未验证 提交 b8b82b72 编写于 作者: T tianshuo78520a 提交者: GitHub

Del cudnn6 code2 (#31986)

上级 a17c3691
...@@ -100,6 +100,11 @@ if(WITH_GPU) ...@@ -100,6 +100,11 @@ if(WITH_GPU)
if(NOT CUDNN_FOUND) if(NOT CUDNN_FOUND)
message(FATAL_ERROR "Paddle needs cudnn to compile") message(FATAL_ERROR "Paddle needs cudnn to compile")
endif() endif()
if(${CUDNN_MAJOR_VERSION} VERSION_LESS 7)
message(FATAL_ERROR "Paddle needs CUDNN >= 7.0 to compile")
endif()
if(CUPTI_FOUND) if(CUPTI_FOUND)
include_directories(${CUPTI_INCLUDE_DIR}) include_directories(${CUPTI_INCLUDE_DIR})
add_definitions(-DPADDLE_WITH_CUPTI) add_definitions(-DPADDLE_WITH_CUPTI)
......
...@@ -24,26 +24,9 @@ void* cudnn_dso_handle = nullptr; ...@@ -24,26 +24,9 @@ void* cudnn_dso_handle = nullptr;
#define DEFINE_WRAP(__name) DynLoad__##__name __name #define DEFINE_WRAP(__name) DynLoad__##__name __name
CUDNN_DNN_ROUTINE_EACH(DEFINE_WRAP); CUDNN_DNN_ROUTINE_EACH(DEFINE_WRAP);
CUDNN_DNN_ROUTINE_EACH_R2(DEFINE_WRAP);
#ifdef CUDNN_DNN_ROUTINE_EACH_AFTER_R3 #ifdef CUDNN_DNN_ROUTINE_EACH_AFTER_R7_LESS_R8
CUDNN_DNN_ROUTINE_EACH_AFTER_R3(DEFINE_WRAP); CUDNN_DNN_ROUTINE_EACH_AFTER_R7_LESS_R8(DEFINE_WRAP);
#endif
#ifdef CUDNN_DNN_ROUTINE_EACH_AFTER_R3_LESS_R8
CUDNN_DNN_ROUTINE_EACH_AFTER_R3_LESS_R8(DEFINE_WRAP);
#endif
#ifdef CUDNN_DNN_ROUTINE_EACH_AFTER_R4
CUDNN_DNN_ROUTINE_EACH_AFTER_R4(DEFINE_WRAP);
#endif
#ifdef CUDNN_DNN_ROUTINE_EACH_R5
CUDNN_DNN_ROUTINE_EACH_R5(DEFINE_WRAP);
#endif
#ifdef CUDNN_DNN_ROUTINE_EACH_R6
CUDNN_DNN_ROUTINE_EACH_R6(DEFINE_WRAP);
#endif #endif
#ifdef CUDNN_DNN_ROUTINE_EACH_R7 #ifdef CUDNN_DNN_ROUTINE_EACH_R7
......
