Add LogSigmoid

Author: lantiga

test/base.py

@@ -32,10 +32,11 @@ def base_test():
     model_2 = lambda x: F.softmax(F.tanh(fc3(x)))
     model_3 = lambda x: F.softmax(F.sigmoid(fc3(x)))
     model_4 = lambda x: F.softmax(F.leaky_relu(fc4(x)))
+    model_5 = lambda x: F.softmax(F.logsigmoid(fc4(x)))
 
     data = Variable(torch.rand(10,10))
 
-    out = model_0(data) + model_1(data) * model_2(data) / model_3(data) / 2.0 + 2.0 * model_4(data) + 1 - 2.0
+    out = model_0(data) + model_1(data) * model_2(data) / model_3(data) / 2.0 + 2.0 * model_4(data) + model_5(data) + 1 - 2.0
 
     out_path = 'out'
     if not os.path.isdir(out_path):

torch2c/emitters.py

@@ -590,6 +590,31 @@ def free_tpl(self):
 register(Sigmoid, torch.autograd._functions.pointwise.Sigmoid)
 
 
+class LogSigmoid(Emitter):
+
+    def __init__(self, obj, prevfns):
+        Emitter.__init__(self, obj, prevfns)
+        self.def_vars({
+            'input': id(prevfns[0]),
+        })
+        self.infer_type_var = 'input'
+
+    def call_tpl(self):
+        return '''
+            TH${T}Tensor *$id = TH${T}Tensor_new();
+            TH${T}Tensor *buffer_$id = TH${T}Tensor_new();
+            THNN_${T}LogSigmoid_updateOutput(NULL,$input,$id,buffer_$id);
+            '''
+
+    def free_tpl(self):
+        return '''
+            TH${T}Tensor_free($id);
+            TH${T}Tensor_free(buffer_$id);
+            '''
+
+register(LogSigmoid, torch.nn._functions.thnn.auto.LogSigmoid)
+
+
 class Noop(Emitter):
 
     def __init__(self, obj, prevfns):