inference-engine/src/cldnn_engine/cldnn_global_custom_kernels/cldnn_global_custom_kernels.xml

   1 <CustomLayer name="PriorBoxClustered" type="SimpleGPU" version="1">
   2     <Kernel entry="prior_box_clustered">
   3         <!-- Path should be relative to executable, so aligning with how it will really put in CI -->
   4         <Source filename="prior_box_clustered.cl"/>
   5         <Define name="min_size_" type="float" param="min_size" default="0.0"/>
   6         <Define name="max_size_" type="float" param="max_size" default="0.0"/>
   7         <Define name="aspect_ratio_" type="float" param="aspect_ratio" default="0.0"/>
   8         <Define name="flip_" type="int" param="flip" default="1"/>
   9         <Define name="clip_" type="int" param="clip" default="0"/>
  10         <Define name="variance_" type="float[]" param="variance" default="0.1"/>
  11         <Define name="img_size_" type="int" param="img_size" default="0"/>
  12         <Define name="img_h_" type="int" param="img_h" default="0"/>
  13         <Define name="img_w_" type="int" param="img_w" default="0"/>
  14         <Define name="step_" type="float" param="step" default="0.0"/>
  15         <Define name="step_w_" type="float" param="step_w" default="0.0"/>
  16         <Define name="step_h_" type="float" param="step_h" default="0.0"/>
  17         <Define name="offset_" type="float" param="offset" default="0.0"/>
  18         <Define name="width_" type="float[]" param="width" default="0.0"/>
  19         <Define name="height_" type="float[]" param="height" default="0.0"/>
  20     </Kernel>
  21     <Buffers>
  22         <Tensor arg-index="0" type="input" port-index="0"  format="BFYX"/>
  23         <Tensor arg-index="1" type="input" port-index="1"  format="BFYX"/>
  24         <Tensor arg-index="2" type="output" port-index="0" format="BFYX"/>
  25     </Buffers>
  26
  27     <CompilerOptions options="-cl-mad-enable"/>
  28     <WorkSizes dim="input" global="Y,X"/>
  29     <!-- define the global worksize. The formulas can use the values of the B,F,Y,X dimensions and contain the operators: +,-,/,*,% (all evaluated in integer arithmetics) Default value: global="B*F*Y*X,1,1"-->
  30 </CustomLayer>
  31
  32 <CustomLayer name="Interp" type="SimpleGPU" version="1">
  33     <Kernel entry="interp">
  34         <Source filename="interp.cl"/>
  35         <Define name="pad_beg_" type="int" param="pad_beg" default="0"/>
  36         <Define name="pad_end_" type="int" param="pad_end" default="0"/>
  37     </Kernel>
  38     <Buffers>
  39         <Tensor arg-index="0" type="input" port-index="0"  format="ANY"/>
  40         <Tensor arg-index="1" type="output" port-index="0" format="ANY"/>
  41     </Buffers>
  42
  43     <CompilerOptions options="-cl-mad-enable"/>
  44     <WorkSizes global="Y,((X + 31)/32)*32" local="1,32"/>
  45 </CustomLayer>
  46
  47 <CustomLayer name="GRN" type="SimpleGPU" version="1">
  48     <Kernel entry="grn">
  49         <Source filename="grn.cl"/>
  50         <Define name="bias_" type="float" param="bias" default="1.0f"/>
  51         <Define name="ACCUMULATOR_TYPE" default="float"/>
  52       </Kernel>
  53     <Buffers>
  54         <Tensor arg-index="0" type="input" port-index="0"  format="BFYX"/>
  55         <Tensor arg-index="1" type="output" port-index="0" format="BFYX"/>
  56     </Buffers>
  57
  58     <CompilerOptions options="-cl-mad-enable"/>
  59     <WorkSizes global="B,Y"/>
  60 </CustomLayer>
  61
  62 <CustomLayer name="Tile" type="SimpleGPU" version="1">
  63     <Kernel entry="tile">
  64         <Source filename="tile.cl"/>
  65         <Define name="axis_" type="int" param="axis" default="1"/>
  66         <Define name="tiles_" type="int" param="tiles"/>
  67     </Kernel>
  68     <Buffers>
  69         <Tensor arg-index="0" type="input" port-index="0"  format="BFYX"/>
  70         <Tensor arg-index="1" type="output" port-index="0" format="BFYX"/>
  71     </Buffers>
  72
  73     <CompilerOptions options="-cl-mad-enable"/>
  74     <WorkSizes global="1,1,1"/>
  75 </CustomLayer>
  76
  77 <CustomLayer name="CTCGreedyDecoder" type="SimpleGPU" version="1">
  78     <Kernel entry="ctc_greedy_decoder">
  79         <Source filename="ctc_greedy_decoder.cl"/>
  80         <Define name="ctc_merge_repeated_" type="int" param="ctc_merge_repeated" default="1"/>
  81       </Kernel>
  82     <Buffers>
  83         <Tensor arg-index="0" type="input" port-index="0"  format="BFYX"/>
  84         <Tensor arg-index="1" type="input" port-index="1"  format="BFYX"/>
  85         <Tensor arg-index="2" type="output" port-index="0" format="BFYX"/>
  86     </Buffers>
  87
  88     <CompilerOptions options="-cl-mad-enable"/>
  89     <WorkSizes global="1,1"/>
  90 </CustomLayer>