| Message ID | 20191003101038.13732-1-stanimir.varbanov@linaro.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v4] venus: venc: Fix enum frameintervals | expand |
I have tested this on db410c with following gst pipeline: gst-launch-1.0 -v videotestsrc ! video/x-raw,format=NV12,width=1280,height=960,framerate=24/1 ! v4l2h264enc extra-controls="controls,h264_profile=4,h264_level="5",video_bitrate=10000000;" ! filesink location=gstenc.h264 Loic, could you give it a try on db820c too? Here is the info on the bug which I try to fix with current patch: https://bugs.96boards.org/show_bug.cgi?id=513 On 10/3/19 1:10 PM, Stanimir Varbanov wrote: > This fixes an issue when setting the encoder framerate because of > missing precision. Now the frameinterval type is changed to > TYPE_CONTINUOUS and step = 1. Also the math is changed when > framerate property is called - the firmware side expects the > framerate in Q16 values. > > Signed-off-by: Stanimir Varbanov <stanimir.varbanov@linaro.org> > --- > > Changes since v3: > Keep min/max numerator one, and divide frate(max/min) to frame > factor (returned framerate max/min capabilities are in range > 1 to 120fps but in Q16 i.e. 65536 to 7864320). > > drivers/media/platform/qcom/venus/venc.c | 17 ++++++++++++----- > 1 file changed, 12 insertions(+), 5 deletions(-) > > diff --git a/drivers/media/platform/qcom/venus/venc.c b/drivers/media/platform/qcom/venus/venc.c > index 1b7fb2d5887c..133ff7eceb83 100644 > --- a/drivers/media/platform/qcom/venus/venc.c > +++ b/drivers/media/platform/qcom/venus/venc.c > @@ -22,6 +22,7 @@ > #include "venc.h" > > #define NUM_B_FRAMES_MAX 4 > +#define FRAMERATE_FACTOR BIT(16) > > /* > * Three resons to keep MPLANE formats (despite that the number of planes > @@ -576,7 +577,7 @@ static int venc_enum_frameintervals(struct file *file, void *fh, > struct venus_inst *inst = to_inst(file); > const struct venus_format *fmt; > > - fival->type = V4L2_FRMIVAL_TYPE_STEPWISE; > + fival->type = V4L2_FRMIVAL_TYPE_CONTINUOUS; > > fmt = find_format(inst, fival->pixel_format, > V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE); > @@ -600,11 +601,11 @@ static int venc_enum_frameintervals(struct file *file, void *fh, > return -EINVAL; > > fival->stepwise.min.numerator = 1; > - fival->stepwise.min.denominator = frate_max(inst); > + fival->stepwise.min.denominator = frate_max(inst) / FRAMERATE_FACTOR; > fival->stepwise.max.numerator = 1; > - fival->stepwise.max.denominator = frate_min(inst); > + fival->stepwise.max.denominator = frate_min(inst) / FRAMERATE_FACTOR; > fival->stepwise.step.numerator = 1; > - fival->stepwise.step.denominator = frate_max(inst); > + fival->stepwise.step.denominator = 1; > > return 0; > } > @@ -649,6 +650,7 @@ static int venc_set_properties(struct venus_inst *inst) > struct hfi_quantization quant; > struct hfi_quantization_range quant_range; > u32 ptype, rate_control, bitrate, profile = 0, level = 0; > + u64 framerate; > int ret; > > ret = venus_helper_set_work_mode(inst, VIDC_WORK_MODE_2); > @@ -659,9 +661,14 @@ static int venc_set_properties(struct venus_inst *inst) > if (ret) > return ret; > > + framerate = inst->timeperframe.denominator * FRAMERATE_FACTOR; > + /* next line is to round up */ > + framerate += inst->timeperframe.numerator - 1; > + do_div(framerate, inst->timeperframe.numerator); > + > ptype = HFI_PROPERTY_CONFIG_FRAME_RATE; > frate.buffer_type = HFI_BUFFER_OUTPUT; > - frate.framerate = inst->fps * (1 << 16); > + frate.framerate = framerate; > > ret = hfi_session_set_property(inst, ptype, &frate); > if (ret) >
Em Thu, 3 Oct 2019 13:10:38 +0300 Stanimir Varbanov <stanimir.varbanov@linaro.org> escreveu: > This fixes an issue when setting the encoder framerate because of > missing precision. Now the frameinterval type is changed to > TYPE_CONTINUOUS and step = 1. Also the math is changed when > framerate property is called - the firmware side expects the > framerate in Q16 values. > > Signed-off-by: Stanimir Varbanov <stanimir.varbanov@linaro.org> Didn't test the patch, but just reviewing the code, this version looks OK on my eyes. > --- > > Changes since v3: > Keep min/max numerator one, and divide frate(max/min) to frame > factor (returned framerate max/min capabilities are in range > 1 to 120fps but in Q16 i.e. 65536 to 7864320). > > drivers/media/platform/qcom/venus/venc.c | 17 ++++++++++++----- > 1 file changed, 12 insertions(+), 5 deletions(-) > > diff --git a/drivers/media/platform/qcom/venus/venc.c b/drivers/media/platform/qcom/venus/venc.c > index 1b7fb2d5887c..133ff7eceb83 100644 > --- a/drivers/media/platform/qcom/venus/venc.c > +++ b/drivers/media/platform/qcom/venus/venc.c > @@ -22,6 +22,7 @@ > #include "venc.h" > > #define NUM_B_FRAMES_MAX 4 > +#define FRAMERATE_FACTOR BIT(16) > > /* > * Three resons to keep MPLANE formats (despite that the number of planes > @@ -576,7 +577,7 @@ static int venc_enum_frameintervals(struct file *file, void *fh, > struct venus_inst *inst = to_inst(file); > const struct venus_format *fmt; > > - fival->type = V4L2_FRMIVAL_TYPE_STEPWISE; > + fival->type = V4L2_FRMIVAL_TYPE_CONTINUOUS; > > fmt = find_format(inst, fival->pixel_format, > V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE); > @@ -600,11 +601,11 @@ static int venc_enum_frameintervals(struct file *file, void *fh, > return -EINVAL; > > fival->stepwise.min.numerator = 1; > - fival->stepwise.min.denominator = frate_max(inst); > + fival->stepwise.min.denominator = frate_max(inst) / FRAMERATE_FACTOR; > fival->stepwise.max.numerator = 1; > - fival->stepwise.max.denominator = frate_min(inst); > + fival->stepwise.max.denominator = frate_min(inst) / FRAMERATE_FACTOR; > fival->stepwise.step.numerator = 1; > - fival->stepwise.step.denominator = frate_max(inst); > + fival->stepwise.step.denominator = 1; > > return 0; > } > @@ -649,6 +650,7 @@ static int venc_set_properties(struct venus_inst *inst) > struct hfi_quantization quant; > struct hfi_quantization_range quant_range; > u32 ptype, rate_control, bitrate, profile = 0, level = 0; > + u64 framerate; > int ret; > > ret = venus_helper_set_work_mode(inst, VIDC_WORK_MODE_2); > @@ -659,9 +661,14 @@ static int venc_set_properties(struct venus_inst *inst) > if (ret) > return ret; > > + framerate = inst->timeperframe.denominator * FRAMERATE_FACTOR; > + /* next line is to round up */ > + framerate += inst->timeperframe.numerator - 1; > + do_div(framerate, inst->timeperframe.numerator); > + > ptype = HFI_PROPERTY_CONFIG_FRAME_RATE; > frate.buffer_type = HFI_BUFFER_OUTPUT; > - frate.framerate = inst->fps * (1 << 16); > + frate.framerate = framerate; > > ret = hfi_session_set_property(inst, ptype, &frate); > if (ret) Thanks, Mauro
Hi Stanimir, On Thu, 3 Oct 2019 at 12:15, Stanimir Varbanov <stanimir.varbanov@linaro.org> wrote: > > I have tested this on db410c with following gst pipeline: > > gst-launch-1.0 -v videotestsrc ! > video/x-raw,format=NV12,width=1280,height=960,framerate=24/1 ! > v4l2h264enc > extra-controls="controls,h264_profile=4,h264_level="5",video_bitrate=10000000;" > ! filesink location=gstenc.h264 > > Loic, could you give it a try on db820c too? > > Here is the info on the bug which I try to fix with current patch: > > https://bugs.96boards.org/show_bug.cgi?id=513 > > On 10/3/19 1:10 PM, Stanimir Varbanov wrote: > > This fixes an issue when setting the encoder framerate because of > > missing precision. Now the frameinterval type is changed to > > TYPE_CONTINUOUS and step = 1. Also the math is changed when > > framerate property is called - the firmware side expects the > > framerate in Q16 values. > > > > Signed-off-by: Stanimir Varbanov <stanimir.varbanov@linaro.org> > > --- > > > > Changes since v3: > > Keep min/max numerator one, and divide frate(max/min) to frame > > factor (returned framerate max/min capabilities are in range > > 1 to 120fps but in Q16 i.e. 65536 to 7864320). > > > > drivers/media/platform/qcom/venus/venc.c | 17 ++++++++++++----- > > 1 file changed, 12 insertions(+), 5 deletions(-) > > > > diff --git a/drivers/media/platform/qcom/venus/venc.c b/drivers/media/platform/qcom/venus/venc.c > > index 1b7fb2d5887c..133ff7eceb83 100644 > > --- a/drivers/media/platform/qcom/venus/venc.c > > +++ b/drivers/media/platform/qcom/venus/venc.c > > @@ -22,6 +22,7 @@ > > #include "venc.h" > > > > #define NUM_B_FRAMES_MAX 4 > > +#define FRAMERATE_FACTOR BIT(16) > > > > /* > > * Three resons to keep MPLANE formats (despite that the number of planes > > @@ -576,7 +577,7 @@ static int venc_enum_frameintervals(struct file *file, void *fh, > > struct venus_inst *inst = to_inst(file); > > const struct venus_format *fmt; > > > > - fival->type = V4L2_FRMIVAL_TYPE_STEPWISE; > > + fival->type = V4L2_FRMIVAL_TYPE_CONTINUOUS; > > > > fmt = find_format(inst, fival->pixel_format, > > V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE); > > @@ -600,11 +601,11 @@ static int venc_enum_frameintervals(struct file *file, void *fh, > > return -EINVAL; > > > > fival->stepwise.min.numerator = 1; > > - fival->stepwise.min.denominator = frate_max(inst); > > + fival->stepwise.min.denominator = frate_max(inst) / FRAMERATE_FACTOR; On 820c frate_max() returns 120 set denominator to 0, and causes gstreamer failure. Regards, Loic
Hi Loic, On 10/17/19 6:08 PM, Loic Poulain wrote: > Hi Stanimir, > > On Thu, 3 Oct 2019 at 12:15, Stanimir Varbanov > <stanimir.varbanov@linaro.org> wrote: >> >> I have tested this on db410c with following gst pipeline: >> >> gst-launch-1.0 -v videotestsrc ! >> video/x-raw,format=NV12,width=1280,height=960,framerate=24/1 ! >> v4l2h264enc >> extra-controls="controls,h264_profile=4,h264_level="5",video_bitrate=10000000;" >> ! filesink location=gstenc.h264 >> >> Loic, could you give it a try on db820c too? >> >> Here is the info on the bug which I try to fix with current patch: >> >> https://bugs.96boards.org/show_bug.cgi?id=513 >> >> On 10/3/19 1:10 PM, Stanimir Varbanov wrote: >>> This fixes an issue when setting the encoder framerate because of >>> missing precision. Now the frameinterval type is changed to >>> TYPE_CONTINUOUS and step = 1. Also the math is changed when >>> framerate property is called - the firmware side expects the >>> framerate in Q16 values. >>> >>> Signed-off-by: Stanimir Varbanov <stanimir.varbanov@linaro.org> >>> --- >>> >>> Changes since v3: >>> Keep min/max numerator one, and divide frate(max/min) to frame >>> factor (returned framerate max/min capabilities are in range >>> 1 to 120fps but in Q16 i.e. 65536 to 7864320). >>> >>> drivers/media/platform/qcom/venus/venc.c | 17 ++++++++++++----- >>> 1 file changed, 12 insertions(+), 5 deletions(-) >>> >>> diff --git a/drivers/media/platform/qcom/venus/venc.c b/drivers/media/platform/qcom/venus/venc.c >>> index 1b7fb2d5887c..133ff7eceb83 100644 >>> --- a/drivers/media/platform/qcom/venus/venc.c >>> +++ b/drivers/media/platform/qcom/venus/venc.c >>> @@ -22,6 +22,7 @@ >>> #include "venc.h" >>> >>> #define NUM_B_FRAMES_MAX 4 >>> +#define FRAMERATE_FACTOR BIT(16) >>> >>> /* >>> * Three resons to keep MPLANE formats (despite that the number of planes >>> @@ -576,7 +577,7 @@ static int venc_enum_frameintervals(struct file *file, void *fh, >>> struct venus_inst *inst = to_inst(file); >>> const struct venus_format *fmt; >>> >>> - fival->type = V4L2_FRMIVAL_TYPE_STEPWISE; >>> + fival->type = V4L2_FRMIVAL_TYPE_CONTINUOUS; >>> >>> fmt = find_format(inst, fival->pixel_format, >>> V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE); >>> @@ -600,11 +601,11 @@ static int venc_enum_frameintervals(struct file *file, void *fh, >>> return -EINVAL; >>> >>> fival->stepwise.min.numerator = 1; >>> - fival->stepwise.min.denominator = frate_max(inst); >>> + fival->stepwise.min.denominator = frate_max(inst) / FRAMERATE_FACTOR; > > On 820c frate_max() returns 120 set denominator to 0, and causes > gstreamer failure. OK, thanks! We have two options - unify frate_min/max() to return in Q16 depending on the hfi version - or move frame_factor in frate_min/max() and return the framerate (1..120)
Hi Stanimir, On Thu, 17 Oct 2019 at 17:47, Stanimir Varbanov <stanimir.varbanov@linaro.org> wrote: > > Hi Loic, > > On 10/17/19 6:08 PM, Loic Poulain wrote: > > Hi Stanimir, > > > > On Thu, 3 Oct 2019 at 12:15, Stanimir Varbanov > > <stanimir.varbanov@linaro.org> wrote: > >> > >> I have tested this on db410c with following gst pipeline: > >> > >> gst-launch-1.0 -v videotestsrc ! > >> video/x-raw,format=NV12,width=1280,height=960,framerate=24/1 ! > >> v4l2h264enc > >> extra-controls="controls,h264_profile=4,h264_level="5",video_bitrate=10000000;" > >> ! filesink location=gstenc.h264 > >> > >> Loic, could you give it a try on db820c too? > >> > >> Here is the info on the bug which I try to fix with current patch: > >> > >> https://bugs.96boards.org/show_bug.cgi?id=513 > >> > >> On 10/3/19 1:10 PM, Stanimir Varbanov wrote: > >>> This fixes an issue when setting the encoder framerate because of > >>> missing precision. Now the frameinterval type is changed to > >>> TYPE_CONTINUOUS and step = 1. Also the math is changed when > >>> framerate property is called - the firmware side expects the > >>> framerate in Q16 values. > >>> > >>> Signed-off-by: Stanimir Varbanov <stanimir.varbanov@linaro.org> > >>> --- > >>> > >>> Changes since v3: > >>> Keep min/max numerator one, and divide frate(max/min) to frame > >>> factor (returned framerate max/min capabilities are in range > >>> 1 to 120fps but in Q16 i.e. 65536 to 7864320). > >>> > >>> drivers/media/platform/qcom/venus/venc.c | 17 ++++++++++++----- > >>> 1 file changed, 12 insertions(+), 5 deletions(-) > >>> > >>> diff --git a/drivers/media/platform/qcom/venus/venc.c b/drivers/media/platform/qcom/venus/venc.c > >>> index 1b7fb2d5887c..133ff7eceb83 100644 > >>> --- a/drivers/media/platform/qcom/venus/venc.c > >>> +++ b/drivers/media/platform/qcom/venus/venc.c > >>> @@ -22,6 +22,7 @@ > >>> #include "venc.h" > >>> > >>> #define NUM_B_FRAMES_MAX 4 > >>> +#define FRAMERATE_FACTOR BIT(16) > >>> > >>> /* > >>> * Three resons to keep MPLANE formats (despite that the number of planes > >>> @@ -576,7 +577,7 @@ static int venc_enum_frameintervals(struct file *file, void *fh, > >>> struct venus_inst *inst = to_inst(file); > >>> const struct venus_format *fmt; > >>> > >>> - fival->type = V4L2_FRMIVAL_TYPE_STEPWISE; > >>> + fival->type = V4L2_FRMIVAL_TYPE_CONTINUOUS; > >>> > >>> fmt = find_format(inst, fival->pixel_format, > >>> V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE); > >>> @@ -600,11 +601,11 @@ static int venc_enum_frameintervals(struct file *file, void *fh, > >>> return -EINVAL; > >>> > >>> fival->stepwise.min.numerator = 1; > >>> - fival->stepwise.min.denominator = frate_max(inst); > >>> + fival->stepwise.min.denominator = frate_max(inst) / FRAMERATE_FACTOR; > > > > On 820c frate_max() returns 120 set denominator to 0, and causes > > gstreamer failure. > > OK, thanks! > > We have two options > - unify frate_min/max() to return in Q16 depending on the hfi version > - or move frame_factor in frate_min/max() and return the framerate (1..120) No strong preference... Regards, Loic
diff --git a/drivers/media/platform/qcom/venus/venc.c b/drivers/media/platform/qcom/venus/venc.c index 1b7fb2d5887c..133ff7eceb83 100644 --- a/drivers/media/platform/qcom/venus/venc.c +++ b/drivers/media/platform/qcom/venus/venc.c @@ -22,6 +22,7 @@ #include "venc.h" #define NUM_B_FRAMES_MAX 4 +#define FRAMERATE_FACTOR BIT(16) /* * Three resons to keep MPLANE formats (despite that the number of planes @@ -576,7 +577,7 @@ static int venc_enum_frameintervals(struct file *file, void *fh, struct venus_inst *inst = to_inst(file); const struct venus_format *fmt; - fival->type = V4L2_FRMIVAL_TYPE_STEPWISE; + fival->type = V4L2_FRMIVAL_TYPE_CONTINUOUS; fmt = find_format(inst, fival->pixel_format, V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE); @@ -600,11 +601,11 @@ static int venc_enum_frameintervals(struct file *file, void *fh, return -EINVAL; fival->stepwise.min.numerator = 1; - fival->stepwise.min.denominator = frate_max(inst); + fival->stepwise.min.denominator = frate_max(inst) / FRAMERATE_FACTOR; fival->stepwise.max.numerator = 1; - fival->stepwise.max.denominator = frate_min(inst); + fival->stepwise.max.denominator = frate_min(inst) / FRAMERATE_FACTOR; fival->stepwise.step.numerator = 1; - fival->stepwise.step.denominator = frate_max(inst); + fival->stepwise.step.denominator = 1; return 0; } @@ -649,6 +650,7 @@ static int venc_set_properties(struct venus_inst *inst) struct hfi_quantization quant; struct hfi_quantization_range quant_range; u32 ptype, rate_control, bitrate, profile = 0, level = 0; + u64 framerate; int ret; ret = venus_helper_set_work_mode(inst, VIDC_WORK_MODE_2); @@ -659,9 +661,14 @@ static int venc_set_properties(struct venus_inst *inst) if (ret) return ret; + framerate = inst->timeperframe.denominator * FRAMERATE_FACTOR; + /* next line is to round up */ + framerate += inst->timeperframe.numerator - 1; + do_div(framerate, inst->timeperframe.numerator); + ptype = HFI_PROPERTY_CONFIG_FRAME_RATE; frate.buffer_type = HFI_BUFFER_OUTPUT; - frate.framerate = inst->fps * (1 << 16); + frate.framerate = framerate; ret = hfi_session_set_property(inst, ptype, &frate); if (ret)
This fixes an issue when setting the encoder framerate because of missing precision. Now the frameinterval type is changed to TYPE_CONTINUOUS and step = 1. Also the math is changed when framerate property is called - the firmware side expects the framerate in Q16 values. Signed-off-by: Stanimir Varbanov <stanimir.varbanov@linaro.org> --- Changes since v3: Keep min/max numerator one, and divide frate(max/min) to frame factor (returned framerate max/min capabilities are in range 1 to 120fps but in Q16 i.e. 65536 to 7864320). drivers/media/platform/qcom/venus/venc.c | 17 ++++++++++++----- 1 file changed, 12 insertions(+), 5 deletions(-)