AI/rocm-systems
2
0 Förgrening 0
Kod Ärenden Pull-förfrågningar Actions Packages Projekt Släpp Wiki Aktiviteter

fix for hipOutOfMem bug when decoding large videos (#189)

Bläddra i källkod 
* fix for issue 441908

[ROCm/rocdecode commit: 51ecd8fccf]
This commit is contained in:
Rajy Rawther authored and GitHub committed 2024-01-19 09:46:02 -08:00
1 förälder 5527ee0b79
incheckning 9afce4a03a
4 files changed
+108 -109
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

projects/rocdecode/samples/videoDecodeFork/videodecodefork.cpp
+1 -10
Visa fil
@@ -59,8 +59,6 @@ void ShowHelpAndExit(const char *option = NULL) {
<< "-f Number of forks (>= 1) - optional; default: 4" << std::endl << "-f Number of forks (>= 1) - optional; default: 4" << std::endl
<< "-d Device ID (>= 0) - optional; default: 0" << std::endl << "-d Device ID (>= 0) - optional; default: 0" << std::endl
<< "-z force_zero_latency (force_zero_latency, Decoded frames will be flushed out for display immediately); optional;" << std::endl << "-z force_zero_latency (force_zero_latency, Decoded frames will be flushed out for display immediately); optional;" << std::endl
<< "-m output_surface_memory_type - decoded surface memory; optional; default - 0"
<< " [0 : OUT_SURFACE_MEM_DEV_INTERNAL/ 1 : OUT_SURFACE_MEM_DEV_COPIED/ 2 : OUT_SURFACE_MEM_HOST_COPIED]" << std::endl;
exit(0); exit(0);
} }
@@ -70,7 +68,7 @@ int main(int argc, char **argv) {
int n_fork = 4; int n_fork = 4;
int device_id = 0; int device_id = 0;
Rect *p_crop_rect = nullptr; Rect *p_crop_rect = nullptr;
OutputSurfaceMemoryType mem_type = OUT_SURFACE_MEM_DEV_INTERNAL; // set to internal OutputSurfaceMemoryType mem_type = OUT_SURFACE_MEM_NOT_MAPPED; // set to unmapped: output frames are not mapped for performance
bool b_force_zero_latency = false; bool b_force_zero_latency = false;
// Parse command-line arguments // Parse command-line arguments
if(argc <= 1) { if(argc <= 1) {
@@ -114,13 +112,6 @@ int main(int argc, char **argv) {
b_force_zero_latency = true; b_force_zero_latency = true;
continue; continue;
} }
if (!strcmp(argv[i], "-m")) {
if (++i == argc) {
ShowHelpAndExit("-m");
}
mem_type = static_cast<OutputSurfaceMemoryType>(atoi(argv[i]));
continue;
}
ShowHelpAndExit(argv[i]); ShowHelpAndExit(argv[i]);
} }
projects/rocdecode/samples/videoDecodePerf/videodecodeperf.cpp
+2 -11
Visa fil
@@ -65,9 +65,7 @@ void ShowHelpAndExit(const char *option = NULL) {
<< "-i Input File Path - required" << std::endl << "-i Input File Path - required" << std::endl
<< "-t Number of threads (>= 1) - optional; default: 4" << std::endl << "-t Number of threads (>= 1) - optional; default: 4" << std::endl
<< "-d Device ID (>= 0) - optional; default: 0" << std::endl << "-d Device ID (>= 0) - optional; default: 0" << std::endl
<< "-z force_zero_latency (force_zero_latency, Decoded frames will be flushed out for display immediately); optional;" << std::endl << "-z force_zero_latency (force_zero_latency, Decoded frames will be flushed out for display immediately); optional;" << std::endl;
<< "-m output_surface_memory_type - decoded surface memory; optional; default - 0"
