remove useless code and fix bugs

source/network/Main.cpp

@@ -175,8 +175,6 @@ void SumDimTest()
     XTensor y;
     XTensor y;
     XTensor z;
     XTensor z;
 
 
-    int loops = 2000;
-
     int a = 5;
     int a = 5;
     int b = 7;
     int b = 7;
     int c = 3;
     int c = 3;

source/network/XBackwardMath.cpp

@@ -118,11 +118,16 @@ void XMathGrad::GradSumDim(XTensor * node)
         if(b->outgo.tailNum > 1){
         if(b->outgo.tailNum > 1){
             XTensor * bGradTMP = NewTensorBuf(b->grad, b->devID, b->mem);
             XTensor * bGradTMP = NewTensorBuf(b->grad, b->devID, b->mem);
             _ReduceSum(node->grad, bGradTMP, 0);
             _ReduceSum(node->grad, bGradTMP, 0);
+            if(beta != 1.0F)
+                _ScaleAndShiftMe(bGradTMP, beta);
             _Sum(bGradTMP, b->grad, b->grad);
             _Sum(bGradTMP, b->grad, b->grad);
             DelTensorBuf(bGradTMP);
             DelTensorBuf(bGradTMP);
         }
         }
-        else
+        else{
             _ReduceSum(node->grad, b->grad, 0);
             _ReduceSum(node->grad, b->grad, 0);
+            if(beta != 1.0F)
+                _ScaleAndShiftMe(b->grad, beta);
+        }
 
 
         node->grad->Reshape(order, dimSize);
         node->grad->Reshape(order, dimSize);
     }
     }
@@ -150,11 +155,16 @@ void XMathGrad::GradSumDim(XTensor * node)
         if(b->outgo.tailNum > 1){
         if(b->outgo.tailNum > 1){
             XTensor * bGradTMP = NewTensorBuf(b->grad, b->devID, b->mem);
             XTensor * bGradTMP = NewTensorBuf(b->grad, b->devID, b->mem);
             _ReduceSum(interGrad, bGradTMP, 0);
             _ReduceSum(interGrad, bGradTMP, 0);
+            if(beta != 1.0F)
+                _ScaleAndShiftMe(bGradTMP, beta);
             _Sum(bGradTMP, b->grad, b->grad);
             _Sum(bGradTMP, b->grad, b->grad);
             DelTensorBuf(bGradTMP);
             DelTensorBuf(bGradTMP);
         }
         }
-        else
+        else{
             _ReduceSum(interGrad, b->grad, 0);
             _ReduceSum(interGrad, b->grad, 0);
+            if(beta != 1.0F)
+                _ScaleAndShiftMe(b->grad, beta);
+        }
 
 
         node->grad->Reshape(order, dimSize);
         node->grad->Reshape(order, dimSize);
 
 

source/sample/fnnlm/FNNLM.cpp

@@ -975,7 +975,6 @@ forward process (with tensor connections)
 */
 */
 void ForwardAutoDiff(XTensor inputs[], XTensor &output, FNNModel &model)
 void ForwardAutoDiff(XTensor inputs[], XTensor &output, FNNModel &model)
 {
 {
-    int batchSize = inputs[0].GetDim(0);
     int n = model.n;
     int n = model.n;
     int depth = model.hDepth;
     int depth = model.hDepth;
 
 

source/sample/transformer/T2TAttention.cpp

@@ -19,6 +19,7 @@
  * $Created by: XIAO Tong (xiaotong@mail.neu.edu.cn) 2018-07-31
  * $Created by: XIAO Tong (xiaotong@mail.neu.edu.cn) 2018-07-31
  */
  */
 
 
+#include <math.h>
 #include "T2TAttention.h"
 #include "T2TAttention.h"
 #include "T2TUtility.h"
 #include "T2TUtility.h"
 #include "../../tensor/core/CHeader.h"
 #include "../../tensor/core/CHeader.h"

source/sample/transformer/T2TEmbedding.cpp

@@ -19,6 +19,7 @@
  * $Created by: XIAO Tong (xiaotong@mail.neu.edu.cn) 2018-08-01
  * $Created by: XIAO Tong (xiaotong@mail.neu.edu.cn) 2018-08-01
  */
  */
 
 
+#include <math.h>
 #include "T2TEmbedding.h"
 #include "T2TEmbedding.h"
 #include "T2TUtility.h"
 #include "T2TUtility.h"
 #include "../../tensor/core/CHeader.h"
 #include "../../tensor/core/CHeader.h"
@@ -51,7 +52,6 @@ void T2TEmbedder::InitModel(int argc, const char ** argv, int myDevID, XMem * my
     devID = myDevID;
     devID = myDevID;
     mem = myMem;
     mem = myMem;
     
     
-    float minmax = 0;
     int maxLength = 0;
     int maxLength = 0;
     int d = 0;
     int d = 0;
 
 

source/sample/transformer/T2TEncoder.cpp

@@ -88,20 +88,20 @@ XTensor * AttEncoder::Make(XTensor * input)
 
 
     for(int i = 0; i < nlayer; i++){
     for(int i = 0; i < nlayer; i++){
         XTensor * att;
         XTensor * att;
-        XTensor * res;
         XTensor * ln;
         XTensor * ln;
         XTensor * fnn;
         XTensor * fnn;
+        XTensor res;
 
 
         /* self attention */
         /* self attention */
         att = attentions[i].Make(x, x, x);
         att = attentions[i].Make(x, x, x);
 
 
         /* residual connection */
         /* residual connection */
-        res = &Sum(*att, *x);
+        res = Sum(*att, *x);
 
 
         /* TODO: dropout */
         /* TODO: dropout */
 
 
         /* layer normalization */
         /* layer normalization */
-        ln = layerNorms[i].Make(res);
+        ln = layerNorms[i].Make(&res);
 
 
         /* input of next layer */
         /* input of next layer */
         x = ln;
         x = ln;
@@ -110,12 +110,12 @@ XTensor * AttEncoder::Make(XTensor * input)
         fnn = fnns[i].Make(x);
         fnn = fnns[i].Make(x);
 
 
         /* residual connection */
         /* residual connection */
-        res = &Sum(*fnn, *x);
+        res = Sum(*fnn, *x);
 
 
         /* TODO: dropout */
         /* TODO: dropout */
 
 
         /* layer normalization */
         /* layer normalization */
-        ln = layerNorms[i].Make(res);
+        ln = layerNorms[i].Make(&res);
 
 
         /* input of next layer */
         /* input of next layer */
         x = ln;
         x = ln;