From 926dc0c9fd6b565e4a1fea5204a03a6aa4689847 Mon Sep 17 00:00:00 2001
From: Jens Koenen <koenen@vr.rwth-aachen.de>
Date: Wed, 9 Nov 2022 17:28:13 +0100
Subject: [PATCH] Worked on nvenc synchronous mode
---
src/encoder/nvidia_encoder.cpp | 140 ++++++++++++---------------------
src/encoder/nvidia_encoder.hpp | 4 +-
2 files changed, 50 insertions(+), 94 deletions(-)
diff --git a/src/encoder/nvidia_encoder.cpp b/src/encoder/nvidia_encoder.cpp
index adb483df..46ca86eb 100644
--- a/src/encoder/nvidia_encoder.cpp
+++ b/src/encoder/nvidia_encoder.cpp
@@ -55,11 +55,6 @@ bool NvidiaEncoder::create(lava::device_ptr device, const lava::renderer& render
return false;
}
- if (!this->create_async_event(frame))
- {
- return false;
- }
-
if (!this->create_semaphore(frame, device))
{
return false;
@@ -74,6 +69,7 @@ bool NvidiaEncoder::create(lava::device_ptr device, const lava::renderer& render
void NvidiaEncoder::destroy()
{
+ this->worker_running = false;
this->worker_pool.stop();
for (NvidiaEncoderFrame::Ptr frame : this->frame_list)
@@ -286,25 +282,9 @@ void NvidiaEncoder::release_frame(NvidiaEncoderFrame::Ptr frame)
void NvidiaEncoder::submit_frame(NvidiaEncoderFrame::Ptr frame, lava::renderer& renderer)
{
-#if defined(_WIN32)
- ResetEvent((HANDLE)frame->event_handle);
-#else
- #error "Not implemented for this platform!"
-#endif
-
- frame->async_event->async_wait([this, frame](const asio::error_code& error_code)
+ asio::post(this->worker_pool, [this, frame]()
{
- if (error_code)
- {
- lava::log()->error("Nvidia Encoder: Can't wait for encoding to finished!");
- this->on_encode_error();
- }
-
- else
- {
- this->read_frame(frame);
- }
-
+ this->read_frame(frame);
this->release_frame(frame);
});
@@ -346,39 +326,57 @@ void NvidiaEncoder::read_frame(NvidiaEncoderFrame::Ptr frame)
frame->on_encode_complete(std::span(parameter_buffer.data(), parameter_size), true);
}
- NV_ENC_LOCK_BITSTREAM lock_stream;
- memset(&lock_stream, 0, sizeof(lock_stream));
- lock_stream.version = NV_ENC_LOCK_BITSTREAM_VER;
- lock_stream.doNotWait = 0;
- lock_stream.getRCStats = 0;
- lock_stream.reservedBitFields = 0;
- lock_stream.outputBitstream = frame->nvenc_output_buffer;
- lock_stream.sliceOffsets = nullptr;
- memset(lock_stream.reserved, 0, sizeof(lock_stream.reserved));
- memset(lock_stream.reserved1, 0, sizeof(lock_stream.reserved1));
- memset(lock_stream.reserved2, 0, sizeof(lock_stream.reserved2));
-
- if (nvenc_functions.nvEncLockBitstream(this->nvenc_session, &lock_stream) != NV_ENC_SUCCESS)
+ while (this->worker_running)
{
- lava::log()->error("Nvidia Encoder: Can't lock bitsteam!");
- this->on_encode_error();
+ NV_ENC_LOCK_BITSTREAM lock_stream;
+ memset(&lock_stream, 0, sizeof(lock_stream));
+ lock_stream.version = NV_ENC_LOCK_BITSTREAM_VER;
+ lock_stream.doNotWait = 1;
+ lock_stream.getRCStats = 0;
+ lock_stream.reservedBitFields = 0;
+ lock_stream.outputBitstream = frame->nvenc_output_buffer;
+ lock_stream.sliceOffsets = nullptr;
+ memset(lock_stream.reserved, 0, sizeof(lock_stream.reserved));
+ memset(lock_stream.reserved1, 0, sizeof(lock_stream.reserved1));
+ memset(lock_stream.reserved2, 0, sizeof(lock_stream.reserved2));
- return;
- }
+ auto a = nvenc_functions.nvEncLockBitstream(this->nvenc_session, &lock_stream);
