cavis/libnd4j/include/loops/cuda/summarystatsreduce.cu

486 lines
21 KiB
Plaintext
Raw Normal View History

2019-06-06 14:21:15 +02:00
/*******************************************************************************
* Copyright (c) 2015-2018 Skymind, Inc.
*
* This program and the accompanying materials are made available under the
* terms of the Apache License, Version 2.0 which is available at
* https://www.apache.org/licenses/LICENSE-2.0.
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
* WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
* License for the specific language governing permissions and limitations
* under the License.
*
* SPDX-License-Identifier: Apache-2.0
******************************************************************************/
//
// @author raver119@gmail.com
//
#include <pointercast.h>
#include <types/types.h>
#include <types/float16.h>
#include <op_boilerplate.h>
#include <loops/summarystatsreduce.h>
#include <helpers/shape.h>
#include <helpers/TAD.h>
#include <dll.h>
#include <Environment.h>
#include <cuda.h>
#include <cuda_runtime.h>
#include <helpers/DebugHelper.h>
#include <specials_cuda.h>
using namespace simdOps;
namespace functions {
namespace summarystats {
template <typename X, typename Z>
void _CUDA_G summaryStatsReduceT(int op, void *dx, Nd4jLong *xShapeInfo, int xRank, void *extraParams, void *z, Nd4jLong *zShapeInfo, int zRank, int *dimension, int dimensionLength, int postProcessOrNot,bool biasCorrected,int *allocationBuffer, void *reductionBuffer, Nd4jLong *tadOnlyShapeInfo, Nd4jLong *tadOffsets) {
functions::summarystats::SummaryStatsReduce<X,Z>::transform(op,dx,xShapeInfo,extraParams,z,zShapeInfo,dimension,dimensionLength,biasCorrected,allocationBuffer,reductionBuffer,tadOnlyShapeInfo,tadOffsets);
}
/**
*
* @param sPartialsRef
* @param tid
* @param extraParams
*/
template<typename X, typename Z>
template<typename OpType>
_CUDA_D void SummaryStatsReduce<X,Z>::aggregatePartials(SummaryStatsData<X> **sPartialsRef, Nd4jLong tid, Nd4jLong numElements, void *vextraParams) {
// start the shared memory loop on the next power of 2 less
// than the block size. If block size is not a power of 2,
// accumulate the intermediate sums in the remainder range.
auto extraParams = static_cast<Z*>(vextraParams);
SummaryStatsData<X> *sPartials = *sPartialsRef;
Nd4jLong floorPow2 = blockDim.x;
if (floorPow2 & (floorPow2 - 1)) {
while (floorPow2 & (floorPow2 - 1)) {
floorPow2 &= floorPow2 - 1;
}
if (tid >= floorPow2) {
SummaryStatsData<X> prev = sPartials[tid - floorPow2];
SummaryStatsData<X> curr = sPartials[tid];
sPartials[tid - floorPow2] = update(prev, curr, extraParams);
}
__syncthreads();
}
for (Nd4jLong activeThreads = floorPow2 >> 1; activeThreads; activeThreads >>= 1) {
if (tid < activeThreads && tid + activeThreads < numElements) {
SummaryStatsData<X> curr = sPartials[tid];
SummaryStatsData<X> next = sPartials[tid + activeThreads];
sPartials[tid] = update(curr, next, extraParams);
}
__syncthreads();
}
};
/**
* @param n n is the number of
* elements to loop through
* @param dx the data to operate on
* @param xVectorInfo the meta data for the vector:
* 0 is the offset
* 1 is the increment/stride
* 2 is the real length of the buffer (n and dx.length won't always be the same)
* 3 is the element wise stride for the buffer
* 4 is the number of elements it takes to get to the next row/column/tensor
* @param gpuInformation
* 0 is the block size
* 1 is the grid size
* 2 is the shared memory size
* @param problemDefinition
* 0 is the number of elements per vector
* 1 is the number of vectors
*/
template<typename X, typename Z>
template<typename OpType>
_CUDA_D void SummaryStatsReduce<X,Z>::transform(void *vx, Nd4jLong *xShapeInfo,
void *vextraParams,
void *vz, Nd4jLong *zShapeInfo,
int *dimension, int dimensionLength,
int postProcessOrNot,
int *allocationBuffer, void *vreductionBuffer,
Nd4jLong *tadOnlyShapeInfo, Nd4jLong *tadOffsets) {
auto dx = static_cast<X*>(vx);
