From: "Juha-Pekka Heikkilä" <juhapekka.heikkila@gmail.com>
To: Luca Coelho <luca@coelho.fi>, igt-dev@lists.freedesktop.org
Subject: Re: [PATCH i-g-t 10/10] tests/kms_plane_scaling: remove legacy subtests covered by max-crc tests
Date: Tue, 10 Nov 2026 12:55:41 +0200 [thread overview]
Message-ID: <3a6a5381-6f7d-4e72-af57-d220f68d4574@gmail.com> (raw)
In-Reply-To: <70a6b6cc10b5056679b60c5ee9d6c99e089c79b6.camel@coelho.fi>
On 10/11/2026 10.02, Luca Coelho wrote:
> On Mon, 2026-11-09 at 20:07 +0200, Juha-Pekka Heikkilä wrote:
>>
>> On 09/11/2026 11.34, Luca Coelho wrote:
>>> On Wed, 2026-10-28 at 16:52 +0200, Juha-Pekka Heikkila wrote:
>>>> The plane-*-max-crc subtests search scaling limits kernel actually accepts
>>>> for each pixel format, modifier and rotation, on all pipes, for primary
>>>> and overlay planes alone and together, and verify result with crc. Legacy
>>>> subtests below only checked that commit with fixed scaling factor was
>>>> accepted, skipping on -EINVAL or -ERANGE, so they never
>>>> verified the scaled output.
>>>>
>>>> The first three removed groups each had six variants: upscale-20x20,
>>>> upscale-factor-0-25, downscale-factor-0-25, downscale-factor-0-5,
>>>> downscale-factor-0-75 and scaler-unity-scaling.
>>>>
>>>> plane-*-with-pixel-format
>>>> Covered by plane-{up/down}scale-max-crc-linear, which tests every
>>>> format at largest accepted factor and checks the crc.
>>>> Unity scaling is reference commit of each max-crc step.
>>>>
>>>> plane-*-with-modifiers
>>>> Covered by per modifier plane-{up/down}scale-max-crc-* subtests.
>>>>
>>>> plane-*-with-rotation
>>>> Linear XRGB8888 walking rotations 0, 90, 180, 270 and returning on the
>>>> first rejected commit. Where 90 is advertised it is rejected with
>>>> linear, so only rotation 0 was ever exercised; Covered by
>>>> plane-{up/down}scale-max-crc-* and plane-{up/down}scale-max-crc-rotation-*
>>>>
>>>> planes-* (24 subtests)
>>>> Covered by the two plane steps of the max-crc subtests on pipes with
>>>> at least two scalers: both planes unscaled as the crc reference, both
>>>> planes at their limits at the same time, and the primary at its limit
>>>> with the overlay scaled 2x in the opposite direction.
>>>>
>>>> Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
>>>> ---
>>>
>>> When looking at this again, I had a feeling some of the tests were
>>> still missing, so I ran it through Claude and it seems that there are
>>> still a bunch of tests dropped, or with reduced coverage. But instead
>>> of just copying its comments here, I think it's more efficient if you
>>> run these changes through an LLM to compare the previous coverage with
>>> the current one.
>>>
>>> Also, it would be nice to have a cover-letter with a small changelog on
>>> what changed between one patchset version and the next, so it's easy to
>>> check what has changed without going over the patches again.
>>>
>>
>> Hi Luca,
>>
>> not sure what was missed. Here's a comparison of old and new coverage.
>>
>> Covered now with crc instead of commit-only: every format, modifier and
>> rotation at the limit; fixed factors (the search probes the whole range
>> with TEST_ONLY and crc-checks the extreme); unity (the reference commit
>> of every step); two planes in the same and in opposite directions. The
>> old rotation tests in reality only ran rotation 0 and nothing else.
>>
>> Differences that are intentional:
>>
>> - Width and height are scaled separately. Each axis has its own limit
>> and testing them one at a time keeps the framebuffers small as well as
>> tells which axis broke.
>> - Only primary and first overlay are used. Scalers are pipe resource and
>> get assigned to whichever plane needs one, so more overlays would run
>> the same scalers again. Per-plane format/modifier support is checked
>> before the overlay is used.
>> - Native mode only instead of walking connector modes until one is
>> accepted. The test is about scaler limits and not about mode selection.
>> - Pipe crc is required. Without crc old tests only showed commit was
>> accepted, which the TEST_ONLY still cover.
>> - Intermediate factors are only probed with TEST_ONLY. Extreme factor is
>> the one that stresses the scaler.
>>
>> If something else is missing that isn't covered by the above, please
>> point it out and I'll take a second look.
