diff --git a/clang/test/CodeGen/RISCV/Float16.c b/clang/test/CodeGen/RISCV/Float16.c
new file mode 100644
--- /dev/null
+++ b/clang/test/CodeGen/RISCV/Float16.c
@@ -0,0 +1,375 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
+// RUN: %clang_cc1 -triple riscv32 -O0 -target-abi ilp32 -emit-llvm %s \
+// RUN:   -o - | FileCheck --check-prefix=CHECK-NOZFH-ILP32 %s
+// RUN: %clang_cc1 -triple riscv32 -target-feature +experimental-zfh -O0 -target-abi ilp32 -emit-llvm %s \
+// RUN:   -o - | FileCheck --check-prefix=CHECK-ZFH-ILP32 %s
+// RUN: %clang_cc1 -triple riscv32 -O0 -target-abi ilp32f -emit-llvm %s \
+// RUN:   -o - | FileCheck --check-prefix=CHECK-NOZFH-ILP32F %s
+// RUN: %clang_cc1 -triple riscv32 -target-feature +experimental-zfh -O0 -target-abi ilp32f -emit-llvm %s \
+// RUN:   -o - | FileCheck --check-prefix=CHECK-ZFH-ILP32F %s
+
+#include <stdint.h>
+
+_Float16 varg(int num, ...);
+
+// CHECK-NOZFH-ILP32-LABEL: @foo(
+// CHECK-NOZFH-ILP32-NEXT:  entry:
+// CHECK-NOZFH-ILP32-NEXT:    [[A_ADDR:%.*]] = alloca half, align 2
+// CHECK-NOZFH-ILP32-NEXT:    [[B_ADDR:%.*]] = alloca half, align 2
+// CHECK-NOZFH-ILP32-NEXT:    store half [[A:%.*]], half* [[A_ADDR]], align 2
+// CHECK-NOZFH-ILP32-NEXT:    store half [[B:%.*]], half* [[B_ADDR]], align 2
+// CHECK-NOZFH-ILP32-NEXT:    [[TMP0:%.*]] = load half, half* [[A_ADDR]], align 2
+// CHECK-NOZFH-ILP32-NEXT:    [[TMP1:%.*]] = load half, half* [[B_ADDR]], align 2
+// CHECK-NOZFH-ILP32-NEXT:    [[ADD:%.*]] = fadd half [[TMP0]], [[TMP1]]
+// CHECK-NOZFH-ILP32-NEXT:    ret half [[ADD]]
+//
+// CHECK-ZFH-ILP32-LABEL: @foo(
+// CHECK-ZFH-ILP32-NEXT:  entry:
+// CHECK-ZFH-ILP32-NEXT:    [[A_ADDR:%.*]] = alloca half, align 2
+// CHECK-ZFH-ILP32-NEXT:    [[B_ADDR:%.*]] = alloca half, align 2
+// CHECK-ZFH-ILP32-NEXT:    store half [[A:%.*]], half* [[A_ADDR]], align 2
+// CHECK-ZFH-ILP32-NEXT:    store half [[B:%.*]], half* [[B_ADDR]], align 2
+// CHECK-ZFH-ILP32-NEXT:    [[TMP0:%.*]] = load half, half* [[A_ADDR]], align 2
+// CHECK-ZFH-ILP32-NEXT:    [[TMP1:%.*]] = load half, half* [[B_ADDR]], align 2
+// CHECK-ZFH-ILP32-NEXT:    [[ADD:%.*]] = fadd half [[TMP0]], [[TMP1]]
+// CHECK-ZFH-ILP32-NEXT:    ret half [[ADD]]
+//
+// CHECK-NOZFH-ILP32F-LABEL: @foo(
+// CHECK-NOZFH-ILP32F-NEXT:  entry:
+// CHECK-NOZFH-ILP32F-NEXT:    [[A_ADDR:%.*]] = alloca half, align 2
+// CHECK-NOZFH-ILP32F-NEXT:    [[B_ADDR:%.*]] = alloca half, align 2
+// CHECK-NOZFH-ILP32F-NEXT:    store half [[A:%.*]], half* [[A_ADDR]], align 2
+// CHECK-NOZFH-ILP32F-NEXT:    store half [[B:%.*]], half* [[B_ADDR]], align 2
+// CHECK-NOZFH-ILP32F-NEXT:    [[TMP0:%.*]] = load half, half* [[A_ADDR]], align 2
+// CHECK-NOZFH-ILP32F-NEXT:    [[TMP1:%.*]] = load half, half* [[B_ADDR]], align 2
+// CHECK-NOZFH-ILP32F-NEXT:    [[ADD:%.*]] = fadd half [[TMP0]], [[TMP1]]
+// CHECK-NOZFH-ILP32F-NEXT:    ret half [[ADD]]
