[fix] use dynamic DLL loading for cuda library

src/media/video/decode/nvcodec/NvDecoder.cpp

@@ -16,6 +16,7 @@
 #include <cmath>
 #include <iostream>
 
+#include "nvcodec_api.h"
 #include "nvcuvid.h"
 
 #define START_TIMER auto start = std::chrono::steady_clock::now();
@@ -31,7 +32,7 @@
     CUresult err__ = call;                                          \
     if (err__ != CUDA_SUCCESS) {                                    \
       const char *szErrName = NULL;                                 \
-      cuGetErrorName(err__, &szErrName);                            \
+      cuGetErrorName_ld(err__, &szErrName);                         \
       std::ostringstream errorLog;                                  \
       errorLog << "CUDA driver API error " << szErrName;            \
       throw NVDECException::makeNVDECException(                     \
@@ -199,9 +200,9 @@ int NvDecoder::HandleVideoSequence(CUVIDEOFORMAT *pVideoFormat) {
   decodecaps.eChromaFormat = pVideoFormat->chroma_format;
   decodecaps.nBitDepthMinus8 = pVideoFormat->bit_depth_luma_minus8;
 
-  CUDA_DRVAPI_CALL(cuCtxPushCurrent(m_cuContext));
-  NVDEC_API_CALL(cuvidGetDecoderCaps(&decodecaps));
-  CUDA_DRVAPI_CALL(cuCtxPopCurrent(NULL));
+  CUDA_DRVAPI_CALL(cuCtxPushCurrent_ld(m_cuContext));
+  NVDEC_API_CALL(cuvidGetDecoderCaps_ld(&decodecaps));
+  CUDA_DRVAPI_CALL(cuCtxPopCurrent_ld(NULL));
 
   if (!decodecaps.bIsSupported) {
     NVDEC_THROW_ERROR("Codec not supported on this GPU",
@@ -242,7 +243,7 @@ int NvDecoder::HandleVideoSequence(CUVIDEOFORMAT *pVideoFormat) {
   }
 
   if (m_nWidth && m_nLumaHeight && m_nChromaHeight) {
-    // cuvidCreateDecoder() has been called before, and now there's possible
+    // cuvidCreateDecoder_ld() has been called before, and now there's possible
     // config change
     return ReconfigureDecoder(pVideoFormat);
   }
@@ -378,9 +379,9 @@ int NvDecoder::HandleVideoSequence(CUVIDEOFORMAT *pVideoFormat) {
                      "Adaptive"}[videoDecodeCreateInfo.DeinterlaceMode];
   m_videoInfo << std::endl;
 
-  CUDA_DRVAPI_CALL(cuCtxPushCurrent(m_cuContext));
-  NVDEC_API_CALL(cuvidCreateDecoder(&m_hDecoder, &videoDecodeCreateInfo));
-  CUDA_DRVAPI_CALL(cuCtxPopCurrent(NULL));
+  CUDA_DRVAPI_CALL(cuCtxPushCurrent_ld(m_cuContext));
+  NVDEC_API_CALL(cuvidCreateDecoder_ld(&m_hDecoder, &videoDecodeCreateInfo));
+  CUDA_DRVAPI_CALL(cuCtxPopCurrent_ld(NULL));
   STOP_TIMER("Session Initialization Time: ");
   return nDecodeSurface;
 }
@@ -506,9 +507,9 @@ int NvDecoder::ReconfigureDecoder(CUVIDEOFORMAT *pVideoFormat) {
   reconfigParams.ulNumDecodeSurfaces = nDecodeSurface;
 
   START_TIMER
-  CUDA_DRVAPI_CALL(cuCtxPushCurrent(m_cuContext));
-  NVDEC_API_CALL(cuvidReconfigureDecoder(m_hDecoder, &reconfigParams));
-  CUDA_DRVAPI_CALL(cuCtxPopCurrent(NULL));
+  CUDA_DRVAPI_CALL(cuCtxPushCurrent_ld(m_cuContext));
+  NVDEC_API_CALL(cuvidReconfigureDecoder_ld(m_hDecoder, &reconfigParams));
+  CUDA_DRVAPI_CALL(cuCtxPopCurrent_ld(NULL));
   STOP_TIMER("Session Reconfigure Time: ");
 
   return nDecodeSurface;
@@ -538,9 +539,9 @@ int NvDecoder::setReconfigParams(const Rect *pCropRect, const Dim *pResizeDim) {
     pFrame = m_vpFrame.back();
     m_vpFrame.pop_back();
     if (m_bUseDeviceFrame) {
-      CUDA_DRVAPI_CALL(cuCtxPushCurrent(m_cuContext));
-      CUDA_DRVAPI_CALL(cuMemFree((CUdeviceptr)pFrame));
-      CUDA_DRVAPI_CALL(cuCtxPopCurrent(NULL));
+      CUDA_DRVAPI_CALL(cuCtxPushCurrent_ld(m_cuContext));
+      CUDA_DRVAPI_CALL(cuMemFree_ld((CUdeviceptr)pFrame));
+      CUDA_DRVAPI_CALL(cuCtxPopCurrent_ld(NULL));
     } else {
       delete pFrame;
     }
@@ -558,9 +559,9 @@ int NvDecoder::HandlePictureDecode(CUVIDPICPARAMS *pPicParams) {
     return false;
   }
   m_nPicNumInDecodeOrder[pPicParams->CurrPicIdx] = m_nDecodePicCnt++;
-  CUDA_DRVAPI_CALL(cuCtxPushCurrent(m_cuContext));
-  NVDEC_API_CALL(cuvidDecodePicture(m_hDecoder, pPicParams));
-  CUDA_DRVAPI_CALL(cuCtxPopCurrent(NULL));
+  CUDA_DRVAPI_CALL(cuCtxPushCurrent_ld(m_cuContext));
+  NVDEC_API_CALL(cuvidDecodePicture_ld(m_hDecoder, pPicParams));
+  CUDA_DRVAPI_CALL(cuCtxPopCurrent_ld(NULL));
   return 1;
 }
 
