328 lines
15 KiB
Plaintext
328 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
|
|
// @author Yurii Shyrma (iuriish@yahoo.com)
|
|
//
|
|
|
|
#include <system/op_boilerplate.h>
|
|
#include <loops/reduce_same.h>
|
|
#include <loops/legacy_ops.h>
|
|
#include <helpers/DebugHelper.h>
|
|
#include <types/types.h>
|
|
#include <execution/LaunchContext.h>
|
|
#include <exceptions/cuda_exception.h>
|
|
#include <loops/scalar.h>
|
|
|
|
|
|
using namespace simdOps;
|
|
|
|
|
|
////////////////////////////////////////////////////////////////////////
|
|
template <typename X, typename OpType>
|
|
__global__ void simpleReduce(void const* x, Nd4jLong const* xShapeInfo,
|
|
void *extraParams,
|
|
void *z, Nd4jLong const* zShapeInfo,
|
|
int *dimension, int dimensionLength,
|
|
void *reductionBuffer,
|
|
Nd4jLong const* tadOnlyShapeInfo, Nd4jLong const* tadOffsets) {
|
|
|
|
functions::reduce::ReduceSameFunction<X>::template transformCudaXD<OpType>(x, xShapeInfo, extraParams, z, zShapeInfo, dimension, dimensionLength, reductionBuffer, tadOnlyShapeInfo, tadOffsets);
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////
|
|
template <typename X, typename OpType>
|
|
__global__ void simpleScalar(void const* x, Nd4jLong const* xShapeInfo,
|
|
void *extraParams,
|
|
void *z, Nd4jLong const* zShapeInfo,
|
|
int *dimension, int dimensionLength,
|
|
void *reductionBuffer, Nd4jLong const* tadOnlyShapeInfo) {
|
|
|
|
functions::reduce::ReduceSameFunction<X>::template execScalarCuda<OpType>(x, xShapeInfo, extraParams, z, zShapeInfo, reductionBuffer, tadOnlyShapeInfo);
|
|
}
|
|
|
|
|
|
namespace functions {
|
|
namespace reduce {
|
|
|
|
////////////////////////////////////////////////////////////////////////
|
|
template <typename X>
|
|
template <typename OpType>
|
|
__device__ void ReduceSameFunction<X>::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 = static_cast<X*>(vsPartials);
|
|
auto extraParams = static_cast<X*>(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>
|
|
template <typename OpType>
|
|
__device__ void ReduceSameFunction<X>::transformCudaXD( void const* vx, Nd4jLong const* xShapeInfo,
|
|
void *vextraParams,
|
|
void *vz, Nd4jLong const* zShapeInfo,
|
|
int *dimension, int dimensionLength,
|
|
void *vreductionBuffer,
|
|
Nd4jLong const* tadOnlyShapeInfo, Nd4jLong const* tadOffsets) {
|
|
|
|
auto x = reinterpret_cast<X const*>(vx);
|
|
auto z = reinterpret_cast<X*>(vz);
|
|
auto extraParams = reinterpret_cast<X*>(vextraParams);
|
|
auto reductionBuffer = reinterpret_cast<X*>(vreductionBuffer);
|
|
|
|
if (OpType::requiresSpecialAccumulation) {
|
|
OpType::execSpecialCuda(x, xShapeInfo, extraParams, z, zShapeInfo, dimension, dimensionLength, reductionBuffer, tadOnlyShapeInfo, tadOffsets);
|
|
return;
|
|
}
|
|
|
|
//shared memory space for storing intermediate results
|
|
__shared__ X* sPartials;
|
|
|
|
__shared__ int tadLength, tadRank, numTads;
|
|
__shared__ Nd4jLong *tadShape, *tadStride;
|
|
__shared__ bool isPlainOutput;
|
|
|
|
if (threadIdx.x == 0) {
|
|
extern __shared__ unsigned char shmem[];
|
|
sPartials = reinterpret_cast<X*>(shmem);
|
|
|
|
isPlainOutput = shape::order(zShapeInfo) == 'c' && shape::elementWiseStride(zShapeInfo) == 1;
|
|
|
|
tadLength = shape::length(tadOnlyShapeInfo);
|
|
tadRank = shape::rank(tadOnlyShapeInfo);
|
|
numTads = shape::length(xShapeInfo) / tadLength;
|
|
tadShape = shape::shapeOf(tadOnlyShapeInfo);
|
|
tadStride = shape::stride(tadOnlyShapeInfo);
|
|
}
|
|
__syncthreads();
|
|
|
|
for (int r = blockIdx.x; r < numTads; r += gridDim.x) {
|
|
|
|
Nd4jLong tadOffsetForBlock = tadOffsets[r];
|
|
sPartials[threadIdx.x] = OpType::startingValue(x + tadOffsetForBlock);
|
|
|
|
for (int i = threadIdx.x; i < tadLength; i += blockDim.x) {
|
|
auto xOffset = tadOffsetForBlock + shape::getIndexOffset(i, tadOnlyShapeInfo);
|
|
sPartials[threadIdx.x] = OpType::update(sPartials[threadIdx.x], OpType::op(x[xOffset], extraParams), extraParams);
|
|
}
|
|
__syncthreads();
|
|
|
|
// aggregate. do NOT reduce for elements > tadLength
|
|
aggregatePartials<OpType>(sPartials, threadIdx.x, sd::math::nd4j_min<int>(blockDim.x, tadLength), extraParams);
|
|
__syncthreads();
|
|
|
|
if (threadIdx.x == 0)
|
|
z[isPlainOutput ? r : shape::getIndexOffset(r, zShapeInfo)] = OpType::postProcess(sPartials[threadIdx.x], tadLength, extraParams);
|
|
}
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////
|
|
template <typename X>
|
|
__device__ void ReduceSameFunction<X>::execScalarCudaLegacy(int opNum, void const* vx, Nd4jLong const* xShapeInfo,
|
|
void *vextraParams,
|
|
void *vz, Nd4jLong const* zShapeInfo,
|
|
void *vreductionBuffer,
|
|
Nd4jLong const* tadOnlyShapeInfo) {
|
|
DISPATCH_BY_OPNUM_T(execScalarCuda, PARAMS(vx, xShapeInfo, vextraParams, vz, zShapeInfo, vreductionBuffer, tadOnlyShapeInfo), REDUCE_SAME_OPS);
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////
|
|
template <typename X>
|
|
template <typename OpType>
|
|
__device__ void ReduceSameFunction<X>::execScalarCuda(void const* vx, Nd4jLong const* xShapeInfo,
|
|
void *vextraParams,
|
|
void * vz, Nd4jLong const* zShapeInfo,
|
|
void *vreductionBuffer,
|
|
Nd4jLong const* tadOnlyShapeInfo) {
|
|
auto x = reinterpret_cast<X const*>(vx);
|
|
auto z = reinterpret_cast<X*>(vz);
|
|
auto extraParams = reinterpret_cast<X*>(vextraParams);
|
|
auto reductionBuffer = reinterpret_cast<X*>(vreductionBuffer);
|
|
|
|
auto tid = blockDim.x * blockIdx.x + threadIdx.x;
|
|
|
|
//shared memory space for storing intermediate results
|
|
__shared__ X* sPartials;
|
|
__shared__ Nd4jLong xEws;
|
|
__shared__ Nd4jLong len;
|
|
|
|
if(threadIdx.x == 0) {
|
|
extern __shared__ unsigned char shmem[];
|
|
sPartials = reinterpret_cast<X*>(shmem);
|
|
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) {
|
|
auto tc = reinterpret_cast<unsigned int *>(reductionBuffer);
|
|
tc[16384] = 0;
|
|
z[0] = OpType::postProcess(sPartials[0], len, extraParams);
|
|
}
|
|
}
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////
|
|
template <typename X>
|
|
template<typename OpType>
|
|
__host__ void ReduceSameFunction<X>::intermediateXD(dim3 launchDims, cudaStream_t *stream, void const* x, Nd4jLong const* xShapeInfo, Nd4jLong const* hXShapeInfo, void *extraParams, void *z, Nd4jLong const* zShapeInfo, Nd4jLong const* hZShapeInfo, int *dimension, int dimensionLength, void *reductionPointer, Nd4jLong const* tadShapeInfo, Nd4jLong const* tadOffsets) {
|
|
|
|
if(shape::isEmpty(hXShapeInfo)) {
|
|
|
|
if(shape::isEmpty(hZShapeInfo))
|
|
return;
|
|
|
|
const auto startingVal = static_cast<X>(OpType::startingValue(reinterpret_cast<const X*>(x)));
|
|
|
|
auto res = cudaMemcpyAsync(sd::LaunchContext::defaultContext()->getScalarPointer(), &startingVal, sizeof(X), cudaMemcpyHostToDevice, *stream);
