Skip to content

Commit

Permalink
Write the output pointer of the RNN functions
Browse files Browse the repository at this point in the history
  • Loading branch information
@jcuda
jcuda committed Jul 6, 2017
1 parent a817592 commit 53de60e
Showing 1 changed file with 46 additions and 46 deletions.
92 changes: 46 additions & 46 deletions JCudnnJNI/src/JCudnn.cpp
View file
Expand Up @@ -884,30 +884,30 @@ JNIEXPORT jint JNICALL Java_jcuda_jcudnn_JCudnn_cudnnGetTensorNdDescriptorNative
}

/**
* <pre>
* PixelOffset( n, c, h, w ) = n *input_stride + c * feature_stride + h * h_stride + w * w_stride
* <pre>
* 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
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
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
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
* </pre>
*/
* </pre>
*/
/** Destroy an instance of Tensor4d descriptor */
JNIEXPORT jint JNICALL Java_jcuda_jcudnn_JCudnn_cudnnDestroyTensorDescriptorNative(JNIEnv *env, jclass cls, jobject tensorDesc)
{
Expand Down Expand Up @@ -2750,8 +2750,8 @@ JNIEXPORT jint JNICALL Java_jcuda_jcudnn_JCudnn_cudnnGetConvolutionForwardAlgori
}

/**
* convolution algorithm (which requires potentially some workspace)
*/
* 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*/
JNIEXPORT jint JNICALL Java_jcuda_jcudnn_JCudnn_cudnnGetConvolutionForwardWorkspaceSizeNative(JNIEnv *env, jclass cls, jobject handle, jobject xDesc, jobject wDesc, jobject convDesc, jobject yDesc, jint algo, jlongArray sizeInBytes)
{
Expand Down Expand Up @@ -3325,8 +3325,8 @@ JNIEXPORT jint JNICALL Java_jcuda_jcudnn_JCudnn_cudnnGetConvolutionBackwardFilte
}

/**
* convolution algorithm (which requires potentially some workspace)
*/
* 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*/
JNIEXPORT jint JNICALL Java_jcuda_jcudnn_JCudnn_cudnnGetConvolutionBackwardFilterWorkspaceSizeNative(JNIEnv *env, jclass cls, jobject handle, jobject xDesc, jobject dyDesc, jobject convDesc, jobject gradDesc, jint algo, jlongArray sizeInBytes)
{
Expand Down Expand Up @@ -5389,11 +5389,11 @@ JNIEXPORT jint JNICALL Java_jcuda_jcudnn_JCudnn_cudnnActivationBackwardNative(JN
}

/**
* <pre>
* <pre>
* 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
* </pre>
*/
* </pre>
*/
JNIEXPORT jint JNICALL Java_jcuda_jcudnn_JCudnn_cudnnCreateLRNDescriptorNative(JNIEnv *env, jclass cls, jobject normDesc)
{
// Null-checks for non-primitive arguments
Expand Down Expand Up @@ -5425,12 +5425,12 @@ JNIEXPORT jint JNICALL Java_jcuda_jcudnn_JCudnn_cudnnCreateLRNDescriptorNative(J
}

/**
* <pre>
* <pre>
* 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.
* </pre>
*/
* </pre>
*/
JNIEXPORT jint JNICALL Java_jcuda_jcudnn_JCudnn_cudnnSetLRNDescriptorNative(JNIEnv *env, jclass cls, jobject normDesc, jint lrnN, jdouble lrnAlpha, jdouble lrnBeta, jdouble lrnK)
{
// Null-checks for non-primitive arguments
Expand Down Expand Up @@ -5478,11 +5478,11 @@ JNIEXPORT jint JNICALL Java_jcuda_jcudnn_JCudnn_cudnnSetLRNDescriptorNative(JNIE
}

/**
* <pre>
* <pre>
* Retrieve the settings currently stored in an LRN layer descriptor
* Any of the provided pointers can be NULL (no corresponding value will be returned)
* </pre>
*/
* </pre>
*/
JNIEXPORT jint JNICALL Java_jcuda_jcudnn_JCudnn_cudnnGetLRNDescriptorNative(JNIEnv *env, jclass cls, jobject normDesc, jintArray lrnN, jdoubleArray lrnAlpha, jdoubleArray lrnBeta, jdoubleArray lrnK)
{
// Null-checks for non-primitive arguments
Expand Down Expand Up @@ -6049,12 +6049,12 @@ JNIEXPORT jint JNICALL Java_jcuda_jcudnn_JCudnn_cudnnDivisiveNormalizationBackwa
}

/**
* <pre>
* <pre>
* 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.
* </pre>
*/
* </pre>
*/
JNIEXPORT jint JNICALL Java_jcuda_jcudnn_JCudnn_cudnnDeriveBNTensorDescriptorNative(JNIEnv *env, jclass cls, jobject derivedBnDesc, jobject xDesc, jint mode)
{
// Null-checks for non-primitive arguments
Expand Down Expand Up @@ -6247,14 +6247,14 @@ JNIEXPORT jint JNICALL Java_jcuda_jcudnn_JCudnn_cudnnBatchNormalizationForwardTr
}

