This is an archive of the discontinued LLVM Phabricator instance.

Differential D9214

[OPENMP] Codegen for 'firstprivate' clause in 'sections' directive.
ClosedPublic

Authored by ABataev on Apr 23 2015, 4:21 AM.

Details

Reviewers
rjmccall
hfinkel
Commits
rG2cb9b95adf7a: [OPENMP] Codegen for 'firstprivate' clause in 'sections' directive.
rC235691: [OPENMP] Codegen for 'firstprivate' clause in 'sections' directive.
rL235691: [OPENMP] Codegen for 'firstprivate' clause in 'sections' directive.
Summary

If there are 2 or more sections in a 'section' directive the following code is generated:

<init for firstprivates>
@__kmpc_cancel_barrier();// To avoid data race in firstprivate init
@__kmpc_for_static_init_4();
<BODY for sections directive>
@__kmpc_for_static_fini()

If there is only one section, the following code is generated:

if (@__kmpc_single()) {
<init for firstprivates>
@__kmpc_end_single();
}
@__kmpc_cancel_barrier(); // To avoid data race in firstprivate init

Diff Detail

Repository
rL LLVM

Event Timeline

ABataev updated this revision to Diff 24283.Apr 23 2015, 4:21 AM
ABataev retitled this revision from to [OPENMP] Coegen for 'firstprivate' clause in 'sections' directive..
ABataev updated this object.
ABataev edited the test plan for this revision. (Show Details)
ABataev added reviewers: rjmccall, hfinkel.
ABataev added subscribers: • fraggamuffin, • ejstotzer, Unknown Object (MLST).
ABataev retitled this revision from [OPENMP] Coegen for 'firstprivate' clause in 'sections' directive. to [OPENMP] Codegen for 'firstprivate' clause in 'sections' directive..Apr 23 2015, 6:53 AM
ABataev updated this object.
rjmccall edited edge metadata.Apr 23 2015, 9:23 AM

LGTM.

Closed by commit rL235691: [OPENMP] Codegen for 'firstprivate' clause in 'sections' directive. (authored by ABataev). · Explain WhyApr 23 2015, 8:40 PM
This revision was automatically updated to reflect the committed changes.

