* - 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>
* working prototype of new CUDA build with cmake
Signed-off-by: raver119 <raver119@gmail.com>
* get rid of older stuff
Signed-off-by: raver119 <raver119@gmail.com>
* remove legacy CUDA debug section
Signed-off-by: raver119 <raver119@gmail.com>
* fPIC for GCC
Signed-off-by: raver119 <raver119@gmail.com>
* - switch to /MD
- make MSVC runtime lib configurable from 1 place
Signed-off-by: raver119 <raver119@gmail.com>
* few last tweaks
Signed-off-by: raver119 <raver119@gmail.com>
* mae static library optional
Signed-off-by: raver119 <raver119@gmail.com>
* typo fixed
Signed-off-by: raver119 <raver119@gmail.com>
* cleaned up bert iterator tests (#110)
Signed-off-by: eraly <susan.eraly@gmail.com>
* Various pre-release fixes (#111)
* Various fixes
Signed-off-by: AlexDBlack <blacka101@gmail.com>
* Fix default dtypes for MaxPoolWithArgmax
Signed-off-by: AlexDBlack <blacka101@gmail.com>
* Small pre-release tweak (#112)
* Log UI address on launch as in previous Play-based UI
Signed-off-by: AlexDBlack <blacka101@gmail.com>
* Logging level tweak for UI
Signed-off-by: AlexDBlack <blacka101@gmail.com>
* http not https
Signed-off-by: AlexDBlack <blacka101@gmail.com>
* datavec python ensure host (#113)
* ensure host
* one more host ensure
* info->debug
* [WIP] reverse improvements (#115)
* initial commit
Signed-off-by: raver119 <raver119@gmail.com>
* reverse draft
Signed-off-by: raver119 <raver119@gmail.com>
* reverse kernel
Signed-off-by: raver119 <raver119@gmail.com>
* reverse kernel
Signed-off-by: raver119 <raver119@gmail.com>
* 2 micro fixes
Signed-off-by: raver119 <raver119@gmail.com>
* Shugeo resize fix5 (#102)
* Refactored resize images ops to use TF-like bool args as input.
* Refactored helpers for cpu implementation of resize_bilinear and resize_nearest_neighbor ops.
* Refactored cuda implementation for image.resize_bilinear and image.resize_nearest_neighbor ops helpers.
* Refactored nearest_neighbor resize op.
* Added a pair of tests for special case of resize_bilinear algorithm.
* Fixed issue with resize_bilinear op.
* Refactored cpu implementation for helpers with resize_nearest_neighbor op.
* Final fixed for resize ops to conform TF v.1.5
* Refactored cuda helpers for resize_neares_neighbor op.
* Fixed resize_bilinear to accept proper data.
* Fixed issue with non-float input for resize_bilinear op.
* Refactored cuda helper for resize_bilinear to proper process non-float inputs.
* Added tests for resize_bilinear to int inputs.
* Fixed ResizeBilinear wrapper
* Tests fixed
* Fixed float and bool constant to avoid overflow for some kind of compilers.
* Corrected float constants with float data type.
* Added f suffix for float constants.
* Corrected float constant to avoid overflow with initializing lists.
* Corrected float initializing list with float input.
* Corrected bool constant with initalizing list.
* Corrected float and bool values with initializing lists.
* Fixed wrong constant.
* Fixed issue with 1x1 input picture for resize.
* ResizeBilinear default values on import fix
Signed-off-by: raver119 <raver119@gmail.com>
* Implementation for non_max_suppression_v3 was added. Initial version
* Added check for overcome threshold.
* Added definition for V3 method.
* java remapping for NonMaxSuppressionV3
Signed-off-by: raver119 <raver119@gmail.com>
* Fixed proporly processing of an empty output and test.
* Refactored op to less threshold data to float.
* Implemented cuda-based helper for non_max_suppression_v3 op.
* Fixed fake_quant_with_min_max_vars op.
* Fixed tests with float numbers.
* - assert now stops execution
- sortByKey/sortByValue now have input validation
Signed-off-by: raver119 <raver119@gmail.com>
* missing var
Signed-off-by: raver119 <raver119@gmail.com>
* Fixed proper processing for zero max_size inputs.
* Refactored kernel callers.
* Fixed return statement for logdet op helper.
* Refactored unsorted segment SqrtN op.
