Welcome to mirror list, hosted at ThFree Co, Russian Federation.

github.com/soumith/cudnn.torch.git - Unnamed repository; edit this file 'description' to name the repository.
summaryrefslogtreecommitdiff
path: root/ffi.lua
diff options
context:
space:
mode:
Diffstat (limited to 'ffi.lua')
-rw-r--r--ffi.lua446
1 files changed, 223 insertions, 223 deletions
diff --git a/ffi.lua b/ffi.lua
index 24eb33e..7f21f63 100644
--- a/ffi.lua
+++ b/ffi.lua
@@ -43,53 +43,51 @@ typedef enum
CUDNN_TENSOR_NHWC = 1 /* feature maps interleaved ( cStride = 1 )*/
} cudnnTensorFormat_t;
-cudnnStatus_t cudnnCreateTensorDescriptor( cudnnTensorDescriptor_t *tensorDesc );
+cudnnStatus_t cudnnCreateTensorDescriptor( cudnnTensorDescriptor_t *tensorDesc);
cudnnStatus_t cudnnSetTensorNdDescriptor( cudnnTensorDescriptor_t tensorDesc,
cudnnDataType_t dataType,
int nbDims,
const int dimA[],
const int strideA[]
);
-cudnnStatus_t cudnnDestroyTensorDescriptor( cudnnTensorDescriptor_t tensorDesc );
+cudnnStatus_t cudnnDestroyTensorDescriptor( cudnnTensorDescriptor_t tensorDesc);
typedef enum
{
CUDNN_ADD_IMAGE = 0,
- CUDNN_ADD_SAME_HW = 0,
+ CUDNN_ADD_SAME_HW = 0,
CUDNN_ADD_FEATURE_MAP = 1,
CUDNN_ADD_SAME_CHW = 1,
CUDNN_ADD_SAME_C = 2,
CUDNN_ADD_FULL_TENSOR = 3
-} cudnnAddMode_t;
-/* Tensor Bias addition : srcDest = alpha * bias + beta * srcDestDesc */
-cudnnStatus_t cudnnAddTensor( cudnnHandle_t handle,
- cudnnAddMode_t mode,
- const void *alpha,
- const cudnnTensorDescriptor_t biasDesc,
- const void *biasData,
- const void *beta,
- cudnnTensorDescriptor_t srcDestDesc,
- void *srcDestData
- );
-
-/* Set all data points of a tensor to a given value : srcDest = value */
+} cudnnAddMode_t;
+
+cudnnStatus_t cudnnAddTensor(cudnnHandle_t handle,
+ cudnnAddMode_t mode,
+ const void *alpha,
+ const cudnnTensorDescriptor_t biasDesc,
+ const void *biasData,
+ const void *beta,
+ cudnnTensorDescriptor_t srcDestDesc,
+ void *srcDestData
+ );
+
cudnnStatus_t cudnnSetTensor( cudnnHandle_t handle,
- const cudnnTensorDescriptor_t srcDestDesc,
- void *srcDestData,
- const void *value
+ const cudnnTensorDescriptor_t srcDestDesc,
+ void *srcDestData,
+ const void *value
);
-/* Set all data points of a tensor to a given value : srcDest = alpha * srcDest */
-cudnnStatus_t cudnnScaleTensor( cudnnHandle_t handle,
- const cudnnTensorDescriptor_t srcDestDesc,
- void *srcDestData,
- const void *alpha
- );
+cudnnStatus_t cudnnScaleTensor(cudnnHandle_t handle,
+ const cudnnTensorDescriptor_t srcDestDesc,
+ void *srcDestData,
+ const void *alpha
+ );
typedef enum
{
- CUDNN_CONVOLUTION = 0,
- CUDNN_CROSS_CORRELATION = 1
+ CUDNN_CONVOLUTION = 0,
+ CUDNN_CROSS_CORRELATION = 1
} cudnnConvolutionMode_t;
typedef enum
@@ -98,139 +96,140 @@ typedef enum
CUDNN_CONVOLUTION_WEIGHT_GRAD = 1, /* Weight Gradient update function */
CUDNN_CONVOLUTION_DATA_GRAD = 2 /* Data Gradient update function */
} cudnnConvolutionPath_t;
