What's Changed

• Revert ovxlib topk kernel modification by @chenfeiyue-cfy in #694
• Fixed typing error in gather test by @chenfeiyue-cfy in #695
• Update internal ovxlib to rel/1.2.14 by @chenfeiyue-cfy in #699
• Update timvx-overview by @chenfeiyue-cfy in #700
• Update timvx_overview.svg by @chenfeiyue-cfy in #701
• Fixed layernorm and logsoftmax ut error by @chenfeiyue-cfy in #702
• Update internal ovxlib to release/1.2.22 by @chenfeiyue-cfy in #706

Full Changelog: v1.2.6...v1.2.22

What's Changed

• Add NBG runner python binding by @Nullkooland in #677
• Fixed layout infer bug when some op is not in op_vector_ by @chenfeiyue-cfy in #676
• Added opjson and bug fix for scatternd_onnx_v16 by @chenfeiyue-cfy in #678
• add unit test for conv2d by @antkillerfarm in #681
• Added Asymmetric perchannel quantization support by @chenfeiyue-cfy in #682
• add a timeout for TIM-VX check by @zhongzhuonan in #687
• add unit case for maxpool by @antkillerfarm in #683
• Add some trigonometric ops by @antkillerfarm in #689
• Updata internal ovxlib to rel1.2.6 by @chenfeiyue-cfy in #690

Full Changelog: v1.2.2...v1.2.6

What's Changed

• Added scalar type support by @chenfeiyue-cfy in #655
• Rebuild prebuild-sdk to adjust lower ubuntu env by @chenfeiyue-cfy in #658
• Refine UnitTest which have acc issue or un-support issue for SP by @chenfeiyue-cfy in #659
• Refine RNNCell/HardSwish/Reduce_sum ut by @chenfeiyue-cfy in #660
• add a case for resize bilinear by @chxin66 in #662
• Fix the instance norm test input size bug in layout infer test by @chenfeiyue-cfy in #661
• Added 2 cases for stack by @chenfeiyue-cfy in #664
• fix slope shpae 1 crash issue by @chxin66 in #663
• fix crash when eletwise inputs are different rank by @chxin66 in #665
• fix const tensor align bug in AlignPermuteVectorForElementWise by @chxin66 in #666
• fix layoutinfer crash when logical op inputs are different rank by @chxin66 in #667
• Add uid() api for class operation by @zhengzhouheng in #668
• Refine layout inference by @Nullkooland in #671
• Add json third party when support node trace db by @zhengzhouheng in #670
• Fixed typing error and added missed opheader by @chenfeiyue-cfy in #673
• Add license declaration of third party by @zhengzhouheng in #672
• Update internal ovxlib to rel/1.2.2 by @chenfeiyue-cfy in #674
• Update ops which has addings in internal by @chenfeiyue-cfy in #675
• Fixed unreasonable type of parameter in broadcast by @chenfeiyue-cfy in #505

