# Conflicts:
#	examples/ios/TNNExamples/TNNCameraPreviewController/TNNViewModel/TNNFaceDetectAlignerViewModel.mm
#	model/download_model.sh

* 'master' of https://github.com/darrenyao87/TNN:
  Feature rknpu support (#387)
  Feature arm yuv2bgra (#381)

# Conflicts:
#	examples/ios/TNNExamples/TNNCameraPreviewController/TNNViewModel/TNNFaceDetectAlignerViewModel.mm
#	examples/ios/TNNExamples/TNNYoutuFaceAlignController/TNNYoutuFaceAlignController.mm

* upstream/master: (31 commits)
  [ARM] fix reduce l2 layer error (#527)
  Patch rknpu (#528)
  [OPENCL] add opencl code gen to make stage (#521)
  Feature issue 475 (#517)
  [ARM][BUG] fix int8 dwconv kernel shape error (#514)
  [OPENCL] change opencl force fp32 to precision mode && fix layer (#522)
  [iOS] fix iPhone and simulator arch conflicts with Xcode12 (#523)
  Set name and type of Reformat LayerParam (#519)
  [DEVICE][OPENCL] 优化3*3卷积和非对称卷积 (#515)
  [iOS] fix tnn iOS&maxOS build error (#508)
  [DEV][UPD] 1. Int8Reformat -> Reformat;
  [ONNX][BUG] fix pool fusion bug (#500)
  Enhance warpaffine nearest (#501)
  [OPENCL] fix chinese comments (#493)
  Feature mat make border (#491)
  Enhance arm int8 (#486)
  [NPU][BUG] fix comiple error due to api change
  Npu fp16 fix (#488)
  Feature fp16 workflow (#482)
  Opencl reduce softmax opt (#443)
  ...

# Conflicts:
#	platforms/ios/tnn.xcodeproj/project.pbxproj

* upstream/master:
  Hotfix opencl select (#555)
  Eff opt (#554)
  [OPENCL] optimize pooling with fine-grained parallelsim (#553)
  create issue templates (#549)
  [OPENCL] fix fp16 select with short condition (#544)
  [Metal] enable CPU N8UC4 Mat in metal ConvertFromMat (#543)
  Add fuse SpaceToDepth and DepthToSpace (#542)
  [TOOL][ADD] add output name param support (#537)
  [CONVERTER][BUG] fix the bug of fuse conv (#529)
  [UPD] update iOS&macos building scripts, check building errors (#531)
  [Metal] fix reshape out-of-bound access bug (#496)
  Feature demo hairsegmentation (#530)
  [Metal] fix bugs in Copy NCHW_FLAOT mat from metal to CPU (#472)
  support SpaceToDepth and DepthToSpace Operator (#526)
  update onnx coreml convert tool (#467)

* upstream/master: (39 commits)
  Feature arm fp16 op (#588)
  [QUANTIZED][BUG]1. add QuantizedUpsample(layer_type.cc); (#644)
  support empty mat (#603)
  [RKNPU][BUG] fix memleak in network init (#640)
  [RKNPU][ADD] add rkmodel cache interface (#633)
  Fix issue 599 (#608)
  fix abstract_layer_acc.cc shadowed variable (#617)
  Fix sign compare warning (#627)
  [OPENCL] fix reduce multi axis kernel (#614)
  Fix metal convacc selection (#609)
  [ONNX2TNN][BUG] 1. fic onnx2tnn convolution input channel; (#611)
  [QUANT][BUG] fix per tensor quantconcat error (#607)
  [CONVERTER][BUG] 1. fix Issue #604; (#605)
  Fix remove squeeze (#571)
  Fix issue #566 (#570)
  Fix typo. (#594)
  [BUG] fix layer resource count error when packing model (#592)
  Fix typo. (#590)
  [OPENCL][FIX]修复IMAC opencl 1.2 兼容错误导致无法运行unit test问题 (#593)
  Feature quant upsample (#589)
  ...

# Conflicts:
#	source/tnn/core/default_network.cc
#	source/tnn/device/cpu/cpu_context.h
#	source/tnn/device/opencl/opencl_context.h

* upstream/master:
  Enhance fp16 blobconverter (#652)
  [TFLITE][ADD] add elu spacetodepth transpose (#645)
  [DOC][UPD] update onnx2tnn usage info (#650)

* upstream/master: (106 commits)
  Stable v0.3 merge master (#709)
  Dev issue 695 (#708)
  Patch rknpu cache (#707)
  [CONVERTER2TNN][BUG]1.fix bug for convert2tnn build.sh (#700)
  [ARM][BUG] fix upsample cubic openmp too many args (#692)
  [CONVERTER2TNN][UPD] 1. default do not clean build directory; (#694)
  Feature fp16 arm32 (#690)
  pose demo (#665)
  [TOOLS][ADD] Model checker support Huawei NPU (#684)
  Fix input name and the bug of optimization model (#674)
  [CUDA][FIX] fix merge error and support cuda scale
  [CUDA]mdf cuda reduce_l2
  [NPU][ADD] 添加NPU ResizeBilinearV2算子的convert (#646)
  fix github action failures
  fix ctest running issue
  fix gcc7.5 compiling issue
  [CUDA]delete redundant md5
  [SCRIPTS][BUILD] close treat warning as error
  [SCRIPTS][BUILD] set openvino build type release
  Feature opencl kernel opt merge (#672)
  ...

# Conflicts:
#	include/tnn/utils/half_utils.h
#	platforms/ios/tnn.xcodeproj/project.pbxproj
#	source/tnn/device/arm/acc/Half8.h
#	source/tnn/device/arm/acc/arm_relu6_layer_acc.cc
#	source/tnn/device/arm/acc/arm_sigmoid_layer_acc.cc
#	source/tnn/device/arm/acc/arm_unary_layer_acc.h
#	source/tnn/device/arm/acc/compute/compute.cc
#	source/tnn/device/arm/arm_blob_converter.cc
#	source/tnn/device/arm/arm_util.cc
#	source/tnn/utils/cpu_utils.cc

* upstream/master: (1117 commits)
  [ARM][UPD]1. arm device support gride_sample layer; (#931)
  [CUDA]fix cuda blob memory alloc in trt init (#941)
  Fix many typos in comment (#938)
  Feature trt conv3d support (#925)
  Dev doc windows cuda (#930)
  [DOC] update install opencv guide in demo doc
  [DOC] update
  [UPD]fix image display ratio
  [DOC] update doc for android demo
  Fix demo (#929)
  [X86] remove register keyword and format for hard_swish and inst_norm (#924)
  [BUG] fix insert_layout and blob_converter errors when build metal with x86 (#923)
  [CPU][COMPILE] cpu device compile option default on (#922)
  [DOC] add reference projects (#921)
  [FIX][BLOB CONVERT] fix blob convert memory illegal (#919)
  [CUDA][UNITTEST] filter unsupported testcases for cuda (#916)
  [API][COMMENT] change for api comment
  remove unused api
  [ARM][UPD]1. optimzie softmax layer for arm device; (#909)
  [CUDA]add cuda benchmark script (#913)
  ...

# Conflicts:
#	source/tnn/core/tnn.cc

# Conflicts:
#	source/tnn/device/cpu/acc/cpu_neg_layer_acc.cc

# Conflicts:
#	source/tnn/device/metal/metal_context.mm

# Conflicts:
#	CMakeLists.txt