1 // REQUIRES: arm-registered-target 2 // REQUIRES: aarch64-registered-target 3 // RUN: %clang_cc1 -triple thumbv7-apple-darwin9 \ 4 // RUN: -target-abi aapcs \ 5 // RUN: -target-cpu cortex-a8 \ 6 // RUN: -mfloat-abi hard \ 7 // RUN: -ffreestanding \ 8 // RUN: -emit-llvm -w -o - %s | FileCheck %s 9 10 // RUN: %clang_cc1 -triple armv7-unknown-nacl-gnueabi \ 11 // RUN: -target-cpu cortex-a8 \ 12 // RUN: -mfloat-abi hard \ 13 // RUN: -ffreestanding \ 14 // RUN: -emit-llvm -w -o - %s | FileCheck %s 15 16 // RUN: %clang_cc1 -triple arm64-apple-darwin9 -target-feature +neon \ 17 // RUN: -ffreestanding \ 18 // RUN: -emit-llvm -w -o - %s | FileCheck -check-prefix=CHECK64 %s 19 20 #ifdef __arm64__ 21 #include <arm_neon.h> 22 #else 23 #include <arm_neon.h> 24 #endif 25 26 struct homogeneous_struct { 27 float f[2]; 28 float f3; 29 float f4; 30 }; 31 // CHECK: define arm_aapcs_vfpcc %struct.homogeneous_struct @test_struct(%struct.homogeneous_struct %{{.*}}) 32 // CHECK64: define %struct.homogeneous_struct @test_struct([4 x float] %{{.*}}) 33 extern struct homogeneous_struct struct_callee(struct homogeneous_struct); 34 struct homogeneous_struct test_struct(struct homogeneous_struct arg) { 35 return struct_callee(arg); 36 } 37 38 // CHECK: define arm_aapcs_vfpcc void @test_struct_variadic(%struct.homogeneous_struct* {{.*}}, ...) 39 struct homogeneous_struct test_struct_variadic(struct homogeneous_struct arg, ...) { 40 return struct_callee(arg); 41 } 42 43 struct nested_array { 44 double d[4]; 45 }; 46 // CHECK: define arm_aapcs_vfpcc void @test_array(%struct.nested_array %{{.*}}) 47 // CHECK64: define void @test_array([4 x double] %{{.*}}) 48 extern void array_callee(struct nested_array); 49 void test_array(struct nested_array arg) { 50 array_callee(arg); 51 } 52 53 extern void complex_callee(__complex__ double); 54 // CHECK: define arm_aapcs_vfpcc void @test_complex({ double, double } %{{.*}}) 55 // CHECK64: define void @test_complex([2 x double] %cd.coerce) 56 void test_complex(__complex__ double cd) { 57 complex_callee(cd); 58 } 59 60 // Long double is the same as double on AAPCS, it should be homogeneous. 61 extern void complex_long_callee(__complex__ long double); 62 // CHECK: define arm_aapcs_vfpcc void @test_complex_long({ double, double } %{{.*}}) 63 void test_complex_long(__complex__ long double cd) { 64 complex_callee(cd); 65 } 66 67 // Structs with more than 4 elements of the base type are not treated 68 // as homogeneous aggregates. Test that. 69 70 struct big_struct { 71 float f1; 72 float f[2]; 73 float f3; 74 float f4; 75 }; 76 // CHECK: define arm_aapcs_vfpcc void @test_big([5 x i32] %{{.*}}) 77 // CHECK64: define void @test_big(%struct.big_struct* %{{.*}}) 78 // CHECK64: call void @llvm.memcpy 79 // CHECK64: call void @big_callee(%struct.big_struct* 80 extern void big_callee(struct big_struct); 81 void test_big(struct big_struct arg) { 82 big_callee(arg); 83 } 84 85 // Make sure that aggregates with multiple base types are not treated as 86 // homogeneous aggregates. 87 88 struct heterogeneous_struct { 89 float f1; 90 int i2; 91 }; 92 // CHECK: define arm_aapcs_vfpcc void @test_hetero([2 x i32] %{{.*}}) 93 // CHECK64: define void @test_hetero(i64 %{{.*}}) 94 extern void hetero_callee(struct heterogeneous_struct); 95 void test_hetero(struct heterogeneous_struct arg) { 96 hetero_callee(arg); 97 } 98 99 // Neon multi-vector types are homogeneous aggregates. 100 // CHECK: define arm_aapcs_vfpcc <16 x i8> @f0(%struct.int8x16x4_t %{{.*}}) 101 // CHECK64: define <16 x i8> @f0([4 x <16 x i8>] %{{.*}}) 102 int8x16_t f0(int8x16x4_t v4) { 103 return vaddq_s8(v4.val[0], v4.val[3]); 104 } 105 106 // ...and it doesn't matter whether the vectors are exactly the same, as long 107 // as they have the same size. 108 109 struct neon_struct { 110 int8x8x2_t v12; 111 int32x2_t v3; 112 int16x4_t v4; 113 }; 114 // CHECK: define arm_aapcs_vfpcc void @test_neon(%struct.neon_struct %{{.