Full Changelog: v1.1.88...v1.2.2

What's Changed

• Fixed multi_device compiling error in gcc 12 by @chenfeiyue-cfy in #501
• add custom op in lenet test by @zhengzhouheng in #504
• Fixed roi_align golden mismatch error by @chxin66 in #509
• Added minimum unit test by @chxin66 in #449
• Update x86_64_linux/include for 22Q3 by @chenfeiyue-cfy in #506
• Set RNN internal dtype by @xuke537 in #511
• Fixed BidirectionalSequenceLSTM bug by @chenfeiyue-cfy in #514
• add rnn by @MESeraph in #513
• Fixed average pool layout infer by @chxin66 in #508
• diabled a failed case in conv2d by @chxin66 in #517
• fix typo for graph_test.cc by @antkillerfarm in #516
• Add BUILD_WITH_BAZEL option, marco of VSI_FEAT_OP_XXX should behind h… by @MercuryChen in #518
• Added layout_inference for UnidirectionalRnn by @chenfeiyue-cfy in #519
• Fixed BidirectionalSequenceRnn bugs by @chenfeiyue-cfy in #520
• Added HashtableLookup Op by @chenfeiyue-cfy in #524
• Modified README.md about rnn&lstm by @chenfeiyue-cfy in #522
• Added invalidate handle marco by @chenfeiyue-cfy in #525
• Fix maxpoolgrad by @MercuryChen in #526
• Added EmbeddingLookup op by @chenfeiyue-cfy in #523
• Added pad_v2 & pad_v2 layout infer & unit_test by @chxin66 in #533
• fixed groupconv2d params in internal by @chenfeiyue-cfy in #535
• Fix groupconv both in TIM-VX by @chenfeiyue-cfy in #532
• Added EmbeddingLookup 4d support internal by @chenfeiyue-cfy in #534
• Fixed (grouped)conv2d layout infer bug by @chxin66 in #527
• Fixed bug for pad_v2 by @chxin66 in #540
• Added a case for fc by @chxin66 in #537
• mapped pool1d by @MESeraph in #536
• Fixed wrong type of hashtable_lookup unit test by @chenfeiyue-cfy in #538
• Added EmbeddingLookup op 4d lut unit test by @chenfeiyue-cfy in #531
• Added reduce_all layoutinfer & a reduce_all case by @chxin66 in #543
• Fixed Invalid read of size 8 reported by Valgrind by @HuanyuCai in #545
• fix depth2space mode enum by @MESeraph in #547
• update copyright information by @antkillerfarm in #549
• Fixed grouped_conv2d layout infer & Added cases by @chxin66 in #546
• complete custom op readme by @zhengzhouheng in #552
• Refine Depth2Space op by @chenfeiyue-cfy in #553
• Introduce CMAKE option TIM_VX_DBG_ENABLE_TENSOR_HNDL=ON by @sunshinemyson in #550
• Fixed deconv2d layout infer bug by @chxin66 in #557
• Added case for hardswish by @chxin66 in #560
• Fix mirror pad param mismatch by @zhengzhouheng in #568
• Vts inputs order bug by @chxin66 in #567
• remove tensor GetDataRef api by @zhengzhouheng in #569
• Fixed l2normalization layout infer bug by @chxin66 in #570
• Support remote platform by gRPC by @liyuenan2333 in #561
• Modified VSILOGE to VSILOGW when check consumed tensor by @chxin66 in #572
• Update internal & prebuilt-sdk for 23Q1 release by @chenfeiyue-cfy in #573
• Share const tensor in graph by @chxin66 in #574
• Update readme in ops by @chenfeiyue-cfy in #575
• Refine prelu layout inference by @chenfeiyue-cfy in #577
• Fixed ovxlib build error on android by @chxin66 in #578
• Add some tensor dtype convert APIs by @SCUWQ in #576
• Support build openssl from local repo by @chxin66 in #582
• Reload "==" operator for quantizations of two tensor by @chenfeiyue-cfy in #583
• Refine UnidirectionalGRU and GRUCell by @chenfeiyue-cfy in #587
• Added missed ops include header by @chenfeiyue-cfy in #584
• Fixed bug for pad test by @chxin66 in #588
• Add a case for local response norm by @chxin66 in #590
• Fixed LayerNormalization eps bug by @shijie-nv in #589
• optimization for tiny_yolov4 by @chxin66 in #591
• fixed instance norm layout bug & add its layoutinfer by @chxin66 in #593
• Added api json for each op to support acuity by @chenfeiyue-cfy in #596
• Added a case for resize_bilinear layoutinfer by @chxin66 in #595
• Added axis support for layernorm by @chenfeiyue-cfy in #602
• fixed reduce layoutinfer bug by @chxin66 in #605
• Fixed openssl android build bug by @chxin66 in #606
• Remove unused value in op layoutinfer by @chenfeiyue-cfy in #607
• Added new_axis_mask param for stridedslice by @chenfeiyue-cfy in #600
• Fixed viplite driver build issue by @chxin66 in #611
• Added case for gather by @chxin66 in #599
• Enable float16 bias convolution model runs on NN by @chenfeiyue-cfy in #612
• Optimize compilation process for openssl by @chxin66 in #613
• fixed yolov4 build issue by @chxin66 in #618
• Update prebuilt && internal for 23Q2 release by @chenfeiyue-cfy in #617
• Added roi_align layoutinfer & cases by @chxin66 in #615
• Fixed transient deconv1d generate wrong output shape bug by @chenfeiyue-cfy in #619
• Added axis param for TopK by @chenfeiyue-cfy in #610
• Remove confusing comment in depthwise conv test by @chenfeiyue-cfy in #621
• Integrate api trace into tim-vx source as an experimental feature. by @MercuryChen in #623
• Fixed conv2d grouped_conv2d deconv2d layoutinfer bug by @chxin66 in #622
• Fixed IExecutable object not bind with DeviceID by @sunshinemyson in #624
• Create self-hosted.yml by @zhongzhuonan in #625
• fixed prelu layoutinfer bug & added cases by @chxin66 in #628
• fix size compute bug in lrn by @chxin66 in #626
• Added op fusion for mean_stddev_normalization by @chenfeiyue-cfy in #629
• Added general Float16 support by @chenfeiyue-cfy in #631
• Code refinement for mean-stddev-normalization fuse by @chenfeiyue-cfy in #632
• Refine api trace code and document by @MercuryChen in #634
• Fixed misleading test case bug in deconv1d by @chenfeiyue-cfy in #633
• Fix data missing when use trace::Graph::CreateTensor by @xie-oritek in #636
• Add CumSum&LRN operator to trace module by @xie-oritek in #639
• Add int4/uint4 definition by @xie-oritek in #638
• Split replayer code from tracer.h by @MercuryChen in #642
• Refine Tensor::SetShape api to avoid compile warning using const ref by @xie-oritek in #640
• export Swap Handle API by @antkillerfarm in #635
• Fix the instance norm test input size bug by @zhengzhouheng in #645
• fixe...

