add cuda kernel.

abs, tanh, elementwise_sub

add cuda ci.

Organize cuda compilation products.

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

add sequence_pool_concat fuse pass

add fuse kernel

optimize search_grnn

* init resnet cuda int8 support
test=develop

* refine cuda unit test
test=develop

* add the forgeted file.
test=develop

* cuda kernel for sequence_topk_avg_pooling and search_fc test=develop

* fix cuda building, test=develop

* remove sequence_pool from cmake because build error, test=develop

* fix x86 search_grnn and add unit test
* add cuda search_grnn and unit test

* add sequence_pool cuda kernel, test=develop

* fix sequence_pool cuda,test=develop

* fix and complete unittest, test=develop

* [X86] add attention_padding_mask op, x86 kernel and unit test

test=develop

* [CUDA] add attention_padding_mask cuda kernel and unit test

test=develop

- add var_conv_2d cuda kernel

- add var_conv_2d cuda kernel unit test

- temporarily set to two input mode, remove input(ROW) and input(COLUMN)

add sequence_pool cuda kernel

* [X86][CUDA] add sequence_arithmetic op , x86 kernel, cuda kernel and unit test

test=develop

* add sequence_arithmetic cuda kernel unit test

test=develop

* add cuda match_matrix_tensor op and test, test=develop

* add search_seq_depadding x86 and cuda
* add match_matrix_tensor x86
* add search_grnn x86, no test

- add sequence_reverse op

- add sequence_reverse kernel for x86 and cuda

- add sequence_reverse_test for x86 and cuda

- add sequence_concat op

- add sequence_concat kernel for x86 and cuda

- add sequence_concat_test for x86 and cuda

add cuda kernel:lookup table

* update framework.proto

* add compatibility check, test=develop

* remove head files, test=develop

* [python api] init add python api test=develop

* add gpu kernel mul pool relu scale softmax dropout bilinear_interp and can run in tx2

* rm GREATER_EQUAL

* cuda add softmax and pool op

* * fix armlinux can find sys/system_properties.h
* conv add padding_algorithm
test=develop

* delete padding_algorithm in op param, test=develop

* fix bugs, test=develop

* add conv int8 support(in condition which the input or output channel not be the times of 4)
add add_kernel for cuda.

* can run yolov3 fp32
test=develop

* 1. fix bug with yolov3 run
test=develop

* can run yolov3 int8 test=develop

* add conv int8 support(in condition which the input or output channel not be the times of 4)
add add_kernel for cuda.

* can run yolov3 fp32
test=develop

* 1. fix bug with yolov3 run
test=develop

add transpose kernel for cuda

* add nearest_interp_cuda kernel, test=develop

* add concat op and elementwise_add op

* remove eigen dependency from nearest_interp cuda kernel, test=develop

* free cuda pointers, test=develop

* add calib cuda kernel.

* add unit test for calib cuda kernel. test=develop

* paddle lite cuda init
can run model with leaky_relu

* add the missing file.
test=develop

* add the load from memory interface.
test=develop

* refine this pr. fix comments
fix ci error
test=develop

* conv impl
fp32:
conv, conv+bais, conv+bias+relu, conv+bias+leaky_relu

int8:
conv, conv+bais+relu(int8 or fp32 output), conv+bias+leaky_relu(int8 or fp32 output)

can run conv+ bias+relu using cxx_api
test=develop

* move the lite/cuda/math to backends/cuda/math
test=develop

add nearest_interp cuda kernel for Paddle-Lite

* add yolo_box_compute cuda

* move multiclass_nms(arm) to host

* add lod in scale op

* add yolo_box_cuda cmake config

* modify shuffle_channel_fuse and transpose_softmax_transpose_fuse to support run ssd model. test=develop

* reshape and transpose op don't have xshape output.

* modify yolo_box_compute_cuda, use tensor to manage cuda memory test=develop

* add yolo_box use kernel test=develop

* paddle lite cuda init
can run model with leaky_relu

* add the missing file.
test=develop