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

   1 // RUN: %clang_cc1 -no-opaque-pointers -triple armv7-apple-darwin -target-abi aapcs -emit-llvm -o - %s | FileCheck %s
   2 // RUN: %clang_cc1 -no-opaque-pointers -triple armv7-apple-darwin -target-abi apcs-gnu -emit-llvm -o - %s | FileCheck -check-prefix=APCS-GNU %s
   3 // RUN: %clang_cc1 -no-opaque-pointers -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: [[ALIGN:%.*]] = and i32 {{%.*}}, -8
  20 // CHECK: [[AP_ALIGN:%.*]] = inttoptr i32 [[ALIGN]] to i8*
  21 // CHECK: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_ALIGN]], i32 8
  22 // CHECK: [[AP_CAST:%.*]] = bitcast i8* [[AP_ALIGN]] to <2 x i32>*
  23 // CHECK: [[VEC:%.*]] = load <2 x i32>, <2 x i32>* [[AP_CAST]], align 8
  24 // CHECK: store <2 x i32> [[VEC]], <2 x i32>* [[VAR]], align 8
  25 // APCS-GNU: varargs_vec_2i
  26 // APCS-GNU: [[VAR:%.*]] = alloca <2 x i32>, align 8
  27 // APCS-GNU: [[AP:%.*]] = load i8*,
  28 // APCS-GNU: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP]], i32 8
  29 // APCS-GNU: [[AP_CAST:%.*]] = bitcast i8* [[AP]] to <2 x i32>*
  30 // APCS-GNU: [[VEC:%.*]] = load <2 x i32>, <2 x i32>* [[AP_CAST]], align 4
  31 // APCS-GNU: store <2 x i32> [[VEC]], <2 x i32>* [[VAR]], align 8
  32 // ANDROID: varargs_vec_2i
  33 // ANDROID: [[VAR:%.*]] = alloca <2 x i32>, align 8
  34 // ANDROID: [[ALIGN:%.*]] = and i32 {{%.*}}, -8
  35 // ANDROID: [[AP_ALIGN:%.*]] = inttoptr i32 [[ALIGN]] to i8*
  36 // ANDROID: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_ALIGN]], i32 8
  37 // ANDROID: [[AP_CAST:%.*]] = bitcast i8* [[AP_ALIGN]] to <2 x i32>*
  38 // ANDROID: [[VEC:%.*]] = load <2 x i32>, <2 x i32>* [[AP_CAST]], align 8
  39 // ANDROID: store <2 x i32> [[VEC]], <2 x i32>* [[VAR]], align 8
  40   va_list ap;
  41   double sum = fixed;
  42   va_start(ap, fixed);
  43   __int2 c3 = va_arg(ap, __int2);
  44   sum = sum + c3.x + c3.y;
  45   va_end(ap);
  46   return sum;
  47 }
  48
  49 double test_2i(__int2 *in) {
  50 // CHECK: test_2i
  51 // CHECK: call arm_aapcscc double (i32, ...) @varargs_vec_2i(i32 noundef 3, <2 x i32> noundef {{%.*}})
  52 // APCS-GNU: test_2i
  53 // APCS-GNU: call double (i32, ...) @varargs_vec_2i(i32 noundef 3, <2 x i32> noundef {{%.*}})
  54 // ANDROID: test_2i
  55 // ANDROID: call double (i32, ...) @varargs_vec_2i(i32 noundef 3, <2 x i32> noundef {{%.*}})
  56   return varargs_vec_2i(3, *in);
  57 }
  58
  59 double varargs_vec_3c(int fixed, ...) {
  60 // CHECK: varargs_vec_3c
  61 // CHECK: alloca <3 x i8>, align 4
  62 // CHECK: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP:%.*]], i32 4
  63 // CHECK: bitcast i8* [[AP]] to <3 x i8>*
  64 // APCS-GNU: varargs_vec_3c
  65 // APCS-GNU: alloca <3 x i8>, align 4
  66 // APCS-GNU: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP:%.*]], i32 4
  67 // APCS-GNU: bitcast i8* [[AP]] to <3 x i8>*
  68 // ANDROID: varargs_vec_3c
  69 // ANDROID: alloca <3 x i8>, align 4
