diff --git a/clang/test/CodeGen/systemz-abi-vector.c b/clang/test/CodeGen/systemz-abi-vector.c
index 6eff24c1e680..06c5f3f8bd66 100644
--- a/clang/test/CodeGen/systemz-abi-vector.c
+++ b/clang/test/CodeGen/systemz-abi-vector.c
@@ -101,37 +101,145 @@ v1f128 pass_v1f128(v1f128 arg) { return arg; }
 
 v1i8 va_v1i8(__builtin_va_list l) { return __builtin_va_arg(l, v1i8); }
 // CHECK-LABEL: define void @va_v1i8(<1 x i8>* noalias sret %{{.*}}, %struct.__va_list_tag* %{{.*}})
-// CHECK: %reg_offset = add i64 %scaled_reg_count, 16
-// CHECK: %raw_mem_addr = getelementptr i8, i8* %overflow_arg_area, i64 0
-// CHECK: %indirect_arg = load <1 x i8>*
+// CHECK: [[REG_COUNT_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 0
+// CHECK: [[REG_COUNT:%[^ ]+]] = load i64, i64* [[REG_COUNT_PTR]]
+// CHECK: [[FITS_IN_REGS:%[^ ]+]] = icmp ult i64 [[REG_COUNT]], 5
+// CHECK: br i1 [[FITS_IN_REGS]],
+// CHECK: [[SCALED_REG_COUNT:%[^ ]+]] = mul i64 [[REG_COUNT]], 8
+// CHECK: [[REG_OFFSET:%[^ ]+]] = add i64 [[SCALED_REG_COUNT]], 16
+// CHECK: [[REG_SAVE_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 3
+// CHECK: [[REG_SAVE_AREA:%[^ ]+]] = load i8*, i8** [[REG_SAVE_AREA_PTR:[^ ]+]]
+// CHECK: [[RAW_REG_ADDR:%[^ ]+]] = getelementptr i8, i8* [[REG_SAVE_AREA]], i64 [[REG_OFFSET]]
+// CHECK: [[REG_ADDR:%[^ ]+]] = bitcast i8* [[RAW_REG_ADDR]] to <1 x i8>**
+// CHECK: [[REG_COUNT1:%[^ ]+]] = add i64 [[REG_COUNT]], 1
+// CHECK: store i64 [[REG_COUNT1]], i64* [[REG_COUNT_PTR]]
+// CHECK: [[OVERFLOW_ARG_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 2
+// CHECK: [[OVERFLOW_ARG_AREA:%[^ ]+]] = load i8*, i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[RAW_MEM_ADDR:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 0
+// CHECK: [[MEM_ADDR:%[^ ]+]] = bitcast i8* [[RAW_MEM_ADDR]] to <1 x i8>**
+// CHECK: [[OVERFLOW_ARG_AREA2:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 8
+// CHECK: store i8* [[OVERFLOW_ARG_AREA2]], i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[VA_ARG_ADDR:%[^ ]+]] = phi <1 x i8>** [ [[REG_ADDR]], %{{.*}} ], [ [[MEM_ADDR]], %{{.*}} ]
+// CHECK: [[INDIRECT_ARG:%[^ ]+]] = load <1 x i8>*, <1 x i8>** [[VA_ARG_ADDR]]
+// CHECK: ret void
 
 v2i8 va_v2i8(__builtin_va_list l) { return __builtin_va_arg(l, v2i8); }
 // CHECK-LABEL: define void @va_v2i8(<2 x i8>* noalias sret %{{.*}}, %struct.__va_list_tag* %{{.*}})
-// CHECK: %reg_offset = add i64 %scaled_reg_count, 16
-// CHECK: %raw_mem_addr = getelementptr i8, i8* %overflow_arg_area, i64 0
-// CHECK: %indirect_arg = load <2 x i8>*
+// CHECK: [[REG_COUNT_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 0
+// CHECK: [[REG_COUNT:%[^ ]+]] = load i64, i64* [[REG_COUNT_PTR]]
+// CHECK: [[FITS_IN_REGS:%[^ ]+]] = icmp ult i64 [[REG_COUNT]], 5
+// CHECK: br i1 [[FITS_IN_REGS]],
+// CHECK: [[SCALED_REG_COUNT:%[^ ]+]] = mul i64 [[REG_COUNT]], 8
+// CHECK: [[REG_OFFSET:%[^ ]+]] = add i64 [[SCALED_REG_COUNT]], 16
+// CHECK: [[REG_SAVE_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 3
+// CHECK: [[REG_SAVE_AREA:%[^ ]+]] = load i8*, i8** [[REG_SAVE_AREA_PTR:[^ ]+]]
+// CHECK: [[RAW_REG_ADDR:%[^ ]+]] = getelementptr i8, i8* [[REG_SAVE_AREA]], i64 [[REG_OFFSET]]
+// CHECK: [[REG_ADDR:%[^ ]+]] = bitcast i8* [[RAW_REG_ADDR]] to <2 x i8>**
+// CHECK: [[REG_COUNT1:%[^ ]+]] = add i64 [[REG_COUNT]], 1
+// CHECK: store i64 [[REG_COUNT1]], i64* [[REG_COUNT_PTR]]
+// CHECK: [[OVERFLOW_ARG_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 2
+// CHECK: [[OVERFLOW_ARG_AREA:%[^ ]+]] = load i8*, i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[RAW_MEM_ADDR:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 0
+// CHECK: [[MEM_ADDR:%[^ ]+]] = bitcast i8* [[RAW_MEM_ADDR]] to <2 x i8>**
+// CHECK: [[OVERFLOW_ARG_AREA2:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 8
+// CHECK: store i8* [[OVERFLOW_ARG_AREA2]], i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[VA_ARG_ADDR:%[^ ]+]] = phi <2 x i8>** [ [[REG_ADDR]], %{{.*}} ], [ [[MEM_ADDR]], %{{.*}} ]
+// CHECK: [[INDIRECT_ARG:%[^ ]+]] = load <2 x i8>*, <2 x i8>** [[VA_ARG_ADDR]]
+// CHECK: ret void
 
 v4i8 va_v4i8(__builtin_va_list l) { return __builtin_va_arg(l, v4i8); }
 // CHECK-LABEL: define void @va_v4i8(<4 x i8>* noalias sret %{{.*}}, %struct.__va_list_tag* %{{.*}})
-// CHECK: %reg_offset = add i64 %scaled_reg_count, 16
-// CHECK: %raw_mem_addr = getelementptr i8, i8* %overflow_arg_area, i64 0
-// CHECK: %indirect_arg = load <4 x i8>*
+// CHECK: [[REG_COUNT_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 0
+// CHECK: [[REG_COUNT:%[^ ]+]] = load i64, i64* [[REG_COUNT_PTR]]
+// CHECK: [[FITS_IN_REGS:%[^ ]+]] = icmp ult i64 [[REG_COUNT]], 5
+// CHECK: br i1 [[FITS_IN_REGS]],
+// CHECK: [[SCALED_REG_COUNT:%[^ ]+]] = mul i64 [[REG_COUNT]], 8
+// CHECK: [[REG_OFFSET:%[^ ]+]] = add i64 [[SCALED_REG_COUNT]], 16
+// CHECK: [[REG_SAVE_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 3
+// CHECK: [[REG_SAVE_AREA:%[^ ]+]] = load i8*, i8** [[REG_SAVE_AREA_PTR:[^ ]+]]
+// CHECK: [[RAW_REG_ADDR:%[^ ]+]] = getelementptr i8, i8* [[REG_SAVE_AREA]], i64 [[REG_OFFSET]]
+// CHECK: [[REG_ADDR:%[^ ]+]] = bitcast i8* [[RAW_REG_ADDR]] to <4 x i8>**
+// CHECK: [[REG_COUNT1:%[^ ]+]] = add i64 [[REG_COUNT]], 1
+// CHECK: store i64 [[REG_COUNT1]], i64* [[REG_COUNT_PTR]]
+// CHECK: [[OVERFLOW_ARG_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 2
+// CHECK: [[OVERFLOW_ARG_AREA:%[^ ]+]] = load i8*, i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[RAW_MEM_ADDR:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 0
+// CHECK: [[MEM_ADDR:%[^ ]+]] = bitcast i8* [[RAW_MEM_ADDR]] to <4 x i8>**
+// CHECK: [[OVERFLOW_ARG_AREA2:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 8
+// CHECK: store i8* [[OVERFLOW_ARG_AREA2]], i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[VA_ARG_ADDR:%[^ ]+]] = phi <4 x i8>** [ [[REG_ADDR]], %{{.*}} ], [ [[MEM_ADDR]], %{{.*}} ]
+// CHECK: [[INDIRECT_ARG:%[^ ]+]] = load <4 x i8>*, <4 x i8>** [[VA_ARG_ADDR]]
+// CHECK: ret void
 
