Skip to content

asahi: admit the measured 30-core T6034 G15C contract - #33

Merged
maralcbr merged 1 commit into
integration/aurorafrom
carry/j514m-mesa
Nov 7, 2026
Merged

maralcbr merged 1 commit into
integration/aurorafrom
carry/j514m-mesa

Conversation

@maralcbr

@maralcbr maralcbr commented Nov 7, 2026

Copy link
Copy Markdown
Member

Summary

Carries Chris Kearney's mesa#32 unchanged, with my sign-off. Chris has left the project. Replaces #32. Its earlier CI run was cancelled after 35 minutes, not failed.

 agx_params_driver_supported()
   case AGX_CHIP_G15C:
-    chip_id == 0x6031
+    chip_id == 0x6031 || agx_params_t6034_supported()
       (0x6034, revision B1 0x11, 10 slots per cluster, 4 non-empty 10-bit masks, exactly 30 cores)
     && 1 die && 4 clusters && USC AGX3 && legacy HAL

J514M's binned M3 Max reports chip 0x6034, which today's T6031-only check rejects. Without this, Hyprland finds no monitor on a cold boot once the J514M kernel and m1n1 hand the GPU over. The T6031 admission is unchanged.

Pairs with omacom/m1n1#32 and the J514M kernel change. Ship order: this Mesa must reach J514M users no later than that kernel and m1n1.

Evidence

  • Host test (asahi-chip-admission):
    • The measured masks [0x1fd, 0x1fd, 0x1ed, 0x1ed] are admitted.
    • Rejected: a wrong chip, revision or slot count; 3 clusters; 2 dies; USC 2; HAL 200; a bit beyond slot 10; an empty mask; 29 or 31 cores.
  • Package build (Chris, b1105-160343-6ea5): all 129 Meson tests pass.
  • Hardware (Scott's lab, Faeric's J514M, j1107-054455-d032, this Mesa staged with the J514M kernel and m1n1): Hyprland on the GPU at eDP-1 3024x1964@120, Vulkan 1389 dispatches with 0 bad.
  • Audit: no Apple code.

Merge Danger

Door: two-way

Blast Radius: J514M only

Only chip 0x6034 with this exact topology is newly admitted. No other GPU behaves differently.

The J514M T6034 reports G15C with four physical ten-slot clusters and 30 fused cores. Admit that exact chip and B1 revision with nonempty ten-bit masks totaling 30 cores, one die, USC3 and the legacy HAL. Preserve the existing T6031 path and all native kernel ABI requirements.

The chip-admission host test checks the measured masks and rejects wrong chip, revision, slots, clusters, dies, ISA, HAL, empty/out-of-range masks and 29/31-core layouts. This is source and host admission evidence; hardware correctness still requires the candidate package on the affected Mac.

Generated-by: LLM

Assisted-by: Codex:GPT-6
Signed-off-by: Marcelo Alcantara <maralc@gmail.com>
@greptile-apps

greptile-apps Bot commented Nov 7, 2026 •

Copy link
Copy Markdown

RetriggerView in GreptileConfidence Score: 5/5

[Medium impact] The PR appears safe to merge; T6034 uses the existing G15C path with narrow admission checks.

Summary

Adds support for the measured 30-core T6034 B1 GPU through the existing G15C path.

  • Mesa now opens the measured 30-core T6034 GPU.

Reviews (1) · Last reviewed commit: "asahi: admit the measured 30-core T6034 ..." · Reviewed by Greptile

@scottjones scottjones left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Same patch as #32 (identical patch-id), which we staged for the J514M runs today: Hyprland on the GPU at 3024x1964@120, Vulkan 1389 dispatches with 0 bad.

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.

3 participants