venv/lib/python3.10/site-packages/nvidia/cudnn/include/cudnn_ops.h

/*
 * Copyright 2014-2023 NVIDIA Corporation.  All rights reserved.
 *
 * NOTICE TO LICENSEE:
 *
 * This source code and/or documentation ("Licensed Deliverables") are
 * subject to NVIDIA intellectual property rights under U.S. and
 * international Copyright laws.
 *
 * These Licensed Deliverables contained herein is PROPRIETARY and
 * CONFIDENTIAL to NVIDIA and is being provided under the terms and
 * conditions of a form of NVIDIA software license agreement by and
 * between NVIDIA and Licensee ("License Agreement") or electronically
 * accepted by Licensee.  Notwithstanding any terms or conditions to
 * the contrary in the License Agreement, reproduction or disclosure
 * of the Licensed Deliverables to any third party without the express
 * written consent of NVIDIA is prohibited.
 *
 * NOTWITHSTANDING ANY TERMS OR CONDITIONS TO THE CONTRARY IN THE
 * LICENSE AGREEMENT, NVIDIA MAKES NO REPRESENTATION ABOUT THE
 * SUITABILITY OF THESE LICENSED DELIVERABLES FOR ANY PURPOSE.  IT IS
 * PROVIDED "AS IS" WITHOUT EXPRESS OR IMPLIED WARRANTY OF ANY KIND.
 * NVIDIA DISCLAIMS ALL WARRANTIES WITH REGARD TO THESE LICENSED
 * DELIVERABLES, INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY,
 * NONINFRINGEMENT, AND FITNESS FOR A PARTICULAR PURPOSE.
 * NOTWITHSTANDING ANY TERMS OR CONDITIONS TO THE CONTRARY IN THE
 * LICENSE AGREEMENT, IN NO EVENT SHALL NVIDIA BE LIABLE FOR ANY
 * SPECIAL, INDIRECT, INCIDENTAL, OR CONSEQUENTIAL DAMAGES, OR ANY
 * DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS,
 * WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS
 * ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE
 * OF THESE LICENSED DELIVERABLES.
 *
 * U.S. Government End Users.  These Licensed Deliverables are a
 * "commercial item" as that term is defined at 48 C.F.R. 2.101 (OCT
 * 1995), consisting of "commercial computer software" and "commercial
 * computer software documentation" as such terms are used in 48
 * C.F.R. 12.212 (SEPT 1995) and is provided to the U.S. Government
 * only as a commercial end item.  Consistent with 48 C.F.R.12.212 and
 * 48 C.F.R. 227.7202-1 through 227.7202-4 (JUNE 1995), all
 * U.S. Government End Users acquire the Licensed Deliverables with
 * only those rights set forth herein.
 *
 * Any use of the Licensed Deliverables in individual and commercial
 * software must include, in the user documentation and internal
 * comments to the code, the above Disclaimer and U.S. Government End
 * Users Notice.
 */

/*
 *  cudnn_ops : cuDNN's basic definitions and basic operations.
 */

#if !defined(CUDNN_OPS_H_)
#define CUDNN_OPS_H_

#include <stdint.h>

#include "cudnn_version.h"
#include "cudnn_graph.h"

/* These version numbers are autogenerated, do not edit manually. */
#define CUDNN_OPS_MAJOR 9
#define CUDNN_OPS_MINOR 10
#define CUDNN_OPS_PATCH 2

#if (CUDNN_OPS_MAJOR != CUDNN_MAJOR) || (CUDNN_OPS_MINOR != CUDNN_MINOR) || (CUDNN_OPS_PATCH != CUDNN_PATCHLEVEL)
#error Version mismatch in cuDNN OPS INFER!!!
#endif

#if defined(__cplusplus)
extern "C" {
#endif

/* Data structures to represent Image/Filter and the Neural Network Layer */
typedef struct cudnnTensorStruct *cudnnTensorDescriptor_t;
typedef struct cudnnPoolingStruct *cudnnPoolingDescriptor_t CUDNN_DEPRECATED;
typedef struct cudnnFilterStruct *cudnnFilterDescriptor_t CUDNN_DEPRECATED;
typedef struct cudnnLRNStruct *cudnnLRNDescriptor_t;
typedef struct cudnnActivationStruct *cudnnActivationDescriptor_t CUDNN_DEPRECATED;
typedef struct cudnnSpatialTransformerStruct *cudnnSpatialTransformerDescriptor_t;
typedef struct cudnnOpTensorStruct *cudnnOpTensorDescriptor_t CUDNN_DEPRECATED;
typedef struct cudnnReduceTensorStruct *cudnnReduceTensorDescriptor_t CUDNN_DEPRECATED;
typedef struct cudnnCTCLossStruct *cudnnCTCLossDescriptor_t;
typedef struct cudnnTensorTransformStruct *cudnnTensorTransformDescriptor_t CUDNN_DEPRECATED;
/*
 * CUDNN Determinism
 */
typedef enum {
    CUDNN_NON_DETERMINISTIC = 0,
    CUDNN_DETERMINISTIC     = 1,
} cudnnDeterminism_t;

/* Create an instance of a generic Tensor descriptor */
cudnnStatus_t CUDNNWINAPI
cudnnCreateTensorDescriptor(cudnnTensorDescriptor_t *tensorDesc);

cudnnStatus_t CUDNNWINAPI
cudnnSetTensor4dDescriptor(cudnnTensorDescriptor_t tensorDesc,
                           cudnnTensorFormat_t format,
                           cudnnDataType_t dataType, /* image data type */
                           int n,                    /* number of inputs (batch size) */
                           int c,                    /* number of input feature maps */
                           int h,                    /* height of input section */
                           int w);                   /* width of input section */

cudnnStatus_t CUDNNWINAPI
cudnnSetTensor4dDescriptorEx(cudnnTensorDescriptor_t tensorDesc,
                             cudnnDataType_t dataType, /* image data type */
                             int n,                    /* number of inputs (batch size) */
                             int c,                    /* number of input feature maps */
                             int h,                    /* height of input section */
                             int w,                    /* width of input section */
                             int nStride,
                             int cStride,
                             int hStride,
                             int wStride);

cudnnStatus_t CUDNNWINAPI
cudnnGetTensor4dDescriptor(const cudnnTensorDescriptor_t tensorDesc,
                           cudnnDataType_t *dataType, /* image data type */
                           int *n,                    /* number of inputs (batch size) */
                           int *c,                    /* number of input feature maps  */
                           int *h,                    /* height of input section */
                           int *w,                    /* width of input section */
                           int *nStride,
                           int *cStride,
                           int *hStride,
                           int *wStride);

cudnnStatus_t CUDNNWINAPI
cudnnSetTensorNdDescriptor(cudnnTensorDescriptor_t tensorDesc,
                           cudnnDataType_t dataType,
                           int nbDims,
                           const int dimA[],
                           const int strideA[]);

cudnnStatus_t CUDNNWINAPI
cudnnSetTensorNdDescriptorEx(cudnnTensorDescriptor_t tensorDesc,
                             cudnnTensorFormat_t format,
                             cudnnDataType_t dataType,
                             int nbDims,
                             const int dimA[]);

cudnnStatus_t CUDNNWINAPI
cudnnGetTensorNdDescriptor(const cudnnTensorDescriptor_t tensorDesc,
                           int nbDimsRequested,
                           cudnnDataType_t *dataType,
                           int *nbDims,
                           int dimA[],
                           int strideA[]);

cudnnStatus_t CUDNNWINAPI
cudnnGetTensorSizeInBytes(const cudnnTensorDescriptor_t tensorDesc, size_t *size);

