Hi rjmccall, hfinkel,
Fixed codegen for reduction operations min, max, && and ||. Codegen for them is
quite similar and I was confused by this similarity.
Also added a call to kmpc_end_reduce() in atomic part of reduction codegen
(call to kmpc_end_reduce_nowait() is not required).
http://reviews.llvm.org/D9513
Files:
lib/CodeGen/CGOpenMPRuntime.cpp
lib/Sema/SemaOpenMP.cpp
test/OpenMP/for_reduction_codegen.cpp
test/OpenMP/parallel_reduction_codegen.cpp
test/OpenMP/sections_reduction_codegen.cpp
EMAIL PREFERENCES
http://reviews.llvm.org/settings/panel/emailpreferences/
Index: test/OpenMP/for_reduction_codegen.cpp
===================================================================
--- test/OpenMP/for_reduction_codegen.cpp
+++ test/OpenMP/for_reduction_codegen.cpp
@@ -95,6 +95,7 @@
// LAMBDA: [[G_PRIV_VAL:%.+]] = load double, double* [[G_PRIVATE_ADDR]]
// LAMBDA: fadd double
// LAMBDA: cmpxchg i64*
+ // LAMBDA: call void @__kmpc_end_reduce(
// LAMBDA: br label %[[REDUCTION_DONE]]
// LAMBDA: [[REDUCTION_DONE]]
// LAMBDA: ret void
@@ -154,6 +155,7 @@
// BLOCKS: [[G_PRIV_VAL:%.+]] = load double, double* [[G_PRIVATE_ADDR]]
// BLOCKS: fadd double
// BLOCKS: cmpxchg i64*
+ // BLOCKS: call void @__kmpc_end_reduce(
// BLOCKS: br label %[[REDUCTION_DONE]]
// BLOCKS: [[REDUCTION_DONE]]
// BLOCKS: ret void
@@ -273,26 +275,25 @@
// var1 = var1.operator &&(var1_reduction);
// CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_REF]])
// CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0
-// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]]
+// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]]
// CHECK: [[TRUE]]
// CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_PRIV]])
// CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0
-// CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]]
-// CHECK: [[TRUE2]]
-// CHECK: br label %[[END2:.+]]
-// CHECK: [[FALSE2]]
// CHECK: br label %[[END2]]
// CHECK: [[END2]]
-// CHECK: [[COND_LVALUE:%.+]] = phi [[S_FLOAT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ]
+// CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ]
+// CHECK: [[CONV:%.+]] = uitofp i1 [[COND_LVALUE]] to float
+// CHECK: call void @{{.+}}([[S_FLOAT_TY]]* [[COND_LVALUE:%.+]], float [[CONV]])
// CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_REF]] to i8*
// CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8*
// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
// t_var1 = min(t_var1, t_var1_reduction);
// CHECK: [[T_VAR1_VAL:%.+]] = load float, float* [[T_VAR1_REF]],
// CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load float, float* [[T_VAR1_PRIV]],
// CHECK: [[CMP:%.+]] = fcmp olt float [[T_VAR1_VAL]], [[T_VAR1_PRIV_VAL]]
-// CHECK: [[UP:%.+]] = uitofp i1 [[CMP]] to float
+// CHECK: br i1 [[CMP]]
+// CHECK: [[UP:%.+]] = phi float
// CHECK: store float [[UP]], float* [[T_VAR1_REF]],
// __kmpc_end_reduce(<loc>, <gtid>, &<lock>);
@@ -333,17 +334,15 @@
// CHECK: call void @__kmpc_critical(
// CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_REF]])
// CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0
-// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]]
+// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]]
// CHECK: [[TRUE]]
// CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_PRIV]])
// CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0
-// CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]]
-// CHECK: [[TRUE2]]
-// CHECK: br label %[[END2:.+]]
-// CHECK: [[FALSE2]]
// CHECK: br label %[[END2]]
// CHECK: [[END2]]
-// CHECK: [[COND_LVALUE:%.+]] = phi [[S_FLOAT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ]
+// CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ]
+// CHECK: [[CONV:%.+]] = uitofp i1 [[COND_LVALUE]] to float
+// CHECK: call void @{{.+}}([[S_FLOAT_TY]]* [[COND_LVALUE:%.+]], float [[CONV]])
// CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_REF]] to i8*
// CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8*
// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
@@ -356,9 +355,10 @@
// CHECK: [[OLD1:%.+]] = bitcast i32 %{{.+}} to float
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
-// CHECK: [[ORIG_OLD:%.+]] = phi float [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %[[CONT]] ]
+// CHECK: [[ORIG_OLD:%.+]] = phi float [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %{{.+}} ]
// CHECK: [[CMP:%.+]] = fcmp olt float
-// CHECK: [[UP:%.+]] = uitofp i1 [[CMP]] to float
+// CHECK: br i1 [[CMP]]
+// CHECK: [[UP:%.+]] = phi float
// CHECK: [[ORIG_OLD_INT:%.+]] = bitcast float [[ORIG_OLD]] to i32
// CHECK: [[UP_INT:%.+]] = bitcast float [[UP]] to i32
// CHECK: [[T_VAR1_REF_INT:%.+]] = bitcast float* [[T_VAR1_REF]] to i32*
@@ -369,6 +369,9 @@
// CHECK: br i1 [[SUCCESS_FAIL]], label %[[ATOMIC_DONE:.+]], label %[[CONT]]
// CHECK: [[ATOMIC_DONE]]
+// __kmpc_end_reduce(<loc>, <gtid>, &<lock>);
+// CHECK: call void @__kmpc_end_reduce(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], [8 x i32]* [[REDUCTION_LOCK]])
+
// break;
// CHECK: br label %[[RED_DONE]]
// CHECK: [[RED_DONE]]
@@ -437,26 +440,25 @@
// var1_lhs = var1_lhs.operator &&(var1_rhs);
// CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_LHS]])
// CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0
-// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]]
+// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]]
// CHECK: [[TRUE]]
// CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_RHS]])
// CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0
-// CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]]
-// CHECK: [[TRUE2]]
-// CHECK: br label %[[END2:.+]]
-// CHECK: [[FALSE2]]
// CHECK: br label %[[END2]]
// CHECK: [[END2]]
-// CHECK: [[COND_LVALUE:%.+]] = phi [[S_FLOAT_TY]]* [ [[VAR1_LHS]], %[[TRUE2]] ], [ [[VAR1_RHS]], %[[FALSE2]] ]
+// CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ]
+// CHECK: [[CONV:%.+]] = uitofp i1 [[COND_LVALUE]] to float
+// CHECK: call void @{{.+}}([[S_FLOAT_TY]]* [[COND_LVALUE:%.+]], float [[CONV]])
// CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_LHS]] to i8*
// CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8*
// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
// t_var1_lhs = min(t_var1_lhs, t_var1_rhs);
// CHECK: [[T_VAR1_LHS_VAL:%.+]] = load float, float* [[T_VAR1_LHS]],
// CHECK: [[T_VAR1_RHS_VAL:%.+]] = load float, float* [[T_VAR1_RHS]],
// CHECK: [[CMP:%.+]] = fcmp olt float [[T_VAR1_LHS_VAL]], [[T_VAR1_RHS_VAL]]
-// CHECK: [[UP:%.+]] = uitofp i1 [[CMP]] to float
+// CHECK: br i1 [[CMP]]
+// CHECK: [[UP:%.+]] = phi float
// CHECK: store float [[UP]], float* [[T_VAR1_LHS]],
// CHECK: ret void
@@ -552,26 +554,25 @@
// var1 = var1.operator &&(var1_reduction);
// CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_REF]])
// CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
-// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]]
+// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]]
// CHECK: [[TRUE]]
// CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_PRIV]])
// CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
-// CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]]
-// CHECK: [[TRUE2]]
-// CHECK: br label %[[END2:.+]]
-// CHECK: [[FALSE2]]
// CHECK: br label %[[END2]]
// CHECK: [[END2]]
-// CHECK: [[COND_LVALUE:%.+]] = phi [[S_INT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ]
+// CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ]
+// CHECK: [[CONV:%.+]] = zext i1 [[COND_LVALUE]] to i32
+// CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]])
// CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8*
// CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8*
// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
// t_var1 = min(t_var1, t_var1_reduction);
// CHECK: [[T_VAR1_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_REF]],
// CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_PRIV]],