/**
* <pre>
* <pre>
* 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.
* </pre>
*/
* </pre>
*/
JNIEXPORT jint JNICALL Java_jcuda_jcudnn_JCudnn_cudnnBatchNormalizationForwardInferenceNative(JNIEnv *env, jclass cls, jobject handle, jint mode, jobject alpha, jobject beta, jobject xDesc, jobject x, jobject yDesc, jobject y, jobject bnScaleBiasMeanVarDesc, jobject bnScale, jobject bnBias, jobject estimatedMean, jobject estimatedVariance, jdouble epsilon)
{
// Null-checks for non-primitive arguments
Expand Down Expand Up @@ -7759,7 +7759,7 @@ JNIEXPORT jint JNICALL Java_jcuda_jcudnn_JCudnn_cudnnGetRNNLinLayerMatrixParamsN
void * w_native = NULL;
int linLayerID_native = 0;
cudnnFilterDescriptor_t linLayerMatDesc_native;
void * * linLayerMat_native = NULL;
void * linLayerMat_native;

// Obtain native variable values
handle_native = (cudnnHandle_t)getNativePointerValue(env, handle);
Expand All @@ -7770,10 +7770,10 @@ JNIEXPORT jint JNICALL Java_jcuda_jcudnn_JCudnn_cudnnGetRNNLinLayerMatrixParamsN
w_native = (void *)getPointer(env, w);
linLayerID_native = (int)linLayerID;
linLayerMatDesc_native = (cudnnFilterDescriptor_t)getNativePointerValue(env, linLayerMatDesc);
linLayerMat_native = (void * *)getPointer(env, linLayerMat);
// linLayerMat is write-only

// Native function call
cudnnStatus_t jniResult_native = cudnnGetRNNLinLayerMatrixParams(handle_native, rnnDesc_native, layer_native, xDesc_native, wDesc_native, w_native, linLayerID_native, linLayerMatDesc_native, linLayerMat_native);
cudnnStatus_t jniResult_native = cudnnGetRNNLinLayerMatrixParams(handle_native, rnnDesc_native, layer_native, xDesc_native, wDesc_native, w_native, linLayerID_native, linLayerMatDesc_native, &linLayerMat_native);

// Write back native variable values
// handle is read-only
Expand All @@ -7784,7 +7784,7 @@ JNIEXPORT jint JNICALL Java_jcuda_jcudnn_JCudnn_cudnnGetRNNLinLayerMatrixParamsN
// w is a native pointer
// linLayerID is primitive
// linLayerMatDesc is read-only
// linLayerMat is a native pointer
setNativePointerValue(env, linLayerMat, (jlong)linLayerMat_native);

// Return the result
jint jniResult = (jint)jniResult_native;
Expand Down Expand Up @@ -7845,7 +7845,7 @@ JNIEXPORT jint JNICALL Java_jcuda_jcudnn_JCudnn_cudnnGetRNNLinLayerBiasParamsNat
void * w_native = NULL;
int linLayerID_native = 0;
cudnnFilterDescriptor_t linLayerBiasDesc_native;
void * * linLayerBias_native = NULL;
void * linLayerBias_native;

// Obtain native variable values
handle_native = (cudnnHandle_t)getNativePointerValue(env, handle);
Expand All @@ -7856,10 +7856,10 @@ JNIEXPORT jint JNICALL Java_jcuda_jcudnn_JCudnn_cudnnGetRNNLinLayerBiasParamsNat
w_native = (void *)getPointer(env, w);
linLayerID_native = (int)linLayerID;
linLayerBiasDesc_native = (cudnnFilterDescriptor_t)getNativePointerValue(env, linLayerBiasDesc);
linLayerBias_native = (void * *)getPointer(env, linLayerBias);
// linLayerBias is write-only

// Native function call
cudnnStatus_t jniResult_native = cudnnGetRNNLinLayerBiasParams(handle_native, rnnDesc_native, layer_native, xDesc_native, wDesc_native, w_native, linLayerID_native, linLayerBiasDesc_native, linLayerBias_native);
cudnnStatus_t jniResult_native = cudnnGetRNNLinLayerBiasParams(handle_native, rnnDesc_native, layer_native, xDesc_native, wDesc_native, w_native, linLayerID_native, linLayerBiasDesc_native, &linLayerBias_native);

// Write back native variable values
// handle is read-only
Expand All @@ -7870,7 +7870,7 @@ JNIEXPORT jint JNICALL Java_jcuda_jcudnn_JCudnn_cudnnGetRNNLinLayerBiasParamsNat
// w is a native pointer
// linLayerID is primitive
// linLayerBiasDesc is read-only
// linLayerBias is a native pointer
setNativePointerValue(env, linLayerBias, (jlong)linLayerBias_native);

// Return the result
jint jniResult = (jint)jniResult_native;
Expand Down

0 comments on commit 53de60e

Please sign in to comment.