Stephen Hines | 6a211c5 | 2014-07-21 00:49:56 -0700 | [diff] [blame] | 1 | Compiler-RT |
| 2 | ================================ |
| 3 | |
| 4 | This directory and its subdirectories contain source code for the compiler |
| 5 | support routines. |
| 6 | |
| 7 | Compiler-RT is open source software. You may freely distribute it under the |
| 8 | terms of the license agreement found in LICENSE.txt. |
| 9 | |
| 10 | ================================ |
| 11 | |
| 12 | This is a replacement library for libgcc. Each function is contained |
| 13 | in its own file. Each function has a corresponding unit test under |
| 14 | test/Unit. |
| 15 | |
| 16 | A rudimentary script to test each file is in the file called |
| 17 | test/Unit/test. |
| 18 | |
| 19 | Here is the specification for this library: |
| 20 | |
| 21 | http://gcc.gnu.org/onlinedocs/gccint/Libgcc.html#Libgcc |
| 22 | |
| 23 | Here is a synopsis of the contents of this library: |
| 24 | |
| 25 | typedef int si_int; |
| 26 | typedef unsigned su_int; |
| 27 | |
| 28 | typedef long long di_int; |
| 29 | typedef unsigned long long du_int; |
| 30 | |
| 31 | // Integral bit manipulation |
| 32 | |
| 33 | di_int __ashldi3(di_int a, si_int b); // a << b |
| 34 | ti_int __ashlti3(ti_int a, si_int b); // a << b |
| 35 | |
| 36 | di_int __ashrdi3(di_int a, si_int b); // a >> b arithmetic (sign fill) |
| 37 | ti_int __ashrti3(ti_int a, si_int b); // a >> b arithmetic (sign fill) |
| 38 | di_int __lshrdi3(di_int a, si_int b); // a >> b logical (zero fill) |
| 39 | ti_int __lshrti3(ti_int a, si_int b); // a >> b logical (zero fill) |
| 40 | |
| 41 | si_int __clzsi2(si_int a); // count leading zeros |
| 42 | si_int __clzdi2(di_int a); // count leading zeros |
| 43 | si_int __clzti2(ti_int a); // count leading zeros |
| 44 | si_int __ctzsi2(si_int a); // count trailing zeros |
| 45 | si_int __ctzdi2(di_int a); // count trailing zeros |
| 46 | si_int __ctzti2(ti_int a); // count trailing zeros |
| 47 | |
| 48 | si_int __ffsdi2(di_int a); // find least significant 1 bit |
| 49 | si_int __ffsti2(ti_int a); // find least significant 1 bit |
| 50 | |
| 51 | si_int __paritysi2(si_int a); // bit parity |
| 52 | si_int __paritydi2(di_int a); // bit parity |
| 53 | si_int __parityti2(ti_int a); // bit parity |
| 54 | |
| 55 | si_int __popcountsi2(si_int a); // bit population |
| 56 | si_int __popcountdi2(di_int a); // bit population |
| 57 | si_int __popcountti2(ti_int a); // bit population |
| 58 | |
| 59 | uint32_t __bswapsi2(uint32_t a); // a byteswapped, arm only |
| 60 | uint64_t __bswapdi2(uint64_t a); // a byteswapped, arm only |
| 61 | |
| 62 | // Integral arithmetic |
| 63 | |
| 64 | di_int __negdi2 (di_int a); // -a |
| 65 | ti_int __negti2 (ti_int a); // -a |
| 66 | di_int __muldi3 (di_int a, di_int b); // a * b |
| 67 | ti_int __multi3 (ti_int a, ti_int b); // a * b |
| 68 | si_int __divsi3 (si_int a, si_int b); // a / b signed |
| 69 | di_int __divdi3 (di_int a, di_int b); // a / b signed |
| 70 | ti_int __divti3 (ti_int a, ti_int b); // a / b signed |
| 71 | su_int __udivsi3 (su_int n, su_int d); // a / b unsigned |
| 72 | du_int __udivdi3 (du_int a, du_int b); // a / b unsigned |
| 73 | tu_int __udivti3 (tu_int a, tu_int b); // a / b unsigned |
| 74 | si_int __modsi3 (si_int a, si_int b); // a % b signed |
| 75 | di_int __moddi3 (di_int a, di_int b); // a % b signed |
| 76 | ti_int __modti3 (ti_int a, ti_int b); // a % b signed |
| 77 | su_int __umodsi3 (su_int a, su_int b); // a % b unsigned |
| 78 | du_int __umoddi3 (du_int a, du_int b); // a % b unsigned |
| 79 | tu_int __umodti3 (tu_int a, tu_int b); // a % b unsigned |
| 80 | du_int __udivmoddi4(du_int a, du_int b, du_int* rem); // a / b, *rem = a % b unsigned |
| 81 | tu_int __udivmodti4(tu_int a, tu_int b, tu_int* rem); // a / b, *rem = a % b unsigned |
| 82 | su_int __udivmodsi4(su_int a, su_int b, su_int* rem); // a / b, *rem = a % b unsigned |
| 83 | si_int __divmodsi4(si_int a, si_int b, si_int* rem); // a / b, *rem = a % b signed |
| 84 | |
| 85 | |
| 86 | |