/* PixelOffset( n, c, h, w ) = n *input_stride + c * feature_stride + h * h_stride + w * w_stride

   1)Example of all images in row major order one batch of features after the other (with an optional padding on row)
   input_stride :  c x h x h_stride
   feature_stride : h x h_stride
   h_stride  :  >= w  ( h_stride = w if no padding)
   w_stride  : 1


   2)Example of all images in row major with features maps interleaved
   input_stride :  c x h x h_stride
   feature_stride : 1
   h_stride  :  w x c
   w_stride  : c

   3)Example of all images in column major order one batch of features after the other (with optional padding on column)
   input_stride :  c x w x w_stride
   feature_stride : w x w_stride
   h_stride  :  1
   w_stride  :  >= h

*/

/* Destroy an instance of Tensor4d descriptor */
cudnnStatus_t CUDNNWINAPI
cudnnDestroyTensorDescriptor(cudnnTensorDescriptor_t tensorDesc);

/* Fold/unfold transforms */
typedef enum {
    CUDNN_TRANSFORM_FOLD   = 0U,
    CUDNN_TRANSFORM_UNFOLD = 1U,
} cudnnFoldingDirection_t;

/** Create a destination descriptor for cudnnTransformTensor */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnInitTransformDest(const cudnnTensorTransformDescriptor_t transformDesc,
                       const cudnnTensorDescriptor_t srcDesc,
                       cudnnTensorDescriptor_t destDesc,
                       size_t *destSizeInBytes);

/** Create an empty tensor transform descriptor */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnCreateTensorTransformDescriptor(cudnnTensorTransformDescriptor_t *transformDesc);

/** Initialize a previously created tensor transform descriptor. */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnSetTensorTransformDescriptor(cudnnTensorTransformDescriptor_t transformDesc,
                                  const uint32_t nbDims,
                                  const cudnnTensorFormat_t destFormat,
                                  const int32_t padBeforeA[],
                                  const int32_t padAfterA[],
                                  const uint32_t foldA[],
                                  const cudnnFoldingDirection_t direction);

/**
 * Retrieves the values stored in a previously initialized tensor transform
 * descriptor.
 */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnGetTensorTransformDescriptor(cudnnTensorTransformDescriptor_t transformDesc,
                                  uint32_t nbDimsRequested,
                                  cudnnTensorFormat_t *destFormat,
                                  int32_t padBeforeA[],
                                  int32_t padAfterA[],
                                  uint32_t foldA[],
                                  cudnnFoldingDirection_t *direction);

/**
 * Destroys a previously created tensor transform descriptor.
 */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnDestroyTensorTransformDescriptor(cudnnTensorTransformDescriptor_t transformDesc);

/* Tensor layout conversion helper (y = alpha * x + beta * y) */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnTransformTensor(cudnnHandle_t handle,
                     const void *alpha,
                     const cudnnTensorDescriptor_t xDesc,
                     const void *x,
                     const void *beta,
                     const cudnnTensorDescriptor_t yDesc,
                     void *y);

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnTransformTensorEx(cudnnHandle_t handle,
                       const cudnnTensorTransformDescriptor_t transDesc,
                       const void *alpha,
                       const cudnnTensorDescriptor_t srcDesc,
                       const void *srcData,
                       const void *beta,
                       const cudnnTensorDescriptor_t destDesc,
                       void *destData);

/* Tensor Bias addition : C = alpha * A + beta * C  */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnAddTensor(cudnnHandle_t handle,
               const void *alpha,
               const cudnnTensorDescriptor_t aDesc,
               const void *A,
               const void *beta,
               const cudnnTensorDescriptor_t cDesc,
               void *C);

/*
 * CUDNN OpTensor op type
 */
typedef enum {
    CUDNN_OP_TENSOR_ADD  = 0,
    CUDNN_OP_TENSOR_MUL  = 1,
    CUDNN_OP_TENSOR_MIN  = 2,
    CUDNN_OP_TENSOR_MAX  = 3,
    CUDNN_OP_TENSOR_SQRT = 4,
    CUDNN_OP_TENSOR_NOT  = 5,
} cudnnOpTensorOp_t;

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnCreateOpTensorDescriptor(cudnnOpTensorDescriptor_t *opTensorDesc);

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnSetOpTensorDescriptor(cudnnOpTensorDescriptor_t opTensorDesc,
                           cudnnOpTensorOp_t opTensorOp,
                           cudnnDataType_t opTensorCompType,
                           cudnnNanPropagation_t opTensorNanOpt);

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnGetOpTensorDescriptor(const cudnnOpTensorDescriptor_t opTensorDesc,
                           cudnnOpTensorOp_t *opTensorOp,
                           cudnnDataType_t *opTensorCompType,
                           cudnnNanPropagation_t *opTensorNanOpt);

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnDestroyOpTensorDescriptor(cudnnOpTensorDescriptor_t opTensorDesc);

/* Tensor operation : C = op( alpha1 * A, alpha2 * B ) + beta * C */
/* B tensor is ignored for CUDNN_OP_TENSOR_SQRT, CUDNN_OP_TENSOR_NOT. */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnOpTensor(cudnnHandle_t handle,
              const cudnnOpTensorDescriptor_t opTensorDesc,
              const void *alpha1,
              const cudnnTensorDescriptor_t aDesc,
              const void *A,
              const void *alpha2,
              const cudnnTensorDescriptor_t bDesc,
              const void *B,
              const void *beta,
              const cudnnTensorDescriptor_t cDesc,
              void *C);

/*
 * CUDNN ReduceTensor indices type
 */
typedef enum {
    CUDNN_REDUCE_TENSOR_NO_INDICES        = 0,
    CUDNN_REDUCE_TENSOR_FLATTENED_INDICES = 1,
} cudnnReduceTensorIndices_t CUDNN_DEPRECATED;

/*
 * CUDNN tensor indices type size (all unsigned)
 * Currently not supported, default is 32 bit unsigned.
 */
typedef enum {
    CUDNN_32BIT_INDICES = 0,
    CUDNN_64BIT_INDICES = 1,
    CUDNN_16BIT_INDICES = 2,
    CUDNN_8BIT_INDICES  = 3,
} cudnnIndicesType_t CUDNN_DEPRECATED;

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnCreateReduceTensorDescriptor(cudnnReduceTensorDescriptor_t *reduceTensorDesc);

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnSetReduceTensorDescriptor(cudnnReduceTensorDescriptor_t reduceTensorDesc,
                               cudnnReduceTensorOp_t reduceTensorOp,
                               cudnnDataType_t reduceTensorCompType,
                               cudnnNanPropagation_t reduceTensorNanOpt,
                               cudnnReduceTensorIndices_t reduceTensorIndices,
                               cudnnIndicesType_t reduceTensorIndicesType);

