Fix buffer exhaustion in the frames_decoder_gpu #4723
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
- frames_decoder_gpu keeps the internal buffer of the num_decode_surfaces_ length as it is the maximal number of frames that needs to be decoded to maintain the display order. The decoder accepts the raw packets as long as there is a free place in the buffer. In some cases, after receiving the packet two frames could be produced (as the N-1 frame may require N-the packet). In this case, we receive more frames and there is spare space. This PR adds functionality to enlarge the buffer if there is an additional frame to keep. Also, it makes sure we don't try to have more than num_decode_surfaces_ frames decoded at the time. Signed-off-by: Janusz Lisiecki <jlisiecki@nvidia.com>
|
!build |
|
CI MESSAGE: [7623615]: BUILD STARTED |
|
CI MESSAGE: [7623615]: BUILD PASSED |
awolant
approved these changes
Mar 18, 2023
Comment on lines
+736
to
+740
| std::vector<BufferedFrame> new_frame_buffer(frame_buffer_.size() + 1); | ||
| for (size_t i = 0; i < frame_buffer_.size(); ++i) { | ||
| new_frame_buffer[i] = std::move(frame_buffer_[i]); | ||
| } | ||
| frame_buffer_ = std::move(new_frame_buffer); |
There was a problem hiding this comment.
I was wondering why you decided to create new buffer and move it in place of the old one instead of push_back new frame to the old buffer?
There was a problem hiding this comment.
Because DeviceBuffer doesn't have assignment operator:
In file included from /Dali/dali/dali/operators/reader/loader/video/frames_decoder_gpu.h:35,
from /Dali/dali/dali/operators/reader/loader/video/frames_decoder_gpu.cc:14:
/Dali/dali/include/dali/core/dev_buffer.h: In instantiation of ‘dali::DeviceBuffer<T>::DeviceBuffer(dali::DeviceBuffer<T>&&) [with T = unsigned char]’:
/Dali/dali/dali/operators/reader/loader/video/frames_decoder_gpu.h:123:8: required from ‘void __gnu_cxx::new_allocator<_Tp>::construct(_Up*, _Args&& ...) [with _Up = dali::BufferedFrame; _Args = {dali::BufferedFrame}; _Tp = dali::BufferedFrame]’
/usr/include/c++/11/bits/alloc_traits.h:516:17: required from ‘static void std::allocator_traits<std::allocator<_Tp1> >::construct(std::allocator_traits<std::allocator<_Tp1> >::allocator_type&, _Up*, _Args&& ...) [with _Up = dali::BufferedFrame; _Args = {dali::BufferedFrame}; _Tp = dali::BufferedFrame; std::allocator_traits<std::allocator<_Tp1> >::allocator_type = std::allocator<dali::BufferedFrame>]’
/usr/include/c++/11/bits/vector.tcc:115:30: required from ‘std::vector<_Tp, _Alloc>::reference std::vector<_Tp, _Alloc>::emplace_back(_Args&& ...) [with _Args = {dali::BufferedFrame}; _Tp = dali::BufferedFrame; _Alloc = std::allocator<dali::BufferedFrame>; std::vector<_Tp, _Alloc>::reference = dali::BufferedFrame&]’
/usr/include/c++/11/bits/stl_vector.h:1204:21: required from ‘void std::vector<_Tp, _Alloc>::push_back(std::vector<_Tp, _Alloc>::value_type&&) [with _Tp = dali::BufferedFrame; _Alloc = std::allocator<dali::BufferedFrame>; std::vector<_Tp, _Alloc>::value_type = dali::BufferedFrame]’
/Dali/dali/dali/operators/reader/loader/video/frames_decoder_gpu.cc:736:26: required from here
/Dali/dali/include/dali/core/dev_buffer.h:73:11: error: use of deleted function ‘dali::DeviceBuffer<unsigned char>& dali::DeviceBuffer<unsigned char>::operator=(const dali::DeviceBuffer<unsigned char>&)’
73 | *this = other;
| ~~~~~~^~~~~~~
/Dali/dali/include/dali/core/dev_buffer.h:70:8: note: ‘dali::DeviceBuffer<unsigned char>& dali::DeviceBuffer<unsigned char>::operator=(const dali::DeviceBuffer<unsigned char>&)’ is implicitly declared as deleted because ‘dali::DeviceBuffer<unsigned char>’ declares a move constructor or move assignment operator
70 | struct DeviceBuffer {
| ^~~~~~~~~~~~
No idea why vector quires it and cannot use move operator insteada.
klecki
approved these changes
Mar 21, 2023
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
length as it is the maximal number of frames that needs to be decoded
to maintain the display order. The decoder accepts the raw packets as
long as there is a free place in the buffer. In some cases, after receiving
the packet two frames could be produced (as the N-1 frame may require N-the packet).
In this case, we receive more frames and there is spare space. This PR
adds functionality to enlarge the buffer if there is an additional frame
to keep. Also, it makes sure we don't try to have more than num_decode_surfaces_
frames decoded at the time.
Relates to #4498
Category:
Bug fix (non-breaking change which fixes an issue)
Description:
length as it is the maximal number of frames that needs to be decoded
to maintain the display order. The decoder accepts the raw packets as
long as there is a free place in the buffer. In some cases, after receiving
the packet two frames could be produced (as the N-1 frame may require N-the packet).
In this case, we receive more frames and there is spare space. This PR
adds functionality to enlarge the buffer if there is an additional frame
to keep. Also, it makes sure we don't try to have more than num_decode_surfaces_
frames decoded at the time.
Additional information:
Affected modules and functionalities:
Key points relevant for the review:
Tests:
Checklist
Documentation
DALI team only
Requirements
REQ IDs: N/A
JIRA TASK: N/A