  70 // ANDROID: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP:%.*]], i32 4
  71 // ANDROID: bitcast i8* [[AP]] to <3 x i8>*
  72   va_list ap;
  73   double sum = fixed;
  74   va_start(ap, fixed);
  75   __char3 c3 = va_arg(ap, __char3);
  76   sum = sum + c3.x + c3.y;
  77   va_end(ap);
  78   return sum;
  79 }
  80
  81 double test_3c(__char3 *in) {
  82 // CHECK: test_3c
  83 // CHECK: call arm_aapcscc double (i32, ...) @varargs_vec_3c(i32 noundef 3, i32 {{%.*}})
  84 // APCS-GNU: test_3c
  85 // APCS-GNU: call double (i32, ...) @varargs_vec_3c(i32 noundef 3, i32 {{%.*}})
  86 // ANDROID: test_3c
  87 // ANDROID: call double (i32, ...) @varargs_vec_3c(i32 noundef 3, <3 x i8> noundef {{%.*}})
  88   return varargs_vec_3c(3, *in);
  89 }
  90
  91 double varargs_vec_5c(int fixed, ...) {
  92 // CHECK: varargs_vec_5c
  93 // CHECK: [[VAR:%.*]] = alloca <5 x i8>, align 8
  94 // CHECK: [[ALIGN:%.*]] = and i32 {{%.*}}, -8
  95 // CHECK: [[AP_ALIGN:%.*]] = inttoptr i32 [[ALIGN]] to i8*
  96 // CHECK: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_ALIGN]], i32 8
  97 // CHECK: [[AP_CAST:%.*]] = bitcast i8* [[AP_ALIGN]] to <5 x i8>*
  98 // CHECK: [[VEC:%.*]] = load <5 x i8>, <5 x i8>* [[AP_CAST]], align 8
  99 // CHECK: store <5 x i8> [[VEC]], <5 x i8>* [[VAR]], align 8
 100 // APCS-GNU: varargs_vec_5c
 101 // APCS-GNU: [[VAR:%.*]] = alloca <5 x i8>, align 8
 102 // APCS-GNU: [[AP:%.*]] = load i8*,
 103 // APCS-GNU: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP]], i32 8
 104 // APCS-GNU: [[AP_CAST:%.*]] = bitcast i8* [[AP]] to <5 x i8>*
 105 // APCS-GNU: [[VEC:%.*]] = load <5 x i8>, <5 x i8>* [[AP_CAST]], align 4
 106 // APCS-GNU: store <5 x i8> [[VEC]], <5 x i8>* [[VAR]], align 8
 107 // ANDROID: varargs_vec_5c
 108 // ANDROID: [[VAR:%.*]] = alloca <5 x i8>, align 8
 109 // ANDROID: [[ALIGN:%.*]] = and i32 {{%.*}}, -8
 110 // ANDROID: [[AP_ALIGN:%.*]] = inttoptr i32 [[ALIGN]] to i8*
 111 // ANDROID: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_ALIGN]], i32 8
 112 // ANDROID: [[AP_CAST:%.*]] = bitcast i8* [[AP_ALIGN]] to <5 x i8>*
 113 // ANDROID: [[VEC:%.*]] = load <5 x i8>, <5 x i8>* [[AP_CAST]], align 8
 114 // ANDROID: store <5 x i8> [[VEC]], <5 x i8>* [[VAR]], align 8
 115   va_list ap;
 116   double sum = fixed;
 117   va_start(ap, fixed);
 118   __char5 c5 = va_arg(ap, __char5);
 119   sum = sum + c5.x + c5.y;
 120   va_end(ap);
 121   return sum;
 122 }
 123
 124 double test_5c(__char5 *in) {
 125 // CHECK: test_5c
 126 // CHECK: call arm_aapcscc double (i32, ...) @varargs_vec_5c(i32 noundef 5, <2 x i32> {{%.*}})
 127 // APCS-GNU: test_5c
 128 // APCS-GNU: call double (i32, ...) @varargs_vec_5c(i32 noundef 5, <2 x i32> {{%.*}})
 129 // ANDROID: test_5c
 130 // ANDROID: call double (i32, ...) @varargs_vec_5c(i32 noundef 5, <2 x i32> {{%.*}})
 131   return varargs_vec_5c(5, *in);
 132 }
 133
 134 double varargs_vec_9c(int fixed, ...) {
 135 // CHECK: varargs_vec_9c
 136 // CHECK: [[VAR:%.*]] = alloca <9 x i8>, align 16
