[clang] 71ffac2 - [OPENMP]Fix PR45132: OpenMP doacross loop nest with a decreasing

[Alexey Bataev via cfe-commits]

Author: Alexey Bataev
Date: 2020-03-10T12:13:58-04:00
New Revision: 71ffac21f7463465c369ba12dc496f9f5338e6b4

URL: https://github.com/llvm/llvm-project/commit/71ffac21f7463465c369ba12dc496f9f5338e6b4
DIFF: https://github.com/llvm/llvm-project/commit/71ffac21f7463465c369ba12dc496f9f5338e6b4.diff

LOG: [OPENMP]Fix PR45132: OpenMP doacross loop nest with a decreasing
induction variable abends.

Used incorrect loop bound when trying to calculate the index in the vec
array for doacross construct in the loops with the reverse order.

Added: 
    

Modified: 
    clang/lib/Sema/SemaOpenMP.cpp
    clang/test/OpenMP/ordered_doacross_codegen.cpp

Removed: 
    


################################################################################
diff  --git a/clang/lib/Sema/SemaOpenMP.cpp b/clang/lib/Sema/SemaOpenMP.cpp
index c0fb56c47371..7a62b2971eb1 100644
--- a/clang/lib/Sema/SemaOpenMP.cpp
+++ b/clang/lib/Sema/SemaOpenMP.cpp
@@ -6884,7 +6884,7 @@ Expr *OpenMPIterationSpaceChecker::buildOrderedLoopData(
     // Upper - Lower
     Expr *Upper = TestIsLessOp.getValue()
                       ? Cnt
-                      : tryBuildCapture(SemaRef, UB, Captures).get();
+                      : tryBuildCapture(SemaRef, LB, Captures).get();
     Expr *Lower = TestIsLessOp.getValue()
                       ? tryBuildCapture(SemaRef, LB, Captures).get()
                       : Cnt;

diff  --git a/clang/test/OpenMP/ordered_doacross_codegen.cpp b/clang/test/OpenMP/ordered_doacross_codegen.cpp
index 2b610a270dd9..836f938fbce6 100644
--- a/clang/test/OpenMP/ordered_doacross_codegen.cpp
+++ b/clang/test/OpenMP/ordered_doacross_codegen.cpp
@@ -84,6 +84,64 @@ int main() {
   return 0;
 }
 
+// CHECK-LABEL: main1
+int main1() {
+// CHECK: [[DIMS:%.+]] = alloca [1 x [[KMP_DIM]]],
+// CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num([[IDENT:%.+]])
+// CHECK: icmp
+// CHECK-NEXT: br i1 %
+// CHECK: [[CAST:%.+]] = bitcast [1 x [[KMP_DIM]]]* [[DIMS]] to i8*
+// CHECK: call void @llvm.memset.p0i8.i64(i8* align 8 [[CAST]], i8 0, i64 24, i1 false)
+// CHECK: [[DIM:%.+]] = getelementptr inbounds [1 x [[KMP_DIM]]], [1 x [[KMP_DIM]]]* [[DIMS]], i64 0, i64 0
+// CHECK: getelementptr inbounds [[KMP_DIM]], [[KMP_DIM]]* [[DIM]], i32 0, i32 1
+// CHECK: store i64 %{{.+}}, i64* %
+// CHECK: getelementptr inbounds [[KMP_DIM]], [[KMP_DIM]]* [[DIM]], i32 0, i32 2
+// CHECK: store i64 1, i64* %
+// CHECK: [[DIM:%.+]] = getelementptr inbounds [1 x [[KMP_DIM]]], [1 x [[KMP_DIM]]]* [[DIMS]], i64 0, i64 0
+// CHECK: [[CAST:%.+]] = bitcast [[KMP_DIM]]* [[DIM]] to i8*
+// CHECK: call void @__kmpc_doacross_init([[IDENT]], i32 [[GTID]], i32 1, i8* [[CAST]])
+// CHECK: call void @__kmpc_for_static_init_4(%struct.ident_t* @{{.+}}, i32 [[GTID]], i32 33, i32* %{{.+}}, i32* %{{.+}}, i32* %{{.+}}, i32* %{{.+}}, i32 1, i32 1)
+#pragma omp for ordered(1)
+  for (int i = n; i > 0; --i) {
+    a[i] = b[i] + 1;
+    foo();
+// CHECK: invoke void [[FOO:.+]](
+// CHECK: [[UB_VAL:%.+]] = load i32, i32* [[UB:%.+]],
+// CHECK-NEXT: [[I_VAL:%.+]] = load i32, i32* [[I:%.+]],
+// CHECK-NEXT: sub nsw i32 [[UB_VAL]], [[I_VAL]]
+// CHECK-NEXT: sdiv i32 %{{.+}}, 1
+// CHECK-NEXT: sext i32 %{{.+}} to i64
+// CHECK-NEXT: [[TMP:%.+]] = getelementptr inbounds [1 x i64], [1 x i64]* [[CNT:%.+]], i64 0, i64 0
+// CHECK-NEXT: store i64 %{{.+}}, i64* [[TMP]],
+// CHECK-NEXT: [[TMP:%.+]] = getelementptr inbounds [1 x i64], [1 x i64]* [[CNT]], i64 0, i64 0
+// CHECK-NEXT: call void @__kmpc_doacross_post([[IDENT]], i32 [[GTID]], i64* [[TMP]])
+#pragma omp ordered depend(source)
+    c[i] = c[i] + 1;
+    foo();
+// CHECK: invoke void [[FOO]]
+// CHECK: [[UB_VAL:%.+]] = load i32, i32* [[UB]],
+// CHECK-NEXT: [[I_VAL:%.+]] = load i32, i32* [[I]],
+// CHECK-NEXT: [[SUB:%.+]] = sub nsw i32 [[I_VAL]], 2
+// CHECK-NEXT: sub nsw i32 [[UB_VAL]], [[SUB]]
+// CHECK-NEXT: sdiv i32 %{{.+}}, 1
+// CHECK-NEXT: sext i32 %{{.+}} to i64
+// CHECK-NEXT: [[TMP:%.+]] = getelementptr inbounds [1 x i64], [1 x i64]* [[CNT:%.+]], i64 0, i64 0
+// CHECK-NEXT: store i64 %{{.+}}, i64* [[TMP]],
+// CHECK-NEXT: [[TMP:%.+]] = getelementptr inbounds [1 x i64], [1 x i64]* [[CNT]], i64 0, i64 0
+// CHECK-NEXT: call void @__kmpc_doacross_wait([[IDENT]], i32 [[GTID]], i64* [[TMP]])
+#pragma omp ordered depend(sink : i - 2)
+    d[i] = a[i - 2];
+  }
+  // CHECK: landingpad
+  // CHECK: call void @__kmpc_doacross_fini([[IDENT]], i32 [[GTID]])
+  // CHECK: br label %
+
+  // CHECK: call void @__kmpc_for_static_fini(
+  // CHECK: call void @__kmpc_doacross_fini([[IDENT]], i32 [[GTID]])
+  // CHECK: ret i32 0
+  return 0;
+}
+
 // CHECK: define {{.+}}TestStruct
 template <typename T>
 struct TestStruct {