[OPENMP] Pass scalar firstprivate vars by value.

For better performance and to unify code with offloading part we pass
scalar firstprivate values by value, instead of by reference. It will
remove some extra copying operations.

llvm-svn: 269751

13 files changed, 185 insertions, 245 deletions

diff --git a/clang/test/OpenMP/for_firstprivate_codegen.cpp b/clang/test/OpenMP/for_firstprivate_codegen.cpp
index 5b1a4069410..e0380e3b02c 100644
--- a/clang/test/OpenMP/for_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/for_firstprivate_codegen.cpp
@@ -263,30 +263,30 @@ int main() {
 // CHECK: define {{.*}} i{{[0-9]+}} [[TMAIN_INT]]()
 // CHECK: [[TEST:%.+]] = alloca [[S_INT_TY]],
 // CHECK: call {{.*}} [[S_INT_TY_DEF_CONSTR:@.+]]([[S_INT_TY]]* [[TEST]])
-// CHECK: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{[0-9]+}} 4, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, i32*, [2 x i32]*, [2 x [[S_INT_TY]]]*, [[S_INT_TY]]*)* [[TMAIN_MICROTASK:@.+]] to void
+// CHECK: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{[0-9]+}} 4, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, i32, [2 x i32]*, [2 x [[S_INT_TY]]]*, [[S_INT_TY]]*)* [[TMAIN_MICROTASK:@.+]] to void
 // CHECK: call {{.*}} [[S_INT_TY_DESTR:@.+]]([[S_INT_TY]]*
 // CHECK: ret
 //
-// CHECK: define internal void [[TMAIN_MICROTASK]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, i32* dereferenceable(4) %{{.+}}, [2 x i32]* dereferenceable(8) %{{.+}}, [2 x [[S_INT_TY]]]* dereferenceable(8) %{{.+}}, [[S_INT_TY]]* dereferenceable(4) %{{.+}})
+// CHECK: define internal void [[TMAIN_MICROTASK]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, i32 {{.*}}%{{.+}}, [2 x i32]* dereferenceable(8) %{{.+}}, [2 x [[S_INT_TY]]]* dereferenceable(8) %{{.+}}, [[S_INT_TY]]* dereferenceable(4) %{{.+}})
 // Skip temp vars for loop
+// CHECK: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}},
 // CHECK: alloca i{{[0-9]+}},
 // CHECK: alloca i{{[0-9]+}},
 // CHECK: alloca i{{[0-9]+}},
 // CHECK: alloca i{{[0-9]+}},
 // CHECK: alloca i{{[0-9]+}},
-// CHECK: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}},
 // CHECK: [[VEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}],
 // CHECK: [[S_ARR_PRIV:%.+]] = alloca [2 x [[S_INT_TY]]],
 // CHECK: [[VAR_PRIV:%.+]] = alloca [[S_INT_TY]],
 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]],
 
-// CHECK: [[T_VAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** %
+// CHECK-NOT: load i{{[0-9]+}}*, i{{[0-9]+}}** %
 // CHECK: [[VEC_REF:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** %
 // CHECK: [[S_ARR:%.+]] = load [2 x [[S_INT_TY]]]*, [2 x [[S_INT_TY]]]** %
+// CHECK: [[VAR:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** %
 
 // firstprivate t_var(t_var)
-// CHECK: [[T_VAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_REF]],
-// CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_PRIV]],
+// CHECK-NOT: load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_REF]],
 
 // firstprivate vec(vec)
 // CHECK: [[VEC_DEST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8*
diff --git a/clang/test/OpenMP/nvptx_target_firstprivate_codegen.cpp b/clang/test/OpenMP/nvptx_target_firstprivate_codegen.cpp
index f7b400e5986..66f8c2f9380 100644
--- a/clang/test/OpenMP/nvptx_target_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_target_firstprivate_codegen.cpp
@@ -30,12 +30,8 @@ int foo(int n, double *ptr) {
   
   // TCHECK:  define void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A_IN:%.+]])
   // TCHECK:  [[A_ADDR:%.+]] = alloca i{{[0-9]+}},
-  // TCHECK:  [[A1:%.+]] = alloca i{{[0-9]+}},
+  // TCHECK-NOT:  alloca i{{[0-9]+}},
   // TCHECK:  store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[A_ADDR]],
-  // TCHECK-64:  [[CONV:%.+]] = bitcast i{{[0-9]+}}* [[A_ADDR]] to i{{[0-9]+}}*
-  // TCHECK-64:  [[A_ADDR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[CONV]],
-  // TCHECK-32:  [[A_ADDR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[A_ADDR]],
-  // TCHECK:  store i{{[0-9]+}} [[A_ADDR_VAL]], i{{[0-9]+}}* [[A1]],
   // TCHECK:  ret void  
 
 #pragma omp target firstprivate(aa,b,c,d)
@@ -54,7 +50,7 @@ int foo(int n, double *ptr) {
   // TCHECK:  [[B_ADDR:%.+]] = alloca [10 x float]*,
   // TCHECK:  [[C_ADDR:%.+]] = alloca [5 x [10 x double]]*,
   // TCHECK:  [[D_ADDR:%.+]] = alloca [[TT]]*,
-  // TCHECK:  [[A2_PRIV:%.+]] = alloca i{{[0-9]+}},
+  // TCHECK-NOT: alloca i{{[0-9]+}},
   // TCHECK:  [[B_PRIV:%.+]] = alloca [10 x float],
   // TCHECK:  [[C_PRIV:%.+]] = alloca [5 x [10 x double]],
   // TCHECK:  [[D_PRIV:%.+]] = alloca [[TT]],
@@ -68,8 +64,6 @@ int foo(int n, double *ptr) {
   // TCHECK:  [[D_ADDR_REF:%.+]] = load [[TT]]*, [[TT]]** [[D_ADDR]],
 
   // firstprivate(aa): a_priv = a_in
-  // TCHECK:  [[A2_CONV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[CONV_A2ADDR]],
-  // TCHECK:  store i{{[0-9]+}} [[A2_CONV_VAL]], i{{[0-9]+}}* [[A2_PRIV]],
 
   //  firstprivate(b): memcpy(b_priv,b_in)
   // TCHECK:  [[B_PRIV_BCAST:%.+]] = bitcast [10 x float]* [[B_PRIV]] to i8*
@@ -86,6 +80,8 @@ int foo(int n, double *ptr) {
   // TCHECK:  [[D_IN_BCAST:%.+]] = bitcast [[TT]]* [[D_ADDR_REF]] to i8*
   // TCHECK:  call void @llvm.memcpy.{{.+}}(i8* [[D_PRIV_BCAST]], i8* [[D_IN_BCAST]],{{.+}})
 
+  // TCHECK: load i16, i16* [[CONV_A2ADDR]],
+
   
   #pragma omp target firstprivate(ptr)
   {
@@ -94,10 +90,10 @@ int foo(int n, double *ptr) {
 
   // TCHECK:  define void @__omp_offloading_{{.+}}(double* [[PTR_IN:%.+]])
   // TCHECK:  [[PTR_ADDR:%.+]] = alloca double*,
-  // TCHECK:  [[PTR_PRIV:%.+]] = alloca double*,
+  // TCHECK-NOT: alloca double*,
   // TCHECK:  store double* [[PTR_IN]], double** [[PTR_ADDR]],
   // TCHECK:  [[PTR_IN_REF:%.+]] = load double*, double** [[PTR_ADDR]],
-  // TCHECK:  store double* [[PTR_IN_REF]], double** [[PTR_PRIV]],
+  // TCHECK-NOT:  store double* [[PTR_IN_REF]], double** [[PTR_PRIV]],
 
   return a;
 }
@@ -137,8 +133,7 @@ int fstatic(int n) {
 // TCHECK:  [[A_ADDR:%.+]] = alloca i{{[0-9]+}},
 // TCHECK:  [[A3_ADDR:%.+]] = alloca i{{[0-9]+}},
 // TCHECK:  [[B_ADDR:%.+]] = alloca [10 x i{{[0-9]+}}]*,
-// TCHECK:  [[A_PRIV:%.+]] = alloca i{{[0-9]+}},
-// TCHECK:  [[A3_PRIV:%.+]] = alloca i{{[0-9]+}},
+// TCHECK-NOT:  alloca i{{[0-9]+}},
 // TCHECK:  [[B_PRIV:%.+]] = alloca [10 x i{{[0-9]+}}],
 // TCHECK:  store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[A_ADDR]],
 // TCHECK:  store i{{[0-9]+}} [[A3_IN]], i{{[0-9]+}}* [[A3_ADDR]],
@@ -148,13 +143,10 @@ int fstatic(int n) {
 // TCHECK:  [[B_ADDR_REF:%.+]] = load [10 x i{{[0-9]+}}]*, [10 x i{{[0-9]+}}]** [[B_ADDR]],
 
 // firstprivate(a): a_priv = a_in
-// TCHECK-64:  [[A_IN_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[A_CONV]],
-// TCHECK-32:  [[A_IN_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[A_ADDR]],
-// TCHECK:  store i{{[0-9]+}} [[A_IN_VAL]], i{{[0-9]+}}* [[A_PRIV]],
 
 // firstprivate(aaa)
-// TCHECK:  [[A3_IN_VAL:%.+]] = load i8, i8* [[A3_CONV]],
-// TCHECK:  store i{{[0-9]+}} [[A3_IN_VAL]], i{{[0-9]+}}* [[A3_PRIV]],
+
+// TCHECK-NOT:  store i{{[0-9]+}} %{{.+}}, i{{[0-9]+}}*
 
 // firstprivate(b)
 // TCHECK:  [[B_PRIV_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_PRIV]] to i8*
@@ -180,7 +172,7 @@ struct S1 {
   // TCHECK: define void @__omp_offloading_{{.+}}([[S1]]* [[TH:%.+]], i{{[0-9]+}} [[B_IN:%.+]])
   // TCHECK:  [[TH_ADDR:%.+]] = alloca [[S1]]*,
   // TCHECK:  [[B_ADDR:%.+]] = alloca i{{[0-9]+}},
-  // TCHECK:  [[B_PRIV:%.+]] = alloca i{{[0-9]+}},
+  // TCHECK-NOT: alloca i{{[0-9]+}},
 
   // TCHECK:  store [[S1]]* [[TH]], [[S1]]** [[TH_ADDR]],
   // TCHECK:  store i{{[0-9]+}} [[B_IN]], i{{[0-9]+}}* [[B_ADDR]],
@@ -188,9 +180,7 @@ struct S1 {
   // TCHECK-64:  [[B_ADDR_CONV:%.+]] = bitcast i{{[0-9]+}}* [[B_ADDR]] to i{{[0-9]+}}*
 
   // firstprivate(b)
-  // TCHECK-64:  [[B_IN_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[B_ADDR_CONV]],
-  // TCHECK-32:  [[B_IN_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[B_ADDR]],
-  // TCHECK:  store i{{[0-9]+}} [[B_IN_VAL]], i{{[0-9]+}}* [[B_PRIV]], 
+  // TCHECK-NOT:  store i{{[0-9]+}} %{{.+}}, i{{[0-9]+}}*
 
   // TCHECK: ret void
 };
@@ -213,7 +203,7 @@ int bar(int n, double *ptr){
 // TCHECK: define void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A_IN:%.+]], [10 x i{{[0-9]+}}]*{{.+}} [[B_IN:%.+]])
 // TCHECK:  [[A_ADDR:%.+]] = alloca i{{[0-9]+}},
 // TCHECK:  [[B_ADDR:%.+]] = alloca [10 x i{{[0-9]+}}]*,
-// TCHECK:  [[A_PRIV:%.+]] = alloca i{{[0-9]+}},
+// TCHECK-NOT: alloca i{{[0-9]+}},
 // TCHECK:  [[B_PRIV:%.+]] = alloca [10 x i{{[0-9]+}}],
 // TCHECK:  store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[A_ADDR]],
 // TCHECK:  store [10 x i{{[0-9]+}}]* [[B_IN]], [10 x i{{[0-9]+}}]** [[B_ADDR]],
@@ -221,9 +211,7 @@ int bar(int n, double *ptr){
 // TCHECK:  [[B_ADDR_REF:%.+]] = load [10 x i{{[0-9]+}}]*, [10 x i{{[0-9]+}}]** [[B_ADDR]],
 
 // firstprivate(a)
-// TCHECK-64:  [[A_IN_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[A_ADDR_CONV]]
-// TCHECK-32:  [[A_IN_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[A_ADDR]]
-// TCHECK:  store i{{[0-9]+}} [[A_IN_VAL]], i{{[0-9]+}}* [[A_PRIV]],
+// TCHECK-NOT:  store i{{[0-9]+}} %{{.+}}, i{{[0-9]+}}*
 
 // firstprivate(b)
 // TCHECK:  [[B_PRIV_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_PRIV]] to i8*
diff --git a/clang/test/OpenMP/nvptx_teams_codegen.cpp b/clang/test/OpenMP/nvptx_teams_codegen.cpp
index f59c7ba3d89..170c1303ec4 100644
--- a/clang/test/OpenMP/nvptx_teams_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_teams_codegen.cpp
@@ -45,13 +45,12 @@ int main (int argc, char **argv) {
 // CK1-NEXT: }
 
