| Message ID | 1540971728-26789-1-git-send-email-mgottam@codeaurora.org (mailing list archive) |
|---|---|
| State | Not Applicable, archived |
| Headers | show |
| Series | media: venus: dynamic handling of bitrate | expand |
Hi Malathi, Thanks for the patch! On 10/31/18 9:42 AM, Malathi Gottam wrote: > Any request for a change in bitrate after both planes > are streamed on is handled by setting the target bitrate > property to hardware. > > Signed-off-by: Malathi Gottam <mgottam@codeaurora.org> > --- > drivers/media/platform/qcom/venus/venc_ctrls.c | 11 +++++++++++ > 1 file changed, 11 insertions(+) > > diff --git a/drivers/media/platform/qcom/venus/venc_ctrls.c b/drivers/media/platform/qcom/venus/venc_ctrls.c > index 45910172..54f310c 100644 > --- a/drivers/media/platform/qcom/venus/venc_ctrls.c > +++ b/drivers/media/platform/qcom/venus/venc_ctrls.c > @@ -79,7 +79,9 @@ static int venc_op_s_ctrl(struct v4l2_ctrl *ctrl) > { > struct venus_inst *inst = ctrl_to_inst(ctrl); > struct venc_controls *ctr = &inst->controls.enc; > + struct hfi_bitrate brate; > u32 bframes; > + u32 ptype; > int ret; > > switch (ctrl->id) { > @@ -88,6 +90,15 @@ static int venc_op_s_ctrl(struct v4l2_ctrl *ctrl) > break; > case V4L2_CID_MPEG_VIDEO_BITRATE: > ctr->bitrate = ctrl->val; > + if (inst->streamon_out && inst->streamon_cap) { Hmm, hfi_session_set_property already checks the instance state so I don't think those checks are needed. Another thing is that we need to take the instance mutex to check the instance state. > + ptype = HFI_PROPERTY_CONFIG_VENC_TARGET_BITRATE; > + brate.bitrate = ctr->bitrate; > + brate.layer_id = 0; > + > + ret = hfi_session_set_property(inst, ptype, &brate); > + if (ret) > + return ret; > + } > break; > case V4L2_CID_MPEG_VIDEO_BITRATE_PEAK: > ctr->bitrate_peak = ctrl->val; >
On 2018-11-01 17:48, Stanimir Varbanov wrote: > Hi Malathi, > > Thanks for the patch! > > On 10/31/18 9:42 AM, Malathi Gottam wrote: >> Any request for a change in bitrate after both planes >> are streamed on is handled by setting the target bitrate >> property to hardware. >> >> Signed-off-by: Malathi Gottam <mgottam@codeaurora.org> >> --- >> drivers/media/platform/qcom/venus/venc_ctrls.c | 11 +++++++++++ >> 1 file changed, 11 insertions(+) >> >> diff --git a/drivers/media/platform/qcom/venus/venc_ctrls.c >> b/drivers/media/platform/qcom/venus/venc_ctrls.c >> index 45910172..54f310c 100644 >> --- a/drivers/media/platform/qcom/venus/venc_ctrls.c >> +++ b/drivers/media/platform/qcom/venus/venc_ctrls.c >> @@ -79,7 +79,9 @@ static int venc_op_s_ctrl(struct v4l2_ctrl *ctrl) >> { >> struct venus_inst *inst = ctrl_to_inst(ctrl); >> struct venc_controls *ctr = &inst->controls.enc; >> + struct hfi_bitrate brate; >> u32 bframes; >> + u32 ptype; >> int ret; >> >> switch (ctrl->id) { >> @@ -88,6 +90,15 @@ static int venc_op_s_ctrl(struct v4l2_ctrl *ctrl) >> break; >> case V4L2_CID_MPEG_VIDEO_BITRATE: >> ctr->bitrate = ctrl->val; >> + if (inst->streamon_out && inst->streamon_cap) { > > Hmm, hfi_session_set_property already checks the instance state so I > don't think those checks are needed. Another thing is that we need to > take the instance mutex to check the instance state. Yes Stan, "hfi_session_set_property" this property check the instance state, but returns EINVAL if this is set at UNINIT instance state. Controls initialization happens much earlier than session init and instance init. So the instance is still in UNINIT state which causes failure while setting. Through this patch we try to meet the client request of changing bitrate only when both planes are streamed on. We have two ways to handle it 1. The way in this patch checks the planes state which will definitely ensure instance is in START state. 