Skip to content
Snippets Groups Projects

Resolves #296 removal of NvPipe

Merged Resolves #296 removal of NvPipe
feature/296/directnvenc into master
Merged Nicolas Pope requested to merge feature/296/directnvenc into master
Viewing commit dcde170a
Prev Next
Show latest version
2 files
+ 29
− 64
Compare changes
  • Side-by-side
  • Inline
Files
2
components/codecs/src/nvidia_decoder.cpp
+ 11
− 64
@@ -66,18 +66,15 @@ bool NvidiaDecoder::_create(const ftl::codecs::Packet &pkt) {
bool is_float_frame = pkt.flags & ftl::codecs::kFlagFloat;
// Check existing decoder is valid first and remove if not
if (nv_decoder_ != nullptr && (last_definition_ != pkt.definition ||
last_codec_ != pkt.codec || is_float_channel_ != is_float_frame ||
width_ != last_width_ || height_ != last_height_)) {
if (nv_decoder_ != nullptr && (last_codec_ != pkt.codec || is_float_channel_ != is_float_frame)) {
//width_ != last_width_ || height_ != last_height_)) {
delete nv_decoder_;
nv_decoder_ = nullptr;
}
if (!nv_decoder_) {
// Ensure we have a CUDA context
LOG(INFO) << "Getting cuda context...";
cudaSafeCall(cudaDeviceSynchronize());
LOG(INFO) << "Have cuda context.";
CUcontext cudaContext;
cuCtxGetCurrent(&cudaContext);
@@ -124,11 +121,6 @@ bool NvidiaDecoder::decode(const ftl::codecs::Packet &pkt, cv::cuda::GpuMat &out
bool islossless = ((pkt.codec == ftl::codecs::codec_t::HEVC || pkt.codec == ftl::codecs::codec_t::H264) && is_float_frame &&
!(pkt.flags & 0x2)) || pkt.codec == ftl::codecs::codec_t::HEVC_LOSSLESS || pkt.codec == ftl::codecs::codec_t::H264_LOSSLESS;
/*if (is_float_frame && !islossless && out.type() != CV_16UC4) {
LOG(ERROR) << "Invalid buffer for lossy float frame";
return false;
}*/
if (is_float_frame && out.type() != CV_32F) {
LOG(ERROR) << "Invalid buffer for float frame";
return false;
@@ -139,51 +131,10 @@ bool NvidiaDecoder::decode(const ftl::codecs::Packet &pkt, cv::cuda::GpuMat &out
return false;
}
int width = ftl::codecs::getWidth(pkt.definition);
int height = ftl::codecs::getHeight(pkt.definition);
auto [tx,ty] = ftl::codecs::chooseTileConfig(pkt.frame_count);
//int exp_height = (is_float_frame && !islossless) ? ty*height+((ty*height)/2) : ty*height;
if (tx*width != out.cols || ty*height != out.rows) {
LOG(ERROR) << "Received frame whose size does not match buffer";
return false;
}
// Is the previous decoder still valid for current resolution and type?
