2021-02-09 13:16:31 +09:00
|
|
|
/*
|
|
|
|
|
* ******************************************************************************
|
|
|
|
|
* *
|
|
|
|
|
* *
|
|
|
|
|
* * 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.
|
|
|
|
|
* *
|
|
|
|
|
* * See the NOTICE file distributed with this work for additional
|
|
|
|
|
* * information regarding copyright ownership.
|
|
|
|
|
* * 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
|
|
|
|
|
* *****************************************************************************
|
|
|
|
|
*/
|
2019-06-06 15:21:15 +03:00
|
|
|
|
|
|
|
|
package org.nd4j.jita.constant;
|
|
|
|
|
|
2022-09-20 15:40:53 +02:00
|
|
|
import lombok.extern.log4j.Log4j2;
|
|
|
|
|
import lombok.extern.slf4j.Slf4j;
|
2019-06-06 15:21:15 +03:00
|
|
|
import lombok.val;
|
|
|
|
|
import org.bytedeco.javacpp.Pointer;
|
2022-09-20 15:40:53 +02:00
|
|
|
import org.nd4j.common.util.ArrayUtil;
|
2019-06-06 15:21:15 +03:00
|
|
|
import org.nd4j.jita.allocator.enums.AllocationStatus;
|
|
|
|
|
import org.nd4j.jita.allocator.impl.AllocationPoint;
|
|
|
|
|
import org.nd4j.jita.allocator.impl.AtomicAllocator;
|
|
|
|
|
import org.nd4j.jita.conf.Configuration;
|
|
|
|
|
import org.nd4j.jita.conf.CudaEnvironment;
|
|
|
|
|
import org.nd4j.jita.flow.FlowController;
|
|
|
|
|
import org.nd4j.linalg.api.buffer.DataBuffer;
|
|
|
|
|
import org.nd4j.linalg.api.buffer.DataType;
|
|
|
|
|
import org.nd4j.linalg.api.memory.AllocationsTracker;
|
2022-09-20 15:40:53 +02:00
|
|
|
import org.nd4j.linalg.api.memory.MemcpyDirection;
|
2019-06-06 15:21:15 +03:00
|
|
|
import org.nd4j.linalg.api.memory.enums.AllocationKind;
|
|
|
|
|
import org.nd4j.linalg.api.ops.performance.PerformanceTracker;
|
|
|
|
|
import org.nd4j.linalg.cache.ArrayDescriptor;
|
|
|
|
|
import org.nd4j.linalg.cache.ConstantHandler;
|
|
|
|
|
import org.nd4j.linalg.exception.ND4JIllegalStateException;
|
|
|
|
|
import org.nd4j.linalg.factory.Nd4j;
|
|
|
|
|
import org.nd4j.linalg.jcublas.buffer.*;
|
|
|
|
|
import org.nd4j.nativeblas.NativeOpsHolder;
|
|
|
|
|
|
|
|
|
|
import java.util.HashMap;
|
|
|
|
|
import java.util.Map;
|
|
|
|
|
import java.util.concurrent.ConcurrentHashMap;
|
|
|
|
|
import java.util.concurrent.Semaphore;
|
|
|
|
|
import java.util.concurrent.atomic.AtomicLong;
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Created by raver on 08.06.2016.
