四、GPU
1、Halcon中使用GPU提速，效果明顯。
Windows開始菜單--運作--輸入dxdiag--顯示，可以看到自己電腦的顯示卡型号。
官方自帶的例程compute_devices.hdev，實作提速的優良效果，必須先關閉裝置：dev_update_off()；
來自官方例程compute_devices.hdev
* This example shows how to use compute devices with HALCON.
*
dev_update_off ()
dev_close_window ()
dev_open_window_fit_size (0, 0, 640, 480, -1, -1, WindowHandle)
set_display_font (WindowHandle, 16, 'mono', 'true', 'false')
*
* Get list of all available compute devices.
query_available_compute_devices (DeviceIdentifier)
*
* End example if no device could be found.
if (|DeviceIdentifier| == 0)
return ()
endif
*
* Display basic information on detected devices.
disp_message (WindowHandle, 'Found ' + |DeviceIdentifier| + ' Compute Device(s):', 'window', 12, 12, 'black', 'true')
for Index := 0 to |DeviceIdentifier| - 1 by 1
get_compute_device_info (DeviceIdentifier[Index], 'name', DeviceName)
get_compute_device_info (DeviceIdentifier[Index], 'vendor', DeviceVendor)
Message[Index] := 'Device #' + Index + ': ' + DeviceVendor + ' ' + DeviceName
endfor
disp_message (WindowHandle, Message, 'window', 42, 12, 'white', 'false')
disp_continue_message (WindowHandle, 'black', 'true')
stop () 2、操作GPU裝置有關的算子：
query_available_compute_devices
get_compute_device_info
open_compute_device
init_compute_device
activate_compute_device
deactivate_compute_device
3、官方自帶的例程get_operator_info.hdev，可以檢視支援GPU加速（OpenCL）的算子；
* Determine all operators that support OpenCL
get_opencl_operators (OpenCLSupport)
* 自定義函數展開之後,有get_operator_info算子
get_operator_name ('', OperatorNames)
get_operator_info (OperatorNames[Index], 'compute_device', Information)
這裡舉例Halcon 19.11版本可以加速的算子有82個：
['abs_diff_image', 'abs_image', 'acos_image', 'add_image', 'affine_trans_image', 'affine_trans_image_size', 'area_center_gray', 'asin_image', 'atan2_image', 'atan_image', 'binocular_disparity_ms', 'binocular_distance_ms', 'binomial_filter', 'cfa_to_rgb', 'change_radial_distortion_image', 'convert_image_type', 'convol_image', 'cos_image', 'crop_domain', 'crop_part', 'crop_rectangle1', 'depth_from_focus', 'derivate_gauss', 'deviation_image', 'div_image', 'edges_image', 'edges_sub_pix', 'exp_image', 'find_ncc_model', 'find_ncc_models', 'gamma_image', 'gauss_filter', 'gauss_image', 'gray_closing_rect', 'gray_closing_shape', 'gray_dilation_rect', 'gray_dilation_shape', 'gray_erosion_rect', 'gray_erosion_shape', 'gray_histo', 'gray_opening_rect', 'gray_opening_shape', 'gray_projections', 'gray_range_rect', 'highpass_image', 'image_to_world_plane', 'invert_image', 'linear_trans_color', 'lines_gauss', 'log_image', 'lut_trans', 'map_image', 'max_image', 'mean_image', 'median_image', 'median_rect', 'min_image', 'mirror_image', 'mult_image', 'points_harris', 'polar_trans_image', 'polar_trans_image_ext', 'polar_trans_image_inv', 'pow_image', 'principal_comp', 'projective_trans_image', 'projective_trans_image_size', 'rgb1_to_gray', 'rgb3_to_gray', 'rotate_image', 'scale_image', 'sin_image', 'sobel_amp', 'sobel_dir', 'sqrt_image', 'sub_image', 'tan_image', 'texture_laws', 'trans_from_rgb', 'trans_to_rgb', 'zoom_image_factor', 'zoom_image_size']
4、官方手冊
C:\Program Files\MVTec\HALCON-19.11-Progress\doc\pdf\reference\reference_hdevelop.pdf
Chapter 25 System --- 25.1 Compute Devices
五、舉例測試
*參考官方例程optimize_aop.hdev;query_aop_info.hdev;simulate_aop.hdev;
*舉例edges_sub_pix算子性能測試
dev_update_off ()//實作提速的優良效果，必須先關閉裝置
dev_close_window ()
dev_open_window_fit_size (0, 0, 640, 480, -1, -1, WindowHandle)
set_display_font (WindowHandle, 16, 'mono', 'true', 'false')
get_system ('processor_num', NumCPUs)
get_system ('parallelize_operators', AOP)
*讀取圖檔
read_image(Image, 'D:/hellowprld/2/1-.jpg')
*彩色轉灰階圖
count_channels (Image, Channels)
if (Channels == 3 or Channels == 4)
rgb1_to_gray (Image, ImageGray)
endif
alpha:=5
low:=10
high:=20
*測試1:去掉AOP,即沒有加速并行處理
set_system ('parallelize_operators', 'false')
get_system ('parallelize_operators', AOP)
count_seconds(T0)
edges_sub_pix (ImageGray, Edges1, 'canny', alpha, low, high)
count_seconds(T1)
Time0:=(T1-T0)*1000
stop()
*測試2:AOP自動加速并行處理
*Halcon的預設值是開啟AOP的,即parallelize_operators值為true
set_system ('parallelize_operators', 'true')
count_seconds(T1)
edges_sub_pix (ImageGray, Edges1, 'canny', alpha, low, high)
count_seconds(T2)
Time1:=(T2-T1)*1000
stop()
*測試3:GPU加速，支援GPU加速的算子Halcon19.11有82個
*GPU加速是先從CPU中将資料拷貝到GPU上處理，處理完成後再将資料從GPU拷貝到CPU上。從CPU到GPU再從GPU到CPU是要花費時間的。
*GPU加速一定會比正常的AOP運算速度快嗎?不一定!結果取決于顯示卡的好壞.
