Modifier and Type | Method and Description |
---|---|
AllocationPoint |
Allocator.allocateMemory(DataBuffer buffer,
AllocationShape requiredMemory,
AllocationStatus location,
boolean initialize)
This method allocates required chunk of memory in specific location
PLEASE NOTE: Do not use this method, unless you're 100% sure what you're doing
|
AllocationPoint |
Allocator.allocateMemory(DataBuffer buffer,
AllocationShape requiredMemory,
boolean initialize)
This method allocates required chunk of memory
|
AllocationPoint |
Allocator.getAllocationPoint(DataBuffer buffer) |
AllocationPoint |
Allocator.getAllocationPoint(INDArray array) |
Modifier and Type | Method and Description |
---|---|
AllocationPoint |
GarbageReference.getPoint() |
Constructor and Description |
---|
GarbageReference(BaseDataBuffer referent,
ReferenceQueue<? super BaseDataBuffer> q,
AllocationPoint point) |
Modifier and Type | Method and Description |
---|---|
AllocationPoint |
AtomicAllocator.allocateMemory(DataBuffer buffer,
AllocationShape requiredMemory,
AllocationStatus location,
boolean initialize)
This method allocates required chunk of memory in specific location
|
AllocationPoint |
AtomicAllocator.allocateMemory(DataBuffer buffer,
AllocationShape requiredMemory,
boolean initialize)
This method allocates required chunk of memory
|
AllocationPoint |
AtomicAllocator.getAllocationPoint(DataBuffer buffer) |
AllocationPoint |
AtomicAllocator.getAllocationPoint(INDArray array) |
protected AllocationPoint |
AtomicAllocator.getAllocationPoint(Long objectId)
This method returns AllocationPoint POJO for specified tracking ID
|
Modifier and Type | Method and Description |
---|---|
protected void |
AtomicAllocator.purgeDeviceObject(Long threadId,
Integer deviceId,
Long objectId,
AllocationPoint point,
boolean copyback)
This method frees native device memory referenced by specified tracking id/AllocationPoint
|
protected void |
AtomicAllocator.purgeZeroObject(Long bucketId,
Long objectId,
AllocationPoint point,
boolean copyback)
This method frees native system memory referenced by specified tracking id/AllocationPoint
|
Modifier and Type | Method and Description |
---|---|
AllocationStatus |
Balancer.makeDemoteDecision(Integer deviceId,
AllocationPoint point,
AllocationShape shape)
Deprecated.
This method checks, if it's worth moving some memory region to host
|
AllocationStatus |
Balancer.makePromoteDecision(Integer deviceId,
AllocationPoint point,
AllocationShape shape)
Deprecated.
This method checks, if it's worth moving some memory region to device
|
Modifier and Type | Method and Description |
---|---|
CudaContext |
FlowController.prepareAction(AllocationPoint result,
AllocationPoint... operands) |
CudaContext |
FlowController.prepareAction(AllocationPoint result,
AllocationPoint... operands) |
void |
FlowController.registerAction(CudaContext context,
AllocationPoint result,
AllocationPoint... operands) |
void |
FlowController.registerAction(CudaContext context,
AllocationPoint result,
AllocationPoint... operands) |
void |
FlowController.synchronizeToHost(AllocationPoint point)
This method ensures, that all asynchronous operations on referenced AllocationPoint are finished, and host memory state is up-to-date
|
void |
FlowController.waitTillFinished(AllocationPoint point)
This method ensures, that all asynchronous operations on referenced AllocationPoint are finished
|
void |
FlowController.waitTillReleased(AllocationPoint point) |
Modifier and Type | Method and Description |
---|---|
protected boolean |
AsynchronousFlowController.hasActiveReads(AllocationPoint point) |
protected int |
AsynchronousFlowController.hasActiveWrite(AllocationPoint point) |
CudaContext |
SynchronousFlowController.prepareAction(AllocationPoint result,
AllocationPoint... operands) |
CudaContext |
SynchronousFlowController.prepareAction(AllocationPoint result,
AllocationPoint... operands) |
CudaContext |
AsynchronousFlowController.prepareAction(AllocationPoint result,
AllocationPoint... operands) |
CudaContext |
AsynchronousFlowController.prepareAction(AllocationPoint result,
AllocationPoint... operands) |
void |
SynchronousFlowController.registerAction(CudaContext context,
AllocationPoint result,
AllocationPoint... operands) |
void |
SynchronousFlowController.registerAction(CudaContext context,
AllocationPoint result,
AllocationPoint... operands) |
void |
AsynchronousFlowController.registerAction(CudaContext context,
AllocationPoint result,
AllocationPoint... operands) |
void |
AsynchronousFlowController.registerAction(CudaContext context,
AllocationPoint result,
AllocationPoint... operands) |
protected void |
AsynchronousFlowController.synchronizeReadLanes(AllocationPoint point) |
void |
SynchronousFlowController.synchronizeToHost(AllocationPoint point) |
void |
AsynchronousFlowController.synchronizeToHost(AllocationPoint point) |
void |
SynchronousFlowController.waitTillFinished(AllocationPoint point) |
void |
AsynchronousFlowController.waitTillFinished(AllocationPoint point) |
void |
SynchronousFlowController.waitTillReleased(AllocationPoint point) |
void |
AsynchronousFlowController.waitTillReleased(AllocationPoint point) |
Modifier and Type | Method and Description |
---|---|
PointersPair |
MemoryHandler.alloc(AllocationStatus targetMode,
AllocationPoint point,
AllocationShape shape,
boolean initialize)
Allocate specified memory chunk on specified device/host
|
void |
MemoryHandler.copyback(AllocationPoint point,
AllocationShape shape)
Copies memory from device to host, if needed.