|
|
|
|
|
*/
|
2020-04-23 01:36:49 +03:00
|
|
|
@Slf4j
|
2019-06-06 15:21:15 +03:00
|
|
|
public class ProtectedCudaConstantHandler implements ConstantHandler {
|
2022-10-21 15:19:32 +02:00
|
|
|
private static final ProtectedCudaConstantHandler ourInstance = new ProtectedCudaConstantHandler();
|
2019-06-06 15:21:15 +03:00
|
|
|
|
|
|
|
|
protected Map<Integer, AtomicLong> constantOffsets = new HashMap<>();
|
|
|
|
|
protected Map<Integer, Semaphore> deviceLocks = new ConcurrentHashMap<>();
|
|
|
|
|
|
|
|
|
|
protected Map<Integer, Map<ArrayDescriptor, DataBuffer>> buffersCache = new HashMap<>();
|
|
|
|
|
protected Map<Integer, Pointer> deviceAddresses = new HashMap<>();
|
|
|
|
|
protected AtomicLong bytes = new AtomicLong(0);
|
|
|
|
|
protected FlowController flowController;
|
|
|
|
|
|
|
|
|
|
protected static final ConstantProtector protector = ConstantProtector.getInstance();
|
|
|
|
|
|
|
|
|
|
private static final int MAX_CONSTANT_LENGTH = 49152;
|
|
|
|
|
private static final int MAX_BUFFER_LENGTH = 272;
|
|
|
|
|
|
|
|
|
|
protected Semaphore lock = new Semaphore(1);
|
|
|
|
|
private boolean resetHappened = false;
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
public static ProtectedCudaConstantHandler getInstance() {
|
|
|
|
|
return ourInstance;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
private ProtectedCudaConstantHandler() {}
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* This method removes all cached constants
|
|
|
|
|
*/
|
|
|
|
|
@Override
|
|
|
|
|
public void purgeConstants() {
|
|
|
|
|
buffersCache = new HashMap<>();
|
|
|
|
|
|
|
|
|
|
protector.purgeProtector();
|
|
|
|
|
|
|
|
|
|
resetHappened = true;
|
2022-09-20 15:40:53 +02:00
|
|
|
log.info("Resetting Constants...");
|
2019-06-06 15:21:15 +03:00
|
|
|
|
|
|
|
|
for (Integer device : constantOffsets.keySet()) {
|
|
|
|
|
constantOffsets.get(device).set(0);
|
|
|
|
|
buffersCache.put(device, new ConcurrentHashMap<ArrayDescriptor, DataBuffer>());
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Method suited for debug purposes only
|
|
|
|
|
*
|
|
|
|
|
* @return
|
|
|
|
|
*/
|
|
|
|
|
protected int amountOfEntries(int deviceId) {
|
|
|
|
|
ensureMaps(deviceId);
|
|
|
|
|
return buffersCache.get(0).size();
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* This method moves specified dataBuffer to CUDA constant memory space.
|
|
|
|
|
*
|
|
|
|
|
* PLEASE NOTE: CUDA constant memory is limited to 48KB per device.
|
|
|
|
|
*
|
|
|
|
|
* @param dataBuffer
|
|
|
|
|
* @return
|
|
|
|
|
*/
|
|
|
|
|
@Override
|
|
|
|
|
public synchronized long moveToConstantSpace(DataBuffer dataBuffer) {
|
|
|
|
|
if (1 > 0)
|
|
|
|
|
throw new RuntimeException("This code shouldn't be called, ever");
|
|
|
|
|
|
|
|
|
|
// now, we move things to constant memory
|
|
|
|
|
Integer deviceId = AtomicAllocator.getInstance().getDeviceId();
|
|
|
|
|
ensureMaps(deviceId);
|
|
|
|
|
|
|
|
|
|
AllocationPoint point = AtomicAllocator.getInstance().getAllocationPoint(dataBuffer);
|
|
|
|
|
|
2020-01-04 13:27:50 +03:00
|
|
|
long requiredMemoryBytes = point.getNumberOfBytes();
|
2019-06-06 15:21:15 +03:00
|
|
|
val originalBytes = requiredMemoryBytes;
|
|
|
|
|
requiredMemoryBytes += 8 - (requiredMemoryBytes % 8);
|
|
|
|
|
|
|
|
|
|
val div = requiredMemoryBytes / 4;
|
|
|
|
|
if (div % 2 != 0)
|
|
|
|
|
requiredMemoryBytes += 4;
|
|
|
|
|
|
|
|
|
|
//logger.info("shape: " + point.getShape());
|
|
|
|
|
// and release device memory :)
|
|
|
|
|
|
|
|
|
|
AllocationsTracker.getInstance().markAllocated(AllocationKind.CONSTANT, deviceId, requiredMemoryBytes);
|
|
|
|
|
|
|
|
|
|
long currentOffset = constantOffsets.get(deviceId).get();
|
[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 16:52:34 +03:00
|
|
|
val context = AtomicAllocator.getInstance().getDeviceContext();
|
2019-06-06 15:21:15 +03:00
|
|
|
if (currentOffset + requiredMemoryBytes >= MAX_CONSTANT_LENGTH || requiredMemoryBytes > MAX_BUFFER_LENGTH) {
|
|
|
|
|
if (point.getAllocationStatus() == AllocationStatus.HOST
|
|
|
|
|
