[update] tensor: add more Opt0 contract func thread

tensor_t/src/tensor_t/tensor_t.c

@@ -788,19 +788,19 @@ void tensorContractnProdOpt0_##type(tensor_##type** MM, tensor_##type *M0, tenso
 \
    \
 \
-    size_t a0_id, a1_id, n0_id, n1_id, begin0, begin1;\
+    size_t a0_id, a1_id, n0_id, n1_id;\
     for (size_t i = 0; i < M->dim->rank; i++) {\
         if(endian){\
           a0_id=i/dSub1->rank;\
           a1_id=i%dSub1->rank;\
-          begin0=a0_id*dM->rank ;\
+          n0_id=a0_id*dM->rank ;\
-          begin1= a1_id ;\
+          n1_id= a1_id ;\
         }\
         else{\
           a0_id=i%dSub0->rank;\
           a1_id=i/dSub0->rank;\
-          begin1= a1_id*dM->rank ;\
+          n1_id= a1_id*dM->rank ;\
-          begin0= a0_id ;\
+          n0_id= a0_id ;\
         }\
         M->x[i] = 0;\
         for (size_t k = 0; k < dM->rank; k++) {\
@@ -808,15 +808,15 @@ void tensorContractnProdOpt0_##type(tensor_##type** MM, tensor_##type *M0, tenso
             /*n0_id= a0_id*dM->rank + k;*/\
             /*n1_id= a1_id + dSub1->rank * k;*/\
             /*M->x[i] += M0->x[begin0++] * M1->x[n1_id];*/\
-            M->x[i] += M0->x[begin0++] * M1->x[begin1];\
+            M->x[i] += M0->x[n0_id++] * M1->x[n1_id];\
-            begin1 +=dSub1->rank ;\
+            n1_id +=dSub1->rank ;\
           }\
           else{\
             /*n0_id= a0_id + dSub0->rank * k;*/\
             /*n1_id= a1_id*dM->rank + k;*/\
             /*M->x[i] += M0->x[n0_id] * M1->x[begin1++];*/\
-            M->x[i] += M0->x[begin0] * M1->x[begin1++];\
+            M->x[i] += M0->x[n0_id] * M1->x[n1_id++];\
-            begin0 += dSub0->rank ;\
+            n0_id += dSub0->rank ;\
           }\
             \
         }\
@@ -1051,6 +1051,103 @@ void tensorContractnProdThread_##type(tensor_##type** MM, tensor_##type *M0, ten
   FREE_dM_S_ ; \
 }\
 \
+\
+void* runProdContractOpt0_thread_##type(void *arg){\
+  struct arg_ProdContract_##type *arg_t = arg;\
+  size_t a0_id, a1_id, n0_id, n1_id;\
+    for (size_t i = arg_t->beginRange; i < arg_t->endRange; i++) {\
+        if(endian){\
+          a0_id=i/ arg_t->dSubRank;\
+          a1_id=i% arg_t->dSubRank;\
+            n0_id= a0_id * arg_t->dMRank ;\
+            n1_id= a1_id ;\
+        }\
+        else{\
+          a0_id=i% arg_t->dSubRank;\
+          a1_id=i/ arg_t->dSubRank;\
+            n0_id= a0_id ;\
+            n1_id= a1_id * arg_t->dMRank ;\
+        }\
+        arg_t->Mx[i] = 0;\
+        for (size_t k = 0; k < arg_t->dMRank; k++) {\
+          if(endian){\
+            arg_t->Mx[i] += arg_t->M0x[n0_id++] * arg_t->M1x[n1_id];\
+            n1_id += arg_t->dSubRank ;\
+          }\
