Re: [PATCH v2] media: iris: add FPS calculation and VPP FW overhead in frequency formula

[Dikshita Agarwal <dikshita.agarwal@oss.qualcomm.com>]

On 3/20/2026 12:02 PM, Vishnu Reddy wrote:
> 
> On 3/13/2026 12:54 PM, Vikash Garodia wrote:
>>
>> On 3/5/2026 7:00 PM, Vishnu Reddy wrote:
>>> The driver was using a fixed default FPS value when calculating the VPU
>>> frequency. This caused wrong frequency requests for high‑frame‑rate
>>> streams, for example 4K at 240 FPS. Because of this, the hardware was
>>> running at a lower frequency than needed.
>>>
>>> Add the FPS measurement based on the decoder input buffer arrival rate.
>>> The measured FPS is stored per instance and used in frequency calculation
>>> instead of the fixed default FPS. The value is clamped so that it does
>>> not exceed platform limits. Add a VPP firmware overhead when running in
>>> STAGE_2.
>>>
>>> Signed-off-by: Vishnu Reddy <busanna.reddy@oss.qualcomm.com>
>>> ---
>>> Changes in v2:
>>> - Replaced div_u64 with mult_frac
>>> - Link to v1:
>>> https://lore.kernel.org/r/20260304-update_fps_calculation-v1-1-4eeac373a504@oss.qualcomm.com
>>> ---
>>>   drivers/media/platform/qcom/iris/iris_instance.h   |  2 ++
>>>   drivers/media/platform/qcom/iris/iris_vdec.c       | 21
>>> +++++++++++++++++++++
>>>   drivers/media/platform/qcom/iris/iris_vpu_common.c |  6 +++++-
>>>   3 files changed, 28 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/media/platform/qcom/iris/iris_instance.h
>>> b/drivers/media/platform/qcom/iris/iris_instance.h
>>> index 16965150f427..180cba36a7f2 100644
>>> --- a/drivers/media/platform/qcom/iris/iris_instance.h
>>> +++ b/drivers/media/platform/qcom/iris/iris_instance.h
>>> @@ -109,6 +109,8 @@ struct iris_inst {
>>>       u32                metadata_idx;
>>>       u32                codec;
>>>       bool                last_buffer_dequeued;
>>> +    u64                last_buf_recv_time_ns;
>>> +    u32                frame_count;
>>>       u32                frame_rate;
>>>       u32                operating_rate;
>>>       u32                hfi_rc_type;
>>> diff --git a/drivers/media/platform/qcom/iris/iris_vdec.c
>>> b/drivers/media/platform/qcom/iris/iris_vdec.c
>>> index 719217399a30..88820060e22a 100644
>>> --- a/drivers/media/platform/qcom/iris/iris_vdec.c
>>> +++ b/drivers/media/platform/qcom/iris/iris_vdec.c
>>> @@ -369,6 +369,8 @@ int iris_vdec_streamon_input(struct iris_inst *inst)
>>>       if (ret)
>>>           return ret;
>>>   +    inst->frame_count = 0;
>>> +
>>>       return iris_process_streamon_input(inst);
>>>   }
>>>   @@ -411,6 +413,7 @@ int iris_vdec_qbuf(struct iris_inst *inst, struct
>>> vb2_v4l2_buffer *vbuf)
>>>   {
>>>       struct iris_buffer *buf = to_iris_buffer(vbuf);
>>>       struct vb2_buffer *vb2 = &vbuf->vb2_buf;
>>> +    u64 cur_buf_recv_time_ns, time_delta_ns;
>>>       struct vb2_queue *q;
>>>       int ret;
>>>   @@ -427,6 +430,24 @@ int iris_vdec_qbuf(struct iris_inst *inst, struct
>>> vb2_v4l2_buffer *vbuf)
>>>           return 0;
>>>       }
>>>   +    if (buf->type == BUF_INPUT) {
>>> +        cur_buf_recv_time_ns = ktime_get_ns();
>>> +
>>> +        if (inst->frame_count == 0) {
>>> +            inst->last_buf_recv_time_ns = cur_buf_recv_time_ns;
>>> +            inst->frame_rate = MAXIMUM_FPS;
>>> +        }
>>> +        time_delta_ns = cur_buf_recv_time_ns -
>>> inst->last_buf_recv_time_ns;
>>> +
>>> +        if (time_delta_ns >= NSEC_PER_SEC) {
>>> +            inst->frame_rate = clamp_t(u32, inst->frame_count,
>>> DEFAULT_FPS,
>>> +                           MAXIMUM_FPS);
>>> +            inst->last_buf_recv_time_ns = cur_buf_recv_time_ns;
>>> +            inst->frame_count = 0;
>>> +        }
>>> +        inst->frame_count++;
>>> +    }
>>> +
>>
>> simplifying it ?
>>
>>     u64 now_ns, delta_ns;
>>
>>     if (buf->type == BUF_INPUT) {
>>         now_ns = ktime_get_ns();
>>
>>         if (!inst->input_frame_count)
>>             inst->buf_fps_start_ts = now_ns;
> 
> Here, the actual required FPS will be found after 1 second. until that,
> the initial FPS need to set for max.