Revision Contents

PathSize
cfe/
trunk/
lib/
CodeGen/
26 lines
test/
OpenMP/
277 lines

Diff 24348

cfe/trunk/lib/CodeGen/CGStmtOpenMP.cpp

Show First 20 Lines • Show All 1,158 Lines • ▼ Show 20 Lines auto &&CodeGen = [&S, CS](CodeGenFunction &CGF) {
auto CaseBB = CGF.createBasicBlock(".omp.sections.case"); auto CaseBB = CGF.createBasicBlock(".omp.sections.case");
CGF.EmitBlock(CaseBB); CGF.EmitBlock(CaseBB);
SwitchStmt->addCase(CGF.Builder.getInt32(CaseNumber), CaseBB); SwitchStmt->addCase(CGF.Builder.getInt32(CaseNumber), CaseBB);
CGF.EmitStmt(*C); CGF.EmitStmt(*C);
CGF.EmitBranch(ExitBB); CGF.EmitBranch(ExitBB);
} }
CGF.EmitBlock(ExitBB, /*IsFinished=*/true); CGF.EmitBlock(ExitBB, /*IsFinished=*/true);
}; };
CodeGenFunction::OMPPrivateScope LoopScope(CGF);
if (CGF.EmitOMPFirstprivateClause(S, LoopScope)) {
// Emit implicit barrier to synchronize threads and avoid data races on
// initialization of firstprivate variables.
CGF.CGM.getOpenMPRuntime().emitBarrierCall(CGF, S.getLocStart(),
OMPD_unknown);
}
(void)LoopScope.Privatize();
// Emit static non-chunked loop. // Emit static non-chunked loop.
CGF.CGM.getOpenMPRuntime().emitForInit( CGF.CGM.getOpenMPRuntime().emitForInit(
CGF, S.getLocStart(), OMPC_SCHEDULE_static, /*IVSize=*/32, CGF, S.getLocStart(), OMPC_SCHEDULE_static, /*IVSize=*/32,
/*IVSigned=*/true, IL.getAddress(), LB.getAddress(), UB.getAddress(), /*IVSigned=*/true, IL.getAddress(), LB.getAddress(), UB.getAddress(),
ST.getAddress()); ST.getAddress());
// UB = min(UB, GlobalUB); // UB = min(UB, GlobalUB);
auto *UBVal = CGF.EmitLoadOfScalar(UB, S.getLocStart()); auto *UBVal = CGF.EmitLoadOfScalar(UB, S.getLocStart());
auto *MinUBGlobalUB = CGF.Builder.CreateSelect( auto *MinUBGlobalUB = CGF.Builder.CreateSelect(
CGF.Builder.CreateICmpSLT(UBVal, GlobalUBVal), UBVal, GlobalUBVal); CGF.Builder.CreateICmpSLT(UBVal, GlobalUBVal), UBVal, GlobalUBVal);
CGF.EmitStoreOfScalar(MinUBGlobalUB, UB); CGF.EmitStoreOfScalar(MinUBGlobalUB, UB);
// IV = LB; // IV = LB;
CGF.EmitStoreOfScalar(CGF.EmitLoadOfScalar(LB, S.getLocStart()), IV); CGF.EmitStoreOfScalar(CGF.EmitLoadOfScalar(LB, S.getLocStart()), IV);
// while (idx <= UB) { BODY; ++idx; } // while (idx <= UB) { BODY; ++idx; }
CGF.EmitOMPInnerLoop(S, /*RequiresCleanup=*/false, &Cond, &Inc, BodyGen, CGF.EmitOMPInnerLoop(S, /*RequiresCleanup=*/false, &Cond, &Inc, BodyGen,
[](CodeGenFunction &) {}); [](CodeGenFunction &) {});
// Tell the runtime we are done. // Tell the runtime we are done.
CGF.CGM.getOpenMPRuntime().emitForStaticFinish(CGF, S.getLocStart()); CGF.CGM.getOpenMPRuntime().emitForStaticFinish(CGF, S.getLocStart());
}; };
CGF.CGM.getOpenMPRuntime().emitInlinedDirective(CGF, CodeGen); CGF.CGM.getOpenMPRuntime().emitInlinedDirective(CGF, CodeGen);
return OMPD_sections; return OMPD_sections;
} }
// If only one section is found - no need to generate loop, emit as a single // If only one section is found - no need to generate loop, emit as a single
// region. // region.
auto &&CodeGen = [Stmt](CodeGenFunction &CGF) { bool HasFirstprivates;
auto &&CodeGen = [Stmt, &S, &HasFirstprivates](CodeGenFunction &CGF) {
CodeGenFunction::OMPPrivateScope SingleScope(CGF);
HasFirstprivates = CGF.EmitOMPFirstprivateClause(S, SingleScope);
(void)SingleScope.Privatize();
CGF.EmitStmt(Stmt); CGF.EmitStmt(Stmt);
CGF.EnsureInsertPoint(); CGF.EnsureInsertPoint();
}; };
CGF.CGM.getOpenMPRuntime().emitSingleRegion(CGF, CodeGen, S.getLocStart(), CGF.CGM.getOpenMPRuntime().emitSingleRegion(CGF, CodeGen, S.getLocStart(),
llvm::None, llvm::None, llvm::None, llvm::None,
llvm::None, llvm::None); llvm::None, llvm::None);
// Emit barrier for firstprivates only if 'sections' directive has 'nowait'
// clause. Otherwise the barrier will be generated by the codegen for the
// directive.
if (HasFirstprivates && S.getSingleClause(OMPC_nowait)) {
// Emit implicit barrier to synchronize threads and avoid data races on
// initialization of firstprivate variables.
CGF.CGM.getOpenMPRuntime().emitBarrierCall(CGF, S.getLocStart(),
OMPD_unknown);
}
return OMPD_single; return OMPD_single;
} }
void CodeGenFunction::EmitOMPSectionsDirective(const OMPSectionsDirective &S) { void CodeGenFunction::EmitOMPSectionsDirective(const OMPSectionsDirective &S) {
LexicalScope Scope(*this, S.getSourceRange()); LexicalScope Scope(*this, S.getSourceRange());
OpenMPDirectiveKind EmittedAs = emitSections(*this, S); OpenMPDirectiveKind EmittedAs = emitSections(*this, S);
// Emit an implicit barrier at the end. // Emit an implicit barrier at the end.
if (!S.getSingleClause(OMPC_nowait)) { if (!S.getSingleClause(OMPC_nowait)) {
▲ Show 20 Lines • Show All 620 Lines • Show Last 20 Lines

