Re: [PATCH] media: venus: Fix internal buffer size calculations for v6.

[Stanimir Varbanov]

On 3/17/21 10:35 AM, Dikshita Agarwal wrote:
> - Update persist buffer size for encoder to 204800.
> - Update persist buffer size calculation for h264 decoder.
> - h264d level 6 support needs update in internal buffer size.
>   update below buffers size
>   - h264 decoder colocated motion vector buffer.
>   - h264 decoder VPP command buffer.
>   - h265 decoder VPP command buffer.
> - Update VP9_NUM_FRAME_INFO_BUF to 32.
> 
> Signed-off-by: Dikshita Agarwal <dikshita@xxxxxxxxxxxxxx>
> ---
>  .../media/platform/qcom/venus/hfi_plat_bufs_v6.c   | 27 ++++++++++++++--------
>  1 file changed, 18 insertions(+), 9 deletions(-)
> 
> diff --git a/drivers/media/platform/qcom/venus/hfi_plat_bufs_v6.c b/drivers/media/platform/qcom/venus/hfi_plat_bufs_v6.c
> index d43d1a5..a41ad63 100644
> --- a/drivers/media/platform/qcom/venus/hfi_plat_bufs_v6.c
> +++ b/drivers/media/platform/qcom/venus/hfi_plat_bufs_v6.c
> @@ -40,7 +40,8 @@
>  
>  #define MAX_TILE_COLUMNS				32 /* 8K/256 */
>  
> -#define NUM_HW_PIC_BUF					10
> +#define VPP_CMD_MAX_SIZE				BIT(20)

This macro represents size, so I think it would be better to use SZ_1M.

> +#define NUM_HW_PIC_BUF					32
>  #define BIN_BUFFER_THRESHOLD				(1280 * 736)
>  #define H264D_MAX_SLICE					1800
>  /* sizeof(h264d_buftab_t) aligned to 256 */
> @@ -90,6 +91,7 @@
>  #define SIZE_SLIST_BUF_H264		512
>  #define LCU_MAX_SIZE_PELS		64
>  #define LCU_MIN_SIZE_PELS		16
> +#define SIZE_SEI_USERDATA		4096
>  
>  #define H265D_MAX_SLICE			600
>  #define SIZE_H265D_HW_PIC_T		SIZE_H264D_HW_PIC_T
> @@ -199,7 +201,7 @@ static inline u32 size_vpxd_lb_se_left_ctrl(u32 width, u32 height)
>  #define VPX_DECODER_FRAME_BIN_RES_BUDGET_RATIO_DEN	2
>  
>  #define VP8_NUM_FRAME_INFO_BUF			(5 + 1)
> -#define VP9_NUM_FRAME_INFO_BUF			(8 + 2 + 1 + 8)
> +#define VP9_NUM_FRAME_INFO_BUF			32
>  #define VP8_NUM_PROBABILITY_TABLE_BUF		VP8_NUM_FRAME_INFO_BUF
>  #define VP9_NUM_PROBABILITY_TABLE_BUF		(VP9_NUM_FRAME_INFO_BUF + 4)
>  #define VP8_PROB_TABLE_SIZE			3840
> @@ -211,7 +213,7 @@ static inline u32 size_vpxd_lb_se_left_ctrl(u32 width, u32 height)
>  
>  #define QMATRIX_SIZE				(sizeof(u32) * 128 + 256)
>  #define MP2D_QPDUMP_SIZE			115200
> -#define HFI_IRIS2_ENC_PERSIST_SIZE		102400
> +#define HFI_IRIS2_ENC_PERSIST_SIZE		204800
>  #define HFI_MAX_COL_FRAME			6
>  #define HFI_VENUS_VENC_TRE_WB_BUFF_SIZE		(65 << 4) /* in Bytes */
>  #define HFI_VENUS_VENC_DB_LINE_BUFF_PER_MB	512
> @@ -467,7 +469,7 @@ static u32 hfi_iris2_h264d_comv_size(u32 width, u32 height,
>  {
>  	u32 frame_width_in_mbs = ((width + 15) >> 4);
>  	u32 frame_height_in_mbs = ((height + 15) >> 4);
> -	u32 col_mv_aligned_width = (frame_width_in_mbs << 6);
> +	u32 col_mv_aligned_width = (frame_width_in_mbs << 7);

So we start align on 256 instead of 128?

>  	u32 col_zero_aligned_width = (frame_width_in_mbs << 2);
>  	u32 col_zero_size = 0, size_colloc = 0, comv_size = 0;
>  
> @@ -499,10 +501,14 @@ static u32 size_h264d_bse_cmd_buf(u32 height)
>  
>  static u32 size_h264d_vpp_cmd_buf(u32 height)
>  {
> +	u32 size = 0;
>  	u32 aligned_height = ALIGN(height, 32);

Add blank line here.

> +	size = min_t(u32, (((aligned_height + 15) >> 4) * 3 * 4), H264D_MAX_SLICE) *
> +		SIZE_H264D_VPP_CMD_PER_BUF;
> +	if (size > VPP_CMD_MAX_SIZE)
> +		size = VPP_CMD_MAX_SIZE;
>  
> -	return min_t(u32, (((aligned_height + 15) >> 4) * 3 * 4),
> -		     H264D_MAX_SLICE) * SIZE_H264D_VPP_CMD_PER_BUF;
> +	return size;
>  }
>  
>  static u32 hfi_iris2_h264d_non_comv_size(u32 width, u32 height,
> @@ -559,8 +565,11 @@ static u32 size_h265d_vpp_cmd_buf(u32 width, u32 height)
>  	size = min_t(u32, size, H265D_MAX_SLICE + 1);
>  	size = ALIGN(size, 4);
>  	size = 2 * size * SIZE_H265D_VPP_CMD_PER_BUF;
> +	size = ALIGN(size, HFI_DMA_ALIGNMENT);
> +	if (size > VPP_CMD_MAX_SIZE)
> +		size = VPP_CMD_MAX_SIZE;
>  
> -	return ALIGN(size, HFI_DMA_ALIGNMENT);
> +	return size;
>  }
>  
>  static u32 hfi_iris2_h265d_comv_size(u32 width, u32 height,
> @@ -1004,8 +1013,8 @@ static u32 enc_persist_size(void)
>  
>  static u32 h264d_persist1_size(void)
>  {
> -	return ALIGN((SIZE_SLIST_BUF_H264 * NUM_SLIST_BUF_H264),
> -		     HFI_DMA_ALIGNMENT);
> +	return ALIGN((SIZE_SLIST_BUF_H264 * NUM_SLIST_BUF_H264
> +		     + NUM_HW_PIC_BUF * SIZE_SEI_USERDATA), HFI_DMA_ALIGNMENT);
>  }
>  
>  static u32 h265d_persist1_size(void)
> 

-- 
regards,
Stan