2. Have a check to ensure that instance is atleast Initialized. I hope the first proposal is good enough for meeting requirement. > >> + ptype = HFI_PROPERTY_CONFIG_VENC_TARGET_BITRATE; >> + brate.bitrate = ctr->bitrate; >> + brate.layer_id = 0; >> + >> + ret = hfi_session_set_property(inst, ptype, &brate); >> + if (ret) >> + return ret; >> + } >> break; >> case V4L2_CID_MPEG_VIDEO_BITRATE_PEAK: >> ctr->bitrate_peak = ctrl->val; >>
On Thu, Nov 1, 2018 at 10:01 PM <mgottam@codeaurora.org> wrote: > > On 2018-11-01 17:48, Stanimir Varbanov wrote: > > Hi Malathi, > > > > Thanks for the patch! > > > > On 10/31/18 9:42 AM, Malathi Gottam wrote: > >> Any request for a change in bitrate after both planes > >> are streamed on is handled by setting the target bitrate > >> property to hardware. > >> > >> Signed-off-by: Malathi Gottam <mgottam@codeaurora.org> > >> --- > >> drivers/media/platform/qcom/venus/venc_ctrls.c | 11 +++++++++++ > >> 1 file changed, 11 insertions(+) > >> > >> diff --git a/drivers/media/platform/qcom/venus/venc_ctrls.c > >> b/drivers/media/platform/qcom/venus/venc_ctrls.c > >> index 45910172..54f310c 100644 > >> --- a/drivers/media/platform/qcom/venus/venc_ctrls.c > >> +++ b/drivers/media/platform/qcom/venus/venc_ctrls.c > >> @@ -79,7 +79,9 @@ static int venc_op_s_ctrl(struct v4l2_ctrl *ctrl) > >> { > >> struct venus_inst *inst = ctrl_to_inst(ctrl); > >> struct venc_controls *ctr = &inst->controls.enc; > >> + struct hfi_bitrate brate; > >> u32 bframes; > >> + u32 ptype; > >> int ret; > >> > >> switch (ctrl->id) { > >> @@ -88,6 +90,15 @@ static int venc_op_s_ctrl(struct v4l2_ctrl *ctrl) > >> break; > >> case V4L2_CID_MPEG_VIDEO_BITRATE: > >> ctr->bitrate = ctrl->val; > >> + if (inst->streamon_out && inst->streamon_cap) { > > > > Hmm, hfi_session_set_property already checks the instance state so I > > don't think those checks are needed. Another thing is that we need to > > take the instance mutex to check the instance state. > > Yes Stan, "hfi_session_set_property" this property check the instance > state, > but returns EINVAL if this is set at UNINIT instance state. > > Controls initialization happens much earlier than session init and > instance init. > So the instance is still in UNINIT state which causes failure while > setting. > > Through this patch we try to meet the client request of changing bitrate > only > when both planes are streamed on. Where does this requirement come from? It should be possible to set the control at any time and it should apply to any encoding happening after the control is set. Best regards, Tomasz
On 2018-11-01 19:23, Tomasz Figa wrote: > On Thu, Nov 1, 2018 at 10:01 PM <mgottam@codeaurora.org> wrote: >> >> On 2018-11-01 17:48, Stanimir Varbanov wrote: >> > Hi Malathi, >> > >> > Thanks for the patch! >> > >> > On 10/31/18 9:42 AM, Malathi Gottam wrote: >> >> Any request for a change in bitrate after both planes >> >> are streamed on is handled by setting the target bitrate >> >> property to hardware. >> >> >> >> Signed-off-by: Malathi Gottam <mgottam@codeaurora.org> >> >> --- >> >> drivers/media/platform/qcom/venus/venc_ctrls.c | 11 +++++++++++ >> >> 1 file changed, 11 insertions(+) >> >> >> >> diff --git a/drivers/media/platform/qcom/venus/venc_ctrls.c >> >> b/drivers/media/platform/qcom/venus/venc_ctrls.c >> >> index 45910172..54f310c 100644 >> >> --- a/drivers/media/platform/qcom/venus/venc_ctrls.c >> >> +++ b/drivers/media/platform/qcom/venus/venc_ctrls.c >> >> @@ -79,7 +79,9 @@ static int venc_op_s_ctrl(struct v4l2_ctrl *ctrl) >> >> { >> >> struct venus_inst *inst = ctrl_to_inst(ctrl); >> >> struct venc_controls *ctr = &inst->controls.enc; >> >> + struct hfi_bitrate brate; >> >> u32 bframes; >> >> + u32 ptype; >> >> int ret; >> >> >> >> switch (ctrl->id) { >> >> @@ -88,6 +90,15 @@ static int venc_op_s_ctrl(struct v4l2_ctrl *ctrl) >> >> break; >> >> case V4L2_CID_MPEG_VIDEO_BITRATE: >> >> ctr->bitrate = ctrl->val; >> >> + if (inst->streamon_out && inst->streamon_cap) { >> > >> > Hmm, hfi_session_set_property already checks the instance state so I >> > don't think those checks are needed. Another thing is that we need to >> > take the instance mutex to check the instance state. >> >> Yes Stan, "hfi_session_set_property" this property check the instance >> state, >> but returns EINVAL if this is set at UNINIT instance state. >> >> Controls initialization happens much earlier than session init and >> instance init. >> So the instance is still in UNINIT state which causes failure while >> setting. >> >> Through this patch we try to meet the client request of changing >> bitrate >> only >> when both planes are streamed on. > > Where does this requirement come from? It should be possible to set > the control at any time and it should apply to any encoding happening > after the control is set. > With the patch, now video driver will set the control whenever client sets and will apply to encoder. Thanks, Vikash
On Thu, Nov 1, 2018 at 11:23 PM Vikash Garodia <vgarodia@codeaurora.org> wrote: > > On 2018-11-01 19:23, Tomasz Figa wrote: > > On Thu, Nov 1, 2018 at 10:01 PM <mgottam@codeaurora.org> wrote: > >> > >> On 2018-11-01 17:48, Stanimir Varbanov wrote: > >> > Hi Malathi, > >> > > >> > Thanks for the patch! > >> > > >> > On 10/31/18 9:42 AM, Malathi Gottam wrote: > >> >> Any request for a change in bitrate after both planes > >> >> are streamed on is handled by setting the target bitrate > >> >> property to hardware. > >> >> > >> >> Signed-off-by: Malathi Gottam <mgottam@codeaurora.org> > >> >> --- > >> >> drivers/media/platform/qcom/venus/venc_ctrls.c | 11 +++++++++++ > >> >> 1 file changed, 11 insertions(+) > >> >> > >> >> diff --git a/drivers/media/platform/qcom/venus/venc_ctrls.c > >> >> b/drivers/media/platform/qcom/venus/venc_ctrls.c > >> >> index 45910172..54f310c 100644 > >> >> --- a/drivers/media/platform/qcom/venus/venc_ctrls.c > >> >> +++ b/drivers/media/platform/qcom/venus/venc_ctrls.c > >> >> @@ -79,7 +79,9 @@ static int venc_op_s_ctrl(struct v4l2_ctrl *ctrl) > >> >> { > >> >> struct venus_inst *inst = ctrl_to_inst(ctrl); > >> >> struct venc_controls *ctr = &inst->controls.enc; > >> >> + struct hfi_bitrate brate; > >> >> u32 bframes; > >> >> + u32 ptype; > >> >> int ret; > >> >> > >> >> switch (ctrl->id) { > >> >> @@ -88,6 +90,15 @@ static int venc_op_s_ctrl(struct v4l2_ctrl *ctrl) > >> >> break; > >> >> case V4L2_CID_MPEG_VIDEO_BITRATE: > >> >> ctr->bitrate = ctrl->val; > >> >> + if (inst->streamon_out && inst->streamon_cap) { > >> > > >> > Hmm, hfi_session_set_property already checks the instance state so I > >> > don't think those checks are needed. Another thing is that we need to > >> > take the instance mutex to check the instance state. > >> > >> Yes Stan, "hfi_session_set_property" this property check