BETA CUDA interface: separate it from the default interface (#931)

Author: NicolasHug

src/torchcodec/_core/BetaCudaDeviceInterface.cpp

@@ -202,6 +202,12 @@ BetaCudaDeviceInterface::BetaCudaDeviceInterface(const torch::Device& device)
   TORCH_CHECK(g_cuda_beta, "BetaCudaDeviceInterface was not registered!");
   TORCH_CHECK(
       device_.type() == torch::kCUDA, "Unsupported device: ", device_.str());
+
+  // Initialize CUDA context with a dummy tensor
+  torch::Tensor dummyTensorForCudaInitialization = torch::empty(
+      {1}, torch::TensorOptions().dtype(torch::kUInt8).device(device_));
+
+  nppCtx_ = getNppStreamContext(device_);
 }
 
 BetaCudaDeviceInterface::~BetaCudaDeviceInterface() {
@@ -222,21 +228,13 @@ BetaCudaDeviceInterface::~BetaCudaDeviceInterface() {
     cuvidDestroyVideoParser(videoParser_);
     videoParser_ = nullptr;
   }
+
+  returnNppStreamContextToCache(device_, std::move(nppCtx_));
 }
 
 void BetaCudaDeviceInterface::initialize(
     const AVStream* avStream,
     const UniqueDecodingAVFormatContext& avFormatCtx) {
-  torch::Tensor dummyTensorForCudaInitialization = torch::empty(
-      {1}, torch::TensorOptions().dtype(torch::kUInt8).device(device_));
-
-  auto cudaDevice = torch::Device(torch::kCUDA);
-  defaultCudaInterface_ =
-      std::unique_ptr<DeviceInterface>(createDeviceInterface(cudaDevice));
-  AVCodecContext dummyCodecContext = {};
-  defaultCudaInterface_->initialize(avStream, avFormatCtx);
-  defaultCudaInterface_->registerHardwareDeviceWithCodec(&dummyCodecContext);
-
   TORCH_CHECK(avStream != nullptr, "AVStream cannot be null");
   timeBase_ = avStream->time_base;
   frameRateAvgFromFFmpeg_ = avStream->r_frame_rate;
@@ -623,15 +621,19 @@ void BetaCudaDeviceInterface::convertAVFrameToFrameOutput(
     UniqueAVFrame& avFrame,
     FrameOutput& frameOutput,
     std::optional<torch::Tensor> preAllocatedOutputTensor) {
+  // TODONVDEC P2: we may need to handle 10bit videos the same way the default
+  // interface does it with maybeConvertAVFrameToNV12OrRGB24().
   TORCH_CHECK(
       avFrame->format == AV_PIX_FMT_CUDA,
       "Expected CUDA format frame from BETA CUDA interface");
 
-  // TODONVDEC P1: we use the 'default' cuda device interface for color
-  // conversion. That's a temporary hack to make things work. we should abstract
-  // the color conversion stuff separately.
-  defaultCudaInterface_->convertAVFrameToFrameOutput(
-      avFrame, frameOutput, preAllocatedOutputTensor);
+  validatePreAllocatedTensorShape(preAllocatedOutputTensor, avFrame);
+
+  at::cuda::CUDAStream nvdecStream =
+      at::cuda::getCurrentCUDAStream(device_.index());
+
+  frameOutput.data = convertNV12FrameToRGB(
+      avFrame, device_, nppCtx_, nvdecStream, preAllocatedOutputTensor);
 }
 
 } // namespace facebook::torchcodec

src/torchcodec/_core/BetaCudaDeviceInterface.h

@@ -15,6 +15,7 @@
 
 #pragma once
 
+#include "src/torchcodec/_core/CUDACommon.h"
 #include "src/torchcodec/_core/Cache.h"
 #include "src/torchcodec/_core/DeviceInterface.h"
 #include "src/torchcodec/_core/FFMPEGCommon.h"
@@ -94,10 +95,8 @@ class BetaCudaDeviceInterface : public DeviceInterface {
 
   UniqueAVBSFContext bitstreamFilter_;
 