cfe/trunk/test/OpenMP/sections_firstprivate_codegen.cpp

PropertyOld ValueNew Value
svn:eol-stylenullnative
svn:keywordsnullAuthor Date Id Rev URL
svn:mime-typenulltext/plain
// RUN: %clang_cc1 -verify -fopenmp=libiomp5 -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck %s
// RUN: %clang_cc1 -fopenmp=libiomp5 -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s
// RUN: %clang_cc1 -fopenmp=libiomp5 -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s
// RUN: %clang_cc1 -verify -fopenmp=libiomp5 -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s
// RUN: %clang_cc1 -verify -fopenmp=libiomp5 -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s
// expected-no-diagnostics
#ifndef HEADER
#define HEADER
struct St {
int a, b;
St() : a(0), b(0) {}
St(const St &st) : a(st.a + st.b), b(0) {}
~St() {}
};
volatile int g = 1212;
template <class T>
struct S {
T f;
S(T a) : f(a + g) {}
S() : f(g) {}
S(const S &s, St t = St()) : f(s.f + t.a) {}
operator T() { return T(); }
~S() {}
};
// CHECK-DAG: [[S_FLOAT_TY:%.+]] = type { float }
// CHECK-DAG: [[S_INT_TY:%.+]] = type { i{{[0-9]+}} }
// CHECK-DAG: [[ST_TY:%.+]] = type { i{{[0-9]+}}, i{{[0-9]+}} }
// CHECK-DAG: [[CAP_TMAIN_TY:%.+]] = type { i{{[0-9]+}}*, [2 x i{{[0-9]+}}]*, [2 x [[S_INT_TY]]]*, [[S_INT_TY]]* }
template <typename T>
T tmain() {
S<T> test;
T t_var = T();
T vec[] = {1, 2};
S<T> s_arr[] = {1, 2};
S<T> var(3);
#pragma omp parallel
#pragma omp sections firstprivate(t_var, vec, s_arr, var)
{
vec[0] = t_var;
#pragma omp section
s_arr[0] = var;
}
return T();
}
// CHECK: [[TEST:@.+]] = global [[S_FLOAT_TY]] zeroinitializer,
S<float> test;
// CHECK-DAG: [[T_VAR:@.+]] = global i{{[0-9]+}} 333,
int t_var = 333;
// CHECK-DAG: [[VEC:@.+]] = global [2 x i{{[0-9]+}}] [i{{[0-9]+}} 1, i{{[0-9]+}} 2],
int vec[] = {1, 2};
// CHECK-DAG: [[S_ARR:@.+]] = global [2 x [[S_FLOAT_TY]]] zeroinitializer,
S<float> s_arr[] = {1, 2};
// CHECK-DAG: [[VAR:@.+]] = global [[S_FLOAT_TY]] zeroinitializer,
S<float> var(3);
// CHECK-DAG: [[IMPLICIT_BARRIER_LOC:@.+]] = private unnamed_addr constant %{{.+}} { i32 0, i32 66, i32 0, i32 0, i8*
// CHECK-DAG: [[SECTIONS_BARRIER_LOC:@.+]] = private unnamed_addr constant %{{.+}} { i32 0, i32 194, i32 0, i32 0, i8*
// CHECK: call {{.*}} [[S_FLOAT_TY_DEF_CONSTR:@.+]]([[S_FLOAT_TY]]* [[TEST]])
// CHECK: ([[S_FLOAT_TY]]*)* [[S_FLOAT_TY_DESTR:@[^ ]+]] {{[^,]+}}, {{.+}}([[S_FLOAT_TY]]* [[TEST]]
int main() {
#ifdef LAMBDA
// LAMBDA: [[G:@.+]] = global i{{[0-9]+}} 1212,
// LAMBDA-LABEL: @main
// LAMBDA: call void [[OUTER_LAMBDA:@.+]](
[&]() {
// LAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]](
// LAMBDA: call void {{.+}} @__kmpc_fork_call({{.+}}, i32 1, {{.+}}* [[OMP_REGION:@.+]] to {{.+}}, i8* %{{.+}})
#pragma omp parallel
#pragma omp sections firstprivate(g)
{
// LAMBDA: define{{.*}} internal{{.*}} void [[OMP_REGION]](i32* %{{.+}}, i32* %{{.+}}, %{{.+}}* [[ARG:%.+]])
