org.nd4j.linalg.factory.Nd4j.getMemoryManager()方法的使用及代码示例

x33g5p2x  于2022-01-24 转载在 其他  
字(12.3k)|赞(0)|评价(0)|浏览(221)

本文整理了Java中org.nd4j.linalg.factory.Nd4j.getMemoryManager()方法的一些代码示例，展示了Nd4j.getMemoryManager()的具体用法。这些代码示例主要来源于Github/Stackoverflow/Maven等平台，是从一些精选项目中提取出来的代码，具有较强的参考意义，能在一定程度帮忙到你。Nd4j.getMemoryManager()方法的具体详情如下：
包路径：org.nd4j.linalg.factory.Nd4j
类名称：Nd4j
方法名：getMemoryManager

Nd4j.getMemoryManager介绍

[英]This method returns backend-specific MemoryManager implementation, for low-level memory management
[中]此方法返回特定于后端的MemoryManager实现，用于低级内存管理

代码示例

代码示例来源：origin: deeplearning4j/nd4j

  1. /**
  2.  * This method notifies given Workspace that new use cycle is starting now
  3.  *
  4.  * @return
  5.  */
  6. @Override
  7. public MemoryWorkspace notifyScopeEntered() {
  8.   parentWorkspace = Nd4j.getMemoryManager().getCurrentWorkspace();
  9.   Nd4j.getMemoryManager().setCurrentWorkspace(null);
  10.   return this;
  11. }

代码示例来源：origin: deeplearning4j/nd4j

  1. @Override
  2. public void close() {
  3.   Nd4j.getMemoryManager().setCurrentWorkspace(parentWorkspace);
  4. }

代码示例来源：origin: deeplearning4j/nd4j

  1. /**
  2.  * This method TEMPORARY enters this workspace, without reset applied
  3.  *
  4.  * @return
  5.  */
  6. @Override
  7. public MemoryWorkspace notifyScopeBorrowed() {
  8.   if (isBorrowed.get())
  9.     throw new ND4JIllegalStateException("Workspace [" + id + "]: Can't borrow from borrowed workspace");
  10.   borrowingWorkspace = Nd4j.getMemoryManager().getCurrentWorkspace();
  11.   isBorrowed.set(true);
  12.   Nd4j.getMemoryManager().setCurrentWorkspace(this);
  13.   return this;
  14. }

代码示例来源：origin: deeplearning4j/nd4j

  1. public static DataBuffer createBuffer(long[] shape, DataBuffer.Type type) {
  2.   long length = ArrayUtil.prodLong(shape);
  3.   if (type == DataBuffer.Type.INT)
  4.     return Nd4j.getMemoryManager().getCurrentWorkspace() == null ? DATA_BUFFER_FACTORY_INSTANCE.createInt(length, true) : DATA_BUFFER_FACTORY_INSTANCE.createInt(length, true, Nd4j.getMemoryManager().getCurrentWorkspace());
  5.   else if (type == DataBuffer.Type.LONG)
  6.     return Nd4j.getMemoryManager().getCurrentWorkspace() == null ? DATA_BUFFER_FACTORY_INSTANCE.createLong(length, true) : DATA_BUFFER_FACTORY_INSTANCE.createLong(length, true, Nd4j.getMemoryManager().getCurrentWorkspace());
  7.   else if (type == DataBuffer.Type.HALF)
  8.     return Nd4j.getMemoryManager().getCurrentWorkspace() == null ? DATA_BUFFER_FACTORY_INSTANCE.createHalf(length, true) : DATA_BUFFER_FACTORY_INSTANCE.createHalf(length, true, Nd4j.getMemoryManager().getCurrentWorkspace());
  9.   else if (type == DataBuffer.Type.DOUBLE)
  10.     return Nd4j.getMemoryManager().getCurrentWorkspace() == null ? DATA_BUFFER_FACTORY_INSTANCE.createDouble(length, true) : DATA_BUFFER_FACTORY_INSTANCE.createDouble(length, true, Nd4j.getMemoryManager().getCurrentWorkspace());
  11.   else
  12.     return Nd4j.getMemoryManager().getCurrentWorkspace() == null ? DATA_BUFFER_FACTORY_INSTANCE.createFloat(length, true) : DATA_BUFFER_FACTORY_INSTANCE.createFloat(length, true, Nd4j.getMemoryManager().getCurrentWorkspace());
  13. }