<< " [0 : OUT_SURFACE_MEM_DEV_INTERNAL/ 1 : OUT_SURFACE_MEM_DEV_COPIED/ 2 : OUT_SURFACE_MEM_HOST_COPIED]" << std::endl;
exit(0); exit(0);
} }
@@ -77,7 +75,7 @@ int main(int argc, char **argv) {
int device_id = 0; int device_id = 0;
int n_thread = 4; int n_thread = 4;
Rect *p_crop_rect = nullptr; Rect *p_crop_rect = nullptr;
OutputSurfaceMemoryType mem_type = OUT_SURFACE_MEM_DEV_INTERNAL; // set to internal OutputSurfaceMemoryType mem_type = OUT_SURFACE_MEM_NOT_MAPPED; // set to decode only for performance
bool b_force_zero_latency = false; bool b_force_zero_latency = false;
// Parse command-line arguments // Parse command-line arguments
if(argc <= 1) { if(argc <= 1) {
@@ -121,13 +119,6 @@ int main(int argc, char **argv) {
b_force_zero_latency = true; b_force_zero_latency = true;
continue; continue;
} }
if (!strcmp(argv[i], "-m")) {
if (++i == argc) {
ShowHelpAndExit("-m");
}
mem_type = static_cast<OutputSurfaceMemoryType>(atoi(argv[i]));
continue;
}
ShowHelpAndExit(argv[i]); ShowHelpAndExit(argv[i]);
} }
projects/rocdecode/utils/rocvideodecode/roc_video_dec.cpp
+103 -87
Visa fil
@@ -79,7 +79,7 @@ RocVideoDecoder::~RocVideoDecoder() {
if (out_mem_type_ == OUT_SURFACE_MEM_DEV_COPIED) { if (out_mem_type_ == OUT_SURFACE_MEM_DEV_COPIED) {
hipError_t hip_status = hipFree(p_frame.frame_ptr); hipError_t hip_status = hipFree(p_frame.frame_ptr);
if (hip_status != hipSuccess) { if (hip_status != hipSuccess) {
std::cout << "ERROR: hipFree failed! (" << hip_status << ")" << std::endl; std::cerr << "ERROR: hipFree failed! (" << hip_status << ")" << std::endl;
} }
} }
else else
@@ -92,7 +92,7 @@ RocVideoDecoder::~RocVideoDecoder() {
hipError_t hip_status = hipSuccess; hipError_t hip_status = hipSuccess;
hip_status = hipStreamDestroy(hip_stream_); hip_status = hipStreamDestroy(hip_stream_);
if (hip_status != hipSuccess) { if (hip_status != hipSuccess) {
std::cout << "ERROR: hipStream_Destroy failed! (" << hip_status << ")" << std::endl; std::cerr << "ERROR: hipStream_Destroy failed! (" << hip_status << ")" << std::endl;
} }
} }
if (fp_out_) { if (fp_out_) {
@@ -375,7 +375,7 @@ int RocVideoDecoder::HandleVideoSequence(RocdecVideoFormat *p_video_format) {
chroma_height_ = (int)(ceil(disp_height_ * GetChromaHeightFactor(video_surface_format_))); chroma_height_ = (int)(ceil(disp_height_ * GetChromaHeightFactor(video_surface_format_)));
num_chroma_planes_ = GetChromaPlaneCount(video_surface_format_); num_chroma_planes_ = GetChromaPlaneCount(video_surface_format_);
if (p_video_format->chroma_format == rocDecVideoChromaFormat_Monochrome) num_chroma_planes_ = 0; if (p_video_format->chroma_format == rocDecVideoChromaFormat_Monochrome) num_chroma_planes_ = 0;
if (out_mem_type_ == OUT_SURFACE_MEM_DEV_INTERNAL) if (out_mem_type_ == OUT_SURFACE_MEM_DEV_INTERNAL || out_mem_type_ == OUT_SURFACE_MEM_NOT_MAPPED)
GetSurfaceStrideInternal(video_surface_format_, p_video_format->coded_width, p_video_format->coded_height, &surface_stride_, &surface_vstride_); GetSurfaceStrideInternal(video_surface_format_, p_video_format->coded_width, p_video_format->coded_height, &surface_stride_, &surface_vstride_);