// Skip temp vars for loop
// LAMBDA: alloca i{{[0-9]+}},
// LAMBDA: alloca i{{[0-9]+}},
// LAMBDA: alloca i{{[0-9]+}},
// LAMBDA: alloca i{{[0-9]+}},
// LAMBDA: alloca i{{[0-9]+}},
// LAMBDA: [[G_PRIVATE_ADDR:%.+]] = alloca i{{[0-9]+}},
// LAMBDA: [[G_VAL:%.+]] = load volatile i{{[0-9]+}}, i{{[0-9]+}}* [[G]]
// LAMBDA: store i{{[0-9]+}} [[G_VAL]], i{{[0-9]+}}* [[G_PRIVATE_ADDR]]
// LAMBDA: call i32 @__kmpc_cancel_barrier(
g = 1;
// LAMBDA: call void @__kmpc_for_static_init_4(
// LAMBDA: store volatile i{{[0-9]+}} 1, i{{[0-9]+}}* [[G_PRIVATE_ADDR]],
// LAMBDA: [[G_PRIVATE_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG:%.+]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
// LAMBDA: store i{{[0-9]+}}* [[G_PRIVATE_ADDR]], i{{[0-9]+}}** [[G_PRIVATE_ADDR_REF]]
// LAMBDA: call void [[INNER_LAMBDA:@.+]](%{{.+}}* [[ARG]])
// LAMBDA: call void @__kmpc_for_static_fini(
// LAMBDA: call i32 @__kmpc_cancel_barrier(
#pragma omp section
[&]() {
// LAMBDA: define {{.+}} void [[INNER_LAMBDA]](%{{.+}}* [[ARG_PTR:%.+]])
// LAMBDA: store %{{.+}}* [[ARG_PTR]], %{{.+}}** [[ARG_PTR_REF:%.+]],
g = 2;
// LAMBDA: [[ARG_PTR:%.+]] = load %{{.+}}*, %{{.+}}** [[ARG_PTR_REF]]
// LAMBDA: [[G_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
// LAMBDA: [[G_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[G_PTR_REF]]
// LAMBDA: store volatile i{{[0-9]+}} 2, i{{[0-9]+}}* [[G_REF]]
}();
}
}();
return 0;
#elif defined(BLOCKS)
// BLOCKS: [[G:@.+]] = global i{{[0-9]+}} 1212,
// BLOCKS-LABEL: @main
// BLOCKS: call void {{%.+}}(i8
^{
// BLOCKS: define{{.*}} internal{{.*}} void {{.+}}(i8*
// BLOCKS: call void {{.+}} @__kmpc_fork_call({{.+}}, i32 1, {{.+}}* [[OMP_REGION:@.+]] to {{.+}}, i8* %{{.+}})
#pragma omp parallel
#pragma omp sections firstprivate(g)
{
// BLOCKS: define{{.*}} internal{{.*}} void [[OMP_REGION]](i32* %{{.+}}, i32* %{{.+}}, %{{.+}}* [[ARG:%.+]])
// Skip temp vars for loop
// BLOCKS: alloca i{{[0-9]+}},
// BLOCKS: alloca i{{[0-9]+}},
// BLOCKS: alloca i{{[0-9]+}},
// BLOCKS: alloca i{{[0-9]+}},
// BLOCKS: alloca i{{[0-9]+}},
// BLOCKS: [[G_PRIVATE_ADDR:%.+]] = alloca i{{[0-9]+}},
// BLOCKS: [[G_VAL:%.+]] = load volatile i{{[0-9]+}}, i{{[0-9]+}}* [[G]]
// BLOCKS: store i{{[0-9]+}} [[G_VAL]], i{{[0-9]+}}* [[G_PRIVATE_ADDR]]
// BLOCKS: call i32 @__kmpc_cancel_barrier(
g = 1;
// BLOCKS: call void @__kmpc_for_static_init_4(
// BLOCKS: store volatile i{{[0-9]+}} 1, i{{[0-9]+}}* [[G_PRIVATE_ADDR]],
// BLOCKS-NOT: [[G]]{{[[^:word:]]}}
// BLOCKS: i{{[0-9]+}}* [[G_PRIVATE_ADDR]]
// BLOCKS-NOT: [[G]]{{[[^:word:]]}}
// BLOCKS: call void {{%.+}}(i8
// BLOCKS: call void @__kmpc_for_static_fini(
// BLOCKS: call i32 @__kmpc_cancel_barrier(
#pragma omp section
^{
// BLOCKS: define {{.+}} void {{@.+}}(i8*
g = 2;
// BLOCKS-NOT: [[G]]{{[[^:word:]]}}
// BLOCKS: store volatile i{{[0-9]+}} 2, i{{[0-9]+}}*
