maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/clang/test/OpenMP/teams_distribute_reduction_...

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

231 lines
13 KiB
C++
Raw Normal View History

[OpenMP] Initial implementation of teams distribute code generation https://reviews.llvm.org/D38371 This patch implements codegen for the combined 'teams distribute" OpenMP pragma and adds regression tests for all its clauses. llvm-svn: 314905
2017-10-04 22:12:09 +08:00
// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64
// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64
// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32
// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32
[OPENMP] Support for -fopenmp-simd option with compilation of simd loops only. Added support for -fopenmp-simd option that allows compilation of simd-based constructs without emission of OpenMP runtime calls. llvm-svn: 321560
2017-12-30 02:07:07 +08:00
// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
// SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
[OpenMP] Initial implementation of teams distribute code generation https://reviews.llvm.org/D38371 This patch implements codegen for the combined 'teams distribute" OpenMP pragma and adds regression tests for all its clauses. llvm-svn: 314905
2017-10-04 22:12:09 +08:00
// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64
// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64
[OPENMP] Support for -fopenmp-simd option with compilation of simd loops only. Added support for -fopenmp-simd option that allows compilation of simd-based constructs without emission of OpenMP runtime calls. llvm-svn: 321560
2017-12-30 02:07:07 +08:00
// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
// SIMD-ONLY1-NOT: {{__kmpc|__tgt}}
[OpenMP] Initial implementation of teams distribute code generation https://reviews.llvm.org/D38371 This patch implements codegen for the combined 'teams distribute" OpenMP pragma and adds regression tests for all its clauses. llvm-svn: 314905
2017-10-04 22:12:09 +08:00
// expected-no-diagnostics
#ifndef HEADER
#define HEADER
template <typename T>
T tmain() {
T t_var = T();
T vec[] = {1, 2};
#pragma omp target
#pragma omp teams distribute reduction(+: t_var)
for (int i = 0; i < 2; ++i) {
t_var += (T) i;
}
return T();
}
int main() {
static int sivar;
#ifdef LAMBDA
// LAMBDA: [[RED_VAR:@.+]] = common global [8 x {{.+}}] zeroinitializer
// LAMBDA-LABEL: @main
// LAMBDA: call void [[OUTER_LAMBDA:@.+]](
[&]() {
// LAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]](
[OPENMP] Emit `__tgt_target_teams` for all teams directives. Previously we emitted `__tgt_target_teams` only for standalone teams directives. This patch allows emit this function for all teams-based directives. llvm-svn: 319585
2017-12-02 05:31:08 +08:00
// LAMBDA: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 1, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* {{.+}}@{{[^,]+}}, i32 0, i32 0), i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), i32 0, i32 0)
[OpenMP] Initial implementation of teams distribute code generation https://reviews.llvm.org/D38371 This patch implements codegen for the combined 'teams distribute" OpenMP pragma and adds regression tests for all its clauses. llvm-svn: 314905