-cudnnStatus_t cudnnCreateFilterDescriptor( cudnnFilterDescriptor_t *filterDesc );
-cudnnStatus_t cudnnSetFilterNdDescriptor( cudnnFilterDescriptor_t filterDesc,
- cudnnDataType_t dataType, // image data type
- int nbDims,
- const int filterDimA[]
- );
+cudnnStatus_t cudnnCreateFilterDescriptor(cudnnFilterDescriptor_t *filterDesc);
+cudnnStatus_t cudnnSetFilterNdDescriptor(cudnnFilterDescriptor_t filterDesc,
+ cudnnDataType_t dataType,
+ int nbDims,
+ const int filterDimA[]
+ );
+
+cudnnStatus_t cudnnDestroyFilterDescriptor( cudnnFilterDescriptor_t filterDesc);
+cudnnStatus_t
+ cudnnCreateConvolutionDescriptor(cudnnConvolutionDescriptor_t *convDesc );
+cudnnStatus_t
+ cudnnSetConvolutionNdDescriptor(cudnnConvolutionDescriptor_t convDesc,
+ int arrayLength, /* nbDims-2 size */
+ const int padA[],
+ const int filterStrideA[],
+ const int upscaleA[],
+ cudnnConvolutionMode_t mode
+ );
+
+cudnnStatus_t
+ cudnnGetConvolutionNdDescriptor(const cudnnConvolutionDescriptor_t convDesc,
+ int arrayLengthRequested,
+ int *arrayLength,
+ int padA[],
+ int strideA[],
+ int upscaleA[],
+ cudnnConvolutionMode_t *mode
+ );
-cudnnStatus_t cudnnDestroyFilterDescriptor( cudnnFilterDescriptor_t filterDesc );
-
-cudnnStatus_t cudnnCreateConvolutionDescriptor( cudnnConvolutionDescriptor_t *convDesc );
-cudnnStatus_t cudnnSetConvolutionNdDescriptor( cudnnConvolutionDescriptor_t convDesc,
- int arrayLength, /* nbDims-2 size */
- const int padA[],
- const int filterStrideA[],
- const int upscaleA[],
- cudnnConvolutionMode_t mode
- );
-
-cudnnStatus_t cudnnGetConvolutionNdDescriptor( const cudnnConvolutionDescriptor_t convDesc,
- int arrayLengthRequested,
- int *arrayLength,
- int padA[],
- int strideA[],
- int upscaleA[],
- cudnnConvolutionMode_t *mode
- );
-
-
-/* Helper function to return the dimensions of the output tensor given a convolution descriptor */
-cudnnStatus_t cudnnGetConvolutionNdForwardOutputDim( const cudnnConvolutionDescriptor_t convDesc,
- const cudnnTensorDescriptor_t inputTensorDesc,
- const cudnnFilterDescriptor_t filterDesc,
- int nbDims,
- int tensorOuputDimA[]
- );
+
+cudnnStatus_t
+ cudnnGetConvolutionNdForwardOutputDim(
+ const cudnnConvolutionDescriptor_t convDesc,
+ const cudnnTensorDescriptor_t inputTensorDesc,
+ const cudnnFilterDescriptor_t filterDesc,
+ int nbDims,
+ int tensorOuputDimA[]
+ );
/* Destroy an instance of convolution descriptor */
-cudnnStatus_t cudnnDestroyConvolutionDescriptor( cudnnConvolutionDescriptor_t convDesc );
+cudnnStatus_t cudnnDestroyConvolutionDescriptor(
+ cudnnConvolutionDescriptor_t convDesc );
typedef enum
{
CUDNN_CONVOLUTION_FWD_NO_WORKSPACE = 0,
CUDNN_CONVOLUTION_FWD_PREFER_FASTEST = 1,
CUDNN_CONVOLUTION_FWD_SPECIFY_WORKSPACE_LIMIT = 2
-} cudnnConvolutionFwdPreference_t;
-
+} cudnnConvolutionFwdPreference_t;
+
typedef enum
{
CUDNN_CONVOLUTION_FWD_ALGO_IMPLICIT_GEMM = 0,
CUDNN_CONVOLUTION_FWD_ALGO_IMPLICIT_PRECOMP_GEMM = 1,
CUDNN_CONVOLUTION_FWD_ALGO_GEMM = 2,
