[PATCH v4 7/7] drm/i915/dsc: Check if vblank is sufficient for dsc prefill

[Nautiyal, Ankit K]

On 1/13/2025 4:43 PM, Nautiyal, Ankit K wrote:
>
> On 1/8/2025 6:05 PM, Mitul Golani wrote:
>> Check if vblank is sufficient for dsc prefill latency.
>>
>> --v2:
>> - Consider chroma downscaling factor in latency calculation. [Ankit]
>> - Replace with appropriate function name.
>>
>> --v3:
>> - Remove FIXME tag.[Ankit]
>> - Replace Ycbcr444 to Ycbcr420.[Anit]
>> - Correct precision. [Ankit]
>> - Use some local valiables like linetime_factor and latency to
>> adjust precision.
>> - Declare latency to 0 initially to avoid returning any garbage values.
>> - Account for second scaler downscaling factor as well. [Ankit]
>>
>> Signed-off-by: Mitul Golani <mitulkumar.ajitkumar.golani at intel.com>
>> ---
>>   drivers/gpu/drm/i915/display/skl_watermark.c | 37 ++++++++++++++++++++
>>   1 file changed, 37 insertions(+)
>>
>> diff --git a/drivers/gpu/drm/i915/display/skl_watermark.c 
>> b/drivers/gpu/drm/i915/display/skl_watermark.c
>> index fe91854e456c..d275ec687748 100644
>> --- a/drivers/gpu/drm/i915/display/skl_watermark.c
>> +++ b/drivers/gpu/drm/i915/display/skl_watermark.c
>> @@ -2292,6 +2292,42 @@ static int icl_build_plane_wm(struct 
>> intel_crtc_state *crtc_state,
>>       return 0;
>>   }
>>   +static int
>> +dsc_prefill_latency(const struct intel_crtc_state *crtc_state)
>> +{
>> +    const struct intel_crtc_scaler_state *scaler_state =
>> +                        &crtc_state->scaler_state;
>> +    int latency = 0;
>> +    int count = hweight32(scaler_state->scaler_users);
>> +    long long hscale_k[2] = {1, 1};
>> +    long long vscale_k[2] = {1, 1};
>> +
>> +    if (!crtc_state->dsc.compression_enable)
>> +        return latency;
>> +
>> +    for (int i = 0; i < count; i++) {
>> +        hscale_k[i] = mul_u32_u32(scaler_state->scalers[i].hscale, 
>> 1000) >> 16;
>> +        vscale_k[i] = mul_u32_u32(scaler_state->scalers[i].vscale, 
>> 1000) >> 16;
>> +    }
>> +
>> +    if (count) {
>> +        int chroma_downscaling_factor =
>> +            crtc_state->output_format == 
>> INTEL_OUTPUT_FORMAT_YCBCR420 ? 2 : 1;
>> +        long long total_scaling_factor;
>> +        int linetime_factor = DIV_ROUND_UP(15 * 
>> crtc_state->linetime, 10);
>> +
>> +        total_scaling_factor  = DIV_ROUND_UP_ULL(hscale_k[0] * 
>> vscale_k[0], 1000000);
>
> I think this might end up being 0.
>
> Consider hscale and vscale to be 0.5 each. hscale_k and vscale_k will 
> be 500 each.
>
> total_scaling_factor will become 500*500/ 10,00,000 or 2,50,000 
> /10,00,000 = 0.
>
> I think you should compute latency for scaler 0 and then if there is 
> another scaler, multiply it with the total scaling factor for scaler 1.

Scratch that.

Just checked the Bspec. We need to have downscale amount <for each 
direction and scaler> = MAX(1, scaler input / scaler output)

So at min the scaling factor will be one. This check is missing and we 
need to consider this while computing the scaling factor.


Regards,

Ankit


>
> Regards,
>
> Ankit
>
>
>> +
>> +        if (count > 1)
>> +            total_scaling_factor *= DIV_ROUND_UP_ULL(hscale_k[1] * 
>> vscale_k[1],
>> +                                 1000000);
>> +
>> +        latency = total_scaling_factor * linetime_factor * 
>> chroma_downscaling_factor;
>> +    }
>> +
>> +    return latency;
>> +}
>> +
>>   static int
>>   scaler_prefill_latency(const struct intel_crtc_state *crtc_state)
>>   {
>> @@ -2333,6 +2369,7 @@ skl_is_vblank_too_short(const struct 
>> intel_crtc_state *crtc_state,
>>       return crtc_state->framestart_delay +
>>           intel_usecs_to_scanlines(adjusted_mode, latency) +
>>           scaler_prefill_latency(crtc_state) +
>> +        dsc_prefill_latency(crtc_state) +
>>           wm0_lines >
>>           adjusted_mode->crtc_vtotal - adjusted_mode->crtc_vblank_start;
>>   }