Commit Graph

8 Commits (2911da061b330f255a8c4b85c4ef4d72705286d3)

Author SHA1 Message Date
Yurii Shyrma 78934c17ad
profiling of stack and unstack ops (#261)
* - profiling of stack and unstack ops

Signed-off-by: Yurii <iuriish@yahoo.com>

* - fix bug in cpu concat op

Signed-off-by: Yurii <iuriish@yahoo.com>

* - correction of cuda stack and unstack

Signed-off-by: Yurii <iuriish@yahoo.com>

* - change shape.h method which operates with unity dimensions strides

Signed-off-by: Yurii <iuriish@yahoo.com>

* - rearrange stack tests

Signed-off-by: Yurii <iuriish@yahoo.com>

* - correct evaluation of smallest stride for moving through contiguous axis

Signed-off-by: Yurii <iuriish@yahoo.com>

* - forgot to update signature of function strideOverContigAxis in cuda concat and split ops

Signed-off-by: Yurii <iuriish@yahoo.com>

* - remove ShapeUtils::shapeAsString method applied before input arrays validations

Signed-off-by: Yurii <iuriish@yahoo.com>

* -  further removing of ShapeUtils::shapeAsString

Signed-off-by: Yurii <iuriish@yahoo.com>

* - take sub-array shapeIndo/offset calculation out of NDArray class
- add possibility of contiguous memory copy in execTransformAny op if opNum == assign

Signed-off-by: Yurii <iuriish@yahoo.com>

* - correct test_empty_scatter_2 in EmptyTests.cpp

Signed-off-by: Yurii <iuriish@yahoo.com>

* - profiling of slice op

Signed-off-by: Yurii <iuriish@yahoo.com>

* - get rid of contiguous memcpy for some cases in concat and split ops

Signed-off-by: Yurii <iuriish@yahoo.com>

* - forgot to declare oid nd4j::SpecialMethods<T>::splitCpuGeneric

Signed-off-by: Yurii <iuriish@yahoo.com>

* - correct typo in calculation of threads in cuda split op

Signed-off-by: Yurii <iuriish@yahoo.com>

* - forgot to correct another set of threads variables in split cuda ops

Signed-off-by: Yurii <iuriish@yahoo.com>

* - further conflicts resolving

Signed-off-by: Yurii <iuriish@yahoo.com>

Co-authored-by: raver119 <raver119@gmail.com>
2020-03-03 07:32:37 +03:00
raver119 63fa3c2ef3
libnd4j polishing (#273)
* initial set of include changes

Signed-off-by: raver119 <raver119@gmail.com>

* one more tweak

Signed-off-by: raver119 <raver119@gmail.com>

* few more rearrangements

Signed-off-by: raver119 <raver119@gmail.com>

* few more rearrangements

Signed-off-by: raver119 <raver119@gmail.com>

* few more rearrangements

Signed-off-by: raver119 <raver119@gmail.com>

* cuda includes rearrangements

Signed-off-by: raver119 <raver119@gmail.com>

* java update

Signed-off-by: raver119 <raver119@gmail.com>

* = namespace changed to sd
- few CMake variables renamed with SD_ prefix

Signed-off-by: raver119 <raver119@gmail.com>

* java update

Signed-off-by: raver119 <raver119@gmail.com>

* LoopKind minor fix

Signed-off-by: raver119 <raver119@gmail.com>

* few more changes

Signed-off-by: raver119 <raver119@gmail.com>

* few more changes

Signed-off-by: raver119 <raver119@gmail.com>

* few more changes

Signed-off-by: raver119 <raver119@gmail.com>

* sanitizer is optional now

Signed-off-by: raver119 <raver119@gmail.com>

* dev tests updated

Signed-off-by: raver119 <raver119@gmail.com>

* few more changes

Signed-off-by: raver119 <raver119@gmail.com>

* last update

Signed-off-by: raver119 <raver119@gmail.com>

* java update

Signed-off-by: raver119 <raver119@gmail.com>
2020-03-02 12:49:41 +03:00
Yurii Shyrma fe47f52896
Oleh tenzor mmul (#231)
* Libnd4j: TensorMMul backprop op #8174, raw implementation