 // target region in template
-// CK1: define {{.*}}void @{{[^,]+}}(i{{.+}}***{{.+}} [[ARGC:%.+]])
+// CK1: define {{.*}}void @{{[^,]+}}(i{{.+}}** [[ARGC:%.+]])
 // CK1: [[ARGCADDR_PTR:%.+]] = alloca i{{.+}}***,
-// CK1: [[ARGCADDR:%.+]] = alloca i{{.+}}***,
-// CK1: store i{{.+}}*** [[ARGC]], i{{.+}}**** [[ARGCADDR]]
-// CK1: [[ARGCADDR_REF:%.+]] = load i{{.+}}***, i{{.+}}**** [[ARGCADDR]],
-// CK1: store i8*** [[ARGCADDR_REF]], i8**** [[ARGCADDR_PTR]],
-// CK1: [[ARGCADDR_PTR_REF:%.+]] = load i{{.+}}***, i{{.+}}**** [[ARGCADDR_PTR]],
+// CK1: [[ARGCADDR:%.+]] = alloca i{{.+}}**,
+// CK1: store i{{.+}}** [[ARGC]], i{{.+}}*** [[ARGCADDR]]
+// CK1: store i8*** [[ARGCADDR]], i8**** [[ARGCADDR_PTR]],
+// CK1: [[ARGCADDR_PTR_REF:%.+]] = load i{{.+}}**, i{{.+}}*** [[ARGCADDR_PTR]],
 // CK1: store i{{[0-9]+}}** null, i{{[0-9]+}}*** [[ARGCADDR_PTR_REF]],
 // CK1-NOT: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(
 // CK1:  ret void
@@ -113,19 +112,16 @@ int main (int argc, char **argv) {
 // CK2-NOT:  call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(
 // CK2: ret
 
-// CK2: define {{.*}}void @{{[^,]+}}(i{{[0-9]+}}*{{.+}} [[A_IN:%.+]], i{{[0-9]+}}*{{.+}} [[BP:%.+]], i{{[0-9]+}}***{{.+}}  [[ARGC:%.+]])
+// CK2: define {{.*}}void @{{[^,]+}}(i{{[0-9]+}} [[A_IN:%.+]], i{{[0-9]+}} [[BP:%.+]], i{{[0-9]+}}** [[ARGC:%.+]])
 // CK2: [[ARGCADDR_PTR:%.+]] = alloca i{{[0-9]+}}***,
-// CK2: [[AADDR:%.+]] = alloca i{{[0-9]+}}*,
-// CK2: [[BADDR:%.+]] = alloca i{{[0-9]+}}*,
-// CK2: [[ARGCADDR:%.+]] = alloca i{{[0-9]+}}***,
+// CK2: [[AADDR:%.+]] = alloca i{{[0-9]+}},
+// CK2: [[BADDR:%.+]] = alloca i{{[0-9]+}},
+// CK2: [[ARGCADDR:%.+]] = alloca i{{[0-9]+}}**,
 // CK2-NOT: {{%.+}} = call i32 @__kmpc_global_thread_num(
-// CK2: store i{{[0-9]+}}* [[A_IN]], i{{[0-9]+}}** [[AADDR]],
-// CK2: store i{{[0-9]+}}* [[B_IN]], i{{[0-9]+}}** [[BADDR]],
-// CK2: store i{{[0-9]+}}*** [[ARGC]], i{{[0-9]+}}**** [[ARGCADDR]],
-// CK2: [[A_ADDR_VAL:%.+]] = load i32*, i32** [[AADDR]]
-// CK2: [[B_ADDR_VAL:%.+]] = load i32*, i32** [[BADDR]]
-// CK2: [[ARGC_ADDR_VAL:%.+]] = load i{{[0-9]+}}***, i{{[0-9]+}}**** [[ARGCADDR]]
-// CK2: store i{{[0-9]+}}*** [[ARGC_ADDR_VAL]], i{{[0-9]+}}**** [[ARGCADDR_PTR]],
+// CK2: store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[AADDR]],
+// CK2: store i{{[0-9]+}} [[B_IN]], i{{[0-9]+}}* [[BADDR]],
+// CK2: store i{{[0-9]+}}** [[ARGC]], i{{[0-9]+}}*** [[ARGCADDR]],
+// CK2: store i{{[0-9]+}}*** [[ARGCADDR]], i{{[0-9]+}}**** [[ARGCADDR_PTR]],
 // CK2: [[ARGCADDR_PTR_REF:%.+]] = load i{{[0-9]+}}***, i{{[0-9]+}}**** [[ARGCADDR_PTR]],
 // CK2: store i{{[0-9]+}}** null, i{{[0-9]+}}*** [[ARGCADDR_PTR_REF]],
 // CK2-NOT: {{.+}} = call i32 @__kmpc_push_num_teams(
diff --git a/clang/test/OpenMP/parallel_firstprivate_codegen.cpp b/clang/test/OpenMP/parallel_firstprivate_codegen.cpp
index 54d5bad54c8..809edf8fc39 100644
--- a/clang/test/OpenMP/parallel_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/parallel_firstprivate_codegen.cpp
@@ -130,15 +130,15 @@ int main() {
     // LAMBDA: getelementptr inbounds [[SS_TY]], [[SS_TY]]* %{{.+}}, i32 0, i32 0
     // LAMBDA: getelementptr inbounds [[SS_TY]], [[SS_TY]]* %{{.+}}, i32 0, i32 1
     // LAMBDA: getelementptr inbounds [[SS_TY]], [[SS_TY]]* %{{.+}}, i32 0, i32 2
-    // LAMBDA: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{[0-9]+}} 5, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, [[SS_TY]]*, i32*, i32*, i32*, [4 x i{{[0-9]+}}]*)* [[SS_MICROTASK:@.+]] to void
+    // LAMBDA: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{[0-9]+}} 5, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, [[SS_TY]]*, i32, i32, i32, [4 x i{{[0-9]+}}]*)* [[SS_MICROTASK:@.+]] to void
     // LAMBDA: ret
 
-    // LAMBDA: define internal void [[SS_MICROTASK]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [[SS_TY]]* %{{.+}}, i32* {{.+}}, i32* {{.+}}, i32* {{.+}}, [4 x i{{[0-9]+}}]* {{.+}})
+    // LAMBDA: define internal void [[SS_MICROTASK]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [[SS_TY]]* %{{.+}}, i32 {{.+}}, i32 {{.+}}, i32 {{.+}}, [4 x i{{[0-9]+}}]* {{.+}})
     // LAMBDA-NOT: getelementptr {{.*}}[[SS_TY]], [[SS_TY]]* %
     // LAMBDA: call{{.*}} void
     // LAMBDA: ret void
 
-    // LAMBDA: define internal void @{{.+}}(i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [[SS_TY]]* %{{.+}}, i32* {{.+}}, i32* {{.+}}, i32* {{.+}})
+    // LAMBDA: define internal void @{{.+}}(i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [[SS_TY]]* %{{.+}}, i32 {{.+}}, i32 {{.+}}, i32 {{.+}})
     // LAMBDA: [[A_PRIV:%.+]] = alloca i{{[0-9]+}},
     // LAMBDA: [[B_PRIV:%.+]] = alloca i{{[0-9]+}},
     // LAMBDA: [[C_PRIV:%.+]] = alloca i{{[0-9]+}},
@@ -157,15 +157,12 @@ int main() {
     // LAMBDA-NEXT: store i{{[0-9]+}} [[DIV]], i{{[0-9]+}}* [[C_PRIV]],
     // LAMBDA-NEXT: ret void
 
-    // LAMBDA: define{{.*}} internal{{.*}} void [[OMP_REGION]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* dereferenceable(4) %{{.+}})
-    // LAMBDA: [[G_PRIVATE_ADDR:%.+]] = alloca i{{[0-9]+}}, align 128
+    // LAMBDA: define{{.*}} internal{{.*}} void [[OMP_REGION]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* dereferenceable(4) %{{.+}}, i32 {{.*}}%{{.+}})
     // LAMBDA: [[SIVAR_PRIVATE_ADDR:%.+]] = alloca i{{[0-9]+}},
+    // LAMBDA: [[G_PRIVATE_ADDR:%.+]] = alloca i{{[0-9]+}}, align 128
     // LAMBDA: [[G_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[G_REF_ADDR:%.+]]
-    // LAMBDA: [[SIVAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[SIVAR_REF_ADDR:%.+]]
     // LAMBDA: [[G_VAL:%.+]] = load volatile i{{[0-9]+}}, i{{[0-9]+}}* [[G_REF]], align 128
     // LAMBDA: store i{{[0-9]+}} [[G_VAL]], i{{[0-9]+}}* [[G_PRIVATE_ADDR]], align 128
-    // LAMBDA: [[SIVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[SIVAR_REF]]
-    // LAMBDA: store i{{[0-9]+}} [[SIVAR_VAL]], i{{[0-9]+}}* [[SIVAR_PRIVATE_ADDR]]
     // LAMBDA-NOT: call {{.*}}void @__kmpc_barrier(
     g = 1;
     sivar = 2;
@@ -201,17 +198,12 @@ int main() {
   // BLOCKS: call {{.*}}void {{.+}} @__kmpc_fork_call({{.+}}, i32 2, {{.+}}* [[OMP_REGION:@.+]] to {{.+}}, i32* [[G]], {{.+}})
 #pragma omp parallel firstprivate(g, sivar)
   {
-    // BLOCKS: define{{.*}} internal{{.*}} void [[OMP_REGION]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* dereferenceable(4) %{{.+}})
-    // BLOCKS: [[G_PRIVATE_ADDR:%.+]] = alloca i{{[0-9]+}}, align 128
+    // BLOCKS: define{{.*}} internal{{.*}} void [[OMP_REGION]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* dereferenceable(4) %{{.+}}, i32 {{.*}}%{{.+}})
     // BLOCKS: [[SIVAR_PRIVATE_ADDR:%.+]] = alloca i{{[0-9]+}},
+    // BLOCKS: [[G_PRIVATE_ADDR:%.+]] = alloca i{{[0-9]+}}, align 128
     // BLOCKS: [[G_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[G_REF_ADDR:%.+]]
-    // BLOCKS: [[SIVAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[SIVAR_REF_ADDR:%.+]]
     // BLOCKS: [[G_VAL:%.+]] = load volatile i{{[0-9]+}}, i{{[0-9]+}}* [[G_REF]], align 128
     // BLOCKS: store i{{[0-9]+}} [[G_VAL]], i{{[0-9]+}}* [[G_PRIVATE_ADDR]], align 128
-    // BLOCK: [[SIVAR_REF_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
-    // BLOCK: [[SIVAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[SIVAR_REF_ADDR]]
-    // BLOCKS: [[SIVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[SIVAR_REF]],
-    // BLOCKS: store i{{[0-9]+}} [[SIVAR_VAL]], i{{[0-9]+}}* [[SIVAR_PRIVATE_ADDR]],
     // BLOCKS-NOT: call {{.*}}void @__kmpc_barrier(
     g = 1;
     sivar = 2;
@@ -248,15 +240,15 @@ int main() {
 // BLOCKS: getelementptr inbounds [[SS_TY]], [[SS_TY]]* %{{.+}}, i32 0, i32 0
 // BLOCKS: getelementptr inbounds [[SS_TY]], [[SS_TY]]* %{{.+}}, i32 0, i32 1
 // BLOCKS: getelementptr inbounds [[SS_TY]], [[SS_TY]]* %{{.+}}, i32 0, i32 2
-// BLOCKS: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{[0-9]+}} 5, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, [[SS_TY]]*, i32*, i32*, i32*, [4 x i{{[0-9]+}}]*)* [[SS_MICROTASK:@.+]] to void
+// BLOCKS: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{[0-9]+}} 5, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, [[SS_TY]]*, i32, i32, i32, [4 x i{{[0-9]+}}]*)* [[SS_MICROTASK:@.+]] to void
 // BLOCKS: ret
 
-// BLOCKS: define internal void [[SS_MICROTASK]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [[SS_TY]]* %{{.+}}, i32* {{.+}}, i32* {{.+}}, i32* {{.+}}, [4 x i{{[0-9]+}}]* {{.+}})
+// BLOCKS: define internal void [[SS_MICROTASK]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [[SS_TY]]* %{{.+}}, i32 {{.+}}, i32 {{.+}}, i32 {{.+}}, [4 x i{{[0-9]+}}]* {{.+}})
 // BLOCKS-NOT: getelementptr {{.*}}[[SS_TY]], [[SS_TY]]* %
 // BLOCKS: call{{.*}} void
 // BLOCKS: ret void
 