代码示例来源：origin: deeplearning4j/nd4j

  1. /**
  2.  * Get the feature wise
  3.  * range for the statistics.
  4.  * Note that this is a lazy getter.
  5.  * It is only computed when needed.
  6.  * @return the feature wise range
  7.  * given the min and max
  8.  */
  9. public INDArray getRange() {
  10.   if (range == null) {
  11.     try (MemoryWorkspace ws = Nd4j.getMemoryManager().scopeOutOfWorkspaces()) {
  12.       range = upper.sub(lower);
  13.     }
  14.   }
  15.   return range;
  16. }

代码示例来源：origin: deeplearning4j/nd4j

  1. /**
  2.  * This method temporary opens block out of any workspace scope.
  3.  * <p>
  4.  * PLEASE NOTE: Do not forget to close this block.
  5.  *
  6.  * @return
  7.  */
  8. @Override
  9. public MemoryWorkspace scopeOutOfWorkspaces() {
  10.   MemoryWorkspace workspace = Nd4j.getMemoryManager().getCurrentWorkspace();
  11.   if (workspace == null)
  12.     return new DummyWorkspace();
  13.   else {
  14.     Nd4j.getMemoryManager().setCurrentWorkspace(null);
  15.     return workspace.tagOutOfScopeUse();
  16.   }
  17. }

代码示例来源：origin: deeplearning4j/nd4j

  1. /**
  2.  * Create a buffer equal of length prod(shape)
  3.  *
  4.  * @param data the shape of the buffer to create
  5.  * @return the created buffer
  6.  */
  7. public static DataBuffer createBuffer(int[] data) {
  8.   DataBuffer ret;
  9.   ret = Nd4j.getMemoryManager().getCurrentWorkspace() == null ? DATA_BUFFER_FACTORY_INSTANCE.createInt(data) : DATA_BUFFER_FACTORY_INSTANCE.createInt(data, Nd4j.getMemoryManager().getCurrentWorkspace());
  10.   logCreationIfNecessary(ret);
  11.   return ret;
  12. }

代码示例来源：origin: deeplearning4j/nd4j

  1. @Override
  2. public DataBuffer decompress(DataBuffer buffer) {
  3.   CompressedDataBuffer comp = (CompressedDataBuffer) buffer;
  4.   DataBuffer result = Nd4j.createBuffer(comp.length(), false);
  5.   Nd4j.getMemoryManager().memcpy(result, buffer);
  6.   return result;
  7. }

代码示例来源：origin: deeplearning4j/nd4j

  1. /**
  2.    * Create a DistributionStats object from the data ingested so far. Can be used multiple times when updating
  3.    * online.
  4.    */
  5.   public MinMaxStats build() {
  6.     if (runningLower == null) {
  7.       throw new RuntimeException("No data was added, statistics cannot be determined");
  8.     }
  9.     try (MemoryWorkspace workspace = Nd4j.getMemoryManager().scopeOutOfWorkspaces()) {
  10.       return new MinMaxStats(runningLower.dup(), runningUpper.dup());
  11.     }
  12.   }
  13. }

代码示例来源：origin: deeplearning4j/nd4j

  1. @Override
  2. public void assertCurrentWorkspace(@NonNull T arrayType, String msg) {
  3.   validateConfig(arrayType);
  4.   MemoryWorkspace curr = Nd4j.getMemoryManager().getCurrentWorkspace();
  5.   if(!scopeOutOfWs.contains(arrayType) && (curr == null || !getWorkspaceName(arrayType).equals(curr.getId()))){
  6.     throw new ND4JWorkspaceException("Assertion failed: expected current workspace to be \"" + getWorkspaceName(arrayType)
  7.         + "\" (for array type " + arrayType + ") - actual current workspace is " + (curr == null ? null : curr.getId())
  8.         + (msg == null ? "" : ": " + msg));
  9.   };
  10. }