- CUDNN_CONVOLUTION_FWD_ALGO_DIRECT = 3
+ CUDNN_CONVOLUTION_FWD_ALGO_DIRECT = 3
} cudnnConvolutionFwdAlgo_t;
-cudnnStatus_t cudnnGetConvolutionForwardAlgorithm( cudnnHandle_t handle,
- const cudnnTensorDescriptor_t srcDesc,
- const cudnnFilterDescriptor_t filterDesc,
- const cudnnConvolutionDescriptor_t convDesc,
- const cudnnTensorDescriptor_t destDesc,
- cudnnConvolutionFwdPreference_t preference,
- size_t memoryLimitInbytes,
- cudnnConvolutionFwdAlgo_t *algo
- );
-
+cudnnStatus_t cudnnGetConvolutionForwardAlgorithm( cudnnHandle_t handle,
+ const cudnnTensorDescriptor_t srcDesc,
+ const cudnnFilterDescriptor_t filterDesc,
+ const cudnnConvolutionDescriptor_t convDesc,
+ const cudnnTensorDescriptor_t destDesc,
+ cudnnConvolutionFwdPreference_t preference,
+ size_t memoryLimitInbytes,
+ cudnnConvolutionFwdAlgo_t *algo
+ );
+
/*
* convolution algorithm (which requires potentially some workspace)
*/
- /* Helper function to return the minimum size of the workspace to be passed to the convolution given an algo*/
-cudnnStatus_t cudnnGetConvolutionForwardWorkspaceSize( cudnnHandle_t handle,
- const cudnnTensorDescriptor_t srcDesc,
- const cudnnFilterDescriptor_t filterDesc,
- const cudnnConvolutionDescriptor_t convDesc,
- const cudnnTensorDescriptor_t destDesc,
- cudnnConvolutionFwdAlgo_t algo,
- size_t *sizeInBytes
- );
-
+cudnnStatus_t cudnnGetConvolutionForwardWorkspaceSize( cudnnHandle_t handle,
+ const cudnnTensorDescriptor_t srcDesc,
+ const cudnnFilterDescriptor_t filterDesc,
+ const cudnnConvolutionDescriptor_t convDesc,
+ const cudnnTensorDescriptor_t destDesc,
+ cudnnConvolutionFwdAlgo_t algo,
+ size_t *sizeInBytes
+ );
-/* Convolution functions: All of the form "output = alpha * Op(inputs) + beta * output" */
/* Function to perform the forward multiconvolution */
-cudnnStatus_t cudnnConvolutionForward( cudnnHandle_t handle,
- const void *alpha,
- const cudnnTensorDescriptor_t srcDesc,
- const void *srcData,
- const cudnnFilterDescriptor_t filterDesc,
- const void *filterData,
- const cudnnConvolutionDescriptor_t convDesc,
- cudnnConvolutionFwdAlgo_t algo,
- void *workSpace,
- size_t workSpaceSizeInBytes,
- const void *beta,
- const cudnnTensorDescriptor_t destDesc,
- void *destData
+cudnnStatus_t cudnnConvolutionForward(cudnnHandle_t handle,
+ const void *alpha,
+ const cudnnTensorDescriptor_t srcDesc,
+ const void *srcData,
+ const cudnnFilterDescriptor_t filterDesc,
+ const void *filterData,
+ const cudnnConvolutionDescriptor_t convDesc,
+ cudnnConvolutionFwdAlgo_t algo,
+ void *workSpace,
+ size_t workSpaceSizeInBytes,
+ const void *beta,
+ const cudnnTensorDescriptor_t destDesc,
+ void *destData
);
/* Functions to perform the backward multiconvolution */
-cudnnStatus_t cudnnConvolutionBackwardBias( cudnnHandle_t handle,
- const void *alpha,
- const cudnnTensorDescriptor_t srcDesc,
- const void *srcData,
- const void *beta,
- const cudnnTensorDescriptor_t destDesc,