the instance > >> state, > >> but returns EINVAL if this is set at UNINIT instance state. > >> > >> Controls initialization happens much earlier than session init and > >> instance init. > >> So the instance is still in UNINIT state which causes failure while > >> setting. > >> > >> Through this patch we try to meet the client request of changing > >> bitrate > >> only > >> when both planes are streamed on. > > > > Where does this requirement come from? It should be possible to set > > the control at any time and it should apply to any encoding happening > > after the control is set. > > > With the patch, now video driver will set the control whenever client > sets > and will apply to encoder. That's good, thanks for clarifying. I guess I misunderstood Malathi's comment. Best regards, Tomasz
On 11/1/18 4:31 PM, Tomasz Figa wrote: > On Thu, Nov 1, 2018 at 11:23 PM Vikash Garodia <vgarodia@codeaurora.org> wrote: >> >> On 2018-11-01 19:23, Tomasz Figa wrote: >>> On Thu, Nov 1, 2018 at 10:01 PM <mgottam@codeaurora.org> wrote: >>>> >>>> On 2018-11-01 17:48, Stanimir Varbanov wrote: >>>>> Hi Malathi, >>>>> >>>>> Thanks for the patch! >>>>> >>>>> On 10/31/18 9:42 AM, Malathi Gottam wrote: >>>>>> Any request for a change in bitrate after both planes >>>>>> are streamed on is handled by setting the target bitrate >>>>>> property to hardware. >>>>>> >>>>>> Signed-off-by: Malathi Gottam <mgottam@codeaurora.org> >>>>>> --- >>>>>> drivers/media/platform/qcom/venus/venc_ctrls.c | 11 +++++++++++ >>>>>> 1 file changed, 11 insertions(+) >>>>>> >>>>>> diff --git a/drivers/media/platform/qcom/venus/venc_ctrls.c >>>>>> b/drivers/media/platform/qcom/venus/venc_ctrls.c >>>>>> index 45910172..54f310c 100644 >>>>>> --- a/drivers/media/platform/qcom/venus/venc_ctrls.c >>>>>> +++ b/drivers/media/platform/qcom/venus/venc_ctrls.c >>>>>> @@ -79,7 +79,9 @@ static int venc_op_s_ctrl(struct v4l2_ctrl *ctrl) >>>>>> { >>>>>> struct venus_inst *inst = ctrl_to_inst(ctrl); >>>>>> struct venc_controls *ctr = &inst->controls.enc; >>>>>> + struct hfi_bitrate brate; >>>>>> u32 bframes; >>>>>> + u32 ptype; >>>>>> int ret; >>>>>> >>>>>> switch (ctrl->id) { >>>>>> @@ -88,6 +90,15 @@ static int venc_op_s_ctrl(struct v4l2_ctrl *ctrl) >>>>>> break; >>>>>> case V4L2_CID_MPEG_VIDEO_BITRATE: >>>>>> ctr->bitrate = ctrl->val; >>>>>> + if (inst->streamon_out && inst->streamon_cap) { >>>>> >>>>> Hmm, hfi_session_set_property already checks the instance state so I >>>>> don't think those checks are needed. Another thing is that we need to >>>>> take the instance mutex to check the instance state. >>>> >>>> Yes Stan, "hfi_session_set_property" this property check the instance >>>> state, >>>> but returns EINVAL if this is set at UNINIT instance state. >>>> >>>> Controls initialization happens much earlier than session init and >>>> instance init. >>>> So the instance is still in UNINIT state which causes failure while >>>> setting. >>>> >>>> Through this patch we try to meet the client request of changing >>>> bitrate >>>> only >>>> when both planes are streamed on. >>> >>> Where does this requirement come from? It should be possible to set >>> the control at any time and it should apply to any encoding happening >>> after the control is set. >>> >> With the patch, now video driver will set the control whenever client >> sets >> and will apply to encoder. > > That's good, thanks for clarifying. I guess I misunderstood Malathi's comment. OK, just wonder about locking when we check streamon_out/cap flags?