2017-10-04 22:12:09 +08:00
// LAMBDA: call void @[[LOFFL1:.+]](
// LAMBDA: ret
#pragma omp target
#pragma omp teams distribute reduction(+: sivar)
for (int i = 0; i < 2; ++i) {
// LAMBDA: define{{.*}} internal{{.*}} void @[[LOFFL1]](i{{64|32}} [[SIVAR_ARG:%.+]])
// LAMBDA: [[SIVAR_ADDR:%.+]] = alloca i{{.+}},
// LAMBDA: store{{.+}} [[SIVAR_ARG]], {{.+}} [[SIVAR_ADDR]],
// LAMBDA: [[SIVAR_CONV:%.+]] = bitcast{{.+}} [[SIVAR_ADDR]] to
// LAMBDA: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[LOUTL1:.+]] to {{.+}}, {{.+}} [[SIVAR_CONV]])
// LAMBDA: ret void
// LAMBDA: define internal void @[[LOUTL1]]({{.+}}, {{.+}}, {{.+}} [[SIVAR_ARG:%.+]])
// Skip global and bound tid vars
// LAMBDA: {{.+}} = alloca i32*,
// LAMBDA: {{.+}} = alloca i32*,
// LAMBDA: [[SIVAR_ADDR:%.+]] = alloca i{{.+}}*,
// LAMBDA: [[SIVAR_PRIV:%.+]] = alloca i{{.+}},
// LAMBDA: [[RED_LIST:%.+]] = alloca [1 x {{.+}}],
// LAMBDA: store{{.+}} [[SIVAR_ARG]], {{.+}} [[SIVAR_ADDR]],
// LAMBDA: [[SIVAR_REF:%.+]] = load{{.+}}, {{.+}} [[SIVAR_ADDR]]
// LAMBDA: store{{.+}} 0, {{.+}} [[SIVAR_PRIV]],
// LAMBDA: call void @__kmpc_for_static_init_4(
// LAMBDA: store{{.+}}, {{.+}} [[SIVAR_PRIV]],
// LAMBDA: call void [[INNER_LAMBDA:@.+]](
// LAMBDA: call void @__kmpc_for_static_fini(
// LAMBDA: [[RED_LIST_GEP:%.+]] = getelementptr{{.+}} [[RED_LIST]],
// LAMBDA: [[SIVAR_PRIV_CAST:%.+]] = bitcast{{.+}} [[SIVAR_PRIV]] to
// LAMBDA: store{{.+}} [[SIVAR_PRIV_CAST]], {{.+}} [[RED_LIST_GEP]],
// LAMBDA: [[RED_LIST_BCAST:%.+]] = bitcast{{.+}} [[RED_LIST]] to
// LAMBDA: [[K_RED_RET:%.+]] = call{{.+}} @__kmpc_reduce({{.+}}, {{.+}}, {{.+}}, {{.+}}, {{.+}} [[RED_LIST_BCAST]], {{.+}} [[RED_FUN:@.+]], {{.+}} [[RED_VAR]])
// LAMBDA: switch{{.+}} [[K_RED_RET]], label{{.+}} [
// LAMBDA: {{.+}}, label %[[CASE1:.+]]
// LAMBDA: {{.+}}, label %[[CASE2:.+]]
// LAMBDA: ]
// LAMBDA: [[CASE1]]:
// LAMBDA-DAG: [[SIVAR_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_REF]],
// LAMBDA-DAG: [[SIVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_PRIV]],
// LAMBDA-DAG: [[SIVAR_INC:%.+]] = add{{.+}} [[SIVAR_VAL]], [[SIVAR_PRIV_VAL]]
// LAMBDA: store{{.+}} [[SIVAR_INC]], {{.+}} [[SIVAR_REF]],
// LAMBDA: call void @__kmpc_end_reduce({{.+}}, {{.+}}, {{.+}} [[RED_VAR]])
// LAMBDA: br
// LAMBDA: [[CASE2]]:
// LAMBDA-DAG: [[SIVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_PRIV]],
// LAMBDA-DAG: [[ATOMIC_RES:%.+]] = atomicrmw add{{.+}} [[SIVAR_REF]], {{.+}} [[SIVAR_PRIV_VAL]]
// LAMBDA: call void @__kmpc_end_reduce({{.+}}, {{.+}}, {{.+}} [[RED_VAR]])
// LAMBDA: br
sivar += i;
[&]() {
// LAMBDA: define {{.+}} void [[INNER_LAMBDA]](%{{.+}}* [[ARG_PTR:%.+]])
// LAMBDA: store %{{.+}}* [[ARG_PTR]], %{{.+}}** [[ARG_PTR_REF:%.+]],
sivar += 4;
// LAMBDA: [[ARG_PTR:%.+]] = load %{{.+}}*, %{{.+}}** [[ARG_PTR_REF]]
// LAMBDA: [[SIVAR_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
// LAMBDA: [[SIVAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[SIVAR_PTR_REF]]
// LAMBDA: [[SIVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[SIVAR_REF]]
// LAMBDA: [[SIVAR_INC:%.+]] = add{{.+}} [[SIVAR_VAL]], 4
// LAMBDA: store i{{[0-9]+}} [[SIVAR_INC]], i{{[0-9]+}}* [[SIVAR_REF]]
}();
}
}();
return 0;
#else
#pragma omp target