@@ -577,15 +578,15 @@ int NvDecoder::HandlePictureDisplay(CUVIDPARSERDISPINFO *pDispInfo) {
 
   CUdeviceptr dpSrcFrame = 0;
   unsigned int nSrcPitch = 0;
-  CUDA_DRVAPI_CALL(cuCtxPushCurrent(m_cuContext));
-  NVDEC_API_CALL(cuvidMapVideoFrame(m_hDecoder, pDispInfo->picture_index,
+  CUDA_DRVAPI_CALL(cuCtxPushCurrent_ld(m_cuContext));
+  NVDEC_API_CALL(cuvidMapVideoFrame64_ld(m_hDecoder, pDispInfo->picture_index,
                                          &dpSrcFrame, &nSrcPitch,
                                          &videoProcessingParameters));
 
   CUVIDGETDECODESTATUS DecodeStatus;
   memset(&DecodeStatus, 0, sizeof(DecodeStatus));
-  CUresult result =
-      cuvidGetDecodeStatus(m_hDecoder, pDispInfo->picture_index, &DecodeStatus);
+  CUresult result = cuvidGetDecodeStatus_ld(
+      m_hDecoder, pDispInfo->picture_index, &DecodeStatus);
   if (result == CUDA_SUCCESS &&
       (DecodeStatus.decodeStatus == cuvidDecodeStatus_Error ||
        DecodeStatus.decodeStatus == cuvidDecodeStatus_Error_Concealed)) {
@@ -602,11 +603,12 @@ int NvDecoder::HandlePictureDisplay(CUVIDPARSERDISPINFO *pDispInfo) {
       uint8_t *pFrame = NULL;
       if (m_bUseDeviceFrame) {
         if (m_bDeviceFramePitched) {
-          CUDA_DRVAPI_CALL(cuMemAllocPitch(
+          CUDA_DRVAPI_CALL(cuMemAllocPitch_ld(
               (CUdeviceptr *)&pFrame, &m_nDeviceFramePitch, GetWidth() * m_nBPP,
               m_nLumaHeight + (m_nChromaHeight * m_nNumChromaPlanes), 16));
         } else {
-          CUDA_DRVAPI_CALL(cuMemAlloc((CUdeviceptr *)&pFrame, GetFrameSize()));
+          CUDA_DRVAPI_CALL(
+              cuMemAlloc_ld((CUdeviceptr *)&pFrame, GetFrameSize()));
         }
       } else {
         pFrame = new uint8_t[GetFrameSize()];
@@ -627,7 +629,7 @@ int NvDecoder::HandlePictureDisplay(CUVIDPARSERDISPINFO *pDispInfo) {
   m.dstPitch = m_nDeviceFramePitch ? m_nDeviceFramePitch : GetWidth() * m_nBPP;
   m.WidthInBytes = GetWidth() * m_nBPP;
   m.Height = m_nLumaHeight;
-  CUDA_DRVAPI_CALL(cuMemcpy2DAsync(&m, m_cuvidStream));
+  CUDA_DRVAPI_CALL(cuMemcpy2DAsync_ld(&m, m_cuvidStream));
 
   // Copy chroma plane
   // NVDEC output has luma height aligned by 2. Adjust chroma offset by aligning
@@ -637,7 +639,7 @@ int NvDecoder::HandlePictureDisplay(CUVIDPARSERDISPINFO *pDispInfo) {
   m.dstDevice =
       (CUdeviceptr)(m.dstHost = pDecodedFrame + m.dstPitch * m_nLumaHeight);
   m.Height = m_nChromaHeight;
-  CUDA_DRVAPI_CALL(cuMemcpy2DAsync(&m, m_cuvidStream));
+  CUDA_DRVAPI_CALL(cuMemcpy2DAsync_ld(&m, m_cuvidStream));
 
   if (m_nNumChromaPlanes == 2) {
     m.srcDevice = (CUdeviceptr)((uint8_t *)dpSrcFrame +
@@ -645,17 +647,17 @@ int NvDecoder::HandlePictureDisplay(CUVIDPARSERDISPINFO *pDispInfo) {
     m.dstDevice = (CUdeviceptr)(m.dstHost = pDecodedFrame +
                                             m.dstPitch * m_nLumaHeight * 2);
     m.Height = m_nChromaHeight;
-    CUDA_DRVAPI_CALL(cuMemcpy2DAsync(&m, m_cuvidStream));
+    CUDA_DRVAPI_CALL(cuMemcpy2DAsync_ld(&m, m_cuvidStream));
   }
-  CUDA_DRVAPI_CALL(cuStreamSynchronize(m_cuvidStream));
-  CUDA_DRVAPI_CALL(cuCtxPopCurrent(NULL));
+  CUDA_DRVAPI_CALL(cuStreamSynchronize_ld(m_cuvidStream));
+  CUDA_DRVAPI_CALL(cuCtxPopCurrent_ld(NULL));
 
   if ((int)m_vTimestamp.size() < m_nDecodedFrame) {
     m_vTimestamp.resize(m_vpFrame.size());
   }
   m_vTimestamp[m_nDecodedFrame - 1] = pDispInfo->timestamp;
 