- void *destData
+cudnnStatus_t cudnnConvolutionBackwardBias( cudnnHandle_t handle,
+ const void *alpha,
+ const cudnnTensorDescriptor_t srcDesc,
+ const void *srcData,
+ const void *beta,
+ const cudnnTensorDescriptor_t destDesc,
+ void *destData
);
-
-
-
-cudnnStatus_t cudnnConvolutionBackwardFilter( cudnnHandle_t handle,
- const void *alpha,
- const cudnnTensorDescriptor_t srcDesc,
- const void *srcData,
- const cudnnTensorDescriptor_t diffDesc,
- const void *diffData,
- const cudnnConvolutionDescriptor_t convDesc,
- const void *beta,
- const cudnnFilterDescriptor_t gradDesc,
- void *gradData
- );
-
-
-cudnnStatus_t cudnnConvolutionBackwardData( cudnnHandle_t handle,
- const void *alpha,
- const cudnnFilterDescriptor_t filterDesc,
- const void *filterData,
- const cudnnTensorDescriptor_t diffDesc,
- const void *diffData,
- const cudnnConvolutionDescriptor_t convDesc,
- const void *beta,
- const cudnnTensorDescriptor_t gradDesc,
- void *gradData
- );
+
+
+
+cudnnStatus_t cudnnConvolutionBackwardFilter( cudnnHandle_t handle,
+ const void *alpha,
+ const cudnnTensorDescriptor_t srcDesc,
+ const void *srcData,
+ const cudnnTensorDescriptor_t diffDesc,
+ const void *diffData,
+ const cudnnConvolutionDescriptor_t convDesc,
+ const void *beta,
+ const cudnnFilterDescriptor_t gradDesc,
+ void *gradData
+ );
+
+
+cudnnStatus_t cudnnConvolutionBackwardData( cudnnHandle_t handle,
+ const void *alpha,
+ const cudnnFilterDescriptor_t filterDesc,
+ const void *filterData,
+ const cudnnTensorDescriptor_t diffDesc,
+ const void *diffData,
+ const cudnnConvolutionDescriptor_t convDesc,
+ const void *beta,
+ const cudnnTensorDescriptor_t gradDesc,
+ void *gradData
+ );
/*
@@ -238,43 +237,41 @@ cudnnStatus_t cudnnConvolutionBackwardData( cudnnHandle_t
*/
typedef enum
{
- CUDNN_SOFTMAX_FAST = 0, /* straightforward implementation */
- CUDNN_SOFTMAX_ACCURATE = 1 /* subtract max from every point to avoid overflow */
+ CUDNN_SOFTMAX_FAST = 0,
+ CUDNN_SOFTMAX_ACCURATE = 1
} cudnnSoftmaxAlgorithm_t;
typedef enum
{
- CUDNN_SOFTMAX_MODE_INSTANCE = 0, /* compute the softmax over all C, H, W for each N */
- CUDNN_SOFTMAX_MODE_CHANNEL = 1 /* compute the softmax over all C for each H, W, N */
+ CUDNN_SOFTMAX_MODE_INSTANCE = 0,
+ CUDNN_SOFTMAX_MODE_CHANNEL = 1
} cudnnSoftmaxMode_t;
-/* Softmax functions: All of the form "output = alpha * Op(inputs) + beta * output" */
-
/* Function to perform forward softmax */
cudnnStatus_t cudnnSoftmaxForward( cudnnHandle_t handle,
- cudnnSoftmaxAlgorithm_t algorithm,
- cudnnSoftmaxMode_t mode,
- const void *alpha,
- const cudnnTensorDescriptor_t srcDesc,
- const void *srcData,
- const void *beta,
- const cudnnTensorDescriptor_t destDesc,
- void *destData
- );
+ cudnnSoftmaxAlgorithm_t algorithm,
+ cudnnSoftmaxMode_t mode,
+ const void *alpha,
+ const cudnnTensorDescriptor_t srcDesc,
+ const void *srcData,
+ const void *beta,