Signed-off-by: Oleg <oleg.semeniv@gmail.com>

* Libnd4j: TensorMMul backprop op #8174 merge master and some corrections

Signed-off-by: Oleg <oleg.semeniv@gmail.com>

* Libnd4j: TensorMMul backprop op #8174 algorithm update, need testing, sync with  master

* Libnd4j: TensorMMul backprop op #8174 fixed incorrect B axes calculation

Signed-off-by: Oleg <oleg.semeniv@gmail.com>

* Libnd4j: TensorMMul backprop op #8174 optimize axes identification and fix bug of indeces overlapping, added first test. need testing with different shapes

Signed-off-by: Oleg <oleg.semeniv@gmail.com>

* Libnd4j: TensorMMul backprop op #8174 some fixes and improvements need more testing

Signed-off-by: Oleg <oleg.semeniv@gmail.com>

* Libnd4j: TensorMMul backprop op #8174 fixed order of matrix multiply

Signed-off-by: Oleg <oleg.semeniv@gmail.com>

* Libnd4j: TensorMMul backprop op #8174 fixed issue of incorrect axes definition, add tests based on TF, need additional testing for case dLdC not equal 1

Signed-off-by: Oleg <oleg.semeniv@gmail.com>

* Libnd4j: TensorMMul backprop op #8174 fixed scalar case add test

Signed-off-by: Oleg <oleg.semeniv@gmail.com>

* Libnd4j: TensorMMul backprop op #8174 fixed bp algorithm, axes definition, need some mode testing with different orders combination f,c; c,f f,f and add some checks for inputs

Signed-off-by: Oleg <oleg.semeniv@gmail.com>

* Libnd4j: TensorMMul backprop op #8174 some checks and corrections added tests, exists the problem with different input orders support A-f B-c and A-f B-f

Signed-off-by: Oleg <oleg.semeniv@gmail.com>

* Libnd4j: TensorMMul backprop op #8174 sync master

Signed-off-by: Oleg <oleg.semeniv@gmail.com>

* - correct bug in MmulHelper::tensorDot(a, b, c, axes_a, axes_b,permutForC)

Signed-off-by: Yurii <iuriish@yahoo.com>

* Libnd4j: TensorMMul backprop op #8174 code clean up and refactoring

Signed-off-by: Oleg <oleg.semeniv@gmail.com>

* - add check for linspase ordered permutations in ShapeUtils::evalShapeForTensorDot

Signed-off-by: Yurii <iuriish@yahoo.com>

* - provide additional code in shape::reshape stuff in order to reduce amount of allocation/copy operations during reshaping procedure

Signed-off-by: Yurii <iuriish@yahoo.com>

* - further work on problem of wrong shape evaluation during permute/reshape procedures

Signed-off-by: Yurii <iuriish@yahoo.com>

* - still looking for bug reason in reshape/permute stuff

Signed-off-by: Yurii <iuriish@yahoo.com>

* - correct bug in transform cuda native ops

Signed-off-by: Yurii <iuriish@yahoo.com>

* - correct bug in NDArray::assign

Signed-off-by: Yurii <iuriish@yahoo.com>

* - remove old shape::reshape stuff

Signed-off-by: Yurii <iuriish@yahoo.com>

* - add possibility to disable copy of old buffer to new buffer during reshape operation in NDArray class

Signed-off-by: Yurii <iuriish@yahoo.com>

* - correct bug in tensorDot which had to do with wrong pointers assigments

Signed-off-by: Yurii <iuriish@yahoo.com>

Co-authored-by: Oleh <oleg.semeniv@gmail.com>
2020-02-13 20:33:54 +03:00
raver119 7783012f39
cuDNN integration (#150)
* initial commit

Signed-off-by: raver119 <raver119@gmail.com>

* one file

Signed-off-by: raver119 <raver119@gmail.com>

* few more includes

Signed-off-by: raver119 <raver119@gmail.com>

* m?