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnGetReduceTensorDescriptor(const cudnnReduceTensorDescriptor_t reduceTensorDesc,
                               cudnnReduceTensorOp_t *reduceTensorOp,
                               cudnnDataType_t *reduceTensorCompType,
                               cudnnNanPropagation_t *reduceTensorNanOpt,
                               cudnnReduceTensorIndices_t *reduceTensorIndices,
                               cudnnIndicesType_t *reduceTensorIndicesType);

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnDestroyReduceTensorDescriptor(cudnnReduceTensorDescriptor_t reduceTensorDesc);

/* Helper function to return the minimum size of the index space to be passed to the reduction given the input and
 * output tensors */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnGetReductionIndicesSize(cudnnHandle_t handle,
                             const cudnnReduceTensorDescriptor_t reduceTensorDesc,
                             const cudnnTensorDescriptor_t aDesc,
                             const cudnnTensorDescriptor_t cDesc,
                             size_t *sizeInBytes);

/* Helper function to return the minimum size of the workspace to be passed to the reduction given the input and output
 * tensors */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnGetReductionWorkspaceSize(cudnnHandle_t handle,
                               const cudnnReduceTensorDescriptor_t reduceTensorDesc,
                               const cudnnTensorDescriptor_t aDesc,
                               const cudnnTensorDescriptor_t cDesc,
                               size_t *sizeInBytes);

/* Tensor operation : C = reduce op( alpha * A ) + beta * C */
/* The NaN propagation enum applies to only the min and max reduce ops; the other reduce ops propagate NaN as usual. */
/* The indices space is ignored for reduce ops other than min or max. */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnReduceTensor(cudnnHandle_t handle,
                  const cudnnReduceTensorDescriptor_t reduceTensorDesc,
                  void *indices,
                  size_t indicesSizeInBytes,
                  void *workspace,
                  size_t workspaceSizeInBytes,
                  const void *alpha,
                  const cudnnTensorDescriptor_t aDesc,
                  const void *A,
                  const void *beta,
                  const cudnnTensorDescriptor_t cDesc,
                  void *C);

/* Set all values of a tensor to a given value : y[i] = value[0] */
cudnnStatus_t CUDNNWINAPI
cudnnSetTensor(cudnnHandle_t handle, const cudnnTensorDescriptor_t yDesc, void *y, const void *valuePtr);

/* Scale all values of a tensor by a given factor : y[i] = alpha * y[i] */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnScaleTensor(cudnnHandle_t handle, const cudnnTensorDescriptor_t yDesc, void *y, const void *alpha);

/* Create an instance of FilterStruct */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnCreateFilterDescriptor(cudnnFilterDescriptor_t *filterDesc);

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnSetFilter4dDescriptor(cudnnFilterDescriptor_t filterDesc,
                           cudnnDataType_t dataType, /* image data type */
                           cudnnTensorFormat_t format,
                           int k,  /* number of output feature maps */
                           int c,  /* number of input feature maps */
                           int h,  /* height of each input filter */
                           int w); /* width of  each input filter */

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnGetFilter4dDescriptor(const cudnnFilterDescriptor_t filterDesc,
                           cudnnDataType_t *dataType, /* image data type */
                           cudnnTensorFormat_t *format,
                           int *k,  /* number of output feature maps */
                           int *c,  /* number of input feature maps */
                           int *h,  /* height of each input filter */
                           int *w); /* width of  each input filter */

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnSetFilterNdDescriptor(cudnnFilterDescriptor_t filterDesc,
                           cudnnDataType_t dataType, /* image data type */
                           cudnnTensorFormat_t format,
                           int nbDims,
                           const int filterDimA[]);

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnGetFilterNdDescriptor(const cudnnFilterDescriptor_t filterDesc,
                           int nbDimsRequested,
                           cudnnDataType_t *dataType, /* image data type */
                           cudnnTensorFormat_t *format,
                           int *nbDims,
                           int filterDimA[]);
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnGetFilterSizeInBytes(const cudnnFilterDescriptor_t filterDesc, size_t *size);

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnTransformFilter(cudnnHandle_t handle,
                     const cudnnTensorTransformDescriptor_t transDesc,
                     const void *alpha,
                     const cudnnFilterDescriptor_t srcDesc,
                     const void *srcData,
                     const void *beta,
                     const cudnnFilterDescriptor_t destDesc,
                     void *destData);

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnDestroyFilterDescriptor(cudnnFilterDescriptor_t filterDesc);

/*
 *  softmax algorithm
 */
typedef enum {
    CUDNN_SOFTMAX_FAST     = 0, /* straightforward implementation */
    CUDNN_SOFTMAX_ACCURATE = 1, /* subtract max from every point to avoid overflow */
    CUDNN_SOFTMAX_LOG      = 2
} 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 */
} cudnnSoftmaxMode_t;

/* Softmax functions: All of the form "output = alpha * Op(inputs) + beta * output" */

/* Function to perform forward softmax */
cudnnStatus_t CUDNNWINAPI
cudnnSoftmaxForward(cudnnHandle_t handle,
                    cudnnSoftmaxAlgorithm_t algo,
                    cudnnSoftmaxMode_t mode,
                    const void *alpha,
                    const cudnnTensorDescriptor_t xDesc,
                    const void *x,
                    const void *beta,
                    const cudnnTensorDescriptor_t yDesc,
                    void *y);

/*
 *  pooling mode
 */
typedef enum {
    CUDNN_POOLING_MAX                           = 0,
    CUDNN_POOLING_AVERAGE_COUNT_INCLUDE_PADDING = 1, /* count for average includes padded values */
    CUDNN_POOLING_AVERAGE_COUNT_EXCLUDE_PADDING = 2, /* count for average does not include padded values */
    CUDNN_POOLING_MAX_DETERMINISTIC             = 3
} cudnnPoolingMode_t CUDNN_DEPRECATED;

/* Create an instance of pooling descriptor */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnCreatePoolingDescriptor(cudnnPoolingDescriptor_t *poolingDesc);

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnSetPooling2dDescriptor(cudnnPoolingDescriptor_t poolingDesc,
                            cudnnPoolingMode_t mode,
                            cudnnNanPropagation_t maxpoolingNanOpt,
                            int windowHeight,
                            int windowWidth,
                            int verticalPadding,
                            int horizontalPadding,
                            int verticalStride,
                            int horizontalStride);

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnGetPooling2dDescriptor(const cudnnPoolingDescriptor_t poolingDesc,
                            cudnnPoolingMode_t *mode,
                            cudnnNanPropagation_t *maxpoolingNanOpt,
                            int *windowHeight,
                            int *windowWidth,
                            int *verticalPadding,
                            int *horizontalPadding,
                            int *verticalStride,
                            int *horizontalStride);

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnSetPoolingNdDescriptor(cudnnPoolingDescriptor_t poolingDesc,
                            const cudnnPoolingMode_t mode,
                            const cudnnNanPropagation_t maxpoolingNanOpt,
                            int nbDims,
                            const int windowDimA[],
                            const int paddingA[],
                            const int strideA[]);

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnGetPoolingNdDescriptor(const cudnnPoolingDescriptor_t poolingDesc,
                            int nbDimsRequested,
                            cudnnPoolingMode_t *mode,
                            cudnnNanPropagation_t *maxpoolingNanOpt,
                            int *nbDims,
                            int windowDimA[],
                            int paddingA[],
                            int strideA[]);