+          else{\
+            arg_t->Mx[i] += arg_t->M0x[n0_id] * arg_t->M1x[n1_id];\
+            n0_id += arg_t->dSubRank ;\
+          }\
+        }\
+    }\
+  return 0;\
+}\
+\
+\
+void tensorContractnProdThreadOpt0_##type(tensor_##type** MM, tensor_##type *M0, tensor_##type *M1, size_t contractionNumber, size_t nbthread) {\
+    if(checkContractProdTensorDim(M0->dim, M1->dim, contractionNumber)==0){\
+      printf("checkContractProdTensorDim %ld contractionNumber\n", contractionNumber);\
+        printDebug_dimension(M0->dim, "M0 dim");\
+        printDebug_dimension(M1->dim, "M1 dim");\
+        getchar();\
+    }\
+    size_t len0 = M0->dim->size - contractionNumber;\
+    size_t len1 = M1->dim->size - contractionNumber;\
+\
+    size_t* tsub0 = malloc(sizeof(size_t) *len0);\
+    size_t* tsub1 = malloc(sizeof(size_t) *len1);\
+    size_t* tDk1 = malloc(sizeof(size_t) *contractionNumber);\
+    size_t* tDk0 = malloc(sizeof(size_t) *contractionNumber);\
+    subArray(tsub0, M0->dim->perm, 0, len0, 0);\
+    subArray(tsub1, M1->dim->perm, 0, len1, contractionNumber);\
+    subArray(tDk1, M1->dim->perm, 0, contractionNumber, 0);\
+    subArray(tDk0, M0->dim->perm, 0, contractionNumber, len0);\
+    dimension *dSub0 = init_dim(tsub0, len0);\
+    dimension *dSub1 = init_dim(tsub1, len1);\
+    dimension *dM1 = init_dim(tDk1, contractionNumber);\
+    dimension *dM0 = init_dim(tDk0, contractionNumber);\
+    dimension *dM;\
+    min_copy_dimension(&dM, dM0, dM1);\
+    \
+    dimension *dd;\
+    add_dimension(&dd, dSub0, dSub1);\
+    updateRankDim(dd);\
+    _RECREATE_TENSOR_IF_NOT_THE_SAME_DIM_OR_NULL_##type(MM,dd);\
+    tensor_##type *M= *MM;\
+\
+\
+    \
+  pthread_t *thrd = malloc(nbthread * sizeof(pthread_t));\
+  struct arg_ProdContract_##type **arg_th = malloc( nbthread * sizeof(struct arg_ProdContract_##type *));\
+\
+  for(size_t i = 0; i < nbthread; ++i){\
+    arg_th[i]=malloc(sizeof(struct arg_ProdContract_##type));\
+    arg_th[i]->M0x=M0->x;\
+    arg_th[i]->M1x=M1->x;\
+    arg_th[i]->Mx=M->x;\
+    arg_th[i]->beginRange = i*(M->dim->rank)/nbthread ;\
+    if(i < nbthread - 1 ) arg_th[i]->endRange = (i+1)*(M->dim->rank)/nbthread ;\
+    else arg_th[i]->endRange = M->dim->rank ;\
+    if(endian){\
+      arg_th[i]->dSubRank = dSub1->rank;\
+    }\
+    else{\
+      arg_th[i]->dSubRank = dSub0->rank;\
+    }\
+    arg_th[i]->dMRank = dM->rank;\
+    pthread_create(&thrd[i], NULL, runProdContractOpt0_thread_##type, (void*)arg_th[i]);\
+  }\
+\
+  for(size_t i=0; i< nbthread; ++i){\
+    pthread_join(thrd[i], NULL);\
+    free(arg_th[i]);\
+  }\
+\
+  free(thrd);\
+  free(arg_th);\
+  FREE_dM_S_ ; \
+}\