代码示例来源：origin: deeplearning4j/nd4j

  1. /**
  2.  * Create a buffer based on the data opType
  3.  *
  4.  * @param data the data to create the buffer with
  5.  * @return the created buffer
  6.  */
  7. public static DataBuffer createBuffer(float[] data) {
  8.   DataBuffer ret;
  9.   if (dataType() == DataBuffer.Type.FLOAT)
  10.     ret = Nd4j.getMemoryManager().getCurrentWorkspace() == null ? DATA_BUFFER_FACTORY_INSTANCE.createFloat(data) : DATA_BUFFER_FACTORY_INSTANCE.createFloat(data, Nd4j.getMemoryManager().getCurrentWorkspace());
  11.   else if (dataType() == DataBuffer.Type.HALF)
  12.     ret = Nd4j.getMemoryManager().getCurrentWorkspace() == null ? DATA_BUFFER_FACTORY_INSTANCE.createHalf(data): DATA_BUFFER_FACTORY_INSTANCE.createHalf(data, Nd4j.getMemoryManager().getCurrentWorkspace());
  13.   else
  14.     ret = Nd4j.getMemoryManager().getCurrentWorkspace() == null ? DATA_BUFFER_FACTORY_INSTANCE.createDouble(ArrayUtil.toDoubles(data)) : DATA_BUFFER_FACTORY_INSTANCE.createDouble(ArrayUtil.toDoubles(data), Nd4j.getMemoryManager().getCurrentWorkspace()) ;
  15.   logCreationIfNecessary(ret);
  16.   return ret;
  17. }

代码示例来源：origin: deeplearning4j/nd4j

  1. /**
  2.  * Create a buffer based on the data opType
  3.  *
  4.  * @param data the data to create the buffer with
  5.  * @return the created buffer
  6.  */
  7. public static DataBuffer createBuffer(float[] data, long offset) {
  8.   DataBuffer ret;
  9.   if (dataType() == DataBuffer.Type.FLOAT)
  10.     ret = Nd4j.getMemoryManager().getCurrentWorkspace() == null ? DATA_BUFFER_FACTORY_INSTANCE.createFloat(offset, data) : DATA_BUFFER_FACTORY_INSTANCE.createFloat(offset, data, Nd4j.getMemoryManager().getCurrentWorkspace());
  11.   else if (dataType() == DataBuffer.Type.HALF)
  12.     ret = Nd4j.getMemoryManager().getCurrentWorkspace() == null ? DATA_BUFFER_FACTORY_INSTANCE.createHalf(offset, data) : DATA_BUFFER_FACTORY_INSTANCE.createHalf(offset, data, Nd4j.getMemoryManager().getCurrentWorkspace());
  13.   else
  14.     ret = Nd4j.getMemoryManager().getCurrentWorkspace() == null ? DATA_BUFFER_FACTORY_INSTANCE.createDouble(offset, ArrayUtil.toDoubles(data)) : DATA_BUFFER_FACTORY_INSTANCE.createDouble(offset, ArrayUtil.toDoubles(data), Nd4j.getMemoryManager().getCurrentWorkspace());
  15.   logCreationIfNecessary(ret);
  16.   return ret;
  17. }

代码示例来源：origin: deeplearning4j/nd4j

  1. /**
  2.  * Create a buffer based on the data opType
  3.  *
  4.  * @param data the data to create the buffer with
  5.  * @return the created buffer
  6.  */
  7. public static DataBuffer createBuffer(double[] data, long offset) {
  8.   DataBuffer ret;
  9.   if (dataType() == DataBuffer.Type.DOUBLE)
  10.     ret = Nd4j.getMemoryManager().getCurrentWorkspace() == null ? DATA_BUFFER_FACTORY_INSTANCE.createDouble(offset, data) : DATA_BUFFER_FACTORY_INSTANCE.createDouble(offset, data, Nd4j.getMemoryManager().getCurrentWorkspace());
  11.   else if (dataType() == DataBuffer.Type.HALF)
  12.     ret = Nd4j.getMemoryManager().getCurrentWorkspace() == null ? DATA_BUFFER_FACTORY_INSTANCE.createHalf(offset, data) : DATA_BUFFER_FACTORY_INSTANCE.createHalf(offset, ArrayUtil.toFloats(data), Nd4j.getMemoryManager().getCurrentWorkspace());
  13.   else
  14.     ret = Nd4j.getMemoryManager().getCurrentWorkspace() == null ? DATA_BUFFER_FACTORY_INSTANCE.createFloat(offset, ArrayUtil.toFloats(data)) : DATA_BUFFER_FACTORY_INSTANCE.createFloat(offset, ArrayUtil.toFloats(data), Nd4j.getMemoryManager().getCurrentWorkspace());
  15.   logCreationIfNecessary(ret);
  16.   return ret;
  17. }

