Re: [PATCH v2 6/7] media: iris: add iris4 specific H265 line buffer calculation

[Dmitry Baryshkov]

On Fri, Feb 27, 2026 at 07:41:22PM +0530, Vikash Garodia wrote:
> The H265 decoder line buffer size calculation for iris4 (VPU4) was
> previously reusing the iris3 formula. While this works for most
> resolutions, certain configurations require a larger buffer size on
> iris4, causing firmware errors during decode. This resolves firmware
> failures seen with specific test vectors on kaanapali (iris4), and fixes
> the following failing fluster tests
> - PICSIZE_C_Bossen_1
> - WPP_E_ericsson_MAIN_2

This reminds me of the commit fixing SC7280 support. Should SC7280 or
all VPU2.0 platforms also use separate formula?

> 
> Co-developed-by: Vishnu Reddy <busanna.reddy@xxxxxxxxxxxxxxxx>
> Signed-off-by: Vishnu Reddy <busanna.reddy@xxxxxxxxxxxxxxxx>
> Signed-off-by: Vikash Garodia <vikash.garodia@xxxxxxxxxxxxxxxx>
> ---
>  drivers/media/platform/qcom/iris/iris_vpu_buffer.c | 51 +++++++++++++++++++++-
>  1 file changed, 50 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/media/platform/qcom/iris/iris_vpu_buffer.c b/drivers/media/platform/qcom/iris/iris_vpu_buffer.c
> index 9270422c16019ba658ee8813940cb9110ad030a1..a4d599c49ce9052b609b9cedf65f669ba78b5407 100644
> --- a/drivers/media/platform/qcom/iris/iris_vpu_buffer.c
> +++ b/drivers/media/platform/qcom/iris/iris_vpu_buffer.c
> @@ -1755,6 +1755,55 @@ static u32 hfi_vpu4x_buffer_line_vp9d(u32 frame_width, u32 frame_height, u32 _yu
>  	return lb_size + dpb_obp_size;
>  }
>  
> +static u32 hfi_vpu4x_buffer_line_h265d(u32 frame_width, u32 frame_height, bool is_opb,
> +				       u32 num_vpp_pipes)
> +{
> +	u32 num_lcu_per_pipe, fe_left_lb, se_left_lb, vsp_left_lb, top_lb, qp_size,
> +	    dpb_obp = 0, lcu_size = 16;
> +
> +	num_lcu_per_pipe = (DIV_ROUND_UP(frame_height, lcu_size) / num_vpp_pipes) +
> +			   (DIV_ROUND_UP(frame_height, lcu_size) % num_vpp_pipes);
> +
> +	fe_left_lb = ALIGN((DMA_ALIGNMENT * num_lcu_per_pipe), DMA_ALIGNMENT) *
> +				FE_LFT_CTRL_LINE_NUMBERS;
> +	fe_left_lb += ALIGN((DMA_ALIGNMENT * 2 * num_lcu_per_pipe), DMA_ALIGNMENT) *
> +				FE_LFT_DB_DATA_LINE_NUMBERS;
> +	fe_left_lb += ALIGN((DMA_ALIGNMENT * num_lcu_per_pipe), DMA_ALIGNMENT);
> +	fe_left_lb += ALIGN((DMA_ALIGNMENT * 2 * num_lcu_per_pipe), DMA_ALIGNMENT);
> +	fe_left_lb += ALIGN((DMA_ALIGNMENT * 8 * num_lcu_per_pipe), DMA_ALIGNMENT) *
> +				FE_LFT_LR_DATA_LINE_NUMBERS;
> +
> +	if (is_opb)
> +		dpb_obp = size_dpb_opb(frame_height, lcu_size) * num_vpp_pipes;
> +
> +	se_left_lb = max_t(u32, (ALIGN(frame_height, BUFFER_ALIGNMENT_16_BYTES) >> 3) *
> +				MAX_SE_NBR_CTRL_LCU16_LINE_BUFFER_SIZE,
> +				max_t(u32, (ALIGN(frame_height, BUFFER_ALIGNMENT_32_BYTES) >> 3) *
> +				MAX_SE_NBR_CTRL_LCU32_LINE_BUFFER_SIZE,
> +				(ALIGN(frame_height, BUFFER_ALIGNMENT_64_BYTES) >> 3) *
> +				MAX_SE_NBR_CTRL_LCU64_LINE_BUFFER_SIZE));
> +
> +	vsp_left_lb = ALIGN(DIV_ROUND_UP(frame_height, BUFFER_ALIGNMENT_64_BYTES) *
> +				H265_NUM_TILE_ROW, DMA_ALIGNMENT);
> +
> +	top_lb = ALIGN((DMA_ALIGNMENT * DIV_ROUND_UP(frame_width, lcu_size)), DMA_ALIGNMENT) *
> +				FE_TOP_CTRL_LINE_NUMBERS;
> +	top_lb += ALIGN(DMA_ALIGNMENT * 2 * DIV_ROUND_UP(frame_width, lcu_size), DMA_ALIGNMENT) *
> +				FE_TOP_DATA_LUMA_LINE_NUMBERS;
> +	top_lb += ALIGN(DMA_ALIGNMENT * 2 * (DIV_ROUND_UP(frame_width, lcu_size) + 1),
> +			DMA_ALIGNMENT) * FE_TOP_DATA_CHROMA_LINE_NUMBERS;
> +	top_lb += ALIGN(ALIGN(frame_width, BUFFER_ALIGNMENT_64_BYTES) * 2, DMA_ALIGNMENT);
> +	top_lb += ALIGN(ALIGN(frame_width, BUFFER_ALIGNMENT_64_BYTES) * 6, DMA_ALIGNMENT);
> +	top_lb += size_h265d_lb_vsp_top(frame_width, frame_height);
> +
> +	qp_size = size_h265d_qp(frame_width, frame_height);
> +
> +	return ((ALIGN(dpb_obp, DMA_ALIGNMENT) + ALIGN(se_left_lb, DMA_ALIGNMENT) +
> +		ALIGN(vsp_left_lb, DMA_ALIGNMENT)) * num_vpp_pipes) +
> +		ALIGN(fe_left_lb, DMA_ALIGNMENT) + ALIGN(top_lb, DMA_ALIGNMENT) +
> +		ALIGN(qp_size, DMA_ALIGNMENT);
> +}
> +
>  static u32 iris_vpu4x_dec_line_size(struct iris_inst *inst)
>  {
>  	u32 num_vpp_pipes = inst->core->iris_platform_data->num_vpp_pipe;
> @@ -1770,7 +1819,7 @@ static u32 iris_vpu4x_dec_line_size(struct iris_inst *inst)
>  	if (inst->codec == V4L2_PIX_FMT_H264)
>  		return hfi_buffer_line_h264d(width, height, is_opb, num_vpp_pipes);
>  	else if (inst->codec == V4L2_PIX_FMT_HEVC)
> -		return hfi_buffer_line_h265d(width, height, is_opb, num_vpp_pipes);
> +		return hfi_vpu4x_buffer_line_h265d(width, height, is_opb, num_vpp_pipes);
>  	else if (inst->codec == V4L2_PIX_FMT_VP9)
>  		return hfi_vpu4x_buffer_line_vp9d(width, height, out_min_count, is_opb,
>  						  num_vpp_pipes);
> 
> -- 
> 2.34.1
> 

-- 
With best wishes
Dmitry