-  // Default CUDA interface for color conversion.
-  // TODONVDEC P2: we shouldn't need to keep a separate instance of the default.
-  // See other TODO there about how interfaces should be completely independent.
-  std::unique_ptr<DeviceInterface> defaultCudaInterface_;
+  // NPP context for color conversion
+  UniqueNppContext nppCtx_;
 };
 
 } // namespace facebook::torchcodec

src/torchcodec/_core/CMakeLists.txt

@@ -99,7 +99,7 @@ function(make_torchcodec_libraries
     )
 
     if(ENABLE_CUDA)
-	    list(APPEND core_sources CudaDeviceInterface.cpp BetaCudaDeviceInterface.cpp NVDECCache.cpp)
+	    list(APPEND core_sources CudaDeviceInterface.cpp BetaCudaDeviceInterface.cpp NVDECCache.cpp CUDACommon.cpp)
     endif()
 
     set(core_library_dependencies

src/torchcodec/_core/CUDACommon.cpp

@@ -0,0 +1,307 @@
+// Copyright (c) Meta Platforms, Inc. and affiliates.
+// All rights reserved.
+//
+// This source code is licensed under the BSD-style license found in the
+// LICENSE file in the root directory of this source tree.
+
+#include "src/torchcodec/_core/CUDACommon.h"
+
+namespace facebook::torchcodec {
+
+namespace {
+
+// Pytorch can only handle up to 128 GPUs.
+// https://github.com/pytorch/pytorch/blob/e30c55ee527b40d67555464b9e402b4b7ce03737/c10/cuda/CUDAMacros.h#L44
+const int MAX_CUDA_GPUS = 128;
+// Set to -1 to have an infinitely sized cache. Set it to 0 to disable caching.
+// Set to a positive number to have a cache of that size.
+const int MAX_CONTEXTS_PER_GPU_IN_CACHE = -1;
+
+PerGpuCache<NppStreamContext> g_cached_npp_ctxs(
+    MAX_CUDA_GPUS,
+    MAX_CONTEXTS_PER_GPU_IN_CACHE);
+
+} // namespace
+
+/* clang-format off */
+// Note: [YUV -> RGB Color Conversion, color space and color range]
+//
+// The frames we get from the decoder (FFmpeg decoder, or NVCUVID) are in YUV
+// format. We need to convert them to RGB. This note attempts to describe this
+// process. There may be some inaccuracies and approximations that experts will
+// notice, but our goal is only to provide a good enough understanding of the
+// process for torchcodec developers to implement and maintain it.
+// On CPU, filtergraph and swscale handle everything for us. With CUDA, we have
+// to do a lot of the heavy lifting ourselves.
+//
+// Color space and color range
+// ---------------------------
+// Two main characteristics of a frame will affect the conversion process:
+// 1. Color space: This basically defines what YUV values correspond to which
+//    physical wavelength. No need to go into details here,the point is that
+//    videos can come in different color spaces, the most common ones being
+//    BT.601 and BT.709, but there are others.
+//    In FFmpeg this is represented with AVColorSpace:
+//    https://ffmpeg.org/doxygen/4.0/pixfmt_8h.html#aff71a069509a1ad3ff54d53a1c894c85
+// 2. Color range: This defines the range of YUV values. There is:
+//    - full range, also called PC range: AVCOL_RANGE_JPEG
+//    - and the "limited" range, also called studio or TV range: AVCOL_RANGE_MPEG
+//    https://ffmpeg.org/doxygen/4.0/pixfmt_8h.html#a3da0bf691418bc22c4bcbe6583ad589a
+//
+// Color space and color range are independent concepts, so we can have a BT.709
+// with full range, and another one with limited range. Same for BT.601.
+//
+// In the first version of this note we'll focus on the full color range. It
+// will later be updated to account for the limited range.