-// BLOCKS: define internal void @{{.+}}(i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [[SS_TY]]* %{{.+}}, i32* {{.+}}, i32* {{.+}}, i32* {{.+}})
+// BLOCKS: define internal void @{{.+}}(i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [[SS_TY]]* %{{.+}}, i32 {{.+}}, i32 {{.+}}, i32 {{.+}})
 // BLOCKS: [[A_PRIV:%.+]] = alloca i{{[0-9]+}},
 // BLOCKS: [[B_PRIV:%.+]] = alloca i{{[0-9]+}},
 // BLOCKS: [[C_PRIV:%.+]] = alloca i{{[0-9]+}},
@@ -295,26 +287,24 @@ int main() {
 // CHECK: define {{.*}}i{{[0-9]+}} @main()
 // CHECK: [[TEST:%.+]] = alloca [[S_FLOAT_TY]],
 // CHECK: call {{.*}} [[S_FLOAT_TY_DEF_CONSTR:@.+]]([[S_FLOAT_TY]]* [[TEST]])
-// CHECK: call {{.*}}void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{[0-9]+}} 5, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, [2 x i32]*, i32*, [2 x [[S_FLOAT_TY]]]*, [[S_FLOAT_TY]]*, i{{[0-9]+}}*)* [[MAIN_MICROTASK:@.+]] to void
+// CHECK: call {{.*}}void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{[0-9]+}} 5, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, [2 x i32]*, i32, [2 x [[S_FLOAT_TY]]]*, [[S_FLOAT_TY]]*, i{{[0-9]+}})* [[MAIN_MICROTASK:@.+]] to void
 // CHECK: = call {{.*}}i{{.+}} [[TMAIN_INT:@.+]]()
 // CHECK: call {{.*}} [[S_FLOAT_TY_DESTR:@.+]]([[S_FLOAT_TY]]*
 // CHECK: ret
 //
-// CHECK: define internal {{.*}}void [[MAIN_MICROTASK]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [2 x i32]* dereferenceable(8) %{{.+}}, i32* dereferenceable(4) %{{.+}}, [2 x [[S_FLOAT_TY]]]* dereferenceable(8) %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}}, i32* dereferenceable(4) [[SIVAR:%.+]])
+// CHECK: define internal {{.*}}void [[MAIN_MICROTASK]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [2 x i32]* dereferenceable(8) %{{.+}}, i32 {{.*}}%{{.+}}, [2 x [[S_FLOAT_TY]]]* dereferenceable(8) %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}}, i32 {{.*}}[[SIVAR:%.+]])
 // CHECK: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}},
+// CHECK: [[SIVAR7_PRIV:%.+]] = alloca i{{[0-9]+}},
 // CHECK: [[VEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}],
 // CHECK: [[S_ARR_PRIV:%.+]] = alloca [2 x [[S_FLOAT_TY]]],
 // CHECK: [[VAR_PRIV:%.+]] = alloca [[S_FLOAT_TY]],
-// CHECK: [[SIVAR7_PRIV:%.+]] = alloca i{{[0-9]+}},
 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]],
 
 // CHECK: [[VEC_REF:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** %
-// CHECK: [[T_VAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** %
+// CHECK-NOT: load i{{[0-9]+}}*, i{{[0-9]+}}** %
 // CHECK: [[S_ARR_REF:%.+]] = load [2 x [[S_FLOAT_TY]]]*, [2 x [[S_FLOAT_TY]]]** %
 // CHECK: [[VAR_REF:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** %
-// CHECK: [[SIVAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** %{{.+}},
-// CHECK: [[T_VAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_REF]],
-// CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_PRIV]],
+// CHECK-NOT: load i{{[0-9]+}}*, i{{[0-9]+}}** %
 // CHECK: [[VEC_DEST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8*
 // CHECK: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_REF]] to i8*
 // CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST]], i8* [[VEC_SRC]],
@@ -332,8 +322,6 @@ int main() {
 // CHECK: call {{.*}} [[S_FLOAT_TY_COPY_CONSTR]]([[S_FLOAT_TY]]* [[VAR_PRIV]], [[S_FLOAT_TY]]* {{.*}} [[VAR_REF]], [[ST_TY]]* [[ST_TY_TEMP]])
 // CHECK: call {{.*}} [[ST_TY_DESTR]]([[ST_TY]]* [[ST_TY_TEMP]])
 
-// CHECK: [[SIVAR_REF_ADDR:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[SIVAR_REF]],
-// CHECK: store i{{[0-9]+}} [[SIVAR_REF_ADDR]], i{{[0-9]+}}* [[SIVAR7_PRIV]],
 // CHECK: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[SIVAR7_PRIV]],
 
 // CHECK-DAG: call {{.*}} [[S_FLOAT_TY_DESTR]]([[S_FLOAT_TY]]* [[VAR_PRIV]])
@@ -355,18 +343,18 @@ int main() {
 // CHECK: getelementptr inbounds [[SS_TY]], [[SS_TY]]* %{{.+}}, i32 0, i32 0
 // CHECK: getelementptr inbounds [[SS_TY]], [[SS_TY]]* %{{.+}}, i32 0, i32 1
 // CHECK: getelementptr inbounds [[SS_TY]], [[SS_TY]]* %{{.+}}, i32 0, i32 2
-// CHECK: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{[0-9]+}} 5, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, [[SS_TY]]*, i32*, i32*, i32*, [4 x i32]*)* [[SS_MICROTASK:@.+]] to void
+// CHECK: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{[0-9]+}} 5, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, [[SS_TY]]*, i32, i32, i32, [4 x i32]*)* [[SS_MICROTASK:@.+]] to void
 // CHECK: ret
 
-// CHECK: define internal void [[SS_MICROTASK]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [[SS_TY]]* %{{.+}}, i32* {{.+}}, i32* {{.+}}, i32* {{.+}}, [4 x i{{[0-9]+}}]* {{.+}})
+// CHECK: define internal void [[SS_MICROTASK]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [[SS_TY]]* %{{.+}}, i32 {{.+}}, i32 {{.+}}, i32 {{.+}}, [4 x i{{[0-9]+}}]* {{.+}})
 // CHECK: [[A_PRIV:%.+]] = alloca i{{[0-9]+}},
 // CHECK: [[B_PRIV:%.+]] = alloca i{{[0-9]+}},
 // CHECK: [[C_PRIV:%.+]] = alloca i{{[0-9]+}},
 // CHECK: [[E_PRIV:%.+]] = alloca [4 x i{{[0-9]+}}],
 // CHECK: store i{{[0-9]+}} {{.+}}, i{{[0-9]+}}* [[A_PRIV]]
-// CHECK: store i{{[0-9]+}}* [[A_PRIV]], i{{[0-9]+}}** [[REFA:%.+]],
 // CHECK: store i{{[0-9]+}} {{.+}}, i{{[0-9]+}}* [[B_PRIV]]
 // CHECK: store i{{[0-9]+}} {{.+}}, i{{[0-9]+}}* [[C_PRIV]]
+// CHECK: store i{{[0-9]+}}* [[A_PRIV]], i{{[0-9]+}}** [[REFA:%.+]],
 // CHECK: store i{{[0-9]+}}* [[C_PRIV]], i{{[0-9]+}}** [[REFC:%.+]],
 // CHECK: bitcast [4 x i{{[0-9]+}}]* [[E_PRIV]] to i8*
 // CHECK: bitcast [4 x i{{[0-9]+}}]* %{{.+}} to i8*
@@ -442,13 +430,13 @@ struct St {
 void array_func(float a[3], St s[2], int n, long double vla1[n]) {
   double vla2[n][n] __attribute__((aligned(128)));
 // ARRAY: @__kmpc_fork_call(
-// ARRAY-DAG: [[PRIV_A:%.+]] = alloca float**,
-// ARRAY-DAG: [[PRIV_S:%.+]] = alloca %struct.St**,
-// ARRAY-DAG: [[PRIV_VLA1:%.+]] = alloca x86_fp80**,
+// ARRAY-DAG: [[PRIV_S:%.+]] = alloca %struct.St*,
+// ARRAY-DAG: [[PRIV_VLA1:%.+]] = alloca x86_fp80*,
+// ARRAY-DAG: [[PRIV_A:%.+]] = alloca float*,
 // ARRAY-DAG: [[PRIV_VLA2:%.+]] = alloca double*,
-// ARRAY-DAG: store float** %{{.+}}, float*** [[PRIV_A]],
-// ARRAY-DAG: store %struct.St** %{{.+}}, %struct.St*** [[PRIV_S]],
-// ARRAY-DAG: store x86_fp80** %{{.+}}, x86_fp80*** [[PRIV_VLA1]],
+// ARRAY-DAG: store %struct.St* %{{.+}}, %struct.St** [[PRIV_S]],
+// ARRAY-DAG: store x86_fp80* %{{.+}}, x86_fp80** [[PRIV_VLA1]],
+// ARRAY-DAG: store float* %{{.+}}, float** [[PRIV_A]],
 // ARRAY-DAG: store double* %{{.+}}, double** [[PRIV_VLA2]],
 // ARRAY: call i8* @llvm.stacksave()
 // ARRAY: [[SIZE:%.+]] = mul nuw i64 %{{.+}}, 8
@@ -460,11 +448,11 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
 
 // ARRAY-LABEL: St_func
 // ARRAY: @__kmpc_fork_call(
-// ARRAY-DAG: [[PRIV_S:%.+]] = alloca %struct.St**,
-// ARRAY-DAG: [[PRIV_VLA1:%.+]] = alloca x86_fp80**,
+// ARRAY-DAG: [[PRIV_VLA1:%.+]] = alloca x86_fp80*,
+// ARRAY-DAG: [[PRIV_S:%.+]] = alloca %struct.St*,
 // ARRAY-DAG: [[PRIV_VLA2:%.+]] = alloca double*,
-// ARRAY-DAG: store %struct.St** %{{.+}}, %struct.St*** [[PRIV_S]],
-// ARRAY-DAG: store x86_fp80** %{{.+}}, x86_fp80*** [[PRIV_VLA1]],
+// ARRAY-DAG: store %struct.St* %{{.+}}, %struct.St** [[PRIV_S]],
+// ARRAY-DAG: store x86_fp80* %{{.+}}, x86_fp80** [[PRIV_VLA1]],
 // ARRAY-DAG: store double* %{{.+}}, double** [[PRIV_VLA2]],
 // ARRAY: call i8* @llvm.stacksave()
 // ARRAY: [[SIZE:%.+]] = mul nuw i64 %{{.+}}, 8
diff --git a/clang/test/OpenMP/sections_firstprivate_codegen.cpp b/clang/test/OpenMP/sections_firstprivate_codegen.cpp
index dd854b57b12..2708872a768 100644
--- a/clang/test/OpenMP/sections_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/sections_firstprivate_codegen.cpp
@@ -256,31 +256,29 @@ int main() {
 // CHECK: define {{.*}} i{{[0-9]+}} [[TMAIN_INT]]()
 // CHECK: [[TEST:%.+]] = alloca [[S_INT_TY]],
 // CHECK: call {{.*}} [[S_INT_TY_DEF_CONSTR:@.+]]([[S_INT_TY]]* [[TEST]])
-// CHECK: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{[0-9]+}} 4, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, i32*, [2 x i32]*, [2 x [[S_INT_TY]]]*, [[S_INT_TY]]*)* [[TMAIN_MICROTASK:@.+]] to void
+// CHECK: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{[0-9]+}} 4, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, i32, [2 x i32]*, [2 x [[S_INT_TY]]]*, [[S_INT_TY]]*)* [[TMAIN_MICROTASK:@.+]] to void
 // CHECK: call {{.*}} [[S_INT_TY_DESTR:@.+]]([[S_INT_TY]]*
 // CHECK: ret
 //
 // CHECK: define internal void [[TMAIN_MICROTASK]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}},
 // Skip temp vars for loop
+// CHECK: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}},
 // CHECK: alloca i{{[0-9]+}},
 // CHECK: alloca i{{[0-9]+}},
 // CHECK: alloca i{{[0-9]+}},
 // CHECK: alloca i{{[0-9]+}},
-// CHECK: alloca i{{[0-9]+}},
-// CHECK: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}},
 // CHECK: [[VEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}],
 // CHECK: [[S_ARR_PRIV:%.+]] = alloca [2 x [[S_INT_TY]]],
 // CHECK: [[VAR_PRIV:%.+]] = alloca [[S_INT_TY]],
 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]],
 
-// CHECK: [[T_VAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** %
+// CHECK-NOT: load i{{[0-9]+}}*, i{{[0-9]+}}** %
 // CHECK: [[VEC_REF:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** %
 // CHECK: [[S_ARR:%.+]] = load [2 x [[S_INT_TY]]]*, [2 x [[S_INT_TY]]]** %
 // CHECK: [[VAR_REF:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** %
 
 // firstprivate t_var(t_var)
-// CHECK: [[T_VAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_REF]],
-// CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_PRIV]],
+// CHECK-NOT: load i{{[0-9]+}}, i{{[0-9]+}}*
 
 // firstprivate vec(vec)
 // CHECK: [[VEC_DEST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8*
diff --git a/clang/test/OpenMP/single_codegen.cpp b/clang/test/OpenMP/single_codegen.cpp
index ca1f80245d9..ab57cf727e5 100644
--- a/clang/test/OpenMP/single_codegen.cpp
+++ b/clang/test/OpenMP/single_codegen.cpp
@@ -225,18 +225,20 @@ void array_func(int n, int a[n], St s[2]) {
 #endif
 
 // CHECK-LABEL:@_ZN2SSC2ERi(
-// CHECK: call void ([[IDENT_T_TY]]*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call([[IDENT_T_TY]]* @{{.+}}, i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, [[SS_TY]]*, i32*, i32*, i32*)* [[SS_MICROTASK:@.+]] to void
+// CHECK: call void ([[IDENT_T_TY]]*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call([[IDENT_T_TY]]* @{{.+}}, i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, [[SS_TY]]*, i32, i32, i32)* [[SS_MICROTASK:@.+]] to void
 // CHECK-NEXT: ret void
 