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnGetPoolingNdForwardOutputDim(const cudnnPoolingDescriptor_t poolingDesc,
                                  const cudnnTensorDescriptor_t inputTensorDesc,
                                  int nbDims,
                                  int outputTensorDimA[]);

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnGetPooling2dForwardOutputDim(const cudnnPoolingDescriptor_t poolingDesc,
                                  const cudnnTensorDescriptor_t inputTensorDesc,
                                  int *n,
                                  int *c,
                                  int *h,
                                  int *w);

/* Destroy an instance of pooling descriptor */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnDestroyPoolingDescriptor(cudnnPoolingDescriptor_t poolingDesc);

/* Pooling functions: All of the form "output = alpha * Op(inputs) + beta * output" */

/* Function to perform forward pooling */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnPoolingForward(cudnnHandle_t handle,
                    const cudnnPoolingDescriptor_t poolingDesc,
                    const void *alpha,
                    const cudnnTensorDescriptor_t xDesc,
                    const void *x,
                    const void *beta,
                    const cudnnTensorDescriptor_t yDesc,
                    void *y);

/* Activation functions: All of the form "output = alpha * Op(inputs) + beta * output" */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnCreateActivationDescriptor(cudnnActivationDescriptor_t *activationDesc);

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnSetActivationDescriptor(cudnnActivationDescriptor_t activationDesc,
                             cudnnActivationMode_t mode,
                             cudnnNanPropagation_t reluNanOpt,
                             double coef); /* ceiling for clipped RELU, alpha for ELU */

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnGetActivationDescriptor(const cudnnActivationDescriptor_t activationDesc,
                             cudnnActivationMode_t *mode,
                             cudnnNanPropagation_t *reluNanOpt,
                             double *coef); /* ceiling for clipped RELU, alpha for ELU */

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnSetActivationDescriptorSwishBeta(cudnnActivationDescriptor_t activationDesc, double swish_beta);

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnGetActivationDescriptorSwishBeta(cudnnActivationDescriptor_t activationDesc, double *swish_beta);

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnDestroyActivationDescriptor(cudnnActivationDescriptor_t activationDesc);

/* Function to perform forward activation  */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnActivationForward(cudnnHandle_t handle,
                       cudnnActivationDescriptor_t activationDesc,
                       const void *alpha,
                       const cudnnTensorDescriptor_t xDesc,
                       const void *x,
                       const void *beta,
                       const cudnnTensorDescriptor_t yDesc,
                       void *y);

/*
 * Create an instance of LRN (Local Response Normalization) descriptor
 * Uses lrnN=5, lrnAlpha=1e-4, lrnBeta=0.75, lrnK=2.0 as defaults from Krizhevsky'12 ImageNet paper
 */
cudnnStatus_t CUDNNWINAPI
cudnnCreateLRNDescriptor(cudnnLRNDescriptor_t *normDesc);

#define CUDNN_LRN_MIN_N 1       /* minimum allowed lrnN */
#define CUDNN_LRN_MAX_N 16      /* maximum allowed lrnN */
#define CUDNN_LRN_MIN_K 1e-5    /* minimum allowed lrnK */
#define CUDNN_LRN_MIN_BETA 0.01 /* minimum allowed lrnBeta */

/* LRN layer mode */
typedef enum {
    CUDNN_LRN_CROSS_CHANNEL_DIM1 = 0, /* Normalize across tensor's dimA[1] dimension */
} cudnnLRNMode_t;

/*
 * Uses a window [center-lookBehind, center+lookAhead], where
 * lookBehind = floor( (lrnN-1)/2 ), lookAhead = lrnN-lookBehind-1.
 * Values of double parameters cast to tensor data type.
 */
cudnnStatus_t CUDNNWINAPI
cudnnSetLRNDescriptor(cudnnLRNDescriptor_t normDesc, unsigned lrnN, double lrnAlpha, double lrnBeta, double lrnK);
/*
 * Retrieve the settings currently stored in an LRN layer descriptor
 * Any of the provided pointers can be NULL (no corresponding value will be returned)
 */
cudnnStatus_t CUDNNWINAPI
cudnnGetLRNDescriptor(cudnnLRNDescriptor_t normDesc, unsigned *lrnN, double *lrnAlpha, double *lrnBeta, double *lrnK);

/* Destroy an instance of LRN descriptor */
cudnnStatus_t CUDNNWINAPI
cudnnDestroyLRNDescriptor(cudnnLRNDescriptor_t lrnDesc);

/* LRN functions: output = alpha * normalize(x) + beta * old_y */

/* LRN cross-channel forward computation. Double parameters cast to tensor data type */
cudnnStatus_t CUDNNWINAPI
cudnnLRNCrossChannelForward(cudnnHandle_t handle,
                            cudnnLRNDescriptor_t normDesc,
                            cudnnLRNMode_t lrnMode,
                            const void *alpha,
                            const cudnnTensorDescriptor_t xDesc,
                            const void *x,
                            const void *beta,
                            const cudnnTensorDescriptor_t yDesc,
                            void *y);

typedef enum {
    CUDNN_DIVNORM_PRECOMPUTED_MEANS = 0,
} cudnnDivNormMode_t;

/* LCN/divisive normalization functions: y = alpha * normalize(x) + beta * y */
cudnnStatus_t CUDNNWINAPI
cudnnDivisiveNormalizationForward(cudnnHandle_t handle,
                                  cudnnLRNDescriptor_t normDesc,
                                  cudnnDivNormMode_t mode,
                                  const void *alpha,
                                  const cudnnTensorDescriptor_t xDesc, /* same desc for means, temp, temp2 */
                                  const void *x,
                                  const void *means, /* if NULL, means are assumed to be zero */
                                  void *temp,
                                  void *temp2,
                                  const void *beta,
                                  const cudnnTensorDescriptor_t yDesc,
                                  void *y);

typedef enum {
    /* bnScale, bnBias tensor dims are 1xCxHxWx.. (one value per CHW...-slice, normalized over N slice) */
    CUDNN_BATCHNORM_PER_ACTIVATION = 0,

    /* bnScale, bnBias tensor dims are 1xCx1x1 (one value per C-dim normalized over Nx1xHxW subtensors) */
    CUDNN_BATCHNORM_SPATIAL = 1,

    /*
     * bnScale, bnBias tensor dims are 1xCx1x1 (one value per C-dim normalized over Nx1xHxW subtensors).
     * May be faster than CUDNN_BATCHNORM_SPATIAL but imposes some limits on the range of values
     */
    CUDNN_BATCHNORM_SPATIAL_PERSISTENT = 2,
} cudnnBatchNormMode_t CUDNN_DEPRECATED;

#define CUDNN_BN_MIN_EPSILON 0.0 /* Minimum epsilon allowed to be used in the Batch Normalization formula */

/*
 * Derives a tensor descriptor from layer data descriptor for BatchNormalization
 * scale, invVariance, bnBias, bnScale tensors. Use this tensor desc for
 * bnScaleBiasMeanVarDesc and bnScaleBiasDiffDesc in Batch Normalization forward and backward functions.
 */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnDeriveBNTensorDescriptor(cudnnTensorDescriptor_t derivedBnDesc,
                              const cudnnTensorDescriptor_t xDesc,
                              cudnnBatchNormMode_t mode);

typedef enum {
    CUDNN_BATCHNORM_OPS_BN                = 0, /* do batch normalization only */
    CUDNN_BATCHNORM_OPS_BN_ACTIVATION     = 1, /* do batchNorm, then activation */
    CUDNN_BATCHNORM_OPS_BN_ADD_ACTIVATION = 2, /* do batchNorm, then elemWiseAdd, then activation */
} cudnnBatchNormOps_t CUDNN_DEPRECATED;