What's Changed

• Added broadcast layout infernece by @chxin66 in #438
• Added cases for reduce sum by @chxin66 in #441
• Rename RoiAlign & RoiPool by @antkillerfarm in #446
• add maxpoolwithargmax2 and maxpoolgrad by @MercuryChen in #444
• Fixed quantize param in reduce_sum by @chxin66 in #445
• update nbg format version by @zhengzhouheng in #440
• update Operators.md by @antkillerfarm in #447
• Add ut configuration for cl only device by @sunshinemyson in #448
• Fixed param compute bug for lrn by @chxin66 in #451
• Fix the build error for clang when export TIM_VX_ENABLE_PLATFORM=ON by @liyuenan2333 in #453
• add readme for ovxlib_bin_build.sh by @antkillerfarm in #452
• Added div int32 unit test by @chxin66 in #455
• Mapped GRUCell & unit test by @chxin66 in #457
• include Topk op's header file by @MercuryChen in #460
• Set graph attributes when compile graph to binary by @xuke537 in #459
• Mapped bidirectional lstm & unit test by @chxin66 in #461
• Update Version to 1.1.50 by @sunshinemyson in #462
• Modified Div_int unit test golden by @chxin66 in #466
• Modified bidirectional_sequence_lstm golden accuracy by @chxin66 in #469
• Mapped unidirectional gru & unit test by @chxin66 in #470
• Feat: disable maxpoolwithargmax2 if no low-level support by @sunshinemyson in #471
• disabled two not supported cases by @chxin66 in #473
• Fix error in feature compatible guard by @sunshinemyson in #472
• Added conv3d unit test by @chxin66 in #468
• disabled two Div cases by @chxin66 in #476
• Update OpenCV usage link by @sunshinemyson in #477
• Disabled a conv3d case by @chxin66 in #479
• fixed some errs on gcc12 by @chenfeiyue-cfy in #483
• supported int16 dfp quantization & added conv2d unit test by @chenfeiyue-cfy in #478
• Replace name direct_map_op with builtin_op by @chxin66 in #481
• added Mod op & Mod unit test by @chenfeiyue-cfy in #484
• added sign & softsign by @chenfeiyue-cfy in #486
• added Rcp op & modified test_utils by @chenfeiyue-cfy in #487
• added MaxPool3d op by @chenfeiyue-cfy in #490
• added cumsum op & added OnBindInputPostProc func by @chenfeiyue-cfy in #489
• Supported composed layout infer & added unit test by @chxin66 in #488
• Added two reduce layout infer unittest by @chxin66 in #491
• Fixed bug when input's index is not 0 by @chxin66 in #493
• Added two cases in strided_slice by @chxin66 in #494
• added transpose_test from https://github.com/VeriSilicon/TIM-VX/issue… by @chenfeiyue-cfy in #495
• fixed bug when broadcast dimensions is negative by @chxin66 in #496
• Fixed tensorflow version in CI by @chenfeiyue-cfy in #499
• Update internal & prebuilt-sdk for 22Q3 release by @chenfeiyue-cfy in #500