 137 // CHECK: [[ALIGN:%.*]] = and i32 {{%.*}}, -8
 138 // CHECK: [[AP_ALIGN:%.*]] = inttoptr i32 [[ALIGN]] to i8*
 139 // CHECK: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_ALIGN]], i32 16
 140 // CHECK: [[AP_CAST:%.*]] = bitcast i8* [[AP_ALIGN]] to <9 x i8>*
 141 // CHECK: [[T0:%.*]] = load <9 x i8>, <9 x i8>* [[AP_CAST]], align 8
 142 // CHECK: store <9 x i8> [[T0]], <9 x i8>* [[VAR]], align 16
 143 // APCS-GNU: varargs_vec_9c
 144 // APCS-GNU: [[VAR:%.*]] = alloca <9 x i8>, align 16
 145 // APCS-GNU: [[AP:%.*]] = load i8*,
 146 // APCS-GNU: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP]], i32 16
 147 // APCS-GNU: [[AP_CAST:%.*]] = bitcast i8* [[AP]] to <9 x i8>*
 148 // APCS-GNU: [[VEC:%.*]] = load <9 x i8>, <9 x i8>* [[AP_CAST]], align 4
 149 // APCS-GNU: store <9 x i8> [[VEC]], <9 x i8>* [[VAR]], align 16
 150 // ANDROID: varargs_vec_9c
 151 // ANDROID: [[VAR:%.*]] = alloca <9 x i8>, align 16
 152 // ANDROID: [[ALIGN:%.*]] = and i32 {{%.*}}, -8
 153 // ANDROID: [[AP_ALIGN:%.*]] = inttoptr i32 [[ALIGN]] to i8*
 154 // ANDROID: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_ALIGN]], i32 16
 155 // ANDROID: [[AP_CAST:%.*]] = bitcast i8* [[AP_ALIGN]] to <9 x i8>*
 156 // ANDROID: [[T0:%.*]] = load <9 x i8>, <9 x i8>* [[AP_CAST]], align 8
 157 // ANDROID: store <9 x i8> [[T0]], <9 x i8>* [[VAR]], align 16
 158   va_list ap;
 159   double sum = fixed;
 160   va_start(ap, fixed);
 161   __char9 c9 = va_arg(ap, __char9);
 162   sum = sum + c9.x + c9.y;
 163   va_end(ap);
 164   return sum;
 165 }
 166
 167 double test_9c(__char9 *in) {
 168 // CHECK: test_9c
 169 // CHECK: call arm_aapcscc double (i32, ...) @varargs_vec_9c(i32 noundef 9, <4 x i32> {{%.*}})
 170 // APCS-GNU: test_9c
 171 // APCS-GNU: call double (i32, ...) @varargs_vec_9c(i32 noundef 9, <4 x i32> {{%.*}})
 172 // ANDROID: test_9c
 173 // ANDROID: call double (i32, ...) @varargs_vec_9c(i32 noundef 9, <4 x i32> {{%.*}})
 174   return varargs_vec_9c(9, *in);
 175 }
 176
 177 double varargs_vec_19c(int fixed, ...) {
 178 // CHECK: varargs_vec_19c
 179 // CHECK: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP:%.*]], i32 4
 180 // CHECK: [[VAR:%.*]] = bitcast i8* [[AP]] to <19 x i8>**
 181 // CHECK: [[VAR2:%.*]] = load <19 x i8>*, <19 x i8>** [[VAR]]
 182 // APCS-GNU: varargs_vec_19c
 183 // APCS-GNU: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP:%.*]], i32 4
 184 // APCS-GNU: [[VAR:%.*]] = bitcast i8* [[AP]] to <19 x i8>**
 185 // APCS-GNU: [[VAR2:%.*]] = load <19 x i8>*, <19 x i8>** [[VAR]]
 186 // ANDROID: varargs_vec_19c
 187 // ANDROID: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP:%.*]], i32 4
 188 // ANDROID: [[VAR:%.*]] = bitcast i8* [[AP]] to <19 x i8>**
 189 // ANDROID: [[VAR2:%.*]] = load <19 x i8>*, <19 x i8>** [[VAR]]
 190   va_list ap;
 191   double sum = fixed;
 192   va_start(ap, fixed);
 193   __char19 c19 = va_arg(ap, __char19);
 194   sum = sum + c19.x + c19.y;
 195   va_end(ap);
 196   return sum;
 197 }
 198
 199 double test_19c(__char19 *in) {
