Re: [PATCH] media: iris: Fix VSP clock calculation for high bitrate streams

[Vikash Garodia <vikash.garodia@oss.qualcomm.com>]

On 10/6/2026 1:46 PM, Vishnu Reddy wrote:
> Frame drops were observed during high-resolution streams decoded at high
> bitrates, as well as concurrent decode sessions. Analysis showed that the
> Video Signal Processor (VSP) clock requirement was underestimated for
> these workloads, resulting in insufficient clock scaling and reduced
> processing throughput.
> 
> Update the VSP frequency calculation to more accurately reflect the
> processing requirements of different codecs and bitrates, ensuring that
> the requested clock frequency is sufficient to sustain the workload and
> prevent frame drops.
> 
> Fixes: 82b96a8d8d6e ("media: iris: Move vpu35 specific api to common to use for vpu4")
> Cc: stable@vger.kernel.org
> Signed-off-by: Vishnu Reddy <busanna.reddy@oss.qualcomm.com>
> ---
>   drivers/media/platform/qcom/iris/iris_vpu_common.c | 29 +++++++++++++++++++++-
>   1 file changed, 28 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/media/platform/qcom/iris/iris_vpu_common.c b/drivers/media/platform/qcom/iris/iris_vpu_common.c
> index 243bd845349f..c4762229a1de 100644
> --- a/drivers/media/platform/qcom/iris/iris_vpu_common.c
> +++ b/drivers/media/platform/qcom/iris/iris_vpu_common.c
> @@ -454,11 +454,32 @@ u64 iris_vpu2_calculate_frequency(struct iris_inst *inst, size_t data_size)
>   	return max(vpp_freq, vsp_freq);
>   }
>   
> +static u64 iris_get_vsp_ref_bitrate(struct iris_inst *inst)
> +{
> +	struct device *dev = inst->core->dev;
> +	u32 codec = inst->codec;
> +
> +	switch (codec) {
> +	case V4L2_PIX_FMT_H264:
> +		return 150;
> +	case V4L2_PIX_FMT_HEVC:
> +		return 140;
> +	case V4L2_PIX_FMT_VP9:
> +		return 70;
> +	case V4L2_PIX_FMT_AV1:
> +		return 100;
> +	default:
> +		dev_err(dev, "Unsupported codec\n");
> +		return 1;
> +	}
> +}
> +
>   u64 iris_vpu3x_vpu4x_calculate_frequency(struct iris_inst *inst, size_t data_size)
>   {
>   	struct platform_inst_caps *caps = inst->core->iris_platform_data->inst_caps;
>   	struct v4l2_format *inp_f = inst->fmt_src;
>   	u32 height, width, mbs_per_second, mbpf;
> +	u64 vsp_ref_bitrate, vsp_ref_freq;
>   	u64 fw_cycles, fw_vpp_cycles;
>   	u64 vsp_cycles, vpp_cycles;
>   	u32 fps = inst->frame_rate;
> @@ -485,7 +506,13 @@ u64 iris_vpu3x_vpu4x_calculate_frequency(struct iris_inst *inst, size_t data_siz
>   		vpp_cycles += div_u64(vpp_cycles * 5, 100);
>   
>   	vsp_cycles = fps * data_size * 8;
> -	vsp_cycles = div_u64(vsp_cycles, 2);
> +	vsp_ref_bitrate = iris_get_vsp_ref_bitrate(inst);
> +	vsp_ref_freq = inst->codec == V4L2_PIX_FMT_AV1 ? 533 : 444;
> +	vsp_cycles = div_u64(vsp_ref_freq * vsp_cycles, vsp_ref_bitrate);

Please check this for encoder if it is calculating the value properly 
considering raw YUV frame size.

Regards,
Vikash

> +
> +	/* VSP FW SW offset */
> +	vsp_cycles = div_u64(vsp_cycles * 1000, 1055);
> +
>   	/* VSP FW overhead 1.05 */
>   	vsp_cycles = div_u64(vsp_cycles * 21, 20);
>   
> 
> ---
> base-commit: f0406245cb9855e6318335a8a223551354291a46
> change-id: 20261005-fix-vsp-freq-calc-19b967b79e62
> 
> Best regards,
> --
> Vishnu Reddy <busanna.reddy@oss.qualcomm.com>