+//
+// Color conversion matrix
+// -----------------------
+// YUV -> RGB conversion is defined as the reverse process of the RGB -> YUV,
+// So this is where we'll start.
+// At the core of a RGB -> YUV conversion are the "luma coefficients", which are
+// specific to a given color space and defined by the color space standard. In
+// FFmpeg they can be found here:
+// https://github.com/FFmpeg/FFmpeg/blob/7d606ef0ccf2946a4a21ab1ec23486cadc21864b/libavutil/csp.c#L46-L56
+//
+// For example, the BT.709 coefficients are: kr=0.2126, kg=0.7152, kb=0.0722
+// Coefficients must sum to 1.
+//
+// Conventionally Y is in [0, 1] range, and U and V are in [-0.5, 0.5] range
+// (that's mathematically, in practice they are represented in integer range).
+// The conversion is defined as:
+// https://en.wikipedia.org/wiki/YCbCr#R'G'B'_to_Y%E2%80%B2PbPr
+// Y = kr*R + kg*G + kb*B
+// U = (B - Y) * 0.5 / (1 - kb) = (B - Y) / u_scale where u_scale = 2 * (1 - kb)
+// V = (R - Y) * 0.5 / (1 - kr) = (R - Y) / v_scale where v_scale = 2 * (1 - kr)
+//
+// Putting all this into matrix form, we get:
+// [Y]   = [kr               kg            kb            ]  [R]
+// [U]     [-kr/u_scale      -kg/u_scale   (1-kb)/u_scale]  [G]
+// [V]     [(1-kr)/v_scale   -kg/v_scale   -kb)/v_scale  ]  [B]
+//
+//
+// Now, to convert YUV to RGB, we just need to invert this matrix:
+// ```py
+// import torch
+// kr, kg, kb = 0.2126, 0.7152, 0.0722  # BT.709  luma coefficients
+// u_scale = 2 * (1 - kb)
+// v_scale = 2 * (1 - kr)
+//
+// rgb_to_yuv = torch.tensor([
+//     [kr, kg, kb],
+//     [-kr/u_scale, -kg/u_scale, (1-kb)/u_scale],
+//     [(1-kr)/v_scale, -kg/v_scale, -kb/v_scale]
+// ])
+//
+// yuv_to_rgb_full = torch.linalg.inv(rgb_to_yuv)
+// print("YUV->RGB matrix (Full Range):")
+// print(yuv_to_rgb_full)
+// ```
+// And we get:
+// tensor([[ 1.0000e+00, -3.3142e-09,  1.5748e+00],
+//         [ 1.0000e+00, -1.8732e-01, -4.6812e-01],
+//         [ 1.0000e+00,  1.8556e+00,  4.6231e-09]])
+//
+// Which matches https://en.wikipedia.org/wiki/YCbCr#ITU-R_BT.709_conversion
+//
+// Color conversion in NPP
+// -----------------------
+// https://docs.nvidia.com/cuda/npp/image_color_conversion.html.
+//
+// NPP provides different ways to convert YUV to RGB:
+// - pre-defined color conversion functions like
+//   nppiNV12ToRGB_709CSC_8u_P2C3R_Ctx and nppiNV12ToRGB_709HDTV_8u_P2C3R_Ctx
+//   which are for BT.709 limited and full range, respectively.
+// - generic color conversion functions that accept a custom color conversion
+//   matrix, called ColorTwist, like nppiNV12ToRGB_8u_ColorTwist32f_P2C3R_Ctx
+//
+// We use the pre-defined functions or the color twist functions depending on
+// which one we find to be closer to the CPU results.
+//
+// The color twist functionality is *partially* described in a section named
+// "YUVToRGBColorTwist". Importantly:
+//
+// - The `nppiNV12ToRGB_8u_ColorTwist32f_P2C3R_Ctx` function takes the YUV data
+//   and the color-conversion matrix as input. The function itself and the
+//   matrix assume different ranges for YUV values:
+// - The **matrix coefficient** must assume that Y is in [0, 1] and U,V are in
+//   [-0.5, 0.5]. That's how we defined our matrix above.
+// - The function `nppiNV12ToRGB_8u_ColorTwist32f_P2C3R_Ctx` however expects all
+//   of the input Y, U, V to be in [0, 255]. That's how the data comes out of
+//   the decoder.
