VideoDecoder -> SingleStreamDecoder (#611)

Author: NicolasHug

src/torchcodec/_core/AVIOContextHolder.h

@@ -32,7 +32,7 @@ namespace facebook::torchcodec {
 //      createAVIOContext(), ideally in their constructor.
 //  3. A generic handle for those that just need to manage having access to an
 //     AVIOContext, but aren't necessarily concerned with how it was customized:
-//     typically, the VideoDecoder.
+//     typically, the SingleStreamDecoder.
 class AVIOContextHolder {
  public:
   virtual ~AVIOContextHolder();

src/torchcodec/_core/CMakeLists.txt

@@ -60,7 +60,7 @@ function(make_torchcodec_libraries
     set(decoder_sources
         AVIOContextHolder.cpp
         FFMPEGCommon.cpp
-        VideoDecoder.cpp
+        SingleStreamDecoder.cpp
     )
 
     if(ENABLE_CUDA)

src/torchcodec/_core/CPUOnlyDevice.cpp

@@ -16,9 +16,10 @@ namespace facebook::torchcodec {
 
 void convertAVFrameToFrameOutputOnCuda(
     const torch::Device& device,
-    [[maybe_unused]] const VideoDecoder::VideoStreamOptions& videoStreamOptions,
+    [[maybe_unused]] const SingleStreamDecoder::VideoStreamOptions&
+        videoStreamOptions,
     [[maybe_unused]] UniqueAVFrame& avFrame,
-    [[maybe_unused]] VideoDecoder::FrameOutput& frameOutput,
+    [[maybe_unused]] SingleStreamDecoder::FrameOutput& frameOutput,
     [[maybe_unused]] std::optional<torch::Tensor> preAllocatedOutputTensor) {
   throwUnsupportedDeviceError(device);
 }

src/torchcodec/_core/CudaDevice.cpp

@@ -6,7 +6,7 @@
 
 #include "src/torchcodec/_core/DeviceInterface.h"
 #include "src/torchcodec/_core/FFMPEGCommon.h"
-#include "src/torchcodec/_core/VideoDecoder.h"
+#include "src/torchcodec/_core/SingleStreamDecoder.h"
 
 extern "C" {
 #include <libavutil/hwcontext_cuda.h>
@@ -20,9 +20,11 @@ namespace {
 // creating a cuda context is expensive. The cache mechanism is as follows:
 // 1. There is a cache of size MAX_CONTEXTS_PER_GPU_IN_CACHE cuda contexts for
 //    each GPU.
-// 2. When we destroy a VideoDecoder instance we release the cuda context to
+// 2. When we destroy a SingleStreamDecoder instance we release the cuda context
+// to
 //    the cache if the cache is not full.
-// 3. When we create a VideoDecoder instance we try to get a cuda context from
+// 3. When we create a SingleStreamDecoder instance we try to get a cuda context
+// from
 //    the cache. If the cache is empty we create a new cuda context.
 
 // Pytorch can only handle up to 128 GPUs.
@@ -189,9 +191,9 @@ void initializeContextOnCuda(
 
 void convertAVFrameToFrameOutputOnCuda(
     const torch::Device& device,
-    const VideoDecoder::VideoStreamOptions& videoStreamOptions,
+    const SingleStreamDecoder::VideoStreamOptions& videoStreamOptions,
     UniqueAVFrame& avFrame,
-    VideoDecoder::FrameOutput& frameOutput,
+    SingleStreamDecoder::FrameOutput& frameOutput,
     std::optional<torch::Tensor> preAllocatedOutputTensor) {
   TORCH_CHECK(
       avFrame->format == AV_PIX_FMT_CUDA,

src/torchcodec/_core/DeviceInterface.h

@@ -11,13 +11,13 @@
 #include <stdexcept>
 #include <string>
 #include "FFMPEGCommon.h"
-#include "src/torchcodec/_core/VideoDecoder.h"
+#include "src/torchcodec/_core/SingleStreamDecoder.h"
 
 namespace facebook::torchcodec {
 
 // Note that all these device functions should only be called if the device is
 // not a CPU device. CPU device functions are already implemented in the
-// VideoDecoder implementation.
+// SingleStreamDecoder implementation.
 // These functions should only be called from within an if block like this:
 // if (device.type() != torch::kCPU) {
 //   deviceFunction(device, ...);
@@ -31,9 +31,9 @@ void initializeContextOnCuda(
 
 void convertAVFrameToFrameOutputOnCuda(
     const torch::Device& device,
-    const VideoDecoder::VideoStreamOptions& videoStreamOptions,
+    const SingleStreamDecoder::VideoStreamOptions& videoStreamOptions,
     UniqueAVFrame& avFrame,
-    VideoDecoder::FrameOutput& frameOutput,
+    SingleStreamDecoder::FrameOutput& frameOutput,
     std::optional<torch::Tensor> preAllocatedOutputTensor = std::nullopt);
 
 void releaseContextOnCuda(