Skip to content

hk: test G17P colour MSAA spill admission - #66

Closed
RyanTheTide wants to merge 1 commit into
feat/g17p-layered-msaa-compressionfrom
feat/g17p-msaa-colour-compression
Closed

RyanTheTide wants to merge 1 commit into
feat/g17p-layered-msaa-compressionfrom
feat/g17p-msaa-colour-compression

Conversation

@RyanTheTide

@RyanTheTide RyanTheTide commented Oct 24, 2026 •

Copy link
Copy Markdown

Small colour MSAA attachments can spill beside wide targets, and stored LOAD attachments or multiview use a narrower tilebuffer allocation. G17P has no in-place decompression kernel for the raw eMRT path, so keep colour MSAA render attachments plain. The existing single-sample UNORM8 compute mip path does not establish a general MSAA conversion.

Add an exhaustive allocator footprint matrix for 1x/2x/4x and both layer shapes under both applicable limits. It checks every attachment in a spilling pass, including resident targets. Expand allocation-policy coverage to all 46 admitted colour formats and add a CPU-only CLI to inspect individual attachment mixes. Allocation logic and the M1/M2 paths are unchanged.

Stacks on #65.

Validation

  • Host GCC and emulated AArch64 SDK policy, native descriptor and metadata-layout suites passed. The policy suite checks 5,879 conditions.
  • The matrix checked 10,290,450 layouts. An unsafe-admission mutation was rejected with NDEBUG enabled.
  • All 224 Gen7/Gen8 descriptors were byte-identical to the base corpus.
  • The shipping AArch64 build passed for Vulkan Asahi, Gallium Asahi/Zink, EGL, GLX and GBM.
  • Target GPU qualification remains pending. Keep this PR in draft until the MSAA render-pass, pipeline and resolve groups and the spilling mixes have run.
  • Meson integration also passed all four registered suites in the AArch64 SDK with the shipping release flags.

Keep colour MSAA attachments plain until eMRT access can preserve their
contents and compression metadata. Even a small attachment can spill next
to wide targets, and stored LOAD attachments and multiview use a narrower
tilebuffer allocation. An image's own footprint cannot prove lifetime
spill safety.

Enumerate every allocator footprint sequence at 1x/2x/4x under both limits
and layer shapes. Require every attachment in a spilling pass to be plain,
including resident targets. Cover all admitted colour formats in the
allocation policy and provide an offline CLI for probing individual mixes.
RyanTheTide added a commit that referenced this pull request Oct 26, 2026
Reuse the public Gen9 layered/MSAA packet and per-level ZLS work from Mesa PRs #65 and #67 on the admitted T6040 profile. Its AGX3B, HAL200, Gen9 descriptor and sampler axes match the native profile established by #55.

Select expanded array/cube/3D colour, transfer-only 2x/4x colour, layered/MSAA depth-stencil and single-sample depth-stencil mip policies by the physical T6040 identity. Preserve the prior Neo policy and Gen10 packet cases. Colour MSAA render attachments stay plain under #66 because a legal MRT mix can spill them; #72 adds unmeasured boundary-copy work and is not silently enabled.

Keep all source-level RT gates unchanged. Retain existing policy regressions and add the public packet, metadata and per-level ZLS checks alongside a separate G16S policy suite. Hardware and performance qualification remain required before merge.

Source: #65
Source: #67
Assisted-by: LLM
Signed-off-by: Ryan Murray <ryan@aurorasilicon.org>
RyanTheTide added a commit that referenced this pull request Oct 26, 2026
Reuse #66 allocation-policy coverage across all46 admitted colour formats, sample/layer counts and input/transient attachment combinations. Transfer-only compressed MSAA must not make any later legal colour-attachment spill depend on an unavailable decompression path.

Source: #66
Assisted-by: LLM
Signed-off-by: Ryan Murray <ryan@aurorasilicon.org>
@RyanTheTide

Copy link
Copy Markdown
Author

Superseded by #87, merged into aurora-wip in the 2026-10-28 consolidation.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant