335 lines
15 KiB
Plaintext
335 lines
15 KiB
Plaintext
/*******************************************************************************
|
|
* 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 <execution/LaunchContext.h>
|
|
#include <exceptions/cuda_exception.h>
|
|
#include <system/op_boilerplate.h>
|
|
#include <loops/reduce_float.h>
|
|
#include <loops/scalar.h>
|
|
#include <loops/legacy_ops.h>
|
|
#include <helpers/DebugHelper.h>
|
|
#include <types/types.h>
|
|
#include <ops/specials_cuda.h>
|
|
|
|
#include <cuda.h>
|
|
#include <cuda_runtime.h>
|
|
|
|
using namespace simdOps;
|
|
|
|
////////////////////////////////////////////////////////////////////////
|
|
template <typename X, typename Z, typename OpType>
|
|
__global__ void simpleReduce(const void *x, const Nd4jLong *outerXTadShapeInfo, const Nd4jLong *innerXTadShapeInfo,
|
|
void *extraParams, void *vreductionBuffer, void *z, const Nd4jLong *zShapeInfo) {
|
|
|
|
functions::reduce::ReduceFloatFunction<X,Z>::template transformCudaXD<OpType>(x, outerXTadShapeInfo, innerXTadShapeInfo, extraParams, vreductionBuffer, z, zShapeInfo);
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////
|
|
template <typename X, typename Z, typename OpType>
|
|
__global__ void simpleScalar(const void *x, const Nd4jLong *xShapeInfo,
|
|
void *extraParams,
|
|
void *z, const Nd4jLong *zShapeInfo,
|
|
int *dimension, int dimensionLength,
|
|
void *reductionBuffer,
|
|
const Nd4jLong *tadOnlyShapeInfo) {
|
|
|
|
functions::reduce::ReduceFloatFunction<X, Z>::template execScalarCuda<OpType>(x, xShapeInfo, extraParams, z, zShapeInfo, reductionBuffer, tadOnlyShapeInfo);
|
|
}
|
|
|
|
namespace functions {
|
|
namespace reduce {
|
|
|
|
////////////////////////////////////////////////////////////////////////
|
|
template <typename X, typename Z>
|
|
template <typename OpType>
|
|
__device__ void ReduceFloatFunction<X,Z>::aggregatePartials(void *vsPartials, Nd4jLong tid, Nd4jLong numItems, 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 sPartials = reinterpret_cast<Z*>(vsPartials);
|
|
auto extraParams = reinterpret_cast<Z*>(vextraParams);
|
|
|
|
Nd4jLong floorPow2 = numItems;
|
|
|
|
if (floorPow2 & (floorPow2 - 1)) {
|
|
|
|
while (floorPow2 & (floorPow2 - 1))
|
|
floorPow2 &= floorPow2 - 1;
|
|
|
|
if (tid >= floorPow2)
|
|
sPartials[tid - floorPow2] = OpType::update(sPartials[tid - floorPow2], sPartials[tid], extraParams);
|
|
|
|
__syncthreads();
|
|
}
|
|
|
|
for (Nd4jLong activeThreads = floorPow2 >> 1; activeThreads; activeThreads >>= 1) {
|
|
if (tid < activeThreads && tid + activeThreads < numItems)
|
|
sPartials[tid] = OpType::update(sPartials[tid], sPartials[tid + activeThreads], extraParams);
|
|
|
|
__syncthreads();
|
|
}
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////
|
|
template <typename X, typename Z>
|
|
template <typename OpType>
|
|
__device__ void ReduceFloatFunction<X,Z>::transformCudaXD(const void *vx, const Nd4jLong *outerXTadShapeInfo, const Nd4jLong *innerXTadShapeInfo,
|
|
void *vextraParams, void *vreductionBuffer,
|
|
void *vz, const Nd4jLong *zShapeInfo) {
|
|
|
|
auto x = reinterpret_cast<const X*>(vx);
|
|
auto z = reinterpret_cast<Z*>(vz);
|
|
auto extraParams = reinterpret_cast<Z*>(vextraParams);
|
|
|
|
//shared memory space for storing intermediate results
|
|
__shared__ Z sPartials[CUDA_BLOCK_SIZE];
|
|
__shared__ int tadLen, numTads;
|
|
__shared__ bool sameOffsets;
|
|
|
|
if (threadIdx.x == 0) {
|
|
sameOffsets = shape::haveSameShapeAndStrides(zShapeInfo, outerXTadShapeInfo);
|
|
|
|
tadLen = shape::length(innerXTadShapeInfo);
|
|
numTads = shape::length(outerXTadShapeInfo);
|
|
}
|
|
__syncthreads();
|
|
|
|
int coords[MAX_RANK];
|
|
|
|
for (int r = blockIdx.x; r < numTads; r += gridDim.x) {
|
|
|
|
shape::index2coords(r, outerXTadShapeInfo, coords);
|
|
const auto outerOffset = shape::getOffset(outerXTadShapeInfo, coords);
|
|
const auto zOffset = sameOffsets ? outerOffset : shape::getOffset(zShapeInfo, coords);
|
|
|
|
const X* xTad = x + outerOffset;
|
|
sPartials[threadIdx.x] = OpType::startingValue(xTad);
|
|
|
|
for (int i = threadIdx.x; i < tadLen; i += blockDim.x)
|
|
sPartials[threadIdx.x] = OpType::update(sPartials[threadIdx.x], OpType::op(xTad[shape::getIndexOffset(i, innerXTadShapeInfo)], extraParams), extraParams);
|
|
__syncthreads();
|
|
|
|
// aggregate. do NOT reduce for elements > tadLen
|
|
aggregatePartials<OpType>(sPartials, threadIdx.x, sd::math::nd4j_min<int>(blockDim.x, tadLen), extraParams);
|
|
__syncthreads();
|
|
|
|
if (threadIdx.x == 0)
|
|
z[zOffset] = OpType::postProcess(sPartials[threadIdx.x], tadLen, extraParams);
|
|
}
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////
|
|
template <typename X, typename Z>
|
|
template <typename OpType>
|
|
__device__ void ReduceFloatFunction<X,Z>::execScalarCuda(const void *vx, const Nd4jLong *xShapeInfo,
|
|
void *vextraParams,
|
|
void *vz, const Nd4jLong *zShapeInfo,
|
|
void *vreductionBuffer,
|
|
const Nd4jLong *tadOnlyShapeInfo) {
|
|
|
|
auto x = reinterpret_cast<const X*>(vx);
|
|
auto z = reinterpret_cast<Z*>(vz);
|
|
auto extraParams = reinterpret_cast<Z*>(vextraParams);
|
|
auto reductionBuffer = reinterpret_cast<Z*>(vreductionBuffer);
|
|
|
|
auto tid = blockDim.x * blockIdx.x + threadIdx.x;
|
|
|
|
//shared memory space for storing intermediate results
|
|
__shared__ Z sPartials[CUDA_BLOCK_SIZE];
|
|
__shared__ Nd4jLong xEws;
|
|
__shared__ Nd4jLong len;
|
|
|
|
if(threadIdx.x == 0) {
|
|
xEws = shape::elementWiseStride(xShapeInfo);
|
|
len = shape::length(xShapeInfo);
|
|
}
|
|
__syncthreads();
|
|
|
|
sPartials[threadIdx.x] = OpType::startingValue(x);
|
|
|
|
if (xEws > 0)
|
|
for (int i = tid; i < len; i += (blockDim.x * gridDim.x))
|
|
sPartials[threadIdx.x] = OpType::update(sPartials[threadIdx.x], OpType::op(x[i * xEws], extraParams), extraParams);
|
|
else
|
|
for (int i = tid; i < len; i += blockDim.x * gridDim.x)
|
|
sPartials[threadIdx.x] = OpType::update(sPartials[threadIdx.x], OpType::op(x[shape::getIndexOffset(i, xShapeInfo)], extraParams), extraParams);
|
|
|
|
__syncthreads();
|
|
aggregatePartials<OpType>(sPartials, threadIdx.x, sd::math::nd4j_min<int>(blockDim.x, len), extraParams);
|
|
__syncthreads();
|
|
|
|
if (gridDim.x > 1) {
|
|
|
|
unsigned int *tc = (unsigned int *)reductionBuffer;
|
|
__shared__ bool amLast;
|
|
|
|
tid = threadIdx.x;
|
|
if (threadIdx.x == 0)
|
|
reductionBuffer[blockIdx.x] = sPartials[0];//this->postProcess(sPartials[0],len,extraParams);
|
|
|
|
__threadfence();
|
|
__syncthreads();
|
|
|
|
if (threadIdx.x == 0) {
|
|
unsigned int ticket = atomicInc(&tc[16384], gridDim.x);
|
|
amLast = (ticket == gridDim.x - 1);
|
|
}
|
|
|
|
__syncthreads();
|
|
|
|
if (amLast) {
|
|
|
|
tc[16384] = 0;
|
|
sPartials[threadIdx.x] = OpType::startingValue(x);
|
|
|
|
for (int i = threadIdx.x; i < gridDim.x; i += blockDim.x)
|
|
sPartials[threadIdx.x] = OpType::update(sPartials[threadIdx.x], reductionBuffer[i], extraParams);
|
|
|
|
__syncthreads();
|
|
aggregatePartials<OpType>(sPartials, threadIdx.x, sd::math::nd4j_min<int>(gridDim.x, blockDim.x), extraParams);
|
|
__syncthreads();
|
|
|
|
if (threadIdx.x == 0) {
|
|
z[0] = OpType::postProcess(sPartials[0], len, extraParams);
|
|
}
|
|
}
|
|
}
|
|
else {
|
|
|
|
if (threadIdx.x == 0) {
|
|
unsigned int *tc = (unsigned *)reductionBuffer;
|
|
tc[16384] = 0;
|
|
z[0] = OpType::postProcess(sPartials[0], len, extraParams);
|
|
}
|
|
}
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////
|
|
template <typename X, typename Z>
|
|
template<typename OpType>
|
|
__host__ void ReduceFloatFunction<X,Z>::intermediateXD(dim3 launchDims, cudaStream_t *stream,
|
|
const void *x, const Nd4jLong *dXShapeInfo, const Nd4jLong *hXShapeInfo,
|
|
void *extraParams, void *vreductionBuffer,
|
|
void *z, const Nd4jLong *dZShapeInfo, const Nd4jLong *hZShapeInfo, const int* dims) {
|
|
|
|
if(shape::isEmpty(hXShapeInfo)) {
|
|
|
|
if(shape::isEmpty(hZShapeInfo))
|
|
return;
|
|
|
|
const auto startingVal = std::is_same<OpType, simdOps::Mean<X,Z>>::value ? sd::DataTypeUtils::nanOrZero<Z>() : static_cast<Z>(OpType::startingValue(reinterpret_cast<const X*>(x)));
|
|
auto res = cudaMemcpyAsync(sd::LaunchContext::defaultContext()->getScalarPointer(), &startingVal, sizeof(Z), cudaMemcpyHostToDevice, *stream);
|
|
if (res != 0)
|
|
throw sd::cuda_exception::build("ReduceFloatFunction<X,Z>::intermediateXD: failed to copy temporary scalar", res);
|
|
|
|
auto ptr = sd::LaunchContext::defaultContext()->getScalarPointer();
|
|
|
|
// scalar assign
|
|
functions::scalar::ScalarTransform<Z, Z, Z>::executeCudaShaped(launchDims, stream, 14, z, dZShapeInfo, hZShapeInfo, z, dZShapeInfo, hZShapeInfo, ptr, nullptr);
|
|
}
|
|
else {
|
|
|
|
const int zRank = shape::rank(hZShapeInfo);
|
|
const int tadRank = shape::rank(hXShapeInfo) - zRank;
|
|
|
|
auto outerPack = sd::ConstantShapeHelper::getInstance().createSubArrShapeInfo(hXShapeInfo, dims, zRank);
|
|
auto innerPack = sd::ConstantShapeHelper::getInstance().createSubArrShapeInfo(hXShapeInfo, dims+zRank, tadRank);
|
|
|
|
simpleReduce<X, Z, OpType><<<launchDims.x, launchDims.y, launchDims.z, *stream>>>(x, reinterpret_cast<Nd4jLong const*>(outerPack.special()), reinterpret_cast<Nd4jLong const*>(innerPack.special()), extraParams, vreductionBuffer, z, dZShapeInfo);
|
|
}
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////
|
|
template <typename X, typename Z>
|
|
template<typename OpType>
|
|
__host__ void ReduceFloatFunction<X,Z>::intermediateScalar(dim3 launchDims, cudaStream_t *stream,
|
|
