with Claude Code

[milescoviello]

Pushed 60ee011. It applies the check MarcZX10 asked for on #11674: the refused-packet buffer is filled with a sentinel first, and after each refusal the test compares all 64 words and cursor_up with the snapshot, instead of only the first word. agc_initialization and agc_commands pass on Linux.

AI-assisted: yes (Claude Code)

[milescoviello]

Windows 11 VM (GT 1030, NVIDIA 582.66, WinLibs GCC 15.2.0, Release, Ninja): this head 60ee0115d0 and its base main abed035 each build agc_command_tests, and agc_initialization and agc_commands pass 3 of 3 on both.

AI-assisted: yes (Claude Code)

[maskym]

Windows + NVIDIA result for head 60ee0115d0, merged into main f759e165d9 (the PR's GPUs tested line names no NVIDIA run on Windows).

This PR's own tests: the libSceAgc packet tests agc_commands (0.15 s), agc_shader_helpers (0.08 s), agc_register_command_tests, agc_dma_data_tests and agc_initialization, which are CPU-side, all passed.

Host: Windows 11 26300, Intel i7-12700K, NVIDIA GeForce RTX 2070 SUPER (vendor 0x10de, device 0x1e84; Vulkan loader 1.4.357 from the NVIDIA driver 621.01), Khronos Validation Layers nightly 1.4.366 (b2b44c0d37) through VK_ADD_LAYER_PATH, WinLibs GCC 15.2.0 (x86_64-ucrt-posix-seh, r6), Release, BUILD_TESTING=ON, Ninja.

Same focused run on main f759e165d9 and on this tree, same machine and GPU: VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation, ANYPS5_NO_SHADER_CACHE=1, ANYPS5_REQUIRE_VULKAN=1, VK_LOADER_DEBUG=error,warn,layer, ctest --test-dir build -R '^(agc|guest_video_out|video_out)' --parallel 4 -V.

• main: 1127/1128 passed (9 skipped), the failure being agc_driver_mesh_barycentric_emulation (AGC: agc_driver_mesh_barycentric_emulation fails since #9193 on devices with mesh shaders and VK_KHR_fragment_shader_barycentric #12121).
• This tree: every pre-existing test passes, including that one (the tree carries test(agc): disable barycentric built-ins in the mesh barycentric emulation test #12062, its fix); the only tests not run are feat(libSceVideoOut): limit the guest display to the host resolution #11948's three, whose executable fails to link on Windows (reported on that PR). The loader inserted VK_LAYER_KHRONOS_validation in every test (998 and 999 insertions), neither log contains was not found but was requested by env var VK_INSTANCE_LAYERS, and both report exactly one VUID, VUID-VkPipelineColorBlendStateCreateInfo-pAttachments-00605 18 times (pre-existing on main, AGC: pipelines use differing per-attachment blend states while independentBlend is never enabled (VUID-VkPipelineColorBlendStateCreateInfo-pAttachments-00605) #11990 / fix(agc): enable independentBlend for color targets with different blend states #6782): this PR adds no new VUID- error.

The tree also carried the open PRs #11948, #12062, #12132, #12070, #11680, #11679, #11677, #11675, #11674, #11605, #11604, #11600, all merged cleanly, so the run is a joint one. Not run on a PS5.

AI-assisted: yes (Claude Code), the runs and the numbers above are from this machine.

[milescoviello]

@maskym thanks for the Windows NVIDIA run with the validation layer. I've added it to Tested.

AI-assisted: yes (Claude Code)