clang/test/CodeGen/aarch64-sve-inline-asm-datatypes.c

   1 // RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve \
   2 // RUN:   -target-feature +neon -S -O1 -o - -emit-llvm %s | FileCheck %s
   3
   4 // Tests to check that all sve datatypes can be passed in as input operands
   5 // and passed out as output operands.
   6
   7 #define SVINT_TEST(DT, KIND)\
   8 DT func_int_##DT##KIND(DT in)\
   9 {\
  10   DT out;\
  11   asm volatile (\
  12     "ptrue p0.b\n"\
  13     "mov %[out]." #KIND ", p0/m, %[in]." #KIND "\n"\
  14     : [out] "=w" (out)\
  15     : [in] "w" (in)\
  16     : "p0"\
  17     );\
  18   return out;\
  19 }
  20
  21 SVINT_TEST(__SVUint8_t,b);
  22 // CHECK: call <vscale x 16 x i8> asm sideeffect "ptrue p0.b\0Amov $0.b, p0/m, $1.b\0A", "=w,w,~{p0}"(<vscale x 16 x i8> %in)
  23 SVINT_TEST(__SVUint8_t,h);
  24 // CHECK: call <vscale x 16 x i8> asm sideeffect "ptrue p0.b\0Amov $0.h, p0/m, $1.h\0A", "=w,w,~{p0}"(<vscale x 16 x i8> %in)
  25 SVINT_TEST(__SVUint8_t,s);
  26 // CHECK: call <vscale x 16 x i8> asm sideeffect "ptrue p0.b\0Amov $0.s, p0/m, $1.s\0A", "=w,w,~{p0}"(<vscale x 16 x i8> %in)
  27 SVINT_TEST(__SVUint8_t,d);
  28 // CHECK: call <vscale x 16 x i8> asm sideeffect "ptrue p0.b\0Amov $0.d, p0/m, $1.d\0A", "=w,w,~{p0}"(<vscale x 16 x i8> %in)
  29
  30 SVINT_TEST(__SVUint16_t,b);
  31 // CHECK: call <vscale x 8 x i16> asm sideeffect "ptrue p0.b\0Amov $0.b, p0/m, $1.b\0A", "=w,w,~{p0}"(<vscale x 8 x i16> %in)
  32 SVINT_TEST(__SVUint16_t,h);
  33 // CHECK: call <vscale x 8 x i16> asm sideeffect "ptrue p0.b\0Amov $0.h, p0/m, $1.h\0A", "=w,w,~{p0}"(<vscale x 8 x i16> %in)
  34 SVINT_TEST(__SVUint16_t,s);
  35 // CHECK: call <vscale x 8 x i16> asm sideeffect "ptrue p0.b\0Amov $0.s, p0/m, $1.s\0A", "=w,w,~{p0}"(<vscale x 8 x i16> %in)
  36 SVINT_TEST(__SVUint16_t,d);
  37 // CHECK: call <vscale x 8 x i16> asm sideeffect "ptrue p0.b\0Amov $0.d, p0/m, $1.d\0A", "=w,w,~{p0}"(<vscale x 8 x i16> %in)
  38
  39 SVINT_TEST(__SVUint32_t,b);
  40 // CHECK: call <vscale x 4 x i32> asm sideeffect "ptrue p0.b\0Amov $0.b, p0/m, $1.b\0A", "=w,w,~{p0}"(<vscale x 4 x i32> %in)
  41 SVINT_TEST(__SVUint32_t,h);
  42 // CHECK: call <vscale x 4 x i32> asm sideeffect "ptrue p0.b\0Amov $0.h, p0/m, $1.h\0A", "=w,w,~{p0}"(<vscale x 4 x i32> %in)
  43 SVINT_TEST(__SVUint32_t,s);
  44 // CHECK: call <vscale x 4 x i32> asm sideeffect "ptrue p0.b\0Amov $0.s, p0/m, $1.s\0A", "=w,w,~{p0}"(<vscale x 4 x i32> %in)
  45 SVINT_TEST(__SVUint32_t,d);