/*
 * Performs Batch Normalization during Inference:
 * y[i] = bnScale[k]*(x[i]-estimatedMean[k])/sqrt(epsilon+estimatedVariance[k]) + bnBias[k]
 * with bnScale, bnBias, runningMean, runningInvVariance tensors indexed
 * according to spatial or per-activation mode. Refer to cudnnBatchNormalizationForwardTraining
 * above for notes on function arguments.
 */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnBatchNormalizationForwardInference(cudnnHandle_t handle,
                                        cudnnBatchNormMode_t mode,
                                        const void *alpha, /* alpha[0] = result blend factor */
                                        const void *beta,  /* beta[0] = dest layer blend factor */
                                        const cudnnTensorDescriptor_t xDesc,
                                        const void *x, /* NxCxHxW */
                                        const cudnnTensorDescriptor_t yDesc,
                                        void *y, /* NxCxHxW */
                                        const cudnnTensorDescriptor_t bnScaleBiasMeanVarDesc,
                                        const void *bnScale,
                                        const void *bnBias,
                                        const void *estimatedMean,
                                        const void *estimatedVariance,
                                        double epsilon);

typedef enum {
    /* bnScale, bnBias tensor dims are 1xCxHxWx.. (one value per CHW...-slice, normalized over N slice) */
    CUDNN_NORM_PER_ACTIVATION = 0,

    /* bnScale, bnBias tensor dims are 1xCx1x1 (one value per C-dim normalized over Nx1xHxW subtensors) */
    CUDNN_NORM_PER_CHANNEL = 1,
} cudnnNormMode_t CUDNN_DEPRECATED;

typedef enum { CUDNN_NORM_ALGO_STANDARD = 0, CUDNN_NORM_ALGO_PERSIST = 1 } cudnnNormAlgo_t CUDNN_DEPRECATED;

/*
 * Derives a tensor descriptor from layer data descriptor for Normalization
 * scale, invVariance, bnBias, bnScale tensors. Use this tensor desc for
 * normScaleBiasMeanVarDesc and normScaleBiasDiffDesc in Normalization forward and backward functions.
 */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnDeriveNormTensorDescriptor(cudnnTensorDescriptor_t derivedNormScaleBiasDesc,
                                cudnnTensorDescriptor_t derivedNormMeanVarDesc,
                                const cudnnTensorDescriptor_t xDesc,
                                cudnnNormMode_t mode,
                                int groupCnt); /* Place hold for future work, should be set to 1 now*/

typedef enum {
    CUDNN_NORM_OPS_NORM                = 0, /* do normalization only */
    CUDNN_NORM_OPS_NORM_ACTIVATION     = 1, /* do Norm, then activation */
    CUDNN_NORM_OPS_NORM_ADD_ACTIVATION = 2, /* do Norm, then elemWiseAdd, then activation */
} cudnnNormOps_t CUDNN_DEPRECATED;

/*
 * Performs Normalization during Inference:
 * y[i] = normScale[k]*(x[i]-estimatedMean[k])/sqrt(epsilon+estimatedVariance[k]) + normBias[k]
 * with normScale, normBias, runningMean, runningInvVariance tensors indexed
 * according to per-channel or per-activation mode. Refer to cudnnNormalizationForwardTraining
 * above for notes on function arguments.
 */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnNormalizationForwardInference(cudnnHandle_t handle,
                                   cudnnNormMode_t mode,
                                   cudnnNormOps_t normOps,
                                   cudnnNormAlgo_t algo,
                                   const void *alpha, /* alpha[0] = result blend factor */
                                   const void *beta,  /* beta[0] = dest layer blend factor */
                                   const cudnnTensorDescriptor_t xDesc,
                                   const void *x, /* NxCxHxW */
                                   const cudnnTensorDescriptor_t normScaleBiasDesc,
                                   const void *normScale,
                                   const void *normBias,
                                   const cudnnTensorDescriptor_t normMeanVarDesc,
                                   const void *estimatedMean,
                                   const void *estimatedVariance,
                                   const cudnnTensorDescriptor_t zDesc,
                                   const void *z,
                                   cudnnActivationDescriptor_t activationDesc,
                                   const cudnnTensorDescriptor_t yDesc,
                                   void *y, /* NxCxHxW */
                                   double epsilon,
                                   int groupCnt); /* Place hold for future work*/

/* APIs for spatial transformer network*/
typedef enum {
    CUDNN_SAMPLER_BILINEAR = 0,
} cudnnSamplerType_t;

cudnnStatus_t CUDNNWINAPI
cudnnCreateSpatialTransformerDescriptor(cudnnSpatialTransformerDescriptor_t *stDesc);

cudnnStatus_t CUDNNWINAPI
cudnnSetSpatialTransformerNdDescriptor(cudnnSpatialTransformerDescriptor_t stDesc,
                                       cudnnSamplerType_t samplerType,
                                       cudnnDataType_t dataType,
                                       const int nbDims,
                                       const int dimA[]);

cudnnStatus_t CUDNNWINAPI
cudnnDestroySpatialTransformerDescriptor(cudnnSpatialTransformerDescriptor_t stDesc);

cudnnStatus_t CUDNNWINAPI
cudnnSpatialTfGridGeneratorForward(cudnnHandle_t handle,
                                   const cudnnSpatialTransformerDescriptor_t stDesc,
                                   const void *theta,
                                   void *grid);

cudnnStatus_t CUDNNWINAPI
cudnnSpatialTfSamplerForward(cudnnHandle_t handle,
                             cudnnSpatialTransformerDescriptor_t stDesc,
                             const void *alpha,
                             const cudnnTensorDescriptor_t xDesc,
                             const void *x,
                             const void *grid,
                             const void *beta,
                             cudnnTensorDescriptor_t yDesc,
                             void *y);

typedef struct cudnnDropoutStruct *cudnnDropoutDescriptor_t;

cudnnStatus_t CUDNNWINAPI
cudnnCreateDropoutDescriptor(cudnnDropoutDescriptor_t *dropoutDesc);

cudnnStatus_t CUDNNWINAPI
cudnnDestroyDropoutDescriptor(cudnnDropoutDescriptor_t dropoutDesc);

/*helper function to determine size of the states to be passed to cudnnSetDropoutDescriptor */
cudnnStatus_t CUDNNWINAPI
cudnnDropoutGetStatesSize(cudnnHandle_t handle, size_t *sizeInBytes);

/*helper function to determine size of the reserve space to be passed to dropout forward/backward calls */
cudnnStatus_t CUDNNWINAPI
cudnnDropoutGetReserveSpaceSize(cudnnTensorDescriptor_t xdesc, size_t *sizeInBytes);

cudnnStatus_t CUDNNWINAPI
cudnnSetDropoutDescriptor(cudnnDropoutDescriptor_t dropoutDesc,
                          cudnnHandle_t handle,
                          float dropout,
                          void *states,
                          size_t stateSizeInBytes,
                          unsigned long long seed);