-// CHECK: define internal void [[SS_MICROTASK]](i32* {{[^,]+}}, i32* {{[^,]+}}, [[SS_TY]]* {{.+}}, i32* {{.+}}, i32* {{.+}}, i32* {{.+}})
+// CHECK: define internal void [[SS_MICROTASK]](i32* {{[^,]+}}, i32* {{[^,]+}}, [[SS_TY]]* {{.+}}, i32 {{.+}}, i32 {{.+}}, i32 {{.+}})
 // Private a
 // CHECK: alloca i32,
-// CHECK: alloca i32*,
 // Private b
 // CHECK: alloca i32,
 // Private c
 // CHECK: alloca i32,
 // CHECK: alloca i32*,
+// CHECK: alloca i32*,
+// CHECK: alloca i32*,
+// CHECK: alloca i32*,
 // CHECK: [[DID_IT:%.+]] = alloca i32,
 // CHECK: store i32 0, i32* [[DID_IT]],
 // CHECK: [[RES:%.+]] = call i32 @__kmpc_single([[IDENT_T_TY]]* @{{.+}}, i32 %{{.+}})
@@ -296,25 +298,30 @@ void array_func(int n, int a[n], St s[2]) {
 // CHECK-NEXT: store i32 %
 // CHECK-NEXT: getelementptr inbounds [[CAP_TY]], [[CAP_TY]]* [[CAP]], i32 0, i32 1
 // CHECK-NEXT: load i32*, i32** %
+// CHECK-NEXT: load i32, i32* %
 // CHECK-NEXT: getelementptr inbounds [[CAP_TY]], [[CAP_TY]]* [[CAP]], i32 0, i32 2
 // CHECK-NEXT: load i32*, i32** %
+// CHECK-NEXT: load i32, i32* %
 // CHECK-NEXT: getelementptr inbounds [[CAP_TY]], [[CAP_TY]]* [[CAP]], i32 0, i32 3
 // CHECK-NEXT: load i32*, i32** %
-// CHECK-NEXT: call void ([[IDENT_T_TY]]*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call([[IDENT_T_TY]]* @{{.+}}, i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, [[SS_TY]]*, i32*, i32*, i32*)* [[SS_MICROTASK1:@.+]] to void
+// CHECK-NEXT: load i32, i32* %
+// CHECK-NEXT: call void ([[IDENT_T_TY]]*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call([[IDENT_T_TY]]* @{{.+}}, i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, [[SS_TY]]*, i32, i32, i32)* [[SS_MICROTASK1:@.+]] to void
 // CHECK-NEXT: ret void
 
 // CHECK: define internal void [[COPY_FUNC]](i8*, i8*)
 // CHECK: ret void
 
-// CHECK: define internal void [[SS_MICROTASK1]](i32* {{[^,]+}}, i32* {{[^,]+}}, [[SS_TY]]* {{.+}}, i32* {{.+}}, i32* {{.+}}, i32* {{.+}})
+// CHECK: define internal void [[SS_MICROTASK1]](i32* {{[^,]+}}, i32* {{[^,]+}}, [[SS_TY]]* {{.+}}, i32 {{.+}}, i32 {{.+}}, i32 {{.+}})
 // Private a
 // CHECK: alloca i32,
-// CHECK: alloca i32*,
 // Private b
 // CHECK: alloca i32,
 // Private c
 // CHECK: alloca i32,
 // CHECK: alloca i32*,
+// CHECK: alloca i32*,
+// CHECK: alloca i32*,
+// CHECK: alloca i32*,
 // CHECK: [[DID_IT:%.+]] = alloca i32,
 // CHECK: [[RES:%.+]] = call i32 @__kmpc_single([[IDENT_T_TY]]* @{{.+}}, i32 %{{.+}})
 // CHECK-NEXT: icmp ne i32 [[RES]], 0
@@ -363,10 +370,11 @@ void array_func(int n, int a[n], St s[2]) {
 // CHECK-NEXT: getelementptr inbounds [[SST_TY]], [[SST_TY]]* %{{.+}}, i32 0, i32 0
 // CHECK-NEXT: store double* %
 // CHECK-NEXT: load double*, double** %
-// CHECK-NEXT: call void ([[IDENT_T_TY]]*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call([[IDENT_T_TY]]* @{{.+}}, i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, [[SST_TY]]*, double*)* [[SST_MICROTASK:@.+]] to void
+// CHECK-NEXT: load double, double* %
+// CHECK-NEXT: call void ([[IDENT_T_TY]]*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call([[IDENT_T_TY]]* @{{.+}}, i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, [[SST_TY]]*, double)* [[SST_MICROTASK:@.+]] to void
 // CHECK-NEXT: ret void
 
-// CHECK: define internal void [[SST_MICROTASK]](i32* {{[^,]+}}, i32* {{[^,]+}}, [[SST_TY]]* {{.+}}, double* {{.+}})
+// CHECK: define internal void [[SST_MICROTASK]](i32* {{[^,]+}}, i32* {{[^,]+}}, [[SST_TY]]* {{.+}}, double {{.+}})
 // CHECK: [[RES:%.+]] = call i32 @__kmpc_single([[IDENT_T_TY]]* @{{.+}}, i32 %{{.+}})
 // CHECK-NEXT: icmp ne i32 [[RES]], 0
 // CHECK-NEXT: br i1
diff --git a/clang/test/OpenMP/single_firstprivate_codegen.cpp b/clang/test/OpenMP/single_firstprivate_codegen.cpp
index 8407f77243c..1e91fd8c843 100644
--- a/clang/test/OpenMP/single_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/single_firstprivate_codegen.cpp
@@ -223,18 +223,18 @@ int main() {
 // CHECK: define {{.*}} i{{[0-9]+}} [[TMAIN_INT]]()
 // CHECK: [[TEST:%.+]] = alloca [[S_INT_TY]],
 // CHECK: call {{.*}} [[S_INT_TY_DEF_CONSTR:@.+]]([[S_INT_TY]]* [[TEST]])
-// CHECK: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{[0-9]+}} 4, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, i32*, [2 x i32]*, [2 x [[S_INT_TY]]]*, [[S_INT_TY]]*)* [[TMAIN_MICROTASK:@.+]] to void
+// CHECK: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{[0-9]+}} 4, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, i32, [2 x i32]*, [2 x [[S_INT_TY]]]*, [[S_INT_TY]]*)* [[TMAIN_MICROTASK:@.+]] to void
 // CHECK: call {{.*}} [[S_INT_TY_DESTR:@.+]]([[S_INT_TY]]*
 // CHECK: ret
 //
-// CHECK: define internal void [[TMAIN_MICROTASK]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, i32* dereferenceable(4) %{{.+}}, [2 x i32]* dereferenceable(8) %{{.+}}, [2 x [[S_INT_TY]]]* dereferenceable(8) %{{.+}}, [[S_INT_TY]]* dereferenceable(4) %{{.+}})
+// CHECK: define internal void [[TMAIN_MICROTASK]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, i32 {{.*}}%{{.+}}, [2 x i32]* dereferenceable(8) %{{.+}}, [2 x [[S_INT_TY]]]* dereferenceable(8) %{{.+}}, [[S_INT_TY]]* dereferenceable(4) %{{.+}})
 // CHECK: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}},
 // CHECK: [[VEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}],
 // CHECK: [[S_ARR_PRIV:%.+]] = alloca [2 x [[S_INT_TY]]],
 // CHECK: [[VAR_PRIV:%.+]] = alloca [[S_INT_TY]],
 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]],
 
-// CHECK: [[T_VAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** %
+// CHECK-NOT: load i{{[0-9]+}}*, i{{[0-9]+}}** %
 // CHECK: [[VEC_REF:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** %
 // CHECK: [[S_ARR:%.+]] = load [2 x [[S_INT_TY]]]*, [2 x [[S_INT_TY]]]** %
 // CHECK: [[VAR_REF:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** %
@@ -244,8 +244,7 @@ int main() {
 // CHECK: call i32 @__kmpc_single(
 
 // firstprivate t_var(t_var)
-// CHECK: [[T_VAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_REF]],
-// CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_PRIV]],
+// CHECK-NOT: load i{{[0-9]+}}, i{{[0-9]+}}*
 
 // firstprivate vec(vec)
 // CHECK: [[VEC_DEST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8*
diff --git a/clang/test/OpenMP/target_firstprivate_codegen.cpp b/clang/test/OpenMP/target_firstprivate_codegen.cpp
index a07522383d0..a3e2b9a3fe9 100644
--- a/clang/test/OpenMP/target_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/target_firstprivate_codegen.cpp
@@ -106,12 +106,9 @@ int foo(int n, double *ptr) {
   
   // TCHECK:  define void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A_IN:%.+]])
   // TCHECK:  [[A_ADDR:%.+]] = alloca i{{[0-9]+}},
-  // TCHECK:  [[A1:%.+]] = alloca i{{[0-9]+}},
+  // TCHECK-NOT: alloca i{{[0-9]+}},
   // TCHECK:  store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[A_ADDR]],
-  // TCHECK-64:  [[CONV:%.+]] = bitcast i{{[0-9]+}}* [[A_ADDR]] to i{{[0-9]+}}*
-  // TCHECK-64:  [[A_ADDR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[CONV]],
-  // TCHECK-32:  [[A_ADDR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[A_ADDR]],
-  // TCHECK: store i{{[0-9]+}} [[A_ADDR_VAL]], i{{[0-9]+}}* [[A1]],
+  // TCHECK-NOT: store i{{[0-9]+}} %
   // TCHECK:  ret void  
 
 #pragma omp target firstprivate(aa,b,bn,c,cn,d)
@@ -236,7 +233,7 @@ int foo(int n, double *ptr) {
   // TCHECK:  [[VLA_ADDR4:%.+]] = alloca i{{[0-9]+}},
   // TCHECK:  [[CN_ADDR:%.+]] = alloca double*,
   // TCHECK:  [[D_ADDR:%.+]] = alloca [[TT]]*,
-  // TCHECK:  [[A2_PRIV:%.+]] = alloca i{{[0-9]+}},
+  // TCHECK-NOT: alloca i{{[0-9]+}},
   // TCHECK:  [[B_PRIV:%.+]] = alloca [10 x float],
   // TCHECK:  [[SSTACK:%.+]] = alloca i8*,
   // TCHECK:  [[C_PRIV:%.+]] = alloca [5 x [10 x double]],
@@ -261,8 +258,7 @@ int foo(int n, double *ptr) {
   // TCHECK:  [[D_ADDR_REF:%.+]] = load [[TT]]*, [[TT]]** [[D_ADDR]],
 
   // firstprivate(aa): a_priv = a_in
-  // TCHECK:  [[A2_CONV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[CONV_A2ADDR]],
-  // TCHECK:  store i{{[0-9]+}} [[A2_CONV_VAL]], i{{[0-9]+}}* [[A2_PRIV]],
+  // TCHECK-NOT:  store i{{[0-9]+}} %
 
   //  firstprivate(b): memcpy(b_priv,b_in)
   // TCHECK:  [[B_PRIV_BCAST:%.+]] = bitcast [10 x float]* [[B_PRIV]] to i8*
@@ -318,10 +314,9 @@ int foo(int n, double *ptr) {
 
   // TCHECK:  define void @__omp_offloading_{{.+}}(double* [[PTR_IN:%.+]])
   // TCHECK:  [[PTR_ADDR:%.+]] = alloca double*,
-  // TCHECK:  [[PTR_PRIV:%.+]] = alloca double*,
+  // TCHECK-NOT: alloca double*,
   // TCHECK:  store double* [[PTR_IN]], double** [[PTR_ADDR]],
-  // TCHECK:  [[PTR_IN_REF:%.+]] = load double*, double** [[PTR_ADDR]],
-  // TCHECK:  store double* [[PTR_IN_REF]], double** [[PTR_PRIV]],
+  // TCHECK-NOT: store double* %
 
   return a;
 }
@@ -361,8 +356,7 @@ int fstatic(int n) {
 // TCHECK:  [[A_ADDR:%.+]] = alloca i{{[0-9]+}},
 // TCHECK:  [[A3_ADDR:%.+]] = alloca i{{[0-9]+}},
 // TCHECK:  [[B_ADDR:%.+]] = alloca [10 x i{{[0-9]+}}]*,
-// TCHECK:  [[A_PRIV:%.+]] = alloca i{{[0-9]+}},
-// TCHECK:  [[A3_PRIV:%.+]] = alloca i{{[0-9]+}},
+// TCHECK-NOT: alloca i{{[0-9]+}},
 // TCHECK:  [[B_PRIV:%.+]] = alloca [10 x i{{[0-9]+}}],
 // TCHECK:  store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[A_ADDR]],
 // TCHECK:  store i{{[0-9]+}} [[A3_IN]], i{{[0-9]+}}* [[A3_ADDR]],
@@ -372,13 +366,9 @@ int fstatic(int n) {
 // TCHECK:  [[B_ADDR_REF:%.+]] = load [10 x i{{[0-9]+}}]*, [10 x i{{[0-9]+}}]** [[B_ADDR]],
 
 // firstprivate(a): a_priv = a_in
-// TCHECK-64:  [[A_IN_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[A_CONV]],
-// TCHECK-32:  [[A_IN_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[A_ADDR]],
-// TCHECK:  store i{{[0-9]+}} [[A_IN_VAL]], i{{[0-9]+}}* [[A_PRIV]],
 