else { else {
surface_stride_ = videoDecodeCreateInfo.ulTargetWidth * byte_per_pixel_; // todo:: check if we need pitched memory for faster copy surface_stride_ = videoDecodeCreateInfo.ulTargetWidth * byte_per_pixel_; // todo:: check if we need pitched memory for faster copy
@@ -396,9 +396,12 @@ int RocVideoDecoder::HandleVideoSequence(RocdecVideoFormat *p_video_format) {
} else if (out_mem_type_ == OUT_SURFACE_MEM_DEV_COPIED) { } else if (out_mem_type_ == OUT_SURFACE_MEM_DEV_COPIED) {
output_surface_info_.output_surface_size_in_bytes = GetFrameSizePitched(); output_surface_info_.output_surface_size_in_bytes = GetFrameSizePitched();
output_surface_info_.mem_type = OUT_SURFACE_MEM_DEV_COPIED; output_surface_info_.mem_type = OUT_SURFACE_MEM_DEV_COPIED;
} else { } else if (out_mem_type_ == OUT_SURFACE_MEM_HOST_COPIED){
output_surface_info_.output_surface_size_in_bytes = GetFrameSize(); output_surface_info_.output_surface_size_in_bytes = GetFrameSize();
output_surface_info_.mem_type = OUT_SURFACE_MEM_HOST_COPIED; output_surface_info_.mem_type = OUT_SURFACE_MEM_HOST_COPIED;
} else {
output_surface_info_.output_surface_size_in_bytes = surface_stride_ * (surface_vstride_ + (chroma_vstride_ * num_chroma_planes_));
output_surface_info_.mem_type = OUT_SURFACE_MEM_NOT_MAPPED;
} }
disp_rect_.b = videoDecodeCreateInfo.display_area.bottom; disp_rect_.b = videoDecodeCreateInfo.display_area.bottom;
@@ -428,7 +431,7 @@ int RocVideoDecoder::HandleVideoSequence(RocdecVideoFormat *p_video_format) {
*/ */
bool RocVideoDecoder::SetReconfigParams(ReconfigParams *p_reconfig_params) { bool RocVideoDecoder::SetReconfigParams(ReconfigParams *p_reconfig_params) {
if (!p_reconfig_params) { if (!p_reconfig_params) {
std::cout << "ERROR: Invalid reconfig struct passed! "<< std::endl; std::cerr << "ERROR: Invalid reconfig struct passed! "<< std::endl;
return false; return false;
} }
//save it //save it
@@ -456,7 +459,6 @@ int RocVideoDecoder::ReconfigureDecoder(RocdecVideoFormat *p_video_format) {
ROCDEC_THROW("Reconfigure Not supported for bit depth change", ROCDEC_NOT_SUPPORTED); ROCDEC_THROW("Reconfigure Not supported for bit depth change", ROCDEC_NOT_SUPPORTED);
return 0; return 0;
} }
bool is_decode_res_changed = !(p_video_format->coded_width == coded_width_ && p_video_format->coded_height == coded_height_); bool is_decode_res_changed = !(p_video_format->coded_width == coded_width_ && p_video_format->coded_height == coded_height_);
bool is_display_rect_changed = !(p_video_format->display_area.bottom == disp_rect_.b && bool is_display_rect_changed = !(p_video_format->display_area.bottom == disp_rect_.b &&
p_video_format->display_area.top == disp_rect_.t && p_video_format->display_area.top == disp_rect_.t &&
@@ -482,7 +484,7 @@ int RocVideoDecoder::ReconfigureDecoder(RocdecVideoFormat *p_video_format) {
if (p_frame->frame_ptr) { if (p_frame->frame_ptr) {
if (out_mem_type_ == OUT_SURFACE_MEM_DEV_COPIED) { if (out_mem_type_ == OUT_SURFACE_MEM_DEV_COPIED) {