#pragma omp teams distribute reduction(+: sivar)
for (int i = 0; i < 2; ++i) {
sivar += i;
}
return tmain<int>();
#endif
}
// CHECK: [[RED_VAR:@.+]] = common global [8 x {{.+}}] zeroinitializer
// CHECK: define {{.*}}i{{[0-9]+}} @main()
[OPENMP] Emit `__tgt_target_teams` for all teams directives. Previously we emitted `__tgt_target_teams` only for standalone teams directives. This patch allows emit this function for all teams-based directives. llvm-svn: 319585
2017-12-02 05:31:08 +08:00
// CHECK: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 1, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* {{.+}}@{{[^,]+}}, i32 0, i32 0), i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), i32 0, i32 0)
[OpenMP] Initial implementation of teams distribute code generation https://reviews.llvm.org/D38371 This patch implements codegen for the combined 'teams distribute" OpenMP pragma and adds regression tests for all its clauses. llvm-svn: 314905
2017-10-04 22:12:09 +08:00
// CHECK: call void @[[OFFL1:.+]](i{{64|32}} %{{.+}})
// CHECK: {{%.+}} = call{{.*}} i32 @[[TMAIN_INT:.+]]()
// CHECK: ret
// CHECK: define{{.*}} void @[[OFFL1]](i{{64|32}} [[SIVAR_ARG:%.+]])
// CHECK: [[SIVAR_ADDR:%.+]] = alloca i{{.+}},
// CHECK: store{{.+}} [[SIVAR_ARG]], {{.+}} [[SIVAR_ADDR]],
// CHECK-64: [[SIVAR_CONV:%.+]] = bitcast{{.+}} [[SIVAR_ADDR]] to
// CHECK-64: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTL1:.+]] to {{.+}}, {{.+}} [[SIVAR_CONV]])
// CHECK-32: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTL1:.+]] to {{.+}}, {{.+}} [[SIVAR_ADDR]])
// CHECK: ret void
// CHECK: define internal void @[[OUTL1]]({{.+}}, {{.+}}, {{.+}} [[SIVAR_ARG:%.+]])
// Skip global and bound tid vars
// CHECK: {{.+}} = alloca i32*,
// CHECK: {{.+}} = alloca i32*,
// CHECK: [[SIVAR_ADDR:%.+]] = alloca i{{.+}}*,
// CHECK: [[SIVAR_PRIV:%.+]] = alloca i{{.+}},
// CHECK: [[RED_LIST:%.+]] = alloca [1 x {{.+}}],
// CHECK: store{{.+}} [[SIVAR_ARG]], {{.+}} [[SIVAR_ADDR]],
// CHECK: [[SIVAR_REF:%.+]] = load{{.+}}, {{.+}} [[SIVAR_ADDR]]
// CHECK: store{{.+}} 0, {{.+}} [[SIVAR_PRIV]],
// CHECK: call void @__kmpc_for_static_init_4(
// CHECK: store{{.+}}, {{.+}} [[SIVAR_PRIV]],
// CHECK: call void @__kmpc_for_static_fini(
// CHECK: [[RED_LIST_GEP:%.+]] = getelementptr{{.+}} [[RED_LIST]],
// CHECK: [[SIVAR_PRIV_CAST:%.+]] = bitcast{{.+}} [[SIVAR_PRIV]] to
// CHECK: store{{.+}} [[SIVAR_PRIV_CAST]], {{.+}} [[RED_LIST_GEP]],
// CHECK: [[RED_LIST_BCAST:%.+]] = bitcast{{.+}} [[RED_LIST]] to
// CHECK: [[K_RED_RET:%.+]] = call{{.+}} @__kmpc_reduce({{.+}}, {{.+}}, {{.+}}, {{.+}}, {{.+}} [[RED_LIST_BCAST]], {{.+}} [[RED_FUN:@.+]], {{.+}} [[RED_VAR]])
// CHECK: switch{{.+}} [[K_RED_RET]], label{{.+}} [
// CHECK: {{.+}}, label %[[CASE1:.+]]
// CHECK: {{.+}}, label %[[CASE2:.+]]
// CHECK: ]
// CHECK: [[CASE1]]:
// CHECK-DAG: [[SIVAR_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_REF]],
// CHECK-DAG: [[SIVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_PRIV]],
// CHECK-DAG: [[SIVAR_INC:%.+]] = add{{.+}} [[SIVAR_VAL]], [[SIVAR_PRIV_VAL]]
// CHECK: store{{.+}} [[SIVAR_INC]], {{.+}} [[SIVAR_REF]],
// CHECK: call void @__kmpc_end_reduce({{.+}}, {{.+}}, {{.+}} [[RED_VAR]])
// CHECK: br
// CHECK: [[CASE2]]:
// CHECK-DAG: [[SIVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_PRIV]],