why max and not default? setting to max would over vote always in the start
of session.

- Dikshita

> 
>>
>>     inst->input_frame_count++;
> If we increment the frame_count here, once frame_count becomes zero (after
> 1 sec),
> In next buffer queue, again the above if condition will be true and FPS
> will be set to max.
> I can add the code like below, Let me know your opinion.
> 
> if (buf->type == BUF_INPUT) {
>     now_ns = ktime_get_ns();
> 
>     if (!inst->frame_count) {
>         inst->last_buf_recv_time_ns = now_ns;
>         inst->frame_rate = MAXIMUM_FPS;
>     }
>     delta_ns = now_ns - inst->buf_fps_start_ts;
> 
>     if (delta_ns >= NSEC_PER_SEC) {
>         inst->frame_rate = clamp_t(u32, inst->frame_count, DEFAULT_FPS,
> MAXIMUM_FPS);
>         inst->buf_fps_start_ts = now_ns;
>         inst->frame_count = 0;
>     }
>     inst->frame_count++;
> }
> 
> Regards,
> Vishnu Reddy
> 
>> delta_ns = now_ns - inst->buf_fps_start_ts;
>>         if (delta_ns < NSEC_PER_SEC)
>>             //skip the further logic;
>>
>>         inst->frame_rate = clamp_t(u32, input_frame_count, DEFAULT_FPS,
>> MAXIMUM_FPS);
>>        inst->input_frame_count = 0;
>>     }
>> }
>>
>>>       iris_scale_power(inst);
>>>         return iris_queue_buffer(inst, buf);
>>> diff --git a/drivers/media/platform/qcom/iris/iris_vpu_common.c
>>> b/drivers/media/platform/qcom/iris/iris_vpu_common.c
>>> index 548e5f1727fd..d621ccffa868 100644
>>> --- a/drivers/media/platform/qcom/iris/iris_vpu_common.c
>>> +++ b/drivers/media/platform/qcom/iris/iris_vpu_common.c
>>> @@ -416,7 +416,7 @@ u64 iris_vpu3x_vpu4x_calculate_frequency(struct
>>> iris_inst *inst, size_t data_siz
>>>       u32 height, width, mbs_per_second, mbpf;
>>>       u64 fw_cycles, fw_vpp_cycles;
>>>       u64 vsp_cycles, vpp_cycles;
>>> -    u32 fps = DEFAULT_FPS;
>>> +    u32 fps = inst->frame_rate;
>>>         width = max(inp_f->fmt.pix_mp.width, inst->crop.width);
>>>       height = max(inp_f->fmt.pix_mp.height, inst->crop.height);
>>> @@ -435,6 +435,10 @@ u64 iris_vpu3x_vpu4x_calculate_frequency(struct
>>> iris_inst *inst, size_t data_siz
>>>       if (inst->fw_caps[PIPE].value > 1)
>>>           vpp_cycles += div_u64(vpp_cycles * 59, 1000);
>>>   +    /* 1.05 is VPP FW overhead */
>>> +    if (inst->fw_caps[STAGE].value == STAGE_2)
>>> +        vpp_cycles += mult_frac(vpp_cycles, 5, 100);
>>> +
>>>       vsp_cycles = fps * data_size * 8;
>>>       vsp_cycles = div_u64(vsp_cycles, 2);
>>>       /* VSP FW overhead 1.05 */
>>>
>>> ---
>>> base-commit: f505e978d1a0442adbbde48aed38c084ddea6d6e
>>> change-id: 20260304-update_fps_calculation-98ee7f7507b1
>>>
>>> Best regards,
>>