...@@ -112,57 +112,29 @@ extern void EnforceCUDNNLoaded(const char* fn_name); ...@@ -112,57 +112,29 @@ extern void EnforceCUDNNLoaded(const char* fn_name);
__macro(cudnnRNNForwardInference); \ __macro(cudnnRNNForwardInference); \
__macro(cudnnDestroyDropoutDescriptor); \ __macro(cudnnDestroyDropoutDescriptor); \
__macro(cudnnDestroyRNNDescriptor); \ __macro(cudnnDestroyRNNDescriptor); \
__macro(cudnnSetTensorNdDescriptorEx); __macro(cudnnSetTensorNdDescriptorEx); \
CUDNN_DNN_ROUTINE_EACH(DECLARE_DYNAMIC_LOAD_CUDNN_WRAP)
#define CUDNN_DNN_ROUTINE_EACH_R2(__macro) \
__macro(cudnnAddTensor); \ __macro(cudnnAddTensor); \
__macro(cudnnConvolutionBackwardData); \ __macro(cudnnConvolutionBackwardData); \
__macro(cudnnConvolutionBackwardFilter); __macro(cudnnConvolutionBackwardFilter); \
CUDNN_DNN_ROUTINE_EACH_R2(DECLARE_DYNAMIC_LOAD_CUDNN_WRAP)
// APIs available after R3:
#if CUDNN_VERSION >= 3000
#define CUDNN_DNN_ROUTINE_EACH_AFTER_R3(__macro) \
__macro(cudnnGetConvolutionBackwardFilterWorkspaceSize); \ __macro(cudnnGetConvolutionBackwardFilterWorkspaceSize); \
__macro(cudnnGetConvolutionBackwardDataWorkspaceSize); __macro(cudnnGetConvolutionBackwardDataWorkspaceSize); \
CUDNN_DNN_ROUTINE_EACH_AFTER_R3(DECLARE_DYNAMIC_LOAD_CUDNN_WRAP)
#endif
// APIs available after R3:
#if CUDNN_VERSION >= 3000 && CUDNN_VERSION < 8000
#define CUDNN_DNN_ROUTINE_EACH_AFTER_R3_LESS_R8(__macro) \
__macro(cudnnGetConvolutionBackwardFilterAlgorithm); \
__macro(cudnnGetConvolutionForwardAlgorithm); \
__macro(cudnnGetConvolutionBackwardDataAlgorithm); \
__macro(cudnnSetRNNDescriptor);
CUDNN_DNN_ROUTINE_EACH_AFTER_R3_LESS_R8(DECLARE_DYNAMIC_LOAD_CUDNN_WRAP)
#endif
// APIs available after R4:
#if CUDNN_VERSION >= 4007
#define CUDNN_DNN_ROUTINE_EACH_AFTER_R4(__macro) \
__macro(cudnnBatchNormalizationForwardTraining); \ __macro(cudnnBatchNormalizationForwardTraining); \
__macro(cudnnBatchNormalizationForwardInference); \ __macro(cudnnBatchNormalizationForwardInference); \
__macro(cudnnBatchNormalizationBackward); __macro(cudnnBatchNormalizationBackward); \
CUDNN_DNN_ROUTINE_EACH_AFTER_R4(DECLARE_DYNAMIC_LOAD_CUDNN_WRAP)
#endif
// APIs in R5
#if CUDNN_VERSION >= 5000
#define CUDNN_DNN_ROUTINE_EACH_R5(__macro) \
__macro(cudnnCreateActivationDescriptor); \ __macro(cudnnCreateActivationDescriptor); \
__macro(cudnnSetActivationDescriptor); \ __macro(cudnnSetActivationDescriptor); \
__macro(cudnnGetActivationDescriptor); \ __macro(cudnnGetActivationDescriptor); \
__macro(cudnnDestroyActivationDescriptor); __macro(cudnnDestroyActivationDescriptor); \
CUDNN_DNN_ROUTINE_EACH_R5(DECLARE_DYNAMIC_LOAD_CUDNN_WRAP) __macro(cudnnSetRNNDescriptor_v6);
#endif CUDNN_DNN_ROUTINE_EACH(DECLARE_DYNAMIC_LOAD_CUDNN_WRAP)
// APIs in R6 #if CUDNN_VERSION >= 7000 && CUDNN_VERSION < 8000
#if CUDNN_VERSION >= 6000 #define CUDNN_DNN_ROUTINE_EACH_AFTER_R7_LESS_R8(__macro) \
#define CUDNN_DNN_ROUTINE_EACH_R6(__macro) __macro(cudnnSetRNNDescriptor_v6); __macro(cudnnGetConvolutionBackwardFilterAlgorithm); \
CUDNN_DNN_ROUTINE_EACH_R6(DECLARE_DYNAMIC_LOAD_CUDNN_WRAP) __macro(cudnnGetConvolutionForwardAlgorithm); \
__macro(cudnnGetConvolutionBackwardDataAlgorithm); \
__macro(cudnnSetRNNDescriptor);
CUDNN_DNN_ROUTINE_EACH_AFTER_R7_LESS_R8(DECLARE_DYNAMIC_LOAD_CUDNN_WRAP)
#endif #endif
#if CUDNN_VERSION >= 7001 #if CUDNN_VERSION >= 7001
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册