Re: [PATCH v4 6/7] media: venus: helpers: update NUM_MBS macro calculation

[dikshita]

Hi Stan,

On 2021-10-02 03:01, Stanimir Varbanov wrote:
> Hi,
>
> On 8/10/21 12:47 PM, Dikshita Agarwal wrote:
> > Consider alignment while calculating NUM_MBS.
> >
> > Co-developed-by: Mansur Alisha Shaik <mansur@xxxxxxxxxxxxxx>
> > Signed-off-by: Dikshita Agarwal <dikshita@xxxxxxxxxxxxxx>
> > ---
> >  drivers/media/platform/qcom/venus/helpers.c | 13 +++++++------
> >  1 file changed, 7 insertions(+), 6 deletions(-)
> >
> > diff --git a/drivers/media/platform/qcom/venus/helpers.c 
> > b/drivers/media/platform/qcom/venus/helpers.c
> > index 60a2775..2db33ba 100644
> > --- a/drivers/media/platform/qcom/venus/helpers.c
> > +++ b/drivers/media/platform/qcom/venus/helpers.c
> > @@ -18,8 +18,8 @@
> >  #include "hfi_platform.h"
> >  #include "hfi_parser.h"
> >
> > -#define NUM_MBS_720P	(((1280 + 15) >> 4) * ((720 + 15) >> 4))
> > -#define NUM_MBS_4K	(((4096 + 15) >> 4) * ((2304 + 15) >> 4))
> > +#define NUM_MBS_720P	(((ALIGN(1280, 16)) >> 4) * ((ALIGN(736, 16)) >> 
> > 4))
> > +#define NUM_MBS_4K	(((ALIGN(4096, 16)) >> 4) * ((ALIGN(2304, 16)) >> 
> > 4))
> >
> >  struct intbuf {
> >  	struct list_head list;
> > @@ -1098,16 +1098,17 @@ static u32 venus_helper_get_work_mode(struct 
> > venus_inst *inst)
> >  	u32 num_mbs;
> >
> >  	mode = VIDC_WORK_MODE_2;
> > +
> >  	if (inst->session_type == VIDC_SESSION_TYPE_DEC) {
> > -		num_mbs = (ALIGN(inst->height, 16) * ALIGN(inst->width, 16)) / 256;
> > +		num_mbs = ((ALIGN(inst->height, 16))/16 * (ALIGN(inst->width, 
> > 16)))/16;
>
> Could you help me understand what is the difference between both
> calculations? IMO this patch should only change NUM_MBS_720P and
> NUM_MBS_4K macros.
We are updating the calculation here to match with the calculation used 
in Macro.
AFAIK, we were seeing a size mismatch for a resolution with the old 
calculation.

> >  		if (inst->hfi_codec == HFI_VIDEO_CODEC_MPEG2 ||
> > -		    inst->pic_struct != HFI_INTERLACE_FRAME_PROGRESSIVE ||
> > -		    num_mbs <= NUM_MBS_720P)
> > +			inst->pic_struct != HFI_INTERLACE_FRAME_PROGRESSIVE ||
> > +			num_mbs <= NUM_MBS_720P)
>
> This change just makes indentation wrong and also it is not related to
> the patch subject.
I think it can be dropped.
> >  			mode = VIDC_WORK_MODE_1;
> >  	} else {
> >  		num_mbs = (ALIGN(inst->out_height, 16) * ALIGN(inst->out_width, 
> > 16)) / 256;
> >  		if (inst->hfi_codec == HFI_VIDEO_CODEC_VP8 &&
> > -		    num_mbs <= NUM_MBS_4K)
> > +			num_mbs <= NUM_MBS_4K)
>
> ditto
Same here.
> >  			mode = VIDC_WORK_MODE_1;
> >  	}