// CHECK-DAG: [[ATOMIC_RES:%.+]] = atomicrmw add{{.+}} [[SIVAR_REF]], {{.+}} [[SIVAR_PRIV_VAL]]
// CHECK: call void @__kmpc_end_reduce({{.+}}, {{.+}}, {{.+}} [[RED_VAR]])
// CHECK: br
// CHECK: define{{.*}} i{{[0-9]+}} @[[TMAIN_INT]]()
[OPENMP] Emit `__tgt_target_teams` for all teams directives. Previously we emitted `__tgt_target_teams` only for standalone teams directives. This patch allows emit this function for all teams-based directives. llvm-svn: 319585
2017-12-02 05:31:08 +08:00
// CHECK: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 1,
[OpenMP] Initial implementation of teams distribute code generation https://reviews.llvm.org/D38371 This patch implements codegen for the combined 'teams distribute" OpenMP pragma and adds regression tests for all its clauses. llvm-svn: 314905
2017-10-04 22:12:09 +08:00
// CHECK: call void @[[TOFFL1:.+]]({{.+}})
// CHECK: ret
// CHECK: define{{.*}} void @[[TOFFL1]](i{{64|32}} [[TVAR_ARG:%.+]])
// CHECK: [[TVAR_ADDR:%.+]] = alloca i{{.+}},
// CHECK: store{{.+}} [[TVAR_ARG]], {{.+}} [[TVAR_ADDR]],
// CHECK-64: [[TVAR_CONV:%.+]] = bitcast{{.+}} [[TVAR_ADDR]] to
// CHECK-64: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[TOUTL1:.+]] to {{.+}}, {{.+}} [[TVAR_CONV]])
// CHECK-32: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[TOUTL1:.+]] to {{.+}}, {{.+}} [[TVAR_ADDR]])
// CHECK: ret void
// CHECK: define internal void @[[TOUTL1]]({{.+}}, {{.+}}, {{.+}} [[TVAR_ARG:%.+]])
// Skip global and bound tid vars
// CHECK: {{.+}} = alloca i32*,
// CHECK: {{.+}} = alloca i32*,
// CHECK: [[TVAR_ADDR:%.+]] = alloca i{{.+}}*,
// CHECK: [[TVAR_PRIV:%.+]] = alloca i{{.+}},
// CHECK: [[RED_LIST:%.+]] = alloca [1 x {{.+}}],
// CHECK: store{{.+}} [[TVAR_ARG]], {{.+}} [[TVAR_ADDR]],
// CHECK: [[TVAR_REF:%.+]] = load{{.+}}, {{.+}} [[TVAR_ADDR]]
// CHECK: store{{.+}} 0, {{.+}} [[TVAR_PRIV]],
// CHECK: call void @__kmpc_for_static_init_4(
// CHECK: store{{.+}}, {{.+}} [[TVAR_PRIV]],
// CHECK: call void @__kmpc_for_static_fini(
// CHECK: [[RED_LIST_GEP:%.+]] = getelementptr{{.+}} [[RED_LIST]],
// CHECK: [[TVAR_PRIV_CAST:%.+]] = bitcast{{.+}} [[TVAR_PRIV]] to
// CHECK: store{{.+}} [[TVAR_PRIV_CAST]], {{.+}} [[RED_LIST_GEP]],
// CHECK: [[RED_LIST_BCAST:%.+]] = bitcast{{.+}} [[RED_LIST]] to
// CHECK: [[K_RED_RET:%.+]] = call{{.+}} @__kmpc_reduce({{.+}}, {{.+}}, {{.+}}, {{.+}}, {{.+}} [[RED_LIST_BCAST]], {{.+}} [[RED_FUN:@.+]], {{.+}} [[RED_VAR]])
// CHECK: switch{{.+}} [[K_RED_RET]], label{{.+}} [
// CHECK: {{.+}}, label %[[CASE1:.+]]
// CHECK: {{.+}}, label %[[CASE2:.+]]
// CHECK: ]
// CHECK: [[CASE1]]:
// CHECK-DAG: [[TVAR_VAL:%.+]] = load{{.+}}, {{.+}} [[TVAR_REF]],
// CHECK-DAG: [[TVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[TVAR_PRIV]],
// CHECK-DAG: [[TVAR_INC:%.+]] = add{{.+}} [[TVAR_VAL]], [[TVAR_PRIV_VAL]]
// CHECK: store{{.+}} [[TVAR_INC]], {{.+}} [[TVAR_REF]],
// CHECK: call void @__kmpc_end_reduce({{.+}}, {{.+}}, {{.+}} [[RED_VAR]])
// CHECK: br
// CHECK: [[CASE2]]:
// CHECK-DAG: [[TVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[TVAR_PRIV]],
// CHECK-DAG: [[ATOMIC_RES:%.+]] = atomicrmw add{{.+}} [[TVAR_REF]], {{.+}} [[TVAR_PRIV_VAL]]
// CHECK: call void @__kmpc_end_reduce({{.+}}, {{.+}}, {{.+}} [[RED_VAR]])
// CHECK: br
#endif