+\
 struct arg_Pro2dContract_##type{\
   type *M0x;\
   type *M1x;\
@@ -1061,6 +1158,7 @@ struct arg_Pro2dContract_##type{\
   size_t dSub0Rank;\
   size_t dSub1Rank;\
 };\
+\
 void* runPro2dContract_thread_##type(void *arg){\
   struct arg_Pro2dContract_##type *arg_t = arg;\
   size_t n0_id, n1_id, l;\
@@ -1152,6 +1250,106 @@ void tensorContractnPro2dThread_##type(tensor_##type** MM, tensor_##type *M0, te
   free(arg_th);\
   FREE_dM_S_ ; \
 }\
+\
+void* runPro2dContractOpt0_thread_##type(void *arg){\
+  struct arg_Pro2dContract_##type *arg_t = arg;\
+  size_t n0_id, n1_id, l;\
+    for (size_t i = arg_t->beginRange; i < arg_t->endRange; i++) {\
+      for (size_t j = 0; j < arg_t->dSub1Rank; j++) {\
+        if(endian){\
+          l = j + arg_t->dSub1Rank * i;\
+            n0_id= i * arg_t->dMRank ;\
+            n1_id= j ;\
+        }else{\
+          l = j * arg_t->dSub0Rank + i;\
+            n0_id= i ;\
+            n1_id= j * arg_t->dMRank ;\
+        }\
+        arg_t->Mx[l] = 0;\
+        for (size_t k = 0; k < arg_t->dMRank; k++) {\
+          if(endian){\
+            /*n0_id= i * arg_t->dMRank + k;\
+            n1_id= j + arg_t->dSub1Rank * k;*/\
+            arg_t->Mx[l] += arg_t->M0x[n0_id++] * arg_t->M1x[n1_id];\
+            n1_id += arg_t->dSub1Rank ;\
+          }\
+          else{\
+            /*n0_id= i + arg_t->dSub0Rank * k;\
+            n1_id= j * arg_t->dMRank + k;*/\
+            arg_t->Mx[l] += arg_t->M0x[n0_id] * arg_t->M1x[n1_id];\
+            n0_id += arg_t->dSub0Rank ;\
+          }\
+        }\
+      }\
+    }\
+  return 0;\
+}\
+/* M[x0,x1,x3..xn] X M[y0,y1,y3..ym] = M[z0,z1...zp] (deep = l > 0) /exists 1<= l<...<l=n /  xl = y0,x{l+1}=y1, x{n}=yl  et zi=xi i<n-l et zj=y{j-(n-l)} j>=n-l alor p=n+m-2l\
+ M[x0,x1,x3..xl x{l+1}...xn] X M[xn,x{n-1},x{n-2}...xl y{l+1} ..ym] = M[x0,x1..xly{l+1}...y{n+m-2l}] (deep = l > 0)\
+M[[i][j]]=sum_{[k]}M0[[i][k]]*M[[k][j]]*/\
+\
+void tensorContractnPro2dThreadOpt0_##type(tensor_##type** MM, tensor_##type *M0, tensor_##type *M1, size_t contractionNumber, size_t nbthread) {\
+    /*if(checkContractProdTensorDim(M0->dim, M1->dim, contractionNumber)==0){\
+      printf("checkContractProdTensorDim %ld contractionNumber\n", contractionNumber);\
+    }*/\
+    if(checkContractProdTensorDim(M0->dim, M1->dim, contractionNumber)==0){\
+      printf("checkContractProdTensorDim %ld contractionNumber\n", contractionNumber);\
+        printDebug_dimension(M0->dim, "M0 dim");\
+        printDebug_dimension(M1->dim, "M1 dim");\
+        getchar();\
+    }\
+\
+    size_t len0 = M0->dim->size - contractionNumber;\
+    size_t len1 = M1->dim->size - contractionNumber;\