&& CudaEnvironment.getInstance().getConfiguration().getMemoryModel() == Configuration.MemoryModel.DELAYED) {
|
2020-01-04 13:27:50 +03:00
|
|
|
//AtomicAllocator.getInstance().getMemoryHandler().alloc(AllocationStatus.DEVICE, point, point.getShape(), false);
|
|
|
|
|
throw new UnsupportedOperationException("Pew-pew");
|
2019-06-06 15:21:15 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
val profD = PerformanceTracker.getInstance().helperStartTransaction();
|
|
|
|
|
|
2020-01-04 13:27:50 +03:00
|
|
|
if (NativeOpsHolder.getInstance().getDeviceNativeOps().memcpyAsync(point.getDevicePointer(), point.getHostPointer(), originalBytes, 1, context.getSpecialStream()) == 0) {
|
2019-06-06 15:21:15 +03:00
|
|
|
throw new ND4JIllegalStateException("memcpyAsync failed");
|
|
|
|
|
}
|
|
|
|
|
flowController.commitTransfer(context.getSpecialStream());
|
|
|
|
|
|
|
|
|
|
PerformanceTracker.getInstance().helperRegisterTransaction(point.getDeviceId(), profD, point.getNumberOfBytes(), MemcpyDirection.HOST_TO_DEVICE);
|
|
|
|
|
|
|
|
|
|
point.setConstant(true);
|
|
|
|
|
point.tickDeviceWrite();
|
|
|
|
|
point.tickHostRead();
|
|
|
|
|
point.setDeviceId(deviceId);
|
|
|
|
|
|
|
|
|
|
protector.persistDataBuffer(dataBuffer);
|
|
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
long bytes = requiredMemoryBytes;
|
|
|
|
|
currentOffset = constantOffsets.get(deviceId).getAndAdd(bytes);
|
|
|
|
|
|
|
|
|
|
if (currentOffset >= MAX_CONSTANT_LENGTH) {
|
|
|
|
|
if (point.getAllocationStatus() == AllocationStatus.HOST
|
|
|
|
|
&& CudaEnvironment.getInstance().getConfiguration().getMemoryModel() == Configuration.MemoryModel.DELAYED) {
|
2020-01-04 13:27:50 +03:00
|
|
|
//AtomicAllocator.getInstance().getMemoryHandler().alloc(AllocationStatus.DEVICE, point, point.getShape(), false);
|
|
|
|
|
throw new UnsupportedOperationException("Pew-pew");
|
2019-06-06 15:21:15 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
val profD = PerformanceTracker.getInstance().helperStartTransaction();
|
|
|
|
|
|
2020-01-04 13:27:50 +03:00
|
|
|
if (NativeOpsHolder.getInstance().getDeviceNativeOps().memcpyAsync(point.getDevicePointer(), point.getHostPointer(), originalBytes, 1, context.getSpecialStream()) == 0) {
|
2019-06-06 15:21:15 +03:00
|
|
|
throw new ND4JIllegalStateException("memcpyAsync failed");
|
|
|
|
|
}
|
|
|
|
|
flowController.commitTransfer(context.getSpecialStream());
|
|
|
|
|
|
|
|
|
|
PerformanceTracker.getInstance().helperRegisterTransaction(point.getDeviceId(), profD, point.getNumberOfBytes(), MemcpyDirection.HOST_TO_DEVICE);
|
|
|
|
|
|
|
|
|
|
point.setConstant(true);
|
|
|
|
|
point.tickDeviceWrite();
|
|
|
|
|
point.tickHostRead();
|
|
|
|
|
point.setDeviceId(deviceId);
|
|
|
|
|
|
|
|
|
|
protector.persistDataBuffer(dataBuffer);
|
|
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-01-04 13:27:50 +03:00
|
|
|
NativeOpsHolder.getInstance().getDeviceNativeOps().memcpyConstantAsync(currentOffset, point.getHostPointer(), originalBytes, 1, context.getSpecialStream());
|
2019-06-06 15:21:15 +03:00
|
|
|
flowController.commitTransfer(context.getSpecialStream());
|
|
|
|
|
|
|
|
|
|
long cAddr = deviceAddresses.get(deviceId).address() + currentOffset;
|
|
|
|
|
|
|
|
|
|
//if (resetHappened)
|
|
|
|
|
// logger.info("copying to constant: {}, bufferLength: {}, bufferDtype: {}, currentOffset: {}, currentAddres: {}", requiredMemoryBytes, dataBuffer.length(), dataBuffer.dataType(), currentOffset, cAddr);
|
|
|
|
|
|
|
|
|
|
point.setAllocationStatus(AllocationStatus.CONSTANT);
|
2020-01-04 13:27:50 +03:00
|
|
|
//point.setDevicePointer(new CudaPointer(cAddr));
|
|
|
|
|
if (1 > 0)
|
|
|
|
|
throw new UnsupportedOperationException("Pew-pew");
|
|
|
|
|
|
2019-06-06 15:21:15 +03:00
|
|
|
point.setConstant(true);
|
|
|
|
|
point.tickDeviceWrite();
|
|
|
|
|
point.setDeviceId(deviceId);
|
|
|
|
|
point.tickHostRead();
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
protector.persistDataBuffer(dataBuffer);
|
|
|
|
|
|
|
|
|
|
return cAddr;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* PLEASE NOTE: This method implementation is hardware-dependant.