// CHECK: [[CMP:%.+]] = icmp slt i{{[0-9]+}} [[T_VAR1_VAL]], [[T_VAR1_PRIV_VAL]]
-// CHECK: [[UP:%.+]] = zext i1 [[CMP]] to i{{[0-9]+}}
+// CHECK: br i1 [[CMP]]
+// CHECK: [[UP:%.+]] = phi i32
// CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR1_REF]],
// __kmpc_end_reduce_nowait(<loc>, <gtid>, &<lock>);
@@ -597,17 +598,15 @@
// CHECK: call void @__kmpc_critical(
// CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_REF]])
// CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
-// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]]
+// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]]
// CHECK: [[TRUE]]
// CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_PRIV]])
// CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
-// CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]]
-// CHECK: [[TRUE2]]
-// CHECK: br label %[[END2:.+]]
-// CHECK: [[FALSE2]]
// CHECK: br label %[[END2]]
// CHECK: [[END2]]
-// CHECK: [[COND_LVALUE:%.+]] = phi [[S_INT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ]
+// CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ]
+// CHECK: [[CONV:%.+]] = zext i1 [[COND_LVALUE]] to i32
+// CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]])
// CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8*
// CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8*
// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
@@ -683,26 +682,25 @@
// var1_lhs = var1_lhs.operator &&(var1_rhs);
// CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_LHS]])
// CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
-// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]]
+// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]]
// CHECK: [[TRUE]]
// CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_RHS]])
// CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
-// CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]]
-// CHECK: [[TRUE2]]
-// CHECK: br label %[[END2:.+]]
-// CHECK: [[FALSE2]]
// CHECK: br label %[[END2]]
// CHECK: [[END2]]
-// CHECK: [[COND_LVALUE:%.+]] = phi [[S_INT_TY]]* [ [[VAR1_LHS]], %[[TRUE2]] ], [ [[VAR1_RHS]], %[[FALSE2]] ]
+// CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ]
+// CHECK: [[CONV:%.+]] = zext i1 [[COND_LVALUE]] to i32
+// CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]])
// CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_LHS]] to i8*
// CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8*
// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
// t_var1_lhs = min(t_var1_lhs, t_var1_rhs);
// CHECK: [[T_VAR1_LHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_LHS]],
// CHECK: [[T_VAR1_RHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_RHS]],
// CHECK: [[CMP:%.+]] = icmp slt i{{[0-9]+}} [[T_VAR1_LHS_VAL]], [[T_VAR1_RHS_VAL]]
-// CHECK: [[UP:%.+]] = zext i1 [[CMP]] to i{{[0-9]+}}
+// CHECK: br i1 [[CMP]]
+// CHECK: [[UP:%.+]] = phi i32
// CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR1_LHS]],
// CHECK: ret void
Index: test/OpenMP/sections_reduction_codegen.cpp
===================================================================
--- test/OpenMP/sections_reduction_codegen.cpp
+++ test/OpenMP/sections_reduction_codegen.cpp
@@ -91,6 +91,7 @@
// LAMBDA: [[G_PRIV_VAL:%.+]] = load double, double* [[G_PRIVATE_ADDR]]
// LAMBDA: fadd double
// LAMBDA: cmpxchg i64*
+ // LAMBDA: call void @__kmpc_end_reduce(
// LAMBDA: br label %[[REDUCTION_DONE]]
// LAMBDA: [[REDUCTION_DONE]]
// LAMBDA: ret void
@@ -151,6 +152,7 @@
// BLOCKS: [[G_PRIV_VAL:%.+]] = load double, double* [[G_PRIVATE_ADDR]]
// BLOCKS: fadd double
// BLOCKS: cmpxchg i64*
+ // BLOCKS: call void @__kmpc_end_reduce(
// BLOCKS: br label %[[REDUCTION_DONE]]
// BLOCKS: [[REDUCTION_DONE]]
// BLOCKS: ret void
@@ -314,26 +316,25 @@
// var1 = var1.operator &&(var1_reduction);
// CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_REF]])
// CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
-// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]]
+// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]]
// CHECK: [[TRUE]]
// CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_PRIV]])
// CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
-// CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]]
-// CHECK: [[TRUE2]]
-// CHECK: br label %[[END2:.+]]
-// CHECK: [[FALSE2]]
// CHECK: br label %[[END2]]
// CHECK: [[END2]]
-// CHECK: [[COND_LVALUE:%.+]] = phi [[S_INT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ]
+// CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ]
+// CHECK: [[CONV:%.+]] = zext i1 [[COND_LVALUE]] to i32
+// CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]])
// CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8*
// CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8*
// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
// t_var1 = min(t_var1, t_var1_reduction);
// CHECK: [[T_VAR1_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_REF]],
// CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_PRIV]],
// CHECK: [[CMP:%.+]] = icmp slt i{{[0-9]+}} [[T_VAR1_VAL]], [[T_VAR1_PRIV_VAL]]
-// CHECK: [[UP:%.+]] = zext i1 [[CMP]] to i{{[0-9]+}}
+// CHECK: br i1 [[CMP]]
+// CHECK: [[UP:%.+]] = phi i32
// CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR1_REF]],
// __kmpc_end_reduce_nowait(<loc>, <gtid>, &<lock>);
@@ -363,13 +364,11 @@
// CHECK: [[TRUE]]
// CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_PRIV]])
// CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
-// CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]]
-// CHECK: [[TRUE2]]
-// CHECK: br label %[[END2:.+]]
-// CHECK: [[FALSE2]]
// CHECK: br label %[[END2]]
// CHECK: [[END2]]
-// CHECK: [[COND_LVALUE:%.+]] = phi [[S_INT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ]
+// CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ]
+// CHECK: [[CONV:%.+]] = zext i1 [[COND_LVALUE]] to i32
+// CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]])
// CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8*
// CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8*
// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
@@ -445,26 +444,25 @@
// var1_lhs = var1_lhs.operator &&(var1_rhs);
// CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_LHS]])
// CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
-// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]]
+// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]]
// CHECK: [[TRUE]]
// CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_RHS]])
// CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
-// CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]]
-// CHECK: [[TRUE2]]
-// CHECK: br label %[[END2:.+]]
-// CHECK: [[FALSE2]]
// CHECK: br label %[[END2]]
// CHECK: [[END2]]
-// CHECK: [[COND_LVALUE:%.+]] = phi [[S_INT_TY]]* [ [[VAR1_LHS]], %[[TRUE2]] ], [ [[VAR1_RHS]], %[[FALSE2]] ]
+// CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ]
+// CHECK: [[CONV:%.+]] = zext i1 [[COND_LVALUE]] to i32
+// CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]])
// CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_LHS]] to i8*
// CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8*
// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
// t_var1_lhs = min(t_var1_lhs, t_var1_rhs);
// CHECK: [[T_VAR1_LHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_LHS]],
// CHECK: [[T_VAR1_RHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_RHS]],
// CHECK: [[CMP:%.+]] = icmp slt i{{[0-9]+}} [[T_VAR1_LHS_VAL]], [[T_VAR1_RHS_VAL]]
-// CHECK: [[UP:%.+]] = zext i1 [[CMP]] to i{{[0-9]+}}
+// CHECK: br i1 [[CMP]]
+// CHECK: [[UP:%.+]] = phi i32
// CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR1_LHS]],
// CHECK: ret void
Index: test/OpenMP/parallel_reduction_codegen.cpp
===================================================================
--- test/OpenMP/parallel_reduction_codegen.cpp
+++ test/OpenMP/parallel_reduction_codegen.cpp
