clang/test/CodeGen/arm-aapcs-vfp.c

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