代码示例来源：origin: deeplearning4j/nd4j

  1. /**
  2.  *
  3.  * @param length
  4.  * @param initialize
  5.  * @return
  6.  */
  7. public static DataBuffer createBuffer(long length, boolean initialize) {
  8.   DataBuffer ret;
  9.   if (dataType() == DataBuffer.Type.FLOAT)
  10.     ret = Nd4j.getMemoryManager().getCurrentWorkspace() == null ? DATA_BUFFER_FACTORY_INSTANCE.createFloat(length, initialize) : DATA_BUFFER_FACTORY_INSTANCE.createFloat(length, initialize, Nd4j.getMemoryManager().getCurrentWorkspace());
  11.   else if (dataType() == DataBuffer.Type.INT)
  12.     ret = DATA_BUFFER_FACTORY_INSTANCE.createInt(length, initialize);
  13.   else if (dataType() == DataBuffer.Type.HALF)
  14.     ret = Nd4j.getMemoryManager().getCurrentWorkspace() == null ? DATA_BUFFER_FACTORY_INSTANCE.createHalf(length, initialize) : DATA_BUFFER_FACTORY_INSTANCE.createHalf(length, initialize, Nd4j.getMemoryManager().getCurrentWorkspace());
  15.   else
  16.     ret = Nd4j.getMemoryManager().getCurrentWorkspace() == null ? DATA_BUFFER_FACTORY_INSTANCE.createDouble(length, initialize) : DATA_BUFFER_FACTORY_INSTANCE.createDouble(length, initialize, Nd4j.getMemoryManager().getCurrentWorkspace());
  17.   logCreationIfNecessary(ret);
  18.   return ret;
  19. }

代码示例来源：origin: deeplearning4j/nd4j

  1. /**
  2.  * Create a buffer based on the data opType
  3.  *
  4.  * @param data the data to create the buffer with
  5.  * @return the created buffer
  6.  */
  7. public static DataBuffer createBuffer(double[] data) {
  8.   DataBuffer ret;
  9.   if (dataType() == DataBuffer.Type.DOUBLE)
  10.     ret = Nd4j.getMemoryManager().getCurrentWorkspace() == null ? DATA_BUFFER_FACTORY_INSTANCE.createDouble(data) : DATA_BUFFER_FACTORY_INSTANCE.createDouble(data, Nd4j.getMemoryManager().getCurrentWorkspace());
  11.   else if (dataType() == DataBuffer.Type.HALF)
  12.     ret = Nd4j.getMemoryManager().getCurrentWorkspace() == null ? DATA_BUFFER_FACTORY_INSTANCE.createHalf(data) : DATA_BUFFER_FACTORY_INSTANCE.createHalf(ArrayUtil.toFloats(data), Nd4j.getMemoryManager().getCurrentWorkspace());
  13.   else
  14.     ret = Nd4j.getMemoryManager().getCurrentWorkspace() == null ? DATA_BUFFER_FACTORY_INSTANCE.createFloat(ArrayUtil.toFloats(data)) : DATA_BUFFER_FACTORY_INSTANCE.createFloat(ArrayUtil.toFloats(data), Nd4j.getMemoryManager().getCurrentWorkspace());
  15.   logCreationIfNecessary(ret);
  16.   return ret;
  17. }