 // firstprivate(aaa)
-// TCHECK:  [[A3_IN_VAL:%.+]] = load i8, i8* [[A3_CONV]],
-// TCHECK:  store i{{[0-9]+}} [[A3_IN_VAL]], i{{[0-9]+}}* [[A3_PRIV]],
+// TCHECK-NOT:  store i{{[0-9]+}} %
 
 // firstprivate(b)
 // TCHECK:  [[B_PRIV_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_PRIV]] to i8*
@@ -462,7 +452,7 @@ struct S1 {
   // TCHECK:  [[VLA_ADDR:%.+]] = alloca i{{[0-9]+}},
   // TCHECK:  [[VLA_ADDR2:%.+]] = alloca i{{[0-9]+}},
   // TCHECK:  [[C_ADDR:%.+]] = alloca i{{[0-9]+}}*,
-  // TCHECK:  [[B_PRIV:%.+]] = alloca i{{[0-9]+}},
+  // TCHECK-NOT: alloca i{{[0-9]+}},
   // TCHECK:  [[SSTACK:%.+]] = alloca i8*,
 
   // TCHECK:  store [[S1]]* [[TH]], [[S1]]** [[TH_ADDR]],
@@ -477,9 +467,7 @@ struct S1 {
   // TCHECK:  [[C_ADDR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[C_ADDR]],
 
   // firstprivate(b)
-  // TCHECK-64:  [[B_IN_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[B_ADDR_CONV]],
-  // TCHECK-32:  [[B_IN_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[B_ADDR]],
-  // TCHECK:  store i{{[0-9]+}} [[B_IN_VAL]], i{{[0-9]+}}* [[B_PRIV]],
+  // TCHECK-NOT:  store i{{[0-9]+}} %
  
   // TCHECK:  [[RET_STACK:%.+]] = call i8* @llvm.stacksave()
   // TCHECK:  store i8* [[RET_STACK:%.+]], i8** [[SSTACK]],
@@ -573,7 +561,7 @@ int bar(int n, double *ptr){
 // TCHECK: define void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A_IN:%.+]], [10 x i{{[0-9]+}}]*{{.+}} [[B_IN:%.+]])
 // TCHECK:  [[A_ADDR:%.+]] = alloca i{{[0-9]+}},
 // TCHECK:  [[B_ADDR:%.+]] = alloca [10 x i{{[0-9]+}}]*,
-// TCHECK:  [[A_PRIV:%.+]] = alloca i{{[0-9]+}},
+// TCHECK-NOT: alloca i{{[0-9]+}},
 // TCHECK:  [[B_PRIV:%.+]] = alloca [10 x i{{[0-9]+}}],
 // TCHECK:  store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[A_ADDR]],
 // TCHECK:  store [10 x i{{[0-9]+}}]* [[B_IN]], [10 x i{{[0-9]+}}]** [[B_ADDR]],
@@ -581,9 +569,7 @@ int bar(int n, double *ptr){
 // TCHECK:  [[B_ADDR_REF:%.+]] = load [10 x i{{[0-9]+}}]*, [10 x i{{[0-9]+}}]** [[B_ADDR]],
 
 // firstprivate(a)
-// TCHECK-64:  [[A_IN_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[A_ADDR_CONV]]
-// TCHECK-32:  [[A_IN_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[A_ADDR]]
-// TCHECK:  store i{{[0-9]+}} [[A_IN_VAL]], i{{[0-9]+}}* [[A_PRIV]],
+// TCHECK-NOT:  store i{{[0-9]+}} %
 
 // firstprivate(b)
 // TCHECK:  [[B_PRIV_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_PRIV]] to i8*
diff --git a/clang/test/OpenMP/task_firstprivate_codegen.cpp b/clang/test/OpenMP/task_firstprivate_codegen.cpp
index c6da687b671..4419263c320 100644
--- a/clang/test/OpenMP/task_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/task_firstprivate_codegen.cpp
@@ -27,7 +27,7 @@ volatile double g;
 // CHECK-DAG: [[KMP_TASK_T_TY:%.+]] = type { i8*, i32 (i32, i8*)*, i32, i32 (i32, i8*)* }
 // CHECK-DAG: [[S_DOUBLE_TY:%.+]] = type { double }
 // CHECK-DAG: [[PRIVATES_MAIN_TY:%.+]] = type {{.?}}{ [2 x [[S_DOUBLE_TY]]], [[S_DOUBLE_TY]], i32, [2 x i32]
-// CHECK-DAG: [[CAP_MAIN_TY:%.+]] = type { [2 x i32]*, i32*, [2 x [[S_DOUBLE_TY]]]*, [[S_DOUBLE_TY]]*, i{{[0-9]+}}* }
+// CHECK-DAG: [[CAP_MAIN_TY:%.+]] = type {{.*}}{ [2 x i32]*, i32, {{.*}}[2 x [[S_DOUBLE_TY]]]*, [[S_DOUBLE_TY]]*, i{{[0-9]+}}
 // CHECK-DAG: [[KMP_TASK_MAIN_TY:%.+]] = type { [[KMP_TASK_T_TY]], [[PRIVATES_MAIN_TY]] }
 // CHECK-DAG: [[S_INT_TY:%.+]] = type { i32 }
 // CHECK-DAG: [[CAP_TMAIN_TY:%.+]] = type { [2 x i32]*, i32*, [2 x [[S_INT_TY]]]*, [[S_INT_TY]]* }
@@ -62,14 +62,12 @@ int main() {
 // LAMBDA: [[PRIVATES:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
 // LAMBDA: [[G_PRIVATE_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[PRIVATES]], i{{.+}} 0, i{{.+}} 0
 // LAMBDA: [[G_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 0
-// LAMBDA: [[G_REF:%.+]] = load double*, double** [[G_ADDR_REF]]
-// LAMBDA: [[G_VAL:%.+]] = load volatile double, double* [[G_REF]]
+// LAMBDA: [[G_VAL:%.+]] = load volatile double, double* [[G_ADDR_REF]]
 // LAMBDA: store volatile double [[G_VAL]], double* [[G_PRIVATE_ADDR]]
 
 // LAMBDA: [[SIVAR_PRIVATE_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[PRIVATES]], i{{.+}} 0, i{{.+}} 1
 // LAMBDA: [[SIVAR_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
-// LAMBDA: [[SIVAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[SIVAR_ADDR_REF]]
-// LAMBDA: [[SIVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[SIVAR_REF]]
+// LAMBDA: [[SIVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[SIVAR_ADDR_REF]]
 // LAMBDA: store i{{[0-9]+}} [[SIVAR_VAL]], i{{[0-9]+}}* [[SIVAR_PRIVATE_ADDR]]
 
 // LAMBDA: call i32 @__kmpc_omp_task(%{{.+}}* @{{.+}}, i32 %{{.+}}, i8* [[RES]])
@@ -108,14 +106,12 @@ int main() {
   // BLOCKS: [[PRIVATES:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
   // BLOCKS: [[G_PRIVATE_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[PRIVATES]], i{{.+}} 0, i{{.+}} 0
   // BLOCKS: [[G_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 0
-  // BLOCKS: [[G_REF:%.+]] = load double*, double** [[G_ADDR_REF]]
-  // BLOCKS: [[G_VAL:%.+]] = load volatile double, double* [[G_REF]]
+  // BLOCKS: [[G_VAL:%.+]] = load volatile double, double* [[G_ADDR_REF]]
   // BLOCKS: store volatile double [[G_VAL]], double* [[G_PRIVATE_ADDR]]
 
   // BLOCKS: [[SIVAR_PRIVATE_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[PRIVATES]], i{{.+}} 0, i{{.+}} 1
   // BLOCKS: [[SIVAR_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
-  // BLOCKS: [[SIVAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[SIVAR_ADDR_REF]]
-  // BLOCKS: [[SIVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[SIVAR_REF]]
+  // BLOCKS: [[SIVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[SIVAR_ADDR_REF]]
   // BLOCKS: store i{{[0-9]+}} [[SIVAR_VAL]], i{{[0-9]+}}* [[SIVAR_PRIVATE_ADDR]]
   // BLOCKS: call i32 @__kmpc_omp_task(%{{.+}}* @{{.+}}, i32 %{{.+}}, i8* [[RES]])
   // BLOCKS: ret
@@ -180,13 +176,15 @@ int main() {
 // CHECK: [[VEC_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 0
 // CHECK: store [2 x i32]* [[VEC_ADDR]], [2 x i32]** [[VEC_REF]],
 // CHECK: [[T_VAR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 1
-// CHECK: store i32* [[T_VAR_ADDR]], i32** [[T_VAR_REF]],
-// CHECK: [[S_ARR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 2
+// CHECK: [[T_VAR:%.+]] = load i32, i32* [[T_VAR_ADDR]],
+// CHECK: store i32 [[T_VAR]], i32* [[T_VAR_REF]],
+// CHECK: [[S_ARR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 3
 // CHECK: store [2 x [[S_DOUBLE_TY]]]* [[S_ARR_ADDR]], [2 x [[S_DOUBLE_TY]]]** [[S_ARR_REF]],
-// CHECK: [[VAR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 3
+// CHECK: [[VAR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 4
 // CHECK: store [[S_DOUBLE_TY]]* [[VAR_ADDR]], [[S_DOUBLE_TY]]** [[VAR_REF]],
-// CHECK: [[SIVAR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 4
-// CHECK: store i{{[0-9]+}}* [[SIVAR]], i{{[0-9]+}}** [[SIVAR_REF]],
+// CHECK: [[SIVAR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 5
+// CHECK: [[SIVAR_VAL:%.+]] = load i32, i32* [[SIVAR]],
+// CHECK: store i{{[0-9]+}} [[SIVAR_VAL]], i{{[0-9]+}}* [[SIVAR_REF]],
 
 // Allocate task.
 // Returns struct kmp_task_t {
@@ -211,7 +209,7 @@ int main() {
 // Constructors for s_arr and var.
 // s_arr;
 // CHECK: [[PRIVATE_S_ARR_REF:%.+]] = getelementptr inbounds [[PRIVATES_MAIN_TY]], [[PRIVATES_MAIN_TY]]* [[PRIVATES]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
-// CHECK: [[S_ARR_ADDR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* [[SHAREDS]], i{{.+}} 0, i{{.+}} 2
+// CHECK: [[S_ARR_ADDR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* [[SHAREDS]], i{{.+}} 0, i{{.+}} 3
 // CHECK: load [2 x [[S_DOUBLE_TY]]]*, [2 x [[S_DOUBLE_TY]]]** [[S_ARR_ADDR_REF]],
 // CHECK: call void [[S_DOUBLE_TY_COPY_CONSTR]]([[S_DOUBLE_TY]]* [[S_ARR_CUR:%[^,]+]],
 // CHECK: getelementptr [[S_DOUBLE_TY]], [[S_DOUBLE_TY]]* [[S_ARR_CUR]], i{{.+}} 1
@@ -221,14 +219,13 @@ int main() {
 
 // var;
 // CHECK: [[PRIVATE_VAR_REF:%.+]] = getelementptr inbounds [[PRIVATES_MAIN_TY]], [[PRIVATES_MAIN_TY]]* [[PRIVATES]], i{{.+}} 0, i{{.+}} 1
-// CHECK: [[VAR_ADDR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* [[SHAREDS]], i{{.+}} 0, i{{.+}} 3
+// CHECK: [[VAR_ADDR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* [[SHAREDS]], i{{.+}} 0, i{{.+}} 4
 // CHECK: [[VAR_REF:%.+]] = load [[S_DOUBLE_TY]]*, [[S_DOUBLE_TY]]** [[VAR_ADDR_REF]],
 // CHECK: call void [[S_DOUBLE_TY_COPY_CONSTR]]([[S_DOUBLE_TY]]* [[PRIVATE_VAR_REF]], [[S_DOUBLE_TY]]* {{.*}}[[VAR_REF]],
 
 // t_var;
 // CHECK: [[PRIVATE_T_VAR_REF:%.+]] = getelementptr inbounds [[PRIVATES_MAIN_TY]], [[PRIVATES_MAIN_TY]]* [[PRIVATES]], i{{.+}} 0, i{{.+}} 2
-// CHECK: [[T_VAR_ADDR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* [[SHAREDS]], i{{.+}} 0, i{{.+}} 1
-// CHECK: [[T_VAR_REF:%.+]] = load i{{.+}}*, i{{.+}}** [[T_VAR_ADDR_REF]],
+// CHECK: [[T_VAR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* [[SHAREDS]], i{{.+}} 0, i{{.+}} 1
 // CHECK: [[T_VAR:%.+]] = load i{{.+}}, i{{.+}}* [[T_VAR_REF]],
 // CHECK: store i32 [[T_VAR]], i32* [[PRIVATE_T_VAR_REF]],
 