  46 // CHECK: call <vscale x 4 x i32> asm sideeffect "ptrue p0.b\0Amov $0.d, p0/m, $1.d\0A", "=w,w,~{p0}"(<vscale x 4 x i32> %in)
  47
  48 SVINT_TEST(__SVUint64_t,b);
  49 // CHECK: call <vscale x 2 x i64> asm sideeffect "ptrue p0.b\0Amov $0.b, p0/m, $1.b\0A", "=w,w,~{p0}"(<vscale x 2 x i64> %in)
  50 SVINT_TEST(__SVUint64_t,h);
  51 // CHECK: call <vscale x 2 x i64> asm sideeffect "ptrue p0.b\0Amov $0.h, p0/m, $1.h\0A", "=w,w,~{p0}"(<vscale x 2 x i64> %in)
  52 SVINT_TEST(__SVUint64_t,s);
  53 // CHECK: call <vscale x 2 x i64> asm sideeffect "ptrue p0.b\0Amov $0.s, p0/m, $1.s\0A", "=w,w,~{p0}"(<vscale x 2 x i64> %in)
  54 SVINT_TEST(__SVUint64_t,d);
  55 // CHECK: call <vscale x 2 x i64> asm sideeffect "ptrue p0.b\0Amov $0.d, p0/m, $1.d\0A", "=w,w,~{p0}"(<vscale x 2 x i64> %in)
  56
  57 SVINT_TEST(__SVInt8_t,b);
  58 // CHECK: call <vscale x 16 x i8> asm sideeffect "ptrue p0.b\0Amov $0.b, p0/m, $1.b\0A", "=w,w,~{p0}"(<vscale x 16 x i8> %in)
  59 SVINT_TEST(__SVInt8_t,h);
  60 // CHECK: call <vscale x 16 x i8> asm sideeffect "ptrue p0.b\0Amov $0.h, p0/m, $1.h\0A", "=w,w,~{p0}"(<vscale x 16 x i8> %in)
  61 SVINT_TEST(__SVInt8_t,s);
  62 // CHECK: call <vscale x 16 x i8> asm sideeffect "ptrue p0.b\0Amov $0.s, p0/m, $1.s\0A", "=w,w,~{p0}"(<vscale x 16 x i8> %in)
  63 SVINT_TEST(__SVInt8_t,d);
  64 // CHECK: call <vscale x 16 x i8> asm sideeffect "ptrue p0.b\0Amov $0.d, p0/m, $1.d\0A", "=w,w,~{p0}"(<vscale x 16 x i8> %in)
  65
  66 SVINT_TEST(__SVInt16_t,b);
  67 // CHECK: call <vscale x 8 x i16> asm sideeffect "ptrue p0.b\0Amov $0.b, p0/m, $1.b\0A", "=w,w,~{p0}"(<vscale x 8 x i16> %in)
  68 SVINT_TEST(__SVInt16_t,h);
  69 // CHECK: call <vscale x 8 x i16> asm sideeffect "ptrue p0.b\0Amov $0.h, p0/m, $1.h\0A", "=w,w,~{p0}"(<vscale x 8 x i16> %in)
  70 SVINT_TEST(__SVInt16_t,s);
  71 // CHECK: call <vscale x 8 x i16> asm sideeffect "ptrue p0.b\0Amov $0.s, p0/m, $1.s\0A", "=w,w,~{p0}"(<vscale x 8 x i16> %in)
  72 SVINT_TEST(__SVInt16_t,d);
  73 // CHECK: call <vscale x 8 x i16> asm sideeffect "ptrue p0.b\0Amov $0.d, p0/m, $1.d\0A", "=w,w,~{p0}"(<vscale x 8 x i16> %in)
  74
  75 SVINT_TEST(__SVInt32_t,b);
  76 // CHECK: call <vscale x 4 x i32> asm sideeffect "ptrue p0.b\0Amov $0.b, p0/m, $1.b\0A", "=w,w,~{p0}"(<vscale x 4 x i32> %in)
  77 SVINT_TEST(__SVInt32_t,h);
  78 // CHECK: call <vscale x 4 x i32> asm sideeffect "ptrue p0.b\0Amov $0.h, p0/m, $1.h\0A", "=w,w,~{p0}"(<vscale x 4 x i32> %in)
  79 SVINT_TEST(__SVInt32_t,s);
  80 // CHECK: call <vscale x 4 x i32> asm sideeffect "ptrue p0.b\0Amov $0.s, p0/m, $1.s\0A", "=w,w,~{p0}"(<vscale x 4 x i32> %in)
