picodet: Paddle-lite on ARM (Linux) + OpenCL failure

[mike-2020]

trafficstars

从这里下载原始模型：https://paddledet.bj.bcebos.com/deploy/Inference/picodet_xs_416_coco_lcnet.tar 使用如下命令将其转换成NB： ./opt_linux --model_dir=picodet_xs_416_coco_lcnet --valid_targets=opencl,arm --optimize_out=picodet_xs_416_coco_lcnet-opencl --optimize_out_type=naive_buffer 使用Paddle-lite v2.11在RK3399(Ubuntu 20.04) （Mali T860）上运行，报告如下错误（错误在最后一行）。只要在opt命令行中把opencl去掉，就可以运行正常了。

[W 6/14 21:54:27.833 ...e-Lite/lite/model_parser/model_parser.cc:889 LoadModelFbsFromFile] warning: the version of opt that transformed this model is not consistent with current Paddle-Lite version. version of opt:v2.11-rc version of current Paddle-Lite:2f131e135 [I 6/14 21:54:27.851 ...e-Lite/lite/backends/opencl/cl_runtime.h:93 OpenCLAvaliableForDevice] need to check fp16 valid:0 [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:69 Init] opencl_lib_found:1 [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:77 Init] dlsym_success:1 [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:537 InitializePlatform] Platform extension: cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_byte_addressable_store cl_khr_3d_image_writes cl_khr_fp64 cl_khr_int64_base_atomics cl_khr_int64_extended_atomics cl_khr_fp16 cl_khr_icd cl_khr_egl_image cl_khr_image2d_from_buffer cl_arm_core_id cl_arm_printf cl_arm_thread_limit_hint cl_arm_non_uniform_work_group_size cl_arm_import_memory [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:85 Init] is_platform_init:1 [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:623 InitializeDevice] Using device: Mali-T860 [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:649 InitializeDevice] CL_DEVICE_VERSION:OpenCL 1.2 v1.r18p0-01rel0.5cb5681058e8e076ff89747c20c32578 [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:656 InitializeDevice] device_type:GPU [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:660 InitializeDevice] The chosen device has 4 compute units. [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:664 InitializeDevice] CL_DEVICE_MAX_CLOCK_FREQUENCY:5 [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:674 InitializeDevice] The local memory size of the chosen device is 32.000000 KB. [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:681 InitializeDevice] CL_DEVICE_GLOBAL_MEM_CACHE_SIZE(KB):256.000000 KB. [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:689 InitializeDevice] CL_DEVICE_GLOBAL_MEM_CACHELINE_SIZE(KB):0.062500 KB. [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:696 InitializeDevice] CL_DEVICE_GLOBAL_MEM_SIZE(KB):3951840.000000 KB. [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:704 InitializeDevice] CL_DEVICE_MAX_WORK_GROUP_SIZE:256 [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:708 InitializeDevice] CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS:3 [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:713 InitializeDevice] max_work_item_sizes[0]:256 [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:713 InitializeDevice] max_work_item_sizes[1]:256 [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:713 InitializeDevice] max_work_item_sizes[2]:256 [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:724 InitializeDevice] CL_DEVICE_MAX_CONSTANT_BUFFER_SIZE:64.000000 [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:735 InitializeDevice] The chosen device supports image processing. [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:739 InitializeDevice] CL_DEVICE_IMAGE2D_MAX_HEIGHT:65536 [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:743 InitializeDevice] CL_DEVICE_IMAGE2D_MAX_WIDTH:65536 [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:757 InitializeDevice] The chosen device supports the half data type. [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:765 InitializeDevice] CL_DEVICE_ADDRESS_BITS:64 [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:769 InitializeDevice] CL_DRIVER_VERSION:1.2 [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:93 Init] is_device_init:1 [I 6/14 21:54:27.889 ...-Lite/lite/backends/opencl/cl_runtime.cc:105 Init] set is_cl_runtime_initialized_ = true [F 6/14 21:54:29.544 ...e/kernels/opencl/matmul_image_compute.cc:342 ReInitWhenNeeded] unsupported input case.

[paddle-bot-old[bot]]

您好，我们已经收到了您的问题，会安排技术人员尽快解答您的问题，请耐心等待。请您再次检查是否提供了清晰的问题描述、复现代码、环境&版本、报错信息等。同时，您也可以通过查看官网文档、常见问题、历史Issue来寻求解答。祝您生活愉快～

Hi! We've received your issue and please be patient to get responded. We will arrange technicians to answer your questions as soon as possible. Please make sure that you have posted enough message to demo your request. You may also check out the API，FAQ and Github Issue to get the answer.Have a nice day!

[sprouteer]

这是因为opencl上matmul v2这个op暂时不支持输入x为2维，输入y为1维的情况，多谢反馈，正在加紧支持

[mike-2020]

谢谢，希望尽快可以得到支持。

这个模型目前无论是320x320还是416x416，都是无法在opencl上跑的。 对于320x320的情况，程序报告calib没有被支持。

[jizhu1023]

@sprouteer 请问大概多久后可以支持?