|
|
|
|
|
* PLEASE NOTE: This method does NOT allow concurrent use of any array
|
|
|
|
|
*
|
|
|
|
|
* @param dataBuffer
|
|
|
|
|
* @return
|
|
|
|
|
*/
|
|
|
|
|
@Override
|
|
|
|
|
public DataBuffer relocateConstantSpace(DataBuffer dataBuffer) {
|
|
|
|
|
// we always assume that data is sync, and valid on host side
|
|
|
|
|
Integer deviceId = AtomicAllocator.getInstance().getDeviceId();
|
|
|
|
|
ensureMaps(deviceId);
|
|
|
|
|
|
|
|
|
|
if (dataBuffer instanceof CudaIntDataBuffer) {
|
|
|
|
|
int[] data = dataBuffer.asInt();
|
|
|
|
|
return getConstantBuffer(data, DataType.INT);
|
|
|
|
|
} else if (dataBuffer instanceof CudaFloatDataBuffer) {
|
|
|
|
|
float[] data = dataBuffer.asFloat();
|
|
|
|
|
return getConstantBuffer(data, DataType.FLOAT);
|
|
|
|
|
} else if (dataBuffer instanceof CudaDoubleDataBuffer) {
|
|
|
|
|
double[] data = dataBuffer.asDouble();
|
|
|
|
|
return getConstantBuffer(data, DataType.DOUBLE);
|
|
|
|
|
} else if (dataBuffer instanceof CudaHalfDataBuffer) {
|
|
|
|
|
float[] data = dataBuffer.asFloat();
|
|
|
|
|
return getConstantBuffer(data, DataType.HALF);
|
|
|
|
|
} else if (dataBuffer instanceof CudaLongDataBuffer) {
|
|
|
|
|
long[] data = dataBuffer.asLong();
|
|
|
|
|
return getConstantBuffer(data, DataType.LONG);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
throw new IllegalStateException("Unknown CudaDataBuffer opType");
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
private void ensureMaps(Integer deviceId) {
|
|
|
|
|
if (!buffersCache.containsKey(deviceId)) {
|
|
|
|
|
if (flowController == null)
|
|
|
|
|
flowController = AtomicAllocator.getInstance().getFlowController();
|
|
|
|
|
|
|
|
|
|
try {
|
|
|
|
|
synchronized (this) {
|
|
|
|
|
if (!buffersCache.containsKey(deviceId)) {
|
|
|
|
|
|
|
|
|
|
// TODO: this op call should be checked
|
|
|
|
|
//nativeOps.setDevice(new CudaPointer(deviceId));
|
|
|
|
|
|
|
|
|
|
buffersCache.put(deviceId, new ConcurrentHashMap<ArrayDescriptor, DataBuffer>());
|
|
|
|
|
constantOffsets.put(deviceId, new AtomicLong(0));
|
|
|
|
|
deviceLocks.put(deviceId, new Semaphore(1));
|
|
|
|
|
|
|
|
|
|
Pointer cAddr = NativeOpsHolder.getInstance().getDeviceNativeOps().getConstantSpace();
|
|
|
|
|
// logger.info("constant pointer: {}", cAddr.address() );
|
|
|
|
|
|
|
|
|
|
deviceAddresses.put(deviceId, cAddr);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
} catch (Exception e) {
|
|
|
|
|
throw new RuntimeException(e);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* This method returns DataBuffer with contant equal to input array.
|
|
|
|
|
*
|
|
|
|
|
* PLEASE NOTE: This method assumes that you'll never ever change values within result DataBuffer
|
|
|
|
|
*
|
|
|
|
|
* @param array
|
|
|
|
|
* @return
|
|
|
|
|
*/
|
|
|
|
|
@Override
|
|
|
|
|
public DataBuffer getConstantBuffer(int[] array, DataType type) {
|
|
|
|
|
return Nd4j.getExecutioner().createConstantBuffer(array, type);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* This method returns DataBuffer with contant equal to input array.
|
|
|
|
|
*
|
|
|
|
|
* PLEASE NOTE: This method assumes that you'll never ever change values within result DataBuffer
|
|
|
|
|
*
|
|
|
|
|
* @param array
|
|
|
|
|
* @return
|
|
|
|
|
*/
|
|
|
|
|
@Override
|
|
|
|
|
public DataBuffer getConstantBuffer(float[] array, DataType type) {
|
|
|
|
|
return Nd4j.getExecutioner().createConstantBuffer(array, type);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* This method returns DataBuffer with contant equal to input array.