  81 SVINT_TEST(__SVInt32_t,d);
  82 // CHECK: call <vscale x 4 x i32> asm sideeffect "ptrue p0.b\0Amov $0.d, p0/m, $1.d\0A", "=w,w,~{p0}"(<vscale x 4 x i32> %in)
  83
  84 SVINT_TEST(__SVInt64_t,b);
  85 // CHECK: call <vscale x 2 x i64> asm sideeffect "ptrue p0.b\0Amov $0.b, p0/m, $1.b\0A", "=w,w,~{p0}"(<vscale x 2 x i64> %in)
  86 SVINT_TEST(__SVInt64_t,h);
  87 // CHECK: call <vscale x 2 x i64> asm sideeffect "ptrue p0.b\0Amov $0.h, p0/m, $1.h\0A", "=w,w,~{p0}"(<vscale x 2 x i64> %in)
  88 SVINT_TEST(__SVInt64_t,s);
  89 // CHECK: call <vscale x 2 x i64> asm sideeffect "ptrue p0.b\0Amov $0.s, p0/m, $1.s\0A", "=w,w,~{p0}"(<vscale x 2 x i64> %in)
  90 SVINT_TEST(__SVInt64_t,d);
  91 // CHECK: call <vscale x 2 x i64> asm sideeffect "ptrue p0.b\0Amov $0.d, p0/m, $1.d\0A", "=w,w,~{p0}"(<vscale x 2 x i64> %in)
  92
  93
  94 //Test that floats can also be used as datatypes for integer instructions
  95 //and check all the variants which would not be possible with a float
  96 //instruction
  97 SVINT_TEST(__SVFloat16_t,b);
  98 // CHECK: call <vscale x 8 x half> asm sideeffect "ptrue p0.b\0Amov $0.b, p0/m, $1.b\0A", "=w,w,~{p0}"(<vscale x 8 x half> %in)
  99 SVINT_TEST(__SVFloat16_t,h);
 100 // CHECK: call <vscale x 8 x half> asm sideeffect "ptrue p0.b\0Amov $0.h, p0/m, $1.h\0A", "=w,w,~{p0}"(<vscale x 8 x half> %in)
 101 SVINT_TEST(__SVFloat16_t,s);
 102 // CHECK: call <vscale x 8 x half> asm sideeffect "ptrue p0.b\0Amov $0.s, p0/m, $1.s\0A", "=w,w,~{p0}"(<vscale x 8 x half> %in)
 103 SVINT_TEST(__SVFloat16_t,d);
 104 // CHECK: call <vscale x 8 x half> asm sideeffect "ptrue p0.b\0Amov $0.d, p0/m, $1.d\0A", "=w,w,~{p0}"(<vscale x 8 x half> %in)
 105
 106 SVINT_TEST(__SVFloat32_t,b);
 107 // CHECK: call <vscale x 4 x float> asm sideeffect "ptrue p0.b\0Amov $0.b, p0/m, $1.b\0A", "=w,w,~{p0}"(<vscale x 4 x float> %in)
 108 SVINT_TEST(__SVFloat32_t,h);
 109 // CHECK: call <vscale x 4 x float> asm sideeffect "ptrue p0.b\0Amov $0.h, p0/m, $1.h\0A", "=w,w,~{p0}"(<vscale x 4 x float> %in)
 110 SVINT_TEST(__SVFloat32_t,s);
 111 // CHECK: call <vscale x 4 x float> asm sideeffect "ptrue p0.b\0Amov $0.s, p0/m, $1.s\0A", "=w,w,~{p0}"(<vscale x 4 x float> %in)
 112 SVINT_TEST(__SVFloat32_t,d);
 113 // CHECK: call <vscale x 4 x float> asm sideeffect "ptrue p0.b\0Amov $0.d, p0/m, $1.d\0A", "=w,w,~{p0}"(<vscale x 4 x float> %in)
 114
 115 SVINT_TEST(__SVFloat64_t,b);
 116 // CHECK: call <vscale x 2 x double> asm sideeffect "ptrue p0.b\0Amov $0.b, p0/m, $1.b\0A", "=w,w,~{p0}"(<vscale x 2 x double> %in)
