clang/test/CodeGen/arm-abi-vector.c

   1 // RUN: %clang_cc1 -triple armv7-apple-darwin -target-abi aapcs -emit-llvm -o - %s | FileCheck %s
   2 // RUN: %clang_cc1 -triple armv7-apple-darwin -target-abi apcs-gnu -emit-llvm -o - %s | FileCheck -check-prefix=APCS-GNU %s
   3 // RUN: %clang_cc1 -triple arm-linux-androideabi -emit-llvm -o - %s | FileCheck -check-prefix=ANDROID %s
   4
   5 #include <stdarg.h>
   6
   7 typedef __attribute__(( ext_vector_type(2) ))  int __int2;
   8 typedef __attribute__(( ext_vector_type(3) ))  char __char3;
   9 typedef __attribute__(( ext_vector_type(5) ))  char __char5;
  10 typedef __attribute__(( ext_vector_type(9) ))  char __char9;
  11 typedef __attribute__(( ext_vector_type(19) )) char __char19;
  12 typedef __attribute__(( ext_vector_type(3) ))  short __short3;
  13 typedef __attribute__(( ext_vector_type(5) ))  short __short5;
  14
  15 // Passing legal vector types as varargs.
  16 double varargs_vec_2i(int fixed, ...) {
  17 // CHECK: varargs_vec_2i
  18 // CHECK: [[VAR:%.*]] = alloca <2 x i32>, align 8
  19 // CHECK: [[AP_ALIGN:%.*]] = call ptr @llvm.ptrmask.p0.i32(ptr {{%.*}}, i32 -8)
  20 // CHECK: [[AP_NEXT:%.*]] = getelementptr inbounds i8, ptr [[AP_ALIGN]], i32 8
  21 // CHECK: [[VEC:%.*]] = load <2 x i32>, ptr [[AP_ALIGN]], align 8
  22 // CHECK: store <2 x i32> [[VEC]], ptr [[VAR]], align 8
  23 // APCS-GNU: varargs_vec_2i
  24 // APCS-GNU: [[VAR:%.*]] = alloca <2 x i32>, align 8
  25 // APCS-GNU: [[AP:%.*]] = load ptr,
  26 // APCS-GNU: [[AP_NEXT:%.*]] = getelementptr inbounds i8, ptr [[AP]], i32 8
  27 // APCS-GNU: [[VEC:%.*]] = load <2 x i32>, ptr [[AP]], align 4
  28 // APCS-GNU: store <2 x i32> [[VEC]], ptr [[VAR]], align 8
  29 // ANDROID: varargs_vec_2i
  30 // ANDROID: [[VAR:%.*]] = alloca <2 x i32>, align 8
  31 // ANDROID: [[AP_ALIGN:%.*]] = call ptr @llvm.ptrmask.p0.i32(ptr {{%.*}}, i32 -8)
  32 // ANDROID: [[AP_NEXT:%.*]] = getelementptr inbounds i8, ptr [[AP_ALIGN]], i32 8
  33 // ANDROID: [[VEC:%.*]] = load <2 x i32>, ptr [[AP_ALIGN]], align 8
  34 // ANDROID: store <2 x i32> [[VEC]], ptr [[VAR]], align 8
  35   va_list ap;
  36   double sum = fixed;
  37   va_start(ap, fixed);
  38   __int2 c3 = va_arg(ap, __int2);
  39   sum = sum + c3.x + c3.y;
  40   va_end(ap);
  41   return sum;
  42 }
  43
  44 double test_2i(__int2 *in) {
  45 // CHECK: test_2i
  46 // CHECK: call arm_aapcscc double (i32, ...) @varargs_vec_2i(i32 noundef 3, <2 x i32> noundef {{%.*}})
  47 // APCS-GNU: test_2i
  48 // APCS-GNU: call double (i32, ...) @varargs_vec_2i(i32 noundef 3, <2 x i32> noundef {{%.*}})
  49 // ANDROID: test_2i
  50 // ANDROID: call double (i32, ...) @varargs_vec_2i(i32 noundef 3, <2 x i32> noundef {{%.*}})