Signed-off-by: raver119 <raver119@gmail.com>

* const

Signed-off-by: raver119 <raver119@gmail.com>

* cudnn linkage in tests

Signed-off-by: raver119 <raver119@gmail.com>

* culibos

Signed-off-by: raver119 <raver119@gmail.com>

* static reminder

Signed-off-by: raver119 <raver119@gmail.com>

* platform engine tag

Signed-off-by: raver119 <raver119@gmail.com>

* HAVE_CUDNN moved to config.h.in

Signed-off-by: raver119 <raver119@gmail.com>

* include

Signed-off-by: raver119 <raver119@gmail.com>

* include

Signed-off-by: raver119 <raver119@gmail.com>

* skip cudnn handle creation if there's not cudnn

Signed-off-by: raver119 <raver119@gmail.com>

* meh

Signed-off-by: raver119 <raver119@gmail.com>

* target device in context

Signed-off-by: raver119 <raver119@gmail.com>

* platform engines

Signed-off-by: raver119 <raver119@gmail.com>

* platform engines

Signed-off-by: raver119 <raver119@gmail.com>

* allow multiple -h args

Signed-off-by: raver119 <raver119@gmail.com>

* allow multiple -h args

Signed-off-by: raver119 <raver119@gmail.com>

* move mkldnn out of CPU block

Signed-off-by: raver119 <raver119@gmail.com>

* link to mkldnn on cuda

Signed-off-by: raver119 <raver119@gmail.com>

* less prints

Signed-off-by: raver119 <raver119@gmail.com>

* minor tweaks

Signed-off-by: raver119 <raver119@gmail.com>

* next step

Signed-off-by: raver119 <raver119@gmail.com>

* conv2d NCHW draft

Signed-off-by: raver119 <raver119@gmail.com>

* conv2d biasAdd

Signed-off-by: raver119 <raver119@gmail.com>

* test for MKL/CUDNN combined use

Signed-off-by: raver119 <raver119@gmail.com>

* - provide additional code for conv2d ff based on cudnn api, not tested yet

Signed-off-by: Yurii <iuriish@yahoo.com>

* - further work on conv2d helper based on using cudnn api

Signed-off-by: Yurii <iuriish@yahoo.com>

* - fixing several cuda bugs which appeared after cudnn lib had been started to use

Signed-off-by: Yurii <iuriish@yahoo.com>

* - implementation of conv2d backprop op based on cudnn api

Signed-off-by: Yurii <iuriish@yahoo.com>

* - implementaion of conv3d and conv3d_bp ops based on cudnn api

Signed-off-by: Yurii <iuriish@yahoo.com>

* - bugs fixing in conv3d/conv3d_bp ops (cudnn in use)

Signed-off-by: Yurii <iuriish@yahoo.com>

* - implementation of depthwiseConv2d (ff/bp) op based on cudnn api

Signed-off-by: Yurii <iuriish@yahoo.com>

* - implementation of batchnorm ff op based on cudnn api

Signed-off-by: Yurii <iuriish@yahoo.com>

* - disable cudnn batchnorm temporary

Signed-off-by: Yurii <iuriish@yahoo.com>

* - add minor change in cmake

Signed-off-by: Yurii <iuriish@yahoo.com>

* engine for depthwise mkldnn

Signed-off-by: raver119 <raver119@gmail.com>

* couple of includes

Signed-off-by: raver119 <raver119@gmail.com>

* - provide permutation to cudnn batchnorm ff when format is NHWC

Signed-off-by: Yurii <iuriish@yahoo.com>

* lgamma fix

Signed-off-by: raver119 <raver119@gmail.com>

* - eliminate memory leak in two tests

Signed-off-by: Yurii <iuriish@yahoo.com>

Co-authored-by: Yurii Shyrma <iuriish@yahoo.com>
2020-01-20 21:32:46 +03:00
Yurii Shyrma 5d9b2a16e5 Shyrma temp (#131)
* - specifying template instantiation for certain types in float16 and bloat16

