Re: [PATCH v2 2/2] media: cedrus: Add HEVC/H.265 decoding support

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi,

On Tue, 2018-11-27 at 09:21 +0100, Maxime Ripard wrote:
> Hi!
> 
> On Fri, Nov 23, 2018 at 02:02:09PM +0100, Paul Kocialkowski wrote:
> > This introduces support for HEVC/H.265 to the Cedrus VPU driver, with
> > both uni-directional and bi-directional prediction modes supported.
> > 
> > Field-coded (interlaced) pictures, custom quantization matrices and
> > 10-bit output are not supported at this point.
> > 
> > Signed-off-by: Paul Kocialkowski <paul.kocialkowski@xxxxxxxxxxx>
> 
> Output from checkpatch:
> total: 0 errors, 68 warnings, 14 checks, 999 lines checked

Looks like many of the "line over 80 chars" are due to macros. I don't
think it would be a good idea to break them down or to change the
macros names since they are directly inherited from the bitstream
elements.

What do you think?

> > +/*
> > + * Note: Neighbor info buffer size is apparently doubled for H6, which may be
> > + * related to 10 bit H265 support.
> > + */
> > +#define CEDRUS_H265_NEIGHBOR_INFO_BUF_SIZE	(397 * SZ_1K)
> > +#define CEDRUS_H265_ENTRY_POINTS_BUF_SIZE	(4 * SZ_1K)
> > +#define CEDRUS_H265_MV_COL_BUF_UNIT_CTB_SIZE	160
> 
> Having some information on where this is coming from would be useful.

Yes, definitely.

> > +static void cedrus_h265_sram_write_data(struct cedrus_dev *dev, u32 *data,
> 
> Since the data pointer is pretty much an opaque structure, you should
> have a void pointer here, that would avoid the type casting you're
> doing when calling that function.

Sure, that would make more sense.

[...]

> > +	/* Output frame. */
> > +
> > +	output_pic_list_index = V4L2_HEVC_DPB_ENTRIES_NUM_MAX;
> > +	pic_order_cnt[0] = pic_order_cnt[1] = slice_params->slice_pic_order_cnt;
> > +	mv_col_buf_addr[0] = cedrus_h265_frame_info_mv_col_buf_addr(ctx,
> > +		run->dst->vb2_buf.index, 0) - PHYS_OFFSET;
> > +	mv_col_buf_addr[1] = cedrus_h265_frame_info_mv_col_buf_addr(ctx,
> > +		run->dst->vb2_buf.index, 1) - PHYS_OFFSET;
> > +	dst_luma_addr = cedrus_dst_buf_addr(ctx, run->dst->vb2_buf.index, 0) -
> > +			PHYS_OFFSET;
> > +	dst_chroma_addr = cedrus_dst_buf_addr(ctx, run->dst->vb2_buf.index, 1) -
> > +			PHYS_OFFSET;
> > +
> > +	cedrus_h265_frame_info_write_single(dev, output_pic_list_index,
> > +					    slice_params->pic_struct != 0,
> > +					    pic_order_cnt, mv_col_buf_addr,
> > +					    dst_luma_addr, dst_chroma_addr);
> 
> You can only pass the run and slice_params pointers to that function.

The point is to make it independent from the context, so that the same
function can be called with either the slice_params or the dpb info.
I don't think making two variants or even two wrappers would bring any
significant benefit.

> > +
> > +	cedrus_write(dev, VE_DEC_H265_OUTPUT_FRAME_IDX, output_pic_list_index);
> > +
> > +	/* Reference picture list 0 (for P/B frames). */
> > +	if (slice_params->slice_type != V4L2_HEVC_SLICE_TYPE_I) {
> > +		cedrus_h265_ref_pic_list_write(dev, slice_params->ref_idx_l0,
> > +			slice_params->num_ref_idx_l0_active_minus1 + 1,
> > +			slice_params->dpb, slice_params->num_active_dpb_entries,
> > +			VE_DEC_H265_SRAM_OFFSET_REF_PIC_LIST0);
> > +
> 
> slice_params is enough.

The rationale is similar to the one above: being able to use the same
helper with either L0 or L1, which implies passing the relevant
elements directly.

> > +		if (pps->weighted_pred_flag || pps->weighted_bipred_flag)
> > +			cedrus_h265_pred_weight_write(dev,
> > +				pred_weight_table->delta_luma_weight_l0,
> > +				pred_weight_table->luma_offset_l0,
> > +				pred_weight_table->delta_chroma_weight_l0,
> > +				pred_weight_table->chroma_offset_l0,
> > +				slice_params->num_ref_idx_l0_active_minus1 + 1,
> > +				VE_DEC_H265_SRAM_OFFSET_PRED_WEIGHT_LUMA_L0,
> > +				VE_DEC_H265_SRAM_OFFSET_PRED_WEIGHT_CHROMA_L0);
> 
> Ditto, that function should only take the pred_weight_table and
> slice_params pointers

And same rational as well.

> > +	}
> > +
> > +	/* Reference picture list 1 (for B frames). */
> > +	if (slice_params->slice_type == V4L2_HEVC_SLICE_TYPE_B) {
> > +		cedrus_h265_ref_pic_list_write(dev, slice_params->ref_idx_l1,
> > +			slice_params->num_ref_idx_l1_active_minus1 + 1,
> > +			slice_params->dpb,
> > +			slice_params->num_active_dpb_entries,
> > +			VE_DEC_H265_SRAM_OFFSET_REF_PIC_LIST1);
> > +
> > +		if (pps->weighted_bipred_flag)
> > +			cedrus_h265_pred_weight_write(dev,
> > +				pred_weight_table->delta_luma_weight_l1,
> > +				pred_weight_table->luma_offset_l1,
> > +				pred_weight_table->delta_chroma_weight_l1,
> > +				pred_weight_table->chroma_offset_l1,
> > +				slice_params->num_ref_idx_l1_active_minus1 + 1,
> > +				VE_DEC_H265_SRAM_OFFSET_PRED_WEIGHT_LUMA_L1,
> > +				VE_DEC_H265_SRAM_OFFSET_PRED_WEIGHT_CHROMA_L1);
> > +	}
> 
> Ditto
> 
> Looks good otherwise, thanks!

Thanks for the review!

Cheers,

Paul

-- 
Paul Kocialkowski, Bootlin
Embedded Linux and kernel engineering
https://bootlin.com

_______________________________________________
devel mailing list
devel@xxxxxxxxxxxxxxxxxxxxxx
http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel



[Index of Archives]     [Linux Driver Backports]     [DMA Engine]     [Linux GPIO]     [Linux SPI]     [Video for Linux]     [Linux USB Devel]     [Linux Coverity]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [Yosemite Backpacking]
  Powered by Linux