@@ -239,8 +236,7 @@ int main() {
 
 // sivar;
 // CHECK: [[PRIVATE_SIVAR_REF:%.+]] = getelementptr inbounds [[PRIVATES_MAIN_TY]], [[PRIVATES_MAIN_TY]]* [[PRIVATES]], i{{.+}} 0, i{{.+}} 4
-// CHECK: [[SIVAR_ADDR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* [[SHAREDS]], i{{.+}} 0, i{{.+}} 4
-// CHECK: [[SIVAR_REF:%.+]] = load i{{.+}}*, i{{.+}}** [[SIVAR_ADDR_REF]],
+// CHECK: [[SIVAR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* [[SHAREDS]], i{{.+}} 0, i{{.+}} 5
 // CHECK: [[SIVAR:%.+]] = load i{{.+}}, i{{.+}}* [[SIVAR_REF]],
 // CHECK: store i32 [[SIVAR]], i32* [[PRIVATE_SIVAR_REF]],
 
diff --git a/clang/test/OpenMP/taskloop_firstprivate_codegen.cpp b/clang/test/OpenMP/taskloop_firstprivate_codegen.cpp
index 6b7923e9b64..8af56fae3b8 100644
--- a/clang/test/OpenMP/taskloop_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/taskloop_firstprivate_codegen.cpp
@@ -27,7 +27,7 @@ volatile double g;
 // CHECK-DAG: [[KMP_TASK_T_TY:%.+]] = type { i8*, i32 (i32, i8*)*, i32, i32 (i32, i8*)*, i64, i64, i64, i32 }
 // CHECK-DAG: [[S_DOUBLE_TY:%.+]] = type { double }
 // CHECK-DAG: [[PRIVATES_MAIN_TY:%.+]] = type {{.?}}{ [2 x [[S_DOUBLE_TY]]], [[S_DOUBLE_TY]], i32, [2 x i32]
-// CHECK-DAG: [[CAP_MAIN_TY:%.+]] = type { [2 x i32]*, i32*, [2 x [[S_DOUBLE_TY]]]*, [[S_DOUBLE_TY]]*, i{{[0-9]+}}* }
+// CHECK-DAG: [[CAP_MAIN_TY:%.+]] = type {{.*}}{ [2 x i32]*, i32, {{.*}}[2 x [[S_DOUBLE_TY]]]*, [[S_DOUBLE_TY]]*, i{{[0-9]+}}
 // CHECK-DAG: [[KMP_TASK_MAIN_TY:%.+]] = type { [[KMP_TASK_T_TY]], [[PRIVATES_MAIN_TY]] }
 // CHECK-DAG: [[S_INT_TY:%.+]] = type { i32 }
 // CHECK-DAG: [[CAP_TMAIN_TY:%.+]] = type { [2 x i32]*, i32*, [2 x [[S_INT_TY]]]*, [[S_INT_TY]]* }
@@ -62,14 +62,12 @@ int main() {
 // LAMBDA: [[PRIVATES:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
 // LAMBDA: [[G_PRIVATE_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[PRIVATES]], i{{.+}} 0, i{{.+}} 0
 // LAMBDA: [[G_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 0
-// LAMBDA: [[G_REF:%.+]] = load double*, double** [[G_ADDR_REF]]
-// LAMBDA: [[G_VAL:%.+]] = load volatile double, double* [[G_REF]]
+// LAMBDA: [[G_VAL:%.+]] = load volatile double, double* [[G_ADDR_REF]]
 // LAMBDA: store volatile double [[G_VAL]], double* [[G_PRIVATE_ADDR]]
 
 // LAMBDA: [[SIVAR_PRIVATE_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[PRIVATES]], i{{.+}} 0, i{{.+}} 1
 // LAMBDA: [[SIVAR_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
-// LAMBDA: [[SIVAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[SIVAR_ADDR_REF]]
-// LAMBDA: [[SIVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[SIVAR_REF]]
+// LAMBDA: [[SIVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[SIVAR_ADDR_REF]]
 // LAMBDA: store i{{[0-9]+}} [[SIVAR_VAL]], i{{[0-9]+}}* [[SIVAR_PRIVATE_ADDR]]
 
 // LAMBDA: call void @__kmpc_taskloop(%{{.+}}* @{{.+}}, i32 %{{.+}}, i8* [[RES]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* null)
@@ -108,14 +106,12 @@ int main() {
   // BLOCKS: [[PRIVATES:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
   // BLOCKS: [[G_PRIVATE_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[PRIVATES]], i{{.+}} 0, i{{.+}} 0
   // BLOCKS: [[G_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 0
-  // BLOCKS: [[G_REF:%.+]] = load double*, double** [[G_ADDR_REF]]
-  // BLOCKS: [[G_VAL:%.+]] = load volatile double, double* [[G_REF]]
+  // BLOCKS: [[G_VAL:%.+]] = load volatile double, double* [[G_ADDR_REF]]
   // BLOCKS: store volatile double [[G_VAL]], double* [[G_PRIVATE_ADDR]]
 
   // BLOCKS: [[SIVAR_PRIVATE_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[PRIVATES]], i{{.+}} 0, i{{.+}} 1
   // BLOCKS: [[SIVAR_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
-  // BLOCKS: [[SIVAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[SIVAR_ADDR_REF]]
-  // BLOCKS: [[SIVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[SIVAR_REF]]
+  // BLOCKS: [[SIVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[SIVAR_ADDR_REF]]
   // BLOCKS: store i{{[0-9]+}} [[SIVAR_VAL]], i{{[0-9]+}}* [[SIVAR_PRIVATE_ADDR]]
   // BLOCKS: call void @__kmpc_taskloop(%{{.+}}* @{{.+}}, i32 %{{.+}}, i8* [[RES]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* null)
   // BLOCKS: ret
@@ -180,13 +176,15 @@ int main() {
 // CHECK: [[VEC_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 0
 // CHECK: store [2 x i32]* [[VEC_ADDR]], [2 x i32]** [[VEC_REF]],
 // CHECK: [[T_VAR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 1
-// CHECK: store i32* [[T_VAR_ADDR]], i32** [[T_VAR_REF]],
-// CHECK: [[S_ARR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 2
+// CHECK: [[T_VAR_VAL:%.+]] = load i32, i32* [[T_VAR_ADDR]],
+// CHECK: store i32 [[T_VAR_VAL]], i32* [[T_VAR_REF]],
+// CHECK: [[S_ARR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 3
 // CHECK: store [2 x [[S_DOUBLE_TY]]]* [[S_ARR_ADDR]], [2 x [[S_DOUBLE_TY]]]** [[S_ARR_REF]],
-// CHECK: [[VAR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 3
+// CHECK: [[VAR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 4
 // CHECK: store [[S_DOUBLE_TY]]* [[VAR_ADDR]], [[S_DOUBLE_TY]]** [[VAR_REF]],
-// CHECK: [[SIVAR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 4
-// CHECK: store i{{[0-9]+}}* [[SIVAR]], i{{[0-9]+}}** [[SIVAR_REF]],
+// CHECK: [[SIVAR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 5
+// CHECK: [[SIVAR_VAL:%.+]] = load i32, i32* [[SIVAR]],
+// CHECK: store i{{[0-9]+}} [[SIVAR_VAL]], i{{[0-9]+}}* [[SIVAR_REF]],
 
 // Allocate task.
 // Returns struct kmp_task_t {
@@ -211,7 +209,7 @@ int main() {
 // Constructors for s_arr and var.
 // s_arr;
 // CHECK: [[PRIVATE_S_ARR_REF:%.+]] = getelementptr inbounds [[PRIVATES_MAIN_TY]], [[PRIVATES_MAIN_TY]]* [[PRIVATES]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
-// CHECK: [[S_ARR_ADDR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* [[SHAREDS]], i{{.+}} 0, i{{.+}} 2
+// CHECK: [[S_ARR_ADDR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* [[SHAREDS]], i{{.+}} 0, i{{.+}} 3
 // CHECK: load [2 x [[S_DOUBLE_TY]]]*, [2 x [[S_DOUBLE_TY]]]** [[S_ARR_ADDR_REF]],
 // CHECK: call void [[S_DOUBLE_TY_COPY_CONSTR]]([[S_DOUBLE_TY]]* [[S_ARR_CUR:%[^,]+]],
 // CHECK: getelementptr [[S_DOUBLE_TY]], [[S_DOUBLE_TY]]* [[S_ARR_CUR]], i{{.+}} 1
@@ -221,15 +219,14 @@ int main() {
 
 // var;
 // CHECK: [[PRIVATE_VAR_REF:%.+]] = getelementptr inbounds [[PRIVATES_MAIN_TY]], [[PRIVATES_MAIN_TY]]* [[PRIVATES]], i{{.+}} 0, i{{.+}} 1
-// CHECK: [[VAR_ADDR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* [[SHAREDS]], i{{.+}} 0, i{{.+}} 3
+// CHECK: [[VAR_ADDR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* [[SHAREDS]], i{{.+}} 0, i{{.+}} 4
 // CHECK: [[VAR_REF:%.+]] = load [[S_DOUBLE_TY]]*, [[S_DOUBLE_TY]]** [[VAR_ADDR_REF]],
 // CHECK: call void [[S_DOUBLE_TY_COPY_CONSTR]]([[S_DOUBLE_TY]]* [[PRIVATE_VAR_REF]], [[S_DOUBLE_TY]]* {{.*}}[[VAR_REF]],
 
 // t_var;
 // CHECK: [[PRIVATE_T_VAR_REF:%.+]] = getelementptr inbounds [[PRIVATES_MAIN_TY]], [[PRIVATES_MAIN_TY]]* [[PRIVATES]], i{{.+}} 0, i{{.+}} 2
 // CHECK: [[T_VAR_ADDR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* [[SHAREDS]], i{{.+}} 0, i{{.+}} 1
-// CHECK: [[T_VAR_REF:%.+]] = load i{{.+}}*, i{{.+}}** [[T_VAR_ADDR_REF]],
-// CHECK: [[T_VAR:%.+]] = load i{{.+}}, i{{.+}}* [[T_VAR_REF]],
+// CHECK: [[T_VAR:%.+]] = load i{{.+}}, i{{.+}}* [[T_VAR_ADDR_REF]],
 // CHECK: store i32 [[T_VAR]], i32* [[PRIVATE_T_VAR_REF]],
 
 // vec;
@@ -239,9 +236,8 @@ int main() {
 
 // sivar;
 // CHECK: [[PRIVATE_SIVAR_REF:%.+]] = getelementptr inbounds [[PRIVATES_MAIN_TY]], [[PRIVATES_MAIN_TY]]* [[PRIVATES]], i{{.+}} 0, i{{.+}} 4
-// CHECK: [[SIVAR_ADDR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* [[SHAREDS]], i{{.+}} 0, i{{.+}} 4
-// CHECK: [[SIVAR_REF:%.+]] = load i{{.+}}*, i{{.+}}** [[SIVAR_ADDR_REF]],
-// CHECK: [[SIVAR:%.+]] = load i{{.+}}, i{{.+}}* [[SIVAR_REF]],
+// CHECK: [[SIVAR_ADDR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* [[SHAREDS]], i{{.+}} 0, i{{.+}} 5
+// CHECK: [[SIVAR:%.+]] = load i{{.+}}, i{{.+}}* [[SIVAR_ADDR_REF]],
 // CHECK: store i32 [[SIVAR]], i32* [[PRIVATE_SIVAR_REF]],
 
 // Provide pointer to destructor function, which will destroy private variables at the end of the task.
diff --git a/clang/test/OpenMP/taskloop_simd_firstprivate_codegen.cpp b/clang/test/OpenMP/taskloop_simd_firstprivate_codegen.cpp
index 2b56d3be0b0..5cb2cd81503 100644
--- a/clang/test/OpenMP/taskloop_simd_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/taskloop_simd_firstprivate_codegen.cpp
@@ -27,7 +27,7 @@ volatile double g;
 // CHECK-DAG: [[KMP_TASK_T_TY:%.+]] = type { i8*, i32 (i32, i8*)*, i32, i32 (i32, i8*)*, i64, i64, i64, i32 }
 // CHECK-DAG: [[S_DOUBLE_TY:%.+]] = type { double }
 // CHECK-DAG: [[PRIVATES_MAIN_TY:%.+]] = type {{.?}}{ [2 x [[S_DOUBLE_TY]]], [[S_DOUBLE_TY]], i32, [2 x i32]
-// CHECK-DAG: [[CAP_MAIN_TY:%.+]] = type { [2 x i32]*, i32*, [2 x [[S_DOUBLE_TY]]]*, [[S_DOUBLE_TY]]*, i{{[0-9]+}}* }
+// CHECK-DAG: [[CAP_MAIN_TY:%.+]] = type {{.*}}{ [2 x i32]*, i32, {{.*}}[2 x [[S_DOUBLE_TY]]]*, [[S_DOUBLE_TY]]*, i{{[0-9]+}}
 // CHECK-DAG: [[KMP_TASK_MAIN_TY:%.+]] = type { [[KMP_TASK_T_TY]], [[PRIVATES_MAIN_TY]] }
 // CHECK-DAG: [[S_INT_TY:%.+]] = type { i32 }
 // CHECK-DAG: [[CAP_TMAIN_TY:%.+]] = type { [2 x i32]*, i32*, [2 x [[S_INT_TY]]]*, [[S_INT_TY]]* }
@@ -62,14 +62,12 @@ int main() {
 // LAMBDA: [[PRIVATES:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
 // LAMBDA: [[G_PRIVATE_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[PRIVATES]], i{{.+}} 0, i{{.+}} 0
 // LAMBDA: [[G_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 0
-// LAMBDA: [[G_REF:%.+]] = load double*, double** [[G_ADDR_REF]]
-// LAMBDA: [[G_VAL:%.+]] = load volatile double, double* [[G_REF]]
+// LAMBDA: [[G_VAL:%.+]] = load volatile double, double* [[G_ADDR_REF]]
 // LAMBDA: store volatile double [[G_VAL]], double* [[G_PRIVATE_ADDR]]
 