+//
+// CHECK-ZFH-ILP32F-LABEL: @foo(
+// CHECK-ZFH-ILP32F-NEXT:  entry:
+// CHECK-ZFH-ILP32F-NEXT:    [[A_ADDR:%.*]] = alloca half, align 2
+// CHECK-ZFH-ILP32F-NEXT:    [[B_ADDR:%.*]] = alloca half, align 2
+// CHECK-ZFH-ILP32F-NEXT:    store half [[A:%.*]], half* [[A_ADDR]], align 2
+// CHECK-ZFH-ILP32F-NEXT:    store half [[B:%.*]], half* [[B_ADDR]], align 2
+// CHECK-ZFH-ILP32F-NEXT:    [[TMP0:%.*]] = load half, half* [[A_ADDR]], align 2
+// CHECK-ZFH-ILP32F-NEXT:    [[TMP1:%.*]] = load half, half* [[B_ADDR]], align 2
+// CHECK-ZFH-ILP32F-NEXT:    [[ADD:%.*]] = fadd half [[TMP0]], [[TMP1]]
+// CHECK-ZFH-ILP32F-NEXT:    ret half [[ADD]]
+//
+_Float16 foo(_Float16 a, _Float16 b) {
+  return a + b;
+}
+
+// CHECK-NOZFH-ILP32-LABEL: @foo1(
+// CHECK-NOZFH-ILP32-NEXT:  entry:
+// CHECK-NOZFH-ILP32-NEXT:    [[A_ADDR:%.*]] = alloca half, align 2
+// CHECK-NOZFH-ILP32-NEXT:    store half [[A:%.*]], half* [[A_ADDR]], align 2
+// CHECK-NOZFH-ILP32-NEXT:    [[TMP0:%.*]] = load half, half* [[A_ADDR]], align 2
+// CHECK-NOZFH-ILP32-NEXT:    [[CALL:%.*]] = call half (i32, ...) @varg(i32 1, half [[TMP0]])
+// CHECK-NOZFH-ILP32-NEXT:    ret half [[CALL]]
+//
+// CHECK-ZFH-ILP32-LABEL: @foo1(
+// CHECK-ZFH-ILP32-NEXT:  entry:
+// CHECK-ZFH-ILP32-NEXT:    [[A_ADDR:%.*]] = alloca half, align 2
+// CHECK-ZFH-ILP32-NEXT:    store half [[A:%.*]], half* [[A_ADDR]], align 2
+// CHECK-ZFH-ILP32-NEXT:    [[TMP0:%.*]] = load half, half* [[A_ADDR]], align 2
+// CHECK-ZFH-ILP32-NEXT:    [[CALL:%.*]] = call half (i32, ...) @varg(i32 1, half [[TMP0]])
+// CHECK-ZFH-ILP32-NEXT:    ret half [[CALL]]
+//
+// CHECK-NOZFH-ILP32F-LABEL: @foo1(
+// CHECK-NOZFH-ILP32F-NEXT:  entry:
+// CHECK-NOZFH-ILP32F-NEXT:    [[A_ADDR:%.*]] = alloca half, align 2
+// CHECK-NOZFH-ILP32F-NEXT:    store half [[A:%.*]], half* [[A_ADDR]], align 2
+// CHECK-NOZFH-ILP32F-NEXT:    [[TMP0:%.*]] = load half, half* [[A_ADDR]], align 2
+// CHECK-NOZFH-ILP32F-NEXT:    [[CALL:%.*]] = call half (i32, ...) @varg(i32 1, half [[TMP0]])
+// CHECK-NOZFH-ILP32F-NEXT:    ret half [[CALL]]
+//
+// CHECK-ZFH-ILP32F-LABEL: @foo1(
+// CHECK-ZFH-ILP32F-NEXT:  entry:
+// CHECK-ZFH-ILP32F-NEXT:    [[A_ADDR:%.*]] = alloca half, align 2
+// CHECK-ZFH-ILP32F-NEXT:    store half [[A:%.*]], half* [[A_ADDR]], align 2
+// CHECK-ZFH-ILP32F-NEXT:    [[TMP0:%.*]] = load half, half* [[A_ADDR]], align 2
+// CHECK-ZFH-ILP32F-NEXT:    [[CALL:%.*]] = call half (i32, ...) @varg(i32 1, half [[TMP0]])
+// CHECK-ZFH-ILP32F-NEXT:    ret half [[CALL]]
+//
+_Float16 foo1(_Float16 a) {
+  return varg(1, a);
+}
+
+struct f16_agg {
+  _Float16 a, b;
+};
+
+// CHECK-NOZFH-ILP32-LABEL: @test_f16_agg_1(
+// CHECK-NOZFH-ILP32-NEXT:  entry:
+// CHECK-NOZFH-ILP32-NEXT:    [[X:%.*]] = alloca [[STRUCT_F16_AGG:%.*]], align 2
+// CHECK-NOZFH-ILP32-NEXT:    [[TMP0:%.*]] = bitcast %struct.f16_agg* [[X]] to i32*
