[wip] Create parallel EncodingDeviceInterface #1212
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/meta-pytorch/torchcodec/1212
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 03db5c9 with merge base a9e1ae3 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
bot
added
the
CLA Signed
|
|
||
| PerGpuCache<AVBufferRef, Deleterp<AVBufferRef, void, av_buffer_unref>> | ||
| g_cached_hw_device_ctxs(MAX_CUDA_GPUS, MAX_CONTEXTS_PER_GPU_IN_CACHE); | ||
|
|
There was a problem hiding this comment.
This code is moved here from CudaDeviceInterface.cpp so that EncodingCudaDeviceInterface.cpp can reuse the context creation and caching.
| TORCH_CHECK( | ||
| avFrame != nullptr, | ||
| "convertTensorToAVFrame failed for frame ", | ||
| i, |
There was a problem hiding this comment.
Encoding interface handles both CPU and CUDA, and removes repeated conditionals:
if (frames_.device().is_cuda && deviceInterface_) { ... }
| getFFMPEGErrorStringFromErrorCode(status)); | ||
| } | ||
|
|
||
| UniqueAVFrame VideoEncoder::convertTensorToAVFrame( |
There was a problem hiding this comment.
CPU specific implementation is removed from Encoder.cpp.
|
|
||
| explicit EncodingDeviceInterfaceKey(torch::DeviceType type) | ||
| : deviceType(type) {} | ||
| }; |
There was a problem hiding this comment.
This is mostly copied over from DeviceInterface.h.
We could move these registration functions to a new header that is shared between EncodingDeviceInterface and DeviceInterface, but I'm not sure that's necessary. For now, I've added this parallel implementation using Encoding... types.
Alternatively, if we never expect to implement a BetaCuda variant for encoding, this can be simplified.
1 participant
No description provided.