auto z = static_cast<Z*>(vz);
auto extraParams = static_cast<Z*>(vextraParams);
auto reductionBuffer = static_cast<Z*>(vreductionBuffer);
int tid = blockIdx.x * blockDim.x + threadIdx.x;
__shared__ volatile int resultScalar;
__shared__ int xElementWiseStride;
int numElements = blockDim.x;
//shared memory space for storing intermediate results
__shared__ SummaryStatsData<X> *sPartials;
if(threadIdx.x == 0) {
extern __shared__ unsigned char shmem[];
sPartials = reinterpret_cast<SummaryStatsData<X>*>(shmem);
}
__syncthreads();
Z startingVal = startingValue(dx);
SummaryStatsData<X> val;
val.initWithValue(startingVal);
val.n = 0;
sPartials[threadIdx.x] = val;
//length for the tad
__shared__ volatile int xLength;
__shared__ volatile int resultLength;
SummaryStatsData<X> reduction;
reduction.initWithValue(0.0);
reduction.n = 0;
if (threadIdx.x == 0) {
if (zShapeInfo != nullptr)
resultLength = shape::length(zShapeInfo);
else resultLength = 1;
if (dimensionLength == 1) {
if (resultLength == 1 && (dimension == nullptr || dimension[0] == MAX_DIMENSION))
resultScalar = 1;
else
resultScalar = 0;
}
else
resultScalar = 0;
if (resultLength == 1)
resultScalar = 1;
auto xStride = shape::stride(xShapeInfo);
auto xOrder = shape::order(xShapeInfo);
if (dimension != nullptr && (dimension[0] != MAX_DIMENSION && dimensionLength == 1)) {
xElementWiseStride = xStride[dimension[0]];
}
else {
xElementWiseStride = shape::elementWiseStride(xShapeInfo);
}
xLength = shape::length(xShapeInfo);
}
__syncthreads();
if (!resultScalar) {
__shared__ int tadLength;
__shared__ int tadEWS;
__shared__ int numTads;
if (threadIdx.x == 0) {
tadLength = shape::length(tadOnlyShapeInfo);//shape::tadLength(xShapeInfo, dimension, dimensionLength);
tadEWS = shape::elementWiseStride(tadOnlyShapeInfo);
numTads = shape::length(xShapeInfo) / tadLength;
}
__syncthreads();
if (tadEWS == 0) {
for (int r = blockIdx.x; r < numTads; r += gridDim.x) {
auto tadOffsetForBlock = tadOffsets[r];
val.initWithValue(startingVal);
val.n = 0;
sPartials[threadIdx.x] = val;
for (int i = threadIdx.x; i < tadLength; i += blockDim.x) {
auto xOffset = tadOffsetForBlock + shape::getIndexOffset(i, tadOnlyShapeInfo, tadLength);
SummaryStatsData<X> indexVal2;
indexVal2.initWithValue(dx[xOffset]);
sPartials[threadIdx.x] = update(sPartials[threadIdx.x], OpType::op(indexVal2, extraParams), extraParams);
}
__syncthreads();
aggregatePartials<OpType>(&sPartials, threadIdx.x, nd4j::math::nd4j_min<int>(blockDim.x, tadLength), extraParams);
__syncthreads();
if (threadIdx.x == 0) {
z[r] = OpType::getValue(postProcessOrNot, sPartials[threadIdx.x]);
}
}
}
else {
for (int i = blockIdx.x; i < numTads; i += gridDim.x) {
auto tadOffsetForBlock = tadOffsets[i];
val.initWithValue(startingVal);
val.n = 0;
sPartials[threadIdx.x] = val;
for (int x = threadIdx.x; x < tadLength; x += blockDim.x) {
auto indexX = tadOffsetForBlock + x * tadEWS;
SummaryStatsData<X> indexVal2;
indexVal2.initWithValue(dx[indexX]);
sPartials[threadIdx.x] = update(sPartials[threadIdx.x], OpType::op(indexVal2, extraParams), extraParams);
}
__syncthreads();
aggregatePartials<OpType>(&sPartials, threadIdx.x, nd4j::math::nd4j_min<int>(blockDim.x, tadLength), extraParams);
__syncthreads();
if (threadIdx.x == 0) {
z[i] = OpType::getValue(postProcessOrNot, sPartials[threadIdx.x]); //postProcess(sPartials[0],tadLength ,extraParams);
}
}
}
}
else if (resultScalar) {
__shared__ int n;
if (threadIdx.x == 0) {
xElementWiseStride = shape::elementWiseStride(xShapeInfo);
n = shape::length(xShapeInfo);
}
__syncthreads();
if (xElementWiseStride >= 1) {
for (Nd4jLong i = tid; i < n; i += (blockDim.x * gridDim.x)) {
SummaryStatsData<X> indexVal2;
indexVal2.initWithValue(dx[i * xElementWiseStride]);
reduction = update(reduction, indexVal2, extraParams);
}
}
else {
for (Nd4jLong i = tid; i < n; i += blockDim.x * gridDim.x) {
auto offset = shape::getIndexOffset(i, xShapeInfo, n);
SummaryStatsData<X> indexVal2;
indexVal2.initWithValue(dx[offset]);