+// CHECK-NOZFH-ILP32-NEXT:    store i32 [[X_COERCE:%.*]], i32* [[TMP0]], align 2
+// CHECK-NOZFH-ILP32-NEXT:    [[B:%.*]] = getelementptr inbounds [[STRUCT_F16_AGG]], %struct.f16_agg* [[X]], i32 0, i32 1
+// CHECK-NOZFH-ILP32-NEXT:    [[TMP1:%.*]] = load half, half* [[B]], align 2
+// CHECK-NOZFH-ILP32-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_F16_AGG]], %struct.f16_agg* [[X]], i32 0, i32 0
+// CHECK-NOZFH-ILP32-NEXT:    [[TMP2:%.*]] = load half, half* [[A]], align 2
+// CHECK-NOZFH-ILP32-NEXT:    [[ADD:%.*]] = fadd half [[TMP2]], [[TMP1]]
+// CHECK-NOZFH-ILP32-NEXT:    store half [[ADD]], half* [[A]], align 2
+// CHECK-NOZFH-ILP32-NEXT:    ret void
+//
+// CHECK-ZFH-ILP32-LABEL: @test_f16_agg_1(
+// CHECK-ZFH-ILP32-NEXT:  entry:
+// CHECK-ZFH-ILP32-NEXT:    [[X:%.*]] = alloca [[STRUCT_F16_AGG:%.*]], align 2
+// CHECK-ZFH-ILP32-NEXT:    [[TMP0:%.*]] = bitcast %struct.f16_agg* [[X]] to i32*
+// CHECK-ZFH-ILP32-NEXT:    store i32 [[X_COERCE:%.*]], i32* [[TMP0]], align 2
+// CHECK-ZFH-ILP32-NEXT:    [[B:%.*]] = getelementptr inbounds [[STRUCT_F16_AGG]], %struct.f16_agg* [[X]], i32 0, i32 1
+// CHECK-ZFH-ILP32-NEXT:    [[TMP1:%.*]] = load half, half* [[B]], align 2
+// CHECK-ZFH-ILP32-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_F16_AGG]], %struct.f16_agg* [[X]], i32 0, i32 0
+// CHECK-ZFH-ILP32-NEXT:    [[TMP2:%.*]] = load half, half* [[A]], align 2
+// CHECK-ZFH-ILP32-NEXT:    [[ADD:%.*]] = fadd half [[TMP2]], [[TMP1]]
+// CHECK-ZFH-ILP32-NEXT:    store half [[ADD]], half* [[A]], align 2
+// CHECK-ZFH-ILP32-NEXT:    ret void
+//
+// CHECK-NOZFH-ILP32F-LABEL: @test_f16_agg_1(
+// CHECK-NOZFH-ILP32F-NEXT:  entry:
+// CHECK-NOZFH-ILP32F-NEXT:    [[X:%.*]] = alloca [[STRUCT_F16_AGG:%.*]], align 2
+// CHECK-NOZFH-ILP32F-NEXT:    [[TMP0:%.*]] = bitcast %struct.f16_agg* [[X]] to i32*
+// CHECK-NOZFH-ILP32F-NEXT:    store i32 [[X_COERCE:%.*]], i32* [[TMP0]], align 2
+// CHECK-NOZFH-ILP32F-NEXT:    [[B:%.*]] = getelementptr inbounds [[STRUCT_F16_AGG]], %struct.f16_agg* [[X]], i32 0, i32 1
+// CHECK-NOZFH-ILP32F-NEXT:    [[TMP1:%.*]] = load half, half* [[B]], align 2
+// CHECK-NOZFH-ILP32F-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_F16_AGG]], %struct.f16_agg* [[X]], i32 0, i32 0
+// CHECK-NOZFH-ILP32F-NEXT:    [[TMP2:%.*]] = load half, half* [[A]], align 2
+// CHECK-NOZFH-ILP32F-NEXT:    [[ADD:%.*]] = fadd half [[TMP2]], [[TMP1]]
+// CHECK-NOZFH-ILP32F-NEXT:    store half [[ADD]], half* [[A]], align 2
+// CHECK-NOZFH-ILP32F-NEXT:    ret void
+//
+// CHECK-ZFH-ILP32F-LABEL: @test_f16_agg_1(
+// CHECK-ZFH-ILP32F-NEXT:  entry:
+// CHECK-ZFH-ILP32F-NEXT:    [[X:%.*]] = alloca [[STRUCT_F16_AGG:%.*]], align 2
+// CHECK-ZFH-ILP32F-NEXT:    [[TMP0:%.*]] = bitcast %struct.f16_agg* [[X]] to i32*
+// CHECK-ZFH-ILP32F-NEXT:    store i32 [[X_COERCE:%.*]], i32* [[TMP0]], align 2