hipError_t hip_status = hipFree(p_frame->frame_ptr); hipError_t hip_status = hipFree(p_frame->frame_ptr);
if (hip_status != hipSuccess) std::cout << "ERROR: hipFree failed! (" << hip_status << ")" << std::endl; if (hip_status != hipSuccess) std::cerr << "ERROR: hipFree failed! (" << hip_status << ")" << std::endl;
} }
else else
delete [] (p_frame->frame_ptr); delete [] (p_frame->frame_ptr);
@@ -521,7 +523,7 @@ int RocVideoDecoder::ReconfigureDecoder(RocdecVideoFormat *p_video_format) {
reconfig_params.target_rect.right = p_video_format->display_area.right; reconfig_params.target_rect.right = p_video_format->display_area.right;
reconfig_params.target_rect.bottom = p_video_format->display_area.bottom; reconfig_params.target_rect.bottom = p_video_format->display_area.bottom;
if (out_mem_type_ == OUT_SURFACE_MEM_DEV_INTERNAL) { if (out_mem_type_ == OUT_SURFACE_MEM_DEV_INTERNAL || out_mem_type_ == OUT_SURFACE_MEM_NOT_MAPPED) {
GetSurfaceStrideInternal(video_surface_format_, p_video_format->coded_width, p_video_format->coded_height, &surface_stride_, &surface_vstride_); GetSurfaceStrideInternal(video_surface_format_, p_video_format->coded_width, p_video_format->coded_height, &surface_stride_, &surface_vstride_);
} else { } else {
surface_stride_ = reconfig_params.ulTargetWidth * byte_per_pixel_; surface_stride_ = reconfig_params.ulTargetWidth * byte_per_pixel_;
@@ -545,9 +547,12 @@ int RocVideoDecoder::ReconfigureDecoder(RocdecVideoFormat *p_video_format) {
} else if (out_mem_type_ == OUT_SURFACE_MEM_DEV_COPIED) { } else if (out_mem_type_ == OUT_SURFACE_MEM_DEV_COPIED) {
output_surface_info_.output_surface_size_in_bytes = GetFrameSizePitched(); output_surface_info_.output_surface_size_in_bytes = GetFrameSizePitched();
output_surface_info_.mem_type = OUT_SURFACE_MEM_DEV_COPIED; output_surface_info_.mem_type = OUT_SURFACE_MEM_DEV_COPIED;
} else { } else if (out_mem_type_ == OUT_SURFACE_MEM_HOST_COPIED) {
output_surface_info_.output_surface_size_in_bytes = GetFrameSize(); output_surface_info_.output_surface_size_in_bytes = GetFrameSize();
output_surface_info_.mem_type = OUT_SURFACE_MEM_HOST_COPIED; output_surface_info_.mem_type = OUT_SURFACE_MEM_HOST_COPIED;
} else {
output_surface_info_.output_surface_size_in_bytes = surface_stride_ * (surface_vstride_ + (chroma_vstride_ * num_chroma_planes_));
output_surface_info_.mem_type = OUT_SURFACE_MEM_NOT_MAPPED;
} }
if (roc_decoder_ == nullptr) { if (roc_decoder_ == nullptr) {
@@ -643,88 +648,97 @@ int RocVideoDecoder::HandlePictureDisplay(RocdecParserDispInfo *pDispInfo) {
sei_message_display_q_[pDispInfo->picture_index].pSEIMessage = NULL; // to avoid double free sei_message_display_q_[pDispInfo->picture_index].pSEIMessage = NULL; // to avoid double free
} }
} }
if (out_mem_type_ != OUT_SURFACE_MEM_NOT_MAPPED) {
void * src_dev_ptr[3] = { 0 }; void * src_dev_ptr[3] = { 0 };
uint32_t src_pitch[3] = { 0 }; uint32_t src_pitch[3] = { 0 };
ROCDEC_API_CALL(rocDecMapVideoFrame(roc_decoder_, pDispInfo->picture_index, src_dev_ptr, src_pitch, &video_proc_params)); ROCDEC_API_CALL(rocDecMapVideoFrame(roc_decoder_, pDispInfo->picture_index, src_dev_ptr, src_pitch, &video_proc_params));