Signed-off-by: Yurii <iuriish@yahoo.com>

* - polishing bfloat16 and float16 member functions template specialization

Signed-off-by: Yurii <iuriish@yahoo.com>

* - rewrite and overload array +-*/ scalar and scalar +-*/ arr in NDAray class

Signed-off-by: Yurii <iuriish@yahoo.com>

* - make corrections which have to do with and rvalue lvalue conversions

Signed-off-by: Yurii <iuriish@yahoo.com>

* - provide move semantic in NDArray operators array +-/* array

Signed-off-by: Yurii <iuriish@yahoo.com>

* float16/bfloat16 tweaks

Signed-off-by: raver119 <raver119@gmail.com>

* one more tweak

Signed-off-by: raver119 <raver119@gmail.com>

* - make float16 and bfloat16 to compile successfully on cuda

Signed-off-by: Yurii <iuriish@yahoo.com>

* - do not use resources of view-like arrays when move semantics is applied

Signed-off-by: Yurii <iuriish@yahoo.com>

* - get rid of pointers in signatures NDArray methods 1

Signed-off-by: Yurii <iuriish@yahoo.com>

* - correction of signature of NDArray::dup method

Signed-off-by: Yurii <iuriish@yahoo.com>

* - correction of signature of NDArray::reduceAlongDimension method

Signed-off-by: Yurii <iuriish@yahoo.com>

* - signature correction of NDArray::applyIndexReduce and applyTrueBroadcast methods

Signed-off-by: Yurii <iuriish@yahoo.com>

* - signature correction of NDArray::applyReduce3 and varianceAlongDimension methods

Signed-off-by: Yurii <iuriish@yahoo.com>

* - signature correction of NDArray::tensorsAlongDimension and diagonal methods

Signed-off-by: Yurii <iuriish@yahoo.com>

* - signature correction of NDArray::allTensorsAlongDimension

Signed-off-by: Yurii <iuriish@yahoo.com>

* - signature correction of NDArray::reduceAlongDimension 2

Signed-off-by: Yurii <iuriish@yahoo.com>

* - signature correction of NDArray::applyTransform 2

Signed-off-by: Yurii <iuriish@yahoo.com>

* - signature correction of NDArray::applyPairwiseTransform 2

Signed-off-by: Yurii <iuriish@yahoo.com>

* - signature correction of NDArray::applyBroadcast 2

Signed-off-by: Yurii <iuriish@yahoo.com>

* - signature correction of NDArray::applyTrueBroadcast 2

Signed-off-by: Yurii <iuriish@yahoo.com>

* - signature correction of NDArray::applyScalar and applyScalarArr

Signed-off-by: Yurii <iuriish@yahoo.com>

* - signature correction of NDArray::lambda methods

Signed-off-by: Yurii <iuriish@yahoo.com>

* - signature correction of NDArray::reduce3 methods 2

Signed-off-by: Yurii <iuriish@yahoo.com>

* - signature correction of following NDArray methods: add/sub/mul/div row/column and fillAsTriangular

Signed-off-by: Yurii <iuriish@yahoo.com>

* - signature correction of NDArray::tileToShape methods

Signed-off-by: Yurii <iuriish@yahoo.com>

* - signature correction of NDArray::isShapeSameStrict method

Signed-off-by: Yurii <iuriish@yahoo.com>

* minor corrections in tests

Signed-off-by: Yurii <iuriish@yahoo.com>

* - replace reduce op in batchnorm mkldnn

Signed-off-by: Yurii <iuriish@yahoo.com>

* - add explicit templates instantiations for operator+(NDArray&&. const scalar)

Signed-off-by: Yurii <iuriish@yahoo.com>

* - corrections of casts in float16/bfloat16

Signed-off-by: Yurii <iuriish@yahoo.com>

* - provide move semantics in following NDArray methods: transform, applyTrueBroadcast, transpose, reshape, permute