query_available_compute_devices(DeviceIdentifiers)
DeviceHandle:=0
for i:=0 to |DeviceIdentifiers|-1 by 1
get_compute_device_info(DeviceIdentifiers[i], 'name', Nmae)
if (Nmae == 'GeForce GT 630')//根據GPU名稱打開GPU
open_compute_device(DeviceIdentifiers[i], DeviceHandle)
break
endif
endfor
if(DeviceHandle#0)
set_compute_device_param (DeviceHandle, 'asynchronous_execution', 'false')
init_compute_device(DeviceHandle, 'edges_sub_pix')
activate_compute_device(DeviceHandle)
endif
*獲得顯示卡的資訊
get_compute_device_param (DeviceHandle, 'buffer_cache_capacity', GenParamValue0)//預設值是顯示卡緩存的1/3
get_compute_device_param (DeviceHandle, 'buffer_cache_used', GenParamValue1)
get_compute_device_param (DeviceHandle, 'image_cache_capacity', GenParamValue2)
get_compute_device_param (DeviceHandle, 'image_cache_used', GenParamValue3)
*GenParamValue0 := GenParamValue0 / 3
*set_compute_device_param (DeviceHandle, 'buffer_cache_capacity', GenParamValue0)
*get_compute_device_param (DeviceHandle, 'buffer_cache_capacity', GenParamValue4)
count_seconds(T3)
*如果顯示卡緩存不夠,會報錯,error #4104 : Out of compute device memory
edges_sub_pix (ImageGray, Edges1, 'canny', alpha, low, high)
count_seconds(T4)
Time2:=(T4-T3)*1000
if(DeviceHandle#0)
deactivate_compute_device(DeviceHandle)
endif
stop()
*測試4:AOP手動優化
set_system ('parallelize_operators', 'true')
get_system ('parallelize_operators', AOP)
*4.1-優化線程數目方法'threshold'
optimize_aop ('edges_sub_pix', 'byte', 'no_file', ['file_mode','model','parameters'], ['nil','threshold','false'])
count_seconds(T5)
edges_sub_pix (ImageGray, Edges1, 'canny', alpha, low, high)
count_seconds(T6)
Time3:=(T6-T5)*1000
*4.2-優化線程數目方法'linear'
optimize_aop ('edges_sub_pix', 'byte', 'no_file', ['file_mode','model','parameters'], ['nil','linear','false'])
count_seconds(T7)
edges_sub_pix (ImageGray, Edges1, 'canny', alpha, low, high)
count_seconds(T8)
Time4:=(T8-T7)*1000
stop()
*4.3-優化線程數目方法'mlp'
optimize_aop ('edges_sub_pix', 'byte', 'no_file', ['file_mode','model','parameters'], ['nil','mlp','false'])
count_seconds(T9)
edges_sub_pix (ImageGray, Edges1, 'canny', alpha, low, high)
count_seconds(T10)
Time5:=(T10-T9)*1000
stop()
dev_clear_window()
Message := 'edges_sub_pix runtimes:'
Message[1] := 'CPU only Time0 without AOP='+Time0+'ms,'
Message[2] := 'CPU only Time1 with AOP='+Time1+'ms,'
Message[3] := 'GPU use Time2='+Time2+'ms,'
Message[4] := 'optimize Time3 threshold='+Time3+'ms'
Message[5] := 'optimize Time4 linear='+Time4+'ms'
Message[6] := 'optimize Time5 mlp='+Time5+'ms'
disp_message (WindowHandle, Message, 'window', 12, 12, 'red', 'false')
stop() edges_sub_pix算子性能測試結果：
rotate_image算子性能測試結果：
得出的結論是：
1、GPU加速是先從CPU中将資料拷貝到GPU上處理，處理完成後再将資料從GPU拷貝到CPU上。從CPU到GPU再從GPU到CPU是要花費時間的。
2、GPU加速一定會比正常的AOP運算速度快嗎?不一定!結果取決于顯示卡的好壞.
3、GPU加速，如果顯示卡緩存不夠,會報錯,error #4104 : Out of compute device memory
完整的*.hdev工程檔案請下載下傳：
https://download.csdn.net/download/libaineu2004/12146529