  51   return varargs_vec_2i(3, *in);
  52 }
  53
  54 double varargs_vec_3c(int fixed, ...) {
  55 // CHECK: varargs_vec_3c
  56 // CHECK: alloca <3 x i8>, align 4
  57 // CHECK: [[AP_NEXT:%.*]] = getelementptr inbounds i8, ptr [[AP:%.*]], i32 4
  58 // APCS-GNU: varargs_vec_3c
  59 // APCS-GNU: alloca <3 x i8>, align 4
  60 // APCS-GNU: [[AP_NEXT:%.*]] = getelementptr inbounds i8, ptr [[AP:%.*]], i32 4
  61 // ANDROID: varargs_vec_3c
  62 // ANDROID: alloca <3 x i8>, align 4
  63 // ANDROID: [[AP_NEXT:%.*]] = getelementptr inbounds i8, ptr [[AP:%.*]], i32 4
  64   va_list ap;
  65   double sum = fixed;
  66   va_start(ap, fixed);
  67   __char3 c3 = va_arg(ap, __char3);
  68   sum = sum + c3.x + c3.y;
  69   va_end(ap);
  70   return sum;
  71 }
  72
  73 double test_3c(__char3 *in) {
  74 // CHECK: test_3c
  75 // CHECK: call arm_aapcscc double (i32, ...) @varargs_vec_3c(i32 noundef 3, i32 {{%.*}})
  76 // APCS-GNU: test_3c
  77 // APCS-GNU: call double (i32, ...) @varargs_vec_3c(i32 noundef 3, i32 {{%.*}})
  78 // ANDROID: test_3c
  79 // ANDROID: call double (i32, ...) @varargs_vec_3c(i32 noundef 3, <3 x i8> noundef {{%.*}})
  80   return varargs_vec_3c(3, *in);
  81 }
  82
  83 double varargs_vec_5c(int fixed, ...) {
  84 // CHECK: varargs_vec_5c
  85 // CHECK: [[VAR:%.*]] = alloca <5 x i8>, align 8
  86 // CHECK: [[AP_ALIGN:%.*]] = call ptr @llvm.ptrmask.p0.i32(ptr {{%.*}}, i32 -8)
  87 // CHECK: [[AP_NEXT:%.*]] = getelementptr inbounds i8, ptr [[AP_ALIGN]], i32 8
  88 // CHECK: [[VEC:%.*]] = load <5 x i8>, ptr [[AP_ALIGN]], align 8
  89 // CHECK: store <5 x i8> [[VEC]], ptr [[VAR]], align 8
  90 // APCS-GNU: varargs_vec_5c
  91 // APCS-GNU: [[VAR:%.*]] = alloca <5 x i8>, align 8
  92 // APCS-GNU: [[AP:%.*]] = load ptr,
  93 // APCS-GNU: [[AP_NEXT:%.*]] = getelementptr inbounds i8, ptr [[AP]], i32 8
  94 // APCS-GNU: [[VEC:%.*]] = load <5 x i8>, ptr [[AP]], align 4
  95 // APCS-GNU: store <5 x i8> [[VEC]], ptr [[VAR]], align 8
  96 // ANDROID: varargs_vec_5c
  97 // ANDROID: [[VAR:%.*]] = alloca <5 x i8>, align 8
  98 // ANDROID: [[AP_ALIGN:%.*]] = call ptr @llvm.ptrmask.p0.i32(ptr {{%.*}}, i32 -8)
  99 // ANDROID: [[AP_NEXT:%.*]] = getelementptr inbounds i8, ptr [[AP_ALIGN]], i32 8
 100 // ANDROID: [[VEC:%.*]] = load <5 x i8>, ptr [[AP_ALIGN]], align 8
 101 // ANDROID: store <5 x i8> [[VEC]], ptr [[VAR]], align 8
 102   va_list ap;
 103   double sum = fixed;
 104   va_start(ap, fixed);
 105   __char5 c5 = va_arg(ap, __char5);
 106   sum = sum + c5.x + c5.y;
 107   va_end(ap);
 108   return sum;
 109 }
 110
 111 double test_5c(__char5 *in) {
 112 // CHECK: test_5c