+// CHECK-ZFH-ILP32F-NEXT:    [[B:%.*]] = getelementptr inbounds [[STRUCT_F16_AGG]], %struct.f16_agg* [[X]], i32 0, i32 1
+// CHECK-ZFH-ILP32F-NEXT:    [[TMP1:%.*]] = load half, half* [[B]], align 2
+// CHECK-ZFH-ILP32F-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_F16_AGG]], %struct.f16_agg* [[X]], i32 0, i32 0
+// CHECK-ZFH-ILP32F-NEXT:    [[TMP2:%.*]] = load half, half* [[A]], align 2
+// CHECK-ZFH-ILP32F-NEXT:    [[ADD:%.*]] = fadd half [[TMP2]], [[TMP1]]
+// CHECK-ZFH-ILP32F-NEXT:    store half [[ADD]], half* [[A]], align 2
+// CHECK-ZFH-ILP32F-NEXT:    ret void
+//
+void test_f16_agg_1(struct f16_agg x) {
+  x.a += x.b;
+}
+
+struct f16_float_agg {
+  _Float16 a;
+  float b;
+};
+
+// CHECK-NOZFH-ILP32-LABEL: @test_f16_agg_2(
+// CHECK-NOZFH-ILP32-NEXT:  entry:
+// CHECK-NOZFH-ILP32-NEXT:    [[A:%.*]] = alloca [[STRUCT_F16_FLOAT_AGG:%.*]], align 4
+// CHECK-NOZFH-ILP32-NEXT:    [[TMP0:%.*]] = bitcast %struct.f16_float_agg* [[A]] to [2 x i32]*
+// CHECK-NOZFH-ILP32-NEXT:    store [2 x i32] [[A_COERCE:%.*]], [2 x i32]* [[TMP0]], align 4
+// CHECK-NOZFH-ILP32-NEXT:    ret void
+//
+// CHECK-ZFH-ILP32-LABEL: @test_f16_agg_2(
+// CHECK-ZFH-ILP32-NEXT:  entry:
+// CHECK-ZFH-ILP32-NEXT:    [[A:%.*]] = alloca [[STRUCT_F16_FLOAT_AGG:%.*]], align 4
+// CHECK-ZFH-ILP32-NEXT:    [[TMP0:%.*]] = bitcast %struct.f16_float_agg* [[A]] to [2 x i32]*
+// CHECK-ZFH-ILP32-NEXT:    store [2 x i32] [[A_COERCE:%.*]], [2 x i32]* [[TMP0]], align 4
+// CHECK-ZFH-ILP32-NEXT:    ret void
+//
+// CHECK-NOZFH-ILP32F-LABEL: @test_f16_agg_2(
+// CHECK-NOZFH-ILP32F-NEXT:  entry:
+// CHECK-NOZFH-ILP32F-NEXT:    [[A:%.*]] = alloca [[STRUCT_F16_FLOAT_AGG:%.*]], align 4
+// CHECK-NOZFH-ILP32F-NEXT:    [[TMP0:%.*]] = bitcast %struct.f16_float_agg* [[A]] to [2 x i32]*
+// CHECK-NOZFH-ILP32F-NEXT:    store [2 x i32] [[A_COERCE:%.*]], [2 x i32]* [[TMP0]], align 4
+// CHECK-NOZFH-ILP32F-NEXT:    ret void
+//
+// CHECK-ZFH-ILP32F-LABEL: @test_f16_agg_2(
+// CHECK-ZFH-ILP32F-NEXT:  entry:
+// CHECK-ZFH-ILP32F-NEXT:    [[A:%.*]] = alloca [[STRUCT_F16_FLOAT_AGG:%.*]], align 4
+// CHECK-ZFH-ILP32F-NEXT:    [[TMP0:%.*]] = bitcast %struct.f16_float_agg* [[A]] to [2 x i32]*
+// CHECK-ZFH-ILP32F-NEXT:    store [2 x i32] [[A_COERCE:%.*]], [2 x i32]* [[TMP0]], align 4
+// CHECK-ZFH-ILP32F-NEXT:    ret void
+//
+void test_f16_agg_2(struct f16_float_agg a) {}
+
+// CHECK-NOZFH-ILP32-LABEL: @test_f16_agg_3(
+// CHECK-NOZFH-ILP32-NEXT:  entry:
+// CHECK-NOZFH-ILP32-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_F16_FLOAT_AGG:%.*]], align 4
+// CHECK-NOZFH-ILP32-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_F16_FLOAT_AGG]], %struct.f16_float_agg* [[RETVAL]], i32 0, i32 0
+// CHECK-NOZFH-ILP32-NEXT:    store half 0xH3C00, half* [[A]], align 4
+// CHECK-NOZFH-ILP32-NEXT:    [[B:%.*]] = getelementptr inbounds [[STRUCT_F16_FLOAT_AGG]], %struct.f16_float_agg* [[RETVAL]], i32 0, i32 1