const void *x, const Nd4jLong *xShapeInfo, const Nd4jLong *hXShapeInfo,
|
|
void *extraParams,
|
|
void *z, const Nd4jLong *dZShapeInfo, const Nd4jLong *hZShapeInfo,
|
|
int *dimension, int dimensionLength,
|
|
void *reductionBuffer,
|
|
const Nd4jLong *tadOnlyShapeInfo) {
|
|
|
|
if (shape::isEmpty(hXShapeInfo)) {
|
|
|
|
if (shape::isEmpty(hZShapeInfo))
|
|
return;
|
|
|
|
const auto startingVal = std::is_same<OpType, simdOps::Mean<X,Z>>::value ? sd::DataTypeUtils::nanOrZero<Z>() : static_cast<Z>(OpType::startingValue(reinterpret_cast<const X*>(x)));
|
|
|
|
auto res = cudaMemcpyAsync(z, &startingVal, sizeof(Z), cudaMemcpyHostToDevice, *stream);
|
|
if (res != 0)
|
|
throw sd::cuda_exception::build("ReduceFloatFunction<X,Z>::intermediateScalar: failed to copy resulting scalar", res);
|
|
}
|
|
else {
|
|
simpleScalar<X, Z, OpType> <<<launchDims.x, launchDims.y, launchDims.z, *stream>>>(x, xShapeInfo, extraParams, z, dZShapeInfo, dimension, dimensionLength, reductionBuffer, tadOnlyShapeInfo);
|
|
}
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////
|
|
template <typename X, typename Y>
|
|
_CUDA_H void ReduceFloatFunction<X,Y>::execReduceScalar(dim3 launchDims, cudaStream_t *stream,
|
|
const int opNum,
|
|
const void *x, const Nd4jLong *xShapeInfo, const Nd4jLong *hXShapeInfo,
|
|
void *extraParams,
|
|
void *z, const Nd4jLong *dZShapeInfo, const Nd4jLong *hZShapeInfo,
|
|
int *dimension, int dimensionLength,
|
|
void *reductionBuffer,
|
|
const Nd4jLong *tadOnlyShapeInfo) {
|
|
|
|
DISPATCH_BY_OPNUM_TT(intermediateScalar, PARAMS(launchDims, stream, x, xShapeInfo, hXShapeInfo, extraParams, z, dZShapeInfo, hZShapeInfo, dimension, dimensionLength, reductionBuffer, tadOnlyShapeInfo), OPS_A(REDUCE_FLOAT_OPS));
|
|
sd::DebugHelper::checkErrorCode(stream, "execReduceScalarFloat(...) failed");
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////
|
|
template <typename X, typename Y>
|
|
_CUDA_H void ReduceFloatFunction<X,Y>::execReduceXD(dim3 launchDims, cudaStream_t *stream, const int opNum,
|
|
const void *x, const Nd4jLong *dXShapeInfo, const Nd4jLong *hXShapeInfo,
|
|
void *extraParams, void *vreductionBuffer,
|
|
void *z, const Nd4jLong *dZShapeInfo, const Nd4jLong *hZShapeInfo, const int *dims) {
|
|
|
|
if(shape::length(hZShapeInfo) == 1) {
|
|
ReduceFloatFunction<X,Y>::execReduceScalar(launchDims, stream, opNum, x, dXShapeInfo, hXShapeInfo, extraParams, z, dZShapeInfo, hZShapeInfo, nullptr, 0, vreductionBuffer, nullptr);
|
|
}
|
|
else {
|
|
DISPATCH_BY_OPNUM_TT(intermediateXD, PARAMS(launchDims, stream, x, dXShapeInfo, hXShapeInfo, extraParams, vreductionBuffer, z, dZShapeInfo, hZShapeInfo, dims), OPS_A(REDUCE_FLOAT_OPS));
|
|
}
|
|
DEBUG_KERNEL(stream, opNum);
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////
|
|
template <typename X>
|
|
__device__ void initializeShared(X *extraParams, X **sPartials, int sMemSize) {
|
|
int sPartialsLength = sMemSize / sizeof(X);
|
|
X *sPartialsDeref = (X *) *sPartials;
|
|
for (int i = 0; i < sPartialsLength; i++)
|
|
sPartialsDeref[i] = extraParams[0];
|
|
|
|
}
|
|
|
|
|
|
//BUILD_DOUBLE_TEMPLATE(template class ND4J_EXPORT ReduceFloatFunction, , LIBND4J_TYPES, FLOAT_TYPES);
|
|
|
|
}
|
|
}
|
|
|