* get back 8 tail bytes on CUDA
Signed-off-by: raver119 <raver119@gmail.com>
* Refactored segment prod ops and helpers for cuda and tests.
* Additional test.
* CudaWorkspace tests updated for 8 tail bytes
Signed-off-by: raver119 <raver119@gmail.com>
* special atomic test
Signed-off-by: raver119 <raver119@gmail.com>
* atomicMul/atomicDiv fix for 16bit values
Signed-off-by: raver119 <raver119@gmail.com>
* Eliminated waste prints.
* - get rid of some copy procedures in mmulHelper ops
Signed-off-by: Yurii <iuriish@yahoo.com>
* - further work on embedding cuda api for batched gemm (cublasGemmBatchedEx) in our mmulHelper class
Signed-off-by: Yurii <iuriish@yahoo.com>
* - further work on cuda batched gamm api
Signed-off-by: Yurii <iuriish@yahoo.com>
* - write own cuda kernel performing batched gemm
Signed-off-by: Yurii <iuriish@yahoo.com>
* missing include in MmulHelper
Signed-off-by: raver119 <raver119@gmail.com>
* - forgot to keep in code previous correct kernels for mmulNxN, since it may happen that new onw will fail for some reason in future
Signed-off-by: Yurii <iuriish@yahoo.com>
* disable old tensordot
Signed-off-by: raver119 <raver119@gmail.com>
* - rewrite cuda kernels for usualGemm and usualGemv
Signed-off-by: Yurii <iuriish@yahoo.com>
* - profiling mmul helpers
Signed-off-by: Yurii <iuriish@yahoo.com>
* - prints to check shapes were added
Signed-off-by: Yurii <iuriish@yahoo.com>
* - correct type of output array Cin mmulNxN
Signed-off-by: Yurii <iuriish@yahoo.com>
* - take into account possible nans in C array
Signed-off-by: Yurii <iuriish@yahoo.com>
* slightly change numThreads message
Signed-off-by: raver119 <raver119@gmail.com>
* - make corrections in accordance to given notes in pr review
Signed-off-by: Yurii <iuriish@yahoo.com>
* - platform helpers can be disabled on per-op basis now via Context::allowHelpers
- java has access to it as well
Signed-off-by: raver119 <raver119@gmail.com>
* global platform-helpers trigger
Signed-off-by: raver119 <raver119@gmail.com>
* few signatures renamed
Signed-off-by: raver119 <raver119@gmail.com>
* - few new env variables to follow
- maxThreads/masterThreads differentiation
Signed-off-by: raver119 <raver119@gmail.com>
* Javadoc update
Signed-off-by: raver119 <raver119@gmail.com>
* Corrected randomuniform declaration.
* Refactored uniform distribution for both cuda and cpu platforms.
* Refactored uniform distribution and tests.
* Fixed type usage with indices.
* Refactored uniform distribution implementation and tests to full conform with TF implementation.
* Refactored gamma function to use type util method.
* Copyright changes and fixes with ConstantHelper.
* Added error checking on allocate cuda device memory and operations.
