Re: [PATCH i-g-t 1/2] test/kms_color: Clamp CTM values to -2.0~2.0 for MediaTek devices

["Ville Syrjälä" <ville.syrjala@linux.intel.com>]

On Sat, Jan 31, 2026 at 11:52:02AM +0000, Jason-JH Lin (林睿祥) wrote:
> On Fri, 2026-01-30 at 17:19 +0200, Ville Syrjälä wrote:
> 
> [snip]
> 
> > > > >  	for (i = 0; i < ARRAY_SIZE(ctm_tests); i++) {
> > > > > +		double fixed_ctm[CTM_SIZE];
> > > > > +		const double *ctm_ptr = ctm_tests[i].ctm;
> > > > > +
> > > > > +		/* The valid value of MediaTek color matrix
> > > > > range
> > > > > is -2.0 ~ 2.0 */
> > > > > +		if (is_mtk_device(data->drm_fd)) {
> > > > > +			int j;
> > > > > +
> > > > > +			for (j = 0; j < CTM_SIZE; j++)
> > > > > +				fixed_ctm[j] =
> > > > > clamp(ctm_tests[i].ctm[j], -2.0, 2.0);
> > > > > +
> > > > > +			ctm_ptr = fixed_ctm;
> > > > > +		}
> > > > 
> > > > The kernel is supposed to clamp this.
> > > 
> > > Thanks for the reviews!
> > > 
> > > The reason I added the clamp in here is that, without it, IGT will
> > > generate a reference framebuffer using CTM=100.0 on the CPU side,
> > > then
> > > send CTM1.0 to kernel.
> > > IGT will generate another framebuffer using CTM=1.0, then send
> > > CTM=100.0 to the kernel.
> > > 
> > > The kernel will clamp the CTM from 100.0 to 2.0 due to hardware
> > > limits,
> > > so the hardware output will not match the CPU-generated reference,
> > > resulting in a CRC mismatch and a test failure.
> > > 
> > > Therefore, I think I need to clamp CTM from 100.0 to 2.0 before IGT
> > > generating a reference framebuffer on CPU side.
> > 
> > IIRC the way the test does things is that with the 100.0 (or
> > whatever >1.0) the colors should end up fully saturated anyway.
> > So whether it used 2.0 or 100.0 shouldn't matter.
> > 
> For MTK hardware, CTM values between 1.0 and 2.0 result in a gain
> effect, while values between 0 and 1.0 cause attenuation. Values above
> 2.0 are not supported by the hardware.
> 
> Therefore, if the reference framebuffer generated by the CPU using CTM
> 2.0 does not match the hardware’s output, the CRC should theoretically
> be different. However, in our current tests, the CRC results for CTM
> 2.0 are actually identical.
> 
> This means that CTM values greater than 1.0 may not be equivalent to
> CTM 1.0 in terms of output.

As I said IIRC the max-ctm test should only generate fully saturated
colors for any value >=1.0. So it shouldn't matter what the hardware
max is as long as it's >=1.0.

-- 
Ville Syrjälä
Intel