New Contributors

• @MercuryChen made their first contribution in #444

Full Changelog: v1.1.50...v1.1.57

What's Changed

• Added param "step" for slice & added unit test by @chxin66 in #352
• Fixed compiler fail for elu by @chxin66 in #358
• update ovxlib virtual_device patch by @lileiigithub in #357
• Supported specifying alpha and beta by @chxin66 in #356
• Fixed layout inference bug for stride_slice by @chxin66 in #329
• refine tim_internal.cmake for ovxlib vip by @lileiigithub in #360
• Added unit test for maxpool by @chxin66 in #361
• Suported specifying CRD_mode & DCR_mode in depthtospace by @chxin66 in #362
• Support specifying pad_mode in pad by @chxin66 in #355
• add BroadcastInDim to internal expand_broadcast op by @antkillerfarm in #364
• Added selu & celu & unit test by @chxin66 in #366
• Add Broadcast op by @antkillerfarm in #365
• Update operator support plan by @sunshinemyson in #367
• Fixed pad layout inference bug by @chxin66 in #370
• CI enhancement - enable benchmark_model and samples by @sunshinemyson in #372
• rename CopyTensorToData to CopyDataFromTensor to align name of tim::v… by @antkillerfarm in #373
• add macro VSI_EXPAND_BROADCAST_ENABLE_DIMENSIONS for ovxlib compatibi… by @antkillerfarm in #374
• add test demo for multi_device by @lileiigithub in #371
• Fix ci crash by @sunshinemyson in #380
• fix bug of param num in custom op by @zhengzhouheng in #385
• Added topk & unit test by @chxin66 in #384
• Added Ceil & unit test by @MESeraph in #381
• Fixed layout inference bug for stack by @chxin66 in #375
• add macro VSI_EXPAND_BROADCAST_ENABLE_DIMENSIONS for unit test compat… by @antkillerfarm in #386
• fix gather_element operation input num issue by @chxin66 in #388
• Added gather_elements & unit test by @chxin66 in #363
• add GetElementNum/GetElementByteSize/GetByteSize for TensorSpec by @antkillerfarm in #393
• Fixed no-output if transpose is last op and can be optimized by @sunshinemyson in #395
• Fix build issue by @sunshinemyson in #397
• feat(tensor): support external buffer when creating input/output tensors by @gdh1995 in #389
• Mapped roi_align & added unit test by @chxin66 in #402
• modify GatherElements by @MESeraph in #406
• Added unidirectional lstm layout inference by @chxin66 in #392
• Mapped roi_pool & added unit test by @MESeraph in #404
• Update tensorflow to v2.9.0 in ci by @sunshinemyson in #403
• add reshape unit test by @antkillerfarm in #416
• remove redefinition of TIM_VX_ENABLE_CUSTOM_OP by @antkillerfarm in #417
• Added grouped conv2d layout inference by @chxin66 in #419
• disabled two failed case by @chxin66 in #422
• Enable SetRoundingPolicy by @liyuenan2333 in #426
• Disabled 3 failed case by @chxin66 in #428
• Fixed transpose layout inference bug by @chxin66 in #430
• Added batch dims in gather by @chxin66 in #435
• Update internal for 22Q2 release by @liyuenan2333 in #432

New Contributors

• @MESeraph made their first contribution in #381

Full Changelog: v1.1.42...v1.1.50

What's Changed

• add alpha & beta parameters for HardSigmoid by @antkillerfarm in #265
• Support that op's all inputs in layout inference are constant by @liyuenan2333 in #264
• Disable fast mode of graph by @onepick in #267
• Update component diagram and README.md by @sunshinemyson in #269
• Support NPU access large memory > 4G by @sunshinemyson in #280
• Fix build error with gcc 6.2.0 by @sunshinemyson in #282
• Enabled bulding with buildroot toolchain. by @SHagerGEL in #281
• [New API] Add compile_option support - relax_mode by @sunshinemyson in #285
• fix compile error in g++5.4 by @yingshengBD in #286
• Install headers to place defined by CMAKE_INSTALL_INCLUDEDIR variable by @robert-kalmar in #291
• enable no bias in FC layout inference by @liyuenan2333 in #294
• Fixed pad bug for grouped_conv1d by @chxin66 in #292
• Added unit test for stack by @chxin66 in #298
• Relax tolerance for div_uint8 case by @sunshinemyson in #303
• Update reshape to reshape2 by @sunshinemyson in #310
• add custom base op and tests by @zhengzhouheng in #315
• Added unit test for batch2space and space2batch by @xuke537 in #321
• fix some comments of Mish and LRN layer by @gdh1995 in #322
• Add document for customized operator by @sunshinemyson in #323
• Fix build warn/error with clang by @sunshinemyson in #326
• Refine customized op support by @sunshinemyson in #327
• Add ArgMax/ArgMin unit tests by @xuke537 in #333
• add cmake option of custom op support by @zhengzhouheng in #335
• OpenCV offical announcement with TIM-VX support by @sunshinemyson in #341
• Add layout inference & layout test for stack by @chxin66 in #337
• support multi virtual devices by @lileiigithub in #331
• Support specifying alpha in elu by @chxin66 in #354

New Contributors

• @SHagerGEL made their first contribution in #281
• @yingshengBD made their first contribution in #286
• @gdh1995 made their first contribution in #322

Full Changelog: v1.1.37...v1.1.42

• Update x86_64 simulation to v6.4.9

• Added Conv3D API inside lowlevel driver without dilation support.

• Fixed strided slice - now work with 5D input

Update to v1.1.34

Update SDK to 6.4.8 release
Update documentation and Op support list
Add utility function and samples for parsing NBG
TVM and TFlite external delegate Support

a broken tag