Signed-off-by: Yurii <iuriish@yahoo.com>

* - get rid of input array A duplicate in svd cuda op

Signed-off-by: Yurii <iuriish@yahoo.com>

* - avoid available bug in svd cuda API

Signed-off-by: Yurii <iuriish@yahoo.com>

* - add temporary global memory buffer in svd cuda when calcUV = false and  m != n

Signed-off-by: Yurii <iuriish@yahoo.com>

* - remove test with blfoat16 type for betainC

Signed-off-by: Yurii <iuriish@yahoo.com>

* - resolve conflicts after master has been merged in

Signed-off-by: Yurii <iuriish@yahoo.com>

* - changed type of affected input array in fused_batch_norm

Signed-off-by: Yurii <iuriish@yahoo.com>

* - add several explicit type castings

Signed-off-by: Yurii <iuriish@yahoo.com>

* - add ND4J_EXPORT to operators

Signed-off-by: Yurii <iuriish@yahoo.com>

* - add explicit template types in instantiations of template arithm operators of NDArray class

Signed-off-by: Yurii <iuriish@yahoo.com>

* - one more test fix

Signed-off-by: Yurii <iuriish@yahoo.com>

Co-authored-by: raver119 <raver119@gmail.com>
2019-12-20 22:35:39 +03:00
Yurii Shyrma d19eeaec52 Shyrma casual conv1d (#90)
* - add causal mode of padding to convolutions

Signed-off-by: Yurii <iuriish@yahoo.com>

* - add additional tests for causal conv1d

Signed-off-by: Yurii <iuriish@yahoo.com>

* - add causal mode for cuda conv kernels

Signed-off-by: Yurii <iuriish@yahoo.com>

* Java side of Conv1D changes

Signed-off-by: raver119 <raver119@gmail.com>

* Add Conv1DDerivative op

Signed-off-by: Alex Black <blacka101@gmail.com>

* Causal Conv1D gradient checks

Signed-off-by: Alex Black <blacka101@gmail.com>

* Tweaks

Signed-off-by: Alex Black <blacka101@gmail.com>

* - add causal padding mode to conv2d_bp

Signed-off-by: Yurii <iuriish@yahoo.com>

* More thorough causal conv1d tests

Signed-off-by: Alex Black <blacka101@gmail.com>
2019-11-29 14:14:30 +03:00
AlexDBlack a66e03355e Merge remote-tracking branch 'fork/master' 2019-09-12 12:20:57 +10:00
raver119 98e2814879
Platform helpers (#8216)
* platform helpers draft