// BLOCKS-NOT: [[G]]{{[[^:word:]]}}
// BLOCKS: ret
}();
}
}();
return 0;
#else
#pragma omp sections firstprivate(t_var, vec, s_arr, var) nowait
{
{
vec[0] = t_var;
s_arr[0] = var;
}
}
return tmain<int>();
#endif
}
// CHECK: define {{.*}}i{{[0-9]+}} @main()
// CHECK: alloca i{{[0-9]+}},
// CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(
// 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_FLOAT_TY]]],
// CHECK: [[VAR_PRIV:%.+]] = alloca [[S_FLOAT_TY]],
// CHECK: call i32 @__kmpc_single(
// firstprivate t_var(t_var)
// CHECK: [[T_VAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR]],
// CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_PRIV]],
// firstprivate vec(vec)
// CHECK: [[VEC_DEST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8*
// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST]], i8* bitcast ([2 x i{{[0-9]+}}]* [[VEC]] to i8*),
// firstprivate s_arr(s_arr)
// CHECK: [[S_ARR_PRIV_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
// CHECK: [[S_ARR_PRIV_END:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_PRIV_BEGIN]], i{{[0-9]+}} 2
// CHECK: [[IS_EMPTY:%.+]] = icmp eq [[S_FLOAT_TY]]* [[S_ARR_PRIV_BEGIN]], [[S_ARR_PRIV_END]]
// CHECK: br i1 [[IS_EMPTY]], label %[[S_ARR_BODY_DONE:.+]], label %[[S_ARR_BODY:.+]]
// CHECK: [[S_ARR_BODY]]
// CHECK: getelementptr inbounds ([2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0)
// CHECK: call {{.*}} [[ST_TY_DEFAULT_CONSTR:@.+]]([[ST_TY]]* [[ST_TY_TEMP:%.+]])
// CHECK: call {{.*}} [[S_FLOAT_TY_COPY_CONSTR:@.+]]([[S_FLOAT_TY]]* {{.+}}, [[S_FLOAT_TY]]* {{.+}}, [[ST_TY]]* [[ST_TY_TEMP]])
// CHECK: call {{.*}} [[ST_TY_DESTR:@.+]]([[ST_TY]]* [[ST_TY_TEMP]])
// CHECK: br i1 {{.+}}, label %{{.+}}, label %[[S_ARR_BODY]]
// firstprivate var(var)
// CHECK: call {{.*}} [[ST_TY_DEFAULT_CONSTR]]([[ST_TY]]* [[ST_TY_TEMP:%.+]])
// CHECK: call {{.*}} [[S_FLOAT_TY_COPY_CONSTR]]([[S_FLOAT_TY]]* [[VAR_PRIV]], [[S_FLOAT_TY]]* {{.*}} [[VAR]], [[ST_TY]]* [[ST_TY_TEMP]])
// CHECK: call {{.*}} [[ST_TY_DESTR]]([[ST_TY]]* [[ST_TY_TEMP]])
// ~(firstprivate var), ~(firstprivate s_arr)
// CHECK-DAG: call {{.*}} [[S_FLOAT_TY_DESTR]]([[S_FLOAT_TY]]* [[VAR_PRIV]])
// CHECK-DAG: call {{.*}} [[S_FLOAT_TY_DESTR]]([[S_FLOAT_TY]]*
// CHECK: call void @__kmpc_end_single(
// CHECK: call i32 @__kmpc_cancel_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i{{[0-9]+}} [[GTID]])
// CHECK: = call {{.*}}i{{.+}} [[TMAIN_INT:@.+]]()
// CHECK: ret void
// 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]+}} 1, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, [[CAP_TMAIN_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]+}}* [[GTID_ADDR:%.+]], i{{[0-9]+}}* %{{.+}}, [[CAP_TMAIN_TY]]* %{{.+}})
// Skip temp vars for loop
// 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:%.+]],
// firstprivate t_var(t_var)
