Skip to content

media: apple: prores: T8132 (M4) engine, gated by its initialization file - #1994

Merged
RyanTheTide merged 5 commits into
aurora-silicon:integration/aurorafrom
IronsideXXVI:j604-prores-t8132
Nov 22, 2026
Merged

RyanTheTide merged 5 commits into
aurora-silicon:integration/aurorafrom
IronsideXXVI:j604-prores-t8132

Conversation

@IronsideXXVI

@IronsideXXVI IronsideXXVI commented Nov 15, 2026 •

Copy link
Copy Markdown
Contributor

What

Adds the ProRes engine of the T8132 (M4) to the Apple ProRes driver, and enables its node on the J604 (MacBook Pro 14", M4).

apple,t8132-prores is in apr_driver's match table, next to the other engines. There is no module parameter and no second platform driver (reworked 2026-11-20 after review). The per-machine initialization file is the gate: without apple/prores-t8132-init.bin the engine stays unbound, as the driver already does for an engine without its rows.

Commits (rebased onto integration/aurora acfcd08)

  1. dt-bindings: media: apple,prores: Name the T8132 initialization table: T8132 takes a per-machine initialization table.
  2. media: apple: prores: Add the T8132 (M4) engine: T6041's register windows and job format, with a 48-row wrapper table instead of 62. Traced under the hypervisor on a J604: 82 SCODEC-system rows, 80 core rows and 48 wrapper rows, the same in two sessions. The rows come from apple/prores-t8132-init.bin (2,560 bytes); T8132 has no built-in rows.
  3. arm64: dts: apple: t8132: Name the ProRes windows; enable ProRes on the J604: reg-names for the three windows; &apr and &dart_apr0 enabled in t8132-j604.dts.

Testing

  • On acfcd08: the ProRes directory builds with W=1, with no warnings and no errors; checkpatch --strict is clean on all three commits; the ProRes KUnit suite passes 19 of 19 under QEMU (18 of 18 at the base, plus the T8132 container case).
  • DTBs: 138 Apple DTBs built at the base and with this branch. 132 are byte-identical. The other five T8132 boards (J623, J624, J713, J715, J773g) gain only reg-names on the ProRes node, and the J604 also has its ProRes and DART nodes enabled.
  • On the J604 the T8132 engine runs jobs in our own kernel series, which binds it by default (MEASURED; our series has the same engine variant but takes its rows from a built-in table rather than the file): 30 frames of 4K ProRes 422 in 85 ms on 1.930, and 518 % of macOS's encode rate in our 1.939 measurement.

Question for review

  • The initialization table is a per-machine file (apple/prores-t8132-init.bin) that the driver does not ship. Where should users get it: a firmware extraction step in the installer, or something else the project prefers?

Assisted-by: LLM (Claude:Opus 5.5, Claude:Fable 5.2)

@RyanTheTide

Copy link
Copy Markdown
Contributor

Thanks for bringing up the M4 ProRes engine and documenting the init-table layout.

One request. The T8132 engine binds only when a new module parameter is set (drivers/media/platform/apple/prores/apple-prores-drv.c:1688-1690, t8132, default off), through a second platform driver with its own match table (:1692-1719) next to the existing apr_driver. We try not to need a boot option to make hardware work, and to avoid a second copy of a driver registration for one chip.

The driver already refuses T8132 without its apple/prores-t8132-init.bin file, so that file is the gate. Could you add { .compatible = "apple,t8132-prores", .data = &apr_t8132 } to the main apr_driver match table and drop the parameter and the extra driver? If the engine's file is missing it stays unbound, as it does today.

The branch also needs a rebase onto the current base. We can run a no-regression ProRes row on our M1 Pro and M4 Pro afterwards.