@@ -267,26 +267,25 @@
// var1 = var1.operator &&(var1_reduction);
// CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_REF]])
// CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0
-// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]]
+// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]]
// CHECK: [[TRUE]]
// CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_PRIV]])
// CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0
-// CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]]
-// CHECK: [[TRUE2]]
-// CHECK: br label %[[END2:.+]]
-// CHECK: [[FALSE2]]
// CHECK: br label %[[END2]]
// CHECK: [[END2]]
-// CHECK: [[COND_LVALUE:%.+]] = phi [[S_FLOAT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ]
+// CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ]
+// CHECK: [[CONV:%.+]] = uitofp i1 [[COND_LVALUE]] to float
+// CHECK: call void @{{.+}}([[S_FLOAT_TY]]* [[COND_LVALUE:%.+]], float [[CONV]])
// CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_REF]] to i8*
// CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8*
// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
// t_var1 = min(t_var1, t_var1_reduction);
// CHECK: [[T_VAR1_VAL:%.+]] = load float, float* [[T_VAR1_REF]],
// CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load float, float* [[T_VAR1_PRIV]],
// CHECK: [[CMP:%.+]] = fcmp olt float [[T_VAR1_VAL]], [[T_VAR1_PRIV_VAL]]
-// CHECK: [[UP:%.+]] = uitofp i1 [[CMP]] to float
+// CHECK: br i1 [[CMP]]
+// CHECK: [[UP:%.+]] = phi float
// CHECK: store float [[UP]], float* [[T_VAR1_REF]],
// __kmpc_end_reduce_nowait(<loc>, <gtid>, &<lock>);
@@ -327,17 +326,15 @@
// CHECK: call void @__kmpc_critical(
// CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_REF]])
// CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0
-// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]]
+// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]]
// CHECK: [[TRUE]]
// CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_PRIV]])
// CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0
-// CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]]
-// CHECK: [[TRUE2]]
-// CHECK: br label %[[END2:.+]]
-// CHECK: [[FALSE2]]
// CHECK: br label %[[END2]]
// CHECK: [[END2]]
-// CHECK: [[COND_LVALUE:%.+]] = phi [[S_FLOAT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ]
+// CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ]
+// CHECK: [[CONV:%.+]] = uitofp i1 [[COND_LVALUE]] to float
+// CHECK: call void @{{.+}}([[S_FLOAT_TY]]* [[COND_LVALUE:%.+]], float [[CONV]])
// CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_REF]] to i8*
// CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8*
// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
@@ -350,9 +347,10 @@
// CHECK: [[OLD1:%.+]] = bitcast i32 %{{.+}} to float
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
-// CHECK: [[ORIG_OLD:%.+]] = phi float [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %[[CONT]] ]
+// CHECK: [[ORIG_OLD:%.+]] = phi float [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %{{.+}} ]
// CHECK: [[CMP:%.+]] = fcmp olt float
-// CHECK: [[UP:%.+]] = uitofp i1 [[CMP]] to float
+// CHECK: br i1 [[CMP]]
+// CHECK: [[UP:%.+]] = phi float
// CHECK: [[ORIG_OLD_INT:%.+]] = bitcast float [[ORIG_OLD]] to i32
// CHECK: [[UP_INT:%.+]] = bitcast float [[UP]] to i32
// CHECK: [[T_VAR1_REF_INT:%.+]] = bitcast float* [[T_VAR1_REF]] to i32*
@@ -430,26 +428,25 @@
// var1_lhs = var1_lhs.operator &&(var1_rhs);
// CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_LHS]])
// CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0
-// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]]
+// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]]
// CHECK: [[TRUE]]
// CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_RHS]])
// CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0
-// CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]]
-// CHECK: [[TRUE2]]