+ const cudnnTensorDescriptor_t destDesc,
+ void *destData
+ );
/* Function to perform backward softmax */
cudnnStatus_t cudnnSoftmaxBackward( cudnnHandle_t handle,
- cudnnSoftmaxAlgorithm_t algorithm,
- cudnnSoftmaxMode_t mode,
- const void *alpha,
- const cudnnTensorDescriptor_t srcDesc,
- const void *srcData,
- const cudnnTensorDescriptor_t srcDiffDesc,
- const void *srcDiffData,
- const void *beta,
- const cudnnTensorDescriptor_t destDiffDesc,
- void *destDiffData
- );
+ cudnnSoftmaxAlgorithm_t algorithm,
+ cudnnSoftmaxMode_t mode,
+ const void *alpha,
+ const cudnnTensorDescriptor_t srcDesc,
+ const void *srcData,
+ const cudnnTensorDescriptor_t srcDiffDesc,
+ const void *srcDiffData,
+ const void *beta,
+ const cudnnTensorDescriptor_t destDiffDesc,
+ void *destDiffData
+ );
@@ -285,56 +282,60 @@ typedef enum
} cudnnPoolingMode_t;
/* Create an instance of pooling descriptor */
-cudnnStatus_t cudnnCreatePoolingDescriptor( cudnnPoolingDescriptor_t *poolingDesc);
-cudnnStatus_t cudnnSetPoolingNdDescriptor( cudnnPoolingDescriptor_t poolingDesc,
- const cudnnPoolingMode_t mode,
- int nbDims,
- const int windowDimA[],
- const int paddingA[],
- const int strideA[]
+cudnnStatus_t cudnnCreatePoolingDescriptor(
+ cudnnPoolingDescriptor_t *poolingDesc);
+cudnnStatus_t cudnnSetPoolingNdDescriptor(
+ cudnnPoolingDescriptor_t poolingDesc,
+ const cudnnPoolingMode_t mode,
+ int nbDims,
+ const int windowDimA[],
+ const int paddingA[],
+ const int strideA[]
);
-cudnnStatus_t cudnnGetPoolingNdDescriptor( const cudnnPoolingDescriptor_t poolingDesc,
- const int nbDimsRequested,
- cudnnPoolingMode_t *mode,
- int *nbDims,
- int windowDimA[],
- int paddingA[],
- int strideA[]
- );
-
-cudnnStatus_t cudnnGetPoolingNdForwardOutputDim( const cudnnPoolingDescriptor_t poolingDesc,
- const cudnnTensorDescriptor_t inputTensorDesc,
- int nbDims,
- int outputTensorDimA[]);
+cudnnStatus_t cudnnGetPoolingNdDescriptor(
+ const cudnnPoolingDescriptor_t poolingDesc,
+ const int nbDimsRequested,
+ cudnnPoolingMode_t *mode,
+ int *nbDims,
+ int windowDimA[],
+ int paddingA[],
+ int strideA[]
+ );
+
+cudnnStatus_t cudnnGetPoolingNdForwardOutputDim(
+ const cudnnPoolingDescriptor_t poolingDesc,
+ const cudnnTensorDescriptor_t inputTensorDesc,
+ int nbDims,
+ int outputTensorDimA[]);
/* Destroy an instance of pooling descriptor */
-cudnnStatus_t cudnnDestroyPoolingDescriptor( cudnnPoolingDescriptor_t poolingDesc );
-/* Pooling functions: All of the form "output = alpha * Op(inputs) + beta * output" */
+cudnnStatus_t cudnnDestroyPoolingDescriptor(
+ cudnnPoolingDescriptor_t poolingDesc );
/* Function to perform forward pooling */
cudnnStatus_t cudnnPoolingForward( cudnnHandle_t handle,
- const cudnnPoolingDescriptor_t poolingDesc,
- const void *alpha,
- const cudnnTensorDescriptor_t srcDesc,
- const void *srcData,
- const void *beta,
- const cudnnTensorDescriptor_t destDesc,
- void *destData
- );