@IronsideXXVI IronsideXXVI changed the title media: apple: prores: T8132 (M4) engine, bound only on request media: apple: prores: T8132 (M4) engine, gated by its initialization file Nov 20, 2026
@IronsideXXVI

Copy link
Copy Markdown
Contributor Author

@RyanTheTide done, and rebased onto integration/aurora acfcd08 (head 5ead177):

  • { .compatible = "apple,t8132-prores", .data = &apr_t8132 } is in apr_driver's table.
  • The t8132 parameter, the second platform driver and the custom init/exit are gone. That commit was dropped completely.
  • Without apple/prores-t8132-init.bin the engine stays unbound.

Checks: the ProRes KUnit suite passes 19/19 under QEMU (18 at the base, plus the T8132 container case). W=1 builds with no warnings. checkpatch --strict is clean. Of the 138 Apple DTBs, 132 are byte-identical. The other T8132 boards gain only reg-names on the ProRes node.

A no-regression row on your M1 Pro and M4 Pro would be great, thank you.

@RyanTheTide

Copy link
Copy Markdown
Contributor

Thanks for 5ead177. With apple,t8132-prores in apr_driver's table, the module parameter and the second driver gone, and the init file as the only gate, this matches what we asked for. The DTB check (132 of 138 identical, reg-names only on the rest) is helpful.

We will run the no-regression rows on our M1 Pro and M4 Pro on 5ead177: the engine binds, and an encode compares with the base. We'll post them here.

@IronsideXXVI

Copy link
Copy Markdown
Contributor Author

@RyanTheTide thanks, and thanks for running the no-regression rows on the M1 Pro and M4 Pro. I'll look at them when they're posted.

@DjDeveloperr DjDeveloperr left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

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

Thanks for the rework. This is what was asked for in the earlier review: the T8132 entry in apr_driver's table, no parameter and no second driver, and the init file as the only gate. The variant mirrors T6041's windows and limits, the 48-row wrapper count is sourced from the J604 trace, and the KUnit container case covers the size, the SoC field and the window limit. Other engines don't change, apart from reg-names on the T8132 node.

Two small things before it can go in:

  1. #1919 merged a few minutes ago, and it also appends a block at the end of t8132-j604.dts, so this now conflicts there. Could you rebase onto integration/aurora and keep both blocks?
  2. While you're there: the DT comment still says the driver binds the T8132 engine "only on request". With the parameter gone, it binds whenever apple/prores-t8132-init.bin is installed.

On your question about where users get the file: it's the same situation as the T6041 and T6050 tables, and packaging it is a separate change. It doesn't hold this PR.

With the rebase this is ready to merge. The M1 Pro and M4 Pro no-regression rows Ryan offered can follow the merge.

T8132 takes a per-machine initialization table like T6041 and T6050.
Name it in the description.

Signed-off-by: Dylan Ironside <dylan@dylanironside.io>
The M4's ProRes engine has T6041's register windows (core, AXI wrapper
and SCODEC system) and job format. macOS brings it up as it does T6041's,
with read-modify-write tables applied before the ring is installed, but
its wrapper table has 48 rows instead of 62 (traced on a J604 under the
hypervisor: 82 SCODEC-system, 80 core and 48 wrapper rows, the same in
two sessions, all below +0x4040 in the wrapper window).

Add the T8132 variant on the T6041 path. Its rows come from
apple/prores-t8132-init.bin, the T6041 container with soc 0x8132 and a
48-entry wrapper block (2,560 bytes). There are no built-in rows, so
without the file the engine stays unbound.

Signed-off-by: Dylan Ironside <dylan@dylanironside.io>
…he J604

The driver maps the ProRes windows by name. Name the T8132 node's three
windows, and enable the engine and its DART on the J604. The engine
binds only on request and with its per-machine initialization table
installed.

Signed-off-by: Dylan Ironside <dylan@dylanironside.io>
@IronsideXXVI

Copy link
Copy Markdown
Contributor Author

Thanks. I'll rebase onto integration/aurora keeping both blocks in t8132-j604.dts, and fix the DT comment so it says the driver binds whenever apple/prores-t8132-init.bin is installed.