-  NVDEC_API_CALL(cuvidUnmapVideoFrame(m_hDecoder, dpSrcFrame));
+  NVDEC_API_CALL(cuvidUnmapVideoFrame64_ld(m_hDecoder, dpSrcFrame));
   return 1;
 }
 
@@ -673,7 +675,7 @@ NvDecoder::NvDecoder(CUcontext cuContext, bool bUseDeviceFrame,
   if (pCropRect) m_cropRect = *pCropRect;
   if (pResizeDim) m_resizeDim = *pResizeDim;
 
-  NVDEC_API_CALL(cuvidCtxLockCreate(&m_ctxLock, cuContext));
+  NVDEC_API_CALL(cuvidCtxLockCreate_ld(&m_ctxLock, cuContext));
 
   CUVIDPARSERPARAMS videoParserParameters = {};
   videoParserParameters.CodecType = eCodec;
@@ -685,32 +687,32 @@ NvDecoder::NvDecoder(CUcontext cuContext, bool bUseDeviceFrame,
   videoParserParameters.pfnDecodePicture = HandlePictureDecodeProc;
   videoParserParameters.pfnDisplayPicture = HandlePictureDisplayProc;
   videoParserParameters.pfnGetOperatingPoint = HandleOperatingPointProc;
-  NVDEC_API_CALL(cuvidCreateVideoParser(&m_hParser, &videoParserParameters));
+  NVDEC_API_CALL(cuvidCreateVideoParser_ld(&m_hParser, &videoParserParameters));
 }
 
 NvDecoder::~NvDecoder() {
   START_TIMER
 
   if (m_hParser) {
-    cuvidDestroyVideoParser(m_hParser);
+    cuvidDestroyVideoParser_ld(m_hParser);
   }
-  cuCtxPushCurrent(m_cuContext);
+  cuCtxPushCurrent_ld(m_cuContext);
   if (m_hDecoder) {
-    cuvidDestroyDecoder(m_hDecoder);
+    cuvidDestroyDecoder_ld(m_hDecoder);
   }
 
   std::lock_guard<std::mutex> lock(m_mtxVPFrame);
 
   for (uint8_t *pFrame : m_vpFrame) {
     if (m_bUseDeviceFrame) {
-      cuMemFree((CUdeviceptr)pFrame);
+      cuMemFree_ld((CUdeviceptr)pFrame);
     } else {
       delete[] pFrame;
     }
   }
-  cuCtxPopCurrent(NULL);
+  cuCtxPopCurrent_ld(NULL);
 
-  cuvidCtxLockDestroy(m_ctxLock);
+  cuvidCtxLockDestroy_ld(m_ctxLock);
 
   STOP_TIMER("Session Deinitialization Time: ");
 }
@@ -727,8 +729,8 @@ int NvDecoder::Decode(const uint8_t *pData, int nSize, int nFlags,
   if (!pData || nSize == 0) {
     packet.flags |= CUVID_PKT_ENDOFSTREAM;
   }
-  // NVDEC_API_CALL(cuvidParseVideoData(m_hParser, &packet));
-  if (CUDA_SUCCESS != cuvidParseVideoData(m_hParser, &packet)) {
+  // NVDEC_API_CALL(cuvidParseVideoData_ld(m_hParser, &packet));
+  if (CUDA_SUCCESS != cuvidParseVideoData_ld(m_hParser, &packet)) {
     return 0;
   }
   m_cuvidStream = 0;

src/media/video/decode/nvcodec/nvidia_video_decoder.cpp

@@ -1,6 +1,7 @@
 #include "nvidia_video_decoder.h"
 
 #include "log.h"
+#include "nvcodec_api.h"
 
 #define SAVE_RECEIVED_H264_STREAM 0
 #define SAVE_DECODED_NV12_STREAM 0
@@ -21,20 +22,20 @@ NvidiaVideoDecoder::~NvidiaVideoDecoder() {
 }
 
 int NvidiaVideoDecoder::Init() {
-  ck(cuInit(0));
+  ck(cuInit_ld(0));
   int nGpu = 0;
   int iGpu = 0;
 
-  ck(cuDeviceGetCount(&nGpu));
+  ck(cuDeviceGetCount_ld(&nGpu));
   if (nGpu < 1) {
     return -1;
   }
 
   CUdevice cuDevice;
-  cuDeviceGet(&cuDevice, iGpu);
+  cuDeviceGet_ld(&cuDevice, iGpu);
 
   CUcontext cuContext = NULL;
-  cuCtxCreate(&cuContext, 0, cuDevice);
+  cuCtxCreate_ld(&cuContext, 0, cuDevice);
   if (!cuContext) {
     return -1;
   }

src/media/video/decode/video_decoder_factory.cpp

@@ -10,6 +10,7 @@
 #endif
 
 #include "log.h"
+#include "nvcodec_api.h"
 
 VideoDecoderFactory::VideoDecoderFactory() {}
 
@@ -44,9 +45,8 @@ bool VideoDecoderFactory::CheckIsHardwareAccerlerationSupported() {
   return false;
 #else
   CUresult cuResult;
-
   CUvideoctxlock cudaCtxLock;
-  cuResult = cuvidCtxLockCreate(&cudaCtxLock, 0);
+  cuResult = cuvidCtxLockCreate_ld(&cudaCtxLock, 0);
   if (cuResult != CUDA_SUCCESS) {
     LOG_WARN(
         "System not support hardware accelerated decode, use default software "

src/media/video/encode/nvcodec/NvEncoder.cpp

@@ -11,6 +11,8 @@
 
 #include "NvEncoder.h"
 