+\
+    size_t* tsub0 = malloc(sizeof(size_t) *len0);\
+    size_t* tsub1 = malloc(sizeof(size_t) *len1);\
+    size_t* tDk1 = malloc(sizeof(size_t) *contractionNumber);\
+    size_t* tDk0 = malloc(sizeof(size_t) *contractionNumber);\
+    subArray(tsub0, M0->dim->perm, 0, len0, 0);\
+    subArray(tsub1, M1->dim->perm, 0, len1, contractionNumber);\
+    subArray(tDk1, M1->dim->perm, 0, contractionNumber, 0);\
+    subArray(tDk0, M0->dim->perm, 0, contractionNumber, len0);\
+    dimension *dSub0 = init_dim(tsub0, len0);\
+    dimension *dSub1 = init_dim(tsub1, len1);\
+    dimension *dM1 = init_dim(tDk1, contractionNumber);\
+    dimension *dM0 = init_dim(tDk0, contractionNumber);\
+    dimension *dM;\
+    min_copy_dimension(&dM, dM0, dM1);\
+    \
+    dimension *dd;\
+    add_dimension(&dd, dSub0, dSub1);\
+    updateRankDim(dd);\
+    _RECREATE_TENSOR_IF_NOT_THE_SAME_DIM_OR_NULL_##type(MM,dd);\
+    tensor_##type *M= *MM;\
+\
+\
+    \
+  pthread_t *thrd = malloc(nbthread * sizeof(pthread_t));\
+  struct arg_Pro2dContract_##type **arg_th = malloc( nbthread * sizeof(struct arg_Pro2dContract_##type *));\
+\
+  for(size_t i = 0; i < nbthread; ++i) {\
+    arg_th[i] = malloc(sizeof(struct arg_Pro2dContract_##type));\
+    arg_th[i]->M0x=M0->x;\
+    arg_th[i]->M1x=M1->x;\
+    arg_th[i]->Mx=M->x;\
+    arg_th[i]->beginRange = i*(dSub0->rank)/nbthread ;\
+    arg_th[i]->endRange = (i+1)*(dSub0->rank)/nbthread ;\
+    arg_th[i]->dSub1Rank = dSub1->rank;\
+    arg_th[i]->dSub0Rank = dSub0->rank;\
+    arg_th[i]->dMRank = dM->rank;\
+    pthread_create(&thrd[i], NULL, runPro2dContractOpt0_thread_##type, (void*)arg_th[i]);\
+  }\
+\
+  for(size_t i=0; i< nbthread; ++i){\
+    pthread_join(thrd[i], NULL);\
+    free(arg_th[i]);\
+  }\
+\
+  free(thrd);\
+  free(arg_th);\
+  FREE_dM_S_ ; \
+}\
 void tensorContractnProdNotOpt_##type(tensor_##type** MM, tensor_##type *M0, tensor_##type *M1, size_t contractionNumber) {\
     if (!checkContractProdTensorDim(M0->dim, M1->dim, contractionNumber)) {\
         printf("error Deep = %ld\n", contractionNumber);\

tensor_t/src/tensor_t/tensor_t.h

@@ -40,7 +40,9 @@ void tensorContractnProd_##type(tensor_##type **MM, tensor_##type *M0, tensor_##
 void tensorProdThread_##type(tensor_##type **MM, tensor_##type *M0, tensor_##type *M1,size_t nbthread); \
 void tensorProdThrea2d_##type(tensor_##type **MM, tensor_##type *M0, tensor_##type *M1,size_t nbthread); \
 void tensorContractnProdThread_##type(tensor_##type **MM, tensor_##type *M0, tensor_##type *M1, size_t contractionNumber, size_t nbthread); \
+void tensorContractnProdThreadOpt0_##type(tensor_##type **MM, tensor_##type *M0, tensor_##type *M1, size_t contractionNumber, size_t nbthread); \