@IronsideXXVI

Copy link
Copy Markdown
Contributor Author

Rebased onto integration/aurora 5b596e6 to clear the merge conflict (head eb2de51). No other change: the only conflict was the new keyboard backlight block next to the ProRes nodes in t8132-j604.dts, both kept. Checkpatch, the host tests, the DT checks and a W=1 build of the ProRes driver give the same results as on the base.

@RyanTheTide

Copy link
Copy Markdown
Contributor

Thanks for the rebase. One leftover from the last review: the comment at arch/arm64/boot/dts/apple/t8132-j604.dts:635 still says the driver binds the T8132 engine "only on request". With the parameter gone it binds whenever apple/prores-t8132-init.bin is installed, so that line could say so. Everything else looks ready from our side, and the M1 Pro and M4 Pro no-regression rows can follow the merge.

The T8132 ProRes engine no longer waits for a module parameter: the
driver binds it whenever apple/prores-t8132-init.bin is installed and
leaves the engine off without it. Say so in the J604 comment.

Signed-off-by: Dylan Ironside <dylan@dylanironside.io>
@IronsideXXVI

Copy link
Copy Markdown
Contributor Author

Thanks, fixed in 3c27dbd: the comment at t8132-j604.dts:635 now says the driver binds the T8132 engine whenever apple/prores-t8132-init.bin is installed and leaves it off without it. Comment only, one commit on top; the J604 dtb is byte-identical and its dtbs_check output is unchanged.

@RyanTheTide

Copy link
Copy Markdown
Contributor

Thanks, 3c27dbd reads right: the J604 comment now says the engine binds whenever apple/prores-t8132-init.bin is installed and stays off without it, which is what the driver does. Comment only, with a byte-identical DTB, so nothing else to re-check. Nothing more from our side; the M1 Pro and M4 Pro no-regression rows will follow the merge.

@RyanTheTide

Copy link
Copy Markdown
Contributor

Merge-train note: your #1895 (cs42l84 init) landed in batch 1 (integration/aurora 606d37e) and touches the same arch/arm64/boot/dts/apple/t8132-j604.dts lines, so this one is conflicting now. Could you merge integration/aurora into the branch? After that it needs a re-approve at the new head (Dj's approval is on 5ead177), then it can go in.

integration/aurora now carries the J604 headset jack node override
(cirrus,ring-sense-detect) at the end of t8132-j604.dts, the same spot
where this branch adds the ProRes nodes.

Conflict in arch/arm64/boot/dts/apple/t8132-j604.dts: both sides append
a block at the end of the file. Keep both, the jack block first and the
ProRes block after it. Nothing else conflicts.

Signed-off-by: Dylan Ironside <dylan@dylanironside.io>
@IronsideXXVI

Copy link
Copy Markdown
Contributor Author

Merged integration/aurora (606d37e) into the branch; head is now 2a68afa.

The only conflict was in t8132-j604.dts, where the jack block from #1895 and the ProRes block landed in the same spot. I kept both, jack first, then ProRes. The diff against base is the same as before: 9 files, +111/-7.

Rebuilt after the merge: the dtb builds, and the driver plus the kunit test build with W=1 and no warnings. GitHub shows it mergeable again, so it only needs Dj's re-approve.

@RyanTheTide RyanTheTide left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

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

Re-checked at 2a68afa. The merge's only resolution is in t8132-j604.dts, and it keeps both blocks unchanged: the &jack_codec ring-sense block from #1895, then the ProRes comment and &apr. Your four commits are the ones reviewed before (the comment fix included), and the diff against the base is the same 9 files. Approving at this head.

Assisted-by: LLM (Claude:Opus 5.5)

@RyanTheTide
RyanTheTide merged commit fa8b339 into aurora-silicon:integration/aurora Nov 22, 2026
8 checks passed
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