 v8i8 va_v8i8(__builtin_va_list l) { return __builtin_va_arg(l, v8i8); }
 // CHECK-LABEL: define void @va_v8i8(<8 x i8>* noalias sret %{{.*}}, %struct.__va_list_tag* %{{.*}})
-// CHECK: %reg_offset = add i64 %scaled_reg_count, 16
-// CHECK: %raw_mem_addr = getelementptr i8, i8* %overflow_arg_area, i64 0
-// CHECK: %indirect_arg = load <8 x i8>*
+// CHECK: [[REG_COUNT_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 0
+// CHECK: [[REG_COUNT:%[^ ]+]] = load i64, i64* [[REG_COUNT_PTR]]
+// CHECK: [[FITS_IN_REGS:%[^ ]+]] = icmp ult i64 [[REG_COUNT]], 5
+// CHECK: br i1 [[FITS_IN_REGS]],
+// CHECK: [[SCALED_REG_COUNT:%[^ ]+]] = mul i64 [[REG_COUNT]], 8
+// CHECK: [[REG_OFFSET:%[^ ]+]] = add i64 [[SCALED_REG_COUNT]], 16
+// CHECK: [[REG_SAVE_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 3
+// CHECK: [[REG_SAVE_AREA:%[^ ]+]] = load i8*, i8** [[REG_SAVE_AREA_PTR:[^ ]+]]
+// CHECK: [[RAW_REG_ADDR:%[^ ]+]] = getelementptr i8, i8* [[REG_SAVE_AREA]], i64 [[REG_OFFSET]]
+// CHECK: [[REG_ADDR:%[^ ]+]] = bitcast i8* [[RAW_REG_ADDR]] to <8 x i8>**
+// CHECK: [[REG_COUNT1:%[^ ]+]] = add i64 [[REG_COUNT]], 1
+// CHECK: store i64 [[REG_COUNT1]], i64* [[REG_COUNT_PTR]]
+// CHECK: [[OVERFLOW_ARG_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 2
+// CHECK: [[OVERFLOW_ARG_AREA:%[^ ]+]] = load i8*, i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[RAW_MEM_ADDR:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 0
+// CHECK: [[MEM_ADDR:%[^ ]+]] = bitcast i8* [[RAW_MEM_ADDR]] to <8 x i8>**
+// CHECK: [[OVERFLOW_ARG_AREA2:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 8
+// CHECK: store i8* [[OVERFLOW_ARG_AREA2]], i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[VA_ARG_ADDR:%[^ ]+]] = phi <8 x i8>** [ [[REG_ADDR]], %{{.*}} ], [ [[MEM_ADDR]], %{{.*}} ]
+// CHECK: [[INDIRECT_ARG:%[^ ]+]] = load <8 x i8>*, <8 x i8>** [[VA_ARG_ADDR]]
+// CHECK: ret void
 
 v16i8 va_v16i8(__builtin_va_list l) { return __builtin_va_arg(l, v16i8); }
 // CHECK-LABEL: define void @va_v16i8(<16 x i8>* noalias sret %{{.*}}, %struct.__va_list_tag* %{{.*}})
-// CHECK: %reg_offset = add i64 %scaled_reg_count, 16
-// CHECK: %raw_mem_addr = getelementptr i8, i8* %overflow_arg_area, i64 0
-// CHECK: %indirect_arg = load <16 x i8>*
+// CHECK: [[REG_COUNT_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 0
+// CHECK: [[REG_COUNT:%[^ ]+]] = load i64, i64* [[REG_COUNT_PTR]]
+// CHECK: [[FITS_IN_REGS:%[^ ]+]] = icmp ult i64 [[REG_COUNT]], 5
+// CHECK: br i1 [[FITS_IN_REGS]],
+// CHECK: [[SCALED_REG_COUNT:%[^ ]+]] = mul i64 [[REG_COUNT]], 8
+// CHECK: [[REG_OFFSET:%[^ ]+]] = add i64 [[SCALED_REG_COUNT]], 16
+// CHECK: [[REG_SAVE_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 3
+// CHECK: [[REG_SAVE_AREA:%[^ ]+]] = load i8*, i8** [[REG_SAVE_AREA_PTR:[^ ]+]]
+// CHECK: [[RAW_REG_ADDR:%[^ ]+]] = getelementptr i8, i8* [[REG_SAVE_AREA]], i64 [[REG_OFFSET]]
+// CHECK: [[REG_ADDR:%[^ ]+]] = bitcast i8* [[RAW_REG_ADDR]] to <16 x i8>**
+// CHECK: [[REG_COUNT1:%[^ ]+]] = add i64 [[REG_COUNT]], 1
+// CHECK: store i64 [[REG_COUNT1]], i64* [[REG_COUNT_PTR]]
+// CHECK: [[OVERFLOW_ARG_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 2
+// CHECK: [[OVERFLOW_ARG_AREA:%[^ ]+]] = load i8*, i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[RAW_MEM_ADDR:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 0
+// CHECK: [[MEM_ADDR:%[^ ]+]] = bitcast i8* [[RAW_MEM_ADDR]] to <16 x i8>**
+// CHECK: [[OVERFLOW_ARG_AREA2:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 8
+// CHECK: store i8* [[OVERFLOW_ARG_AREA2]], i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[VA_ARG_ADDR:%[^ ]+]] = phi <16 x i8>** [ [[REG_ADDR]], %{{.*}} ], [ [[MEM_ADDR]], %{{.*}} ]
+// CHECK: [[INDIRECT_ARG:%[^ ]+]] = load <16 x i8>*, <16 x i8>** [[VA_ARG_ADDR]]
+// CHECK: ret void
 
 v32i8 va_v32i8(__builtin_va_list l) { return __builtin_va_arg(l, v32i8); }
 // CHECK-LABEL: define void @va_v32i8(<32 x i8>* noalias sret %{{.*}}, %struct.__va_list_tag* %{{.*}})
-// CHECK: %reg_offset = add i64 %scaled_reg_count, 16
-// CHECK: %raw_mem_addr = getelementptr i8, i8* %overflow_arg_area, i64 0
-// CHECK: %indirect_arg = load <32 x i8>*
+// CHECK: [[REG_COUNT_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 0
+// CHECK: [[REG_COUNT:%[^ ]+]] = load i64, i64* [[REG_COUNT_PTR]]
+// CHECK: [[FITS_IN_REGS:%[^ ]+]] = icmp ult i64 [[REG_COUNT]], 5
+// CHECK: br i1 [[FITS_IN_REGS]],
+// CHECK: [[SCALED_REG_COUNT:%[^ ]+]] = mul i64 [[REG_COUNT]], 8
+// CHECK: [[REG_OFFSET:%[^ ]+]] = add i64 [[SCALED_REG_COUNT]], 16
+// CHECK: [[REG_SAVE_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 3
+// CHECK: [[REG_SAVE_AREA:%[^ ]+]] = load i8*, i8** [[REG_SAVE_AREA_PTR:[^ ]+]]
+// CHECK: [[RAW_REG_ADDR:%[^ ]+]] = getelementptr i8, i8* [[REG_SAVE_AREA]], i64 [[REG_OFFSET]]
+// CHECK: [[REG_ADDR:%[^ ]+]] = bitcast i8* [[RAW_REG_ADDR]] to <32 x i8>**
+// CHECK: [[REG_COUNT1:%[^ ]+]] = add i64 [[REG_COUNT]], 1
+// CHECK: store i64 [[REG_COUNT1]], i64* [[REG_COUNT_PTR]]
+// CHECK: [[OVERFLOW_ARG_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 2
+// CHECK: [[OVERFLOW_ARG_AREA:%[^ ]+]] = load i8*, i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[RAW_MEM_ADDR:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 0
+// CHECK: [[MEM_ADDR:%[^ ]+]] = bitcast i8* [[RAW_MEM_ADDR]] to <32 x i8>**
+// CHECK: [[OVERFLOW_ARG_AREA2:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 8
+// CHECK: store i8* [[OVERFLOW_ARG_AREA2]], i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[VA_ARG_ADDR:%[^ ]+]] = phi <32 x i8>** [ [[REG_ADDR]], %{{.*}} ], [ [[MEM_ADDR]], %{{.*}} ]
+// CHECK: [[INDIRECT_ARG:%[^ ]+]] = load <32 x i8>*, <32 x i8>** [[VA_ARG_ADDR]]
+// CHECK: ret void
 
diff --git a/clang/test/CodeGen/systemz-abi.c b/clang/test/CodeGen/systemz-abi.c
index 4de1d9806012..5599c79e9005 100644
--- a/clang/test/CodeGen/systemz-abi.c
+++ b/clang/test/CodeGen/systemz-abi.c
@@ -138,115 +138,449 @@ struct agg_nofloat3 pass_agg_nofloat3(struct agg_nofloat3 arg) { return arg; }
 
 int va_int(__builtin_va_list l) { return __builtin_va_arg(l, int); }
 // CHECK-LABEL: define signext i32 @va_int(%struct.__va_list_tag* %{{.*}})
-// CHECK: %reg_offset = add i64 %scaled_reg_count, 20
-// CHECK: %raw_mem_addr = getelementptr i8, i8* %overflow_arg_area, i64 4
-// CHECK-NOT: %indirect_arg
+// CHECK: [[REG_COUNT_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 0
+// CHECK: [[REG_COUNT:%[^ ]+]] = load i64, i64* [[REG_COUNT_PTR]]
+// CHECK: [[FITS_IN_REGS:%[^ ]+]] = icmp ult i64 [[REG_COUNT]], 5
+// CHECK: br i1 [[FITS_IN_REGS]],
+// CHECK: [[SCALED_REG_COUNT:%[^ ]+]] = mul i64 [[REG_COUNT]], 8
+// CHECK: [[REG_OFFSET:%[^ ]+]] = add i64 [[SCALED_REG_COUNT]], 20
+// CHECK: [[REG_SAVE_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 3
+// CHECK: [[REG_SAVE_AREA:%[^ ]+]] = load i8*, i8** [[REG_SAVE_AREA_PTR:[^ ]+]]
+// CHECK: [[RAW_REG_ADDR:%[^ ]+]] = getelementptr i8, i8* [[REG_SAVE_AREA]], i64 [[REG_OFFSET]]
+// CHECK: [[REG_ADDR:%[^ ]+]] = bitcast i8* [[RAW_REG_ADDR]] to i32*
+// CHECK: [[REG_COUNT1:%[^ ]+]] = add i64 [[REG_COUNT]], 1
+// CHECK: store i64 [[REG_COUNT1]], i64* [[REG_COUNT_PTR]]
+// CHECK: [[OVERFLOW_ARG_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 2
+// CHECK: [[OVERFLOW_ARG_AREA:%[^ ]+]] = load i8*, i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[RAW_MEM_ADDR:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 4
+// CHECK: [[MEM_ADDR:%[^ ]+]] = bitcast i8* [[RAW_MEM_ADDR]] to i32*
+// CHECK: [[OVERFLOW_ARG_AREA2:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 8
+// CHECK: store i8* [[OVERFLOW_ARG_AREA2]], i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[VA_ARG_ADDR:%[^ ]+]] = phi i32* [ [[REG_ADDR]], %{{.*}} ], [ [[MEM_ADDR]], %{{.*}} ]
+// CHECK: [[RET:%[^ ]+]] = load i32, i32* [[VA_ARG_ADDR]]
+// CHECK: ret i32 [[RET]]
 
 long va_long(__builtin_va_list l) { return __builtin_va_arg(l, long); }
 // CHECK-LABEL: define i64 @va_long(%struct.__va_list_tag* %{{.*}})
-// CHECK: %reg_offset = add i64 %scaled_reg_count, 16
-// CHECK: %raw_mem_addr = getelementptr i8, i8* %overflow_arg_area, i64 0
-// CHECK-NOT: %indirect_arg
+// CHECK: [[REG_COUNT_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 0
+// CHECK: [[REG_COUNT:%[^ ]+]] = load i64, i64* [[REG_COUNT_PTR]]
+// CHECK: [[FITS_IN_REGS:%[^ ]+]] = icmp ult i64 [[REG_COUNT]], 5
+// CHECK: br i1 [[FITS_IN_REGS]],
+// CHECK: [[SCALED_REG_COUNT:%[^ ]+]] = mul i64 [[REG_COUNT]], 8
+// CHECK: [[REG_OFFSET:%[^ ]+]] = add i64 [[SCALED_REG_COUNT]], 16
+// CHECK: [[REG_SAVE_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 3
+// CHECK: [[REG_SAVE_AREA:%[^ ]+]] = load i8*, i8** [[REG_SAVE_AREA_PTR:[^ ]+]]
+// CHECK: [[RAW_REG_ADDR:%[^ ]+]] = getelementptr i8, i8* [[REG_SAVE_AREA]], i64 [[REG_OFFSET]]
+// CHECK: [[REG_ADDR:%[^ ]+]] = bitcast i8* [[RAW_REG_ADDR]] to i64*
+// CHECK: [[REG_COUNT1:%[^ ]+]] = add i64 [[REG_COUNT]], 1
+// CHECK: store i64 [[REG_COUNT1]], i64* [[REG_COUNT_PTR]]
+// CHECK: [[OVERFLOW_ARG_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 2
+// CHECK: [[OVERFLOW_ARG_AREA:%[^ ]+]] = load i8*, i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[RAW_MEM_ADDR:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 0
+// CHECK: [[MEM_ADDR:%[^ ]+]] = bitcast i8* [[RAW_MEM_ADDR]] to i64*
+// CHECK: [[OVERFLOW_ARG_AREA2:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 8
+// CHECK: store i8* [[OVERFLOW_ARG_AREA2]], i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[VA_ARG_ADDR:%[^ ]+]] = phi i64* [ [[REG_ADDR]], %{{.*}} ], [ [[MEM_ADDR]], %{{.*}} ]
+// CHECK: [[RET:%[^ ]+]] = load i64, i64* [[VA_ARG_ADDR]]
+// CHECK: ret i64 [[RET]]
 
 long long va_longlong(__builtin_va_list l) { return __builtin_va_arg(l, long long); }
 // CHECK-LABEL: define i64 @va_longlong(%struct.__va_list_tag* %{{.*}})
-// CHECK: %reg_offset = add i64 %scaled_reg_count, 16
-// CHECK: %raw_mem_addr = getelementptr i8, i8* %overflow_arg_area, i64 0
-// CHECK-NOT: %indirect_arg
+// CHECK: [[REG_COUNT_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 0
+// CHECK: [[REG_COUNT:%[^ ]+]] = load i64, i64* [[REG_COUNT_PTR]]
+// CHECK: [[FITS_IN_REGS:%[^ ]+]] = icmp ult i64 [[REG_COUNT]], 5
+// CHECK: br i1 [[FITS_IN_REGS]],
+// CHECK: [[SCALED_REG_COUNT:%[^ ]+]] = mul i64 [[REG_COUNT]], 8
+// CHECK: [[REG_OFFSET:%[^ ]+]] = add i64 [[SCALED_REG_COUNT]], 16
+// CHECK: [[REG_SAVE_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 3
+// CHECK: [[REG_SAVE_AREA:%[^ ]+]] = load i8*, i8** [[REG_SAVE_AREA_PTR:[^ ]+]]
+// CHECK: [[RAW_REG_ADDR:%[^ ]+]] = getelementptr i8, i8* [[REG_SAVE_AREA]], i64 [[REG_OFFSET]]
+// CHECK: [[REG_ADDR:%[^ ]+]] = bitcast i8* [[RAW_REG_ADDR]] to i64*
+// CHECK: [[REG_COUNT1:%[^ ]+]] = add i64 [[REG_COUNT]], 1
+// CHECK: store i64 [[REG_COUNT1]], i64* [[REG_COUNT_PTR]]
+// CHECK: [[OVERFLOW_ARG_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 2
+// CHECK: [[OVERFLOW_ARG_AREA:%[^ ]+]] = load i8*, i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[RAW_MEM_ADDR:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 0
+// CHECK: [[MEM_ADDR:%[^ ]+]] = bitcast i8* [[RAW_MEM_ADDR]] to i64*
+// CHECK: [[OVERFLOW_ARG_AREA2:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 8
+// CHECK: store i8* [[OVERFLOW_ARG_AREA2]], i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[VA_ARG_ADDR:%[^ ]+]] = phi i64* [ [[REG_ADDR]], %{{.*}} ], [ [[MEM_ADDR]], %{{.*}} ]
+// CHECK: [[RET:%[^ ]+]] = load i64, i64* [[VA_ARG_ADDR]]
+// CHECK: ret i64 [[RET]]
 