| 87 | // Integral arithmetic with trapping overflow |
| 88 | |
| 89 | si_int __absvsi2(si_int a); // abs(a) |
| 90 | di_int __absvdi2(di_int a); // abs(a) |
| 91 | ti_int __absvti2(ti_int a); // abs(a) |
| 92 | |
| 93 | si_int __negvsi2(si_int a); // -a |
| 94 | di_int __negvdi2(di_int a); // -a |
| 95 | ti_int __negvti2(ti_int a); // -a |
| 96 | |
| 97 | si_int __addvsi3(si_int a, si_int b); // a + b |
| 98 | di_int __addvdi3(di_int a, di_int b); // a + b |
| 99 | ti_int __addvti3(ti_int a, ti_int b); // a + b |
| 100 | |
| 101 | si_int __subvsi3(si_int a, si_int b); // a - b |
| 102 | di_int __subvdi3(di_int a, di_int b); // a - b |
| 103 | ti_int __subvti3(ti_int a, ti_int b); // a - b |
| 104 | |
| 105 | si_int __mulvsi3(si_int a, si_int b); // a * b |
| 106 | di_int __mulvdi3(di_int a, di_int b); // a * b |
| 107 | ti_int __mulvti3(ti_int a, ti_int b); // a * b |
| 108 | |
| 109 | |
| 110 | // Integral arithmetic which returns if overflow |
| 111 | |
| 112 | si_int __mulosi4(si_int a, si_int b, int* overflow); // a * b, overflow set to one if result not in signed range |
| 113 | di_int __mulodi4(di_int a, di_int b, int* overflow); // a * b, overflow set to one if result not in signed range |
| 114 | ti_int __muloti4(ti_int a, ti_int b, int* overflow); // a * b, overflow set to |
| 115 | one if result not in signed range |
| 116 | |
| 117 | |
| 118 | // Integral comparison: a < b -> 0 |
| 119 | // a == b -> 1 |
| 120 | // a > b -> 2 |
| 121 | |
| 122 | si_int __cmpdi2 (di_int a, di_int b); |
| 123 | si_int __cmpti2 (ti_int a, ti_int b); |
| 124 | si_int __ucmpdi2(du_int a, du_int b); |
| 125 | si_int __ucmpti2(tu_int a, tu_int b); |
| 126 | |
| 127 | // Integral / floating point conversion |
| 128 | |
| 129 | di_int __fixsfdi( float a); |
| 130 | di_int __fixdfdi( double a); |
| 131 | di_int __fixxfdi(long double a); |
| 132 | |
| 133 | ti_int __fixsfti( float a); |
| 134 | ti_int __fixdfti( double a); |
| 135 | ti_int __fixxfti(long double a); |
| 136 | uint64_t __fixtfdi(long double input); // ppc only, doesn't match documentation |
| 137 | |
| 138 | su_int __fixunssfsi( float a); |
| 139 | su_int __fixunsdfsi( double a); |
| 140 | su_int __fixunsxfsi(long double a); |
| 141 | |
| 142 | du_int __fixunssfdi( float a); |
| 143 | du_int __fixunsdfdi( double a); |
| 144 | du_int __fixunsxfdi(long double a); |
| 145 | |
| 146 | tu_int __fixunssfti( float a); |
| 147 | tu_int __fixunsdfti( double a); |
| 148 | tu_int __fixunsxfti(long double a); |
| 149 | uint64_t __fixunstfdi(long double input); // ppc only |
| 150 | |
| 151 | float __floatdisf(di_int a); |
| 152 | double __floatdidf(di_int a); |
| 153 | long double __floatdixf(di_int a); |
| 154 | long double __floatditf(int64_t a); // ppc only |
| 155 | |
| 156 | float __floattisf(ti_int a); |
| 157 | double __floattidf(ti_int a); |
| 158 | long double __floattixf(ti_int a); |
| 159 | |
| 160 | float __floatundisf(du_int a); |
| 161 | double __floatundidf(du_int a); |
| 162 | long double __floatundixf(du_int a); |
| 163 | long double __floatunditf(uint64_t a); // ppc only |
| 164 | |
| 165 | float __floatuntisf(tu_int a); |
| 166 | double __floatuntidf(tu_int a); |
| 167 | long double __floatuntixf(tu_int a); |
| 168 | |
| 169 | // Floating point raised to integer power |
| 170 | |
| 171 | float __powisf2( float a, si_int b); // a ^ b |
| 172 | double __powidf2( double a, si_int b); // a ^ b |
| 173 | long double __powixf2(long double a, si_int b); // a ^ b |
| 174 | long double __powitf2(long double a, si_int b); // ppc only, a ^ b |
| 175 | |
| 176 | // Complex arithmetic |
| 177 | |
| 178 | // (a + ib) * (c + id) |
| 179 | |
| 180 | float _Complex __mulsc3( float a, float b, float c, float d); |
| 181 | double _Complex __muldc3(double a, double b, double c, double d); |
| 182 | long double _Complex __mulxc3(long double a, long double b, |
| 183 | long double c, long double d); |
| 184 | long double _Complex __multc3(long double a, long double b, |
| 185 | long double c, long double d); // ppc only |
| 186 | |
| 187 | // (a + ib) / (c + id) |