* - provide correct call NDArray::applyBroadcast inside of NDArray::applyTrueBroadcast
Signed-off-by: Yurii <yurii@skymind.io>
* - provide new trueBroadcast helper
Signed-off-by: Yurii <yurii@skymind.io>
* example for yurii
Signed-off-by: raver119 <raver119@gmail.com>
* - provide new trueBroadcast helper for cpu
Signed-off-by: Yurii <yurii@skymind.io>
* - start working on new trueBroadcat helper for cuda
Signed-off-by: Yurii <yurii@skymind.io>
* - further work on trueBroadcast for cuda
Signed-off-by: Yurii <yurii@skymind.io>
* - fix bugs in cuda helper trueBroadcast
Signed-off-by: Yurii <yurii@skymind.io>
* - profiling bias_add op
- add some docementation
Signed-off-by: Yurii <yurii@skymind.io>
* - minor change
Signed-off-by: Yurii <yurii@skymind.io>
* - provide addBias cuda kernel
Signed-off-by: Yurii <yurii@skymind.io>
* - improve shape::getIndexOfffset and change its signature
Signed-off-by: Yurii <yurii@skymind.io>
* - same as previous
Signed-off-by: Yurii <yurii@skymind.io>
* - improve and change signature in some shape:: stuff which has to do with calculation of offsets for array elements
Signed-off-by: Yurii <yurii@skymind.io>
* - minor changes in flatten
Signed-off-by: Yurii <shyrma@skymind.io>
* - add function shape::getIndexOffsetOrdered
Signed-off-by: Yurii <shyrma@skymind.io>
* - correct shape::getIndexOffsetOrdered()
Signed-off-by: Yurii <shyrma@skymind.io>
* - move getIndexOffsetOrdered to flatten.h header in order to isolate this function
Signed-off-by: Yurii <shyrma@skymind.io>
* Rename flatbuffers DataType enum to DType
Signed-off-by: Alex Black <blacka101@gmail.com>
* Rename flatbuffers DataType enum to DType
Signed-off-by: Alex Black <blacka101@gmail.com>
* Updates for flatbuffers datatype enum renaming
Signed-off-by: Alex Black <blacka101@gmail.com>
* one test for alex
Signed-off-by: raver119 <raver119@gmail.com>
* fix
Signed-off-by: raver119 <raver119@gmail.com>
* get rid of safety offset in cpp
Signed-off-by: raver119 <raver119@gmail.com>
* bfloat16
Signed-off-by: raver119 <raver119@gmail.com>
* minor test rearrangement to fastpath launch
Signed-off-by: raver119 <raver119@gmail.com>
* - atomicAdd/Mul/Div fix for float16/bfloat16 misalignment
- one special test for maxpoolbp java
- safety offset of 8 bytes is back to libnd4j legacy
Signed-off-by: raver119 <raver119@gmail.com>
* one noop test
Signed-off-by: raver119 <raver119@gmail.com>
* skip input validation for no-input ops
Signed-off-by: raver119 <raver119@gmail.com>
* - one more noop empty test
- one more validation before sync
Signed-off-by: raver119 <raver119@gmail.com>
* typo
Signed-off-by: raver119 <raver119@gmail.com>
* one more validation fix
Signed-off-by: raver119 <raver119@gmail.com>
* CUDA empty reductions java side
Signed-off-by: raver119 <raver119@gmail.com>
* one svd test
Signed-off-by: raver119 <raver119@gmail.com>
* Corrected segment_mean helpers and added another test.
* Refactored segment_mean kernels to avoid race_condition.
* CUDA empty reduction
Signed-off-by: raver119 <raver119@gmail.com>
* - listdiff synchronization fix for CUDA
- listdiff test
Signed-off-by: raver119 <raver119@gmail.com>
* - IndexReduce ops now allow INDEXING_TYPES output
- topK op accepts only INDEXING_TYPES as output
Signed-off-by: raver119 <raver119@gmail.com>
* one test for maxpool2d_bp
Signed-off-by: raver119 <raver119@gmail.com>
* - maxpool2d_bp cuda fix for NaNs
- streamSync after each custom op execution
Signed-off-by: raver119 <raver119@gmail.com>
* - start working on space_to_batch_nd
Signed-off-by: Yurii <yurii@skymind.io>
* - provide cpu helper for space_to_batch_nd op
Signed-off-by: Yurii <yurii@skymind.io>
* few typos fixed
Signed-off-by: raver119 <raver119@gmail.com>
* - add tests for space_to_batch and correct bugs
Signed-off-by: Yurii <yurii@skymind.io>
* - write cuda kernel for space_to_batch op
Signed-off-by: Yurii <yurii@skymind.io>
* - add order argument to shape::index2coords method in convolution cuda ops
Signed-off-by: Yurii <yurii@skymind.io>
* - restore some previous code
Signed-off-by: Yurii <yurii@skymind.io>
* old col2im kernel activated
Signed-off-by: raver119 <raver119@gmail.com>
* - change coords calculation in col2im kernel
Signed-off-by: Yurii <yurii@skymind.io>
* - restore old col2im kernel
Signed-off-by: Yurii <yurii@skymind.io>
* - add custom op for batch_to_space
Signed-off-by: Yurii <yurii@skymind.io>
* - provide cpu version for batch_to_space_nd op
Signed-off-by: Yurii <yurii@skymind.io>
* - provide cuda kernel for batch_to_space_nd op
Signed-off-by: Yurii <yurii@skymind.io>
raver119
Yurii Shyrma
shugeo
Alexander Stoyakin
Alex Black