 // LAMBDA: [[SIVAR_PRIVATE_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[PRIVATES]], i{{.+}} 0, i{{.+}} 1
 // LAMBDA: [[SIVAR_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
-// LAMBDA: [[SIVAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[SIVAR_ADDR_REF]]
-// LAMBDA: [[SIVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[SIVAR_REF]]
+// LAMBDA: [[SIVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[SIVAR_ADDR_REF]]
 // LAMBDA: store i{{[0-9]+}} [[SIVAR_VAL]], i{{[0-9]+}}* [[SIVAR_PRIVATE_ADDR]]
 
 // LAMBDA: call void @__kmpc_taskloop(%{{.+}}* @{{.+}}, i32 %{{.+}}, i8* [[RES]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* null)
@@ -108,14 +106,12 @@ int main() {
   // BLOCKS: [[PRIVATES:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
   // BLOCKS: [[G_PRIVATE_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[PRIVATES]], i{{.+}} 0, i{{.+}} 0
   // BLOCKS: [[G_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 0
-  // BLOCKS: [[G_REF:%.+]] = load double*, double** [[G_ADDR_REF]]
-  // BLOCKS: [[G_VAL:%.+]] = load volatile double, double* [[G_REF]]
+  // BLOCKS: [[G_VAL:%.+]] = load volatile double, double* [[G_ADDR_REF]]
   // BLOCKS: store volatile double [[G_VAL]], double* [[G_PRIVATE_ADDR]]
 
   // BLOCKS: [[SIVAR_PRIVATE_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[PRIVATES]], i{{.+}} 0, i{{.+}} 1
   // BLOCKS: [[SIVAR_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
-  // BLOCKS: [[SIVAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[SIVAR_ADDR_REF]]
-  // BLOCKS: [[SIVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[SIVAR_REF]]
+  // BLOCKS: [[SIVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[SIVAR_ADDR_REF]]
   // BLOCKS: store i{{[0-9]+}} [[SIVAR_VAL]], i{{[0-9]+}}* [[SIVAR_PRIVATE_ADDR]]
   // BLOCKS: call void @__kmpc_taskloop(%{{.+}}* @{{.+}}, i32 %{{.+}}, i8* [[RES]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* null)
   // BLOCKS: ret
@@ -180,13 +176,15 @@ int main() {
 // CHECK: [[VEC_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 0
 // CHECK: store [2 x i32]* [[VEC_ADDR]], [2 x i32]** [[VEC_REF]],
 // CHECK: [[T_VAR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 1
-// CHECK: store i32* [[T_VAR_ADDR]], i32** [[T_VAR_REF]],
-// CHECK: [[S_ARR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 2
+// CHECK: [[T_VAR_VAL:%.+]] = load i32, i32* [[T_VAR_ADDR]],
+// CHECK: store i32 [[T_VAR_VAL]], i32* [[T_VAR_REF]],
+// CHECK: [[S_ARR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 3
 // CHECK: store [2 x [[S_DOUBLE_TY]]]* [[S_ARR_ADDR]], [2 x [[S_DOUBLE_TY]]]** [[S_ARR_REF]],
-// CHECK: [[VAR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 3
+// CHECK: [[VAR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 4
 // CHECK: store [[S_DOUBLE_TY]]* [[VAR_ADDR]], [[S_DOUBLE_TY]]** [[VAR_REF]],
-// CHECK: [[SIVAR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 4
-// CHECK: store i{{[0-9]+}}* [[SIVAR]], i{{[0-9]+}}** [[SIVAR_REF]],
+// CHECK: [[SIVAR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 5
+// CHECK: [[SIVAR_VAL:%.+]] = load i32, i32* [[SIVAR]],
+// CHECK: store i{{[0-9]+}} [[SIVAR_VAL]], i{{[0-9]+}}* [[SIVAR_REF]],
 
 // Allocate task.
 // Returns struct kmp_task_t {
@@ -211,7 +209,7 @@ int main() {
 // Constructors for s_arr and var.
 // s_arr;
 // CHECK: [[PRIVATE_S_ARR_REF:%.+]] = getelementptr inbounds [[PRIVATES_MAIN_TY]], [[PRIVATES_MAIN_TY]]* [[PRIVATES]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
-// CHECK: [[S_ARR_ADDR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* [[SHAREDS]], i{{.+}} 0, i{{.+}} 2
+// CHECK: [[S_ARR_ADDR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* [[SHAREDS]], i{{.+}} 0, i{{.+}} 3
 // CHECK: load [2 x [[S_DOUBLE_TY]]]*, [2 x [[S_DOUBLE_TY]]]** [[S_ARR_ADDR_REF]],
 // CHECK: call void [[S_DOUBLE_TY_COPY_CONSTR]]([[S_DOUBLE_TY]]* [[S_ARR_CUR:%[^,]+]],
 // CHECK: getelementptr [[S_DOUBLE_TY]], [[S_DOUBLE_TY]]* [[S_ARR_CUR]], i{{.+}} 1
@@ -221,15 +219,14 @@ int main() {
 
 // var;
 // CHECK: [[PRIVATE_VAR_REF:%.+]] = getelementptr inbounds [[PRIVATES_MAIN_TY]], [[PRIVATES_MAIN_TY]]* [[PRIVATES]], i{{.+}} 0, i{{.+}} 1
-// CHECK: [[VAR_ADDR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* [[SHAREDS]], i{{.+}} 0, i{{.+}} 3
+// CHECK: [[VAR_ADDR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* [[SHAREDS]], i{{.+}} 0, i{{.+}} 4
 // CHECK: [[VAR_REF:%.+]] = load [[S_DOUBLE_TY]]*, [[S_DOUBLE_TY]]** [[VAR_ADDR_REF]],
 // CHECK: call void [[S_DOUBLE_TY_COPY_CONSTR]]([[S_DOUBLE_TY]]* [[PRIVATE_VAR_REF]], [[S_DOUBLE_TY]]* {{.*}}[[VAR_REF]],
 
 // t_var;
 // CHECK: [[PRIVATE_T_VAR_REF:%.+]] = getelementptr inbounds [[PRIVATES_MAIN_TY]], [[PRIVATES_MAIN_TY]]* [[PRIVATES]], i{{.+}} 0, i{{.+}} 2
 // CHECK: [[T_VAR_ADDR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* [[SHAREDS]], i{{.+}} 0, i{{.+}} 1
-// CHECK: [[T_VAR_REF:%.+]] = load i{{.+}}*, i{{.+}}** [[T_VAR_ADDR_REF]],
-// CHECK: [[T_VAR:%.+]] = load i{{.+}}, i{{.+}}* [[T_VAR_REF]],
+// CHECK: [[T_VAR:%.+]] = load i{{.+}}, i{{.+}}* [[T_VAR_ADDR_REF]],
 // CHECK: store i32 [[T_VAR]], i32* [[PRIVATE_T_VAR_REF]],
 
 // vec;
@@ -239,9 +236,8 @@ int main() {
 
 // sivar;
 // CHECK: [[PRIVATE_SIVAR_REF:%.+]] = getelementptr inbounds [[PRIVATES_MAIN_TY]], [[PRIVATES_MAIN_TY]]* [[PRIVATES]], i{{.+}} 0, i{{.+}} 4
-// CHECK: [[SIVAR_ADDR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* [[SHAREDS]], i{{.+}} 0, i{{.+}} 4
-// CHECK: [[SIVAR_REF:%.+]] = load i{{.+}}*, i{{.+}}** [[SIVAR_ADDR_REF]],
-// CHECK: [[SIVAR:%.+]] = load i{{.+}}, i{{.+}}* [[SIVAR_REF]],
+// CHECK: [[SIVAR_ADDR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* [[SHAREDS]], i{{.+}} 0, i{{.+}} 5
+// CHECK: [[SIVAR:%.+]] = load i{{.+}}, i{{.+}}* [[SIVAR_ADDR_REF]],
 // CHECK: store i32 [[SIVAR]], i32* [[PRIVATE_SIVAR_REF]],
 
 // Provide pointer to destructor function, which will destroy private variables at the end of the task.
diff --git a/clang/test/OpenMP/teams_codegen.cpp b/clang/test/OpenMP/teams_codegen.cpp
index c26b5862575..74ae0b87901 100644
--- a/clang/test/OpenMP/teams_codegen.cpp
+++ b/clang/test/OpenMP/teams_codegen.cpp
@@ -260,11 +260,10 @@ int main (int argc, char **argv) {
 // CK4:  ret void
 // CK4-NEXT: }
 
-// CK4:  define {{.*}}void @{{[^,]+}}(i8*** dereferenceable({{.}}) [[ARGC1:%.+]])
-// CK4:  [[ARGCADDR1:%.+]] = alloca i8***
-// CK4:  store i8*** [[ARGC1]], i8**** [[ARGCADDR1]]
-// CK4:  [[CONV1:%.+]] = load i8***, i8**** [[ARGCADDR1]]
-// CK4:  call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC_0]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i8***)* {{.+}} to void (i32*, i32*, ...)*), i8*** [[CONV1]])
+// CK4:  define {{.*}}void @{{[^,]+}}(i8** [[ARGC1:%.+]])
+// CK4:  [[ARGCADDR1:%.+]] = alloca i8**
+// CK4:  store i8** [[ARGC1]], i8*** [[ARGCADDR1]]
+// CK4:  call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC_0]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i8***)* {{.+}} to void (i32*, i32*, ...)*), i8*** [[ARGCADDR1]])
 
 
 #endif // CK4
@@ -330,21 +329,23 @@ int main (int argc, char **argv) {
 // CK5-64:  call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC_0]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*)* @.omp_outlined. to void (i32*, i32*, ...)*), i32* [[CONV]])
 // CK5-32:  call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC_0]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*)* @.omp_outlined. to void (i32*, i32*, ...)*), i32* [[ARGCADDR]])
 
-// CK5:  define {{.*}}void @{{[^,]+}}(i{{.+}} dereferenceable({{.+}}) [[AP:%.+]], i{{.+}} dereferenceable({{.+}}) [[BP:%.+]], i{{.+}} dereferenceable({{.+}}) [[ARGC:%.+]])
+// CK5:  define {{.*}}void @{{[^,]+}}(i{{.+}} [[AP:%.+]], i{{.+}} [[BP:%.+]], i{{.+}}** [[ARGC:%.+]])
 // CK5:  [[AADDR:%.+]] = alloca i{{.+}}
 // CK5:  [[BADDR:%.+]] = alloca i{{.+}}
-// CK5:  [[ARGCADDR:%.+]] = alloca i{{.+}}
+// CK5:  [[ARGCADDR:%.+]] = alloca i{{.+}}**
 // CK5:  [[GBL_TH_NUM:%.+]] = call i32 @__kmpc_global_thread_num(%ident_t* [[DEF_LOC_0]])
 // CK5:  store i{{.+}} [[AP]], i{{.+}}* [[AADDR]]
 // CK5:  store i{{.+}} [[BP]], i{{.+}}* [[BADDR]]
-// CK5:  store i{{.+}} [[ARGC]], i{{.+}}* [[ARGCADDR]]
-// CK5:  [[A_ADDR_VAL:%.+]] = load i32*, i32** [[AADDR]]
-// CK5:  [[B_ADDR_VAL:%.+]] = load i32*, i32** [[BADDR]]
-// CK5:  [[ARGC_ADDR_VAL:%.+]] = load i{{.+}}, i{{.+}}* [[ARGCADDR]]
-// CK5:  [[A_VAL:%.+]] = load i32, i32* [[A_ADDR_VAL]]
-// CK5:  [[B_VAL:%.+]] = load i32, i32* [[B_ADDR_VAL]]
-// CK5:  {{.+}} = call i32 @__kmpc_push_num_teams(%ident_t* [[DEF_LOC_0]], i32 [[GBL_TH_NUM]], i32 [[A_VAL]], i32 [[B_VAL]])
-// CK5:  call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC_0]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i{{.+}})* @.omp_outlined.{{.+}} to void (i32*, i32*, ...)*), i{{.+}} [[ARGC_ADDR_VAL]])
+// CK5:  store i{{.+}}** [[ARGC]], i{{.+}}*** [[ARGCADDR]]
+// CK5-64:  [[ACONV:%.+]] = bitcast i64* [[AADDR]] to i32*
+// CK5-64:  [[BCONV:%.+]] = bitcast i64* [[BADDR]] to i32*
+// CK5-64:  [[ACONVVAL:%.+]] = load i32, i32* [[ACONV]]
+// CK5-64:  [[BCONVVAL:%.+]] = load i32, i32* [[BCONV]]
+// CK5-64:  {{.+}} = call i32 @__kmpc_push_num_teams(%ident_t* [[DEF_LOC_0]], i32 [[GBL_TH_NUM]], i32 [[ACONVVAL]], i32 [[BCONVVAL]])
+// CK5-32:  [[A_VAL:%.+]] = load i32, i32* [[AADDR]]
+// CK5-32:  [[B_VAL:%.+]] = load i32, i32* [[BADDR]]
+// CK5-32:  {{.+}} = call i32 @__kmpc_push_num_teams(%ident_t* [[DEF_LOC_0]], i32 [[GBL_TH_NUM]], i32 [[A_VAL]], i32 [[B_VAL]])
+// CK5:  call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC_0]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i{{.+}})* @.omp_outlined.{{.+}} to void (i32*, i32*, ...)*), i{{.+}}*** [[ARGCADDR]])
 // CK5:  ret void
 // CK5-NEXT: }
 