 void tensorContractnPro2dThread_##type(tensor_##type **MM, tensor_##type *M0, tensor_##type *M1, size_t contractionNumber, size_t nbthread); \
+void tensorContractnPro2dThreadOpt0_##type(tensor_##type **MM, tensor_##type *M0, tensor_##type *M1, size_t contractionNumber, size_t nbthread); \
 void tensorContractnProdNotOpt_##type(tensor_##type **MM, tensor_##type *M0, tensor_##type *M1, size_t contractionNumber); \
 void tensorContractnProdOpt0_##type(tensor_##type **MM, tensor_##type *M0, tensor_##type *M1, size_t contractionNumber); \
 type scalarProduct_dep_contractProd_##type(tensor_##type *M0, tensor_##type *M1, size_t nbthreads ,void (*tensorContractVar)(tensor_##type **MM, tensor_##type *M0, tensor_##type *M1, size_t contractionNumber, size_t nbthread ));\

tensor_t/test/is_good.c

@@ -1204,6 +1204,309 @@ TEST(tensorContractnProdOpt0_TYPE_FLOAT2 ){
   free_tensor_TYPE_FLOAT(M1);
 
 }
+
+
+TEST(tensorContractnProdThreadOpt0_TYPE_FLOAT2 ){
+  dimension *d0=create_dim(3);
+  dimension *d1=create_dim(3);
+#if VALGRIND_
+  d0->perm[0]=5;
+  d0->perm[1]=2; //3;
+  d0->perm[2]=3;
+
+  d1->perm[0]=2;
+  d1->perm[1]=3;//3;
+  d1->perm[2]=8;
+
+#else
+
+
+  d0->perm[0]=335;
+  d0->perm[1]=32; //3;
+  d0->perm[2]=43;
+
+  d1->perm[0]=32;
+  d1->perm[1]=43;//3;
+  d1->perm[2]=244;
+#endif
+
+  updateRankDim(d0);
+  updateRankDim(d1);
+
+
+  tensor_TYPE_FLOAT *M0 = CREATE_TENSOR_TYPE_FLOAT(d0);
+  tensor_TYPE_FLOAT *M1 = CREATE_TENSOR_TYPE_FLOAT(d1);
+
+  LOG("M0->dim->rank = %ld\n",M0->dim->rank);
+  LOG("M1->dim->rank = %ld\n",M1->dim->rank);
+  for(size_t i=0; i<M0->dim->rank;++i) M0->x[i]=i*0.1 +1;
+  for(size_t i=0; i<M1->dim->rank;++i) M1->x[i]=i*0.003 + 2;
+
+//  print_tensor_float(M0,"M0");
+//  print_tensor_float(M1,"M1");
+
+  tensor_TYPE_FLOAT *M=NULL;
+  tensor_TYPE_FLOAT *MnO=NULL;
+
+  tensorContractnProdThreadOpt0_TYPE_FLOAT(&M, M0,M1,2,8);
+//  print_tensor_float(M,"M");
+  tensorContractnProdNotOpt_TYPE_FLOAT(&MnO, M0,M1,2);
+
+
+//  print_tensor_float(MnO,"MnO");
+ 
+  // for(size_t i=0;i<M->dim->rank;++i)
+  //    EXPECT_EQ_TYPE_FLOAT(M->x[i],MnO->x[i]);
+    
+  EXPECT_ARRAY_EQ_TYPE_FLOAT(M->x,M->dim->rank,MnO->x,MnO->dim->rank);
+
+  free_tensor_TYPE_FLOAT(M);
+  free_tensor_TYPE_FLOAT(MnO);
+  free_tensor_TYPE_FLOAT(M0);
+  free_tensor_TYPE_FLOAT(M1);
+
+}
+
+TEST(tensorContractnPro2dThreadOpt0_TYPE_FLOAT2 ){
+  dimension *d0=create_dim(3);
+  dimension *d1=create_dim(3);
+#if VALGRIND_
+  d0->perm[0]=5;
+  d0->perm[1]=2; //3;
+  d0->perm[2]=3;
+
+  d1->perm[0]=2;
+  d1->perm[1]=3;//3;
+  d1->perm[2]=8;
+
+#else
+
+
+  d0->perm[0]=335;
+  d0->perm[1]=32; //3;