// CHECK: [[T_VAR_PTR_REF:%.+]] = getelementptr inbounds [[CAP_TMAIN_TY]], [[CAP_TMAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 0
// CHECK: [[T_VAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[T_VAR_PTR_REF]],
// 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]],
// firstprivate vec(vec)
// CHECK: [[VEC_PTR_REF:%.+]] = getelementptr inbounds [[CAP_TMAIN_TY]], [[CAP_TMAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 1
// CHECK: [[VEC_REF:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** [[VEC_PTR_REF:%.+]],
// 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]],
// firstprivate s_arr(s_arr)
// CHECK: [[S_ARR_REF:%.+]] = getelementptr inbounds [[CAP_TMAIN_TY]], [[CAP_TMAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 2
// CHECK: [[S_ARR:%.+]] = load [2 x [[S_INT_TY]]]*, [2 x [[S_INT_TY]]]** [[S_ARR_REF]],
// CHECK: [[S_ARR_PRIV_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
// CHECK: [[S_ARR_PRIV_END:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_PRIV_BEGIN]], i{{[0-9]+}} 2
// CHECK: [[IS_EMPTY:%.+]] = icmp eq [[S_INT_TY]]* [[S_ARR_PRIV_BEGIN]], [[S_ARR_PRIV_END]]
// CHECK: br i1 [[IS_EMPTY]], label %[[S_ARR_BODY_DONE:.+]], label %[[S_ARR_BODY:.+]]
// CHECK: [[S_ARR_BODY]]
// CHECK: call {{.*}} [[ST_TY_DEFAULT_CONSTR:@.+]]([[ST_TY]]* [[ST_TY_TEMP:%.+]])
// CHECK: call {{.*}} [[S_INT_TY_COPY_CONSTR:@.+]]([[S_INT_TY]]* {{.+}}, [[S_INT_TY]]* {{.+}}, [[ST_TY]]* [[ST_TY_TEMP]])
// CHECK: call {{.*}} [[ST_TY_DESTR:@.+]]([[ST_TY]]* [[ST_TY_TEMP]])
// CHECK: br i1 {{.+}}, label %{{.+}}, label %[[S_ARR_BODY]]
// firstprivate var(var)
// CHECK: [[VAR_REF_PTR:%.+]] = getelementptr inbounds [[CAP_TMAIN_TY]], [[CAP_TMAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 3
// CHECK: [[VAR_REF:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[VAR_REF_PTR]],
// CHECK: call {{.*}} [[ST_TY_DEFAULT_CONSTR]]([[ST_TY]]* [[ST_TY_TEMP:%.+]])
// CHECK: call {{.*}} [[S_INT_TY_COPY_CONSTR]]([[S_INT_TY]]* [[VAR_PRIV]], [[S_INT_TY]]* {{.*}} [[VAR_REF]], [[ST_TY]]* [[ST_TY_TEMP]])
// CHECK: call {{.*}} [[ST_TY_DESTR]]([[ST_TY]]* [[ST_TY_TEMP]])
// Synchronization for initialization.
// CHECK: [[GTID_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_ADDR]]
// CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_REF]]
// CHECK: call i32 @__kmpc_cancel_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i{{[0-9]+}} [[GTID]])
// CHECK: call void @__kmpc_for_static_init_4(
// CHECK: call void @__kmpc_for_static_fini(
// ~(firstprivate var), ~(firstprivate s_arr)
// CHECK-DAG: call {{.*}} [[S_INT_TY_DESTR]]([[S_INT_TY]]* [[VAR_PRIV]])
// CHECK-DAG: call {{.*}} [[S_INT_TY_DESTR]]([[S_INT_TY]]*
// CHECK: [[GTID_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_ADDR]]
// CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_REF]]
// CHECK: call i32 @__kmpc_cancel_barrier(%{{.+}}* [[SECTIONS_BARRIER_LOC]], i{{[0-9]+}} [[GTID]])
// CHECK: ret void
#endif