RocdecDecodeStatus dec_status; RocdecDecodeStatus dec_status;
memset(&dec_status, 0, sizeof(dec_status)); memset(&dec_status, 0, sizeof(dec_status));
rocDecStatus result = rocDecGetDecodeStatus(roc_decoder_, pDispInfo->picture_index, &dec_status); rocDecStatus result = rocDecGetDecodeStatus(roc_decoder_, pDispInfo->picture_index, &dec_status);
if (result == ROCDEC_SUCCESS && (dec_status.decodeStatus == rocDecodeStatus_Error || dec_status.decodeStatus == rocDecodeStatus_Error_Concealed)) { if (result == ROCDEC_SUCCESS && (dec_status.decodeStatus == rocDecodeStatus_Error || dec_status.decodeStatus == rocDecodeStatus_Error_Concealed)) {
std::cerr << "Decode Error occurred for picture: " << pic_num_in_dec_order_[pDispInfo->picture_index] << std::endl; std::cerr << "Decode Error occurred for picture: " << pic_num_in_dec_order_[pDispInfo->picture_index] << std::endl;
}
if (out_mem_type_ == OUT_SURFACE_MEM_DEV_INTERNAL) {
DecFrameBuffer dec_frame = { 0 };
dec_frame.frame_ptr = (uint8_t *)(src_dev_ptr[0]);
dec_frame.pts = pDispInfo->pts;
dec_frame.picture_index = pDispInfo->picture_index;
std::lock_guard<std::mutex> lock(mtx_vp_frame_);
vp_frames_q_.push(dec_frame);
decoded_frame_cnt_++;
} else {
// copy the decoded surface info device or host
uint8_t *p_dec_frame = nullptr;
{
std::lock_guard<std::mutex> lock(mtx_vp_frame_);
// if not enough frames in stock, allocate
if ((unsigned)++decoded_frame_cnt_ > vp_frames_.size()) {
num_alloced_frames_++;
DecFrameBuffer dec_frame = { 0 };
if (out_mem_type_ == OUT_SURFACE_MEM_DEV_COPIED) {
// allocate device memory
HIP_API_CALL(hipMalloc((void **)&dec_frame.frame_ptr, GetFrameSizePitched()));
} else {
dec_frame.frame_ptr = new uint8_t[GetFrameSize()];
}
dec_frame.pts = pDispInfo->pts;
dec_frame.picture_index = pDispInfo->picture_index;
vp_frames_.push_back(dec_frame);
}
p_dec_frame = vp_frames_[decoded_frame_cnt_ - 1].frame_ptr;
} }
// Copy luma data if (out_mem_type_ == OUT_SURFACE_MEM_DEV_INTERNAL) {
int dst_pitch = surface_stride_; DecFrameBuffer dec_frame = { 0 };
if (out_mem_type_ == OUT_SURFACE_MEM_DEV_COPIED) { dec_frame.frame_ptr = (uint8_t *)(src_dev_ptr[0]);
if (src_pitch[0] == dst_pitch) { dec_frame.pts = pDispInfo->pts;
int luma_size = src_pitch[0] * coded_height_; dec_frame.picture_index = pDispInfo->picture_index;
HIP_API_CALL(hipMemcpyDtoDAsync(p_dec_frame, src_dev_ptr[0], luma_size, hip_stream_)); std::lock_guard<std::mutex> lock(mtx_vp_frame_);
} else { vp_frames_q_.push(dec_frame);
// use 2d copy to copy an ROI decoded_frame_cnt_++;
HIP_API_CALL(hipMemcpy2DAsync(p_dec_frame, dst_pitch, src_dev_ptr[0], src_pitch[0], coded_width_ * byte_per_pixel_, coded_height_, hipMemcpyDeviceToDevice, hip_stream_)); } else {
// copy the decoded surface info device or host
uint8_t *p_dec_frame = nullptr;
{
std::lock_guard<std::mutex> lock(mtx_vp_frame_);
// if not enough frames in stock, allocate
if ((unsigned)++decoded_frame_cnt_ > vp_frames_.size()) {
num_alloced_frames_++;
DecFrameBuffer dec_frame = { 0 };
if (out_mem_type_ == OUT_SURFACE_MEM_DEV_COPIED) {