 double va_double(__builtin_va_list l) { return __builtin_va_arg(l, double); }
 // CHECK-LABEL: define double @va_double(%struct.__va_list_tag* %{{.*}})
-// CHECK: %reg_offset = add i64 %scaled_reg_count, 128
-// CHECK: %raw_mem_addr = getelementptr i8, i8* %overflow_arg_area, i64 0
-// CHECK-NOT: %indirect_arg
+// CHECK: [[REG_COUNT_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 1
+// CHECK: [[REG_COUNT:%[^ ]+]] = load i64, i64* [[REG_COUNT_PTR]]
+// CHECK: [[FITS_IN_REGS:%[^ ]+]] = icmp ult i64 [[REG_COUNT]], 4
+// CHECK: br i1 [[FITS_IN_REGS]],
+// CHECK: [[SCALED_REG_COUNT:%[^ ]+]] = mul i64 [[REG_COUNT]], 8
+// CHECK: [[REG_OFFSET:%[^ ]+]] = add i64 [[SCALED_REG_COUNT]], 128
+// CHECK: [[REG_SAVE_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 3
+// CHECK: [[REG_SAVE_AREA:%[^ ]+]] = load i8*, i8** [[REG_SAVE_AREA_PTR:[^ ]+]]
+// CHECK: [[RAW_REG_ADDR:%[^ ]+]] = getelementptr i8, i8* [[REG_SAVE_AREA]], i64 [[REG_OFFSET]]
+// CHECK: [[REG_ADDR:%[^ ]+]] = bitcast i8* [[RAW_REG_ADDR]] to double*
+// CHECK: [[REG_COUNT1:%[^ ]+]] = add i64 [[REG_COUNT]], 1
+// CHECK: store i64 [[REG_COUNT1]], i64* [[REG_COUNT_PTR]]
+// CHECK: [[OVERFLOW_ARG_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 2
+// CHECK: [[OVERFLOW_ARG_AREA:%[^ ]+]] = load i8*, i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[RAW_MEM_ADDR:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 0
+// CHECK: [[MEM_ADDR:%[^ ]+]] = bitcast i8* [[RAW_MEM_ADDR]] to double*
+// CHECK: [[OVERFLOW_ARG_AREA2:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 8
+// CHECK: store i8* [[OVERFLOW_ARG_AREA2]], i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[VA_ARG_ADDR:%[^ ]+]] = phi double* [ [[REG_ADDR]], %{{.*}} ], [ [[MEM_ADDR]], %{{.*}} ]
+// CHECK: [[RET:%[^ ]+]] = load double, double* [[VA_ARG_ADDR]]
+// CHECK: ret double [[RET]]
 
 long double va_longdouble(__builtin_va_list l) { return __builtin_va_arg(l, long double); }
 // CHECK-LABEL: define void @va_longdouble(fp128* noalias sret %{{.*}}, %struct.__va_list_tag* %{{.*}})
-// CHECK: %reg_offset = add i64 %scaled_reg_count, 16
-// CHECK: %raw_mem_addr = getelementptr i8, i8* %overflow_arg_area, i64 0
-// CHECK: %indirect_arg = load fp128*
+// CHECK: [[REG_COUNT_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 0
+// CHECK: [[REG_COUNT:%[^ ]+]] = load i64, i64* [[REG_COUNT_PTR]]
+// CHECK: [[FITS_IN_REGS:%[^ ]+]] = icmp ult i64 [[REG_COUNT]], 5
+// CHECK: br i1 [[FITS_IN_REGS]],
+// CHECK: [[SCALED_REG_COUNT:%[^ ]+]] = mul i64 [[REG_COUNT]], 8
+// CHECK: [[REG_OFFSET:%[^ ]+]] = add i64 [[SCALED_REG_COUNT]], 16
+// CHECK: [[REG_SAVE_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 3
+// CHECK: [[REG_SAVE_AREA:%[^ ]+]] = load i8*, i8** [[REG_SAVE_AREA_PTR:[^ ]+]]
+// CHECK: [[RAW_REG_ADDR:%[^ ]+]] = getelementptr i8, i8* [[REG_SAVE_AREA]], i64 [[REG_OFFSET]]
+// CHECK: [[REG_ADDR:%[^ ]+]] = bitcast i8* [[RAW_REG_ADDR]] to fp128**
+// CHECK: [[REG_COUNT1:%[^ ]+]] = add i64 [[REG_COUNT]], 1
+// CHECK: store i64 [[REG_COUNT1]], i64* [[REG_COUNT_PTR]]
+// CHECK: [[OVERFLOW_ARG_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 2
+// CHECK: [[OVERFLOW_ARG_AREA:%[^ ]+]] = load i8*, i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[RAW_MEM_ADDR:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 0
+// CHECK: [[MEM_ADDR:%[^ ]+]] = bitcast i8* [[RAW_MEM_ADDR]] to fp128**
+// CHECK: [[OVERFLOW_ARG_AREA2:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 8
+// CHECK: store i8* [[OVERFLOW_ARG_AREA2]], i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[VA_ARG_ADDR:%[^ ]+]] = phi fp128** [ [[REG_ADDR]], %{{.*}} ], [ [[MEM_ADDR]], %{{.*}} ]
+// CHECK: [[INDIRECT_ARG:%[^ ]+]] = load fp128*, fp128** [[VA_ARG_ADDR]]
+// CHECK: [[RET:%[^ ]+]] = load fp128, fp128* [[INDIRECT_ARG]]
+// CHECK: store fp128 [[RET]], fp128* %{{.*}} 
+// CHECK: ret void
 
 _Complex char va_complex_char(__builtin_va_list l) { return __builtin_va_arg(l, _Complex char); }
 // CHECK-LABEL: define void @va_complex_char({ i8, i8 }* noalias sret %{{.*}}, %struct.__va_list_tag* %{{.*}}
-// CHECK: %reg_offset = add i64 %scaled_reg_count, 16
-// CHECK: %raw_mem_addr = getelementptr i8, i8* %overflow_arg_area, i64 0
-// CHECK: %indirect_arg = load { i8, i8 }*
+// CHECK: [[REG_COUNT_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 0
+// CHECK: [[REG_COUNT:%[^ ]+]] = load i64, i64* [[REG_COUNT_PTR]]
+// CHECK: [[FITS_IN_REGS:%[^ ]+]] = icmp ult i64 [[REG_COUNT]], 5
+// CHECK: br i1 [[FITS_IN_REGS]],
+// CHECK: [[SCALED_REG_COUNT:%[^ ]+]] = mul i64 [[REG_COUNT]], 8
+// CHECK: [[REG_OFFSET:%[^ ]+]] = add i64 [[SCALED_REG_COUNT]], 16
+// CHECK: [[REG_SAVE_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 3
+// CHECK: [[REG_SAVE_AREA:%[^ ]+]] = load i8*, i8** [[REG_SAVE_AREA_PTR:[^ ]+]]
+// CHECK: [[RAW_REG_ADDR:%[^ ]+]] = getelementptr i8, i8* [[REG_SAVE_AREA]], i64 [[REG_OFFSET]]
+// CHECK: [[REG_ADDR:%[^ ]+]] = bitcast i8* [[RAW_REG_ADDR]] to { i8, i8 }**
+// CHECK: [[REG_COUNT1:%[^ ]+]] = add i64 [[REG_COUNT]], 1
+// CHECK: store i64 [[REG_COUNT1]], i64* [[REG_COUNT_PTR]]
+// CHECK: [[OVERFLOW_ARG_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 2
+// CHECK: [[OVERFLOW_ARG_AREA:%[^ ]+]] = load i8*, i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[RAW_MEM_ADDR:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 0
+// CHECK: [[MEM_ADDR:%[^ ]+]] = bitcast i8* [[RAW_MEM_ADDR]] to { i8, i8 }**
+// CHECK: [[OVERFLOW_ARG_AREA2:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 8
+// CHECK: store i8* [[OVERFLOW_ARG_AREA2]], i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[VA_ARG_ADDR:%[^ ]+]] = phi { i8, i8 }** [ [[REG_ADDR]], %{{.*}} ], [ [[MEM_ADDR]], %{{.*}} ]
+// CHECK: [[INDIRECT_ARG:%[^ ]+]] = load { i8, i8 }*, { i8, i8 }** [[VA_ARG_ADDR]]
+// CHECK: ret void
 