| 188 | |
| 189 | float _Complex __divsc3( float a, float b, float c, float d); |
| 190 | double _Complex __divdc3(double a, double b, double c, double d); |
| 191 | long double _Complex __divxc3(long double a, long double b, |
| 192 | long double c, long double d); |
| 193 | long double _Complex __divtc3(long double a, long double b, |
| 194 | long double c, long double d); // ppc only |
| 195 | |
| 196 | |
| 197 | // Runtime support |
| 198 | |
| 199 | // __clear_cache() is used to tell process that new instructions have been |
| 200 | // written to an address range. Necessary on processors that do not have |
| 201 | // a unified instruction and data cache. |
| 202 | void __clear_cache(void* start, void* end); |
| 203 | |
| 204 | // __enable_execute_stack() is used with nested functions when a trampoline |
| 205 | // function is written onto the stack and that page range needs to be made |
| 206 | // executable. |
| 207 | void __enable_execute_stack(void* addr); |
| 208 | |
| 209 | // __gcc_personality_v0() is normally only called by the system unwinder. |
| 210 | // C code (as opposed to C++) normally does not need a personality function |
| 211 | // because there are no catch clauses or destructors to be run. But there |
| 212 | // is a C language extension __attribute__((cleanup(func))) which marks local |
| 213 | // variables as needing the cleanup function "func" to be run when the |
| 214 | // variable goes out of scope. That includes when an exception is thrown, |
| 215 | // so a personality handler is needed. |
| 216 | _Unwind_Reason_Code __gcc_personality_v0(int version, _Unwind_Action actions, |
| 217 | uint64_t exceptionClass, struct _Unwind_Exception* exceptionObject, |
| 218 | _Unwind_Context_t context); |
| 219 | |
| 220 | // for use with some implementations of assert() in <assert.h> |
| 221 | void __eprintf(const char* format, const char* assertion_expression, |
| 222 | const char* line, const char* file); |
| 223 | |
| 224 | |
| 225 | |
| 226 | // Power PC specific functions |
| 227 | |
| 228 | // There is no C interface to the saveFP/restFP functions. They are helper |
| 229 | // functions called by the prolog and epilog of functions that need to save |
| 230 | // a number of non-volatile float point registers. |
| 231 | saveFP |
| 232 | restFP |
| 233 | |
| 234 | // PowerPC has a standard template for trampoline functions. This function |
| 235 | // generates a custom trampoline function with the specific realFunc |
| 236 | // and localsPtr values. |
| 237 | void __trampoline_setup(uint32_t* trampOnStack, int trampSizeAllocated, |
| 238 | const void* realFunc, void* localsPtr); |
| 239 | |
| 240 | // adds two 128-bit double-double precision values ( x + y ) |
| 241 | long double __gcc_qadd(long double x, long double y); |
| 242 | |
| 243 | // subtracts two 128-bit double-double precision values ( x - y ) |
| 244 | long double __gcc_qsub(long double x, long double y); |
| 245 | |
| 246 | // multiples two 128-bit double-double precision values ( x * y ) |
| 247 | long double __gcc_qmul(long double x, long double y); |
| 248 | |
| 249 | // divides two 128-bit double-double precision values ( x / y ) |
| 250 | long double __gcc_qdiv(long double a, long double b); |
| 251 | |
| 252 | |
| 253 | // ARM specific functions |
| 254 | |
| 255 | // There is no C interface to the switch* functions. These helper functions |
| 256 | // are only needed by Thumb1 code for efficient switch table generation. |
| 257 | switch16 |
| 258 | switch32 |
| 259 | switch8 |
| 260 | switchu8 |
| 261 | |
| 262 | // There is no C interface to the *_vfp_d8_d15_regs functions. There are |
| 263 | // called in the prolog and epilog of Thumb1 functions. When the C++ ABI use |
| 264 | // SJLJ for exceptions, each function with a catch clause or destuctors needs |
| 265 | // to save and restore all registers in it prolog and epliog. But there is |
| 266 | // no way to access vector and high float registers from thumb1 code, so the |
| 267 | // compiler must add call outs to these helper functions in the prolog and |