 200 // CHECK: test_19c
 201 // CHECK: call arm_aapcscc double (i32, ...) @varargs_vec_19c(i32 noundef 19, <19 x i8>* noundef {{%.*}})
 202 // APCS-GNU: test_19c
 203 // APCS-GNU: call double (i32, ...) @varargs_vec_19c(i32 noundef 19, <19 x i8>* noundef {{%.*}})
 204 // ANDROID: test_19c
 205 // ANDROID: call double (i32, ...) @varargs_vec_19c(i32 noundef 19, <19 x i8>* noundef {{%.*}})
 206   return varargs_vec_19c(19, *in);
 207 }
 208
 209 double varargs_vec_3s(int fixed, ...) {
 210 // CHECK: varargs_vec_3s
 211 // CHECK: alloca <3 x i16>, align 8
 212 // CHECK: [[ALIGN:%.*]] = and i32 {{%.*}}, -8
 213 // CHECK: [[AP_ALIGN:%.*]] = inttoptr i32 [[ALIGN]] to i8*
 214 // CHECK: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_ALIGN]], i32 8
 215 // CHECK: bitcast i8* [[AP_ALIGN]] to <3 x i16>*
 216 // APCS-GNU: varargs_vec_3s
 217 // APCS-GNU: [[VAR:%.*]] = alloca <3 x i16>, align 8
 218 // APCS-GNU: [[AP:%.*]] = load i8*,
 219 // APCS-GNU: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP]], i32 8
 220 // APCS-GNU: [[AP_CAST:%.*]] = bitcast i8* [[AP]] to <3 x i16>*
 221 // APCS-GNU: [[VEC:%.*]] = load <3 x i16>, <3 x i16>* [[AP_CAST]], align 4
 222 // ANDROID: varargs_vec_3s
 223 // ANDROID: alloca <3 x i16>, align 8
 224 // ANDROID: [[ALIGN:%.*]] = and i32 {{%.*}}, -8
 225 // ANDROID: [[AP_ALIGN:%.*]] = inttoptr i32 [[ALIGN]] to i8*
 226 // ANDROID: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_ALIGN]], i32 8
 227 // ANDROID: bitcast i8* [[AP_ALIGN]] to <3 x i16>*
 228   va_list ap;
 229   double sum = fixed;
 230   va_start(ap, fixed);
 231   __short3 c3 = va_arg(ap, __short3);
 232   sum = sum + c3.x + c3.y;
 233   va_end(ap);
 234   return sum;
 235 }
 236
 237 double test_3s(__short3 *in) {
 238 // CHECK: test_3s
 239 // CHECK: call arm_aapcscc double (i32, ...) @varargs_vec_3s(i32 noundef 3, <2 x i32> {{%.*}})
 240 // APCS-GNU: test_3s
 241 // APCS-GNU: call double (i32, ...) @varargs_vec_3s(i32 noundef 3, <2 x i32> {{%.*}})
 242 // ANDROID: test_3s
 243 // ANDROID: call double (i32, ...) @varargs_vec_3s(i32 noundef 3, <3 x i16> noundef {{%.*}})
 244   return varargs_vec_3s(3, *in);
 245 }
 246
 247 double varargs_vec_5s(int fixed, ...) {
 248 // CHECK: varargs_vec_5s
 249 // CHECK: [[VAR_ALIGN:%.*]] = alloca <5 x i16>, align 16
 250 // CHECK: [[ALIGN:%.*]] = and i32 {{%.*}}, -8
 251 // CHECK: [[AP_ALIGN:%.*]] = inttoptr i32 [[ALIGN]] to i8*
 252 // CHECK: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_ALIGN]], i32 16
 253 // CHECK: [[AP_CAST:%.*]] = bitcast i8* [[AP_ALIGN]] to <5 x i16>*
 254 // CHECK: [[VEC:%.*]] = load <5 x i16>, <5 x i16>* [[AP_CAST]], align 8
 255 // CHECK: store <5 x i16> [[VEC]], <5 x i16>* [[VAR_ALIGN]], align 16
 256 // APCS-GNU: varargs_vec_5s
 257 // APCS-GNU: [[VAR:%.*]] = alloca <5 x i16>, align 16
 258 // APCS-GNU: [[AP:%.*]] = load i8*,
 259 // APCS-GNU: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP]], i32 16
 260 // APCS-GNU: [[AP_CAST:%.*]] = bitcast i8* [[AP]] to <5 x i16>*