 struct agg_1byte va_agg_1byte(__builtin_va_list l) { return __builtin_va_arg(l, struct agg_1byte); }
 // CHECK-LABEL: define void @va_agg_1byte(%struct.agg_1byte* noalias sret %{{.*}}, %struct.__va_list_tag* %{{.*}}
-// CHECK: %reg_offset = add i64 %scaled_reg_count, 23
-// CHECK: %raw_mem_addr = getelementptr i8, i8* %overflow_arg_area, i64 7
-// CHECK-NOT: %indirect_arg
+// CHECK: [[REG_COUNT_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 0
+// CHECK: [[REG_COUNT:%[^ ]+]] = load i64, i64* [[REG_COUNT_PTR]]
+// CHECK: [[FITS_IN_REGS:%[^ ]+]] = icmp ult i64 [[REG_COUNT]], 5
+// CHECK: br i1 [[FITS_IN_REGS]],
+// CHECK: [[SCALED_REG_COUNT:%[^ ]+]] = mul i64 [[REG_COUNT]], 8
+// CHECK: [[REG_OFFSET:%[^ ]+]] = add i64 [[SCALED_REG_COUNT]], 23
+// CHECK: [[REG_SAVE_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 3
+// CHECK: [[REG_SAVE_AREA:%[^ ]+]] = load i8*, i8** [[REG_SAVE_AREA_PTR:[^ ]+]]
+// CHECK: [[RAW_REG_ADDR:%[^ ]+]] = getelementptr i8, i8* [[REG_SAVE_AREA]], i64 [[REG_OFFSET]]
+// CHECK: [[REG_ADDR:%[^ ]+]] = bitcast i8* [[RAW_REG_ADDR]] to %struct.agg_1byte*
+// CHECK: [[REG_COUNT1:%[^ ]+]] = add i64 [[REG_COUNT]], 1
+// CHECK: store i64 [[REG_COUNT1]], i64* [[REG_COUNT_PTR]]
+// CHECK: [[OVERFLOW_ARG_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 2
+// CHECK: [[OVERFLOW_ARG_AREA:%[^ ]+]] = load i8*, i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[RAW_MEM_ADDR:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 7
+// CHECK: [[MEM_ADDR:%[^ ]+]] = bitcast i8* [[RAW_MEM_ADDR]] to %struct.agg_1byte*
+// CHECK: [[OVERFLOW_ARG_AREA2:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 8
+// CHECK: store i8* [[OVERFLOW_ARG_AREA2]], i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[VA_ARG_ADDR:%[^ ]+]] = phi %struct.agg_1byte* [ [[REG_ADDR]], %{{.*}} ], [ [[MEM_ADDR]], %{{.*}} ]
+// CHECK: ret void
 
 struct agg_2byte va_agg_2byte(__builtin_va_list l) { return __builtin_va_arg(l, struct agg_2byte); }
 // CHECK-LABEL: define void @va_agg_2byte(%struct.agg_2byte* noalias sret %{{.*}}, %struct.__va_list_tag* %{{.*}}
-// CHECK: %reg_offset = add i64 %scaled_reg_count, 22
-// CHECK: %raw_mem_addr = getelementptr i8, i8* %overflow_arg_area, i64 6
-// CHECK-NOT: %indirect_arg
+// CHECK: [[REG_COUNT_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 0
+// CHECK: [[REG_COUNT:%[^ ]+]] = load i64, i64* [[REG_COUNT_PTR]]
+// CHECK: [[FITS_IN_REGS:%[^ ]+]] = icmp ult i64 [[REG_COUNT]], 5
+// CHECK: br i1 [[FITS_IN_REGS]],
+// CHECK: [[SCALED_REG_COUNT:%[^ ]+]] = mul i64 [[REG_COUNT]], 8
+// CHECK: [[REG_OFFSET:%[^ ]+]] = add i64 [[SCALED_REG_COUNT]], 22
+// CHECK: [[REG_SAVE_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 3
+// CHECK: [[REG_SAVE_AREA:%[^ ]+]] = load i8*, i8** [[REG_SAVE_AREA_PTR:[^ ]+]]
+// CHECK: [[RAW_REG_ADDR:%[^ ]+]] = getelementptr i8, i8* [[REG_SAVE_AREA]], i64 [[REG_OFFSET]]
+// CHECK: [[REG_ADDR:%[^ ]+]] = bitcast i8* [[RAW_REG_ADDR]] to %struct.agg_2byte*
+// CHECK: [[REG_COUNT1:%[^ ]+]] = add i64 [[REG_COUNT]], 1
+// CHECK: store i64 [[REG_COUNT1]], i64* [[REG_COUNT_PTR]]
+// CHECK: [[OVERFLOW_ARG_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 2
+// CHECK: [[OVERFLOW_ARG_AREA:%[^ ]+]] = load i8*, i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[RAW_MEM_ADDR:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 6
+// CHECK: [[MEM_ADDR:%[^ ]+]] = bitcast i8* [[RAW_MEM_ADDR]] to %struct.agg_2byte*
+// CHECK: [[OVERFLOW_ARG_AREA2:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 8
+// CHECK: store i8* [[OVERFLOW_ARG_AREA2]], i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[VA_ARG_ADDR:%[^ ]+]] = phi %struct.agg_2byte* [ [[REG_ADDR]], %{{.*}} ], [ [[MEM_ADDR]], %{{.*}} ]
+// CHECK: ret void
 
 struct agg_3byte va_agg_3byte(__builtin_va_list l) { return __builtin_va_arg(l, struct agg_3byte); }
 // CHECK-LABEL: define void @va_agg_3byte(%struct.agg_3byte* noalias sret %{{.*}}, %struct.__va_list_tag* %{{.*}}
-// CHECK: %reg_offset = add i64 %scaled_reg_count, 16
-// CHECK: %raw_mem_addr = getelementptr i8, i8* %overflow_arg_area, i64 0
-// CHECK: %indirect_arg = load %struct.agg_3byte*
+// CHECK: [[REG_COUNT_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 0
+// CHECK: [[REG_COUNT:%[^ ]+]] = load i64, i64* [[REG_COUNT_PTR]]
+// CHECK: [[FITS_IN_REGS:%[^ ]+]] = icmp ult i64 [[REG_COUNT]], 5
+// CHECK: br i1 [[FITS_IN_REGS]],
+// CHECK: [[SCALED_REG_COUNT:%[^ ]+]] = mul i64 [[REG_COUNT]], 8
+// CHECK: [[REG_OFFSET:%[^ ]+]] = add i64 [[SCALED_REG_COUNT]], 16
+// CHECK: [[REG_SAVE_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 3
+// CHECK: [[REG_SAVE_AREA:%[^ ]+]] = load i8*, i8** [[REG_SAVE_AREA_PTR:[^ ]+]]
+// CHECK: [[RAW_REG_ADDR:%[^ ]+]] = getelementptr i8, i8* [[REG_SAVE_AREA]], i64 [[REG_OFFSET]]
+// CHECK: [[REG_ADDR:%[^ ]+]] = bitcast i8* [[RAW_REG_ADDR]] to %struct.agg_3byte**
+// CHECK: [[REG_COUNT1:%[^ ]+]] = add i64 [[REG_COUNT]], 1
+// CHECK: store i64 [[REG_COUNT1]], i64* [[REG_COUNT_PTR]]
+// CHECK: [[OVERFLOW_ARG_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 2
+// CHECK: [[OVERFLOW_ARG_AREA:%[^ ]+]] = load i8*, i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[RAW_MEM_ADDR:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 0
+// CHECK: [[MEM_ADDR:%[^ ]+]] = bitcast i8* [[RAW_MEM_ADDR]] to %struct.agg_3byte**
+// CHECK: [[OVERFLOW_ARG_AREA2:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 8
+// CHECK: store i8* [[OVERFLOW_ARG_AREA2]], i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[VA_ARG_ADDR:%[^ ]+]] = phi %struct.agg_3byte** [ [[REG_ADDR]], %{{.*}} ], [ [[MEM_ADDR]], %{{.*}} ]
+// CHECK: [[INDIRECT_ARG:%[^ ]+]] = load %struct.agg_3byte*, %struct.agg_3byte** [[VA_ARG_ADDR]]
+// CHECK: ret void
 
 struct agg_4byte va_agg_4byte(__builtin_va_list l) { return __builtin_va_arg(l, struct agg_4byte); }
 // CHECK-LABEL: define void @va_agg_4byte(%struct.agg_4byte* noalias sret %{{.*}}, %struct.__va_list_tag* %{{.*}}
-// CHECK: %reg_offset = add i64 %scaled_reg_count, 20
-// CHECK: %raw_mem_addr = getelementptr i8, i8* %overflow_arg_area, i64 4
-// CHECK-NOT: %indirect_arg
+// CHECK: [[REG_COUNT_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 0
+// CHECK: [[REG_COUNT:%[^ ]+]] = load i64, i64* [[REG_COUNT_PTR]]
+// CHECK: [[FITS_IN_REGS:%[^ ]+]] = icmp ult i64 [[REG_COUNT]], 5
+// CHECK: br i1 [[FITS_IN_REGS]],
+// CHECK: [[SCALED_REG_COUNT:%[^ ]+]] = mul i64 [[REG_COUNT]], 8
+// CHECK: [[REG_OFFSET:%[^ ]+]] = add i64 [[SCALED_REG_COUNT]], 20
+// CHECK: [[REG_SAVE_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 3
+// CHECK: [[REG_SAVE_AREA:%[^ ]+]] = load i8*, i8** [[REG_SAVE_AREA_PTR:[^ ]+]]
+// CHECK: [[RAW_REG_ADDR:%[^ ]+]] = getelementptr i8, i8* [[REG_SAVE_AREA]], i64 [[REG_OFFSET]]
+// CHECK: [[REG_ADDR:%[^ ]+]] = bitcast i8* [[RAW_REG_ADDR]] to %struct.agg_4byte*
+// CHECK: [[REG_COUNT1:%[^ ]+]] = add i64 [[REG_COUNT]], 1
+// CHECK: store i64 [[REG_COUNT1]], i64* [[REG_COUNT_PTR]]
+// CHECK: [[OVERFLOW_ARG_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 2
+// CHECK: [[OVERFLOW_ARG_AREA:%[^ ]+]] = load i8*, i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[RAW_MEM_ADDR:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 4
+// CHECK: [[MEM_ADDR:%[^ ]+]] = bitcast i8* [[RAW_MEM_ADDR]] to %struct.agg_4byte*
+// CHECK: [[OVERFLOW_ARG_AREA2:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 8
+// CHECK: store i8* [[OVERFLOW_ARG_AREA2]], i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[VA_ARG_ADDR:%[^ ]+]] = phi %struct.agg_4byte* [ [[REG_ADDR]], %{{.*}} ], [ [[MEM_ADDR]], %{{.*}} ]
+// CHECK: ret void
 
 struct agg_8byte va_agg_8byte(__builtin_va_list l) { return __builtin_va_arg(l, struct agg_8byte); }
 // CHECK-LABEL: define void @va_agg_8byte(%struct.agg_8byte* noalias sret %{{.*}}, %struct.__va_list_tag* %{{.*}}
-// CHECK: %reg_offset = add i64 %scaled_reg_count, 16
-// CHECK: %raw_mem_addr = getelementptr i8, i8* %overflow_arg_area, i64 0
-// CHECK-NOT: %indirect_arg
+// CHECK: [[REG_COUNT_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 0
+// CHECK: [[REG_COUNT:%[^ ]+]] = load i64, i64* [[REG_COUNT_PTR]]
+// CHECK: [[FITS_IN_REGS:%[^ ]+]] = icmp ult i64 [[REG_COUNT]], 5
+// CHECK: br i1 [[FITS_IN_REGS]],
+// CHECK: [[SCALED_REG_COUNT:%[^ ]+]] = mul i64 [[REG_COUNT]], 8
+// CHECK: [[REG_OFFSET:%[^ ]+]] = add i64 [[SCALED_REG_COUNT]], 16
+// CHECK: [[REG_SAVE_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 3
+// CHECK: [[REG_SAVE_AREA:%[^ ]+]] = load i8*, i8** [[REG_SAVE_AREA_PTR:[^ ]+]]
+// CHECK: [[RAW_REG_ADDR:%[^ ]+]] = getelementptr i8, i8* [[REG_SAVE_AREA]], i64 [[REG_OFFSET]]
+// CHECK: [[REG_ADDR:%[^ ]+]] = bitcast i8* [[RAW_REG_ADDR]] to %struct.agg_8byte*
+// CHECK: [[REG_COUNT1:%[^ ]+]] = add i64 [[REG_COUNT]], 1
+// CHECK: store i64 [[REG_COUNT1]], i64* [[REG_COUNT_PTR]]
+// CHECK: [[OVERFLOW_ARG_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 2
+// CHECK: [[OVERFLOW_ARG_AREA:%[^ ]+]] = load i8*, i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[RAW_MEM_ADDR:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 0
+// CHECK: [[MEM_ADDR:%[^ ]+]] = bitcast i8* [[RAW_MEM_ADDR]] to %struct.agg_8byte*
+// CHECK: [[OVERFLOW_ARG_AREA2:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 8
+// CHECK: store i8* [[OVERFLOW_ARG_AREA2]], i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[VA_ARG_ADDR:%[^ ]+]] = phi %struct.agg_8byte* [ [[REG_ADDR]], %{{.*}} ], [ [[MEM_ADDR]], %{{.*}} ]
+// CHECK: ret void
 
 struct agg_float va_agg_float(__builtin_va_list l) { return __builtin_va_arg(l, struct agg_float); }
 // CHECK-LABEL: define void @va_agg_float(%struct.agg_float* noalias sret %{{.*}}, %struct.__va_list_tag* %{{.*}}
-// CHECK: %reg_offset = add i64 %scaled_reg_count, 128
-// CHECK: %raw_mem_addr = getelementptr i8, i8* %overflow_arg_area, i64 4
-// CHECK-NOT: %indirect_arg
+// CHECK: [[REG_COUNT_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 1
+// CHECK: [[REG_COUNT:%[^ ]+]] = load i64, i64* [[REG_COUNT_PTR]]
+// CHECK: [[FITS_IN_REGS:%[^ ]+]] = icmp ult i64 [[REG_COUNT]], 4
+// CHECK: br i1 [[FITS_IN_REGS]],
+// CHECK: [[SCALED_REG_COUNT:%[^ ]+]] = mul i64 [[REG_COUNT]], 8
+// CHECK: [[REG_OFFSET:%[^ ]+]] = add i64 [[SCALED_REG_COUNT]], 128
+// CHECK: [[REG_SAVE_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 3
+// CHECK: [[REG_SAVE_AREA:%[^ ]+]] = load i8*, i8** [[REG_SAVE_AREA_PTR:[^ ]+]]
+// CHECK: [[RAW_REG_ADDR:%[^ ]+]] = getelementptr i8, i8* [[REG_SAVE_AREA]], i64 [[REG_OFFSET]]
+// CHECK: [[REG_ADDR:%[^ ]+]] = bitcast i8* [[RAW_REG_ADDR]] to %struct.agg_float*
+// CHECK: [[REG_COUNT1:%[^ ]+]] = add i64 [[REG_COUNT]], 1
+// CHECK: store i64 [[REG_COUNT1]], i64* [[REG_COUNT_PTR]]
+// CHECK: [[OVERFLOW_ARG_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 2
+// CHECK: [[OVERFLOW_ARG_AREA:%[^ ]+]] = load i8*, i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[RAW_MEM_ADDR:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 4
+// CHECK: [[MEM_ADDR:%[^ ]+]] = bitcast i8* [[RAW_MEM_ADDR]] to %struct.agg_float*
+// CHECK: [[OVERFLOW_ARG_AREA2:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 8
+// CHECK: store i8* [[OVERFLOW_ARG_AREA2]], i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[VA_ARG_ADDR:%[^ ]+]] = phi %struct.agg_float* [ [[REG_ADDR]], %{{.*}} ], [ [[MEM_ADDR]], %{{.*}} ]
+// CHECK: ret void
 
 struct agg_double va_agg_double(__builtin_va_list l) { return __builtin_va_arg(l, struct agg_double); }
 // CHECK-LABEL: define void @va_agg_double(%struct.agg_double* noalias sret %{{.*}}, %struct.__va_list_tag* %{{.*}}
-// CHECK: %reg_offset = add i64 %scaled_reg_count, 128
-// CHECK: %raw_mem_addr = getelementptr i8, i8* %overflow_arg_area, i64 0
-// CHECK-NOT: %indirect_arg
+// CHECK: [[REG_COUNT_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 1
+// CHECK: [[REG_COUNT:%[^ ]+]] = load i64, i64* [[REG_COUNT_PTR]]
+// CHECK: [[FITS_IN_REGS:%[^ ]+]] = icmp ult i64 [[REG_COUNT]], 4
+// CHECK: br i1 [[FITS_IN_REGS]],
+// CHECK: [[SCALED_REG_COUNT:%[^ ]+]] = mul i64 [[REG_COUNT]], 8
+// CHECK: [[REG_OFFSET:%[^ ]+]] = add i64 [[SCALED_REG_COUNT]], 128
+// CHECK: [[REG_SAVE_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 3
+// CHECK: [[REG_SAVE_AREA:%[^ ]+]] = load i8*, i8** [[REG_SAVE_AREA_PTR:[^ ]+]]
+// CHECK: [[RAW_REG_ADDR:%[^ ]+]] = getelementptr i8, i8* [[REG_SAVE_AREA]], i64 [[REG_OFFSET]]
+// CHECK: [[REG_ADDR:%[^ ]+]] = bitcast i8* [[RAW_REG_ADDR]] to %struct.agg_double*
+// CHECK: [[REG_COUNT1:%[^ ]+]] = add i64 [[REG_COUNT]], 1
+// CHECK: store i64 [[REG_COUNT1]], i64* [[REG_COUNT_PTR]]
+// CHECK: [[OVERFLOW_ARG_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 2
+// CHECK: [[OVERFLOW_ARG_AREA:%[^ ]+]] = load i8*, i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[RAW_MEM_ADDR:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 0
+// CHECK: [[MEM_ADDR:%[^ ]+]] = bitcast i8* [[RAW_MEM_ADDR]] to %struct.agg_double*
+// CHECK: [[OVERFLOW_ARG_AREA2:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 8
+// CHECK: store i8* [[OVERFLOW_ARG_AREA2]], i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[VA_ARG_ADDR:%[^ ]+]] = phi %struct.agg_double* [ [[REG_ADDR]], %{{.*}} ], [ [[MEM_ADDR]], %{{.*}} ]
+// CHECK: ret void
 