>
> This is, very shortly, what Claude found:
>
> * Other plane pairs: these still aren't covered, which JP agreed to on
> v1, yet the commit message says planes-* is covered.
you're right, the commit message overstates it. Both scalers are pipe
resources, so pairs other than primary + first overlay use the same two
scalers. I'll reword the message to say only primary + first overlay are
tested and why.
>
> * Full-screen scaling: two full-screen overlapping planes became two
> 64-pixel strips side by side, so the expensive bandwidth case isn't
> exercised anymore.
Bands still reach the peak pixel rate since that depends on the scale
ratio and not on the height. What they don't stress is memory bandwidth,
which isn't a scaler property; kms_bw covers that case. I'd rather keep
this test about scaler correctness.
>
> * Fixed factors: 0.25, 0.5 and 0.75 used to be committed for real. Now
> only the limit is, and the binary search only uses TEST_ONLY commits,
> which never reach the hardware.
True, TEST_ONLY probes don't reach hardware. Those old fixed-factor
commits were only displayed, never verified: without crc they could only
catch kernel warning or underrun. The max-crc tests do a real commit at
the limit which is the most demanding factor and check output with crc.
Real commit at smaller factor wouldn't add any check that the limit
commit doesn't already do.
>
> I'm not sure these make sense, so I'll leave them for you to comment.
> In any case, I don't want to block this, so if you think this is all
> fine, you have my:
>
> Reviewed-by: Luca Coelho <luciano.coelho@intel.com>
Thanks!
/Juha-Pekka
next prev parent reply other threads:[~2026-11-10 10:57 UTC|newest]
Thread overview: 25+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-10-28 14:51 [PATCH i-g-t 01/10] lib/igt_fb: add igt_format_get_subsampling() function Juha-Pekka Heikkila
2026-10-28 14:51 ` [PATCH i-g-t 02/10] tests/kms_plane_scaling: add search for scaling limit Juha-Pekka Heikkila
2026-10-28 14:51 ` [PATCH i-g-t 03/10] tests/kms_plane_scaling: add plane-downscale-max-crc subtest Juha-Pekka Heikkila
2026-11-08 12:56 ` Luca Coelho
2026-10-28 14:51 ` [PATCH i-g-t 04/10] tests/kms_plane_scaling: change plane-downscale-max-crc to do variations Juha-Pekka Heikkila
2026-10-28 14:51 ` [PATCH i-g-t 05/10] tests/kms_plane_scaling: add plane-upscale-max-crc subtest Juha-Pekka Heikkila
2026-10-28 14:51 ` [PATCH i-g-t 06/10] tests/kms_plane_scaling: run plane-*-max-crc on all pipes Juha-Pekka Heikkila
2026-11-08 13:30 ` Luca Coelho
2026-10-28 14:51 ` [PATCH i-g-t 07/10] tests/kms_plane_scaling: add rotated plane-*-max-crc subtests Juha-Pekka Heikkila
2026-10-28 14:52 ` [PATCH i-g-t 08/10] tests/kms_plane_scaling: test overlay plane in plane-*-max-crc Juha-Pekka Heikkila
2026-11-08 13:38 ` Luca Coelho
2026-10-28 14:52 ` [PATCH i-g-t 09/10] tests/kms_plane_scaling: scale overlay opposite to primary in max-crc Juha-Pekka Heikkila
2026-11-08 13:39 ` Luca Coelho
2026-10-28 14:52 ` [PATCH i-g-t 10/10] tests/kms_plane_scaling: remove legacy subtests covered by max-crc tests Juha-Pekka Heikkila
2026-11-09 9:34 ` Luca Coelho
2026-11-09 18:07 ` Juha-Pekka Heikkilä
2026-11-10 8:02 ` Luca Coelho
2026-11-10 10:55 ` Juha-Pekka Heikkilä [this message]
2026-10-28 22:46 ` ✓ Xe.CI.BAT: success for series starting with [i-g-t,01/10] lib/igt_fb: add igt_format_get_subsampling() function Patchwork
2026-10-28 23:02 ` ✗ i915.CI.BAT: failure " Patchwork
2026-10-29 10:31 ` ✗ Xe.CI.FULL: " Patchwork
2026-10-29 16:42 ` ✓ Xe.CI.BAT: success for series starting with [i-g-t,01/10] lib/igt_fb: add igt_format_get_subsampling() function (rev2) Patchwork
2026-10-29 16:58 ` ✓ i915.CI.BAT: " Patchwork
2026-10-30 8:01 ` ✗ i915.CI.Full: failure " Patchwork
2026-10-30 10:43 ` ✗ Xe.CI.FULL: " Patchwork
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=3a6a5381-6f7d-4e72-af57-d220f68d4574@gmail.com \
--to=juhapekka.heikkila@gmail.com \
--cc=igt-dev@lists.freedesktop.org \
--cc=luca@coelho.fi \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox