This commit adds the case handling if the asynchronous execution
of a request fails by checking the exit status of the thread when
joining before starting another execution. On failure, it does the
cleanup as well.
Signed-off-by: Shubhanshu Saxena <shubhanshu.e01@gmail.com>
The frame allocation and filling the TaskItem with execution
parameters is common in the three backends. This commit shifts
this logic to dnn_backend_common.
Signed-off-by: Shubhanshu Saxena <shubhanshu.e01@gmail.com>
Since requests are running in parallel, there is inconsistency in
the status of the execution. To resolve it, we avoid using mutex
as it would result in single TF_Session running at a time. So add
TF_Status to the TFRequestItem
Signed-off-by: Shubhanshu Saxena <shubhanshu.e01@gmail.com>
This patch adds error handling for cases where the execute_model_tf
fails, clears the used memory in the TFRequestItem and finally pushes
it back to the request queue.
Signed-off-by: Shubhanshu Saxena <shubhanshu.e01@gmail.com>
This commit enables async execution in the TensorFlow backend
and adds function to flush extra frames.
The async execution mechanism executes the TFInferRequests on
a separate thread which is joined before the next execution of
same TFRequestItem/while freeing the model.
The following is the comparison of this mechanism with the existing
sync mechanism on TensorFlow C API 2.5 CPU variant.
Async Mode: 4m32.846s
Sync Mode: 5m17.582s
The above was performed on super resolution filter using SRCNN model.
Signed-off-by: Shubhanshu Saxena <shubhanshu.e01@gmail.com>
This commit adds a function for execution of TFInferRequest and documentation
for functions related to TFInferRequest.
Signed-off-by: Shubhanshu Saxena <shubhanshu.e01@gmail.com>
This commit adds an async execution mechanism for common use
in the TensorFlow and Native backends.
This commit also adds the documentation of typedefs and functions in
the async module for common use in DNN backends.
Signed-off-by: Shubhanshu Saxena <shubhanshu.e01@gmail.com>
The function ff_qsvvpp_filter_frame should return a FFmpeg error code if
there is an error. However it might return a SDK error code without this
patch.
Reviewed-by: Soft Works <softworkz@hotmail.com>
The flite filter apparently only wanted to declare a struct,
but mistakenly also defined an unused and zero-initialized element
with external linkage.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
The function definition used float *data_buf[14], although there are
only 13 elements (and only 13 are used); the declaration used 13.
Given that the type will be converted to float **data_buf anyway,
this is not in violation of the C specs, but nevertheless a bug.
GCC 11 has a new warning for this -Warray-parameter.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
The reasons for including them don't exist any longer: ff_tlog() has
been moved to libavutil/internal.h and FF_QSCALE_TYPE_* has been moved
to qp_table.h.
Reviewed-by: Nicolas George <george@nsup.org>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
It is the natural header for it.
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
It is unused since 02aa0701ae.
The corresponding size field is write-only since then.
Reviewed-by: Nicolas George <george@nsup.org>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
8b83dad825 added another potentially used
video enhancement filter without increasing a define for the number of
such options which is used as the size of stack array. This can lead to
a buffer overrun if all filters are used simultaneously. So increase
said number.
Fixes Coverity ticket #1489775.
Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
They are not used by the header at all and only used by very few files;
so include the headers in their users instead of in internal.h.
Reviewed-by: Nicolas George <george@nsup.org>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
The magic constants come from the unofficial "ITU-R BS.1770-1 filter
specifications"¹ by Raiden (libebur128) which relies on "Parameter
Quantization in Direct-Form Recursive Audio Filters"² by Brian
Neunaber.
The constants seem to include a quantization bias, for example:
- Vb is supposed to be exactly √Vh in a high shelf filter
- the Pre-filter Gain should likely be 4dB
- Pre Q and RLB Q are respectively very close to √½ and ½
Those are not adjusted to prevent the values from drifting away from
the official specifications.
An alternative to this approach would be to requantize on the fly as
proposed by pbelkner³, where the 48kHz code path would use the exact
specifications constants while derivating constants for other
frequencies.
[1]: https://www.scribd.com/document/49991813/ITU-R-BS-1770-1-filters
[2]: https://www.scribd.com/document/6531763/Direct-Form-Filter-Parameter-Quantization
[3]: https://hydrogenaud.io/index.php?topic=86116.msg740092#msg740092
Andreas Rheinhardt
Paul B Mahol
Shubhanshu Saxena
Haihao Xiang
James Almer
Linjie Fu
Martin Storsjö
Fei Wang