Repository navigation
hk: add bounded G17P layered and MSAA compression - #65
Closed
RyanTheTide wants to merge 1 commit into
Closed
RyanTheTide wants to merge 1 commit into
RyanTheTide wants to merge 1 commit into
Conversation
Accept GPU-tiled lossless PBE packets for arrays, 3D slices and 2x/4x MSAA. Enable single-level array, cube and 3D colour images, along with layered and multisample depth/stencil images. Use six faces for the G17P cube host-copy performance query's representative image. Multisample colour images used only for transfers and sampling may also compress. Keep colour MSAA attachments plain because eMRT can spill them and G17P has no in-place decompression kernel. Keep layered colour mip chains plain because the compressed secondary stride aliases the mip fields. Ordinary 2D mip policy and AGX2 descriptors are unchanged. Add offline descriptor and metadata layout checks, including address and layer bounds, sample expansion, cube face slots and 3D metadata reserves. Eight samples and nonzero lossless codec/footprint selectors stay closed. Assisted-by: LLM Signed-off-by: Ryan Murray <ryan@aurorasilicon.org>
This was referenced Oct 24, 2026
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>
Author
|
Superseded by #87, merged into aurora-wip in the 2026-10-28 consolidation. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This adds Gen9 lossless PBE packing for GPU-tiled arrays, 3D slices and 2x/4x MSAA. G17P can now compress single-level colour arrays, cubes and 3D images, layered depth/stencil images, and 2x/4x depth/stencil images. Multisample colour images used only for transfers and sampling can also compress. Cube host-copy performance queries use a six-face representative image so they report the compressed device layout correctly.
This is a bounded part of the layered/MSAA work. Colour MSAA attachments remain plain because legal render passes can spill them to raw eMRT storage, which needs a G17P decompression path. Layered colour mip chains remain plain because compressed PBE secondary-stride fields overlap the mip fields. The existing ordinary 2D mip policy is preserved. Eight samples and nonzero lossless codec/footprint selectors remain disabled.
Offline validation passes for the compression policy, descriptor packing/unpacking, and metadata allocation. The layout checks cover sample expansion, array and cube face slots, 3D metadata reserves, mip offsets and page alignment. A base-versus-candidate corpus has identical bytes for 224 Gen7/Gen8 descriptors.
The shipping AArch64 build passes with
buildtype=plain,b_ndebug=true, Asahi/Zink Gallium and Asahi Vulkan. The policy, descriptor and metadata tests also pass in the emulated AArch64 SDK. A process-local overlay with Vulkan and Zink launch wrappers is ready. This stays a draft until target testing is available. There are no performance results yet.The hardware test plan is to compare against the base on the same boot using a process-local driver overlay. Run Vulkan image tests, copy/blit tests, renderpass and pipeline MSAA groups, cube/array texture groups, and Zink GL samples. Check selected layers, cube faces, 3D slices, clears, copies and resolves, then measure transfer and sampling workloads with repeated before/after runs. Do not merge before these checks pass.