+ const cudnnPoolingDescriptor_t poolingDesc,
+ const void *alpha,
+ const cudnnTensorDescriptor_t srcDesc,
+ const void *srcData,
+ const void *beta,
+ const cudnnTensorDescriptor_t destDesc,
+ void *destData
+ );
/* Function to perform backward pooling */
cudnnStatus_t cudnnPoolingBackward( cudnnHandle_t handle,
- const cudnnPoolingDescriptor_t poolingDesc,
- const void *alpha,
- const cudnnTensorDescriptor_t srcDesc,
- const void *srcData,
- const cudnnTensorDescriptor_t srcDiffDesc,
- const void *srcDiffData,
- const cudnnTensorDescriptor_t destDesc,
- const void *destData,
- const void *beta,
- const cudnnTensorDescriptor_t destDiffDesc,
- void *destDiffData
+ const cudnnPoolingDescriptor_t poolingDesc,
+ const void *alpha,
+ const cudnnTensorDescriptor_t srcDesc,
+ const void *srcData,
+ const cudnnTensorDescriptor_t srcDiffDesc,
+ const void *srcDiffData,
+ const cudnnTensorDescriptor_t destDesc,
+ const void *destData,
+ const void *beta,
+ const cudnnTensorDescriptor_t destDiffDesc,
+ void *destDiffData
);
typedef enum
@@ -346,36 +347,35 @@ typedef enum
/* Function to perform forward activation */
cudnnStatus_t cudnnActivationForward( cudnnHandle_t handle,
- cudnnActivationMode_t mode,
- const void *alpha,
- const cudnnTensorDescriptor_t srcDesc,
- const void *srcData,
- const void *beta,
- const cudnnTensorDescriptor_t destDesc,
- void *destData
- );
+ cudnnActivationMode_t mode,
+ const void *alpha,
+ const cudnnTensorDescriptor_t srcDesc,
+ const void *srcData,
+ const void *beta,
+ const cudnnTensorDescriptor_t destDesc,
+ void *destData
+ );
/* Function to perform backward activation */
cudnnStatus_t cudnnActivationBackward( cudnnHandle_t handle,
- cudnnActivationMode_t mode,
- const void *alpha,
- const cudnnTensorDescriptor_t srcDesc,
- const void *srcData,
- const cudnnTensorDescriptor_t srcDiffDesc,
- const void *srcDiffData,
- const cudnnTensorDescriptor_t destDesc,
- const void *destData,
- const void *beta,
- const cudnnTensorDescriptor_t destDiffDesc,
- void *destDiffData
- );
+ cudnnActivationMode_t mode,
+ const void *alpha,
+ const cudnnTensorDescriptor_t srcDesc,
+ const void *srcData,
+ const cudnnTensorDescriptor_t srcDiffDesc,
+ const void *srcDiffData,
+ const cudnnTensorDescriptor_t destDesc,
+ const void *destData,
+ const void *beta,
+ const cudnnTensorDescriptor_t destDiffDesc,
+ void *destDiffData
+ );
]]
-local ok
-ok,err = pcall(function() cudnn.C = ffi.load('libcudnn') end)
+local ok,err = pcall(function() cudnn.C = ffi.load('libcudnn') end)
if not ok then
print(err)
- error([['libcudnn.so not found in library path.
+ error([['libcudnn.so not found in library path.
Please install CuDNN from https://developer.nvidia.com/cuDNN
Then make sure all the files named as libcudnn.so* are placed in your library load path (for example /usr/local/lib , or manually add a path to LD_LIBRARY_PATH)
]])
generated by cgit v1.2.3 (git 2.25.1) at 2024-07-09 22:52:49 +0300