On 2018-11-01 20:33, Stanimir Varbanov wrote: > On 11/1/18 4:31 PM, Tomasz Figa wrote: >> On Thu, Nov 1, 2018 at 11:23 PM Vikash Garodia >> <vgarodia@codeaurora.org> wrote: >>> >>> On 2018-11-01 19:23, Tomasz Figa wrote: >>>> On Thu, Nov 1, 2018 at 10:01 PM <mgottam@codeaurora.org> wrote: >>>>> >>>>> On 2018-11-01 17:48, Stanimir Varbanov wrote: >>>>>> Hi Malathi, >>>>>> >>>>>> Thanks for the patch! >>>>>> >>>>>> On 10/31/18 9:42 AM, Malathi Gottam wrote: >>>>>>> Any request for a change in bitrate after both planes >>>>>>> are streamed on is handled by setting the target bitrate >>>>>>> property to hardware. >>>>>>> >>>>>>> Signed-off-by: Malathi Gottam <mgottam@codeaurora.org> >>>>>>> --- >>>>>>> drivers/media/platform/qcom/venus/venc_ctrls.c | 11 +++++++++++ >>>>>>> 1 file changed, 11 insertions(+) >>>>>>> >>>>>>> diff --git a/drivers/media/platform/qcom/venus/venc_ctrls.c >>>>>>> b/drivers/media/platform/qcom/venus/venc_ctrls.c >>>>>>> index 45910172..54f310c 100644 >>>>>>> --- a/drivers/media/platform/qcom/venus/venc_ctrls.c >>>>>>> +++ b/drivers/media/platform/qcom/venus/venc_ctrls.c >>>>>>> @@ -79,7 +79,9 @@ static int venc_op_s_ctrl(struct v4l2_ctrl >>>>>>> *ctrl) >>>>>>> { >>>>>>> struct venus_inst *inst = ctrl_to_inst(ctrl); >>>>>>> struct venc_controls *ctr = &inst->controls.enc; >>>>>>> + struct hfi_bitrate brate; >>>>>>> u32 bframes; >>>>>>> + u32 ptype; >>>>>>> int ret; >>>>>>> >>>>>>> switch (ctrl->id) { >>>>>>> @@ -88,6 +90,15 @@ static int venc_op_s_ctrl(struct v4l2_ctrl >>>>>>> *ctrl) >>>>>>> break; >>>>>>> case V4L2_CID_MPEG_VIDEO_BITRATE: >>>>>>> ctr->bitrate = ctrl->val; >>>>>>> + if (inst->streamon_out && inst->streamon_cap) { >>>>>> >>>>>> Hmm, hfi_session_set_property already checks the instance state so >>>>>> I >>>>>> don't think those checks are needed. Another thing is that we need >>>>>> to >>>>>> take the instance mutex to check the instance state. >>>>> >>>>> Yes Stan, "hfi_session_set_property" this property check the >>>>> instance >>>>> state, >>>>> but returns EINVAL if this is set at UNINIT instance state. >>>>> >>>>> Controls initialization happens much earlier than session init and >>>>> instance init. >>>>> So the instance is still in UNINIT state which causes failure while >>>>> setting. >>>>> >>>>> Through this patch we try to meet the client request of changing >>>>> bitrate >>>>> only >>>>> when both planes are streamed on. >>>> >>>> Where does this requirement come from? It should be possible to set >>>> the control at any time and it should apply to any encoding >>>> happening >>>> after the control is set. >>>> >>> With the patch, now video driver will set the control whenever client >>> sets >>> and will apply to encoder. >> >> That's good, thanks for clarifying. I guess I misunderstood Malathi's >> comment. > > OK, just wonder about locking when we check streamon_out/cap flags? Yes Stan, I agree that we should add an instance lock for these flags check. Posting the updated patch. Please review.
diff --git a/drivers/media/platform/qcom/venus/venc_ctrls.c b/drivers/media/platform/qcom/venus/venc_ctrls.c index 45910172..54f310c 100644 --- a/drivers/media/platform/qcom/venus/venc_ctrls.c +++ b/drivers/media/platform/qcom/venus/venc_ctrls.c @@ -79,7 +79,9 @@ static int venc_op_s_ctrl(struct v4l2_ctrl *ctrl) { struct venus_inst *inst = ctrl_to_inst(ctrl); struct venc_controls *ctr = &inst->controls.enc; + struct hfi_bitrate brate; u32 bframes; + u32 ptype; int ret; switch (ctrl->id) { @@ -88,6 +90,15 @@ static int venc_op_s_ctrl(struct v4l2_ctrl *ctrl) break; case V4L2_CID_MPEG_VIDEO_BITRATE: ctr->bitrate = ctrl->val; + if (inst->streamon_out && inst->streamon_cap) { + ptype = HFI_PROPERTY_CONFIG_VENC_TARGET_BITRATE; + brate.bitrate = ctr->bitrate; + brate.layer_id = 0; + + ret = hfi_session_set_property(inst, ptype, &brate); + if (ret) + return ret; + } break; case V4L2_CID_MPEG_VIDEO_BITRATE_PEAK: ctr->bitrate_peak = ctrl->val;
Any request for a change in bitrate after both planes are streamed on is handled by setting the target bitrate property to hardware. Signed-off-by: Malathi Gottam <mgottam@codeaurora.org> --- drivers/media/platform/qcom/venus/venc_ctrls.c | 11 +++++++++++ 1 file changed, 11 insertions(+)