v0.9.0

Feature

• Feature: Update toolkit to 3.6.1. #795
• Feature: add syntax highlight. #799
• Feature: Update binary-ops to 1.20.0rc1 with toolkit3.6.1. #802
• Feature: new op roi_pooling_forward. #806
• Feature: Update binary ops to v1.20.0. #807
• Feature: new op roi_pooling_backward. #819
• Feature: add coverage test doc and revise env descriptions doc. #817
• Feature: Update binary-ops to v1.20.2 and fix docs. #824
• Feature: Update binary ops to v1.20.3temp. #832
• Feature: nms_rotated fix code of iou_threshold<0 cases. https://github## What's Changed
• Feature: Update toolkit to 3.6.1. #795
• Feature: add syntax highlight. #799
• Feature: Update binary-ops to 1.20.0rc1 with toolkit3.6.1. #802
• Feature: new op roi_pooling_forward. #806
• Feature: Update binary ops to v1.20.0. #807
• Feature: new op roi_pooling_backward. #819
• Feature: add coverage test doc and revise env descriptions doc. #817
• Feature: Update binary-ops to v1.20.2 and fix docs. #824
• Feature: Update binary ops to v1.20.3temp. #832
• Feature: nms_rotated fix code of iou_threshold<0 cases. #822
• Feature: Update binary ops to 1.20.3. #834
• Feature: perf voxelization. #815
• Feature: upgrade to toolkit3.7.0. #840
• Feature: add Sybn op. #844
• Feature: roiaware_pool3d add max_pts_each_voxel paramcheck by in #847
• Feature: Roiaware_pool3d specify 'paramcheck' by in #848
• Feature: add maximum input size check for box_iou_rotated. #850
• Feature: Fix bugs of ms_deform_attn_forward and align na…. #852
• Feature: add strided_slice binary operator. #849
• Feature: add transform binary oprator. #839
• Feature: use fill_v3 instead of fill_zeros by in #858
• Feature: Update binary ops to v1.21.0rc2. #860

Bug Fixes

• Fix: revise sample compile shell script for local compiling user friendly. #794
• Fix: do not remove build/test dir when compiling. #796
• Fix: add scale limitation in mlu_op.h. #801
• Fix: fix warnings when compiling nms_rotated by in #814
• Fix: add large tensor num check for psamask. #818
• Fix: clear build/mlu_op_gtest when build. #816
• Fix: perf voxel_pooling_forward and fix bug. #829
• Fix: revise sample compile shell script for local compiling user friendly. #794
• Fix: do not remove build/test dir when compiling. #796
• Fix: add scale limitation in mlu_op.h. #801
• Fix: fix warnings when compiling nms_rotated by in #814
• Fix: add large tensor num check for psamask. #818
• Fix: clear build/mlu_op_gtest when build. #816
• Fix: perf voxel_pooling_forward and fix bug. #829
• Fix: check kernel launch and specify header file by in #813
• Fix: roiaware add Param Check and use _cn_scalar_sin* by in #838
• Fix: roiaware add annotation and modify docs by in #842
• Fix: fix polynms doc's descriptor. #855

Documentations

• Docs: update design_docs and development-Process by in #800
• Docs: Fix a bug of docs. #809
• Docs: Update toolkit description in docs. #811
• Docs: update pr_template. #821
• Docs: fix env var guide link error. #831
• Docs: Fix docs for v0.8.1. #826
• Docs: update design_docs and development-Process by in #800
• Docs: Fix a bug of docs. #809
• Docs: Update toolkit description in docs. #811
• Docs: update pr_template. #821
• Docs: fix env var guide link error. #831
• Docs: Fix docs for v0.8.1. #826
• Docs: Add mlu-ops BANGC-OPS-OpList.md. #830
• Docs: update README.md. #828
• Docs: rm some words, such as ipu. #836
• Docs: modify the documentation comments of mlu_op.h. #837
• Docs: Update performance acceptance standard. #820
• Docs: Update accuracy standard. #823
• Docs: modify design docs. #843
• Docs: update release-note. #859
• Docs: update installer version message. #846

Test

• TEST: add border_align api test by @wangrt1 in #804

Contributors

@dcmee336 @yuer-cn @TingyuZhangSie @duzekunKTH @PetrelYy @mahxn0 @Unireverse @yuexiaof @DanieeelLiu @chqy99 @xwulin @zhengleiZL @baicaixmj

Full Changelog: v0.8.1...v0.9.0

v0.8.1

Feature

• Feature: Update binary-ops to v1.20.2 and fix docs. #825
• Feature: Update binary ops to v1.20.3temp. #833
• Feature: Update binary ops to 1.20.3. #835

Documentations

• Docs: Fix a bug of docs. #810
• Docs: Update toolkit description in docs. #812
• Docs: Fix docs for v0.8.1. #827

Contributors

@Unireverse

Full Changelog: v0.8.0...v0.8.1

v0.8.0

Features

• Feature:add new op masked_col2im_forward #649
• Feature:add new op tin_shift forward and backward #666
• Feature: fix yolo_box doc #696
• Feature: box_iou_rotated support nan and inf. #701
• Feature: replace math.h by cmath #717
• Feature:add new op masked_im2col_forward #659
• Feature:add file name format check #726
• Feature:prior_box support nan/inf. #730
• Feature: Add border_align op #677
• Feature: Add concurrent build jobs #732
• Feature: update FAQ in user guide #740
• Feature:voxelization support nan/inf. #738
• Feature:bbox_overlaps support nan/inf test #739
• Feature: nms_rotated boxes support naninf #742
• Feature: rois and offset support nan inf #746
• Feature:voxelization support voxels largetensor. #751
• Feature: recorrect tin_shift api guide #754
• Feature:bbox_overlaps opt #735
• Feature: Add bang memcheck #757
• Feature:Correct the scale limitation #759
• Feature: add fool-check for tensor size #758
• Feature:generate_proposals_v2 support nan/inf cases. #766
• Feature:Modify scale limit from 0.5G num to 2G num #774
• Feature: Redefine sizeof in mlu file #789
• Feature: Update toolkit to 3.6.1. #797
• Feature: update binary ops to v1.20.0 #808

Bug Fixes

• Fix: correct release docs for v0.7.0 #691
• Fix: Delete masked_col2im_forward docs information #695
• Fix: update clang version for coverage test. #692
• Fix: Fix binary ops memory leak. #697
• Fix: fix mutual_information doc #700
• Fix: fix focal_loss_sigmoid_backward gamma range precheck #705
• Fix: change param name gard_feats to grad_feats #702
• Fix: add ms_deform_attn_forward/backward into the real d… #704
• Fix: support grad_voxel_feats zero element & fix constra… #714
• Fix: change dynamic_point_to_voxel_backward mlu file‘s name #716
• Fix: Fix binary ops memory leak. #721
• Fix: fix memory leak #724
• Fix: fix add_n in uesr_guide #736
• Fix: fix api gtest #737
• Fix: update ci code format check. #744
• Fix: fix check_log_error space error. #752
• Fix: fix focal_loss log #753
• Fix:voxelization zero tensor bug fix. #755
• Fix:ms_deform_attn_backward memcheck warning fix. #761
• Fix: fix device memory leak of box_iou_rotated. #760
• Fix:fix mmcheck warning for nms_rotated #762
• Fix: fix focal_loss nan,inf bug #728
• Fix: fix interface-note #767
• Fix:fix missing return value of interface #772
• Fix:Repair interface description error. #773
• Fix: fix roipointpool3d interface description error. #775
• Fix: delete unused api return message on indice_convolution_backward_data api #777
• Fix: fix roipoint_pool3d device memcheck buffer overflow. #776
• Fix:Repair the problem of gdram memory reading and writing coverage. #778
• Fix: fix focalLossSigmoidBackward buffer overflow #780
• Fix: fix border_align_backward bug, sup nan/inf #765
• Fix: fix memory leak of dynamic_point_to_voxel_backward #782
• Fix: fix return-status in kernel-func #784
• Fix: fix border_align #779
• Fix:Fix generate_proposals_v2 buffer overflow. #786
• Fix:Fix 500 series platform max_num_priors exceed the limit problem. #785
• Fix: support half datatype nan/inf. #787
• Fix: add large tensor num check for bbox_overlaps. #790