+// CHECK-NOZFH-ILP32-NEXT:    store float 2.000000e+00, float* [[B]], align 4
+// CHECK-NOZFH-ILP32-NEXT:    [[TMP0:%.*]] = bitcast %struct.f16_float_agg* [[RETVAL]] to [2 x i32]*
+// CHECK-NOZFH-ILP32-NEXT:    [[TMP1:%.*]] = load [2 x i32], [2 x i32]* [[TMP0]], align 4
+// CHECK-NOZFH-ILP32-NEXT:    ret [2 x i32] [[TMP1]]
+//
+// CHECK-ZFH-ILP32-LABEL: @test_f16_agg_3(
+// CHECK-ZFH-ILP32-NEXT:  entry:
+// CHECK-ZFH-ILP32-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_F16_FLOAT_AGG:%.*]], align 4
+// CHECK-ZFH-ILP32-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_F16_FLOAT_AGG]], %struct.f16_float_agg* [[RETVAL]], i32 0, i32 0
+// CHECK-ZFH-ILP32-NEXT:    store half 0xH3C00, half* [[A]], align 4
+// CHECK-ZFH-ILP32-NEXT:    [[B:%.*]] = getelementptr inbounds [[STRUCT_F16_FLOAT_AGG]], %struct.f16_float_agg* [[RETVAL]], i32 0, i32 1
+// CHECK-ZFH-ILP32-NEXT:    store float 2.000000e+00, float* [[B]], align 4
+// CHECK-ZFH-ILP32-NEXT:    [[TMP0:%.*]] = bitcast %struct.f16_float_agg* [[RETVAL]] to [2 x i32]*
+// CHECK-ZFH-ILP32-NEXT:    [[TMP1:%.*]] = load [2 x i32], [2 x i32]* [[TMP0]], align 4
+// CHECK-ZFH-ILP32-NEXT:    ret [2 x i32] [[TMP1]]
+//
+// CHECK-NOZFH-ILP32F-LABEL: @test_f16_agg_3(
+// CHECK-NOZFH-ILP32F-NEXT:  entry:
+// CHECK-NOZFH-ILP32F-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_F16_FLOAT_AGG:%.*]], align 4
+// CHECK-NOZFH-ILP32F-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_F16_FLOAT_AGG]], %struct.f16_float_agg* [[RETVAL]], i32 0, i32 0
+// CHECK-NOZFH-ILP32F-NEXT:    store half 0xH3C00, half* [[A]], align 4
+// CHECK-NOZFH-ILP32F-NEXT:    [[B:%.*]] = getelementptr inbounds [[STRUCT_F16_FLOAT_AGG]], %struct.f16_float_agg* [[RETVAL]], i32 0, i32 1
+// CHECK-NOZFH-ILP32F-NEXT:    store float 2.000000e+00, float* [[B]], align 4
+// CHECK-NOZFH-ILP32F-NEXT:    [[TMP0:%.*]] = bitcast %struct.f16_float_agg* [[RETVAL]] to [2 x i32]*
+// CHECK-NOZFH-ILP32F-NEXT:    [[TMP1:%.*]] = load [2 x i32], [2 x i32]* [[TMP0]], align 4
+// CHECK-NOZFH-ILP32F-NEXT:    ret [2 x i32] [[TMP1]]
+//
+// CHECK-ZFH-ILP32F-LABEL: @test_f16_agg_3(
+// CHECK-ZFH-ILP32F-NEXT:  entry:
+// CHECK-ZFH-ILP32F-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_F16_FLOAT_AGG:%.*]], align 4
+// CHECK-ZFH-ILP32F-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_F16_FLOAT_AGG]], %struct.f16_float_agg* [[RETVAL]], i32 0, i32 0
+// CHECK-ZFH-ILP32F-NEXT:    store half 0xH3C00, half* [[A]], align 4
+// CHECK-ZFH-ILP32F-NEXT:    [[B:%.*]] = getelementptr inbounds [[STRUCT_F16_FLOAT_AGG]], %struct.f16_float_agg* [[RETVAL]], i32 0, i32 1
+// CHECK-ZFH-ILP32F-NEXT:    store float 2.000000e+00, float* [[B]], align 4
+// CHECK-ZFH-ILP32F-NEXT:    [[TMP0:%.*]] = bitcast %struct.f16_float_agg* [[RETVAL]] to [2 x i32]*
+// CHECK-ZFH-ILP32F-NEXT:    [[TMP1:%.*]] = load [2 x i32], [2 x i32]* [[TMP0]], align 4