-// CHECK: br label %[[END2:.+]]
-// CHECK: [[FALSE2]]
// CHECK: br label %[[END2]]
// CHECK: [[END2]]
-// CHECK: [[COND_LVALUE:%.+]] = phi [[S_FLOAT_TY]]* [ [[VAR1_LHS]], %[[TRUE2]] ], [ [[VAR1_RHS]], %[[FALSE2]] ]
+// CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ]
+// CHECK: [[CONV:%.+]] = uitofp i1 [[COND_LVALUE]] to float
+// CHECK: call void @{{.+}}([[S_FLOAT_TY]]* [[COND_LVALUE:%.+]], float [[CONV]])
// CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_LHS]] to i8*
// CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8*
// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
// t_var1_lhs = min(t_var1_lhs, t_var1_rhs);
// CHECK: [[T_VAR1_LHS_VAL:%.+]] = load float, float* [[T_VAR1_LHS]],
// CHECK: [[T_VAR1_RHS_VAL:%.+]] = load float, float* [[T_VAR1_RHS]],
// CHECK: [[CMP:%.+]] = fcmp olt float [[T_VAR1_LHS_VAL]], [[T_VAR1_RHS_VAL]]
-// CHECK: [[UP:%.+]] = uitofp i1 [[CMP]] to float
+// CHECK: br i1 [[CMP]]
+// CHECK: [[UP:%.+]] = phi float
// CHECK: store float [[UP]], float* [[T_VAR1_LHS]],
// CHECK: ret void
@@ -538,26 +535,24 @@
// var1 = var1.operator &&(var1_reduction);
// CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_REF]])
// CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
-// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]]
+// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]]
// CHECK: [[TRUE]]
// CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_PRIV]])
// CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
-// CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]]
-// CHECK: [[TRUE2]]
-// CHECK: br label %[[END2:.+]]
-// CHECK: [[FALSE2]]
-// CHECK: br label %[[END2]]
// CHECK: [[END2]]
-// CHECK: [[COND_LVALUE:%.+]] = phi [[S_INT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ]
+// CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ]
+// CHECK: [[CONV:%.+]] = zext i1 [[COND_LVALUE]] to i32
+// CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]])
// CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8*
// CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8*
// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
// t_var1 = min(t_var1, t_var1_reduction);
// CHECK: [[T_VAR1_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_REF]],
// CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_PRIV]],
// CHECK: [[CMP:%.+]] = icmp slt i{{[0-9]+}} [[T_VAR1_VAL]], [[T_VAR1_PRIV_VAL]]
-// CHECK: [[UP:%.+]] = zext i1 [[CMP]] to i{{[0-9]+}}
+// CHECK: br i1 [[CMP]]
+// CHECK: [[UP:%.+]] = phi i32
// CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR1_REF]],
// __kmpc_end_reduce_nowait(<loc>, <gtid>, &<lock>);
@@ -583,17 +578,15 @@
// CHECK: call void @__kmpc_critical(
// CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_REF]])
// CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
-// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]]
+// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]]
// CHECK: [[TRUE]]
// CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_PRIV]])
// CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
-// CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]]
-// CHECK: [[TRUE2]]
-// CHECK: br label %[[END2:.+]]
-// CHECK: [[FALSE2]]
// CHECK: br label %[[END2]]
// CHECK: [[END2]]
-// CHECK: [[COND_LVALUE:%.+]] = phi [[S_INT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ]
+// CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ]
+// CHECK: [[CONV:%.+]] = zext i1 [[COND_LVALUE]] to i32
+// CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]])
// CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8*
// CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8*
// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
@@ -670,26 +663,25 @@
// var1_lhs = var1_lhs.operator &&(var1_rhs);
// CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_LHS]])
// CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
-// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]]
+// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]]
// CHECK: [[TRUE]]
// CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_RHS]])
// CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
-// CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]]
-// CHECK: [[TRUE2]]
-// CHECK: br label %[[END2:.+]]
-// CHECK: [[FALSE2]]
// CHECK: br label %[[END2]]
// CHECK: [[END2]]
-// CHECK: [[COND_LVALUE:%.+]] = phi [[S_INT_TY]]* [ [[VAR1_LHS]], %[[TRUE2]] ], [ [[VAR1_RHS]], %[[FALSE2]] ]
+// CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ]
+// CHECK: [[CONV:%.+]] = zext i1 [[COND_LVALUE]] to i32
+// CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]])
// CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_LHS]] to i8*
// CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8*
// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
// t_var1_lhs = min(t_var1_lhs, t_var1_rhs);
// CHECK: [[T_VAR1_LHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_LHS]],
// CHECK: [[T_VAR1_RHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_RHS]],
// CHECK: [[CMP:%.+]] = icmp slt i{{[0-9]+}} [[T_VAR1_LHS_VAL]], [[T_VAR1_RHS_VAL]]
-// CHECK: [[UP:%.+]] = zext i1 [[CMP]] to i{{[0-9]+}}
+// CHECK: br i1 [[CMP]]
+// CHECK: [[UP:%.+]] = phi i32
// CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR1_LHS]],
// CHECK: ret void
Index: lib/Sema/SemaOpenMP.cpp
===================================================================
--- lib/Sema/SemaOpenMP.cpp
+++ lib/Sema/SemaOpenMP.cpp
@@ -5667,7 +5667,7 @@
BuildBinOp(DSAStack->getCurScope(), ReductionId.getLocStart(), BOK,
LHSDRE, RHSDRE);
if (ReductionOp.isUsable()) {
- if (BOK != BO_LOr && BOK != BO_LAnd) {
+ if (BOK != BO_LT && BOK != BO_GT) {
ReductionOp =
BuildBinOp(DSAStack->getCurScope(), ReductionId.getLocStart(),
BO_Assign, LHSDRE, ReductionOp.get());
Index: lib/CodeGen/CGOpenMPRuntime.cpp
===================================================================
--- lib/CodeGen/CGOpenMPRuntime.cpp
+++ lib/CodeGen/CGOpenMPRuntime.cpp
@@ -2121,6 +2121,7 @@
// ...
// Atomic(<LHSExprs>[i] = RedOp<i>(*<LHSExprs>[i], *<RHSExprs>[i]));
// ...
+ // [__kmpc_end_reduce(<loc>, <gtid>, &<lock>);]
// break;
// default:;
// }
@@ -2221,28 +2222,43 @@
{
CodeGenFunction::RunCleanupsScope Scope(CGF);
+ if (!WithNowait) {
+ // Add emission of __kmpc_end_reduce(<loc>, <gtid>, &<lock>);
+ llvm::Value *EndArgs[] = {
+ IdentTLoc, // ident_t *<loc>
+ ThreadId, // i32 <gtid>
+ Lock // kmp_critical_name *&<lock>
+ };
+ CGF.EHStack
+ .pushCleanup<CallEndCleanup<std::extent<decltype(EndArgs)>::value>>(
+ NormalAndEHCleanup,
+ createRuntimeFunction(OMPRTL__kmpc_end_reduce),
+ llvm::makeArrayRef(EndArgs));
+ }
auto I = LHSExprs.begin();
for (auto *E : ReductionOps) {
const Expr *XExpr = nullptr;
const Expr *EExpr = nullptr;
const Expr *UpExpr = nullptr;
BinaryOperatorKind BO = BO_Comma;
- // Try to emit update expression as a simple atomic.
- if (auto *ACO = dyn_cast<AbstractConditionalOperator>(E)) {
- // If this is a conditional operator, analyze it's condition for
- // min/max reduction operator.
- E = ACO->getCond();
- }
if (auto *BO = dyn_cast<BinaryOperator>(E)) {
if (BO->getOpcode() == BO_Assign) {
XExpr = BO->getLHS();
UpExpr = BO->getRHS();
}
}
- // Analyze RHS part of the whole expression.
- if (UpExpr) {
+ // Try to emit update expression as a simple atomic.
+ auto *RHSExpr = UpExpr;
+ if (RHSExpr) {
+ // Analyze RHS part of the whole expression.
+ if (auto *ACO = dyn_cast<AbstractConditionalOperator>(
+ RHSExpr->IgnoreParenImpCasts())) {
+ // If this is a conditional operator, analyze it's condition for
+ // min/max reduction operator.
+ RHSExpr = ACO->getCond();
+ }
if (auto *BORHS =
- dyn_cast<BinaryOperator>(UpExpr->IgnoreParenImpCasts())) {
+ dyn_cast<BinaryOperator>(RHSExpr->IgnoreParenImpCasts())) {
EExpr = BORHS->getRHS();
BO = BORHS->getOpcode();
}
_______________________________________________
cfe-commits mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