Performance

• Perf: Optimize ms_deform_attn_backward on MLU590 #661

Documentations

• Docs:update mlu-only mode #720
• Docs: Update docs for v0.8.0 #793

Test

• TEST: add mutual_information_backward api test #675
• TEST:add diff_iou_rotated_sort_vertices_forward api test #710
• TEST: add focal_loss_sigmoid api test #709
• TEST: add dynamic_point_to_voxel_forward api test #711
• TEST: add mutual_information_backward api test #718
• TEST: add mutual_information_forward api test #712
• TEST: add tin_shift api test #756
• TEST: fix error in diff_iou_rotated_sort_vertices_forward #770
• TEST: fix error in indice_conv_backard_filter and indice_conv_forward #771
• TEST: add masked_col2im_forward/masked_im2col_forward api test #781

Contributors

@wangrt1 @pingjiang123 @GuangpingYuan @defei-coder @duzekunKTH @Unireverse @PetrelYy @DanieeelLiu @mahxn0 @Wickyzheng @baicaixmj @Devin-D-U @xwulin @stulai @zhoujiangmin

Full Changelog: v0.7.1...v0.8.0

v0.7.1

Features

• Feature: replace math.h by cmath #719
• Feature: add concurrent build jobs #733
• Feature: upgrade binary to v1.19.0 and fix bug #74

Bug Fixes

• Fix: support grad_voxel_feats zero element & fix constraint in header #713
• Fix: change dynamic_point_to_voxel_backward mlu file name #715
• Fix: fix memory leak in binary ops #722
• Fix: dynamic_point_to_voxel fix mm leak #725
• Fix: update ci code format check r0.7 #734

Documentations

• Docs: update docs for v0.7.1 #749

Contributors

@WangYuan-cn @duzekunKTH @tudejiang79 @LittleReal @mahxn0 @ZhangLearning @DanieeelLiu @liuyuan1-v @wushaoqiang2168 @Unireverse @starkhu @Devin-D-U @Shin-Wang @FRYao @njdragonknight @yangdian0315 @Wickyzheng @tcd-fang @AquamarineSu @xichangzun

Full Changelog: v0.7.0...v0.7.1

v0.7.0

Features

• Feature: new op focal_loss_sigmoid #590
• Feature: add ms_deform_attn_backward fast kernel #629
• Feature: improve focal_loss_sigmoid_backward precision #641
• Feature: three_interpolate supports the tensor num near 2G #637
• Feature: new op mutual_information_backward #648
• Feature: new op dynamic_point_to_voxel_forward #634
• Feature: ball_query support the tensor num near to 2G #653
• Feature: new op diff_iou_rotated_sort_vertices_forward #625
• Feature: format func in common.h #672
• Feature: new op mutual_information_forward #668
• Feature: update toolkit3.5.0 #660
• Feature: new op dynamic_point_to_voxel_backward #662
• Feature: upgrade binary ops to v1.18.1 #686

Bug Fixes

• Fix: unify carafe_forward and carafe_backward op's proto #607
• Fix: replace MLULOG to LOG(ERROR) #621
• Fix: fix dynamic_point_to_voxel_forward bug #663
• Fix: fix psroipool docs #665
• Fix: fix ms_deform_attn_backward bug #650
• Fix: fix nms_rotated bug #669
• Fix: fix focal_loss_sigmoid_backward mpu read overflow #676
• Fix: correct dynamic_point_to_voxel_backward call scatter_nd shape #678
• Fix: update clang version to fix coverage test on r0.7 #693
• Fix: fix binary ops memory leak #698
• Fix: fix focal_loss_sigmoid_backward gamma range precheck #706
• Fix: add ms_deform_attn_forward/backward into the real data relied op list #703

Test

• TEST: add ms_deform_attn_backward api check cases #633
• TEST: add ms_deform_attn_forward api test #652

Performance

• Perf: optimize ms_deform_attn_backward on MLU590 #684

Documentations

• Docs: fix doc problems #608
• Docs: fix userguide copyright year #620
• Docs: delete get_indice_pairs sensitive words #636
• Docs: fix code coverage doc #646
• Docs: update the License for nms and roialign_forward #657
• Docs: revise terminology #655
• Docs: fix warnings in sphinx #667

Contributors

@WangYuan-cn @duzekunKTH @tudejiang79 @LittleReal @mahxn0 @ZhangLearning @DanieeelLiu @liuyuan1-v @wushaoqiang2168 @Unireverse @starkhu @Devin-D-U @Shin-Wang @FRYao @njdragonknight @yangdian0315 @Wickyzheng @tcd-fang @AquamarineSu @xichangzun

Full Changelog: v0.6.0...v0.7.0

v0.6.0

Features

• Feature: new op points_in_boxes #560
• Feature: move the symbol_visibility.map file to build dir #570
• Feature: roi_crop_backward support nan/inf #573
• Feature: new op nms binary #568
• Feature: add submodules proto #578
• Feature: rm mlu_op_kerner.h and modify some ops code #569
• Feature: new op roi_align_backward binary #575
• Feature: modify yolo_box kernel code to support nan/inf #586
• Feature: new op ms_deform_attn_backward #533
• Feature: new op ms_deform_attn_forward source code. #587
• Feature: modify binary to source code #592
• Feature: add test_list #595
• Feature: update cntoolkit 3.4.1 and docker ci yaml. #594
• Feature: update read.me, pr and develop process with submodule info. #593
• Feature: add nms userguide & modify roi_align_forward userguide #598
• Feature: roiaware_pool3d_forward support mlu590 nan inf #599
• Feature: modify asm sync and coreId. #610

Bug Fixes

• Fix: fix gen_case #563
• Fix: fix ballquery gencase macro function #562
• Fix: modify CMakeList to solve the single-operator link error #547
• Fix: fix roi_crop_forward mlu_op.h #574
• Fix: remove old install package when extract cntoolkit in independent_build. #544
• Fix: add libprotobuf version restriction info to read.me. in #579
• Fix: fix psroipool __float2int_rn to __float2int_rd #582
• Fix: add arch sperating setting in operators to support 290 compile. #600
• Fix: fix a host selection error of ms_deform_attn_forward #605
• Fix: unify carafe_forward and carafe_backward op's proto. #617

Test

• TEST:add roipoint_pool3d api check cases #596

Documentations

• Docs: fix unique/transpose docs #564
• Docs: upgrade relase-note in master #567
• Docs: fix points_in_boxes index rst #601
• Docs: update docs for v0.6.0 #603
• Docs: fix doc problems #618

Contributors

@WangYuan-cn @duzekunKTH @tudejiang79 @LittleReal @mahxn0 @ZhangLearning @DanieeelLiu @liuyuan1-v @wushaoqiang2168 @Unireverse @starkhu @Devin-D-U @Shin-Wang @FRYao @njdragonknight @yangdian0315 @Wickyzheng

Full Changelog: v0.5.1...v0.6.0

v0.5.1

Features

• Feature: Add new op nms_rotated. #472
• Feature: add foolcheck for indice-convbpdata workspace-size ptr. #492
• Feature: sup mlu590. #497
• Feature: add new op moe_dispatch_backward_gate. #517
• Feature: add moe_dispatch_backward_data new op. in #524
• Feature: Add ops GTEST and upgrade binary ops. #503
• Feature: add new op moe-dispatch-forward. #530
• Featrue use fillv3 op to fill -1 in host. #500