- frame->on_encode_complete(std::span((uint8_t*)lock_stream.bitstreamBufferPtr, lock_stream.bitstreamSizeInBytes), false);
+ if (a == NV_ENC_ERR_LOCK_BUSY)
+ {
+ continue;
+ }
- if (nvenc_functions.nvEncUnlockBitstream(this->nvenc_session, frame->nvenc_output_buffer) != NV_ENC_SUCCESS)
- {
- lava::log()->error("Nvidia Encoder: Can't unlock bistream!");
- this->on_encode_error();
+ else if (a != NV_ENC_SUCCESS)
+ {
+ lava::log()->error("Nvidia Encoder: Can't lock bitsteam!");
+ lava::log()->error("{} {}", a, nvenc_functions.nvEncGetLastErrorString(this->nvenc_session));
+ this->on_encode_error();
- return;
+ return;
+ }
+
+ frame->on_encode_complete(std::span((uint8_t*)lock_stream.bitstreamBufferPtr, lock_stream.bitstreamSizeInBytes), false);
+
+ if (nvenc_functions.nvEncUnlockBitstream(this->nvenc_session, frame->nvenc_output_buffer) != NV_ENC_SUCCESS)
+ {
+ lava::log()->error("Nvidia Encoder: Can't unlock bistream!");
+ this->on_encode_error();
+
+ return;
+ }
+
+ if (lock_stream.hwEncodeStatus == 2)
+ {
+ break;
+ }
}
- if (nvenc_functions.nvEncUnmapInputResource(this->nvenc_session, frame->nvenc_mapped_buffer) != NV_ENC_SUCCESS)
+ auto a = nvenc_functions.nvEncUnmapInputResource(this->nvenc_session, frame->nvenc_mapped_buffer);
+
+ if (a != NV_ENC_SUCCESS)
{
- lava::log()->error("Nvidia Encoder: Can't unmap input resource!");
+ lava::log()->error("Nvidia Encoder: Can't unmap input resource! {}, {}", a, nvenc_functions.nvEncGetLastErrorString(this->nvenc_session));
this->on_encode_error();
return;
@@ -534,7 +532,7 @@ void NvidiaEncoder::submit_encode_task(NvidiaEncoderFrame::Ptr frame)
encode_parameters.inputDuration = 0;
encode_parameters.inputBuffer = frame->nvenc_mapped_buffer;
encode_parameters.outputBitstream = frame->nvenc_output_buffer;
- encode_parameters.completionEvent = frame->event_handle;
+ encode_parameters.completionEvent = nullptr;
encode_parameters.bufferFmt = map_info.mappedBufferFmt;
encode_parameters.pictureStruct = NV_ENC_PIC_STRUCT_FRAME;
encode_parameters.pictureType = (NV_ENC_PIC_TYPE)0;
@@ -714,10 +712,10 @@ bool NvidiaEncoder::create_session(const glm::uvec2& size)
this->nvenc_session_config.darHeight = size.y;
this->nvenc_session_config.frameRateNum = this->frame_rate;
this->nvenc_session_config.frameRateDen = 1;
- this->nvenc_session_config.enableEncodeAsync = 1;
+ this->nvenc_session_config.enableEncodeAsync = 0;
this->nvenc_session_config.enablePTD = 1;
this->nvenc_session_config.reportSliceOffsets = 0;
- this->nvenc_session_config.enableSubFrameWrite = 0;
+ this->nvenc_session_config.enableSubFrameWrite = 1;
this->nvenc_session_config.enableExternalMEHints = 0;
this->nvenc_session_config.enableMEOnlyMode = 0;
this->nvenc_session_config.enableWeightedPrediction = 0;
@@ -961,32 +959,6 @@ bool NvidiaEncoder::create_output_buffer(NvidiaEncoderFrame::Ptr frame)
return true;
}
-bool NvidiaEncoder::create_async_event(NvidiaEncoderFrame::Ptr frame)
-{
-#if defined(_WIN32)
- frame->event_handle = CreateEvent(nullptr, true, false, nullptr);
- frame->async_event = NvidiaEncoderFrame::EventType(this->worker_pool.get_executor(), frame->event_handle);
-#else
- #error "Not implemented for this platform!"