|
void |
MemoryHandler.copyforward(AllocationPoint point,
AllocationShape shape)
Copies memory from host buffer to device.
|
void |
MemoryHandler.fallback(AllocationPoint point,
AllocationShape shape)
Copies memory from device to zero-copy memory
|
void |
MemoryHandler.free(AllocationPoint point,
AllocationStatus target)
This method frees memory chunk specified by pointer
|
void |
MemoryHandler.purgeDeviceObject(Long threadId,
Integer deviceId,
Long objectId,
AllocationPoint point,
boolean copyback)
This method removes specific previously allocated object from device memory
|
void |
MemoryHandler.purgeZeroObject(Long bucketId,
Long objectId,
AllocationPoint point,
boolean copyback)
This method removes specific previously allocated object from host memory
|
void |
MemoryHandler.relocate(AllocationStatus currentStatus,
AllocationStatus targetStatus,
AllocationPoint point,
AllocationShape shape,
CudaContext context)
Relocates specific chunk of memory from one storage to another
|
void |
MemoryHandler.synchronizeThreadDevice(Long threadId,
Integer deviceId,
AllocationPoint point)
This method causes memory synchronization on host side.
|
Modifier and Type | Method and Description |
---|---|
PointersPair |
CudaZeroHandler.alloc(AllocationStatus targetMode,
AllocationPoint point,
AllocationShape shape,
boolean initialize)
Allocate specified memory chunk on specified device/host
|
void |
CudaZeroHandler.copyback(AllocationPoint point,
AllocationShape shape)
Deprecated.
|
void |
CudaZeroHandler.copyforward(AllocationPoint point,
AllocationShape shape)
Deprecated.
|
void |
CudaZeroHandler.fallback(AllocationPoint point,
AllocationShape shape)
Deprecated.
|
void |
CudaZeroHandler.free(AllocationPoint point,
AllocationStatus target)
This method frees memory chunk specified by pointer and location
|
boolean |
CudaZeroHandler.promoteObject(Long trackingPoint,
AllocationPoint point,
AllocationShape shape)
This method moves specific object from zero-copy memory to device memory
PLEASE NOTE: DO NOT EVER USE THIS METHOD MANUALLY, UNLESS YOU 100% HAVE TO
|
void |
CudaZeroHandler.purgeDeviceObject(Long threadId,
Integer deviceId,
Long objectId,
AllocationPoint point,
boolean copyback)
This method explicitly removes object from device memory.
|
void |
CudaZeroHandler.purgeZeroObject(Long bucketId,
Long objectId,
AllocationPoint point,
boolean copyback)
This method explicitly removes object from zero-copy memory.
|
void |
CudaZeroHandler.relocate(AllocationStatus currentStatus,
AllocationStatus targetStatus,
AllocationPoint point,
AllocationShape shape,
CudaContext context)
Copies specific chunk of memory from one storage to another
Possible directions: HOST -> DEVICE, DEVICE -> HOST
|
void |
CudaZeroHandler.synchronizeThreadDevice(Long threadId,
Integer deviceId,
AllocationPoint point)
This method causes memory synchronization on host side.
|
Modifier and Type | Method and Description |
---|---|
void |
MemoryProvider.free(AllocationPoint point)
This method frees specific chunk of memory, described by AllocationPoint passed in
|
PointersPair |
MemoryProvider.malloc(AllocationShape shape,
AllocationPoint point,
AllocationStatus location)
This method provides PointersPair to memory chunk specified by AllocationShape
|
Modifier and Type | Method and Description |
---|---|
void |
CudaFullCachingProvider.free(AllocationPoint point) |
void |
CudaDirectProvider.free(AllocationPoint point)
This method frees specific chunk of memory, described by AllocationPoint passed in
|
void |
CudaCachingZeroProvider.free(AllocationPoint point)
This method frees specific chunk of memory, described by AllocationPoint passed in.
|
PointersPair |
CudaFullCachingProvider.malloc(AllocationShape shape,
AllocationPoint point,
AllocationStatus location) |
PointersPair |
CudaDirectProvider.malloc(AllocationShape shape,
AllocationPoint point,
AllocationStatus location)
This method provides PointersPair to memory chunk specified by AllocationShape
|
PointersPair |
CudaCachingZeroProvider.malloc(AllocationShape shape,
AllocationPoint point,
AllocationStatus location)
This method provides PointersPair to memory chunk specified by AllocationShape
PLEASE NOTE: This method can actually ignore malloc request, and give out previously cached free memory chunk with equal shape.
|
Modifier and Type | Field and Description |
---|---|
protected AllocationPoint |
BaseCudaDataBuffer.allocationPoint |
Copyright © 2016. All Rights Reserved.