reduction = update(reduction, indexVal2, extraParams);
}
}
sPartials[threadIdx.x] = reduction;
__syncthreads();
aggregatePartials<OpType>(&sPartials, threadIdx.x, blockDim.x, extraParams);
__syncthreads();
if (gridDim.x > 1) {
__shared__ bool amLast;
unsigned int *tc = (unsigned int *)reductionBuffer;
tid = threadIdx.x;
if (threadIdx.x == 0) {
SummaryStatsData<X> *pBuffer = (SummaryStatsData<X>*) reductionBuffer;
pBuffer[blockIdx.x] = sPartials[0];
}
__syncthreads();
__threadfence();
if (tid == 0) {
unsigned int ticket = atomicInc(&tc[16384], gridDim.x);
amLast = (ticket == gridDim.x - 1);
}
__syncthreads();
if (amLast) {
tc[16384] = 0;
SummaryStatsData<X>* pBuffer = (SummaryStatsData<X>*) reductionBuffer;
Z startingVal = startingValue(dx);
SummaryStatsData<X> val;
val.initWithValue(startingVal);
val.n = 0;
sPartials[threadIdx.x] = val;
for (int i = threadIdx.x; i < gridDim.x; i += blockDim.x) {
sPartials[threadIdx.x] = update(sPartials[threadIdx.x], pBuffer[i], extraParams);
}
__syncthreads();
aggregatePartials<OpType>(&sPartials, threadIdx.x, gridDim.x, extraParams);
__syncthreads();
if (tid == 0) {
z[0] = OpType::getValue(postProcessOrNot, sPartials[0]);
}
}
}
else {
if (tid == 0) {
unsigned int *tc = (unsigned *)reductionBuffer;
tc[16384] = 0;
z[0] = z[0] = OpType::getValue(postProcessOrNot, sPartials[0]);
}
}
}
};
template <typename X, typename Y>
_CUDA_D void SummaryStatsReduce<X,Y>::transform(const int opNum, void *dx, Nd4jLong *xShapeInfo, void *extraParams, void *z, Nd4jLong *zShapeInfo, int *dimension, int dimensionLength, int postProcessOrNot, int *allocationBuffer, void *reductionBuffer, Nd4jLong *tadOnlyShapeInfo, Nd4jLong *tadOffsets) {
DISPATCH_BY_OPNUM_TT(transform, PARAMS(dx, xShapeInfo, extraParams, z, zShapeInfo, dimension, dimensionLength, postProcessOrNot, allocationBuffer, reductionBuffer, tadOnlyShapeInfo, tadOffsets), SUMMARY_STATS_OPS);
};
template <typename X, typename Z>
_CUDA_H void SummaryStatsReduce<X,Z>::execSummaryStatsReduceScalar(dim3& launchDims, cudaStream_t *stream, int opNum, void *vx, Nd4jLong *xShapeInfo, Nd4jLong *hxShapeInfo, void *vextraParams, void *vz, Nd4jLong *zShapeInfo, Nd4jLong *hzShapeInfo, Nd4jLong *tadShapeInfo, Nd4jLong *tadOffsets, bool biasCorrected, void *reductionBuffer) {
auto x = static_cast<X*>(vx);
auto extraParams = static_cast<Z*>(vextraParams);
auto z = reinterpret_cast<Z*>(vz);
auto reductionPointerA = reinterpret_cast<Z*>(reductionBuffer);
if (nd4j::Environment::getInstance()->isDebugAndVerbose())
printf("D16 opNum:[%i]\n", opNum);
summaryStatsReduceT<X,Z><<<launchDims.x,launchDims.y,launchDims.z, *stream>>>(
opNum,
x,
xShapeInfo, shape::rank(hxShapeInfo),
extraParams,
z,
zShapeInfo, shape::rank(hzShapeInfo),
nullptr,
1,
1,biasCorrected, nullptr, reductionPointerA, tadShapeInfo, tadOffsets);
// this is blocking method since method should return scalar
nd4j::DebugHelper::checkErrorCode(stream, "execSSReduceScalar(...) failed");
}
template <typename X, typename Z>
_CUDA_H void SummaryStatsReduce<X,Z>::execSummaryStatsReduce(dim3& launchDims, cudaStream_t *stream, int opNum, void *vx, Nd4jLong *xShapeInfo, Nd4jLong *hxShapeInfo, void *vextraParams, void *vz, Nd4jLong *zShapeInfo, Nd4jLong *hzShapeInfo, Nd4jLong *tadShapeInfo, Nd4jLong *tadOffsets, bool biasCorrected, void *reductionBuffer) {
auto x = static_cast<X*>(vx);
auto z = static_cast<Z*>(vz);
auto extraParams = static_cast<Z*>(vextraParams);
if (nd4j::Environment::getInstance()->isDebugAndVerbose())
printf("F17 opNum:[%i]\n", opNum);
auto reductionPointerA = reinterpret_cast<Z*>(reductionBuffer);
summaryStatsReduceT<X,Z><<<launchDims.x,launchDims.y,launchDims.z, *stream>>>(
opNum,
x,
xShapeInfo, shape::rank(hxShapeInfo),
extraParams,
z,
zShapeInfo, shape::rank(hzShapeInfo),
nullptr,
1,
1,biasCorrected, nullptr, reductionPointerA, tadShapeInfo, tadOffsets);
DEBUG_KERNEL(stream, opNum);
}
template<typename X, typename Z>