-#endif
-
- NV_ENC_EVENT_PARAMS event_info;
- event_info.version = NV_ENC_EVENT_PARAMS_VER;
- event_info.reserved = 0;
- event_info.completionEvent = frame->event_handle;
- memset(event_info.reserved1, 0, sizeof(event_info.reserved1));
- memset(event_info.reserved2, 0, sizeof(event_info.reserved2));
-
- if (nvenc_functions.nvEncRegisterAsyncEvent(this->nvenc_session, &event_info) != NV_ENC_SUCCESS)
- {
- lava::log()->error("Nvidia Encoder: Can't register async event!");
-
- return false;
- }
-
- return true;
-}
-
bool NvidiaEncoder::create_semaphore(NvidiaEncoderFrame::Ptr frame, lava::device_ptr device)
{
VkExportSemaphoreCreateInfo semaphore_export_info;
@@ -1054,20 +1026,6 @@ void NvidiaEncoder::destroy_frame(NvidiaEncoderFrame::Ptr frame)
frame->nvenc_output_buffer = nullptr;
}
- if (frame->event_handle != nullptr)
- {
- NV_ENC_EVENT_PARAMS event_info;
- event_info.version = NV_ENC_EVENT_PARAMS_VER;
- event_info.reserved = 0;
- event_info.completionEvent = frame->event_handle;
- memset(event_info.reserved1, 0, sizeof(event_info.reserved1));
- memset(event_info.reserved2, 0, sizeof(event_info.reserved2));
-
- nvenc_functions.nvEncUnregisterAsyncEvent(this->nvenc_session, &event_info);
- frame->event_handle = nullptr;
- frame->async_event.reset();
- }
-
if (frame->cuda_buffer != 0)
{
cuMemFree(frame->cuda_buffer);
diff --git a/src/encoder/nvidia_encoder.hpp b/src/encoder/nvidia_encoder.hpp
index 78060a72..b735a4d6 100644
--- a/src/encoder/nvidia_encoder.hpp
+++ b/src/encoder/nvidia_encoder.hpp
@@ -33,8 +33,6 @@ public:
void* memory_handle = nullptr;
void* semaphore_handle = nullptr;
- void* event_handle = nullptr;
- std::optional<EventType> async_event;
CUexternalMemory cuda_external_memory = nullptr;
CUexternalSemaphore cuda_external_semaphore = nullptr;
@@ -93,7 +91,6 @@ private:
bool create_input_buffer(NvidiaEncoderFrame::Ptr frame, lava::device_ptr device, const glm::uvec2& size, EncoderFormat format);
bool create_output_buffer(NvidiaEncoderFrame::Ptr frame);
- bool create_async_event(NvidiaEncoderFrame::Ptr frame);
bool create_semaphore(NvidiaEncoderFrame::Ptr frame, lava::device_ptr device);
void destroy_frame(NvidiaEncoderFrame::Ptr frame);
@@ -115,6 +112,7 @@ private:
private:
asio::thread_pool worker_pool;
+ bool worker_running = true;
lava::device_ptr device;
CUdevice cuda_device = 0;
--
GitLab