代码示例来源：origin: deeplearning4j/nd4j

  1. @Override
  2.   public MemoryWorkspace scopeOutOfWorkspaces() {
  3.     MemoryWorkspace workspace = Nd4j.getMemoryManager().getCurrentWorkspace();
  4.     if (workspace == null)
  5.       return new DummyWorkspace();
  6.     else {
  7.       //Nd4j.getMemoryManager().setCurrentWorkspace(null);
  8.       return new DummyWorkspace().notifyScopeEntered();//workspace.tagOutOfScopeUse();
  9.     }
  10.   }
  11. }

代码示例来源：origin: deeplearning4j/nd4j

  1. /**
  2.  * Assert that the specified workspace is open, active, and is the current workspace
  3.  *
  4.  * @param ws       Name of the workspace to assert open/active/current
  5.  * @param errorMsg Message to include in the exception, if required
  6.  */
  7. public static void assertOpenActiveAndCurrent(@NonNull String ws, @NonNull String errorMsg) throws ND4JWorkspaceException {
  8.   if (!Nd4j.getWorkspaceManager().checkIfWorkspaceExistsAndActive(ws)) {
  9.     throw new ND4JWorkspaceException(errorMsg + " - workspace is not open and active");
  10.   }
  11.   MemoryWorkspace currWs = Nd4j.getMemoryManager().getCurrentWorkspace();
  12.   if (currWs == null || !ws.equals(currWs.getId())) {
  13.     throw new ND4JWorkspaceException(errorMsg + " - not the current workspace (current workspace: "
  14.         + (currWs == null ? null : currWs.getId()));
  15.   }
  16. }

代码示例来源：origin: deeplearning4j/nd4j

  1. /**
  2.  * This method migrates this DataSet into current Workspace (if any)
  3.  */
  4. @Override
  5. public void migrate() {
  6.   if (Nd4j.getMemoryManager().getCurrentWorkspace() != null) {
  7.     if (features != null)
  8.       features = features.migrate();
  9.     if (labels != null)
  10.       labels = labels.migrate();
  11.     if (featuresMask != null)
  12.       featuresMask = featuresMask.migrate();
  13.     if (labelsMask != null)
  14.       labelsMask = labelsMask.migrate();
  15.   }
  16. }

代码示例来源：origin: deeplearning4j/nd4j

  1. @Override
  2. public DataBuffer compress(DataBuffer buffer) {
  3.   CompressionDescriptor descriptor = new CompressionDescriptor(buffer, this);
  4.   BytePointer ptr = new BytePointer(buffer.length() * buffer.getElementSize());
  5.   CompressedDataBuffer result = new CompressedDataBuffer(ptr, descriptor);
  6.   Nd4j.getMemoryManager().memcpy(result, buffer);
  7.   return result;
  8. }

代码示例来源：origin: deeplearning4j/nd4j

  1. @Override
  2. public void migrate() {
  3.   if (Nd4j.getMemoryManager().getCurrentWorkspace() != null) {
  4.     if (features != null)
  5.       for (int e = 0; e < features.length; e++)
  6.         features[e] = features[e].migrate();
  7.     if (labels != null)
  8.       for (int e = 0; e < labels.length; e++)
  9.         labels[e] = labels[e].migrate();
  10.     if (featuresMaskArrays != null)
  11.       for (int e = 0; e < featuresMaskArrays.length; e++)
  12.         featuresMaskArrays[e] = featuresMaskArrays[e].migrate();
  13.     if (labelsMaskArrays != null)
  14.       for (int e = 0; e < labelsMaskArrays.length; e++)
  15.         labelsMaskArrays[e] = labelsMaskArrays[e].migrate();
  16.   }
  17. }

内容来源于网络，如有侵权，请联系作者删除！

Nd4j

相关文章

    查看更多

    热门标签

    更多
    JavaquerypythonNode开发语言requestUtil数据库Table后端算法LoggerMessageElementParser

    最新文章

    更多

    Nd4j类方法

    • 蜀ICP备13028337号-1 大数据知识库 https://www.saoniuhuo.com © All rights reserved
    • 本站内容来源互联网,如果侵犯您的权益请联系我们删除, 联系方式：448109455@qq.com