Package org.nd4j.autodiff.samediff.optimize.optimizations

Class CuDNNFunctionOptimizations.CudnnConv2dNCHWtoNHWCConversion

java.lang.Object

org.nd4j.autodiff.samediff.optimize.optimizations.CuDNNFunctionOptimizations.CudnnConv2dNCHWtoNHWCConversion

All Implemented Interfaces:

Optimizer

Enclosing class:

CuDNNFunctionOptimizations

public static class CuDNNFunctionOptimizations.CudnnConv2dNCHWtoNHWCConversion
extends Object
implements Optimizer

https://docs.nvidia.com/deeplearning/sdk/dl-performance-guide/index.html#tensor-layout For tensor cores: we want NHWC layout: Section 7.3.1 "Layout choice has an effect on performance, as convolutions implemented for Tensor Cores require NHWC layout and are fastest when input tensors are laid out in NHWC." "To maximize performance, we recommend using NHWC tensor layout." As for weights format: cuDNN docs are vague - but TF uses NCHW+OIHW or NHWC+OHWI

Constructor Summary

Constructors

Constructor	Description
CudnnConv2dNCHWtoNHWCConversion()

Method Summary

Modifier and Type	Method	Description
boolean	checkAndApply(SameDiff sd, OptimizationHelper helper, SameDiffOp op, ArrayHolder constantArrays, ArrayHolder variablesArrays)

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

CudnnConv2dNCHWtoNHWCConversion

public CudnnConv2dNCHWtoNHWCConversion()

Method Detail

checkAndApply

public boolean checkAndApply(SameDiff sd,
                             OptimizationHelper helper,
                             SameDiffOp op,
                             ArrayHolder constantArrays,
                             ArrayHolder variablesArrays)

Specified by:

checkAndApply in interface Optimizer

Parameters:

sd - Current SameDiff instance to optimize

helper - Helper class for optimization

op - Operation to check for optimization

constantArrays - Array holder for constant arrays

variablesArrays - Array holder for variable arrays

Returns:

True if the optimization was applied