// allocate device memory
HIP_API_CALL(hipMalloc((void **)&dec_frame.frame_ptr, GetFrameSizePitched()));
} else {
dec_frame.frame_ptr = new uint8_t[GetFrameSize()];
}
dec_frame.pts = pDispInfo->pts;
dec_frame.picture_index = pDispInfo->picture_index;
vp_frames_.push_back(dec_frame);
}
p_dec_frame = vp_frames_[decoded_frame_cnt_ - 1].frame_ptr;
} }
} else // Copy luma data
HIP_API_CALL(hipMemcpy2DAsync(p_dec_frame, coded_width_ * byte_per_pixel_, src_dev_ptr[0], src_pitch[0], coded_width_ * byte_per_pixel_, coded_height_, hipMemcpyDeviceToHost, hip_stream_)); int dst_pitch = surface_stride_;
// Copy chroma plane ( )
// rocDec output gives pointer to luma and chroma pointers seperated for the decoded frame
uint8_t *p_frame_uv = p_dec_frame + dst_pitch * coded_height_;
if (out_mem_type_ == OUT_SURFACE_MEM_DEV_COPIED) {
if (src_pitch[1] == dst_pitch) {
int chroma_size = chroma_height_ * dst_pitch;
HIP_API_CALL(hipMemcpyDtoDAsync(p_frame_uv, src_dev_ptr[1], chroma_size, hip_stream_));
} else {
// use 2d copy to copy an ROI
HIP_API_CALL(hipMemcpy2DAsync(p_frame_uv, dst_pitch, src_dev_ptr[1], src_pitch[1], coded_width_ * byte_per_pixel_, chroma_height_, hipMemcpyDeviceToDevice, hip_stream_));
}
} else
HIP_API_CALL(hipMemcpy2DAsync(p_frame_uv, dst_pitch, src_dev_ptr[1], src_pitch[1], coded_width_ * byte_per_pixel_, chroma_height_, hipMemcpyDeviceToHost, hip_stream_));
if (num_chroma_planes_ == 2) {
uint8_t *p_frame_uv = p_dec_frame + dst_pitch * (coded_height_ + chroma_height_);
if (out_mem_type_ == OUT_SURFACE_MEM_DEV_COPIED) { if (out_mem_type_ == OUT_SURFACE_MEM_DEV_COPIED) {
if (src_pitch[2] == dst_pitch) { if (src_pitch[0] == dst_pitch) {
int chroma_size = chroma_height_ * dst_pitch; int luma_size = src_pitch[0] * coded_height_;
HIP_API_CALL(hipMemcpyDtoDAsync(p_frame_uv, src_dev_ptr[2], chroma_size, hip_stream_)); HIP_API_CALL(hipMemcpyDtoDAsync(p_dec_frame, src_dev_ptr[0], luma_size, hip_stream_));
} else { } else {
// use 2d copy to copy an ROI // use 2d copy to copy an ROI
HIP_API_CALL(hipMemcpy2DAsync(p_frame_uv, dst_pitch, src_dev_ptr[2], src_pitch[2], coded_width_ * byte_per_pixel_, chroma_height_, hipMemcpyDeviceToDevice, hip_stream_)); HIP_API_CALL(hipMemcpy2DAsync(p_dec_frame, dst_pitch, src_dev_ptr[0], src_pitch[0], coded_width_ * byte_per_pixel_, coded_height_, hipMemcpyDeviceToDevice, hip_stream_));
} }
} else } else
HIP_API_CALL(hipMemcpy2DAsync(p_frame_uv, dst_pitch, src_dev_ptr[2], src_pitch[2], coded_width_ * byte_per_pixel_, chroma_height_, hipMemcpyDeviceToHost, hip_stream_)); HIP_API_CALL(hipMemcpy2DAsync(p_dec_frame, coded_width_ * byte_per_pixel_, src_dev_ptr[0], src_pitch[0], coded_width_ * byte_per_pixel_, coded_height_, hipMemcpyDeviceToHost, hip_stream_));
}
HIP_API_CALL(hipStreamSynchronize(hip_stream_)); // Copy chroma plane ( )
ROCDEC_API_CALL(rocDecUnMapVideoFrame(roc_decoder_, pDispInfo->picture_index)); // rocDec output gives pointer to luma and chroma pointers seperated for the decoded frame
uint8_t *p_frame_uv = p_dec_frame + dst_pitch * coded_height_;
if (out_mem_type_ == OUT_SURFACE_MEM_DEV_COPIED) {
