This reverts commit 670fcc3c.

- 增加cuda c++ demo.
- 考虑到检测模型尾部一般是multiclass_nms，该kernel为host，如果fetch kernel为cuda的话，则会在此处插入无用的io_copy(host->cuda)，由于该原因，注释掉fetch的cuda kernel. 默认使用host的fetch kernel. 此处暗中进行的行为：每次predictor run完，都会默认把数据从cuda拷贝到cpu

* feature: buffer sharing, test=develop

* fix the warning of buffer.h, test=develop

* test cuda only if necessary, test=develop

* [NPU] subgraph's precision register to kAny

* get the precision from the real tensor or tensor_array of base_scope in arena/framwork

* register assign and assign_value to kAny

* [MLU] support resnet50 on MLU

* [MLU] support resnet50 on MLU

* fix: fix nlp ops input and output type
* fix: add elementwise x_dims>y_dims case

* Fix memory_resuse for opencl. test=develop

* remove useless code. test=develop

* test=devellop

* add cl file into resulted lib test=develop

* test=develop

* test=develop

* [LITE][OPENCL] Support video-sr feature using OpenCL FP16 Image. test=develop

* optimize image2d_to_buffer_with_post255. test=develop

* add def debug in cl kernel. test=develop

* remove conv image code in conv buffer. test=develop

* Support lac model, fix bug for argmax lod_reset, test=develop

[LITE][OPENCL] support fp16 for cl_image_converter, layout, activation all OpenCL image kernel. test=develop (#2964)

* [LITE][OPENCL] support fp16 for cl_image_converter, layout, activation image kernel. test=develop

* add conv, depthwise and UT. test=develop

* add pool, conv, nearest_interp kernel. test=develop

* support fp16 for scale, reshape, concat, fc buffer opencl kernel. test=develop

* refactor for mul opencl buffer kernel. test=develop

* support fp16 for elementwise_mul opecl image kernel. test=develop

* support fp16 for elementwise_mul opencl image kernel. test=develop

* support fp16 for ele_add, fuse_ele_add_act opencl kernel. test=develop

* rename io_copy. test=develop

* mobilenetv1,v2 passed on 855. test=develop

* fix opt for opencl. test=develop

skip fusing quantized conv2d + relu6 for now

Optimize cuda kernel and remove io_copy added by default due to missing fetch_cuda kernel

fix fpga lite_tensor compile bug
     add fake quantize_abs_max op
     test=develop

* fix con+relu6/leakyRelu fusion in Fp32, test=develop

* note m=397 in sgemv_int8 ut, test=develop

* fix ios build error. test=develop

* Optimize the InferShape of several operators.
test=develop

* Remove the new function, resize and CheckPositive in DDim.
test=develop

* Fix a bug in fc_op's InferShape.
test=develop

test=develop

* optimize quant_dequant_fuse_pass, test=develop

* update, test=develop

* update, test=develop

* fix bug for accessing the removed node, test=develop

* set the bias of int8 conv as float, test=develop

* support weight quantization, test=develop

* up, test=develop

* up, test=develop

* up, test=develop

1. add conv_5x5s2_dw to support any padding
2. add 1x1s2pooling impl
3. fix conv dw 3x3 s1p01 bug