Signed-off-by: raver119 <raver119@gmail.com>

* typo

Signed-off-by: raver119 <raver119@gmail.com>

* disable platform cmake

Signed-off-by: raver119 <raver119@gmail.com>

* another draft

Signed-off-by: raver119 <raver119@gmail.com>

* mkldnn convolution refactored

Signed-off-by: raver119 <raver119@gmail.com>

* minor tweaks

Signed-off-by: raver119 <raver119@gmail.com>

* one more safety check

Signed-off-by: raver119 <raver119@gmail.com>

* prototype works

Signed-off-by: raver119 <raver119@gmail.com>

* meh

Signed-off-by: raver119 <raver119@gmail.com>

* force static library mode for mkldnn

Signed-off-by: raver119 <raver119@gmail.com>

* - ismax fix
- experimental arg fix
- don't enforce openblas on Apple hardware

Signed-off-by: raver119 <raver119@gmail.com>

* bunch of small fixes

Signed-off-by: raver119@gmail.com <raver119@gmail.com>

* declare concurrent

Signed-off-by: raver119@gmail.com <raver119@gmail.com>

* - MKLDNN version upgrade to 1.0.2
- avgpool2d/maxpool2d APIs update

Signed-off-by: raver119 <raver119@gmail.com>

* - avgpool2d_bp/maxpool2d_bp APIs update

Signed-off-by: raver119 <raver119@gmail.com>

* - conv2d/batchnorm APIs update

Signed-off-by: raver119 <raver119@gmail.com>

* - lrn/conv2d_bp/conv3d/conv3d_bp APIs update

Signed-off-by: raver119 <raver119@gmail.com>

* all ops converted to MKLDNN 1.x

Signed-off-by: raver119 <raver119@gmail.com>

* bunch of tweaks

Signed-off-by: raver119 <raver119@gmail.com>

* namespace for platform helpers

Signed-off-by: raver119 <raver119@gmail.com>

* make sure platform helpers aren't opimized out

Signed-off-by: raver119 <raver119@gmail.com>

* build cpu_features on x86 systems

Signed-off-by: raver119 <raver119@gmail.com>

* build cpu_features on x86 systems

Signed-off-by: raver119 <raver119@gmail.com>

* more of cpu_features

Signed-off-by: raver119 <raver119@gmail.com>

* - mkldnn removed from java
- cpu_features checks in CpuNDArrayFactory

Signed-off-by: raver119 <raver119@gmail.com>

* F16C definition renamed

Signed-off-by: raver119 <raver119@gmail.com>

* some mkldnn rearrangements

Signed-off-by: raver119 <raver119@gmail.com>

* check supported instructions before doing anything

Signed-off-by: raver119 <raver119@gmail.com>

* typo

Signed-off-by: raver119 <raver119@gmail.com>

* missied impl

Signed-off-by: raver119 <raver119@gmail.com>

* BUILD_PIC option

Signed-off-by: raver119 <raver119@gmail.com>

* conv2d fix

Signed-off-by: raver119 <raver119@gmail.com>

* avgpool3d fix

Signed-off-by: raver119 <raver119@gmail.com>

* avgpool3d_bp fix

Signed-off-by: raver119 <raver119@gmail.com>

* avgpool2d_bp leak fix

Signed-off-by: raver119 <raver119@gmail.com>

* avgpool3d_bp leak fix

Signed-off-by: raver119 <raver119@gmail.com>

* maxpool bp leaks fixed

Signed-off-by: raver119 <raver119@gmail.com>

* printf removed

Signed-off-by: raver119 <raver119@gmail.com>

* batchnorm fix

Signed-off-by: raver119 <raver119@gmail.com>

* AVX warning/error polishing

Signed-off-by: AlexDBlack <blacka101@gmail.com>

* Fix

Signed-off-by: AlexDBlack <blacka101@gmail.com>

* More polish

Signed-off-by: AlexDBlack <blacka101@gmail.com>

* Polish

Signed-off-by: AlexDBlack <blacka101@gmail.com>

* remove previous MKL-DNN support layer

Signed-off-by: raver119 <raver119@gmail.com>

* avx2 tweak

Signed-off-by: raver119 <raver119@gmail.com>

* allow static for apple

Signed-off-by: raver119@gmail.com <raver119@gmail.com>

* exclude mkldnn in one more place

Signed-off-by: raver119 <raver119@gmail.com>

* exclude mkldnn in one more place

Signed-off-by: raver119 <raver119@gmail.com>

* restore OPENBLAS_PATH use

Signed-off-by: raver119 <raver119@gmail.com>

* add runtime check for avx/avx2 support

Signed-off-by: raver119 <raver119@gmail.com>

* convolution_auto

Signed-off-by: raver119 <raver119@gmail.com>

* Add logic for helper argument

* minor test fix

Signed-off-by: raver119 <raver119@gmail.com>

* few tweaks

Signed-off-by: raver119 <raver119@gmail.com>

* few tweaks

Signed-off-by: raver119 <raver119@gmail.com>

* skip OpTracker props for non-x86 builds

Signed-off-by: raver119 <raver119@gmail.com>

* linux arm isn't x86 :)

Signed-off-by: raver119 <raver119@gmail.com>

* avx-512

Signed-off-by: raver119 <raver119@gmail.com>

* CUDA presets fix

Signed-off-by: raver119 <raver119@gmail.com>

* BUILD_PIC

Signed-off-by: raver119 <raver119@gmail.com>

* prefetchw for avx2

Signed-off-by: raver119 <raver119@gmail.com>

* BUILD_PIC again

Signed-off-by: raver119 <raver119@gmail.com>
2019-09-11 21:50:28 +03:00