|
|
|
|
|
*
|
|
|
|
|
* PLEASE NOTE: This method assumes that you'll never ever change values within result DataBuffer
|
|
|
|
|
*
|
|
|
|
|
* @param array
|
|
|
|
|
* @return
|
|
|
|
|
*/
|
|
|
|
|
@Override
|
|
|
|
|
public DataBuffer getConstantBuffer(double[] array, DataType type) {
|
|
|
|
|
return Nd4j.getExecutioner().createConstantBuffer(array, type);
|
|
|
|
|
/*
|
|
|
|
|
ArrayDescriptor descriptor = new ArrayDescriptor(array, type);
|
|
|
|
|
|
|
|
|
|
Integer deviceId = AtomicAllocator.getInstance().getDeviceId();
|
|
|
|
|
|
|
|
|
|
ensureMaps(deviceId);
|
|
|
|
|
|
|
|
|
|
if (!buffersCache.get(deviceId).containsKey(descriptor)) {
|
|
|
|
|
// we create new databuffer
|
|
|
|
|
//logger.info("Creating new constant buffer...");
|
|
|
|
|
DataBuffer buffer = Nd4j.createTypedBufferDetached(array, type);
|
|
|
|
|
|
|
|
|
|
if (constantOffsets.get(deviceId).get() + (array.length * Nd4j.sizeOfDataType()) < MAX_CONSTANT_LENGTH) {
|
|
|
|
|
buffer.setConstant(true);
|
|
|
|
|
// now we move data to constant memory, and keep happy
|
|
|
|
|
moveToConstantSpace(buffer);
|
|
|
|
|
|
|
|
|
|
buffersCache.get(deviceId).put(descriptor, buffer);
|
|
|
|
|
|
|
|
|
|
bytes.addAndGet(array.length * Nd4j.sizeOfDataType());
|
|
|
|
|
}
|
|
|
|
|
return buffer;
|
|
|
|
|
} //else logger.info("Reusing constant buffer...");
|
|
|
|
|
|
|
|
|
|
return buffersCache.get(deviceId).get(descriptor);
|
|
|
|
|
*/
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
@Override
|
|
|
|
|
public DataBuffer getConstantBuffer(long[] array, DataType type) {
|
|
|
|
|
return Nd4j.getExecutioner().createConstantBuffer(array, type);
|
|
|
|
|
/*
|
|
|
|
|
// logger.info("getConstantBuffer(int[]) called");
|
|
|
|
|
ArrayDescriptor descriptor = new ArrayDescriptor(array, type);
|
|
|
|
|
|
|
|
|
|
Integer deviceId = AtomicAllocator.getInstance().getDeviceId();
|
|
|
|
|
|
|
|
|
|
ensureMaps(deviceId);
|
|
|
|
|
|
|
|
|
|
if (!buffersCache.get(deviceId).containsKey(descriptor)) {
|
|
|
|
|
// we create new databuffer
|
|
|
|
|
//logger.info("Creating new constant buffer...");
|
|
|
|
|
DataBuffer buffer = Nd4j.createTypedBufferDetached(array, type);
|
|
|
|
|
|
|
|
|
|
if (constantOffsets.get(deviceId).get() + (array.length * 8) < MAX_CONSTANT_LENGTH) {
|
|
|
|
|
buffer.setConstant(true);
|
|
|
|
|
// now we move data to constant memory, and keep happy
|
|
|
|
|
moveToConstantSpace(buffer);
|
|
|
|
|
|
|
|
|
|
buffersCache.get(deviceId).put(descriptor, buffer);
|
|
|
|
|
|
|
|
|
|
bytes.addAndGet(array.length * 8);
|
|
|
|
|
}
|
|
|
|
|
return buffer;
|
|
|
|
|
} //else logger.info("Reusing constant buffer...");
|
|
|
|
|
|
|
|
|
|
return buffersCache.get(deviceId).get(descriptor);
|
|
|
|
|
*/
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
@Override
|
|
|
|
|
public DataBuffer getConstantBuffer(boolean[] array, DataType dataType) {
|
|
|
|
|
return getConstantBuffer(ArrayUtil.toLongs(array), dataType);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
@Override
|
|
|
|
|
public long getCachedBytes() {
|
|
|
|
|
return bytes.get();
|
|
|
|
|
}
|
|
|
|
|
}
|