URL
https://opencores.org/ocsvn/openrisc_me/openrisc_me/trunk
Subversion Repositories openrisc_me
[/] [openrisc/] [trunk/] [gnu-src/] [gcc-4.5.1/] [gcc/] [testsuite/] [gcc.dg/] [torture/] [builtin-math-6.c] - Rev 298
Compare with Previous | Blame | View Log
/* Copyright (C) 2009 Free Software Foundation. Verify that folding of built-in complex math functions with constant arguments is correctly performed by the compiler. Origin: Kaveh R. Ghazi, January 28, 2009. */ /* { dg-do link } */ /* All references to link_error should go away at compile-time. The first number is the line number and the second is the value number among several tests. These appear in the tree dump file and aid in debugging. */ extern void link_error(int, int); #define CONJ(X) __builtin_conjf(X) /* Return TRUE if the signs of floating point values X and Y are not equal. This is important when comparing signed zeros. */ #define CKSGN_F(X,Y) \ (__builtin_copysignf(1,(X)) != __builtin_copysignf(1,(Y))) #define CKSGN(X,Y) \ (__builtin_copysign(1,(X)) != __builtin_copysign(1,(Y))) #define CKSGN_L(X,Y) \ (__builtin_copysignl(1,(X)) != __builtin_copysignl(1,(Y))) /* Return TRUE if signs of the real parts, and the signs of the imaginary parts, of X and Y are not equal. */ #define COMPLEX_CKSGN_F(X,Y) \ (CKSGN_F(__real__ (X), __real__ (Y)) || CKSGN_F (__imag__ (X), __imag__ (Y))) #define COMPLEX_CKSGN(X,Y) \ (CKSGN(__real__ (X), __real__ (Y)) || CKSGN (__imag__ (X), __imag__ (Y))) #define COMPLEX_CKSGN_L(X,Y) \ (CKSGN_L(__real__ (X), __real__ (Y)) || CKSGN_L (__imag__ (X), __imag__ (Y))) /* For complex numbers, test that FUNC(ARG) == (RES). */ #define TESTIT_COMPLEX(VAL_NUM, FUNC, ARG, RES) do { \ if (__builtin_##FUNC##f(ARG) != (RES) \ || COMPLEX_CKSGN_F(__builtin_##FUNC##f(ARG), (RES))) \ link_error(__LINE__, VAL_NUM); \ if (__builtin_##FUNC(ARG) != (RES) \ || COMPLEX_CKSGN(__builtin_##FUNC(ARG), (RES))) \ link_error(__LINE__, VAL_NUM); \ if (__builtin_##FUNC##l(ARG) != (RES) \ || COMPLEX_CKSGN_L(__builtin_##FUNC##l(ARG), (RES))) \ link_error(__LINE__, VAL_NUM); \ } while (0) /* For complex numbers, call the TESTIT_COMPLEX macro for all combinations of neg and conj. */ #define TESTIT_COMPLEX_ALLNEG(FUNC, ARG, RES1, RES2, RES3, RES4) do { \ TESTIT_COMPLEX(1, FUNC, (_Complex float)(ARG), RES1); \ TESTIT_COMPLEX(2, FUNC, -CONJ(ARG), RES2); \ TESTIT_COMPLEX(3, FUNC, CONJ(ARG), RES3); \ TESTIT_COMPLEX(4, FUNC, -(_Complex float)(ARG), RES4); \ } while (0) /* For complex numbers, call the TESTIT_COMPLEX_R macro for all combinations of neg and conj. */ #define TESTIT_COMPLEX_R_ALLNEG(FUNC, ARG, RES1, RES2, RES3, RES4) do { \ TESTIT_COMPLEX_R(1, FUNC, (_Complex float)(ARG), RES1); \ TESTIT_COMPLEX_R(2, FUNC, -CONJ(ARG), RES2); \ TESTIT_COMPLEX_R(3, FUNC, CONJ(ARG), RES3); \ TESTIT_COMPLEX_R(4, FUNC, -(_Complex