 117 SVINT_TEST(__SVFloat64_t,h);
 118 // CHECK: call <vscale x 2 x double> asm sideeffect "ptrue p0.b\0Amov $0.h, p0/m, $1.h\0A", "=w,w,~{p0}"(<vscale x 2 x double> %in)
 119 SVINT_TEST(__SVFloat64_t,s);
 120 // CHECK: call <vscale x 2 x double> asm sideeffect "ptrue p0.b\0Amov $0.s, p0/m, $1.s\0A", "=w,w,~{p0}"(<vscale x 2 x double> %in)
 121 SVINT_TEST(__SVFloat64_t,d);
 122 // CHECK: call <vscale x 2 x double> asm sideeffect "ptrue p0.b\0Amov $0.d, p0/m, $1.d\0A", "=w,w,~{p0}"(<vscale x 2 x double> %in)
 123
 124
 125 #define SVBOOL_TEST(KIND)\
 126 __SVBool_t func_bool_##KIND(__SVBool_t in1, __SVBool_t in2)\
 127 {\
 128   __SVBool_t out;\
 129   asm volatile (\
 130     "zip1 %[out]." #KIND ", %[in1]." #KIND ", %[in2]." #KIND "\n"\
 131     : [out] "=Upa" (out)\
 132     :  [in1] "Upa" (in1),\
 133       [in2] "Upa" (in2)\
 134     :);\
 135   return out;\
 136 }
 137
 138 SVBOOL_TEST(b) ;
 139 // CHECK: call <vscale x 16 x i1> asm sideeffect "zip1 $0.b, $1.b, $2.b\0A", "=@3Upa,@3Upa,@3Upa"(<vscale x 16 x i1> %in1, <vscale x 16 x i1> %in2)
 140 SVBOOL_TEST(h) ;
 141 // CHECK: call <vscale x 16 x i1> asm sideeffect "zip1 $0.h, $1.h, $2.h\0A", "=@3Upa,@3Upa,@3Upa"(<vscale x 16 x i1> %in1, <vscale x 16 x i1> %in2)
 142 SVBOOL_TEST(s) ;
 143 // CHECK: call <vscale x 16 x i1> asm sideeffect "zip1 $0.s, $1.s, $2.s\0A", "=@3Upa,@3Upa,@3Upa"(<vscale x 16 x i1> %in1, <vscale x 16 x i1> %in2)
 144 SVBOOL_TEST(d) ;
 145 // CHECK: call <vscale x 16 x i1> asm sideeffect "zip1 $0.d, $1.d, $2.d\0A", "=@3Upa,@3Upa,@3Upa"(<vscale x 16 x i1> %in1, <vscale x 16 x i1> %in2)
 146
 147
 148 #define SVBOOL_TEST_UPL(DT, KIND)\
 149 __SVBool_t func_bool_upl_##KIND(__SVBool_t in1, DT in2, DT in3)\
 150 {\
 151   __SVBool_t out;\
 152   asm volatile (\
 153     "fadd %[out]." #KIND ", %[in1]." #KIND ", %[in2]." #KIND ", %[in3]." #KIND "\n"\
 154     : [out] "=w" (out)\
 155     :  [in1] "Upl" (in1),\
 156       [in2] "w" (in2),\
 157       [in3] "w" (in3)\
 158     :);\
 159   return out;\
 160 }
 161
 162 SVBOOL_TEST_UPL(__SVInt8_t, b) ;
 163 // CHECK: call <vscale x 16 x i1> asm sideeffect "fadd $0.b, $1.b, $2.b, $3.b\0A", "=w,@3Upl,w,w"(<vscale x 16 x i1> %in1, <vscale x 16 x i8> %in2, <vscale x 16 x i8> %in3)
 164 SVBOOL_TEST_UPL(__SVInt16_t, h) ;
 165 // CHECK: call <vscale x 16 x i1> asm sideeffect "fadd $0.h, $1.h, $2.h, $3.h\0A", "=w,@3Upl,w,w"(<vscale x 16 x i1> %in1, <vscale x 8 x i16> %in2, <vscale x 8 x i16> %in3)
 166 SVBOOL_TEST_UPL(__SVInt32_t, s) ;
