Fix StreamWriter regression around RGB0/BGR0 (#3428)

Summary:
- Add RGB0/BGR0 support to CPU encoder
- Allow to pass RGB/BGR when expectged format is RGB0/BGR0

Pull Request resolved: https://github.com/pytorch/audio/pull/3428

Differential Revision: D47274370

Pulled By: mthrok

fbshipit-source-id: d34d940e04b07673bb86f518fe895c0735912444

torchaudio/csrc/ffmpeg/stream_writer/encode_process.cpp

@@ -96,33 +96,30 @@ enum AVSampleFormat get_src_sample_fmt(const std::string& src) {
       ".");
 }
 
+const std::set<AVPixelFormat> SUPPORTED_PIX_FMTS{
+    AV_PIX_FMT_GRAY8,
+    AV_PIX_FMT_RGB0,
+    AV_PIX_FMT_BGR0,
+    AV_PIX_FMT_RGB24,
+    AV_PIX_FMT_BGR24,
+    AV_PIX_FMT_YUV444P};
+
 enum AVPixelFormat get_src_pix_fmt(const std::string& src) {
   AVPixelFormat fmt = av_get_pix_fmt(src.c_str());
-  switch (fmt) {
-    case AV_PIX_FMT_GRAY8:
-    case AV_PIX_FMT_RGB24:
-    case AV_PIX_FMT_BGR24:
-    case AV_PIX_FMT_YUV444P:
-      return fmt;
-    default:;
-  }
   TORCH_CHECK(
-      false,
+      SUPPORTED_PIX_FMTS.count(fmt),
       "Unsupported pixel format (",
       src,
       ") was provided. Valid values are ",
       []() -> std::string {
         std::vector<std::string> ret;
-        for (const auto& fmt :
-             {AV_PIX_FMT_GRAY8,
-              AV_PIX_FMT_RGB24,
-              AV_PIX_FMT_BGR24,
-              AV_PIX_FMT_YUV444P}) {
+        for (const auto& fmt : SUPPORTED_PIX_FMTS) {
           ret.emplace_back(av_get_pix_fmt_name(fmt));
         }
         return c10::Join(", ", ret);
       }(),
       ".");
+  return fmt;
 }
 
 ////////////////////////////////////////////////////////////////////////////////

torchaudio/csrc/ffmpeg/stream_writer/tensor_converter.cpp

@@ -8,6 +8,8 @@ namespace torchaudio::io {
 
 namespace {
 
+using namespace torch::indexing;
+
 using InitFunc = TensorConverter::InitFunc;
 using ConvertFunc = TensorConverter::ConvertFunc;
 
@@ -111,6 +113,28 @@ void validate_video_input(
       t.sizes());
 }
 
+// Special case where encode pixel format is RGB0/BGR0 but the tensor is RGB/BGR
+void validate_rgb0(const torch::Tensor& t, AVFrame* buffer) {
+  if (buffer->hw_frames_ctx) {
+    TORCH_CHECK(t.device().is_cuda(), "Input tensor has to be on CUDA.");
+  } else {
+    TORCH_CHECK(t.device().is_cpu(), "Input tensor has to be on CPU.");
+  }
+  TORCH_CHECK(
+      t.dtype().toScalarType() == c10::ScalarType::Byte,
+      "Expected Tensor of uint8 type.");
+
+  TORCH_CHECK(t.dim() == 4, "Input Tensor has to be 4D.");
+  TORCH_CHECK(
+      t.size(2) == buffer->height && t.size(3) == buffer->width,
+      "Expected tensor with shape (N, 3, ",
+      buffer->height,
+      ", ",
+      buffer->width,
+      ") (NCHW format). Found ",
+      t.sizes());
+}
+
 // NCHW ->NHWC, ensure contiguous
 torch::Tensor init_interlaced(const torch::Tensor& tensor) {
   TORCH_INTERNAL_ASSERT_DEBUG_ONLY(tensor.dim() == 4);
@@ -276,16 +300,20 @@ std::pair<InitFunc, ConvertFunc> get_video_func(AVFrame* buffer) {
     auto frames_ctx = (AVHWFramesContext*)(buffer->hw_frames_ctx->data);
     auto sw_pix_fmt = frames_ctx->sw_format;
     switch (sw_pix_fmt) {
-      // Note:
-      // RGB0 / BGR0 expects 4 channel, but neither
-      // av_pix_fmt_desc_get(pix_fmt)->nb_components
-      // or av_pix_fmt_count_planes(pix_fmt) returns 4.
       case AV_PIX_FMT_RGB0:
       case AV_PIX_FMT_BGR0: {
         ConvertFunc convert_func = [](const torch::Tensor& t, AVFrame* f) {
           write_interlaced_video_cuda(t, f, 4);
         };
         InitFunc init_func = [](const torch::Tensor& t, AVFrame* f) {
+          // Special treatment for the case user pass regular RGB/BGR tensor.
+          if (t.dim() == 4 && t.size(1) == 3) {
+            validate_rgb0(t, f);
+            auto tmp =
+                torch::empty({t.size(0), t.size(2), t.size(3), 4}, t.options());
+            tmp.index_put_({"...", Slice(0, 3)}, t.permute({0, 2, 3, 1}));
+            return tmp;
+          }
           validate_video_input(t, f, 4);
           return init_interlaced(t);
         };
@@ -327,6 +355,24 @@ std::pair<InitFunc, ConvertFunc> get_video_func(AVFrame* buffer) {
       };
       return {init_func, convert_func};
     }
+    case AV_PIX_FMT_RGB0:
+    case AV_PIX_FMT_BGR0: {
+      InitFunc init_func = [](const torch::Tensor& t, AVFrame* f) {
+        if (t.dim() == 4 && t.size(1) == 3) {
+          validate_rgb0(t, f);
+          auto tmp =
+              torch::empty({t.size(0), t.size(2), t.size(3), 4}, t.options());
+          tmp.index_put_({"...", Slice(0, 3)}, t.permute({0, 2, 3, 1}));
+          return tmp;
+        }
+        validate_video_input(t, f, 4);
+        return init_interlaced(t);
+      };
+      ConvertFunc convert_func = [](const torch::Tensor& t, AVFrame* f) {
+        write_interlaced_video(t, f, 4);
+      };
+      return {init_func, convert_func};
+    }
     case AV_PIX_FMT_YUV444P: {
       InitFunc init_func = [](const torch::Tensor& t, AVFrame* f) {
         validate_video_input(t, f, 3);