/* Restores the dropout descriptor to a previously saved-off state */
cudnnStatus_t CUDNNWINAPI
cudnnRestoreDropoutDescriptor(cudnnDropoutDescriptor_t dropoutDesc,
                              cudnnHandle_t handle,
                              float dropout,
                              void *states,
                              size_t stateSizeInBytes,
                              unsigned long long seed);

cudnnStatus_t CUDNNWINAPI
cudnnGetDropoutDescriptor(cudnnDropoutDescriptor_t dropoutDesc,
                          cudnnHandle_t handle,
                          float *dropout,
                          void **states,
                          unsigned long long *seed);

cudnnStatus_t CUDNNWINAPI
cudnnDropoutForward(cudnnHandle_t handle,
                    const cudnnDropoutDescriptor_t dropoutDesc,
                    const cudnnTensorDescriptor_t xdesc,
                    const void *x,
                    const cudnnTensorDescriptor_t ydesc,
                    void *y,
                    void *reserveSpace,
                    size_t reserveSpaceSizeInBytes);

/* TODO: move these enums out to the appropriate submodule */
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_FFT                   = 4,
    CUDNN_CONVOLUTION_FWD_ALGO_FFT_TILING            = 5,
    CUDNN_CONVOLUTION_FWD_ALGO_WINOGRAD              = 6,
    CUDNN_CONVOLUTION_FWD_ALGO_WINOGRAD_NONFUSED     = 7,
    CUDNN_CONVOLUTION_FWD_ALGO_COUNT                 = 8
} cudnnConvolutionFwdAlgo_t;

typedef enum {
    CUDNN_CONVOLUTION_BWD_FILTER_ALGO_0                 = 0, /* non-deterministic */
    CUDNN_CONVOLUTION_BWD_FILTER_ALGO_1                 = 1,
    CUDNN_CONVOLUTION_BWD_FILTER_ALGO_FFT               = 2,
    CUDNN_CONVOLUTION_BWD_FILTER_ALGO_3                 = 3, /* non-deterministic */
    CUDNN_CONVOLUTION_BWD_FILTER_ALGO_WINOGRAD          = 4, /* not implemented */
    CUDNN_CONVOLUTION_BWD_FILTER_ALGO_WINOGRAD_NONFUSED = 5,
    CUDNN_CONVOLUTION_BWD_FILTER_ALGO_FFT_TILING        = 6,
    CUDNN_CONVOLUTION_BWD_FILTER_ALGO_COUNT             = 7
} cudnnConvolutionBwdFilterAlgo_t;

typedef enum {
    CUDNN_CONVOLUTION_BWD_DATA_ALGO_0                 = 0, /* non-deterministic */
    CUDNN_CONVOLUTION_BWD_DATA_ALGO_1                 = 1,
    CUDNN_CONVOLUTION_BWD_DATA_ALGO_FFT               = 2,
    CUDNN_CONVOLUTION_BWD_DATA_ALGO_FFT_TILING        = 3,
    CUDNN_CONVOLUTION_BWD_DATA_ALGO_WINOGRAD          = 4,
    CUDNN_CONVOLUTION_BWD_DATA_ALGO_WINOGRAD_NONFUSED = 5,
    CUDNN_CONVOLUTION_BWD_DATA_ALGO_COUNT             = 6
} cudnnConvolutionBwdDataAlgo_t;

typedef enum { CUDNN_CTC_LOSS_ALGO_DETERMINISTIC = 0, CUDNN_CTC_LOSS_ALGO_NON_DETERMINISTIC = 1 } cudnnCTCLossAlgo_t;

/*
 * \brief Cross-library version checker.
 * This function is implemented differently in each sub-library. Each sublib
 * checks whether its own version matches that of its dependencies.
 * \returns CUDNN_STATUS_SUCCESS if the version check passes,
 *          CUDNN_STATUS_SUBLIBRARY_VERSION_MISMATCH if the versions are inconsistent.
 */
cudnnStatus_t CUDNNWINAPI
cudnnOpsVersionCheck(void);

/* Function to perform backward softmax */
cudnnStatus_t CUDNNWINAPI
cudnnSoftmaxBackward(cudnnHandle_t handle,
                     cudnnSoftmaxAlgorithm_t algo,
                     cudnnSoftmaxMode_t mode,
                     const void *alpha,
                     const cudnnTensorDescriptor_t yDesc,
                     const void *y,
                     const cudnnTensorDescriptor_t dyDesc,
                     const void *dy,
                     const void *beta,
                     const cudnnTensorDescriptor_t dxDesc,
                     void *dx);

/* Function to perform backward pooling */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnPoolingBackward(cudnnHandle_t handle,
                     const cudnnPoolingDescriptor_t poolingDesc,
                     const void *alpha,
                     const cudnnTensorDescriptor_t yDesc,
                     const void *y,
                     const cudnnTensorDescriptor_t dyDesc,
                     const void *dy,
                     const cudnnTensorDescriptor_t xDesc,
                     const void *x,
                     const void *beta,
                     const cudnnTensorDescriptor_t dxDesc,
                     void *dx);

/* Function to perform backward activation  */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnActivationBackward(cudnnHandle_t handle,
                        cudnnActivationDescriptor_t activationDesc,
                        const void *alpha,
                        const cudnnTensorDescriptor_t yDesc,
                        const void *y,
                        const cudnnTensorDescriptor_t dyDesc,
                        const void *dy,
                        const cudnnTensorDescriptor_t xDesc,
                        const void *x,
                        const void *beta,
                        const cudnnTensorDescriptor_t dxDesc,
                        void *dx);

/* LRN cross-channel backward computation. Double parameters cast to tensor data type */
cudnnStatus_t CUDNNWINAPI
cudnnLRNCrossChannelBackward(cudnnHandle_t handle,
                             cudnnLRNDescriptor_t normDesc,
                             cudnnLRNMode_t lrnMode,
                             const void *alpha,
                             const cudnnTensorDescriptor_t yDesc,
                             const void *y,
                             const cudnnTensorDescriptor_t dyDesc,
                             const void *dy,
                             const cudnnTensorDescriptor_t xDesc,
                             const void *x,
                             const void *beta,
                             const cudnnTensorDescriptor_t dxDesc,
                             void *dx);

cudnnStatus_t CUDNNWINAPI
cudnnDivisiveNormalizationBackward(cudnnHandle_t handle,
                                   cudnnLRNDescriptor_t normDesc,
                                   cudnnDivNormMode_t mode,
                                   const void *alpha,
                                   const cudnnTensorDescriptor_t xDesc, /* same desc for x, means, dy, temp, temp2 */
                                   const void *x,
                                   const void *means, /* if NULL, means are assumed to be zero */
                                   const void *dy,
                                   void *temp,
                                   void *temp2,
                                   const void *beta,
                                   const cudnnTensorDescriptor_t dXdMeansDesc, /* same desc for dx, dMeans */
                                   void *dx,                                   /* output x differential */
                                   void *dMeans); /* output means differential, can be NULL */

