Repository navigation
Conversation
|
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.
|
Current dependency/composition status, 2026-10-23@MarcZX10 The remaining sampled-layout change from your original The maintained root qualification now records 77 passes across five exact composed heads and literal-main reference checks on RTX5080/Radeon Graphics, with no validation warnings/errors/VUIDs, skips or timeouts. This validates the adapted stack, not your unchanged standalone head or Intel behavior. #8223's requested retirement work remains separate and first in its agreed order. Your last status says this branch is intentionally unchanged to avoid duplicating the composed work, while #6973 still declares #6971 first. Could you and the maintainers confirm the intended landing plan: merge standalone #6971 first, or land its attributed commits through #6973 and close this as superseded? The tested fork adaptation above is available for either plan; I have not changed the declared order. AI-assisted validation/reporting: yes (Codex); author responsible. Earlier coordination at its original pins@MarcZX100 Follow-up to the overlap coordination: #6973's published composition 82bfd5250 now contains exact #6635 ebc6ea3 and your #6971 f65cecd, with authorship preserved. The maintained validation report records the initial-upload/regular-and-array-sampling, attachment/storage reuse and retirement checks on NVIDIA, AMD and llvmpipe, including the recorder's separate #6225 adapter-selection composition. The matched initializer-removal control fails on all three devices; restoration passes. Four focused llvmpipe runs with Khronos 1.4.363.0 are free of VUIDs/validation errors. This validates the adapted composition; it does not retest your unchanged standalone head or establish the Intel UHD 770 fix. The sequence remains #6635, then #6971, then #6973. No new competing layout PR opened. AI-assisted: Codex; author responsible. Landing clarification received,2026-10-23: oneandonlydean confirmed that #6971 is carried by main plus the remaining #6973 stack and should close after #6973 merges. I have removed #6971 as a separate pending prerequisite in our current descriptions, retaining its attributed commits and the existing source-specific qualification. #8223-first and the requested owner-retirement publication gate remain. No source or foreign branch was changed. AI-assisted maintenance: yes(Codex). |
|
Thanks for confirming the composition and preserved authorship. I verified that #6973 carries the #6971 commits, while #6635 and #6973 remain open. I will keep #6971 in draft pending the requested #6635 order and the composition review. The reported validation is useful for the combined changes, but does not establish the Intel UHD 770 result for #6957. |
|
UHD 770 result, as asked in #6957: this fixes Windows 11 Pro 26300, WinLibs GCC 15.2.0 r7, Release, every execution test built, Tests: the 14 multisample tests on
Full CTest with this PR,
AI-assisted: yes (Claude Code) |
|
Retest on the affected GPU (#6957): Intel UHD Graphics 770 (driver 32.0.101.6129), with an RTX 4090 (NVIDIA 616.92) in the same machine for comparison. Windows 11, WinLibs GCC 15.2, Release. This PR merged onto main
Multisample-related group (
So the read-only layout transition fixes the Intel failure, with nothing newly broken in that group on either GPU. AI-assisted (Claude Code). |
|
Thanks for the two independent UHD 770 results. Both show the reported AI-assisted: yes (Codex); reports reviewed. |
Retain original boykopovar#6971 f65cecd ancestry. Adapt its layout tracking to current shared ownership, device checks, CMASK handling and array/swapped views for focused dependency validation.
Consolidate the previously published initializer and its fault/regression coverage onto the refreshed layout/uploader foundations and the credited boykopovar#6971 dependency composition. Preserve the full snapshot and supported uniform/nonuniform paths, restrict borrowed sampled-view usage and keep current main ownership/device/CMASK handling. Original initializer and test work is preserved at published8068 and its archived history; boykopovar#6971 f65cecd remains an original ancestor. Written with AI assistance.
|
Status update: #6635 is merged, and open #6973 now contains the multisampled-array support and the read-only sampled-layout commits from this PR, with their authorship preserved. Its current head |
Consolidate the previously published initializer and its fault/regression coverage onto the refreshed layout/uploader foundations and the credited boykopovar#6971 dependency composition. Preserve the full snapshot and supported uniform/nonuniform paths, restrict borrowed sampled-view usage and keep current main ownership/device/CMASK handling. Original initializer and test work is preserved at published8068 and its archived history; boykopovar#6971 f65cecd remains an original ancestor. Written with AI assistance.
What
Fixes #6957 by tracking a borrowed multisampled target's current image layout. After storage writes, sampled reads transition the image from
GENERALtoSHADER_READ_ONLY_OPTIMAL; before later attachment or storage use, it transitions back toGENERAL. This targets the Intel UHD 770 behavior reported in the issue while leaving other image access paths intact.The sampled view keeps the requested
VkImageViewTypein its cache key and supports a restricted view usage, so this change composes with #6635's multisampled array views and #6683's sample-order storage view. Their sample-order pass writes through the tracked storage path, so the next sampled read uses the read-only transition. The author of #6635 requested that it land first; this PR remains draft until then. New PR #6973 also changesMultisampleTarget.cppand the sampled view path to initialize native multisample color contents. I have flagged the overlap there and will verify that the layout tracking preserves its initial upload before marking this PR ready.Tested
agc_driver_graphics_tests,agc_driver_recorder_tests, andagc_driver_sample_multisample_target_tests. GCC 12 required a temporarytemplatekeyword workaround in unrelatedSrtDescriptorEvaluation.cpp; I restored that file, and the workaround is not part of this PR.MultisampleTarget.cpp,Texture.cpp,SampleOrder.cpp, andRecorder.cpparray-view regression, and linked the graphics, recorder, multisample-target, and sample-order test executables.agc_driver_sample_multisample_target,agc_multisample_guest_access,agc_driver_sample_order, andagc_driver_sample_order_planningaborts before assertions on this host because its Xeon E5-1620 v2 lacks LZCNT, BMI1/2, MOVBE, AVX2, and FMA. No runtime or Intel UHD 770 result is claimed.python3 tools/check_conventions.py --base upstream/mainandgit diff --checkpass.Checklist
main; no duplicate PR implements this layout fix