+#include "nvcodec_api.h"
+
 #ifndef _WIN32
 #include <cstring>
 static inline bool operator==(const GUID &guid1, const GUID &guid2) {
@@ -60,7 +62,7 @@ void NvEncoder::LoadNvEncApi() {
   uint32_t version = 0;
   uint32_t currentVersion =
       (NVENCAPI_MAJOR_VERSION << 4) | NVENCAPI_MINOR_VERSION;
-  NVENC_API_CALL(NvEncodeAPIGetMaxSupportedVersion(&version));
+  NVENC_API_CALL(NvEncodeAPIGetMaxSupportedVersion_ld(&version));
   if (currentVersion > version) {
     NVENC_THROW_ERROR(
         "Current Driver Version does not support this NvEncodeAPI version, "
@@ -69,7 +71,7 @@ void NvEncoder::LoadNvEncApi() {
   }
 
   m_nvenc = {NV_ENCODE_API_FUNCTION_LIST_VER};
-  NVENC_API_CALL(NvEncodeAPICreateInstance(&m_nvenc));
+  NVENC_API_CALL(NvEncodeAPICreateInstance_ld(&m_nvenc));
 }
 
 NvEncoder::~NvEncoder() { DestroyHWEncoder(); }

src/media/video/encode/nvcodec/NvEncoderCuda.cpp

@@ -11,6 +11,8 @@
 
 #include "NvEncoderCuda.h"
 
+#include "nvcodec_api.h"
+
 NvEncoderCuda::NvEncoderCuda(CUcontext cuContext, uint32_t nWidth,
                              uint32_t nHeight,
                              NV_ENC_BUFFER_FORMAT eBufferFormat,
@@ -44,7 +46,7 @@ void NvEncoderCuda::AllocateInputBuffers(int32_t numInputBuffers) {
   int numCount = m_bMotionEstimationOnly ? 2 : 1;
 
   for (int count = 0; count < numCount; count++) {
-    CUDA_DRVAPI_CALL(cuCtxPushCurrent(m_cuContext));
+    CUDA_DRVAPI_CALL(cuCtxPushCurrent_ld(m_cuContext));
     std::vector<void *> inputFrames;
     for (int i = 0; i < numInputBuffers; i++) {
       CUdeviceptr pDeviceFrame;
@@ -54,13 +56,13 @@ void NvEncoderCuda::AllocateInputBuffers(int32_t numInputBuffers) {
       if (GetPixelFormat() == NV_ENC_BUFFER_FORMAT_YV12 ||
           GetPixelFormat() == NV_ENC_BUFFER_FORMAT_IYUV)
         chromaHeight = GetChromaHeight(GetPixelFormat(), GetMaxEncodeHeight());
-      CUDA_DRVAPI_CALL(cuMemAllocPitch(
+      CUDA_DRVAPI_CALL(cuMemAllocPitch_ld(
           (CUdeviceptr *)&pDeviceFrame, &m_cudaPitch,
           GetWidthInBytes(GetPixelFormat(), GetMaxEncodeWidth()),
           GetMaxEncodeHeight() + chromaHeight, 16));
       inputFrames.push_back((void *)pDeviceFrame);
     }
-    CUDA_DRVAPI_CALL(cuCtxPopCurrent(NULL));
+    CUDA_DRVAPI_CALL(cuCtxPopCurrent_ld(NULL));
 
     RegisterInputResources(
         inputFrames, NV_ENC_INPUT_RESOURCE_TYPE_CUDADEVICEPTR,
@@ -88,23 +90,24 @@ void NvEncoderCuda::ReleaseCudaResources() {
 
   UnregisterInputResources();
 
-  cuCtxPushCurrent(m_cuContext);
+  cuCtxPushCurrent_ld(m_cuContext);
 
   for (uint32_t i = 0; i < m_vInputFrames.size(); ++i) {
     if (m_vInputFrames[i].inputPtr) {
-      cuMemFree(reinterpret_cast<CUdeviceptr>(m_vInputFrames[i].inputPtr));
+      cuMemFree_ld(reinterpret_cast<CUdeviceptr>(m_vInputFrames[i].inputPtr));
     }
   }
   m_vInputFrames.clear();
 
   for (uint32_t i = 0; i < m_vReferenceFrames.size(); ++i) {
     if (m_vReferenceFrames[i].inputPtr) {
-      cuMemFree(reinterpret_cast<CUdeviceptr>(m_vReferenceFrames[i].inputPtr));
+      cuMemFree_ld(
+          reinterpret_cast<CUdeviceptr>(m_vReferenceFrames[i].inputPtr));
     }
   }
   m_vReferenceFrames.clear();
 
-  cuCtxPopCurrent(NULL);
+  cuCtxPopCurrent_ld(NULL);
   m_cuContext = nullptr;
 }
 
@@ -120,7 +123,7 @@ void NvEncoderCuda::CopyToDeviceFrame(
                       NV_ENC_ERR_INVALID_PARAM);
   }
 
