media: venus: add support for selection rectangles
Subsystems (4)
PATCH v3 1/1 media: venus: add support for selection rectangles
From: Malathi Gottam <>
Handles target type crop by setting the new active rectangle to hardware. The new rectangle should be within YUV size.
This was taken from: https://lkml.org/lkml/2018/11/9/899
--- drivers/media/platform/qcom/venus/venc.c | 24 ++++++++++++++++++++---- 1 file changed, 20 insertions(+), 4 deletions(-)
drivers/media/platform/qcom/venus/venc.c +20 -4
@@ -479,10 +479,26 @@ venc_s_selection(struct file *file, void *fh, struct v4l2_selection *s)switch (s->target) {case V4L2_SEL_TGT_CROP : -if (s->r.width != inst->out_width || - s->r.height != inst->out_height || - s->r.top !=0 || s->r.left !=0 ) -return -EINVAL ; +if (s->r.left !=0 ) { + s->r.width += s->r.left; + s->r.left =0 ; + } + +if (s->r.top !=0 ) { + s->r.height += s->r.top; + s->r.top =0 ; + } + +if (s->r.width > inst->width) + s->r.width = inst->width; +else + inst->width = s->r.width; + +if (s->r.height > inst->height) + s->r.height = inst->height; +else + inst->height = s->r.height; +break ;default :return -EINVAL ;
Signature · 1 line
2.25.0.265.gbab2e86ba0-goog
(again, sorry for duplicate to some people, didn't know the plain text email rule before)
There were a few comments made when this patch was originally posted that were not addressed. I left the patch as it last stood, except for removing the unused variables.
The outstanding comments were:
1. In venc_init_session, it is only using the inst->width/height rectangles and not the inst->out_width/out_height rectangles. So there was a question about whether the OUTUPT rectangle should be set to out_width/out_height. I'm looking for feedback here as I'm not familiar enough with this driver code yet.
2. We should return EBUSY if the selection rectangles are changed after we are initialized (I will update it to do this)
3. Support for non-zero top/left parameters. I'm suspicious that the HFI_INDEX_EXTRADATA_INPUT_CROP property is what should be used for that, but it's not currently used anywhere. Does anybody have details on if that's what it's for?
Quoted 52 lines · Jeffrey Kardatzke
From: Malathi Gottam <>
Handles target type crop by setting the new active rectangle to hardware. The new rectangle should be within YUV size.
This was taken from: https://lkml.org/lkml/2018/11/9/899
Signed-off-by Malathi Gottam <[email protected]>Signed-off-by Jeffrey Kardatzke <[email protected]>--- drivers/media/platform/qcom/venus/venc.c | 24 ++++++++++++++++++++---- 1 file changed, 20 insertions(+), 4 deletions(-)drivers/media/platform/qcom/venus/venc.c +20
-5@@ -479,10 +479,26 @@ venc_s_selection(struct file *file, void *fh, struct v4l2_selection *s)switch (s->target) {case V4L2_SEL_TGT_CROP : -if (s->r.width != inst->out_width || - s->r.height != inst->out_height || - s->r.top !=0 || s->r.left !=0 ) -return -EINVAL ; +if (s->r.left !=0 ) { + s->r.width += s->r.left; + s->r.left =0 ; + } + +if (s->r.top !=0 ) { + s->r.height += s->r.top; + s->r.top =0 ; + } + +if (s->r.width > inst->width) + s->r.width = inst->width; +else + inst->width = s->r.width; + +if (s->r.height > inst->height) + s->r.height = inst->height; +else + inst->height = s->r.height; +break ;default :return -EINVAL ; --2.25.0.265.gbab2e86ba0-goog
Hi Jeff,
Quoted 6 lines · Jeffrey Kardatzke
(again, sorry for duplicate to some people, didn't know the plain text email rule before)
There were a few comments made when this patch was originally posted that were not addressed. I left the patch as it last stood, except for removing the unused variables.
Quoted 6 lines
The outstanding comments were:
1. Invenc_init_session, it is only using the inst->width/height rectangles and not the inst->out_width/out_height rectangles. So there was a question about whether the OUTUPT rectangle should be set to out_width/out_height. I'm looking for feedback here as I'm not familiar enough with this driver code yet.
Yes, the observant reader will see that out_width/height (the encoder input resolution) is not changed in set_selection method, and that is simply because currently the implementation for crop is missing. And it is missing because the so-called extradata is not implemented and not used in the driver, yet.
In fact the proposed patch will just fake the userspace application to think that it sets the crop rectangle but the cropping will not happen at all.
Quoted 2 lines
2. We should return EBUSY if the selection rectangles are changed after we are initialized (I will update it to do this)
Not sure we need that.
Quoted 4 lines
3. Support for non-zero top/left parameters. I'm suspicious that the
HFI_INDEX_EXTRADATA_INPUT_CROPproperty is what should be used for that, but it's not currently used anywhere. Does anybody have details on if that's what it's for?
This is the right property to be used to set encoder crop on the input, but unfortunately it is not that simple.
Actually this property is used to enable the crop extradata handling in the firmware side, but we need to take care of that extradata population in the v4l2 driver. For that purpose we need an extradata buffer which should be filled with extradata header plus the extradata itself.
I'm not sure how to implement that, still. There is a bunch of extradata types where some of them could be filled by the driver but the others could be filled by userspace.
For reference you can see how the crop is set on Android [1].
Quoted 56 lines
Quoted 52 lines · Jeffrey Kardatzke
From: Malathi Gottam <>
Handles target type crop by setting the new active rectangle to hardware. The new rectangle should be within YUV size.
This was taken from: https://lkml.org/lkml/2018/11/9/899
Signed-off-by Malathi Gottam <[email protected]>Signed-off-by Jeffrey Kardatzke <[email protected]>--- drivers/media/platform/qcom/venus/venc.c | 24 ++++++++++++++++++++---- 1 file changed, 20 insertions(+), 4 deletions(-)drivers/media/platform/qcom/venus/venc.c +20
-5@@ -479,10 +479,26 @@ venc_s_selection(struct file *file, void *fh, struct v4l2_selection *s)switch (s->target) {case V4L2_SEL_TGT_CROP : -if (s->r.width != inst->out_width || - s->r.height != inst->out_height || - s->r.top !=0 || s->r.left !=0 ) -return -EINVAL ; +if (s->r.left !=0 ) { + s->r.width += s->r.left; + s->r.left =0 ; + } + +if (s->r.top !=0 ) { + s->r.height += s->r.top; + s->r.top =0 ; + } + +if (s->r.width > inst->width) + s->r.width = inst->width; +else + inst->width = s->r.width; + +if (s->r.height > inst->height) + s->r.height = inst->height; +else + inst->height = s->r.height; +break ;default :return -EINVAL ; --2.25.0.265.gbab2e86ba0-goog
Signature · 2 lines
regards, Stan