+// - But *internally*, `nppiNV12ToRGB_8u_ColorTwist32f_P2C3R_Ctx` needs U and V to
+//   be centered around 0, i.e. in [-128, 127]. So we need to apply a -128
+//   offset to U and V. Y doesn't need to be offset. The offset can be applied
+//   by adding a 4th column to the matrix.
+//
+//
+// So our conversion matrix becomes the following, with new offset column:
+// tensor([[ 1.0000e+00, -3.3142e-09,  1.5748e+00,     0]
+//         [ 1.0000e+00, -1.8732e-01, -4.6812e-01,     -128]
+//         [ 1.0000e+00,  1.8556e+00,  4.6231e-09 ,    -128]])
+//
+// And that's what we need to pass for BT701, full range.
+/* clang-format on */
+
+// BT.709 full range color conversion matrix for YUV to RGB conversion.
+// See Note [YUV -> RGB Color Conversion, color space and color range]
+const Npp32f bt709FullRangeColorTwist[3][4] = {
+    {1.0f, 0.0f, 1.5748f, 0.0f},
+    {1.0f, -0.187324273f, -0.468124273f, -128.0f},
+    {1.0f, 1.8556f, 0.0f, -128.0f}};
+
+torch::Tensor convertNV12FrameToRGB(
+    UniqueAVFrame& avFrame,
+    const torch::Device& device,
+    const UniqueNppContext& nppCtx,
+    at::cuda::CUDAStream nvdecStream,
+    std::optional<torch::Tensor> preAllocatedOutputTensor) {
+  auto frameDims = FrameDims(avFrame->height, avFrame->width);
+  torch::Tensor dst;
+  if (preAllocatedOutputTensor.has_value()) {
+    dst = preAllocatedOutputTensor.value();
+  } else {
+    dst = allocateEmptyHWCTensor(frameDims, device);
+  }
+
+  // We need to make sure NVDEC has finished decoding a frame before
+  // color-converting it with NPP.
+  // So we make the NPP stream wait for NVDEC to finish.
+  at::cuda::CUDAStream nppStream =
+      at::cuda::getCurrentCUDAStream(device.index());
+  at::cuda::CUDAEvent nvdecDoneEvent;
+  nvdecDoneEvent.record(nvdecStream);
+  nvdecDoneEvent.block(nppStream);
+
+  nppCtx->hStream = nppStream.stream();
+  cudaError_t err = cudaStreamGetFlags(nppCtx->hStream, &nppCtx->nStreamFlags);
+  TORCH_CHECK(
+      err == cudaSuccess,
+      "cudaStreamGetFlags failed: ",
+      cudaGetErrorString(err));
+
+  NppiSize oSizeROI = {frameDims.width, frameDims.height};
+  Npp8u* yuvData[2] = {avFrame->data[0], avFrame->data[1]};
+
+  NppStatus status;
+
+  // For background, see
+  // Note [YUV -> RGB Color Conversion, color space and color range]
+  if (avFrame->colorspace == AVColorSpace::AVCOL_SPC_BT709) {
+    if (avFrame->color_range == AVColorRange::AVCOL_RANGE_JPEG) {
+      // NPP provides a pre-defined color conversion function for BT.709 full
+      // range: nppiNV12ToRGB_709HDTV_8u_P2C3R_Ctx. But it's not closely
+      // matching the results we have on CPU. So we're using a custom color
+      // conversion matrix, which provides more accurate results. See the note
+      // mentioned above for details, and headaches.
+
+      int srcStep[2] = {avFrame->linesize[0], avFrame->linesize[1]};
+
+      status = nppiNV12ToRGB_8u_ColorTwist32f_P2C3R_Ctx(
+          yuvData,
+          srcStep,
+          static_cast<Npp8u*>(dst.data_ptr()),
+          dst.stride(0),
+          oSizeROI,
+          bt709FullRangeColorTwist,
+          *nppCtx);
+    } else {
+      // If not full range, we assume studio limited range.
+      // The color conversion matrix for BT.709 limited range should be:
+      // static const Npp32f bt709LimitedRangeColorTwist[3][4] = {
+      //   {1.16438356f, 0.0f, 1.79274107f, -16.0f},
+      //   {1.16438356f, -0.213248614f, -0.5329093290f, -128.0f},
