dropout:ompsimd+cuda的实现

Author: miaobyte

doc/excuter/op-mem-ompsimd/list.md

@@ -49,6 +49,7 @@
 
 | Operation | Author |  Math Formula | IR Instruction |
 |-----------|--------|--------------|----------------|
+| dropout | miaobyte | A.dropout(p,seed)->C | dropout(tensor<any> A, var<float32> p, var<int32> seed)->(tensor<any> C) |
 | switch | miaobyte | C=switch([tensors],case) | switch(listtensor<any> tensors, tensor<int8> cases)->(tensor<any> C) |
 | greaterscalar | miaobyte | mask=greater(T1,scalar) | greaterscalar(tensor<any> A, var<any> scalar)->(tensor<bool> mask) |
 | equalscalar | miaobyte | mask=equal(T1,scalar) | equalscalar(tensor<any> A, var<any> scalar)->(tensor<bool> mask) |

excuter/op-mem-ompsimd/src/client/tfs.cpp

@@ -465,6 +465,17 @@ namespace deepx::tf
                                                                 {
                                                                     Param("C", DataCategory::Tensor, Precision::Any),
                                                                 })));
+        // dropout author=miaobyte
+        tffactory.add_tf(std::make_shared<Dropout<miaobyte>>(vector<Param>(
+                                                                 {
+                                                                     Param("A", DataCategory::Tensor, Precision::Any),
+                                                                     Param("p", DataCategory::Var, Precision::Float32),
+                                                                     Param("seed", DataCategory::Var, Precision::Int32),
+                                                                 }),
+                                                                 vector<Param>(
+                                                                     {
+                                                                         Param("C", DataCategory::Tensor, Precision::Any),
+                                                                     }) ));
     }
     // matmul
     void register_matmul(TfFactory &tffactory)