Skip to content

drm/asahi: g17: four-cluster tiling PTC cap as macOS writes it - #2171

Draft
AxxzyWasTaken wants to merge 5 commits into
aurora-silicon:integration/aurorafrom
AxxzyWasTaken:pr/t6041-tiling-ptc-cap
Draft

AxxzyWasTaken wants to merge 5 commits into
aurora-silicon:integration/aurorafrom
AxxzyWasTaken:pr/t6041-tiling-ptc-cap

Conversation

@AxxzyWasTaken

@AxxzyWasTaken AxxzyWasTaken commented Nov 17, 2026 •

Copy link
Copy Markdown
Contributor

Depends on #2021. This branch sits on #2021's branch, so GitHub shows #2021's four commits first; only the last commit belongs to this PR. I'll rebase it once #2021 lands.

0x0a5a1 carries the tiling PTC cap in bits 32 and up. The 25G83 tiling array writes 0x60 there, the two-cluster M4 Pro value. In my macOS 26.6.2 (25G83) capture on this J616C, taken under a hypervisor while macOS ran a game, the clustered passes have 0x7f, 0x97 or 0xaf there. This gives clustered four-cluster passes 0xaf_0040_0020, the largest of those. TA mode 1, two clusters and other builds keep 0x60.

I don't know how macOS picks among the three values per pass, so taking the largest is my own choice.

Testing

All on a J616C (32-core M4 Max), 25G83:

  • My daily module since 16 Nov, with drm/asahi: g17: tile clustered on the 32-core M4 Max (T6041 25G83) #2021 and other work that isn't in this PR. That build ran Roblox (Vulkan, 11+ minutes, 0 GPU recoveries), Minecraft and, since today's fix for the metadata copies, SCP:SL.
  • Never A/B tested on its own: I have no measurement showing a gain or a fix from it. While I was tracing today's SCP:SL failure, going back to 0x60 failed the same way, so it isn't that failure's cause either.
  • This branch: built (M=drivers/gpu/drm/asahi), not booted as-is. tools/asahi/run-host-tests.py --only g17: 90/90. I couldn't run KUnit (no QEMU here).

Assisted-by: LLM (Anthropic:Claude Opus 5.5)

@RyanTheTide

Copy link
Copy Markdown
Contributor

The production boot stub pins at macOS 27.2 (26B123), so please retarget this to 27.2. The tiling PTC cap here comes from the 26.6.2 (25G83) GPU firmware; please take it from the T6041 GPU image the 27.2 stub loads.

T6041 ran its 25G83 tiling passes in TA mode 1 (0x1c890 = 1), which
tiles non-clustered: one of the four clusters runs the tiling and
vertex work of a pass. Vertex shaders reached a quarter of the ALU rate
of the GPU (2.95 vs 12.3 TFLOPS), and geometry-bound games were capped
by it.

Tile clustered instead, with what a four-cluster pass needs:

- Clustering metadata sized for more than two clusters, as the M2
  Pro/Max path sizes it (meta1 per block of tiles, meta2-4 fixed).
- The four-cluster segment words of cluster_control(): 0x1c890 =
  0x280540, 0x1c918 = 4, and 32 + 16 x clusters = 0x60 free parameter
  buffer pages in 0x1c9a8. The two-cluster capture (0x200440 / 4 /
  0x3c) stays for two clusters.
- A per-build tiler heap normal maximum: the four-cluster 25G83 build
  may use every slot of the block table. A clustered RD32 Minecraft
  frame needs about 1600-2200 blocks; at the 0x45e normal maximum the
  pass never starts and the firmware recovers with reason 0.

25F84 keeps TA mode 1, and every other build keeps its values.

Measured on the J616C (32-core M4 Max, 25G83): vertex-shader ALU 2.95
-> 10-11.5 TFLOPS; Sodium-style terrain 1.93 -> 0.68 ms per frame with
identical images; Minecraft RD32 105 -> ~300 fps, with no GPU
recoveries over several sessions, heavy and concurrent stress included.
G17 host tests 31/31.