+      //   {1.16438356f, 2.11240179f, 0.0f, -128.0f}
+      // };
+      // We get very close results to CPU with that, but using the pre-defined
+      // nppiNV12ToRGB_709CSC_8u_P2C3R_Ctx seems to be even more accurate.
+      status = nppiNV12ToRGB_709CSC_8u_P2C3R_Ctx(
+          yuvData,
+          avFrame->linesize[0],
+          static_cast<Npp8u*>(dst.data_ptr()),
+          dst.stride(0),
+          oSizeROI,
+          *nppCtx);
+    }
+  } else {
+    // TODO we're assuming BT.601 color space (and probably limited range) by
+    // calling nppiNV12ToRGB_8u_P2C3R_Ctx. We should handle BT.601 full range,
+    // and other color-spaces like 2020.
+    status = nppiNV12ToRGB_8u_P2C3R_Ctx(
+        yuvData,
+        avFrame->linesize[0],
+        static_cast<Npp8u*>(dst.data_ptr()),
+        dst.stride(0),
+        oSizeROI,
+        *nppCtx);
+  }
+  TORCH_CHECK(status == NPP_SUCCESS, "Failed to convert NV12 frame.");
+
+  return dst;
+}
+
+UniqueNppContext getNppStreamContext(const torch::Device& device) {
+  torch::DeviceIndex nonNegativeDeviceIndex = getNonNegativeDeviceIndex(device);
+
+  UniqueNppContext nppCtx = g_cached_npp_ctxs.get(device);
+  if (nppCtx) {
+    return nppCtx;
+  }
+
+  // From 12.9, NPP recommends using a user-created NppStreamContext and using
+  // the `_Ctx()` calls:
+  // https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html#npp-release-12-9-update-1
+  // And the nppGetStreamContext() helper is deprecated. We are explicitly
+  // supposed to create the NppStreamContext manually from the CUDA device
+  // properties:
+  // https://github.com/NVIDIA/CUDALibrarySamples/blob/d97803a40fab83c058bb3d68b6c38bd6eebfff43/NPP/README.md?plain=1#L54-L72
+
+  nppCtx = std::make_unique<NppStreamContext>();
+  cudaDeviceProp prop{};
+  cudaError_t err = cudaGetDeviceProperties(&prop, nonNegativeDeviceIndex);
+  TORCH_CHECK(
+      err == cudaSuccess,
+      "cudaGetDeviceProperties failed: ",
+      cudaGetErrorString(err));
+
+  nppCtx->nCudaDeviceId = nonNegativeDeviceIndex;
+  nppCtx->nMultiProcessorCount = prop.multiProcessorCount;
+  nppCtx->nMaxThreadsPerMultiProcessor = prop.maxThreadsPerMultiProcessor;
+  nppCtx->nMaxThreadsPerBlock = prop.maxThreadsPerBlock;
+  nppCtx->nSharedMemPerBlock = prop.sharedMemPerBlock;
+  nppCtx->nCudaDevAttrComputeCapabilityMajor = prop.major;
+  nppCtx->nCudaDevAttrComputeCapabilityMinor = prop.minor;
+
+  return nppCtx;
+}
+
+void returnNppStreamContextToCache(
+    const torch::Device& device,
+    UniqueNppContext nppCtx) {
+  if (nppCtx) {
+    g_cached_npp_ctxs.addIfCacheHasCapacity(device, std::move(nppCtx));
+  }
+}
+
+void validatePreAllocatedTensorShape(
+    const std::optional<torch::Tensor>& preAllocatedOutputTensor,
+    const UniqueAVFrame& avFrame) {
+  // Note that CUDA does not yet support transforms, so the only possible
+  // frame dimensions are the raw decoded frame's dimensions.
+  auto frameDims = FrameDims(avFrame->height, avFrame->width);
+
+  if (preAllocatedOutputTensor.has_value()) {
+    auto shape = preAllocatedOutputTensor.value().sizes();
+    TORCH_CHECK(
+        (shape.size() == 3) && (shape[0] == frameDims.height) &&
+            (shape[1] == frameDims.width) && (shape[2] == 3),
+        "Expected tensor of shape ",
+        frameDims.height,
+        "x",
+        frameDims.width,
+        "x3, got ",
+        shape);
+  }
+}
+
+} // namespace facebook::torchcodec