if (src_pitch[1] == dst_pitch) {
int chroma_size = chroma_height_ * dst_pitch;
HIP_API_CALL(hipMemcpyDtoDAsync(p_frame_uv, src_dev_ptr[1], chroma_size, hip_stream_));
} else {
// use 2d copy to copy an ROI
HIP_API_CALL(hipMemcpy2DAsync(p_frame_uv, dst_pitch, src_dev_ptr[1], src_pitch[1], coded_width_ * byte_per_pixel_, chroma_height_, hipMemcpyDeviceToDevice, hip_stream_));
}
} else
HIP_API_CALL(hipMemcpy2DAsync(p_frame_uv, dst_pitch, src_dev_ptr[1], src_pitch[1], coded_width_ * byte_per_pixel_, chroma_height_, hipMemcpyDeviceToHost, hip_stream_));
if (num_chroma_planes_ == 2) {
uint8_t *p_frame_uv = p_dec_frame + dst_pitch * (coded_height_ + chroma_height_);
if (out_mem_type_ == OUT_SURFACE_MEM_DEV_COPIED) {
if (src_pitch[2] == dst_pitch) {
int chroma_size = chroma_height_ * dst_pitch;
HIP_API_CALL(hipMemcpyDtoDAsync(p_frame_uv, src_dev_ptr[2], chroma_size, hip_stream_));
} else {
// use 2d copy to copy an ROI
HIP_API_CALL(hipMemcpy2DAsync(p_frame_uv, dst_pitch, src_dev_ptr[2], src_pitch[2], coded_width_ * byte_per_pixel_, chroma_height_, hipMemcpyDeviceToDevice, hip_stream_));
}
} else
HIP_API_CALL(hipMemcpy2DAsync(p_frame_uv, dst_pitch, src_dev_ptr[2], src_pitch[2], coded_width_ * byte_per_pixel_, chroma_height_, hipMemcpyDeviceToHost, hip_stream_));
}
HIP_API_CALL(hipStreamSynchronize(hip_stream_));
ROCDEC_API_CALL(rocDecUnMapVideoFrame(roc_decoder_, pDispInfo->picture_index));
}
} else {
RocdecDecodeStatus dec_status;
memset(&dec_status, 0, sizeof(dec_status));
rocDecStatus result = rocDecGetDecodeStatus(roc_decoder_, pDispInfo->picture_index, &dec_status);
if (result == ROCDEC_SUCCESS && (dec_status.decodeStatus == rocDecodeStatus_Error || dec_status.decodeStatus == rocDecodeStatus_Error_Concealed)) {
std::cerr << "Decode Error occurred for picture: " << pic_num_in_dec_order_[pDispInfo->picture_index] << std::endl;
}
decoded_frame_cnt_++;
} }
return 1; return 1;
@@ -789,7 +803,7 @@ uint8_t* RocVideoDecoder::GetFrame(int64_t *pts) {
DecFrameBuffer *fb = &vp_frames_q_.front(); DecFrameBuffer *fb = &vp_frames_q_.front();
if (pts) *pts = fb->pts; if (pts) *pts = fb->pts;
return fb->frame_ptr; return fb->frame_ptr;
} else { } else if (vp_frames_.size() > 0){
if (pts) *pts = vp_frames_[decoded_frame_cnt_ret_].pts; if (pts) *pts = vp_frames_[decoded_frame_cnt_ret_].pts;
return vp_frames_[decoded_frame_cnt_ret_++].frame_ptr; return vp_frames_[decoded_frame_cnt_ret_++].frame_ptr;
} }
@@ -808,6 +822,8 @@ uint8_t* RocVideoDecoder::GetFrame(int64_t *pts) {
*/ */
bool RocVideoDecoder::ReleaseFrame(int64_t pTimestamp, bool b_flushing) { bool RocVideoDecoder::ReleaseFrame(int64_t pTimestamp, bool b_flushing) {
if (out_mem_type_ == OUT_SURFACE_MEM_NOT_MAPPED)
return true; // nothing to do
if (out_mem_type_ != OUT_SURFACE_MEM_DEV_INTERNAL) { if (out_mem_type_ != OUT_SURFACE_MEM_DEV_INTERNAL) {
if (!b_flushing) // if not flushing the buffers are re-used, so keep them if (!b_flushing) // if not flushing the buffers are re-used, so keep them
return true; // nothing to do return true; // nothing to do
