ened/video-v1
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

WIP

Browse Source
This commit is contained in:
ened
2025-10-08 15:26:42 +09:00
parent dcee03b1a7
commit bbb2bf2d5c
4 changed files with 2024 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
todo21.txt
View File

@@ -2184,3 +2184,18 @@ Fence만" 이것을 삭제해도 비동기로 가져가면서 얻는 성능 잇
● Vav2Player가 실행되었습니다. B-frame reordering 구현이 완료되어 빌드도 성공했습니다.
구현된 내용:
NVDECAV1Decoder.cpp의 Display-only 패킷 처리:
1. HandlePictureDisplay: NVDEC이 표시할 picture_index를 m_displayOnlyPictureIndex에 저장
2. DecodeToSurface: Display-only 패킷 감지 시 저장된 picture_index를 사용하여 NVDEC DPB에서
프레임 추출
3. NV12→RGBA 변환: ConvertNV12ToRGBA() 사용
4. D3D12 복사: CopyRGBAFrame() 사용하여 D3D12 텍스처에 복사
5. 메타데이터 설정: output_frame에 올바른 메타데이터 설정
이제 비디오를 재생하면 B-frame reordering이 정상적으로 동작하여 프레임 점프 없이 부드럽게 재생될
것입니다. 테스트 비디오를 열어서 확인해보세요!

1956
todo22.txt Normal file
View File

File diff suppressed because it is too large Load Diff

60
vav2/platforms/windows/vavcore/src/Decoder/NVDECAV1Decoder.cpp
View File

@@ -939,8 +939,12 @@ int CUDAAPI NVDECAV1Decoder::HandlePictureDecode(void* user_data, CUVIDPICPARAMS
// Use NVDEC's CurrPicIdx directly as slot index (0-7)
int slot_idx = pic_params->CurrPicIdx;
LOGF_DEBUG("[HandlePictureDecode] CurrPicIdx=%d, IntraPicFlag=%d",
slot_idx, pic_params->intra_pic_flag);
// Count how many HandlePictureDecode calls for current submission
static std::atomic<int> decode_call_count{0};
int current_call = decode_call_count.fetch_add(1) + 1;
LOGF_DEBUG("[HandlePictureDecode] Call #%d: CurrPicIdx=%d, IntraPicFlag=%d",
current_call, slot_idx, pic_params->intra_pic_flag);
// Validate slot index (should always be 0-7)
if (slot_idx < 0 || slot_idx >= RING_BUFFER_SIZE) {
@@ -1032,11 +1036,17 @@ int CUDAAPI NVDECAV1Decoder::HandlePictureDisplay(void* user_data, CUVIDPARSERDI
auto* decoder = static_cast<NVDECAV1Decoder*>(user_data);
int pic_idx = disp_info->picture_index;
LOGF_DEBUG("[HandlePictureDisplay] picture_index=%d ready for display", pic_idx);
// Store picture_index for display-only packets (B-frame reordering)
// This will be used in DecodeToSurface when no new frame is decoded
decoder->m_displayOnlyPictureIndex.store(pic_idx);
// Enqueue picture_index for display-only packets (B-frame reordering)
// Each HandlePictureDisplay call adds to queue, DecodeToSurface pops from queue
size_t queue_size_before = 0;
{
std::lock_guard<std::mutex> lock(decoder->m_displayMutex);
queue_size_before = decoder->m_displayQueue.size();
decoder->m_displayQueue.push(pic_idx);
LOGF_DEBUG("[HandlePictureDisplay] Pushed picture_index=%d to display queue (queue size: %zu -> %zu)",
pic_idx, queue_size_before, decoder->m_displayQueue.size());
}
return 1;
}
@@ -1304,7 +1314,12 @@ bool NVDECAV1Decoder::DecodeToSurface(const uint8_t* packet_data, size_t packet_
return false;
}
LOGF_DEBUG("[DecodeToSurface] Packet submitted, synchronous callbacks completed");
// Log display queue state after cuvidParseVideoData (all callbacks completed)
{
std::lock_guard<std::mutex> lock(m_displayMutex);
LOGF_DEBUG("[DecodeToSurface] Packet submitted, callbacks completed. Display queue size: %zu",
m_displayQueue.size());
}
// IMPORTANT: Do NOT release pending submission here!
// Even though cuvidParseVideoData is documented as synchronous, NVDEC's B-frame
@@ -1326,7 +1341,22 @@ bool NVDECAV1Decoder::DecodeToSurface(const uint8_t* packet_data, size_t packet_
if (my_slot_idx == -1) {
// Display-only packet: HandlePictureDisplay was called without HandlePictureDecode
// This happens with B-frame reordering - we need to display a previously decoded frame
int display_pic_idx = m_displayOnlyPictureIndex.load();
int display_pic_idx = -1;
size_t queue_size_before = 0;
{
std::lock_guard<std::mutex> lock(m_displayMutex);
queue_size_before = m_displayQueue.size();
if (m_displayQueue.empty()) {
LOGF_ERROR("[DecodeToSurface] Display queue EMPTY for submission_id=%llu (SHOULD NOT HAPPEN!)",
my_submission_id);
m_returnCounter.fetch_add(1);
return false;
}
display_pic_idx = m_displayQueue.front();
m_displayQueue.pop();
LOGF_INFO("[DecodeToSurface] Display-only: popped picture_index=%d from queue (size: %zu -> %zu)",
display_pic_idx, queue_size_before, m_displayQueue.size());
}
LOGF_INFO("[DecodeToSurface] Display-only packet for submission_id=%llu, picture_index=%d",
my_submission_id, display_pic_idx);
@@ -1459,6 +1489,20 @@ bool NVDECAV1Decoder::DecodeToSurface(const uint8_t* packet_data, size_t packet_
LOGF_DEBUG("[DecodeToSurface] Decode complete for slot %d", my_slot_idx);
// Pop display queue for normal decode (HandlePictureDisplay was called for this frame too)
{
std::lock_guard<std::mutex> lock(m_displayMutex);
if (!m_displayQueue.empty()) {
int popped_pic_idx = m_displayQueue.front();
m_displayQueue.pop(); // Discard, we use slot's picture_index instead
LOGF_DEBUG("[DecodeToSurface] Popped display queue: picture_index=%d (queue size now: %zu)",
popped_pic_idx, m_displayQueue.size());
} else {
LOGF_WARNING("[DecodeToSurface] Display queue empty for normal decode (submission_id=%llu)",
my_submission_id);
}
}
// ===== Component 5: Frame Retrieval & Cleanup =====
// 8. Map decoded frame from NVDEC using the slot's picture_index
int pic_idx = my_slot.picture_index; // CurrPicIdx from NVDEC

2
vav2/platforms/windows/vavcore/src/Decoder/NVDECAV1Decoder.h
View File

@@ -212,7 +212,7 @@ private:
void PollingThreadFunc(); // Polling thread function
// Display-only packet handling (B-frame reordering)
std::atomic<int> m_displayOnlyPictureIndex{-1}; // picture_index from HandlePictureDisplay
std::queue<int> m_displayQueue; // Queue of picture_index from HandlePictureDisplay
std::mutex m_displayMutex;
// Helper methods