Re: venus: venc: Fix enum frameintervals - was: [GIT PULL for v5.5] Venus updates, take 2

[Stanimir Varbanov <stanimir.varbanov@linaro.org>]

Hi Mauro,

On 10/2/19 5:09 PM, Mauro Carvalho Chehab wrote:
> Em Wed,  2 Oct 2019 14:29:53 +0300
> Stanimir Varbanov <stanimir.varbanov@linaro.org> escreveu:
> 
>> Hi Mauro,
>>
>> The Venus driver updates include:
>>
>> * three fixes: fail to suspend, enum frameinterval issue with encoder
>> and frequency table modifications for v3 to handle performance issues.
>> * two new features: interconnect bandwidth support on v4 and more precise
>> clock-scaling on v4.
>>
>> Please pull.
>>
>> Stanimir Varbanov (3):
>>       venus: venc: Fix enum frameintervals
> 
>> From c430fca8f2b9b7274a1186f85b69c469378dbd8a Mon Sep 17 00:00:00 2001
>> From: Stanimir Varbanov <stanimir.varbanov@linaro.org>
>> Date: Tue, 22 Jan 2019 12:53:22 +0200
>> Subject: venus: venc: Fix enum frameintervals
>> To: Linux Media Mailing List <linux-media@vger.kernel.org>
>> Cc: Mauro Carvalho Chehab <mchehab@infradead.org>
>>
>> This fixes an issue when setting the encoder framerate because of
>> missing precision. Now the frameinterval type is changed to
>> TYPE_CONTINUOUS and step = 1. Also the math is changed when
>> framerate property is called - the firmware side expects the
>> framerate in Q16 values.
>>
>> Signed-off-by: Loic Poulain <loic.poulain@linaro.org>
>> Signed-off-by: Stanimir Varbanov <stanimir.varbanov@linaro.org>
>> ---
>>  drivers/media/platform/qcom/venus/venc.c | 23 ++++++++++++++++-------
>>  1 file changed, 16 insertions(+), 7 deletions(-)
>>
>> diff --git a/drivers/media/platform/qcom/venus/venc.c b/drivers/media/platform/qcom/venus/venc.c
>> index 1b7fb2d5887c..d0a97754ef18 100644
>> --- a/drivers/media/platform/qcom/venus/venc.c
>> +++ b/drivers/media/platform/qcom/venus/venc.c
>> @@ -22,6 +22,7 @@
>>  #include "venc.h"
>>  
>>  #define NUM_B_FRAMES_MAX	4
>> +#define FRAMERATE_FACTOR	BIT(16)
>>  
>>  /*
>>   * Three resons to keep MPLANE formats (despite that the number of planes
>> @@ -576,7 +577,7 @@ static int venc_enum_frameintervals(struct file *file, void *fh,
>>  	struct venus_inst *inst = to_inst(file);
>>  	const struct venus_format *fmt;
>>  
>> -	fival->type = V4L2_FRMIVAL_TYPE_STEPWISE;
>> +	fival->type = V4L2_FRMIVAL_TYPE_CONTINUOUS;
>>  
>>  	fmt = find_format(inst, fival->pixel_format,
>>  			  V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE);
>> @@ -599,12 +600,12 @@ static int venc_enum_frameintervals(struct file *file, void *fh,
>>  	    fival->height < frame_height_min(inst))
>>  		return -EINVAL;
>>  
>> -	fival->stepwise.min.numerator = 1;
>> -	fival->stepwise.min.denominator = frate_max(inst);
>> -	fival->stepwise.max.numerator = 1;
>> -	fival->stepwise.max.denominator = frate_min(inst);
>> +	fival->stepwise.min.numerator = FRAMERATE_FACTOR;
>> +	fival->stepwise.min.denominator = frate_max(inst) * FRAMERATE_FACTOR;
>> +	fival->stepwise.max.numerator = FRAMERATE_FACTOR;
>> +	fival->stepwise.max.denominator = frate_min(inst) * FRAMERATE_FACTOR;
> 
> Hmm... this change seems plain wrong to me... Why do you want to change
> the numerator? I mean:
> 
> 	1/frame_min(inst)
> 
> is equal to:
> 
> 	(const * 1) / (const * frame_min(inst))
> 
> Also, on every other driver, the returned fractions are normalized.

Agree, this looks suspicious. I will drop this patch from pull request.

I will look into the other drivers once again.

> 
>>  	fival->stepwise.step.numerator = 1;
>> -	fival->stepwise.step.denominator = frate_max(inst);
>> +	fival->stepwise.step.denominator = 1;
>>  
>>  	return 0;
>>  }
>> @@ -649,6 +650,7 @@ static int venc_set_properties(struct venus_inst *inst)
>>  	struct hfi_quantization quant;
>>  	struct hfi_quantization_range quant_range;
>>  	u32 ptype, rate_control, bitrate, profile = 0, level = 0;
>> +	u64 framerate;
>>  	int ret;
>>  
>>  	ret = venus_helper_set_work_mode(inst, VIDC_WORK_MODE_2);
>> @@ -659,9 +661,16 @@ static int venc_set_properties(struct venus_inst *inst)
>>  	if (ret)
>>  		return ret;
>>  
>> +	framerate = inst->timeperframe.denominator * FRAMERATE_FACTOR;
>> +	/* next line is to round up */
>> +	framerate += inst->timeperframe.numerator - 1;
>> +	do_div(framerate, inst->timeperframe.numerator);
>> +
>>  	ptype = HFI_PROPERTY_CONFIG_FRAME_RATE;
>>  	frate.buffer_type = HFI_BUFFER_OUTPUT;
>> -	frate.framerate = inst->fps * (1 << 16);
>> +	frate.framerate = framerate;
>> +	if (frate.framerate > frate_max(inst) * FRAMERATE_FACTOR)
>> +		frate.framerate = frate_max(inst) * FRAMERATE_FACTOR;
> 
> You should not assume that userspace will be multiplying by the
> frame factor. I mean, the driver should work the same way, no matter
> if userspace is setting the framerate as:
> 
> 	1/30, 2/60, ... n/(n *30)

Agreed.

-- 
regards,
Stan