@@ -845,7 +861,7 @@ bool RocVideoDecoder::ReleaseFrame(int64_t pTimestamp, bool b_flushing) {
* @return false - falied * @return false - falied
*/ */
bool RocVideoDecoder::ReleaseInternalFrames() { bool RocVideoDecoder::ReleaseInternalFrames() {
if (out_mem_type_ != OUT_SURFACE_MEM_DEV_INTERNAL) if (out_mem_type_ != OUT_SURFACE_MEM_DEV_INTERNAL || out_mem_type_ == OUT_SURFACE_MEM_NOT_MAPPED)
return true; // nothing to do return true; // nothing to do
// only needed when using internal mapped buffer // only needed when using internal mapped buffer
while (!vp_frames_q_.empty()) { while (!vp_frames_q_.empty()) {
@@ -869,7 +885,7 @@ void RocVideoDecoder::SaveFrameToFile(std::string output_file_name, void *surf_m
hipError_t hip_status = hipSuccess; hipError_t hip_status = hipSuccess;
hip_status = hipMemcpyDtoH((void *)hst_ptr, surf_mem, output_image_size); hip_status = hipMemcpyDtoH((void *)hst_ptr, surf_mem, output_image_size);
if (hip_status != hipSuccess) { if (hip_status != hipSuccess) {
std::cout << "ERROR: hipMemcpyDtoH failed! (" << hip_status << ")" << std::endl; std::cerr << "ERROR: hipMemcpyDtoH failed! (" << hip_status << ")" << std::endl;
delete [] hst_ptr; delete [] hst_ptr;
return; return;
} }
@@ -978,7 +994,7 @@ void RocVideoDecoder::UpdateMd5ForFrame(void *surf_mem, OutputSurfaceInfo *surf_
hipError_t hip_status = hipSuccess; hipError_t hip_status = hipSuccess;
hip_status = hipMemcpyDtoH((void *)hst_ptr, surf_mem, output_image_size); hip_status = hipMemcpyDtoH((void *)hst_ptr, surf_mem, output_image_size);
if (hip_status != hipSuccess) { if (hip_status != hipSuccess) {
std::cout << "ERROR: hipMemcpyDtoH failed! (" << hip_status << ")" << std::endl; std::cerr << "ERROR: hipMemcpyDtoH failed! (" << hip_status << ")" << std::endl;
delete [] hst_ptr; delete [] hst_ptr;
return; return;
} }
projects/rocdecode/utils/rocvideodecode/roc_video_dec.h
+2 -1
Visa fil
@@ -61,7 +61,8 @@ typedef enum SeiH264HevcPayloadType_enum {
typedef enum OutputSurfaceMemoryType_enum { typedef enum OutputSurfaceMemoryType_enum {
OUT_SURFACE_MEM_DEV_INTERNAL = 0, /**< Internal interopped decoded surface memory(original mapped decoded surface) */ OUT_SURFACE_MEM_DEV_INTERNAL = 0, /**< Internal interopped decoded surface memory(original mapped decoded surface) */
OUT_SURFACE_MEM_DEV_COPIED = 1, /**< decoded output will be copied to a separate device memory (the user doesn't need to call release) **/ OUT_SURFACE_MEM_DEV_COPIED = 1, /**< decoded output will be copied to a separate device memory (the user doesn't need to call release) **/
OUT_SURFACE_MEM_HOST_COPIED = 2 /**< decoded output will be copied to a separate host memory (the user doesn't need to call release) **/ OUT_SURFACE_MEM_HOST_COPIED = 2, /**< decoded output will be copied to a separate host memory (the user doesn't need to call release) **/
OUT_SURFACE_MEM_NOT_MAPPED = 3 /**< < decoded output is not available (interop won't be used): useful for decode only performance app*/
} OutputSurfaceMemoryType; } OutputSurfaceMemoryType;
#define TOSTR(X) std::to_string(static_cast<int>(X)) #define TOSTR(X) std::to_string(static_cast<int>(X))