|
|
if (res != 0)
|
|
throw sd::cuda_exception::build("ReduceSameFunction<X,Z>::intermediateXD: failed to copy temporary scalar", res);
|
|
|
|
auto ptr = sd::LaunchContext::defaultContext()->getScalarPointer();
|
|
|
|
// scalar assign
|
|
functions::scalar::ScalarTransform<X, X, X>::executeCudaShaped(launchDims, stream, 14, z, zShapeInfo, hXShapeInfo, z, zShapeInfo, hZShapeInfo, ptr, nullptr);
|
|
}
|
|
else {
|
|
simpleReduce<X, OpType><<<launchDims.x, launchDims.y, launchDims.z, *stream>>>(x, xShapeInfo, extraParams, z, zShapeInfo, dimension, dimensionLength, reductionPointer, tadShapeInfo, tadOffsets);
|
|
}
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////
|
|
template <typename X>
|
|
template<typename OpType>
|
|
__host__ void ReduceSameFunction<X>::intermediateScalar(dim3 launchDims, cudaStream_t *stream, void const* x, Nd4jLong const* xShapeInfo, Nd4jLong const* hXShapeInfo, void *extraParams, void *z, Nd4jLong const* zShapeInfo, Nd4jLong const* hZShapeInfo, int *dimension, int dimensionLength, void *reductionBuffer, Nd4jLong const* tadOnlyShapeInfo) {
|
|
|
|
if (shape::isEmpty(hXShapeInfo)) {
|
|
|
|
if (shape::isEmpty(hZShapeInfo))
|
|
return;
|
|
|
|
const auto startingVal = static_cast<X>(OpType::startingValue(reinterpret_cast<const X*>(x)));
|
|
|
|
auto res = cudaMemcpyAsync(z, &startingVal, sizeof(X), cudaMemcpyHostToDevice, *stream);
|
|
if (res != 0)
|
|
throw sd::cuda_exception::build("ReduceSameFunction<X>::intermediateScalar: failed to copy resulting scalar", res);
|
|
}
|
|
else {
|
|
simpleScalar<X, OpType><<<launchDims.x, launchDims.y, launchDims.z, *stream>>>(x, xShapeInfo, extraParams, z, zShapeInfo, dimension, dimensionLength, reductionBuffer, tadOnlyShapeInfo);
|
|
}
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////
|
|
template <typename X>
|
|
_CUDA_H void ReduceSameFunction<X>::execReduceScalar(dim3 launchDims, cudaStream_t *stream, int opNum, void const* x, Nd4jLong const* xShapeInfo, Nd4jLong const* hXShapeInfo, void *extraParams, void *z, Nd4jLong const* zShapeInfo, Nd4jLong const* hZShapeInfo, int *dimension, int dimensionLength, void *reductionBuffer, Nd4jLong const* tadOnlyShapeInfo) {
|
|
|
|
DISPATCH_BY_OPNUM_T(intermediateScalar, PARAMS(launchDims, stream, x, xShapeInfo, hXShapeInfo, extraParams, z, zShapeInfo, hZShapeInfo, dimension, dimensionLength, reductionBuffer, tadOnlyShapeInfo), REDUCE_SAME_OPS);
|
|
sd::DebugHelper::checkErrorCode(stream, "execReduceScalarSame(...) failed");
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////
|
|
template <typename X>
|
|
_CUDA_H void ReduceSameFunction<X>::execReduceXD(dim3 launchDims, cudaStream_t *stream, int opNum, int rank, void const* x, Nd4jLong const* xShapeInfo, Nd4jLong const* hXShapeInfo, void *extraParams, void *z, Nd4jLong const* zShapeInfo, Nd4jLong const* hZShapeInfo, int *dimension, int dimensionLength, void *reductionPointer, Nd4jLong const* tadShapeInfo, Nd4jLong const* tadOffsets) {
|
|
|
|
DISPATCH_BY_OPNUM_T(intermediateXD, PARAMS(launchDims, stream, x, xShapeInfo, hXShapeInfo, extraParams, z, zShapeInfo, hZShapeInfo, dimension, dimensionLength, reductionPointer, tadShapeInfo, tadOffsets), REDUCE_SAME_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_SINGLE_TEMPLATE(template class ND4J_EXPORT ReduceSameFunction, , LIBND4J_TYPES);
|
|
|
|
}
|
|
} |