 struct agg_longdouble va_agg_longdouble(__builtin_va_list l) { return __builtin_va_arg(l, struct agg_longdouble); }
 // CHECK-LABEL: define void @va_agg_longdouble(%struct.agg_longdouble* noalias sret %{{.*}}, %struct.__va_list_tag* %{{.*}}
-// CHECK: %reg_offset = add i64 %scaled_reg_count, 16
-// CHECK: %raw_mem_addr = getelementptr i8, i8* %overflow_arg_area, i64 0
-// CHECK: %indirect_arg = load %struct.agg_longdouble*
+// CHECK: [[REG_COUNT_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 0
+// CHECK: [[REG_COUNT:%[^ ]+]] = load i64, i64* [[REG_COUNT_PTR]]
+// CHECK: [[FITS_IN_REGS:%[^ ]+]] = icmp ult i64 [[REG_COUNT]], 5
+// CHECK: br i1 [[FITS_IN_REGS]],
+// CHECK: [[SCALED_REG_COUNT:%[^ ]+]] = mul i64 [[REG_COUNT]], 8
+// CHECK: [[REG_OFFSET:%[^ ]+]] = add i64 [[SCALED_REG_COUNT]], 16
+// CHECK: [[REG_SAVE_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 3
+// CHECK: [[REG_SAVE_AREA:%[^ ]+]] = load i8*, i8** [[REG_SAVE_AREA_PTR:[^ ]+]]
+// CHECK: [[RAW_REG_ADDR:%[^ ]+]] = getelementptr i8, i8* [[REG_SAVE_AREA]], i64 [[REG_OFFSET]]
+// CHECK: [[REG_ADDR:%[^ ]+]] = bitcast i8* [[RAW_REG_ADDR]] to %struct.agg_longdouble**
+// CHECK: [[REG_COUNT1:%[^ ]+]] = add i64 [[REG_COUNT]], 1
+// CHECK: store i64 [[REG_COUNT1]], i64* [[REG_COUNT_PTR]]
+// CHECK: [[OVERFLOW_ARG_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 2
+// CHECK: [[OVERFLOW_ARG_AREA:%[^ ]+]] = load i8*, i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[RAW_MEM_ADDR:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 0
+// CHECK: [[MEM_ADDR:%[^ ]+]] = bitcast i8* [[RAW_MEM_ADDR]] to %struct.agg_longdouble**
+// CHECK: [[OVERFLOW_ARG_AREA2:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 8
+// CHECK: store i8* [[OVERFLOW_ARG_AREA2]], i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[VA_ARG_ADDR:%[^ ]+]] = phi %struct.agg_longdouble** [ [[REG_ADDR]], %{{.*}} ], [ [[MEM_ADDR]], %{{.*}} ]
+// CHECK: [[INDIRECT_ARG:%[^ ]+]] = load %struct.agg_longdouble*, %struct.agg_longdouble** [[VA_ARG_ADDR]]
+// CHECK: ret void
 
 struct agg_float_a8 va_agg_float_a8(__builtin_va_list l) { return __builtin_va_arg(l, struct agg_float_a8); }
 // CHECK-LABEL: define void @va_agg_float_a8(%struct.agg_float_a8* noalias sret %{{.*}}, %struct.__va_list_tag* %{{.*}}
-// CHECK: %reg_offset = add i64 %scaled_reg_count, 128
-// CHECK: %raw_mem_addr = getelementptr i8, i8* %overflow_arg_area, i64 0
-// CHECK-NOT: %indirect_arg
+// CHECK: [[REG_COUNT_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 1
+// CHECK: [[REG_COUNT:%[^ ]+]] = load i64, i64* [[REG_COUNT_PTR]]
+// CHECK: [[FITS_IN_REGS:%[^ ]+]] = icmp ult i64 [[REG_COUNT]], 4
+// CHECK: br i1 [[FITS_IN_REGS]],
+// CHECK: [[SCALED_REG_COUNT:%[^ ]+]] = mul i64 [[REG_COUNT]], 8
+// CHECK: [[REG_OFFSET:%[^ ]+]] = add i64 [[SCALED_REG_COUNT]], 128
+// CHECK: [[REG_SAVE_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 3
+// CHECK: [[REG_SAVE_AREA:%[^ ]+]] = load i8*, i8** [[REG_SAVE_AREA_PTR:[^ ]+]]
+// CHECK: [[RAW_REG_ADDR:%[^ ]+]] = getelementptr i8, i8* [[REG_SAVE_AREA]], i64 [[REG_OFFSET]]
+// CHECK: [[REG_ADDR:%[^ ]+]] = bitcast i8* [[RAW_REG_ADDR]] to %struct.agg_float_a8*
+// CHECK: [[REG_COUNT1:%[^ ]+]] = add i64 [[REG_COUNT]], 1
+// CHECK: store i64 [[REG_COUNT1]], i64* [[REG_COUNT_PTR]]
+// CHECK: [[OVERFLOW_ARG_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 2
+// CHECK: [[OVERFLOW_ARG_AREA:%[^ ]+]] = load i8*, i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[RAW_MEM_ADDR:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 0
+// CHECK: [[MEM_ADDR:%[^ ]+]] = bitcast i8* [[RAW_MEM_ADDR]] to %struct.agg_float_a8*
+// CHECK: [[OVERFLOW_ARG_AREA2:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 8
+// CHECK: store i8* [[OVERFLOW_ARG_AREA2]], i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[VA_ARG_ADDR:%[^ ]+]] = phi %struct.agg_float_a8* [ [[REG_ADDR]], %{{.*}} ], [ [[MEM_ADDR]], %{{.*}} ]
+// CHECK: ret void
 
 struct agg_float_a16 va_agg_float_a16(__builtin_va_list l) { return __builtin_va_arg(l, struct agg_float_a16); }
 // CHECK-LABEL: define void @va_agg_float_a16(%struct.agg_float_a16* noalias sret %{{.*}}, %struct.__va_list_tag* %{{.*}}
-// CHECK: %reg_offset = add i64 %scaled_reg_count, 16
-// CHECK: %raw_mem_addr = getelementptr i8, i8* %overflow_arg_area, i64 0
-// CHECK: %indirect_arg = load %struct.agg_float_a16*
+// CHECK: [[REG_COUNT_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 0
+// CHECK: [[REG_COUNT:%[^ ]+]] = load i64, i64* [[REG_COUNT_PTR]]
+// CHECK: [[FITS_IN_REGS:%[^ ]+]] = icmp ult i64 [[REG_COUNT]], 5
+// CHECK: br i1 [[FITS_IN_REGS]],
+// CHECK: [[SCALED_REG_COUNT:%[^ ]+]] = mul i64 [[REG_COUNT]], 8
+// CHECK: [[REG_OFFSET:%[^ ]+]] = add i64 [[SCALED_REG_COUNT]], 16
+// CHECK: [[REG_SAVE_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 3
+// CHECK: [[REG_SAVE_AREA:%[^ ]+]] = load i8*, i8** [[REG_SAVE_AREA_PTR:[^ ]+]]
+// CHECK: [[RAW_REG_ADDR:%[^ ]+]] = getelementptr i8, i8* [[REG_SAVE_AREA]], i64 [[REG_OFFSET]]
+// CHECK: [[REG_ADDR:%[^ ]+]] = bitcast i8* [[RAW_REG_ADDR]] to %struct.agg_float_a16**
+// CHECK: [[REG_COUNT1:%[^ ]+]] = add i64 [[REG_COUNT]], 1
+// CHECK: store i64 [[REG_COUNT1]], i64* [[REG_COUNT_PTR]]
+// CHECK: [[OVERFLOW_ARG_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 2
+// CHECK: [[OVERFLOW_ARG_AREA:%[^ ]+]] = load i8*, i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[RAW_MEM_ADDR:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 0
+// CHECK: [[MEM_ADDR:%[^ ]+]] = bitcast i8* [[RAW_MEM_ADDR]] to %struct.agg_float_a16**
+// CHECK: [[OVERFLOW_ARG_AREA2:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 8
+// CHECK: store i8* [[OVERFLOW_ARG_AREA2]], i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[VA_ARG_ADDR:%[^ ]+]] = phi %struct.agg_float_a16** [ [[REG_ADDR]], %{{.*}} ], [ [[MEM_ADDR]], %{{.*}} ]
+// CHECK: [[INDIRECT_ARG:%[^ ]+]] = load %struct.agg_float_a16*, %struct.agg_float_a16** [[VA_ARG_ADDR]]
+// CHECK: ret void
 