 167 // CHECK: call <vscale x 16 x i1> asm sideeffect "fadd $0.s, $1.s, $2.s, $3.s\0A", "=w,@3Upl,w,w"(<vscale x 16 x i1> %in1, <vscale x 4 x i32> %in2, <vscale x 4 x i32> %in3)
 168 SVBOOL_TEST_UPL(__SVInt64_t, d) ;
 169 // CHECK: call <vscale x 16 x i1> asm sideeffect "fadd $0.d, $1.d, $2.d, $3.d\0A", "=w,@3Upl,w,w"(<vscale x 16 x i1> %in1, <vscale x 2 x i64> %in2, <vscale x 2 x i64> %in3)
 170
 171 #define SVFLOAT_TEST(DT,KIND)\
 172 DT func_float_##DT##KIND(DT inout1, DT in2)\
 173 {\
 174   asm volatile (\
 175     "ptrue p0." #KIND ", #1 \n"\
 176     "fsub %[inout1]." #KIND ", p0/m, %[inout1]." #KIND ", %[in2]." #KIND "\n"\
 177     : [inout1] "=w" (inout1)\
 178     : "[inout1]" (inout1),\
 179       [in2] "w" (in2)\
 180     : "p0");\
 181   return inout1 ;\
 182 }\
 183
 184 SVFLOAT_TEST(__SVFloat16_t,s);
 185 // CHECK: call <vscale x 8 x half> asm sideeffect "ptrue p0.s, #1 \0Afsub $0.s, p0/m, $0.s, $2.s\0A", "=w,0,w,~{p0}"(<vscale x 8 x half> %inout1, <vscale x 8 x half> %in2)
 186 SVFLOAT_TEST(__SVFloat16_t,d);
 187 // CHECK: call <vscale x 8 x half> asm sideeffect "ptrue p0.d, #1 \0Afsub $0.d, p0/m, $0.d, $2.d\0A", "=w,0,w,~{p0}"(<vscale x 8 x half> %inout1, <vscale x 8 x half> %in2)
 188
 189 SVFLOAT_TEST(__SVFloat32_t,s);
 190 // CHECK: call <vscale x 4 x float> asm sideeffect "ptrue p0.s, #1 \0Afsub $0.s, p0/m, $0.s, $2.s\0A", "=w,0,w,~{p0}"(<vscale x 4 x float> %inout1, <vscale x 4 x float> %in2)
 191 SVFLOAT_TEST(__SVFloat32_t,d);
 192 // CHECK: call <vscale x 4 x float> asm sideeffect "ptrue p0.d, #1 \0Afsub $0.d, p0/m, $0.d, $2.d\0A", "=w,0,w,~{p0}"(<vscale x 4 x float> %inout1, <vscale x 4 x float> %in2)
 193
 194 SVFLOAT_TEST(__SVFloat64_t,s);
 195 // CHECK: call <vscale x 2 x double> asm sideeffect "ptrue p0.s, #1 \0Afsub $0.s, p0/m, $0.s, $2.s\0A", "=w,0,w,~{p0}"(<vscale x 2 x double> %inout1, <vscale x 2 x double> %in2)
 196 SVFLOAT_TEST(__SVFloat64_t,d);
 197 // CHECK: call <vscale x 2 x double> asm sideeffect "ptrue p0.d, #1 \0Afsub $0.d, p0/m, $0.d, $2.d\0A", "=w,0,w,~{p0}"(<vscale x 2 x double> %inout1, <vscale x 2 x double> %in2)
 198
 199 #define SVFLOAT_TEST_Y(DT, KIND)\
 200 __SVBool_t func_float_y_##KIND(DT in1, DT in2)\
 201 {\
 202   __SVBool_t out;\
 203   asm volatile (\
 204     "fmul %[out]." #KIND ", %[in1]." #KIND ", %[in2]." #KIND "\n"\
 205     : [out] "=w" (out)\
 206     :  [in1] "w" (in1),\
 207       [in2] "y" (in2)\
 208     :);\
 209   return out;\
 210 }
 211
 212 SVFLOAT_TEST_Y(__SVFloat16_t,h);