 113 // CHECK: call arm_aapcscc double (i32, ...) @varargs_vec_5c(i32 noundef 5, <2 x i32> {{%.*}})
 114 // APCS-GNU: test_5c
 115 // APCS-GNU: call double (i32, ...) @varargs_vec_5c(i32 noundef 5, <2 x i32> {{%.*}})
 116 // ANDROID: test_5c
 117 // ANDROID: call double (i32, ...) @varargs_vec_5c(i32 noundef 5, <2 x i32> {{%.*}})
 118   return varargs_vec_5c(5, *in);
 119 }
 120
 121 double varargs_vec_9c(int fixed, ...) {
 122 // CHECK: varargs_vec_9c
 123 // CHECK: [[VAR:%.*]] = alloca <9 x i8>, align 16
 124 // CHECK: [[AP_ALIGN:%.*]] = call ptr @llvm.ptrmask.p0.i32(ptr {{%.*}}, i32 -8)
 125 // CHECK: [[AP_NEXT:%.*]] = getelementptr inbounds i8, ptr [[AP_ALIGN]], i32 16
 126 // CHECK: [[T0:%.*]] = load <9 x i8>, ptr [[AP_ALIGN]], align 8
 127 // CHECK: store <9 x i8> [[T0]], ptr [[VAR]], align 16
 128 // APCS-GNU: varargs_vec_9c
 129 // APCS-GNU: [[VAR:%.*]] = alloca <9 x i8>, align 16
 130 // APCS-GNU: [[AP:%.*]] = load ptr,
 131 // APCS-GNU: [[AP_NEXT:%.*]] = getelementptr inbounds i8, ptr [[AP]], i32 16
 132 // APCS-GNU: [[VEC:%.*]] = load <9 x i8>, ptr [[AP]], align 4
 133 // APCS-GNU: store <9 x i8> [[VEC]], ptr [[VAR]], align 16
 134 // ANDROID: varargs_vec_9c
 135 // ANDROID: [[VAR:%.*]] = alloca <9 x i8>, align 16
 136 // ANDROID: [[AP_ALIGN:%.*]] = call ptr @llvm.ptrmask.p0.i32(ptr {{%.*}}, i32 -8)
 137 // ANDROID: [[AP_NEXT:%.*]] = getelementptr inbounds i8, ptr [[AP_ALIGN]], i32 16
 138 // ANDROID: [[T0:%.*]] = load <9 x i8>, ptr [[AP_ALIGN]], align 8
 139 // ANDROID: store <9 x i8> [[T0]], ptr [[VAR]], align 16
 140   va_list ap;
 141   double sum = fixed;
 142   va_start(ap, fixed);
 143   __char9 c9 = va_arg(ap, __char9);
 144   sum = sum + c9.x + c9.y;
 145   va_end(ap);
 146   return sum;
 147 }
 148
 149 double test_9c(__char9 *in) {
 150 // CHECK: test_9c
 151 // CHECK: call arm_aapcscc double (i32, ...) @varargs_vec_9c(i32 noundef 9, <4 x i32> {{%.*}})
 152 // APCS-GNU: test_9c
 153 // APCS-GNU: call double (i32, ...) @varargs_vec_9c(i32 noundef 9, <4 x i32> {{%.*}})
 154 // ANDROID: test_9c
 155 // ANDROID: call double (i32, ...) @varargs_vec_9c(i32 noundef 9, <4 x i32> {{%.*}})
 156   return varargs_vec_9c(9, *in);
 157 }
 158
 159 double varargs_vec_19c(int fixed, ...) {
 160 // CHECK: varargs_vec_19c
 161 // CHECK: [[AP_NEXT:%.*]] = getelementptr inbounds i8, ptr [[AP:%.*]], i32 4
 162 // CHECK: [[VAR2:%.*]] = load ptr, ptr [[AP]]
 163 // APCS-GNU: varargs_vec_19c
 164 // APCS-GNU: [[AP_NEXT:%.*]] = getelementptr inbounds i8, ptr [[AP:%.*]], i32 4
 165 // APCS-GNU: [[VAR2:%.*]] = load ptr, ptr [[AP]]
 166 // ANDROID: varargs_vec_19c
 167 // ANDROID: [[AP_NEXT:%.*]] = getelementptr inbounds i8, ptr [[AP:%.*]], i32 4
 168 // ANDROID: [[VAR2:%.*]] = load ptr, ptr [[AP]]