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnGetBatchNormalizationForwardTrainingExWorkspaceSize(cudnnHandle_t handle,
                                                         cudnnBatchNormMode_t mode,
                                                         cudnnBatchNormOps_t bnOps,
                                                         const cudnnTensorDescriptor_t xDesc,
                                                         const cudnnTensorDescriptor_t zDesc,
                                                         const cudnnTensorDescriptor_t yDesc,
                                                         const cudnnTensorDescriptor_t bnScaleBiasMeanVarDesc,
                                                         const cudnnActivationDescriptor_t activationDesc,
                                                         size_t *sizeInBytes);

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnGetBatchNormalizationBackwardExWorkspaceSize(cudnnHandle_t handle,
                                                  cudnnBatchNormMode_t mode,
                                                  cudnnBatchNormOps_t bnOps,
                                                  const cudnnTensorDescriptor_t xDesc,
                                                  const cudnnTensorDescriptor_t yDesc,
                                                  const cudnnTensorDescriptor_t dyDesc,
                                                  const cudnnTensorDescriptor_t dzDesc,
                                                  const cudnnTensorDescriptor_t dxDesc,
                                                  const cudnnTensorDescriptor_t dBnScaleBiasDesc,
                                                  const cudnnActivationDescriptor_t activationDesc,
                                                  size_t *sizeInBytes);

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnGetBatchNormalizationTrainingExReserveSpaceSize(cudnnHandle_t handle,
                                                     cudnnBatchNormMode_t mode,
                                                     cudnnBatchNormOps_t bnOps,
                                                     const cudnnActivationDescriptor_t activationDesc,
                                                     const cudnnTensorDescriptor_t xDesc,
                                                     size_t *sizeInBytes);

/* Computes y = BN(x). Also accumulates moving averages of mean and inverse variances */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnBatchNormalizationForwardTraining(
    cudnnHandle_t handle,
    cudnnBatchNormMode_t mode,

    const void *alpha, /* alpha[0] = result blend factor */
    const void *beta,  /* beta[0] = dest layer blend factor */

    const cudnnTensorDescriptor_t xDesc,
    const void *x, /* NxCxHxW */
    const cudnnTensorDescriptor_t yDesc,
    void *y, /* NxCxHxW */

    /* Shared desc for the next 6 tensors in the argument list.
       Data type to be set as follows:
       type = (typeOf(x) == double) ? double : float
       Dimensions for this descriptor depend on normalization mode
       - Spatial Normalization : tensors are expected to have dims 1xCx1x1
        (normalization is performed across NxHxW)
       - Per-Activation Normalization : tensors are expected to have dims of 1xCxHxW
        (normalization is performed across N) */
    const cudnnTensorDescriptor_t bnScaleBiasMeanVarDesc,

    /* 'Gamma' and 'Beta' respectively in Ioffe and Szegedy's paper's notation */
    const void *bnScale,
    const void *bnBias,

    /* MUST use factor=1 in the very first call of a complete training cycle.
       Use a factor=1/(1+n) at N-th call to the function to get
       Cumulative Moving Average (CMA) behavior
       CMA[n] = (x[1]+...+x[n])/n
       Since CMA[n+1] = (n*CMA[n]+x[n+1])/(n+1) =
       ((n+1)*CMA[n]-CMA[n])/(n+1) + x[n+1]/(n+1) =
       CMA[n]*(1-1/(n+1)) + x[n+1]*1/(n+1) */
    double exponentialAverageFactor,

    /* Used in Training phase only.
       runningMean = newMean*factor + runningMean*(1-factor) */
    void *resultRunningMean,
    /* Output in training mode, input in inference. Is the moving average
       of  variance[x] (factor is applied in the same way as for runningMean) */
    void *resultRunningVariance,

    /* Has to be >= CUDNN_BN_MIN_EPSILON. Should be the same in forward and backward functions. */
    double epsilon,

    /* Optionally save intermediate results from the forward pass here
       - can be reused to speed up backward pass. NULL if unused */
    void *resultSaveMean,
    void *resultSaveInvVariance);

/* Computes y = relu(BN(x) + z). Also accumulates moving averages of mean and inverse variances */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnBatchNormalizationForwardTrainingEx(
    cudnnHandle_t handle,
    cudnnBatchNormMode_t mode,
    cudnnBatchNormOps_t bnOps,

    const void *alpha, /* alpha[0] = result blend factor */
    const void *beta,  /* beta[0] = dest layer blend factor */

    const cudnnTensorDescriptor_t xDesc,
    const void *xData,
    const cudnnTensorDescriptor_t zDesc,
    const void *zData,
    const cudnnTensorDescriptor_t yDesc,
    void *yData,

    const cudnnTensorDescriptor_t bnScaleBiasMeanVarDesc,
    const void *bnScale,
    const void *bnBias,

    double exponentialAverageFactor,
    void *resultRunningMean,
    void *resultRunningVariance,

    /* Has to be >= CUDNN_BN_MIN_EPSILON. Should be the same in forward and backward functions. */
    double epsilon,

    /* Optionally save intermediate results from the forward pass here
       - can be reused to speed up backward pass. NULL if unused */
    void *resultSaveMean,
    void *resultSaveInvVariance,

    cudnnActivationDescriptor_t activationDesc,
    void *workspace,
    size_t workSpaceSizeInBytes,
    void *reserveSpace,
    size_t reserveSpaceSizeInBytes);

/* Performs backward pass of Batch Normalization layer. Returns x gradient,
 * bnScale gradient and bnBias gradient */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnBatchNormalizationBackward(cudnnHandle_t handle,
                                cudnnBatchNormMode_t mode,
                                const void *alphaDataDiff,
                                const void *betaDataDiff,
                                const void *alphaParamDiff,
                                const void *betaParamDiff,
                                const cudnnTensorDescriptor_t xDesc, /* same desc for x, dx, dy */
                                const void *x,
                                const cudnnTensorDescriptor_t dyDesc,
                                const void *dy,
                                const cudnnTensorDescriptor_t dxDesc,
                                void *dx,
                                /* Shared tensor desc for the 4 tensors below */
                                const cudnnTensorDescriptor_t dBnScaleBiasDesc,
                                const void *bnScale, /* bnBias doesn't affect backpropagation */
                                /* scale and bias diff are not backpropagated below this layer */
                                void *dBnScaleResult,
                                void *dBnBiasResult,
                                /* Same epsilon as forward pass */
                                double epsilon,

                                /* Optionally cached intermediate results from
                                   forward pass */
                                const void *savedMean,
                                const void *savedInvVariance);

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnBatchNormalizationBackwardEx(cudnnHandle_t handle,
                                  cudnnBatchNormMode_t mode,
                                  cudnnBatchNormOps_t bnOps,

                                  const void *alphaDataDiff,
                                  const void *betaDataDiff,
                                  const void *alphaParamDiff,
                                  const void *betaParamDiff,
                                  const cudnnTensorDescriptor_t xDesc,
                                  const void *xData,
                                  const cudnnTensorDescriptor_t yDesc,
                                  const void *yData,
                                  const cudnnTensorDescriptor_t dyDesc,
                                  const void *dyData,
                                  const cudnnTensorDescriptor_t dzDesc,
                                  void *dzData,
                                  const cudnnTensorDescriptor_t dxDesc,
                                  void *dxData,