 213 // CHECK: call <vscale x 16 x i1> asm sideeffect "fmul $0.h, $1.h, $2.h\0A", "=w,w,y"(<vscale x 8 x half> %in1, <vscale x 8 x half> %in2)
 214 SVFLOAT_TEST_Y(__SVFloat32_t,s);
 215 // CHECK: call <vscale x 16 x i1> asm sideeffect "fmul $0.s, $1.s, $2.s\0A", "=w,w,y"(<vscale x 4 x float> %in1, <vscale x 4 x float> %in2)
 216 SVFLOAT_TEST_Y(__SVFloat64_t,d);
 217 // CHECK: call <vscale x 16 x i1> asm sideeffect "fmul $0.d, $1.d, $2.d\0A", "=w,w,y"(<vscale x 2 x double> %in1, <vscale x 2 x double> %in2)
 218
 219
 220 // Another test for floats to include h suffix
 221
 222 #define SVFLOAT_CVT_TEST(DT1,KIND1,DT2,KIND2)\
 223 DT1 func_float_cvt_##DT1##KIND1##DT2##KIND2(DT2 in1)\
 224 {\
 225   DT1 out1 ;\
 226   asm volatile (\
 227     "ptrue p0." #KIND2 ", #1 \n"\
 228     "fcvt %[out1]." #KIND1 ", p0/m, %[in1]." #KIND2 "\n"\
 229     : [out1] "=w" (out1)\
 230     : [in1] "w" (in1)\
 231     : "p0");\
 232   return out1 ;\
 233 }\
 234
 235 SVFLOAT_CVT_TEST(__SVFloat64_t,d,__SVFloat32_t,s);
 236 // CHECK: call <vscale x 2 x double> asm sideeffect "ptrue p0.s, #1 \0Afcvt $0.d, p0/m, $1.s\0A", "=w,w,~{p0}"(<vscale x 4 x float> %in1)
 237 SVFLOAT_CVT_TEST(__SVFloat64_t,d,__SVFloat16_t,h);
 238 // CHECK: call <vscale x 2 x double> asm sideeffect "ptrue p0.h, #1 \0Afcvt $0.d, p0/m, $1.h\0A", "=w,w,~{p0}"(<vscale x 8 x half> %in1)
 239 SVFLOAT_CVT_TEST(__SVFloat32_t,s,__SVFloat16_t,h);
 240 // CHECK: call <vscale x 4 x float> asm sideeffect "ptrue p0.h, #1 \0Afcvt $0.s, p0/m, $1.h\0A", "=w,w,~{p0}"(<vscale x 8 x half> %in1)
 241 SVFLOAT_CVT_TEST(__SVFloat32_t,s,__SVFloat64_t,d);
 242 // CHECK: call <vscale x 4 x float> asm sideeffect "ptrue p0.d, #1 \0Afcvt $0.s, p0/m, $1.d\0A", "=w,w,~{p0}"(<vscale x 2 x double> %in1)
 243 SVFLOAT_CVT_TEST(__SVFloat16_t,h,__SVFloat64_t,d);
 244 // CHECK: call <vscale x 8 x half> asm sideeffect "ptrue p0.d, #1 \0Afcvt $0.h, p0/m, $1.d\0A", "=w,w,~{p0}"(<vscale x 2 x double> %in1)
 245 SVFLOAT_CVT_TEST(__SVFloat16_t,h,__SVFloat32_t,s);
 246 // CHECK: call <vscale x 8 x half> asm sideeffect "ptrue p0.s, #1 \0Afcvt $0.h, p0/m, $1.s\0A", "=w,w,~{p0}"(<vscale x 4 x float> %in1)
 247
 248 //Test a mix of float and ints
 249 SVFLOAT_CVT_TEST(__SVInt16_t,h,__SVFloat32_t,s);
 250 // CHECK: call <vscale x 8 x i16> asm sideeffect "ptrue p0.s, #1 \0Afcvt $0.h, p0/m, $1.s\0A", "=w,w,~{p0}"(<vscale x 4 x float> %in1)
 251 SVFLOAT_CVT_TEST(__SVFloat16_t,s,__SVUint32_t,d);
 252 // CHECK: call <vscale x 8 x half> asm sideeffect "ptrue p0.d, #1 \0Afcvt $0.s, p0/m, $1.d\0A", "=w,w,~{p0}"(<vscale x 4 x i32> %in1)