*}}) 115 // CHECK64: define void @test_neon([4 x <8 x i8>] %{{.*}}) 116 extern void neon_callee(struct neon_struct); 117 void test_neon(struct neon_struct arg) { 118 neon_callee(arg); 119 } 120 121 // CHECK-LABEL: define arm_aapcs_vfpcc void @f33(%struct.s33* byval align 4 %s) 122 struct s33 { char buf[32*32]; }; 123 void f33(struct s33 s) { } 124 125 typedef struct { long long x; int y; } struct_long_long_int; 126 // CHECK: define arm_aapcs_vfpcc void @test_vfp_stack_gpr_split_1(double %a, double %b, double %c, double %d, double %e, double %f, double %g, double %h, double %i, i32 %j, i64 %k, i32 %l) 127 void test_vfp_stack_gpr_split_1(double a, double b, double c, double d, double e, double f, double g, double h, double i, int j, long long k, int l) {} 128 129 // CHECK: define arm_aapcs_vfpcc void @test_vfp_stack_gpr_split_2(double %a, double %b, double %c, double %d, double %e, double %f, double %g, double %h, double %i, i32 %j, [2 x i64] %k.coerce) 130 void test_vfp_stack_gpr_split_2(double a, double b, double c, double d, double e, double f, double g, double h, double i, int j, struct_long_long_int k) {} 131 132 // CHECK: define arm_aapcs_vfpcc void @test_vfp_stack_gpr_split_3(%struct.struct_long_long_int* noalias sret %agg.result, double %a, double %b, double %c, double %d, double %e, double %f, double %g, double %h, double %i, [2 x i64] %k.coerce) 133 struct_long_long_int test_vfp_stack_gpr_split_3(double a, double b, double c, double d, double e, double f, double g, double h, double i, struct_long_long_int k) {} 134 135 typedef struct { int a; int b:4; int c; } struct_int_bitfield_int; 136 // CHECK: define arm_aapcs_vfpcc void @test_test_vfp_stack_gpr_split_bitfield(double %a, double %b, double %c, double %d, double %e, double %f, double %g, double %h, double %i, i32 %j, i32 %k, [3 x i32] %l.coerce) 137 void test_test_vfp_stack_gpr_split_bitfield(double a, double b, double c, double d, double e, double f, double g, double h, double i, int j, int k, struct_int_bitfield_int l) {} 138 139 // Note: this struct requires internal padding 140 typedef struct { int x; long long y; } struct_int_long_long; 141 // CHECK: define arm_aapcs_vfpcc void @test_vfp_stack_gpr_split_4(double %a, double %b, double %c, double %d, double %e, double %f, double %g, double %h, double %i, i32 %j, [2 x i64] %k.coerce) 142 void test_vfp_stack_gpr_split_4(double a, double b, double c, double d, double e, double f, double g, double h, double i, int j, struct_int_long_long k) {} 143 144 // This very large struct (passed byval) uses up the GPRs, so no padding is needed 145 typedef struct { int x[17]; } struct_seventeen_ints; 146 typedef struct { int x[4]; } struct_four_ints; 147 // CHECK: define arm_aapcs_vfpcc void @test_vfp_stack_gpr_split_5(%struct.struct_seventeen_ints* byval align 4 %a, double %b, double %c, double %d, double %e, double %f, double %g, double %h, double %i, double %j, [4 x i32] %k.coerce) 148 void test_vfp_stack_gpr_split_5(struct_seventeen_ints a, double b, double c, double d, double e, double f, double g, double h, double i, double j, struct_four_ints k) {} 149 150 // Here, parameter k would need padding to prevent it from being split, but it 151 // is passed ByVal (due to being > 64 bytes), so the backend handles this instead. 152 void test_vfp_stack_gpr_split_6(double a, double b, double c, double d, double e, double f, double g, double h, double i, int j, struct_seventeen_ints k) {} 153 // CHECK: define arm_aapcs_vfpcc void @test_vfp_stack_gpr_split_6(double %a, double %b, double %c, double %d, double %e, double %f, double %g, double %h, double %i, i32 %j, %struct.struct_seventeen_ints* byval align 4 %k) 154