-  CUDA_DRVAPI_CALL(cuCtxPushCurrent(device));
+  CUDA_DRVAPI_CALL(cuCtxPushCurrent_ld(device));
 
   uint32_t srcPitch =
       nSrcPitch ? nSrcPitch : NvEncoder::GetWidthInBytes(pixelFormat, width);
@@ -138,10 +141,10 @@ void NvEncoderCuda::CopyToDeviceFrame(
   m.WidthInBytes = NvEncoder::GetWidthInBytes(pixelFormat, width);
   m.Height = height;
   if (bUnAlignedDeviceCopy && srcMemoryType == CU_MEMORYTYPE_DEVICE) {
-    CUDA_DRVAPI_CALL(cuMemcpy2DUnaligned(&m));
+    CUDA_DRVAPI_CALL(cuMemcpy2DUnaligned_ld(&m));
   } else {
-    CUDA_DRVAPI_CALL(stream == NULL ? cuMemcpy2D(&m)
-                                    : cuMemcpy2DAsync(&m, stream));
+    CUDA_DRVAPI_CALL(stream == NULL ? cuMemcpy2D_ld(&m)
+                                    : cuMemcpy2DAsync_ld(&m, stream));
   }
 
   std::vector<uint32_t> srcChromaOffsets;
@@ -167,14 +170,14 @@ void NvEncoderCuda::CopyToDeviceFrame(
       m.WidthInBytes = chromaWidthInBytes;
       m.Height = chromaHeight;
       if (bUnAlignedDeviceCopy && srcMemoryType == CU_MEMORYTYPE_DEVICE) {
-        CUDA_DRVAPI_CALL(cuMemcpy2DUnaligned(&m));
+        CUDA_DRVAPI_CALL(cuMemcpy2DUnaligned_ld(&m));
       } else {
-        CUDA_DRVAPI_CALL(stream == NULL ? cuMemcpy2D(&m)
-                                        : cuMemcpy2DAsync(&m, stream));
+        CUDA_DRVAPI_CALL(stream == NULL ? cuMemcpy2D_ld(&m)
+                                        : cuMemcpy2DAsync_ld(&m, stream));
       }
     }
   }
-  CUDA_DRVAPI_CALL(cuCtxPopCurrent(NULL));
+  CUDA_DRVAPI_CALL(cuCtxPopCurrent_ld(NULL));
 }
 
 void NvEncoderCuda::CopyToDeviceFrame(
@@ -189,7 +192,7 @@ void NvEncoderCuda::CopyToDeviceFrame(
                       NV_ENC_ERR_INVALID_PARAM);
   }
 
-  CUDA_DRVAPI_CALL(cuCtxPushCurrent(device));
+  CUDA_DRVAPI_CALL(cuCtxPushCurrent_ld(device));
 
   uint32_t srcPitch =
       nSrcPitch ? nSrcPitch : NvEncoder::GetWidthInBytes(pixelFormat, width);
@@ -207,9 +210,9 @@ void NvEncoderCuda::CopyToDeviceFrame(
   m.WidthInBytes = NvEncoder::GetWidthInBytes(pixelFormat, width);
   m.Height = height;
   if (bUnAlignedDeviceCopy && srcMemoryType == CU_MEMORYTYPE_DEVICE) {
-    CUDA_DRVAPI_CALL(cuMemcpy2DUnaligned(&m));
+    CUDA_DRVAPI_CALL(cuMemcpy2DUnaligned_ld(&m));
   } else {
-    CUDA_DRVAPI_CALL(cuMemcpy2D(&m));
+    CUDA_DRVAPI_CALL(cuMemcpy2D_ld(&m));
   }
 
   std::vector<uint32_t> srcChromaOffsets;
@@ -234,11 +237,11 @@ void NvEncoderCuda::CopyToDeviceFrame(
       m.WidthInBytes = chromaWidthInBytes;
       m.Height = chromaHeight;
       if (bUnAlignedDeviceCopy && srcMemoryType == CU_MEMORYTYPE_DEVICE) {
-        CUDA_DRVAPI_CALL(cuMemcpy2DUnaligned(&m));
+        CUDA_DRVAPI_CALL(cuMemcpy2DUnaligned_ld(&m));
       } else {
-        CUDA_DRVAPI_CALL(cuMemcpy2D(&m));
+        CUDA_DRVAPI_CALL(cuMemcpy2D_ld(&m));
       }
     }
   }
-  CUDA_DRVAPI_CALL(cuCtxPopCurrent(NULL));
+  CUDA_DRVAPI_CALL(cuCtxPopCurrent_ld(NULL));
 }

src/media/video/encode/nvcodec/NvEncoderCuda.h

@@ -24,7 +24,7 @@
     CUresult err__ = call;                                            \
     if (err__ != CUDA_SUCCESS) {                                      \
       const char* szErrName = NULL;                                   \
-      cuGetErrorName(err__, &szErrName);                              \
+      cuGetErrorName_ld(err__, &szErrName);                           \
       std::ostringstream errorLog;                                    \
       errorLog << "CUDA driver API error " << szErrName;              \
       throw NVENCException::makeNVENCException(                       \

src/media/video/encode/nvcodec/nvidia_video_encoder.cpp

@@ -3,6 +3,7 @@
 #include <chrono>
 
 #include "log.h"
+#include "nvcodec_api.h"
 