_CUDA_H void SummaryStatsReduce<X,Z>::execSummaryStatsReduce(dim3& launchDims, cudaStream_t *stream, int opNum, void *vx, Nd4jLong *xShapeInfo, Nd4jLong *hxShapeInfo, void *vextraParams, void *vz, Nd4jLong *zShapeInfo, Nd4jLong *hzShapeInfo, int *dimension, int dimensionLength, Nd4jLong *tadShapeInfo, Nd4jLong *tadOffsets, bool biasCorrected, void *reductionBuffer) {
auto x = static_cast<X*>(vx);
auto z = static_cast<Z*>(vz);
auto extraParams = static_cast<Z*>(vextraParams);
if (nd4j::Environment::getInstance()->isDebugAndVerbose())
printf("D18 opNum:[%i]\n", opNum);
summaryStatsReduceT<X, Z><<<launchDims.x,launchDims.y,launchDims.z, *stream>>>(
opNum,
x,
xShapeInfo, shape::rank(hxShapeInfo),
extraParams,
z,
zShapeInfo, shape::rank(hzShapeInfo),
dimension,
dimensionLength,
1, biasCorrected, nullptr, reinterpret_cast<Z*>(reductionBuffer), tadShapeInfo, tadOffsets);
DEBUG_KERNEL(stream, opNum);
}
[WIP] multi-device support (#80) * fix pad javadoc and @see links. (#72) Signed-off-by: Robert Altena <Rob@Ra-ai.com> * [WIP] More fixes (#73) * special tests for ConstantTadHelper/ConstantShapeHelper Signed-off-by: raver119 <raver119@gmail.com> * release methods for data buffers Signed-off-by: raver119 <raver119@gmail.com> * delete temporary buffer Java side Signed-off-by: raver119 <raver119@gmail.com> * delete temporary buffer Java side Signed-off-by: raver119 <raver119@gmail.com> * delete temporary TadPack C++/Java side (#74) Signed-off-by: raver119 <raver119@gmail.com> * Zoo model TF import test updates (#75) * argLine fix, update compression_gru comment * updated comment for xception * undid but commented argLine change * updated xlnet comment * copyright headers * - new NDArray methods like()/ulike() (#77) - fix for depthwise_conv2d_bp + special test Signed-off-by: raver119 <raver119@gmail.com> * upsampling2d fix CUDA Signed-off-by: raver119 <raver119@gmail.com> * DL4J trace logging (#79) * MLN/CG trace logging for debugging Signed-off-by: AlexDBlack <blacka101@gmail.com> * Tiny tweak Signed-off-by: AlexDBlack <blacka101@gmail.com> * strided_slice_bp shape fn leak fix Signed-off-by: raver119 <raver119@gmail.com> * SameDiff fixes and naming (#78) * remove SDVariable inplace methods * import methods * npe fix in OpVal * removed SameDiff inplace ops from tests * Naming updates, moved to centralized methods in SameDiff, should use op_#:# for everything * quick fixes * javadoc * SDVariable eval with placeholders * use regex match * better matching * initial commit Signed-off-by: raver119 <raver119@gmail.com> * initial commit Signed-off-by: raver119 <raver119@gmail.com> * fix javadoc. (#76) * fix javadoc. Signed-off-by: Robert Altena <Rob@Ra-ai.com> * replace most @see with @link s. Signed-off-by: Robert Altena <Rob@Ra-ai.com> * 4 additional tests Signed-off-by: raver119 <raver119@gmail.com> * launch context reorganization Signed-off-by: raver119 <raver119@gmail.com> * LaunchContext reorganization Signed-off-by: raver119 <raver119@gmail.com> * per-device LaunchContext Signed-off-by: raver119 <raver119@gmail.com> * Various DL4J/ND4J fixes (#81) * #7954 Force refresh of UI when switching tabs on overview page Signed-off-by: AlexDBlack <blacka101@gmail.com> * #8017 Concurrent modification exception (synchronize) fix Signed-off-by: AlexDBlack <blacka101@gmail.com> * #8033 Don't initialize updater in middle of writing memory crash dump Signed-off-by: AlexDBlack <blacka101@gmail.com> * #8208 Fix shape checks for ND4J int[] creator methods Signed-off-by: AlexDBlack <blacka101@gmail.com> * #6385 #7992 Keras import naming fixes + cleanup Signed-off-by: AlexDBlack <blacka101@gmail.com> * #8016 Upsampling3D - add NDHWC format support Signed-off-by: AlexDBlack <blacka101@gmail.com> * ContextBuffers as separate entity Signed-off-by: raver119 <raver119@gmail.com> * Refactor NativeOps.h to export C functions * Actually export functions from NativeOps.h * Adapt the Java wrappers in ND4J generated with JavaCPP * Create C wrappers for some of the C++ classes currently used by ND4J * ContextBuffers as separate entity Signed-off-by: raver119 <raver119@gmail.com> * remove duplicate code in createBufferDetached. (#83) Signed-off-by: Robert Altena <Rob@Ra-ai.com> * Keras model import - updater lr fix (#84) * Keras model import - updater lr fix Signed-off-by: eraly <susan.eraly@gmail.com> * Keras model import - updater lr fix, cleanup Signed-off-by: eraly <susan.eraly@gmail.com> * ContextBuffers as separate entity Signed-off-by: raver119 <raver119@gmail.com> * ContextBuffers as separate entity Signed-off-by: raver119 <raver119@gmail.com> * Fix functions of OpaqueVariablesSet * thread-local buffers/affinity Signed-off-by: raver119 <raver119@gmail.com> * thread safety for LaunchContext Signed-off-by: raver119 <raver119@gmail.com> * more of thread safety Signed-off-by: raver119 <raver119@gmail.com> * one more multi threaded test Signed-off-by: raver119 <raver119@gmail.com> * SameDiff Convolution Config validation, better output methods (#82) * Conv Config validation & tests Signed-off-by: Ryan Nett <rnett@skymind.io> * stackOutputs utility method Signed-off-by: Ryan Nett <rnett@skymind.io> * use constructor for validation, support negative kernel sizes (infered from weights) Signed-off-by: Ryan Nett <rnett@skymind.io> * better output methods Signed-off-by: Ryan Nett <rnett@skymind.io> * move output to be with fit and evaluate Signed-off-by: Ryan Nett <rnett@skymind.io> * fixes Signed-off-by: Ryan Nett <rnett@skymind.io> * more fixes Signed-off-by: Ryan Nett <rnett@skymind.io> * refactor duplicate code from pad methods. (#86) * refactor duplicate code from pad methods. Signed-off-by: Robert Altena <Rob@Ra-ai.com> * replace switch with if. Signed-off-by: Robert Altena <Rob@Ra-ai.com> * Various ND4J/DL4J fixes and improvements (#87) * Reshape and reallocate - small fixes Signed-off-by: AlexDBlack <blacka101@gmail.com> * Reshape and reallocate - small fixes Signed-off-by: AlexDBlack <blacka101@gmail.com> * #6488 ElementWiseVertex broadcast support Signed-off-by: AlexDBlack <blacka101@gmail.com> * Constructors and broadcast supported it Transforms.max/min Signed-off-by: AlexDBlack <blacka101@gmail.com> * #8054 ElementWiseVertex now supports broadcast inputs Signed-off-by: AlexDBlack <blacka101@gmail.com> * #8057 Nd4j.create overload dtype fix Signed-off-by: AlexDBlack <blacka101@gmail.com> * #7551 ND4J Shape validation fix Signed-off-by: AlexDBlack <blacka101@gmail.com> * [WIP] Numpy boolean import (#91) * numpy bool type Signed-off-by: raver119 <raver119@gmail.com> * numpy bool java side Signed-off-by: raver119 <raver119@gmail.com> * remove create method with unused parameter. (#89) * remove create method with unused parameter. * removed more unused methods. Signed-off-by: Robert Altena <Rob@Ra-ai.com> * removing more unused code. Signed-off-by: Robert Altena <Rob@Ra-ai.com> * last removal of unused code. Signed-off-by: Robert Altena <Rob@Ra-ai.com> * remove createSparse methods. (#92) Signed-off-by: Robert Altena <Rob@Ra-ai.com> * Various ND4J/DL4J fixes (#90) * Deprecate Old*Op instances Signed-off-by: AlexDBlack <blacka101@gmail.com> * #8063 #8054 Broadcast exceptions + cleanup inplace ops Signed-off-by: AlexDBlack <blacka101@gmail.com> * Small fix Signed-off-by: AlexDBlack <blacka101@gmail.com> * Remove bad test condition Signed-off-by: AlexDBlack <blacka101@gmail.com> * #7993 Fix shape function issue in crop_and_resize op Signed-off-by: AlexDBlack <blacka101@gmail.com> * DL4J SameDiff lambda layer fix Signed-off-by: AlexDBlack <blacka101@gmail.com> * #8029 Fix for pnorm backprop math Signed-off-by: AlexDBlack <blacka101@gmail.com> * #8038 Fix Op profiler NaN/Inf triggering + add tests (#93) Signed-off-by: AlexDBlack <blacka101@gmail.com> * createUninitializedDetached refactoring. (#94) * wip * update interface, add null implementations. * Breaking one test in a weird way. Signed-off-by: Robert Altena <Rob@Ra-ai.com> * createUninitializedDetached