//if (nv_decoder_ != nullptr && (last_definition_ != pkt.definition || last_codec_ != pkt.codec || is_float_channel_ != is_float_frame)) {
// NvPipe_Destroy(nv_decoder_);
// nv_decoder_ = nullptr;
//}
// Create an appropriate NvDecoder instance
width_ = tx*width;
height_ = ty*height;
//if (islossless && is_float_frame) width_ *= 2; // 16bit = double width 8 bit
_create(pkt);
is_float_channel_ = is_float_frame;
last_definition_ = pkt.definition;
last_codec_ = pkt.codec;
last_width_ = width_;
last_height_ = height_;
// Build a decoder instance of the correct kind
/*if (nv_decoder_ == nullptr) {
nv_decoder_ = NvPipe_CreateDecoder(
(is_float_frame) ? (islossless) ? NVPIPE_UINT16 : NVPIPE_NV12_10bit : NVPIPE_RGBA32,
(pkt.codec == codec_t::HEVC || pkt.codec == ftl::codecs::codec_t::HEVC_LOSSLESS) ? NVPIPE_HEVC : NVPIPE_H264,
out.cols,
out.rows);
if (!nv_decoder_) {
//LOG(INFO) << "Bitrate=" << (int)bitrate << " width=" << ABRController::getColourWidth(bitrate);
LOG(FATAL) << "Could not create decoder: " << NvPipe_GetError(NULL);
}
seen_iframe_ = false;
}*/
//tmp_.create(cv::Size(ftl::codecs::getWidth(pkt.definition),ftl::codecs::getHeight(pkt.definition)), (!is_float_frame) ? CV_8UC4 : (islossless) ? CV_16U : CV_16UC4);
// Final checks for validity
if (pkt.data.size() == 0) { // || !ftl::codecs::hevc::validNAL(pkt.data)) {
@@ -212,21 +163,22 @@ bool NvidiaDecoder::decode(const ftl::codecs::Packet &pkt, cv::cuda::GpuMat &out
}
decodedPtr = _decode(ptr, size);
//rc = NvPipe_Decode(nv_decoder_, ptr, size, out.data, tx*width, ty*height, out.step);
//if (rc == 0) LOG(ERROR) << "NvPipe decode error: " << NvPipe_GetError(nv_decoder_);
ptr += size;
}
//LOG(WARNING) << "Decode of multiple frames: " << count;
} else {
if (!_checkIFrame(pkt.codec, pkt.data.data(), pkt.data.size())) {
LOG(WARNING) << "P-Frame without I-Frame in decoder: " << pkt.data.size();
return false;
}
decodedPtr = _decode(pkt.data.data(), pkt.data.size());
//LOG(INFO) << "Decoded size = " << nv_decoder_->GetWidth() << "x" << nv_decoder_->GetHeight() << " - " << nv_decoder_->GetBPP();
//rc = NvPipe_Decode(nv_decoder_, pkt.data.data(), pkt.data.size(), out.data, tx*width, ty*height, out.step);
//if (rc == 0) LOG(ERROR) << "NvPipe decode error: " << NvPipe_GetError(nv_decoder_);
}
width_ = nv_decoder_->GetWidth();
height_ = nv_decoder_->GetHeight();
if (out.cols != ((is_float_frame && islossless) ? width_/2 : width_) || out.rows != height_) {
LOG(ERROR) << "Decoded frame not same size as buffer";
return false;
}
// OpenCV GpuMat for YCbCr 4:2:0
@@ -241,21 +193,16 @@ bool NvidiaDecoder::decode(const ftl::codecs::Packet &pkt, cv::cuda::GpuMat &out
if (!islossless) {
cv::cuda::GpuMat sroi = surface(cv::Rect(0,0,width_, height_));
cv::cuda::GpuMat csroi = surface(cv::Rect(0,height_,width_, height_/2));
//cv::cuda::cvtColor(tmp_, tmp_, cv::COLOR_RGB2YUV, 4, stream_);
//LOG(INFO) << "Depth convert: " << out.cols << ", " << out.rows << ", " << out.type();
//ftl::cuda::vuya_to_depth(out, tmp_, 16.0f, stream_);
ftl::cuda::vuya_to_depth(out, sroi, csroi, 16.0f, cvstream);
} else {
//tmp_.convertTo(out, CV_32FC1, 1.0f/1000.0f, stream_);
ftl::cuda::nv12_to_float(decodedPtr, width_*2, (float*)out.data, out.step1(), width_, height_, stream_);
}
} else {
// Flag 0x1 means frame is in RGB so needs conversion to BGR
if (pkt.flags & 0x1) {
//cv::cuda::cvtColor(surface, out, cv::COLOR_YUV2BGRA_NV12, 0, cvstream);
Nv12ToColor32<BGRA32>(decodedPtr, width_, out.data, out.step1(), width_, height_, 0, stream_);
} else {
//cv::cuda::cvtColor(surface, out, cv::COLOR_YUV2RGBA_NV12, 0, cvstream);
Nv12ToColor32<RGBA32>(decodedPtr, width_, out.data, out.step1(), width_, height_, 0, stream_);
}
}