                                  /* Shared tensor desc for the 4 tensors below */
                                  const cudnnTensorDescriptor_t dBnScaleBiasDesc,
                                  const void *bnScaleData,
                                  const void *bnBiasData, /* needed if there is activation */
                                  void *dBnScaleData,
                                  void *dBnBiasData,
                                  double epsilon, /* Same epsilon as forward pass */

                                  /* Optionally cached intermediate results from
                                     forward pass */
                                  const void *savedMean,
                                  const void *savedInvVariance,
                                  cudnnActivationDescriptor_t activationDesc,
                                  void *workSpace,
                                  size_t workSpaceSizeInBytes,
                                  void *reserveSpace,
                                  size_t reserveSpaceSizeInBytes);

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnGetNormalizationForwardTrainingWorkspaceSize(cudnnHandle_t handle,
                                                  cudnnNormMode_t mode,
                                                  cudnnNormOps_t normOps,
                                                  cudnnNormAlgo_t algo,
                                                  const cudnnTensorDescriptor_t xDesc,
                                                  const cudnnTensorDescriptor_t zDesc,
                                                  const cudnnTensorDescriptor_t yDesc,
                                                  const cudnnTensorDescriptor_t normScaleBiasDesc,
                                                  const cudnnActivationDescriptor_t activationDesc,
                                                  const cudnnTensorDescriptor_t normMeanVarDesc,
                                                  size_t *sizeInBytes,
                                                  int groupCnt); /* Place hold for future work, should be set to 1 now*/

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnGetNormalizationBackwardWorkspaceSize(cudnnHandle_t handle,
                                           cudnnNormMode_t mode,
                                           cudnnNormOps_t normOps,
                                           cudnnNormAlgo_t algo,
                                           const cudnnTensorDescriptor_t xDesc,
                                           const cudnnTensorDescriptor_t yDesc,
                                           const cudnnTensorDescriptor_t dyDesc,
                                           const cudnnTensorDescriptor_t dzDesc,
                                           const cudnnTensorDescriptor_t dxDesc,
                                           const cudnnTensorDescriptor_t dNormScaleBiasDesc,
                                           const cudnnActivationDescriptor_t activationDesc,
                                           const cudnnTensorDescriptor_t normMeanVarDesc,
                                           size_t *sizeInBytes,
                                           int groupCnt); /* Place hold for future work, should be set to 1 now*/

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnGetNormalizationTrainingReserveSpaceSize(cudnnHandle_t handle,
                                              cudnnNormMode_t mode,
                                              cudnnNormOps_t normOps,
                                              cudnnNormAlgo_t algo,
                                              const cudnnActivationDescriptor_t activationDesc,
                                              const cudnnTensorDescriptor_t xDesc,
                                              size_t *sizeInBytes,
                                              int groupCnt); /* Place hold for future work, should be set to 1 now*/

/* Computes y = relu(Norm(x) + z). Also accumulates moving averages of mean and inverse variances */
CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnNormalizationForwardTraining(cudnnHandle_t handle,
                                  cudnnNormMode_t mode,
                                  cudnnNormOps_t normOps,
                                  cudnnNormAlgo_t algo,
                                  const void *alpha, /* alpha[0] = result blend factor */
                                  const void *beta,  /* beta[0] = dest layer blend factor */
                                  const cudnnTensorDescriptor_t xDesc,
                                  const void *xData,
                                  const cudnnTensorDescriptor_t normScaleBiasDesc,
                                  const void *normScale,
                                  const void *normBias,
                                  double exponentialAverageFactor,
                                  const cudnnTensorDescriptor_t normMeanVarDesc,
                                  void *resultRunningMean,
                                  void *resultRunningVariance,
                                  /* Has to be >= 0. Should be the same in forward and backward functions. */
                                  double epsilon,
                                  /* Optionally save intermediate results from the forward pass here
                                     - can be reused to speed up backward pass. NULL if unused */
                                  void *resultSaveMean,
                                  void *resultSaveInvVariance,
                                  cudnnActivationDescriptor_t activationDesc,
                                  const cudnnTensorDescriptor_t zDesc,
                                  const void *zData,
                                  const cudnnTensorDescriptor_t yDesc,
                                  void *yData,
                                  void *workspace,
                                  size_t workSpaceSizeInBytes,
                                  void *reserveSpace,
                                  size_t reserveSpaceSizeInBytes,
                                  int groupCnt); /* Place hold for future work, should be set to 1 now*/

CUDNN_DEPRECATED cudnnStatus_t CUDNNWINAPI
cudnnNormalizationBackward(cudnnHandle_t handle,
                           cudnnNormMode_t mode,
                           cudnnNormOps_t normOps,
                           cudnnNormAlgo_t algo,
                           const void *alphaDataDiff,
                           const void *betaDataDiff,
                           const void *alphaParamDiff,
                           const void *betaParamDiff,
                           const cudnnTensorDescriptor_t xDesc,
                           const void *xData,
                           const cudnnTensorDescriptor_t yDesc,
                           const void *yData,
                           const cudnnTensorDescriptor_t dyDesc,
                           const void *dyData,
                           const cudnnTensorDescriptor_t dzDesc,
                           void *dzData,
                           const cudnnTensorDescriptor_t dxDesc,
                           void *dxData,
                           /* Shared tensor desc for the 4 tensors below */
                           const cudnnTensorDescriptor_t dNormScaleBiasDesc,
                           const void *normScaleData,
                           const void *normBiasData, /* needed if there is activation */
                           void *dNormScaleData,
                           void *dNormBiasData,
                           double epsilon, /* Same epsilon as forward pass */
                           const cudnnTensorDescriptor_t normMeanVarDesc,
                           /* Optionally cached intermediate results from
                              forward pass */
                           const void *savedMean,
                           const void *savedInvVariance,
                           cudnnActivationDescriptor_t activationDesc,
                           void *workSpace,
                           size_t workSpaceSizeInBytes,
                           void *reserveSpace,
                           size_t reserveSpaceSizeInBytes,
                           int groupCnt); /* Place hold for future work, should be set to 1 now*/

cudnnStatus_t CUDNNWINAPI
cudnnSpatialTfGridGeneratorBackward(cudnnHandle_t handle,
                                    const cudnnSpatialTransformerDescriptor_t stDesc,
                                    const void *dgrid,
                                    void *dtheta);

cudnnStatus_t CUDNNWINAPI
cudnnSpatialTfSamplerBackward(cudnnHandle_t handle,
                              cudnnSpatialTransformerDescriptor_t stDesc,
                              const void *alpha,
                              const cudnnTensorDescriptor_t xDesc,
                              const void *x,
                              const void *beta,
                              const cudnnTensorDescriptor_t dxDesc,
                              void *dx,
                              const void *alphaDgrid,
                              const cudnnTensorDescriptor_t dyDesc,
                              const void *dy,
                              const void *grid,
                              const void *betaDgrid,
                              void *dgrid);

cudnnStatus_t CUDNNWINAPI
cudnnDropoutBackward(cudnnHandle_t handle,
                     const cudnnDropoutDescriptor_t dropoutDesc,
                     const cudnnTensorDescriptor_t dydesc,
                     const void *dy,
                     const cudnnTensorDescriptor_t dxdesc,
                     void *dx,
                     void *reserveSpace,
                     size_t reserveSpaceSizeInBytes);

#if defined(__cplusplus)
}
#endif

#endif /* CUDNN_OPS_H_ */