 #define SAVE_RECEIVED_NV12_STREAM 0
 #define SAVE_ENCODED_H264_STREAM 0
@@ -32,11 +33,12 @@ int NvidiaVideoEncoder::Init() {
   int num_of_GPUs = 0;
   CUdevice cuda_device;
   bool cuda_ctx_succeed =
-      (index_of_GPU >= 0 && cuInit(0) == CUresult::CUDA_SUCCESS &&
-       cuDeviceGetCount(&num_of_GPUs) == CUresult::CUDA_SUCCESS &&
+      (index_of_GPU >= 0 && cuInit_ld(0) == CUresult::CUDA_SUCCESS &&
+       cuDeviceGetCount_ld(&num_of_GPUs) == CUresult::CUDA_SUCCESS &&
        (num_of_GPUs > 0 && index_of_GPU < num_of_GPUs) &&
-       cuDeviceGet(&cuda_device, index_of_GPU) == CUresult::CUDA_SUCCESS &&
-       cuCtxCreate(&cuda_context_, 0, cuda_device) == CUresult::CUDA_SUCCESS);
+       cuDeviceGet_ld(&cuda_device, index_of_GPU) == CUresult::CUDA_SUCCESS &&
+       cuCtxCreate_ld(&cuda_context_, 0, cuda_device) ==
+           CUresult::CUDA_SUCCESS);
   if (!cuda_ctx_succeed) {
   }
 

src/media/video/encode/video_encoder_factory.cpp

@@ -10,6 +10,7 @@
 #endif
 
 #include "log.h"
+#include "nvcodec_api.h"
 
 VideoEncoderFactory::VideoEncoderFactory() {}
 
@@ -46,7 +47,7 @@ bool VideoEncoderFactory::CheckIsHardwareAccerlerationSupported() {
   CUresult cuResult;
   NV_ENCODE_API_FUNCTION_LIST functionList = {NV_ENCODE_API_FUNCTION_LIST_VER};
 
-  cuResult = cuInit(0);
+  cuResult = cuInit_ld(0);
   if (cuResult != CUDA_SUCCESS) {
     LOG_WARN(
         "System not support hardware accelerated encode, use default software "
@@ -54,7 +55,7 @@ bool VideoEncoderFactory::CheckIsHardwareAccerlerationSupported() {
     return false;
   }
 
-  NVENCSTATUS nvEncStatus = NvEncodeAPICreateInstance(&functionList);
+  NVENCSTATUS nvEncStatus = NvEncodeAPICreateInstance_ld(&functionList);
   if (nvEncStatus != NV_ENC_SUCCESS) {
     LOG_WARN(
         "System not support hardware accelerated encode, use default software "

src/pc/peer_connection.cpp

@@ -6,6 +6,7 @@
 #include "common.h"
 #include "log.h"
 #include "nlohmann/json.hpp"
+#include "nvcodec_api.h"
 
 using nlohmann::json;
 
@@ -227,6 +228,7 @@ int PeerConnection::CreateVideoCodec(bool hardware_acceleration) {
         "MacOS not support hardware acceleration, use default software codec");
   }
 #else
+  InitNvCodecApi();
 #endif
 