 169   va_list ap;
 170   double sum = fixed;
 171   va_start(ap, fixed);
 172   __char19 c19 = va_arg(ap, __char19);
 173   sum = sum + c19.x + c19.y;
 174   va_end(ap);
 175   return sum;
 176 }
 177
 178 double test_19c(__char19 *in) {
 179 // CHECK: test_19c
 180 // CHECK: call arm_aapcscc double (i32, ...) @varargs_vec_19c(i32 noundef 19, ptr noundef {{%.*}})
 181 // APCS-GNU: test_19c
 182 // APCS-GNU: call double (i32, ...) @varargs_vec_19c(i32 noundef 19, ptr noundef {{%.*}})
 183 // ANDROID: test_19c
 184 // ANDROID: call double (i32, ...) @varargs_vec_19c(i32 noundef 19, ptr noundef {{%.*}})
 185   return varargs_vec_19c(19, *in);
 186 }
 187
 188 double varargs_vec_3s(int fixed, ...) {
 189 // CHECK: varargs_vec_3s
 190 // CHECK: alloca <3 x i16>, align 8
 191 // CHECK: [[AP_ALIGN:%.*]] = call ptr @llvm.ptrmask.p0.i32(ptr {{%.*}}, i32 -8)
 192 // CHECK: [[AP_NEXT:%.*]] = getelementptr inbounds i8, ptr [[AP_ALIGN]], i32 8
 193 // APCS-GNU: varargs_vec_3s
 194 // APCS-GNU: [[VAR:%.*]] = alloca <3 x i16>, align 8
 195 // APCS-GNU: [[AP:%.*]] = load ptr,
 196 // APCS-GNU: [[AP_NEXT:%.*]] = getelementptr inbounds i8, ptr [[AP]], i32 8
 197 // APCS-GNU: [[VEC:%.*]] = load <4 x i16>, ptr [[AP]], align 4
 198 // ANDROID: varargs_vec_3s
 199 // ANDROID: alloca <3 x i16>, align 8
 200 // ANDROID: [[AP_ALIGN:%.*]] = call ptr @llvm.ptrmask.p0.i32(ptr {{%.*}}, i32 -8)
 201 // ANDROID: [[AP_NEXT:%.*]] = getelementptr inbounds i8, ptr [[AP_ALIGN]], i32 8
 202   va_list ap;
 203   double sum = fixed;
 204   va_start(ap, fixed);
 205   __short3 c3 = va_arg(ap, __short3);
 206   sum = sum + c3.x + c3.y;
 207   va_end(ap);
 208   return sum;
 209 }
 210
 211 double test_3s(__short3 *in) {
 212 // CHECK: test_3s
 213 // CHECK: call arm_aapcscc double (i32, ...) @varargs_vec_3s(i32 noundef 3, <2 x i32> {{%.*}})
 214 // APCS-GNU: test_3s
 215 // APCS-GNU: call double (i32, ...) @varargs_vec_3s(i32 noundef 3, <2 x i32> {{%.*}})
 216 // ANDROID: test_3s
 217 // ANDROID: call double (i32, ...) @varargs_vec_3s(i32 noundef 3, <3 x i16> noundef {{%.*}})
 218   return varargs_vec_3s(3, *in);
 219 }
 220
 221 double varargs_vec_5s(int fixed, ...) {
 222 // CHECK: varargs_vec_5s
 223 // CHECK: [[VAR_ALIGN:%.*]] = alloca <5 x i16>, align 16
 224 // CHECK: [[AP_ALIGN:%.*]] = call ptr @llvm.ptrmask.p0.i32(ptr {{%.*}}, i32 -8)
 225 // CHECK: [[AP_NEXT:%.*]] = getelementptr inbounds i8, ptr [[AP_ALIGN]], i32 16
 226 // CHECK: [[VEC:%.*]] = load <5 x i16>, ptr [[AP_ALIGN]], align 8
 227 // CHECK: store <5 x i16> [[VEC]], ptr [[VAR_ALIGN]], align 16
 228 // APCS-GNU: varargs_vec_5s
 229 // APCS-GNU: [[VAR:%.*]] = alloca <5 x i16>, align 16
 230 // APCS-GNU: [[AP:%.*]] = load ptr,