refactored. Signed-off-by: Robert Altena <Rob@Ra-ai.com> * cuda build fix for issues introduced by recent refactoring Signed-off-by: raver119 <raver119@gmail.com> * [WIP] More of CUDA (#95) * initial commit Signed-off-by: raver119 <raver119@gmail.com> * Implementation of hashcode cuda helper. Working edition. * Fixed parallel test input arangements. * Fixed tests for hashcode op. * Fixed shape calculation for image:crop_and_resize op and test. * NativeOps tests. Initial test suite. * Added tests for indexReduce methods. * Added test on execBroadcast with NDArray as dimensions. * Added test on execBroadcastBool with NDArray as dimensions. * Added tests on execPairwiseTransform and execPairwiseTransofrmBool. * Added tests for execReduce with scalar results. * Added reduce tests for non-empty dims array. * Added tests for reduce3. * Added tests for execScalar. * Added tests for execSummaryStats. * - provide cpu/cuda code for batch_to_space - testing it Signed-off-by: Yurii <yurii@skymind.io> * - remove old test for batch_to_space (had wrong format and numbers were not checked) Signed-off-by: Yurii <yurii@skymind.io> * Fixed complilation errors with test. * Added test for execTransformFloat. * Added test for execTransformSame. * Added test for execTransformBool. * Added test for execTransformStrict. * Added tests for execScalar/execScalarBool with TADs. * Added test for flatten. * - provide cpu/cuda code for space_to_Batch operaion Signed-off-by: Yurii <yurii@skymind.io> * Added test for concat. * comment unnecessary stuff in s_t_b Signed-off-by: Yurii <yurii@skymind.io> * Added test for specialConcat. * Added tests for memcpy/set routines. * Fixed pullRow cuda test. * Added pullRow test. * Added average test. * - correct typo in NDArray::applyPairwiseTransform(nd4j::pairwise::BoolOps op...) Signed-off-by: Yurii <yurii@skymind.io> * - debugging and fixing cuda tests in JavaInteropTests file Signed-off-by: Yurii <yurii@skymind.io> * - correct some tests Signed-off-by: Yurii <yurii@skymind.io> * Added test for shuffle. * Fixed ops declarations. * Restored omp and added shuffle test. * Added convertTypes test. * Added tests for execRandom. Eliminated usage of RandomBuffer with NativeOps. * Added sort tests. * Added tests for execCustomOp. * - further debuging and fixing tests terminated with crash Signed-off-by: Yurii <yurii@skymind.io> * Added tests for calculateOutputShapes. * Addded Benchmarks test. * Commented benchmark tests. * change assertion Signed-off-by: raver119 <raver119@gmail.com> * Added tests for apply_sgd op. Added cpu helper for that op. * Implement cuda helper for aplly_sgd op. Fixed tests for NativeOps. * Added test for assign broadcastable. * Added tests for assign_bp op. * Added tests for axpy op. * - assign/execScalar/execTransformAny signature change - minor test fix Signed-off-by: raver119 <raver119@gmail.com> * Fixed axpy op. * meh Signed-off-by: raver119 <raver119@gmail.com> * - fix tests for nativeOps::concat Signed-off-by: Yurii <yurii@skymind.io> * sequential transform/scalar Signed-off-by: raver119 <raver119@gmail.com> * allow nested parallelism Signed-off-by: raver119 <raver119@gmail.com> * assign_bp leak fix Signed-off-by: raver119 <raver119@gmail.com> * block setRNG fix Signed-off-by: raver119 <raver119@gmail.com> * enable parallelism by default Signed-off-by: raver119 <raver119@gmail.com> * enable nested parallelism by default Signed-off-by: raver119 <raver119@gmail.com> * Added cuda implementation for row_count helper. * Added implementation for tnse gains op helper. * - take into account possible situations when input arrays are empty in reduce_ cuda stuff Signed-off-by: Yurii <yurii@skymind.io> * Implemented tsne/edge_forces op cuda-based helper. Parallelized cpu-based helper for edge_forces. * Added kernel for tsne/symmetrized op heleper. * Implementation of tsne/symmetrized op cuda helper. Working edition. * Eliminated waste printfs. * Added test for