   if (av1_encoding_) {

thirdparty/nvcodec/Interface/nvcodec_api.h

@@ -0,0 +1,367 @@
+/*
+ * @Author: DI JUNKUN
+ * @Date: 2024-08-12
+ * Copyright (c) 2024 by DI JUNKUN, All Rights Reserved.
+ */
+
+#ifndef _NVCODEC_API_H_
+#define _NVCODEC_API_H_
+
+#include <Windows.h>
+
+#include <iostream>
+
+#include "cuda.h"
+#include "cuviddec.h"
+#include "nvEncodeAPI.h"
+#include "nvcuvid.h"
+
+typedef CUresult (*TcuInit)(unsigned int Flags);
+
+typedef CUresult (*TcuDeviceGet)(CUdevice *device, int ordinal);
+
+typedef CUresult (*TcuDeviceGetCount)(int *count);
+
+typedef CUresult (*TcuCtxCreate)(CUcontext *pctx, unsigned int flags,
+                                 CUdevice dev);
+
+typedef CUresult (*TcuGetErrorName)(CUresult error, const char **pStr);
+
+typedef CUresult (*TcuCtxPushCurrent)(CUcontext ctx);
+
+typedef CUresult (*TcuCtxPopCurrent)(CUcontext *pctx);
+
+typedef CUresult (*TcuMemAlloc)(CUdeviceptr *dptr, size_t bytesize);
+
+typedef CUresult (*TcuMemAllocPitch)(CUdeviceptr *dptr, size_t *pPitch,
+                                     size_t WidthInBytes, size_t Height,
+                                     unsigned int ElementSizeBytes);
+
+typedef CUresult (*TcuMemFree)(CUdeviceptr dptr);
+
+typedef CUresult (*TcuMemcpy2DAsync)(const CUDA_MEMCPY2D *pCopy,
+                                     CUstream hStream);
+
+typedef CUresult (*TcuStreamSynchronize)(CUstream hStream);
+
+typedef CUresult (*TcuMemcpy2D)(const CUDA_MEMCPY2D *pCopy);
+
+typedef CUresult (*TcuMemcpy2DUnaligned)(const CUDA_MEMCPY2D *pCopy);
+
+// API
+static TcuInit cuInit_ld;
+static TcuDeviceGet cuDeviceGet_ld;
+static TcuDeviceGetCount cuDeviceGetCount_ld;
+static TcuCtxCreate cuCtxCreate_ld;
+static TcuGetErrorName cuGetErrorName_ld;
+static TcuCtxPushCurrent cuCtxPushCurrent_ld;
+static TcuCtxPopCurrent cuCtxPopCurrent_ld;
+static TcuMemAlloc cuMemAlloc_ld;
+static TcuMemAllocPitch cuMemAllocPitch_ld;
+static TcuMemFree cuMemFree_ld;
+static TcuMemcpy2DAsync cuMemcpy2DAsync_ld;
+static TcuStreamSynchronize cuStreamSynchronize_ld;
+static TcuMemcpy2D cuMemcpy2D_ld;
+static TcuMemcpy2DUnaligned cuMemcpy2DUnaligned_ld;
+
+//
+typedef CUresult (*TcuvidCtxLockCreate)(CUvideoctxlock *pLock, CUcontext ctx);
+typedef CUresult (*TcuvidGetDecoderCaps)(CUVIDDECODECAPS *pdc);
+typedef CUresult (*TcuvidCreateDecoder)(CUvideodecoder *phDecoder,
+                                        CUVIDDECODECREATEINFO *pdci);
+typedef CUresult (*TcuvidDestroyDecoder)(CUvideodecoder hDecoder);
+typedef CUresult (*TcuvidDecodePicture)(CUvideodecoder hDecoder,
+                                        CUVIDPICPARAMS *pPicParams);
+typedef CUresult (*TcuvidGetDecodeStatus)(CUvideodecoder hDecoder, int nPicIdx,
+                                          CUVIDGETDECODESTATUS *pDecodeStatus);
+typedef CUresult (*TcuvidReconfigureDecoder)(
+    CUvideodecoder hDecoder, CUVIDRECONFIGUREDECODERINFO *pDecReconfigParams);
+typedef CUresult (*TcuvidMapVideoFrame64)(CUvideodecoder hDecoder, int nPicIdx,
+                                          unsigned long long *pDevPtr,
+                                          unsigned int *pPitch,
+                                          CUVIDPROCPARAMS *pVPP);
+typedef CUresult (*TcuvidUnmapVideoFrame64)(CUvideodecoder hDecoder,
+                                            unsigned long long DevPtr);
+typedef CUresult (*TcuvidCtxLockDestroy)(CUvideoctxlock lck);
+typedef CUresult (*TcuvidCreateVideoParser)(CUvideoparser *pObj,
+                                            CUVIDPARSERPARAMS *pParams);
+typedef CUresult (*TcuvidParseVideoData)(CUvideoparser obj,
+                                         CUVIDSOURCEDATAPACKET *pPacket);
+typedef CUresult (*TcuvidDestroyVideoParser)(CUvideoparser obj);
+
+//
+static TcuvidCtxLockCreate cuvidCtxLockCreate_ld;
+static TcuvidGetDecoderCaps cuvidGetDecoderCaps_ld;
+static TcuvidCreateDecoder cuvidCreateDecoder_ld;
+static TcuvidDestroyDecoder cuvidDestroyDecoder_ld;
+static TcuvidDecodePicture cuvidDecodePicture_ld;
+static TcuvidGetDecodeStatus cuvidGetDecodeStatus_ld;
+static TcuvidReconfigureDecoder cuvidReconfigureDecoder_ld;
+static TcuvidMapVideoFrame64 cuvidMapVideoFrame64_ld;
+static TcuvidUnmapVideoFrame64 cuvidUnmapVideoFrame64_ld;
+static TcuvidCtxLockDestroy cuvidCtxLockDestroy_ld;
+static TcuvidCreateVideoParser cuvidCreateVideoParser_ld;
+static TcuvidParseVideoData cuvidParseVideoData_ld;
+static TcuvidDestroyVideoParser cuvidDestroyVideoParser_ld;
+
+//
+typedef NVENCSTATUS (*TNvEncodeAPICreateInstance)(
+    NV_ENCODE_API_FUNCTION_LIST *functionList);
+typedef NVENCSTATUS (*TNvEncodeAPIGetMaxSupportedVersion)(uint32_t *version);
+
+//
+static TNvEncodeAPICreateInstance NvEncodeAPICreateInstance_ld;
+static TNvEncodeAPIGetMaxSupportedVersion NvEncodeAPIGetMaxSupportedVersion_ld;
+
+static int InitNvCodecApi() {
+  // Load library
+  HMODULE nvcuda_dll = LoadLibrary(TEXT("nvcuda.dll"));
+  if (nvcuda_dll == NULL) {
+    std::cerr << "Unable to load nvcuda.dll!" << std::endl;
+    return 1;
+  }
+
+  cuInit_ld = (TcuInit)GetProcAddress(nvcuda_dll, "cuInit");
+  if (cuInit_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  cuDeviceGet_ld = (TcuDeviceGet)GetProcAddress(nvcuda_dll, "cuDeviceGet");
+  if (cuDeviceGet_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  cuDeviceGetCount_ld =
+      (TcuDeviceGetCount)GetProcAddress(nvcuda_dll, "cuDeviceGetCount");
+  if (cuDeviceGetCount_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  cuCtxCreate_ld = (TcuCtxCreate)GetProcAddress(nvcuda_dll, "cuCtxCreate");