| 268 | // epilog. |
| 269 | restore_vfp_d8_d15_regs |
| 270 | save_vfp_d8_d15_regs |
| 271 | |
| 272 | |
| 273 | // Note: long ago ARM processors did not have floating point hardware support. |
| 274 | // Floating point was done in software and floating point parameters were |
| 275 | // passed in integer registers. When hardware support was added for floating |
| 276 | // point, new *vfp functions were added to do the same operations but with |
| 277 | // floating point parameters in floating point registers. |
| 278 | |
| 279 | // Undocumented functions |
| 280 | |
| 281 | float __addsf3vfp(float a, float b); // Appears to return a + b |
| 282 | double __adddf3vfp(double a, double b); // Appears to return a + b |
| 283 | float __divsf3vfp(float a, float b); // Appears to return a / b |
| 284 | double __divdf3vfp(double a, double b); // Appears to return a / b |
| 285 | int __eqsf2vfp(float a, float b); // Appears to return one |
| 286 | // iff a == b and neither is NaN. |
| 287 | int __eqdf2vfp(double a, double b); // Appears to return one |
| 288 | // iff a == b and neither is NaN. |
| 289 | double __extendsfdf2vfp(float a); // Appears to convert from |
| 290 | // float to double. |
| 291 | int __fixdfsivfp(double a); // Appears to convert from |
| 292 | // double to int. |
| 293 | int __fixsfsivfp(float a); // Appears to convert from |
| 294 | // float to int. |
| 295 | unsigned int __fixunssfsivfp(float a); // Appears to convert from |
| 296 | // float to unsigned int. |
| 297 | unsigned int __fixunsdfsivfp(double a); // Appears to convert from |
| 298 | // double to unsigned int. |
| 299 | double __floatsidfvfp(int a); // Appears to convert from |
| 300 | // int to double. |
| 301 | float __floatsisfvfp(int a); // Appears to convert from |
| 302 | // int to float. |
| 303 | double __floatunssidfvfp(unsigned int a); // Appears to convert from |
| 304 | // unisgned int to double. |
| 305 | float __floatunssisfvfp(unsigned int a); // Appears to convert from |
| 306 | // unisgned int to float. |
| 307 | int __gedf2vfp(double a, double b); // Appears to return __gedf2 |
| 308 | // (a >= b) |
| 309 | int __gesf2vfp(float a, float b); // Appears to return __gesf2 |
| 310 | // (a >= b) |
| 311 | int __gtdf2vfp(double a, double b); // Appears to return __gtdf2 |
| 312 | // (a > b) |
| 313 | int __gtsf2vfp(float a, float b); // Appears to return __gtsf2 |
| 314 | // (a > b) |
| 315 | int __ledf2vfp(double a, double b); // Appears to return __ledf2 |
| 316 | // (a <= b) |
| 317 | int __lesf2vfp(float a, float b); // Appears to return __lesf2 |
| 318 | // (a <= b) |
| 319 | int __ltdf2vfp(double a, double b); // Appears to return __ltdf2 |
| 320 | // (a < b) |
| 321 | int __ltsf2vfp(float a, float b); // Appears to return __ltsf2 |
| 322 | // (a < b) |
| 323 | double __muldf3vfp(double a, double b); // Appears to return a * b |
| 324 | float __mulsf3vfp(float a, float b); // Appears to return a * b |
| 325 | int __nedf2vfp(double a, double b); // Appears to return __nedf2 |
| 326 | // (a != b) |
| 327 | double __negdf2vfp(double a); // Appears to return -a |
| 328 | float __negsf2vfp(float a); // Appears to return -a |
| 329 | float __negsf2vfp(float a); // Appears to return -a |
| 330 | double __subdf3vfp(double a, double b); // Appears to return a - b |
| 331 | float __subsf3vfp(float a, float b); // Appears to return a - b |
| 332 | float __truncdfsf2vfp(double a); // Appears to convert from |
| 333 | // double to float. |
| 334 | int __unorddf2vfp(double a, double b); // Appears to return __unorddf2 |
| 335 | int __unordsf2vfp(float a, float b); // Appears to return __unordsf2 |
| 336 | |
| 337 | |
| 338 | Preconditions are listed for each function at the definition when there are any. |
| 339 | Any preconditions reflect the specification at |
| 340 | http://gcc.gnu.org/onlinedocs/gccint/Libgcc.html#Libgcc. |
| 341 | |
| 342 | Assumptions are listed in "int_lib.h", and in individual files. Where possible |
| 343 | assumptions are checked at compile time. |