+// CHECK-ZFH-ILP32F-NEXT:    ret [2 x i32] [[TMP1]]
+//
+struct f16_float_agg test_f16_agg_3() { return (struct f16_float_agg){1.0, 2.0}; }
+
+struct f16_int_agg {
+  _Float16 a;
+  int32_t b;
+};
+
+// CHECK-NOZFH-ILP32-LABEL: @test_f16_agg_4(
+// CHECK-NOZFH-ILP32-NEXT:  entry:
+// CHECK-NOZFH-ILP32-NEXT:    [[A:%.*]] = alloca [[STRUCT_F16_INT_AGG:%.*]], align 4
+// CHECK-NOZFH-ILP32-NEXT:    [[TMP0:%.*]] = bitcast %struct.f16_int_agg* [[A]] to [2 x i32]*
+// CHECK-NOZFH-ILP32-NEXT:    store [2 x i32] [[A_COERCE:%.*]], [2 x i32]* [[TMP0]], align 4
+// CHECK-NOZFH-ILP32-NEXT:    ret void
+//
+// CHECK-ZFH-ILP32-LABEL: @test_f16_agg_4(
+// CHECK-ZFH-ILP32-NEXT:  entry:
+// CHECK-ZFH-ILP32-NEXT:    [[A:%.*]] = alloca [[STRUCT_F16_INT_AGG:%.*]], align 4
+// CHECK-ZFH-ILP32-NEXT:    [[TMP0:%.*]] = bitcast %struct.f16_int_agg* [[A]] to [2 x i32]*
+// CHECK-ZFH-ILP32-NEXT:    store [2 x i32] [[A_COERCE:%.*]], [2 x i32]* [[TMP0]], align 4
+// CHECK-ZFH-ILP32-NEXT:    ret void
+//
+// CHECK-NOZFH-ILP32F-LABEL: @test_f16_agg_4(
+// CHECK-NOZFH-ILP32F-NEXT:  entry:
+// CHECK-NOZFH-ILP32F-NEXT:    [[A:%.*]] = alloca [[STRUCT_F16_INT_AGG:%.*]], align 4
+// CHECK-NOZFH-ILP32F-NEXT:    [[TMP0:%.*]] = bitcast %struct.f16_int_agg* [[A]] to [2 x i32]*
+// CHECK-NOZFH-ILP32F-NEXT:    store [2 x i32] [[A_COERCE:%.*]], [2 x i32]* [[TMP0]], align 4
+// CHECK-NOZFH-ILP32F-NEXT:    ret void
+//
+// CHECK-ZFH-ILP32F-LABEL: @test_f16_agg_4(
+// CHECK-ZFH-ILP32F-NEXT:  entry:
+// CHECK-ZFH-ILP32F-NEXT:    [[A:%.*]] = alloca [[STRUCT_F16_INT_AGG:%.*]], align 4
+// CHECK-ZFH-ILP32F-NEXT:    [[TMP0:%.*]] = bitcast %struct.f16_int_agg* [[A]] to [2 x i32]*
+// CHECK-ZFH-ILP32F-NEXT:    store [2 x i32] [[A_COERCE:%.*]], [2 x i32]* [[TMP0]], align 4
+// CHECK-ZFH-ILP32F-NEXT:    ret void
+//
+void test_f16_agg_4(struct f16_int_agg a) {}
+
+// CHECK-NOZFH-ILP32-LABEL: @test_f16_agg_5(
+// CHECK-NOZFH-ILP32-NEXT:  entry:
+// CHECK-NOZFH-ILP32-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_F16_INT_AGG:%.*]], align 4
+// CHECK-NOZFH-ILP32-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_F16_INT_AGG]], %struct.f16_int_agg* [[RETVAL]], i32 0, i32 0
+// CHECK-NOZFH-ILP32-NEXT:    store half 0xH3C00, half* [[A]], align 4
+// CHECK-NOZFH-ILP32-NEXT:    [[B:%.*]] = getelementptr inbounds [[STRUCT_F16_INT_AGG]], %struct.f16_int_agg* [[RETVAL]], i32 0, i32 1
+// CHECK-NOZFH-ILP32-NEXT:    store i32 2, i32* [[B]], align 4
+// CHECK-NOZFH-ILP32-NEXT:    [[TMP0:%.*]] = bitcast %struct.f16_int_agg* [[RETVAL]] to [2 x i32]*
+// CHECK-NOZFH-ILP32-NEXT:    [[TMP1:%.*]] = load [2 x i32], [2 x i32]* [[TMP0]], align 4
+// CHECK-NOZFH-ILP32-NEXT:    ret [2 x i32] [[TMP1]]
+//
+// CHECK-ZFH-ILP32-LABEL: @test_f16_agg_5(
+// CHECK-ZFH-ILP32-NEXT:  entry:
+// CHECK-ZFH-ILP32-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_F16_INT_AGG:%.*]], align 4