+  if (cuCtxCreate_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  cuGetErrorName_ld =
+      (TcuGetErrorName)GetProcAddress(nvcuda_dll, "cuGetErrorName");
+  if (cuGetErrorName_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  cuCtxPushCurrent_ld =
+      (TcuCtxPushCurrent)GetProcAddress(nvcuda_dll, "cuCtxPushCurrent");
+  if (cuCtxPushCurrent_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  cuCtxPopCurrent_ld =
+      (TcuCtxPopCurrent)GetProcAddress(nvcuda_dll, "cuCtxPopCurrent");
+  if (cuCtxPopCurrent_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+  cuMemAlloc_ld = (TcuMemAlloc)GetProcAddress(nvcuda_dll, "cuMemAlloc");
+  if (cuMemAlloc_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  cuMemAllocPitch_ld =
+      (TcuMemAllocPitch)GetProcAddress(nvcuda_dll, "cuMemAllocPitch");
+  if (cuMemAllocPitch_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  cuMemFree_ld = (TcuMemFree)GetProcAddress(nvcuda_dll, "cuMemFree");
+  if (cuMemFree_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  cuMemcpy2DAsync_ld =
+      (TcuMemcpy2DAsync)GetProcAddress(nvcuda_dll, "cuMemcpy2DAsync");
+  if (cuMemcpy2DAsync_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  cuStreamSynchronize_ld =
+      (TcuStreamSynchronize)GetProcAddress(nvcuda_dll, "cuStreamSynchronize");
+  if (cuStreamSynchronize_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  cuMemcpy2D_ld = (TcuMemcpy2D)GetProcAddress(nvcuda_dll, "cuMemcpy2D");
+  if (cuMemcpy2D_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  cuMemcpy2DUnaligned_ld =
+      (TcuMemcpy2DUnaligned)GetProcAddress(nvcuda_dll, "cuMemcpy2DUnaligned");
+  if (cuMemcpy2DUnaligned_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  //
+  HMODULE nvcuvid_dll = LoadLibrary(TEXT("nvcuvid.dll"));
+  if (nvcuvid_dll == NULL) {
+    std::cerr << "Unable to load nvcuvid.dll!" << std::endl;
+    return 1;
+  }
+
+  cuvidCtxLockCreate_ld =
+      (TcuvidCtxLockCreate)GetProcAddress(nvcuda_dll, "cuvidCtxLockCreate");
+  if (cuvidCtxLockCreate_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  cuvidGetDecoderCaps_ld =
+      (TcuvidGetDecoderCaps)GetProcAddress(nvcuda_dll, "cuvidGetDecoderCaps");
+  if (cuvidGetDecoderCaps_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  cuvidCreateDecoder_ld =
+      (TcuvidCreateDecoder)GetProcAddress(nvcuda_dll, "cuvidCreateDecoder");
+  if (cuvidCreateDecoder_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  cuvidDestroyDecoder_ld =
+      (TcuvidDestroyDecoder)GetProcAddress(nvcuda_dll, "cuvidDestroyDecoder");
+  if (cuvidDestroyDecoder_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  cuvidDecodePicture_ld =
+      (TcuvidDecodePicture)GetProcAddress(nvcuda_dll, "cuvidDecodePicture");
+  if (cuvidDecodePicture_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  cuvidGetDecodeStatus_ld =
+      (TcuvidGetDecodeStatus)GetProcAddress(nvcuda_dll, "cuvidGetDecodeStatus");
+  if (cuvidGetDecodeStatus_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  cuvidReconfigureDecoder_ld = (TcuvidReconfigureDecoder)GetProcAddress(
+      nvcuda_dll, "cuvidReconfigureDecoder");
+  if (cuvidReconfigureDecoder_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  cuvidMapVideoFrame64_ld =
+      (TcuvidMapVideoFrame64)GetProcAddress(nvcuda_dll, "cuvidMapVideoFrame64");
+  if (cuvidMapVideoFrame64_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  cuvidUnmapVideoFrame64_ld = (TcuvidUnmapVideoFrame64)GetProcAddress(
+      nvcuda_dll, "cuvidUnmapVideoFrame64");
+  if (cuvidUnmapVideoFrame64_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  cuvidCtxLockDestroy_ld =
+      (TcuvidCtxLockDestroy)GetProcAddress(nvcuda_dll, "cuvidCtxLockDestroy");
+  if (cuvidCtxLockDestroy_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  cuvidCreateVideoParser_ld = (TcuvidCreateVideoParser)GetProcAddress(
+      nvcuda_dll, "cuvidCreateVideoParser");
+  if (cuvidCreateVideoParser_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  cuvidParseVideoData_ld =
+      (TcuvidParseVideoData)GetProcAddress(nvcuda_dll, "cuvidParseVideoData");
+  if (cuvidParseVideoData_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  cuvidDestroyVideoParser_ld = (TcuvidDestroyVideoParser)GetProcAddress(
+      nvcuda_dll, "cuvidDestroyVideoParser");
+  if (cuvidDestroyVideoParser_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  //
+  HMODULE nvEncodeAPI64_dll = LoadLibrary(TEXT("nvEncodeAPI64.dll"));
+  if (nvEncodeAPI64_dll == NULL) {
+    std::cerr << "Unable to load nvEncodeAPI64.dll!" << std::endl;
+    return 1;
+  }
+
+  NvEncodeAPICreateInstance_ld = (TNvEncodeAPICreateInstance)GetProcAddress(
+      nvcuda_dll, "NvEncodeAPICreateInstance");
+  if (NvEncodeAPICreateInstance_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  NvEncodeAPIGetMaxSupportedVersion_ld =
+      (TNvEncodeAPIGetMaxSupportedVersion)GetProcAddress(
+          nvcuda_dll, "NvEncodeAPIGetMaxSupportedVersion");
+  if (NvEncodeAPIGetMaxSupportedVersion_ld == NULL) {
+    std::cerr << "Unable to find function!" << std::endl;
+    FreeLibrary(nvcuda_dll);
+    return 1;
+  }
+
+  return 0;
+}
+
+#endif

xmake.lua

@@ -192,7 +192,6 @@ target("projectx")
         "pcre2-8", "pcre2-16", "pcre2-32", "pcre2-posix", 
         "zlib", "ffi", "libcrypto", "libssl", "intl", "iconv", 
         "Shell32", "Advapi32", "Dnsapi", "Shlwapi", "Crypt32", 
-        "cuda", "nvencodeapi", "nvcuvid",
         "ws2_32", "Bcrypt", "windowsapp", "User32", "Strmiids", "Mfuuid",
         "Secur32", "Bcrypt")
     elseif is_os(("linux")) then