Bug Fixes

• Fix: improve compile and pre-commit. #505
• Fix: nms_rotated not support for large tensor. #509
• Fix: have ci work seprately for r0.5. #513
• Fix: transfer ci to new local for r0.5. #520
• Fix: Fix bug and spelling errors. #489
• Fix: Fix moe_dispatch_backward_data api guide. #528
• Fix: Repair spelling errors in notes. #484

Test

• TEST: add nms_rotated api cases. #531

Documentations

• Docs: Fix doc. #460
• Docs: Fix mlu_op.h. #479
• Docs: Fix r0.5 docs. #519
• Docs: relase-note 0.5.302. in #537
• Docs: upgrade release-note-v0.5.1. #488

Contributors

@WangYuan-cn @tudejiang79 @LittleReal @mahxn0 @ZhangLearning @DanieeelLiu @liuyuan1-v @duzekunKTH @wushaoqiang2168 @Unireverse @starkhu @Devin-D-U

Full Changelog: v0.5.0...v0.5.1

v0.5.302

Special Edition

Features

• Feature: Add new op nms_rotated. #472
• Feature: add foolcheck for indice-convbpdata workspace-size ptr. #492
• Feature: sup mlu590. #497
• Feature: add new op moe_dispatch_backward_gate. #517
• Feature: add moe_dispatch_backward_data new op. in #524
• Feature: Add ops GTEST and upgrade binary ops. #503
• Feature: add new op moe-dispatch-forward. #530
• Featrue use fillv3 op to fill -1 in host. #500

Bug Fixes

• Fix: improve compile and pre-commit. #505
• Fix: nms_rotated not support for large tensor. #509
• Fix: have ci work seprately for r0.5. #513
• Fix: transfer ci to new local for r0.5. #520
• Fix: Fix bug and spelling errors. #489
• Fix: Fix moe_dispatch_backward_data api guide. #528
• Fix: Repair spelling errors in notes. #484

Test

• TEST: add nms_rotated api cases. #531

Documentations

• Docs: Fix doc. #460
• Docs: Fix mlu_op.h. #479
• Docs: Fix r0.5 docs. #519
• Docs: relase-note 0.5.302. in #537
• Docs: upgrade release-note-v0.5.1. #488

Contributors

@WangYuan-cn @tudejiang79 @LittleReal @mahxn0 @ZhangLearning @DanieeelLiu @liuyuan1-v @duzekunKTH @wushaoqiang2168 @Unireverse @starkhu @Devin-D-U

v0.4.2

Features

• Feature: Add new op box_iou_rotated. #396
• Feature: Add new op nms_rotated. #470
• Feature: Make voxelization operator supported mlu200 series. #463
• Feature: Add bangc-ops independent_build for r0.4 and remove bangpy ci from r0.4. #465
• Feature: Add namespace and symbol visibility. #480
• Feature: Add new op tensor_stride_process. #486

Bug Fixes

• Fix: Repair spelling errors in notes. #483

Test

• None

Documentations

• Docs: Fixed doc problems. #473
• Docs: Fixed mlu_op.h. #476
• Docs: Upgrade release-note-v0.4.2. #478

Contributors

@ZhangLearning @mahxn0 @DanieeelLiu @liuyuan1-v @tudejiang79 @wushaoqiang2168 @WangYuan-cn @GraceGuanH