+// CHECK-ZFH-ILP32-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_F16_INT_AGG]], %struct.f16_int_agg* [[RETVAL]], i32 0, i32 0
+// CHECK-ZFH-ILP32-NEXT:    store half 0xH3C00, half* [[A]], align 4
+// CHECK-ZFH-ILP32-NEXT:    [[B:%.*]] = getelementptr inbounds [[STRUCT_F16_INT_AGG]], %struct.f16_int_agg* [[RETVAL]], i32 0, i32 1
+// CHECK-ZFH-ILP32-NEXT:    store i32 2, i32* [[B]], align 4
+// CHECK-ZFH-ILP32-NEXT:    [[TMP0:%.*]] = bitcast %struct.f16_int_agg* [[RETVAL]] to [2 x i32]*
+// CHECK-ZFH-ILP32-NEXT:    [[TMP1:%.*]] = load [2 x i32], [2 x i32]* [[TMP0]], align 4
+// CHECK-ZFH-ILP32-NEXT:    ret [2 x i32] [[TMP1]]
+//
+// CHECK-NOZFH-ILP32F-LABEL: @test_f16_agg_5(
+// CHECK-NOZFH-ILP32F-NEXT:  entry:
+// CHECK-NOZFH-ILP32F-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_F16_INT_AGG:%.*]], align 4
+// CHECK-NOZFH-ILP32F-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_F16_INT_AGG]], %struct.f16_int_agg* [[RETVAL]], i32 0, i32 0
+// CHECK-NOZFH-ILP32F-NEXT:    store half 0xH3C00, half* [[A]], align 4
+// CHECK-NOZFH-ILP32F-NEXT:    [[B:%.*]] = getelementptr inbounds [[STRUCT_F16_INT_AGG]], %struct.f16_int_agg* [[RETVAL]], i32 0, i32 1
+// CHECK-NOZFH-ILP32F-NEXT:    store i32 2, i32* [[B]], align 4
+// CHECK-NOZFH-ILP32F-NEXT:    [[TMP0:%.*]] = bitcast %struct.f16_int_agg* [[RETVAL]] to [2 x i32]*
+// CHECK-NOZFH-ILP32F-NEXT:    [[TMP1:%.*]] = load [2 x i32], [2 x i32]* [[TMP0]], align 4
+// CHECK-NOZFH-ILP32F-NEXT:    ret [2 x i32] [[TMP1]]
+//
+// CHECK-ZFH-ILP32F-LABEL: @test_f16_agg_5(
+// CHECK-ZFH-ILP32F-NEXT:  entry:
+// CHECK-ZFH-ILP32F-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_F16_INT_AGG:%.*]], align 4
+// CHECK-ZFH-ILP32F-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_F16_INT_AGG]], %struct.f16_int_agg* [[RETVAL]], i32 0, i32 0
+// CHECK-ZFH-ILP32F-NEXT:    store half 0xH3C00, half* [[A]], align 4
+// CHECK-ZFH-ILP32F-NEXT:    [[B:%.*]] = getelementptr inbounds [[STRUCT_F16_INT_AGG]], %struct.f16_int_agg* [[RETVAL]], i32 0, i32 1
+// CHECK-ZFH-ILP32F-NEXT:    store i32 2, i32* [[B]], align 4
+// CHECK-ZFH-ILP32F-NEXT:    [[TMP0:%.*]] = bitcast %struct.f16_int_agg* [[RETVAL]] to [2 x i32]*
+// CHECK-ZFH-ILP32F-NEXT:    [[TMP1:%.*]] = load [2 x i32], [2 x i32]* [[TMP0]], align 4
+// CHECK-ZFH-ILP32F-NEXT:    ret [2 x i32] [[TMP1]]
+//
+struct f16_int_agg test_f16_agg_5() { return (struct f16_int_agg){1.0, 2}; }
+
+// CHECK-NOZFH-ILP32-LABEL: @f16_scalar_stack_1(
+// CHECK-NOZFH-ILP32-NEXT:  entry:
+// CHECK-NOZFH-ILP32-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
+// CHECK-NOZFH-ILP32-NEXT:    [[B_ADDR:%.*]] = alloca i64, align 8
+// CHECK-NOZFH-ILP32-NEXT:    [[C_ADDR:%.*]] = alloca float, align 4
+// CHECK-NOZFH-ILP32-NEXT:    [[D_ADDR:%.*]] = alloca half, align 2
+// CHECK-NOZFH-ILP32-NEXT:    store i32 [[A:%.*]], i32* [[A_ADDR]], align 4