broadcastgradientargs op. * host-only fallback for empty reduce float Signed-off-by: raver119 <raver119@gmail.com> * - some tests fixes Signed-off-by: Yurii <yurii@skymind.io> * - correct the rest of reduce_ stuff Signed-off-by: Yurii <yurii@skymind.io> * - further correction of reduce_ stuff Signed-off-by: Yurii <yurii@skymind.io> * Added test for Cbow op. Also added cuda implementation for cbow helpers. * - improve code of stack operation for scalar case Signed-off-by: Yurii <yurii@skymind.io> * - provide cuda kernel for gatherND operation Signed-off-by: Yurii <yurii@skymind.io> * Implementation of cbow helpers with cuda kernels. * minor tests tweaks Signed-off-by: raver119 <raver119@gmail.com> * minor tests tweaks Signed-off-by: raver119 <raver119@gmail.com> * - further correction of cuda stuff Signed-off-by: Yurii <yurii@skymind.io> * Implementatation of cbow op helper with cuda kernels. Working edition. * Skip random testing for cudablas case. * lstmBlockCell context fix Signed-off-by: raver119 <raver119@gmail.com> * Added tests for ELU and ELU_BP ops. * Added tests for eq_scalar, gt_scalar, gte_scalar and lte_scalar ops. * Added tests for neq_scalar. * Added test for noop. * - further work on clipbynorm_bp Signed-off-by: Yurii <yurii@skymind.io> * - get rid of concat op call, use instead direct concat helper call Signed-off-by: Yurii <yurii@skymind.io> * lstmBlockCell context fix Signed-off-by: raver119 <raver119@gmail.com> * Added tests for lrelu and lrelu_bp. * Added tests for selu and selu_bp. * Fixed lrelu derivative helpers. * - some corrections in lstm Signed-off-by: Yurii <yurii@skymind.io> * operator * result shape fix Signed-off-by: raver119 <raver119@gmail.com> * - correct typo in lstmCell Signed-off-by: Yurii <yurii@skymind.io> * few tests fixed Signed-off-by: raver119 <raver119@gmail.com> * CUDA inverse broadcast bool fix Signed-off-by: raver119 <raver119@gmail.com> * disable MMAP test for CUDA Signed-off-by: raver119 <raver119@gmail.com> * BooleanOp syncToDevice Signed-off-by: raver119 <raver119@gmail.com> * meh Signed-off-by: raver119 <raver119@gmail.com> * additional data types for im2col/col2im Signed-off-by: raver119 <raver119@gmail.com> * Added test for firas_sparse op. * one more RandomBuffer test excluded Signed-off-by: raver119 <raver119@gmail.com> * Added tests for flatten op. * Added test for Floor op. * bunch of tests fixed Signed-off-by: raver119 <raver119@gmail.com> * mmulDot tests fixed Signed-off-by: raver119 <raver119@gmail.com> * more tests fixed Signed-off-by: raver119 <raver119@gmail.com> * Implemented floordiv_bp op and tests. * Fixed scalar case with cuda implementation for bds. * - work on cuda kernel for clip_by_norm backprop op is completed Signed-off-by: Yurii <yurii@skymind.io> * Eliminate cbow crach. * more tests fixed Signed-off-by: raver119 <raver119@gmail.com> * more tests fixed Signed-off-by: raver119 <raver119@gmail.com> * Eliminated abortion with batched nlp test. * more tests fixed Signed-off-by: raver119 <raver119@gmail.com> * Fixed shared flag initializing. * disabled bunch of cpu workspaces tests Signed-off-by: raver119 <raver119@gmail.com> * scalar operators fix: missing registerSpecialUse call Signed-off-by: raver119 <raver119@gmail.com> * Fixed logdet for cuda and tests. * - correct clipBynorm_bp Signed-off-by: Yurii <yurii@skymind.io> * Fixed crop_and_resize shape datatype. * - correct some mmul tests Signed-off-by: Yurii <yurii@skymind.io> * build fix Signed-off-by: raver119 <raver119@gmail.com> * exclude two methods for JNI Signed-off-by: raver119 <raver119@gmail.com> * exclude two methods for JNI Signed-off-by: raver119 <raver119@gmail.com> * exclude two methods for JNI (#97) Signed-off-by: raver119 <raver119@gmail.com> * temporary stack fix Signed-off-by: raver119 <raver119@gmail.com> * round robin affinity test Signed-off-by: raver119 <raver119@gmail.com> * get rid of legacy CudaContext methods Signed-off-by: raver119 <raver119@gmail.com> * get rid of legacy ContextPool classes/methods Signed-off-by: raver119 <raver119@gmail.com> * one legacy test removed Signed-off-by: raver119 <raver119@gmail.com> * few more fields rearranged Signed-off-by: raver119 <raver119@gmail.com> * OpaqueLaunchContext Signed-off-by: raver119 <raver119@gmail.com> * OpaqueLaunchContext++ Signed-off-by: raver119 <raver119@gmail.com> * more of OpaqueLaunchContext methods Signed-off-by: raver119 <raver119@gmail.com> * LaunchContext -> CudaContext Signed-off-by: raver119 <raver119@gmail.com> * AffinityManger changes Signed-off-by: raver119 <raver119@gmail.com> * AffinityManger changes Signed-off-by: raver119 <raver119@gmail.com> * cusolver handles Signed-off-by: raver119 <raver119@gmail.com> * typo Signed-off-by: raver119 <raver119@gmail.com> * cusolver method Signed-off-by: raver119 <raver119@gmail.com> * cusolver handle propagated Signed-off-by: raver119 <raver119@gmail.com> * blas/solver handles Signed-off-by: raver119 <raver119@gmail.com> * one more test Signed-off-by: raver119 <raver119@gmail.com> * legacy concat implementations replaced with new CustomOp Signed-off-by: raver119 <raver119@gmail.com> * one more test Signed-off-by: raver119 <raver119@gmail.com> * concat now uses way more blocks Signed-off-by: raver119 <raver119@gmail.com> * print Signed-off-by: raver119 <raver119@gmail.com> * no more triple template mmul Signed-off-by: raver119 <raver119@gmail.com> * bunch of kernels have dtypes reconsidered Signed-off-by: raver119 <raver119@gmail.com> * bunch of kernels have dtypes reconsidered Signed-off-by: raver119 <raver119@gmail.com> * bitonic sort reorganized Signed-off-by: raver119 <raver119@gmail.com> * bunch of cpu stuff removed from cuda scope Signed-off-by: raver119 <raver119@gmail.com> * bunch of cpu stuff removed from cuda scope Signed-off-by: raver119 <raver119@gmail.com> * type conversions moved to generic impl Signed-off-by: raver119 <raver119@gmail.com> * cpu data types pass Signed-off-by: raver119 <raver119@gmail.com> * non_max_suppression Signed-off-by: raver119 <raver119@gmail.com> * sortByValue fix Signed-off-by: raver119 <raver119@gmail.com> * ignore all mixed datatype tests for mmul Signed-off-by: raver119 <raver119@gmail.com> * special handling of OpProfiler exceptions Signed-off-by: raver119 <raver119@gmail.com> * - one failing concat test in cpp - Nd4j.tile now uses op internally Signed-off-by: raver119 <raver119@gmail.com> * get back dtype exception for legacy arrays deserialization Signed-off-by: raver119 <raver119@gmail.com>
2019-08-14 15:52:34 +02:00
template <typename X, typename Y>
Y SummaryStatsReduce<X,Y>::execScalar(int opNum,
bool biasCorrected,
void *x,
Nd4jLong *xShapeInfo,
void *extraParams) {
return 0;
}
template <typename X, typename Y>
void SummaryStatsReduce<X,Y>::execScalar(int opNum,
bool biasCorrected,
void *x,
Nd4jLong *xShapeInfo,
void *extraParams,
void *vz,
Nd4jLong *resultShapeInfoBuffer) {
}
template <typename X, typename Y>
void SummaryStatsReduce<X,Y>::exec(int opNum,
bool biasCorrected,
void *x,
Nd4jLong *xShapeInfo,
void *extraParams,
void *vz,
Nd4jLong *resultShapeInfoBuffer,
int *dimension, int dimensionLength) {
}
template <typename X, typename Y>
template<typename OpType>
Y SummaryStatsReduce<X,Y>::execScalar(bool biasCorrected,
void *x,
Nd4jLong *xShapeInfo,
void *extraParams) {
return 0;
}
template <typename X, typename Y>
template<typename OpType>
void SummaryStatsReduce<X,Y>::execScalar(bool biasCorrected,
void *x,
Nd4jLong *xShapeInfo,
void *extraParams,
void *vz,
Nd4jLong *resultShapeInfoBuffer) {
//
}
template <typename X, typename Y>
template<typename OpType>
void SummaryStatsReduce<X,Y>::exec(bool biasCorrected,
void *x,
Nd4jLong *xShapeInfo,
void *extraParams,
void *vz,
Nd4jLong *resultShapeInfoBuffer,
int *dimension,
int dimensionLength) {
}
2019-06-06 14:21:15 +02:00
BUILD_DOUBLE_TEMPLATE(template class ND4J_EXPORT SummaryStatsReduce, , LIBND4J_TYPES, FLOAT_TYPES);
}
}