[thuanquang]

What

agc_driver_gpu_map_on_fiber fails on Windows on current main with a Vulkan device:

AGC graphics: GPU mapping on a fiber started 1 host threads for its import instead of importing on the fiber's host stack

The test runs its fiber on a 64 KiB context. Since #6003 a fiber gets a host stack only when its context is smaller than 64 KiB (size_context < FIBER_GUEST_STACK_MIN), so this fiber now runs on the context itself on Windows too. StackHoldsImport finds less than 256 KiB of stack and the driver imports the mapping on a host thread, which is correct; the Windows-only expectation of no thread came from 8136438, when every Windows fiber had a host stack.

The test now maps on two fibers:

• a 16 KiB fiber, which keeps a host stack: on Windows the import runs on it with no thread, as 8136438 checked; elsewhere one thread, as before;
• the 64 KiB fiber, which runs on its context: one import thread on every platform.

Both still check that the import used at most 8 KiB of the fiber's context, and the mapping on an ordinary thread still starts no thread.

CI does not show the failure: in run 38993324986 (main f643e61d9, after #6003) Windows builds the test as ExecutionTestNotBuilt.cpp and skips it, and Linux, which already expected one thread, passes.

Tested

Windows 11 26200, AMD Radeon RX 6700 XT (driver 32.0.21045.5002), WinLibs GCC 15.2.0 r7, Release, BUILD_TESTING=ON, ANYPS5_REQUIRE_VULKAN=1, on main 9bf01db66:

• main: fails 3/3 with the line above.
• This change: passes 5/5 (A GPU mapping made on a fiber used 0 bytes of its context with a host stack and 3104 running on the context); fiber passes.
• This change with fix(libSceFiber): run a fiber on its own context when the context is large #6003's size_context < FIBER_GUEST_STACK_MIN condition removed locally: fails with GPU mapping on a fiber running on its context started 0 host threads for its import instead of one, so the test follows fix(libSceFiber): run a fiber on its own context when the context is large #6003's limit.
• Full ctest --parallel 4 --timeout 120 on main: 1352 tests, 3 fail: this one, agc_driver_lds_mixed_width_atomics (agc_* tests: 3 of 483 fail on AMD Radeon RX 6400, pass on NVIDIA RTX 4080 SUPER #5271) and agc_driver_layered_volume_vertex_no_gpl (AMD Windows: GPL-mode cache failures (alpha/wave32 pass after local driver update) #4768, fixed by fix(agc): key shader modules by the pipeline library mode #6436).
• python tools/check_conventions.py --base origin/main with this title and body passes.
• Not tested: Linux and macOS.

Checklist

• Based on current main; no other open PR implements the same functions
• One topic per PR; follow-ups go in a new PR
• No comments in code except technical debt
• Unimplemented paths throw (NotImplemented_nid_no_patch); silent stubs are listed under Silent stubs in their library's technical debt file
• No notes, investigation .md files or images added to the repository (attach them to this PR)
• New third-party code is a submodule built from source
• Generic behaviour, not specific to one title
• Depends on: none
• Unblocks: none
• GPUs tested: AMD Radeon RX 6700 XT, driver 32.0.21045.5002, Windows 11 26200
• AI-assisted: yes (Claude Code)

🤖 Generated with Claude Code

[johnmaia]

Confirmed on NVIDIA and Intel too: i7-13700K, RTX 4080 (617.42) and Intel UHD 770 (101.6129), Windows, WinLibs GCC 15.2.0, Release, ANYPS5_REQUIRE_VULKAN=1.

• main f643e61: agc_driver_gpu_map_on_fiber fails on both GPUs with the same message ("started 1 host threads for its import instead of importing on the fiber's host stack").
• This PR merged onto that main: passes 3/3 on each GPU.

AI-assisted: yes (Claude Code)