+  d0->perm[2]=43;
+
+  d1->perm[0]=32;
+  d1->perm[1]=43;//3;
+  d1->perm[2]=244;
+#endif
+
+  updateRankDim(d0);
+  updateRankDim(d1);
+
+
+  tensor_TYPE_FLOAT *M0 = CREATE_TENSOR_TYPE_FLOAT(d0);
+  tensor_TYPE_FLOAT *M1 = CREATE_TENSOR_TYPE_FLOAT(d1);
+
+  LOG("M0->dim->rank = %ld\n",M0->dim->rank);
+  LOG("M1->dim->rank = %ld\n",M1->dim->rank);
+  for(size_t i=0; i<M0->dim->rank;++i) M0->x[i]=i*0.1 +1;
+  for(size_t i=0; i<M1->dim->rank;++i) M1->x[i]=i*0.003 + 2;
+
+//  print_tensor_float(M0,"M0");
+//  print_tensor_float(M1,"M1");
+
+  tensor_TYPE_FLOAT *M=NULL;
+  tensor_TYPE_FLOAT *MnO=NULL;
+
+  tensorContractnPro2dThreadOpt0_TYPE_FLOAT(&M, M0,M1,2,8);
+//  print_tensor_float(M,"M");
+  tensorContractnProdNotOpt_TYPE_FLOAT(&MnO, M0,M1,2);
+
+
+//  print_tensor_float(MnO,"MnO");
+ 
+  // for(size_t i=0;i<M->dim->rank;++i)
+  //    EXPECT_EQ_TYPE_FLOAT(M->x[i],MnO->x[i]);
+    
+  EXPECT_ARRAY_EQ_TYPE_FLOAT(M->x,M->dim->rank,MnO->x,MnO->dim->rank);
+
+  free_tensor_TYPE_FLOAT(M);
+  free_tensor_TYPE_FLOAT(MnO);
+  free_tensor_TYPE_FLOAT(M0);
+  free_tensor_TYPE_FLOAT(M1);
+
+}
+
+TEST(tensorContractnPro2dThread_TYPE_FLOAT2 ){
+  dimension *d0=create_dim(3);
+  dimension *d1=create_dim(3);
+#if VALGRIND_
+  d0->perm[0]=5;
+  d0->perm[1]=2; //3;
+  d0->perm[2]=3;
+
+  d1->perm[0]=2;
+  d1->perm[1]=3;//3;
+  d1->perm[2]=8;
+
+#else
+
+
+  d0->perm[0]=335;
+  d0->perm[1]=32; //3;
+  d0->perm[2]=43;
+
+  d1->perm[0]=32;
+  d1->perm[1]=43;//3;
+  d1->perm[2]=244;
+#endif
+
+  updateRankDim(d0);
+  updateRankDim(d1);
+
+
+  tensor_TYPE_FLOAT *M0 = CREATE_TENSOR_TYPE_FLOAT(d0);
+  tensor_TYPE_FLOAT *M1 = CREATE_TENSOR_TYPE_FLOAT(d1);
+
+  LOG("M0->dim->rank = %ld\n",M0->dim->rank);
+  LOG("M1->dim->rank = %ld\n",M1->dim->rank);
+  for(size_t i=0; i<M0->dim->rank;++i) M0->x[i]=i*0.1 +1;
+  for(size_t i=0; i<M1->dim->rank;++i) M1->x[i]=i*0.003 + 2;
+
+//  print_tensor_float(M0,"M0");
+//  print_tensor_float(M1,"M1");
+
+  tensor_TYPE_FLOAT *M=NULL;
+  tensor_TYPE_FLOAT *MnO=NULL;
+
+  tensorContractnPro2dThread_TYPE_FLOAT(&M, M0,M1,2,8);
+//  print_tensor_float(M,"M");
+  tensorContractnProdNotOpt_TYPE_FLOAT(&MnO, M0,M1,2);
+
+
+//  print_tensor_float(MnO,"MnO");
+ 
+  // for(size_t i=0;i<M->dim->rank;++i)
+  //    EXPECT_EQ_TYPE_FLOAT(M->x[i],MnO->x[i]);
+    
+  EXPECT_ARRAY_EQ_TYPE_FLOAT(M->x,M->dim->rank,MnO->x,MnO->dim->rank);
+
+  free_tensor_TYPE_FLOAT(M);
+  free_tensor_TYPE_FLOAT(MnO);
+  free_tensor_TYPE_FLOAT(M0);
+  free_tensor_TYPE_FLOAT(M1);
+
+}
+
+
+
+TEST(tensorContractnProdThread_TYPE_FLOAT2 ){
+  dimension *d0=create_dim(3);
+  dimension *d1=create_dim(3);
+#if VALGRIND_
+  d0->perm[0]=5;
+  d0->perm[1]=2; //3;