 struct agg_nofloat1 va_agg_nofloat1(__builtin_va_list l) { return __builtin_va_arg(l, struct agg_nofloat1); }
 // CHECK-LABEL: define void @va_agg_nofloat1(%struct.agg_nofloat1* noalias sret %{{.*}}, %struct.__va_list_tag* %{{.*}}
-// CHECK: %reg_offset = add i64 %scaled_reg_count, 16
-// CHECK: %raw_mem_addr = getelementptr i8, i8* %overflow_arg_area, i64 0
-// CHECK-NOT: %indirect_arg
+// CHECK: [[REG_COUNT_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 0
+// CHECK: [[REG_COUNT:%[^ ]+]] = load i64, i64* [[REG_COUNT_PTR]]
+// CHECK: [[FITS_IN_REGS:%[^ ]+]] = icmp ult i64 [[REG_COUNT]], 5
+// CHECK: br i1 [[FITS_IN_REGS]],
+// CHECK: [[SCALED_REG_COUNT:%[^ ]+]] = mul i64 [[REG_COUNT]], 8
+// CHECK: [[REG_OFFSET:%[^ ]+]] = add i64 [[SCALED_REG_COUNT]], 16
+// CHECK: [[REG_SAVE_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 3
+// CHECK: [[REG_SAVE_AREA:%[^ ]+]] = load i8*, i8** [[REG_SAVE_AREA_PTR:[^ ]+]]
+// CHECK: [[RAW_REG_ADDR:%[^ ]+]] = getelementptr i8, i8* [[REG_SAVE_AREA]], i64 [[REG_OFFSET]]
+// CHECK: [[REG_ADDR:%[^ ]+]] = bitcast i8* [[RAW_REG_ADDR]] to %struct.agg_nofloat1*
+// CHECK: [[REG_COUNT1:%[^ ]+]] = add i64 [[REG_COUNT]], 1
+// CHECK: store i64 [[REG_COUNT1]], i64* [[REG_COUNT_PTR]]
+// CHECK: [[OVERFLOW_ARG_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 2
+// CHECK: [[OVERFLOW_ARG_AREA:%[^ ]+]] = load i8*, i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[RAW_MEM_ADDR:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 0
+// CHECK: [[MEM_ADDR:%[^ ]+]] = bitcast i8* [[RAW_MEM_ADDR]] to %struct.agg_nofloat1*
+// CHECK: [[OVERFLOW_ARG_AREA2:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 8
+// CHECK: store i8* [[OVERFLOW_ARG_AREA2]], i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[VA_ARG_ADDR:%[^ ]+]] = phi %struct.agg_nofloat1* [ [[REG_ADDR]], %{{.*}} ], [ [[MEM_ADDR]], %{{.*}} ]
+// CHECK: ret void
 
 struct agg_nofloat2 va_agg_nofloat2(__builtin_va_list l) { return __builtin_va_arg(l, struct agg_nofloat2); }
 // CHECK-LABEL: define void @va_agg_nofloat2(%struct.agg_nofloat2* noalias sret %{{.*}}, %struct.__va_list_tag* %{{.*}}
-// CHECK: %reg_offset = add i64 %scaled_reg_count, 16
-// CHECK: %raw_mem_addr = getelementptr i8, i8* %overflow_arg_area, i64 0
-// CHECK-NOT: %indirect_arg
+// CHECK: [[REG_COUNT_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 0
+// CHECK: [[REG_COUNT:%[^ ]+]] = load i64, i64* [[REG_COUNT_PTR]]
+// CHECK: [[FITS_IN_REGS:%[^ ]+]] = icmp ult i64 [[REG_COUNT]], 5
+// CHECK: br i1 [[FITS_IN_REGS]],
+// CHECK: [[SCALED_REG_COUNT:%[^ ]+]] = mul i64 [[REG_COUNT]], 8
+// CHECK: [[REG_OFFSET:%[^ ]+]] = add i64 [[SCALED_REG_COUNT]], 16
+// CHECK: [[REG_SAVE_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 3
+// CHECK: [[REG_SAVE_AREA:%[^ ]+]] = load i8*, i8** [[REG_SAVE_AREA_PTR:[^ ]+]]
+// CHECK: [[RAW_REG_ADDR:%[^ ]+]] = getelementptr i8, i8* [[REG_SAVE_AREA]], i64 [[REG_OFFSET]]
+// CHECK: [[REG_ADDR:%[^ ]+]] = bitcast i8* [[RAW_REG_ADDR]] to %struct.agg_nofloat2*
+// CHECK: [[REG_COUNT1:%[^ ]+]] = add i64 [[REG_COUNT]], 1
+// CHECK: store i64 [[REG_COUNT1]], i64* [[REG_COUNT_PTR]]
+// CHECK: [[OVERFLOW_ARG_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 2
+// CHECK: [[OVERFLOW_ARG_AREA:%[^ ]+]] = load i8*, i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[RAW_MEM_ADDR:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 0
+// CHECK: [[MEM_ADDR:%[^ ]+]] = bitcast i8* [[RAW_MEM_ADDR]] to %struct.agg_nofloat2*
+// CHECK: [[OVERFLOW_ARG_AREA2:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 8
+// CHECK: store i8* [[OVERFLOW_ARG_AREA2]], i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[VA_ARG_ADDR:%[^ ]+]] = phi %struct.agg_nofloat2* [ [[REG_ADDR]], %{{.*}} ], [ [[MEM_ADDR]], %{{.*}} ]
+// CHECK: ret void
 
 struct agg_nofloat3 va_agg_nofloat3(__builtin_va_list l) { return __builtin_va_arg(l, struct agg_nofloat3); }
 // CHECK-LABEL: define void @va_agg_nofloat3(%struct.agg_nofloat3* noalias sret %{{.*}}, %struct.__va_list_tag* %{{.*}}
-// CHECK: %reg_offset = add i64 %scaled_reg_count, 20
-// CHECK: %raw_mem_addr = getelementptr i8, i8* %overflow_arg_area, i64 4
-// CHECK-NOT: %indirect_arg
+// CHECK: [[REG_COUNT_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 0
+// CHECK: [[REG_COUNT:%[^ ]+]] = load i64, i64* [[REG_COUNT_PTR]]
+// CHECK: [[FITS_IN_REGS:%[^ ]+]] = icmp ult i64 [[REG_COUNT]], 5
+// CHECK: br i1 [[FITS_IN_REGS]],
+// CHECK: [[SCALED_REG_COUNT:%[^ ]+]] = mul i64 [[REG_COUNT]], 8
+// CHECK: [[REG_OFFSET:%[^ ]+]] = add i64 [[SCALED_REG_COUNT]], 20
+// CHECK: [[REG_SAVE_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 3
+// CHECK: [[REG_SAVE_AREA:%[^ ]+]] = load i8*, i8** [[REG_SAVE_AREA_PTR:[^ ]+]]
+// CHECK: [[RAW_REG_ADDR:%[^ ]+]] = getelementptr i8, i8* [[REG_SAVE_AREA]], i64 [[REG_OFFSET]]
+// CHECK: [[REG_ADDR:%[^ ]+]] = bitcast i8* [[RAW_REG_ADDR]] to %struct.agg_nofloat3*
+// CHECK: [[REG_COUNT1:%[^ ]+]] = add i64 [[REG_COUNT]], 1
+// CHECK: store i64 [[REG_COUNT1]], i64* [[REG_COUNT_PTR]]
+// CHECK: [[OVERFLOW_ARG_AREA_PTR:%[^ ]+]] = getelementptr inbounds %struct.__va_list_tag, %struct.__va_list_tag* %{{.*}}, i32 0, i32 2
+// CHECK: [[OVERFLOW_ARG_AREA:%[^ ]+]] = load i8*, i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[RAW_MEM_ADDR:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 4
+// CHECK: [[MEM_ADDR:%[^ ]+]] = bitcast i8* [[RAW_MEM_ADDR]] to %struct.agg_nofloat3*
+// CHECK: [[OVERFLOW_ARG_AREA2:%[^ ]+]] = getelementptr i8, i8* [[OVERFLOW_ARG_AREA]], i64 8
+// CHECK: store i8* [[OVERFLOW_ARG_AREA2]], i8** [[OVERFLOW_ARG_AREA_PTR]]
+// CHECK: [[VA_ARG_ADDR:%[^ ]+]] = phi %struct.agg_nofloat3* [ [[REG_ADDR]], %{{.*}} ], [ [[MEM_ADDR]], %{{.*}} ]
+// CHECK: ret void