diff --git a/clang/test/OpenMP/teams_firstprivate_codegen.cpp b/clang/test/OpenMP/teams_firstprivate_codegen.cpp
index c6f084385f7..505d76313a4 100644
--- a/clang/test/OpenMP/teams_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/teams_firstprivate_codegen.cpp
@@ -76,23 +76,20 @@ int main() {
   #pragma omp target
   #pragma omp teams firstprivate(g, sivar)
   {
-    // LAMBDA: define{{.*}} internal{{.*}} void [[OMP_REGION]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* dereferenceable(4) [[G_IN:%.+]], i32* dereferenceable(4) [[SIVAR_IN:%.+]])
+    // LAMBDA: define{{.*}} internal{{.*}} void [[OMP_REGION]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* dereferenceable(4) [[G_IN:%.+]], i32 {{.*}}[[SIVAR_IN:%.+]])
     // LAMBDA: store i{{[0-9]+}}* [[G_IN]], i{{[0-9]+}}** [[G_ADDR:%.+]],
-    // LAMBDA: store i{{[0-9]+}}* [[SIVAR_IN]], i{{[0-9]+}}** [[SIVAR_ADDR:%.+]],
+    // LAMBDA: store i{{[0-9]+}} [[SIVAR_IN]], i{{[0-9]+}}* [[SIVAR_ADDR:%.+]],
     // LAMBDA: [[G_ADDR_VAL:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[G_ADDR]],
-    // LAMBDA: [[SIVAR_ADDR_VAL:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[SIVAR_ADDR]],
     // LAMBDA: [[G_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[G_ADDR_VAL]],
     // LAMBDA: store i{{[0-9]+}} [[G_VAL]], i{{[0-9]+}}* [[G_LOCAL:%.+]],
-    // LAMBDA: [[SIVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[SIVAR_ADDR_VAL]],
-    // LAMBDA: store i{{[0-9]+}} [[SIVAR_VAL]], i{{[0-9]+}}* [[SIVAR_LOCAL:%.+]],
     g = 1;
     sivar = 2;
     // LAMBDA: store i{{[0-9]+}} 1, i{{[0-9]+}}* [[G_LOCAL]],
-    // LAMBDA: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[SIVAR_LOCAL]],
+    // LAMBDA: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[SIVAR_ADDR]],
     // LAMBDA: [[G_PRIVATE_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG:%.+]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
     // LAMBDA: store i{{[0-9]+}}* [[G_LOCAL]], i{{[0-9]+}}** [[G_PRIVATE_ADDR_REF]]
     // LAMBDA: [[SIVAR_PRIVATE_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG:%.+]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
-    // LAMBDA: store i{{[0-9]+}}* [[SIVAR_LOCAL]], i{{[0-9]+}}** [[SIVAR_PRIVATE_ADDR_REF]]
+    // LAMBDA: store i{{[0-9]+}}* [[SIVAR_ADDR]], i{{[0-9]+}}** [[SIVAR_PRIVATE_ADDR_REF]]
     // LAMBDA: call{{.*}} void [[INNER_LAMBDA:@.+]](%{{.+}}* [[ARG]])
     [&]() {
       // LAMBDA: define {{.+}} void [[INNER_LAMBDA]](%{{.+}}* [[ARG_PTR:%.+]])
@@ -130,24 +127,20 @@ int main() {
 }
 
 // CHECK: define internal {{.*}}void [[OMP_OFFLOADING:@.+]](
-// CHECK: call {{.*}}void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_teams(%{{.+}}* @{{.+}}, i{{[0-9]+}} 5, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, [2 x i32]*, i32*, [2 x [[S_FLOAT_TY]]]*, [[S_FLOAT_TY]]*, i{{[0-9]+}}*)* [[OMP_OUTLINED:@.+]] to void
+// CHECK: call {{.*}}void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_teams(%{{.+}}* @{{.+}}, i{{[0-9]+}} 5, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, [2 x i32]*, i32, [2 x [[S_FLOAT_TY]]]*, [[S_FLOAT_TY]]*, i{{[0-9]+}})* [[OMP_OUTLINED:@.+]] to void
 // CHECK: ret
 //
-// CHECK: define internal {{.*}}void [[OMP_OUTLINED]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [2 x i32]* dereferenceable(8) %{{.+}}, i32* dereferenceable(4) %{{.+}}, [2 x [[S_FLOAT_TY]]]* dereferenceable(8) %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}}, i32* dereferenceable(4) [[SIVAR:%.+]])
+// CHECK: define internal {{.*}}void [[OMP_OUTLINED]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [2 x i32]* dereferenceable(8) %{{.+}}, i32 {{.*}}%{{.+}}, [2 x [[S_FLOAT_TY]]]* dereferenceable(8) %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}}, i32 {{.*}}[[SIVAR:%.+]])
 // CHECK: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}},
+// CHECK: [[SIVAR7_PRIV:%.+]] = alloca i{{[0-9]+}},
 // CHECK: [[VEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}],
 // CHECK: [[S_ARR_PRIV:%.+]] = alloca [2 x [[S_FLOAT_TY]]],
 // CHECK: [[VAR_PRIV:%.+]] = alloca [[S_FLOAT_TY]],
-// CHECK: [[SIVAR7_PRIV:%.+]] = alloca i{{[0-9]+}},
 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]],
 
 // CHECK: [[VEC_REF:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** %
-// CHECK: [[T_VAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** %
 // CHECK: [[S_ARR_REF:%.+]] = load [2 x [[S_FLOAT_TY]]]*, [2 x [[S_FLOAT_TY]]]** %
 // CHECK: [[VAR_REF:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** %
-// CHECK: [[SIVAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** %{{.+}},
-// CHECK: [[T_VAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_REF]],
-// CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_PRIV]],
 // CHECK: [[VEC_DEST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8*
 // CHECK: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_REF]] to i8*
 // CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST]], i8* [[VEC_SRC]],
@@ -165,8 +158,6 @@ int main() {
 // CHECK: call {{.*}} [[S_FLOAT_TY_COPY_CONSTR]]([[S_FLOAT_TY]]* [[VAR_PRIV]], [[S_FLOAT_TY]]* {{.*}} [[VAR_REF]], [[ST_TY]]* [[ST_TY_TEMP]])
 // CHECK: call {{.*}} [[ST_TY_DESTR]]([[ST_TY]]* [[ST_TY_TEMP]])
 
-// CHECK: [[SIVAR_REF_ADDR:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[SIVAR_REF]],
-// CHECK: store i{{[0-9]+}} [[SIVAR_REF_ADDR]], i{{[0-9]+}}* [[SIVAR7_PRIV]],
 // CHECK: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[SIVAR7_PRIV]],
 
 // CHECK-DAG: call {{.*}} [[S_FLOAT_TY_DESTR:@.+]]([[S_FLOAT_TY]]* [[VAR_PRIV]])
@@ -174,15 +165,12 @@ int main() {
 // CHECK: ret void
 
 // CHECK: define internal {{.*}}void [[OMP_OFFLOADING_1:@.+]](
-// CHECK: call {{.*}}void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_teams(%{{.+}}* @{{.+}}, i{{[0-9]+}} 1, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, i{{[0-9]+}}*)* [[OMP_OUTLINED_1:@.+]] to void
+// CHECK: call {{.*}}void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_teams(%{{.+}}* @{{.+}}, i{{[0-9]+}} 1, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, i{{[0-9]+}})* [[OMP_OUTLINED_1:@.+]] to void
 // CHECK: ret
 
-// CHECK: define internal {{.*}}void [[OMP_OUTLINED_1]](i{{[0-9]+}}* noalias {{%.+}}, i{{[0-9]+}}* noalias {{%.+}}, i32* dereferenceable(4) [[T_VAR:%.+]])
+// CHECK: define internal {{.*}}void [[OMP_OUTLINED_1]](i{{[0-9]+}}* noalias {{%.+}}, i{{[0-9]+}}* noalias {{%.+}}, i32 {{.*}}[[T_VAR:%.+]])
 // CHECK: [[T_VAR_LOC:%.+]] = alloca i{{[0-9]+}},
-// CHECK: store i{{[0-9]+}}* [[T_VAR]], i{{[0-9]+}}** [[T_VAR_ADDR:%.+]],
-// CHECK: [[T_VAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[T_VAR_ADDR]],
-// CHECK: [[T_VAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_REF]],
-// CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_LOC]],
+// CHECK: store i{{[0-9]+}} [[T_VAR]], i{{[0-9]+}}* [[T_VAR_LOC]],
 // CHECK: ret
 
 // CHECK: define internal {{.*}}void [[OMP_OFFLOADING_2:@.+]](i{{[0-9]+}}* {{.+}} {{%.+}}, [2 x i32]* {{.+}} {{%.+}}, [2 x [[S_INT_TY]]]* {{.+}} {{%.+}}, [[S_INT_TY]]* {{.+}} {{%.+}})
@@ -229,7 +217,7 @@ int main() {
 // CHECK: ret
 
 // CHECK: define internal {{.*}}void [[OMP_OUTLINED_3]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, i32* dereferenceable(4) [[T_VAR:%.+]])
-// CHECK [[T_VAR_LOC:%.+]] = alloca i{{[0-9]+}},
+// CHECK: [[T_VAR_LOC:%.+]] = alloca i{{[0-9]+}},
 // CHECK: store i{{[0-9]+}}* [[T_VAR]], i{{[0-9]+}}** [[T_VAR_ADDR:%.+]],
 // CHECK: [[T_VAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[T_VAR_ADDR]],
 // CHECK: [[T_VAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_REF]],
@@ -254,14 +242,14 @@ struct St {
 void array_func(float a[3], St s[2], int n, long double vla1[n]) {
   double vla2[n][n] __attribute__((aligned(128)));
 // ARRAY: call {{.+}} @__kmpc_fork_teams(
-// ARRAY-DAG: [[PRIV_A:%.+]] = alloca float**,
-// ARRAY-DAG: [[PRIV_S:%.+]] = alloca %struct.St**,
-// ARRAY-64-DAG: [[PRIV_VLA1:%.+]] = alloca ppc_fp128**,
+// ARRAY-DAG: [[PRIV_S:%.+]] = alloca %struct.St*,
+// ARRAY-64-DAG: [[PRIV_VLA1:%.+]] = alloca ppc_fp128*,
 // ARRAY-32-DAG: [[PRIV_VLA1:%.+]] = alloca x86_fp80*,
+// ARRAY-DAG: [[PRIV_A:%.+]] = alloca float*,
 // ARRAY-DAG: [[PRIV_VLA2:%.+]] = alloca double*,
-// ARRAY-DAG: store float** %{{.+}}, float*** [[PRIV_A]],
-// ARRAY-DAG: store %struct.St** %{{.+}}, %struct.St*** [[PRIV_S]],
-// ARRAY-64-DAG: store ppc_fp128** %{{.+}}, ppc_fp128*** [[PRIV_VLA1]],
+// ARRAY-DAG: store float* %{{.+}}, float** [[PRIV_A]],
+// ARRAY-DAG: store %struct.St* %{{.+}}, %struct.St** [[PRIV_S]],
+// ARRAY-64-DAG: store ppc_fp128* %{{.+}}, ppc_fp128** [[PRIV_VLA1]],
 // ARRAY-32-DAG: store x86_fp80* %{{.+}}, x86_fp80** [[PRIV_VLA1]],
 // ARRAY-DAG: store double* %{{.+}}, double** [[PRIV_VLA2]],
 // ARRAY: call i8* @llvm.stacksave()
@@ -274,13 +262,13 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
 }
 
 // ARRAY: @__kmpc_fork_teams(
-// ARRAY-DAG: [[PRIV_S:%.+]] = alloca %struct.St**,
-// ARRAY-64-DAG: [[PRIV_VLA1:%.+]] = alloca ppc_fp128**,
-// ARRAY-32-DAG: [[PRIV_VLA1:%.+]] = alloca x86_fp80**,
+// ARRAY-DAG: [[PRIV_S:%.+]] = alloca %struct.St*,
+// ARRAY-64-DAG: [[PRIV_VLA1:%.+]] = alloca ppc_fp128*,
+// ARRAY-32-DAG: [[PRIV_VLA1:%.+]] = alloca x86_fp80*,
 // ARRAY-DAG: [[PRIV_VLA2:%.+]] = alloca double*,
-// ARRAY-DAG: store %struct.St** %{{.+}}, %struct.St*** [[PRIV_S]],
-// ARRAY-64-DAG: store ppc_fp128** %{{.+}}, ppc_fp128*** [[PRIV_VLA1]],
-// ARRAY-32-DAG: store x86_fp80** %{{.+}}, x86_fp80*** [[PRIV_VLA1]],
+// ARRAY-DAG: store %struct.St* %{{.+}}, %struct.St** [[PRIV_S]],
+// ARRAY-64-DAG: store ppc_fp128* %{{.+}}, ppc_fp128** [[PRIV_VLA1]],
+// ARRAY-32-DAG: store x86_fp80* %{{.+}}, x86_fp80** [[PRIV_VLA1]],
 // ARRAY-DAG: store double* %{{.+}}, double** [[PRIV_VLA2]],
 // ARRAY: call i8* @llvm.stacksave()
 // ARRAY: [[SIZE:%.+]] = mul nuw i{{[0-9]+}} %{{.+}}, 8