 261 // APCS-GNU: [[VEC:%.*]] = load <5 x i16>, <5 x i16>* [[AP_CAST]], align 4
 262 // ANDROID: varargs_vec_5s
 263 // ANDROID: [[VAR_ALIGN:%.*]] = alloca <5 x i16>, align 16
 264 // ANDROID: [[ALIGN:%.*]] = and i32 {{%.*}}, -8
 265 // ANDROID: [[AP_ALIGN:%.*]] = inttoptr i32 [[ALIGN]] to i8*
 266 // ANDROID: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_ALIGN]], i32 16
 267 // ANDROID: [[AP_CAST:%.*]] = bitcast i8* [[AP_ALIGN]] to <5 x i16>*
 268 // ANDROID: [[VEC:%.*]] = load <5 x i16>, <5 x i16>* [[AP_CAST]], align 8
 269 // ANDROID: store <5 x i16> [[VEC]], <5 x i16>* [[VAR_ALIGN]], align 16
 270   va_list ap;
 271   double sum = fixed;
 272   va_start(ap, fixed);
 273   __short5 c5 = va_arg(ap, __short5);
 274   sum = sum + c5.x + c5.y;
 275   va_end(ap);
 276   return sum;
 277 }
 278
 279 double test_5s(__short5 *in) {
 280 // CHECK: test_5s
 281 // CHECK: call arm_aapcscc double (i32, ...) @varargs_vec_5s(i32 noundef 5, <4 x i32> {{%.*}})
 282 // APCS-GNU: test_5s
 283 // APCS-GNU: call double (i32, ...) @varargs_vec_5s(i32 noundef 5, <4 x i32> {{%.*}})
 284 // ANDROID: test_5s
 285 // ANDROID: call double (i32, ...) @varargs_vec_5s(i32 noundef 5, <4 x i32> {{%.*}})
 286   return varargs_vec_5s(5, *in);
 287 }
 288
 289 // Pass struct as varargs.
 290 typedef struct
 291 {
 292   __int2 i2;
 293   float f;
 294 } StructWithVec;
 295
 296 double varargs_struct(int fixed, ...) {
 297 // CHECK: varargs_struct
 298 // CHECK: [[ALIGN:%.*]] = and i32 {{%.*}}, -8
 299 // CHECK: [[AP_ALIGN:%.*]] = inttoptr i32 [[ALIGN]] to i8*
 300 // CHECK: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_ALIGN]], i32 16
 301 // CHECK: bitcast i8* [[AP_ALIGN]] to %struct.StructWithVec*
 302 // APCS-GNU: varargs_struct
 303 // APCS-GNU: [[VAR_ALIGN:%.*]] = alloca %struct.StructWithVec
 304 // APCS-GNU: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* {{%.*}}, i32 16
 305 // APCS-GNU: bitcast %struct.StructWithVec* [[VAR_ALIGN]] to i8*
 306 // APCS-GNU: call void @llvm.memcpy
 307 // ANDROID: varargs_struct
 308 // ANDROID: [[ALIGN:%.*]] = and i32 {{%.*}}, -8
 309 // ANDROID: [[AP_ALIGN:%.*]] = inttoptr i32 [[ALIGN]] to i8*
 310 // ANDROID: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_ALIGN]], i32 16
 311 // ANDROID: bitcast i8* [[AP_ALIGN]] to %struct.StructWithVec*
 312   va_list ap;
 313   double sum = fixed;
 314   va_start(ap, fixed);
 315   StructWithVec c3 = va_arg(ap, StructWithVec);
 316   sum = sum + c3.i2.x + c3.i2.y + c3.f;
 317   va_end(ap);
 318   return sum;
 319 }
 320
 321 double test_struct(StructWithVec* d) {
 322 // CHECK: test_struct
 323 // CHECK: call arm_aapcscc double (i32, ...) @varargs_struct(i32 noundef 3, [2 x i64] {{%.*}})
 324 // APCS-GNU: test_struct
 325 // APCS-GNU: call double (i32, ...) @varargs_struct(i32 noundef 3, [2 x i64] {{%.*}})
 326 // ANDROID: test_struct
 327 // ANDROID: call double (i32, ...) @varargs_struct(i32 noundef 3, [2 x i64] {{%.*}})
 328   return varargs_struct(3, *d);
 329 }