bug fix

source/sample/fnnlm/FNNLM.cpp

@@ -1108,10 +1108,6 @@ void Test(const char * test, const char * result, FNNModel &model)
         /* the gold standard */
         XTensor gold;
         
-        if (!autoDiff) {
-            /* prepare an empty network for building the fnn */
-            FNNNet net;
-
         /* make the input tensor for position i */
         for (int i = 0; i < model.n - 1; i++)
             MakeWordBatch(inputs[i], ngrams, ngramNum, i, model.vSize, model.devID, model.mem);
@@ -1119,6 +1115,10 @@ void Test(const char * test, const char * result, FNNModel &model)
         /* make the gold tensor */
         MakeWordBatch(gold, ngrams, ngramNum, model.n - 1, model.vSize, model.devID, model.mem);
 
+        if (!autoDiff) {
+            /* prepare an empty network for building the fnn */
+            FNNNet net;
+
             /* forward computation */
             Forward(inputs, output, model, net);
         }

source/tensor/core/math/Unary.cpp

@@ -122,9 +122,9 @@ _SIMPLE_UNARY_FUNCTION(_Tan, tan)
 _SIMPLE_UNARY_FUNCTION_ME(_TanMe, _Tan)
 SIMPLE_UNARY_FUNCTION(Tan, _Tan, MATH_TAN)
 
-_SIMPLE_UNARY_FUNCTION(_Round, round)
+/*_SIMPLE_UNARY_FUNCTION(_Round, round)
 _SIMPLE_UNARY_FUNCTION_ME(_RoundMe, _Round)
-SIMPLE_UNARY_FUNCTION(Round, _Round, MATH_ROUND)
+SIMPLE_UNARY_FUNCTION(Round, _Round, MATH_ROUND)*/
 #endif
 
 }
\ No newline at end of file

source/tensor/core/math/Unary.cu

@@ -57,6 +57,6 @@ SIMPLE_UNARY_FUNCTION_GPU(Log, log)
 SIMPLE_UNARY_FUNCTION_GPU(Sin, sin)
 SIMPLE_UNARY_FUNCTION_GPU(Cos, cos)
 SIMPLE_UNARY_FUNCTION_GPU(Tan, tan)
-SIMPLE_UNARY_FUNCTION_GPU(Round, round)
+//SIMPLE_UNARY_FUNCTION_GPU(Round, round)
 
 }
\ No newline at end of file

source/tensor/core/math/Unary.cuh

@@ -84,13 +84,13 @@ void KernelTan(__half * a, __half * b, int size);
 void _CudaTan(const XTensor * a, XTensor * b);
 
 /* set each entry to its round value (CUDA Kernel) */
-__global__
-void KernelRound(DTYPE * a, DTYPE * b, int size);
+//__global__
+//void KernelRound(DTYPE * a, DTYPE * b, int size);
 /* set each entry to its round value (CUDA Kernel) with float16 data type*/
-__global__
-void KernelRound(__half * a, __half * b, int size);
+//__global__
+//void KernelRound(__half * a, __half * b, int size);
 /* set each entry to its round value */
-void _CudaRound(const XTensor * a, XTensor * b);
+//void _CudaRound(const XTensor * a, XTensor * b);
 
 #endif // USE_CUDA