V0.5.0

Features

• Feature: Add new op roi_align_rotated. #307
• Feature: Add new op rotated_feature_align. #331
• Feature: Add new op active_rotated_filter_forward. #361
• Feature: Add new op roiawarepool3d. #352
• Feature: Add new operator bbox_overlaps. #363
• Feature: Add new op deform_roi_pool. #380
• Feature: Add new op indices_convolution_backward_data. #391
• Feature: Add new op three_nn_forward. #381
• Feature: Add new op get_indice_pairs. #388
• Feature: Add new operator carafe. #364
• Feature: Add new operator box_iou_rotated. #346
• Feature: Add new operator three_interpolate_backward. #332
• Feature: Add new operator indice convolution backward filter. #392
• Feature: support binary kernels in MLUOPS. #360
• Feature: update cntookit 3.3.0. #369
• Feature: Add some base function. #375
• Feature: pts_feature support inf/nan on MLU590. #382
• Feature: Fix binary-ops core dump problem. #387
• Feature: voxel_pooling_forward Add gen_case op param. #389
• Feature: update binary-ops version to 1.15.2. #385
• Feature: Add Add_n files. #390
• Feature: support bangc-ops build with user-defined cntoolkit. in #402
• Feature: update indice_convolution_backward_data docs and code. #405
• Feature: Add new op indice_convolution_forward. #406
• Feature: binary ops Add gtest. #410
• Feature: update release docs and ci. #409
• Feature: adjust mlu_op_kernel.h. #431
• Feature: update platform-arch check for indice convbpdata. #452
• Feature: Get indice pairs paltform. #453

Bug Fixes

• Fix: mluOpVoxelPoolingForward check prams. #341
• Fix: Canceled the 0 element check. #343
• Fix: Fix perf_analyse readme and pr template. #345
• Fix: Fix nueware_home environment variable name. #344
• Fix: Fix bangc-ops samples build. #377
• Fix: sync debian installer info. #376
• Fix: Fix mlu_ops to mlu-ops. #348
• Fix: Fix coverage test failed. #347
• Fix: Fix op support nan/inf. #351
• Fix: del some vlog. #357
• Fix: Fix libexternal_lib.so not found error in daily test. #367
• Fix: Fix unique ops core dump. #458
• Fix: Fix indice_convolution_backward_filter param check for r0.5. #436
• Fix: Fix roi_align_rotated cpu compute. #427
• Fix: Fix indice_convolution_backward_filter param check. #430
• Fix: Add data type check, workspacesize nullptr check to indice_convolution_forward. #434
• Fix: place indice_convolution_forward zero element before nullptr check. #446
• Fix: update foolcheck for indice_convolution_backward_data. #429
• Fix: optimize indice_convolution_forward. #422
• Fix: Fix indice_convolution_forward bug. #416
• Fix :Fix get_indice_pairs bug. #413
• Fix: Fix get indice pairs gencase. #418
• Fix: Fix indice convolution backward filter getWorkspace…. #417
• Fix: get num_act_out from sparse_conv_desc. #419
• Fix: Fix the log error. #404
• Fix: Fix libmluops.so file symbol visibility. #379
• Fix: refine include directories searching under mlu_op_gtest. #368
• Fix: remove libexternal.so related package info. #374
• Fix: Fix mlu_op.h format. #386

Test

• TEST: Add api check cases for voxelization. #327
• TEST:Add api check cases for voxel_poolking_forward. #315
• TEST: Add sync. #359
• TEST:Add api check cases for three_interpolate_backward. #371
• TEST: Add api check cases. #384
• [TEST](three_nn_forward, active_rotated_filter_forward, deform_roi_pool): Add api check…. #412

Documentations

• Docs: Fix bangc-ops user guide info. #383
• Docs: Fix pr style. #395
• Docs: Fix spconv_bpfilter docs problem. #400
• Docs: update BANGC-OPS-Operator-Design-Doc-Template.md. #420
• Docs: Fixed the doc. #454
• Docs: Fixed the 0.5 doc. #437
• Docs: Fix some word description. #358
• Docs: update indice_convpbdata api comments. #415
• Docs: Add get indice pairs docs. #423
• Docs: Improved the doc. #366
• Docs: update indice_convbpdata user guide. #408

Contributors

@FRYao @njdragonknight @LittleReal @FRYao @ling08 @Shin-Wang @DanieeelLiu @jerryjiahaha @wushaoqiang2168 @tudejiang79 @alex-xuwenming @defei-coder @Devin-D-U @duzekunKTH @NekHan @ZhangLearning @WangYuan-cn @GraceGuanH @Unireverse @mahxn0 @njdragonknight @ArtIntAI