Signed-off-by: AxxzyWasTaken <axxzytte@gmail.com>
With four-cluster tiling the 32-core M4 Max runs vertex shaders on every
cluster at once. Roblox (Vulkan) then hangs the GPU about once a minute:
the firmware kills a small depth/stencil render as no-progress (reason
0), with no MMU fault and a nearly empty tiler heap. The same hang showed
up on this Mac with the earlier driver until its render pool had 96
blocks.

Give firmware variants a smallest render pool, 44 blocks as before, and
96 for T6041 25G83. Owners grow to it on their first render, as they grow
for ray-query scratch.

Signed-off-by: AxxzyWasTaken <axxzytte@gmail.com>
…them

macOS 26.6.2 on the 32-core M4 Max (25G83), captured under a hypervisor
while running a game, spaces a clustered pass's tiler metadata 2, 3 and 4
0x1180, 0x1400 and 0x1600 apart and writes 0x5c free parameter-buffer
pages. Metadata 2 was 0xc00, the M2 Pro/Max size, so the tiler could
write past it into metadata 3. Use 0x1180 and 0x5c; metadata 3 and 4 keep
their larger sizes.

Signed-off-by: AxxzyWasTaken <axxzytte@gmail.com>
Four-cluster passes size clustering metadata 1 as the M2 Pro/Max path
does, 0x44 bytes per block of tiles, with 0x80, the two-cluster M4 Pro
size, as the floor. Metadata 1 is the last sub-buffer, and the macOS
26.6.2 capture on the M4 Max does not show its size. Give it a 16 KiB
floor instead.

The per-block size stays the M2 Pro/Max formula; the floor is our own
choice, not a macOS value.

Signed-off-by: AxxzyWasTaken <axxzytte@gmail.com>
0x0a5a1 carries the tiling PTC cap in bits 32 and up. The 25G83 array
wrote 0x60, the two-cluster M4 Pro value. macOS 26.6.2 on the 32-core
M4 Max writes 0x7f, 0x97 or 0xaf for its passes. Give clustered
four-cluster passes 0xaf, the largest of those.

Signed-off-by: AxxzyWasTaken <axxzytte@gmail.com>
@AxxzyWasTaken
AxxzyWasTaken force-pushed the pr/t6041-tiling-ptc-cap branch from 7da6f6f to 845bd7f Compare November 21, 2026 09:48
@RyanTheTide

Copy link
Copy Markdown
Contributor

Blockers: 2.

  1. drivers/gpu/drm/asahi/g17/fw/render.rs:708 and :802: 0xaf is picked as "the largest of three values seen", and the PR says it was never A/B tested and showed no gain and no fix (going back to 0x60 failed the same way). That makes it a chosen value with no measured effect on a register whose meaning isn't known. Either find what the per-pass value depends on and compute it from that (the capture already has three values per pass to correlate against the pass geometry and cluster count), or drop this commit until a same-Mac A/B shows it matters.
  2. drivers/gpu/drm/asahi/g17/fw/render.rs:708: the constant hardcodes the low 32 bits (0x0040_0020) a second time instead of taking them from UNK_0A599 (g17/fw/mod.rs:102). If the shared value ever changes, clustered four-cluster passes silently keep the old low bits. Build it as (UNK_0A599 & 0xffff_ffff) | (0xaf << 32), or give the cap field a named shift.

Also checked: multi (render.rs:760) is only Some for non-TA-mode-1 passes with more than two clusters, so the M4 Pro and the TA-mode-1 builds keep 0x60. The branch sits 178 commits behind integration/aurora on top of #2021, which is still open. A test merge with the current base takes the 26B123 entries cleanly, and there T6041 keeps ta_mode_1: true (hw/t6041.rs, BUILD_26B123), so this override never runs on that firmware. CI host-tests pass; the commit has a human Signed-off-by and the body ends with an Assisted-by line.

HW row: J616C (M4 Max), this commit rebased on the landed #2021 and booted as-is, same-Mac A/B with 0x60 against the new value on the firmware the boot stub loads: Roblox 10+ min, Minecraft RD32 fps, 0 GPU recoveries, KUnit asahi_g17_ta_mode passing, 0 AVC, 0 failed units, 0 new kernel errors.

Assisted-by: LLM (Claude:Opus 5.5)

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.

2 participants