+  d0->perm[2]=3;
+
+  d1->perm[0]=2;
+  d1->perm[1]=3;//3;
+  d1->perm[2]=8;
+
+#else
+
+
+  d0->perm[0]=335;
+  d0->perm[1]=32; //3;
+  d0->perm[2]=43;
+
+  d1->perm[0]=32;
+  d1->perm[1]=43;//3;
+  d1->perm[2]=244;
+#endif
+
+  updateRankDim(d0);
+  updateRankDim(d1);
+
+
+  tensor_TYPE_FLOAT *M0 = CREATE_TENSOR_TYPE_FLOAT(d0);
+  tensor_TYPE_FLOAT *M1 = CREATE_TENSOR_TYPE_FLOAT(d1);
+
+  LOG("M0->dim->rank = %ld\n",M0->dim->rank);
+  LOG("M1->dim->rank = %ld\n",M1->dim->rank);
+  for(size_t i=0; i<M0->dim->rank;++i) M0->x[i]=i*0.1 +1;
+  for(size_t i=0; i<M1->dim->rank;++i) M1->x[i]=i*0.003 + 2;
+
+//  print_tensor_float(M0,"M0");
+//  print_tensor_float(M1,"M1");
+
+  tensor_TYPE_FLOAT *M=NULL;
+  tensor_TYPE_FLOAT *MnO=NULL;
+
+  tensorContractnProdThread_TYPE_FLOAT(&M, M0,M1,2,8);
+//  print_tensor_float(M,"M");
+  tensorContractnProdNotOpt_TYPE_FLOAT(&MnO, M0,M1,2);
+
+
+//  print_tensor_float(MnO,"MnO");
+ 
+  // for(size_t i=0;i<M->dim->rank;++i)
+  //    EXPECT_EQ_TYPE_FLOAT(M->x[i],MnO->x[i]);
+    
+  EXPECT_ARRAY_EQ_TYPE_FLOAT(M->x,M->dim->rank,MnO->x,MnO->dim->rank);
+
+  free_tensor_TYPE_FLOAT(M);
+  free_tensor_TYPE_FLOAT(MnO);
+  free_tensor_TYPE_FLOAT(M0);
+  free_tensor_TYPE_FLOAT(M1);
+
+}
+TEST(tensorContractnProd_TYPE_DOUBLE_2_1 ){
+  dimension *d0=create_dim(2);
+  dimension *d1=create_dim(1);
+#if VALGRIND_
+  d0->perm[0]=4;
+  d0->perm[1]=2; //3;
+
+  d1->perm[0]=2;
+
+#else
+
+  d0->perm[0]=125;
+  d0->perm[1]=52; //3;
+
+  d1->perm[0]=52;
+#endif
+
+  updateRankDim(d0);
+  updateRankDim(d1);
+
+
+  tensor_TYPE_DOUBLE *M0 = CREATE_TENSOR_TYPE_DOUBLE(d0);
+  tensor_TYPE_DOUBLE *M1 = CREATE_TENSOR_TYPE_DOUBLE(d1);
+
+  LOG("M0->dim->rank = %ld\n",M0->dim->rank);
+  LOG("M1->dim->rank = %ld\n",M1->dim->rank);
+  for(size_t i=0; i<M0->dim->rank;++i) M0->x[i]=i*0.1 +1;
+  for(size_t i=0; i<M1->dim->rank;++i) M1->x[i]=i*0.003 + 2;
+
+  print_tensor_double(M0,"M0");
+  print_tensor_double(M1,"M1");
+
+  tensor_TYPE_DOUBLE *M=NULL;
+  tensor_TYPE_DOUBLE *MnO=NULL;
+
+  tensorContractnProd_TYPE_DOUBLE(&M, M0,M1,1);
+  //print_tensor_double(M,"M");
+  //cl_tensorContractnProd_TYPE_DOUBLE(&MnO, M0,M1,2);
+  tensorContractnProdNotOpt_TYPE_DOUBLE(&MnO, M0,M1,1);
+
+  print_tensor_double(MnO,"MnO");
+ 
+  // for(size_t i=0;i<M->dim->rank;++i)
+  //    EXPECT_EQ_TYPE_DOUBLE(M->x[i],MnO->x[i]);
+    
+  EXPECT_ARRAY_EQ_TYPE_DOUBLE(M->x,M->dim->rank,MnO->x,MnO->dim->rank);
+
+  free_tensor_TYPE_DOUBLE(M);
+  free_tensor_TYPE_DOUBLE(MnO);
+  free_tensor_TYPE_DOUBLE(M0);
+  free_tensor_TYPE_DOUBLE(M1);
+
+}
+
+
+
 TEST(tensorContractnProd_TYPE_DOUBLE_2_1 ){
   dimension *d0=create_dim(2);
   dimension *d1=create_dim(1);