 231 // APCS-GNU: [[AP_NEXT:%.*]] = getelementptr inbounds i8, ptr [[AP]], i32 16
 232 // APCS-GNU: [[VEC:%.*]] = load <5 x i16>, ptr [[AP]], align 4
 233 // ANDROID: varargs_vec_5s
 234 // ANDROID: [[VAR_ALIGN:%.*]] = alloca <5 x i16>, align 16
 235 // ANDROID: [[AP_ALIGN:%.*]] = call ptr @llvm.ptrmask.p0.i32(ptr {{%.*}}, i32 -8)
 236 // ANDROID: [[AP_NEXT:%.*]] = getelementptr inbounds i8, ptr [[AP_ALIGN]], i32 16
 237 // ANDROID: [[VEC:%.*]] = load <5 x i16>, ptr [[AP_ALIGN]], align 8
 238 // ANDROID: store <5 x i16> [[VEC]], ptr [[VAR_ALIGN]], align 16
 239   va_list ap;
 240   double sum = fixed;
 241   va_start(ap, fixed);
 242   __short5 c5 = va_arg(ap, __short5);
 243   sum = sum + c5.x + c5.y;
 244   va_end(ap);
 245   return sum;
 246 }
 247
 248 double test_5s(__short5 *in) {
 249 // CHECK: test_5s
 250 // CHECK: call arm_aapcscc double (i32, ...) @varargs_vec_5s(i32 noundef 5, <4 x i32> {{%.*}})
 251 // APCS-GNU: test_5s
 252 // APCS-GNU: call double (i32, ...) @varargs_vec_5s(i32 noundef 5, <4 x i32> {{%.*}})
 253 // ANDROID: test_5s
 254 // ANDROID: call double (i32, ...) @varargs_vec_5s(i32 noundef 5, <4 x i32> {{%.*}})
 255   return varargs_vec_5s(5, *in);
 256 }
 257
 258 // Pass struct as varargs.
 259 typedef struct
 260 {
 261   __int2 i2;
 262   float f;
 263 } StructWithVec;
 264
 265 double varargs_struct(int fixed, ...) {
 266 // CHECK: varargs_struct
 267 // CHECK: [[AP_ALIGN:%.*]] = call ptr @llvm.ptrmask.p0.i32(ptr {{%.*}}, i32 -8)
 268 // CHECK: [[AP_NEXT:%.*]] = getelementptr inbounds i8, ptr [[AP_ALIGN]], i32 16
 269 // APCS-GNU: varargs_struct
 270 // APCS-GNU: [[VAR_ALIGN:%.*]] = alloca %struct.StructWithVec
 271 // APCS-GNU: [[AP_NEXT:%.*]] = getelementptr inbounds i8, ptr {{%.*}}, i32 16
 272 // APCS-GNU: call void @llvm.memcpy
 273 // ANDROID: varargs_struct
 274 // ANDROID: [[AP_ALIGN:%.*]] = call ptr @llvm.ptrmask.p0.i32(ptr {{%.*}}, i32 -8)
 275 // ANDROID: [[AP_NEXT:%.*]] = getelementptr inbounds i8, ptr [[AP_ALIGN]], i32 16
 276   va_list ap;
 277   double sum = fixed;
 278   va_start(ap, fixed);
 279   StructWithVec c3 = va_arg(ap, StructWithVec);
 280   sum = sum + c3.i2.x + c3.i2.y + c3.f;
 281   va_end(ap);
 282   return sum;
 283 }
 284
 285 double test_struct(StructWithVec* d) {
 286 // CHECK: test_struct
 287 // CHECK: call arm_aapcscc double (i32, ...) @varargs_struct(i32 noundef 3, [2 x i64] {{%.*}})
 288 // APCS-GNU: test_struct
 289 // APCS-GNU: call double (i32, ...) @varargs_struct(i32 noundef 3, [2 x i64] {{%.*}})
 290 // ANDROID: test_struct
 291 // ANDROID: call double (i32, ...) @varargs_struct(i32 noundef 3, [2 x i64] {{%.*}})
 292   return varargs_struct(3, *d);
 293 }