+// CHECK-NOZFH-ILP32-NEXT:    store i64 [[B:%.*]], i64* [[B_ADDR]], align 8
+// CHECK-NOZFH-ILP32-NEXT:    store float [[C:%.*]], float* [[C_ADDR]], align 4
+// CHECK-NOZFH-ILP32-NEXT:    store half [[D:%.*]], half* [[D_ADDR]], align 2
+// CHECK-NOZFH-ILP32-NEXT:    [[TMP0:%.*]] = load half, half* [[D_ADDR]], align 2
+// CHECK-NOZFH-ILP32-NEXT:    ret half [[TMP0]]
+//
+// CHECK-ZFH-ILP32-LABEL: @f16_scalar_stack_1(
+// CHECK-ZFH-ILP32-NEXT:  entry:
+// CHECK-ZFH-ILP32-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
+// CHECK-ZFH-ILP32-NEXT:    [[B_ADDR:%.*]] = alloca i64, align 8
+// CHECK-ZFH-ILP32-NEXT:    [[C_ADDR:%.*]] = alloca float, align 4
+// CHECK-ZFH-ILP32-NEXT:    [[D_ADDR:%.*]] = alloca half, align 2
+// CHECK-ZFH-ILP32-NEXT:    store i32 [[A:%.*]], i32* [[A_ADDR]], align 4
+// CHECK-ZFH-ILP32-NEXT:    store i64 [[B:%.*]], i64* [[B_ADDR]], align 8
+// CHECK-ZFH-ILP32-NEXT:    store float [[C:%.*]], float* [[C_ADDR]], align 4
+// CHECK-ZFH-ILP32-NEXT:    store half [[D:%.*]], half* [[D_ADDR]], align 2
+// CHECK-ZFH-ILP32-NEXT:    [[TMP0:%.*]] = load half, half* [[D_ADDR]], align 2
+// CHECK-ZFH-ILP32-NEXT:    ret half [[TMP0]]
+//
+// CHECK-NOZFH-ILP32F-LABEL: @f16_scalar_stack_1(
+// CHECK-NOZFH-ILP32F-NEXT:  entry:
+// CHECK-NOZFH-ILP32F-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
+// CHECK-NOZFH-ILP32F-NEXT:    [[B_ADDR:%.*]] = alloca i64, align 8
+// CHECK-NOZFH-ILP32F-NEXT:    [[C_ADDR:%.*]] = alloca float, align 4
+// CHECK-NOZFH-ILP32F-NEXT:    [[D_ADDR:%.*]] = alloca half, align 2
+// CHECK-NOZFH-ILP32F-NEXT:    store i32 [[A:%.*]], i32* [[A_ADDR]], align 4
+// CHECK-NOZFH-ILP32F-NEXT:    store i64 [[B:%.*]], i64* [[B_ADDR]], align 8
+// CHECK-NOZFH-ILP32F-NEXT:    store float [[C:%.*]], float* [[C_ADDR]], align 4
+// CHECK-NOZFH-ILP32F-NEXT:    store half [[D:%.*]], half* [[D_ADDR]], align 2
+// CHECK-NOZFH-ILP32F-NEXT:    [[TMP0:%.*]] = load half, half* [[D_ADDR]], align 2
+// CHECK-NOZFH-ILP32F-NEXT:    ret half [[TMP0]]
+//
+// CHECK-ZFH-ILP32F-LABEL: @f16_scalar_stack_1(
+// CHECK-ZFH-ILP32F-NEXT:  entry:
+// CHECK-ZFH-ILP32F-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
+// CHECK-ZFH-ILP32F-NEXT:    [[B_ADDR:%.*]] = alloca i64, align 8
+// CHECK-ZFH-ILP32F-NEXT:    [[C_ADDR:%.*]] = alloca float, align 4
+// CHECK-ZFH-ILP32F-NEXT:    [[D_ADDR:%.*]] = alloca half, align 2
+// CHECK-ZFH-ILP32F-NEXT:    store i32 [[A:%.*]], i32* [[A_ADDR]], align 4
+// CHECK-ZFH-ILP32F-NEXT:    store i64 [[B:%.*]], i64* [[B_ADDR]], align 8
+// CHECK-ZFH-ILP32F-NEXT:    store float [[C:%.*]], float* [[C_ADDR]], align 4
+// CHECK-ZFH-ILP32F-NEXT:    store half [[D:%.*]], half* [[D_ADDR]], align 2
+// CHECK-ZFH-ILP32F-NEXT:    [[TMP0:%.*]] = load half, half* [[D_ADDR]], align 2
+// CHECK-ZFH-ILP32F-NEXT:    ret half [[TMP0]]
+//
+_Float16 f16_scalar_stack_1(int32_t a, int64_t b, float c, _Float16 d) {
+  return d;
+}