float)(ARG), RES4); \ } while (0) /* For complex numbers, test that FUNC(ARG0, ARG1) == (RES). */ #define TESTIT_COMPLEX2(VAL_NUM, FUNC, ARG0, ARG1, RES) do { \ if (__builtin_##FUNC##f(ARG0, ARG1) != (RES) \ || COMPLEX_CKSGN_F(__builtin_##FUNC##f(ARG0, ARG1), (RES))) \ link_error(__LINE__, VAL_NUM); \ if (__builtin_##FUNC(ARG0, ARG1) != (RES) \ || COMPLEX_CKSGN(__builtin_##FUNC(ARG0, ARG1), (RES))) \ link_error(__LINE__, VAL_NUM); \ if (__builtin_##FUNC##l(ARG0, ARG1) != (RES) \ || COMPLEX_CKSGN_L(__builtin_##FUNC##l(ARG0, ARG1), (RES))) \ link_error(__LINE__, VAL_NUM); \ } while (0) /* For complex numbers, call the TESTIT_COMPLEX2 macro for all combinations of neg and conj. */ #define TESTIT_COMPLEX2_ALLNEG(FUNC, ARG0, ARG1, RES1, RES2, RES3, RES4, RES5,\ RES6, RES7, RES8, RES9, RES10, RES11, RES12, RES13, RES14, RES15, RES16) do{ \ TESTIT_COMPLEX2(1, FUNC, (_Complex float)(ARG0),(_Complex float)(ARG1), RES1);\ TESTIT_COMPLEX2(2, FUNC, (_Complex float)(ARG0),CONJ(ARG1), RES2); \ TESTIT_COMPLEX2(3, FUNC, (_Complex float)(ARG0),-(_Complex float)(ARG1), RES3); \ TESTIT_COMPLEX2(4, FUNC, (_Complex float)(ARG0),-CONJ(ARG1), RES4); \ TESTIT_COMPLEX2(5, FUNC, -(_Complex float)(ARG0),(_Complex float)(ARG1), RES5); \ TESTIT_COMPLEX2(6, FUNC, -(_Complex float)(ARG0),CONJ(ARG1), RES6); \ TESTIT_COMPLEX2(7, FUNC, -(_Complex float)(ARG0),-(_Complex float)(ARG1), RES7); \ TESTIT_COMPLEX2(8, FUNC, -(_Complex float)(ARG0),-CONJ(ARG1), RES8); \ TESTIT_COMPLEX2(9, FUNC, CONJ(ARG0),(_Complex float)(ARG1), RES9); \ TESTIT_COMPLEX2(10, FUNC, CONJ(ARG0),CONJ(ARG1), RES10); \ TESTIT_COMPLEX2(11, FUNC, CONJ(ARG0),-(_Complex float)(ARG1), RES11); \ TESTIT_COMPLEX2(12, FUNC, CONJ(ARG0),-CONJ(ARG1), RES12); \ TESTIT_COMPLEX2(13, FUNC, -CONJ(ARG0),(_Complex float)(ARG1), RES13); \ TESTIT_COMPLEX2(14, FUNC, -CONJ(ARG0),CONJ(ARG1), RES14); \ TESTIT_COMPLEX2(15, FUNC, -CONJ(ARG0),-(_Complex float)(ARG1), RES15); \ TESTIT_COMPLEX2(16, FUNC, -CONJ(ARG0),-CONJ(ARG1), RES16); \ } while (0) /* Return TRUE if X differs from EXPECTED by more than 1%. If EXPECTED is zero, then any difference may return TRUE. We don't worry about signed zeros. */ #define DIFF1PCT_F(X,EXPECTED) \ (__builtin_fabsf((X)-(EXPECTED)) * 100 > __builtin_fabsf(EXPECTED)) #define DIFF1PCT(X,EXPECTED) \ (__builtin_fabs((X)-(EXPECTED)) * 100 > __builtin_fabs(EXPECTED)) #define DIFF1PCT_L(X,EXPECTED) \ (__builtin_fabsl((X)-(EXPECTED)) * 100 > __builtin_fabsl(EXPECTED)) /* Return TRUE if complex value X differs from EXPECTED by more than 1% in either the real or imaginary parts. */ #define COMPLEX_DIFF1PCT_F(X,EXPECTED) \ (DIFF1PCT_F(__real__ (X), __real__ (EXPECTED)) \ || DIFF1PCT_F(__imag__ (X), __imag__ (EXPECTED))) #define COMPLEX_DIFF1PCT(X,EXPECTED) \ (DIFF1PCT(__real__ (X), __real__ (EXPECTED)) \ || DIFF1PCT(__imag__ (X), __imag__ (EXPECTED))) #define COMPLEX_DIFF1PCT_L(X,EXPECTED) \ (DIFF1PCT_L(__real__ (X), __real__ (EXPECTED)) \ || DIFF1PCT_L(__imag__ (X), __imag__ (EXPECTED))) /* Range test, for complex numbers check that FUNC(ARG) is within 1% of RES. This is NOT a test for accuracy to the last-bit, we're merely checking that we get relatively sane results. I.e. the GCC builtin is hooked up to the correct MPC function call. We first check the magnitude and then the sign. */ #define TESTIT_COMPLEX_R(VAL_NUM, FUNC, ARG, RES) do { \ if (COMPLEX_DIFF1PCT_F (__builtin_##FUNC##f(ARG), (RES)) \ || COMPLEX_CKSGN_F(__builtin_##FUNC##f(ARG), (RES))) \ link_error(__LINE__, VAL_NUM); \ if (COMPLEX_DIFF1PCT (__builtin_##FUNC(ARG), (RES)) \ || COMPLEX_CKSGN(__builtin_##FUNC(ARG), (RES))) \ link_error(__LINE__, VAL_NUM); \ if (COMPLEX_DIFF1PCT (__builtin_##FUNC(ARG), (RES)) \ || COMPLEX_CKSGN(__builtin_##FUNC(ARG), (RES))) \ link_error(__LINE__, VAL_NUM); \ } while (0) /* Range test, for complex numbers check that FUNC(ARG0, ARG1) is within 1% of RES. This is NOT a test for accuracy to the last-bit, we're merely checking that we get relatively sane results. I.e. the GCC builtin is hooked up to the correct MPC function call. We first check the magnitude and then the sign. */ #define TESTIT_COMPLEX_R2(VAL_NUM, FUNC, ARG0, ARG1, RES) do { \ if (COMPLEX_DIFF1PCT_F (__builtin_##FUNC##f(ARG0, ARG1), (RES)) \ || COMPLEX_CKSGN_F (__builtin_##FUNC##f(ARG0, ARG1), (RES))) \ link_error(__LINE__, VAL_NUM); \ if (COMPLEX_DIFF1PCT (__builtin_##FUNC(ARG0, ARG1), (RES)) \ || COMPLEX_CKSGN (__builtin_##FUNC(ARG0, ARG1), (RES))) \ link_error(__LINE__, VAL_NUM); \ if (COMPLEX_DIFF1PCT_L (__builtin_##FUNC##l(ARG0, ARG1), (RES)) \ || COMPLEX_CKSGN_L (__builtin_##FUNC##l(ARG0, ARG1), (RES))) \ link_error(__LINE__, VAL_NUM); \ } while (0) /* For complex numbers, call the TESTIT_COMPLEX_R2 macro for all combinations of neg and conj. */ #define TESTIT_COMPLEX_R2_ALLNEG(FUNC, ARG0, ARG1, RES1, RES2, RES3, RES4, RES5,\ RES6, RES7, RES8, RES9, RES10, RES11, RES12, RES13, RES14, RES15, RES16) do{ \ TESTIT_COMPLEX_R2(1, FUNC, (_Complex float)(ARG0),(_Complex float)(ARG1), RES1);\ TESTIT_COMPLEX_R2(2, FUNC, (_Complex float)(ARG0),CONJ(ARG1), RES2); \ TESTIT_COMPLEX_R2(3, FUNC, (_Complex float)(ARG0),-(_Complex float)(ARG1), RES3); \ TESTIT_COMPLEX_R2(4, FUNC, (_Complex float)(ARG0),-CONJ(ARG1), RES4); \ TESTIT_COMPLEX_R2(5, FUNC, -(_Complex float)(ARG0),(_Complex float)(ARG1), RES5); \ TESTIT_COMPLEX_R2(6, FUNC, -(_Complex float)(ARG0),CONJ(ARG1), RES6); \ TESTIT_COMPLEX_R2(7, FUNC, -(_Complex float)(ARG0),-(_Complex float)(ARG1), RES7); \ TESTIT_COMPLEX_R2(8, FUNC, -(_Complex float)(ARG0),-CONJ(ARG1), RES8); \ TESTIT_COMPLEX_R2(9, FUNC, CONJ(ARG0),(_Complex float)(ARG1), RES9); \ TESTIT_COMPLEX_R2(10, FUNC, CONJ(ARG0),CONJ(ARG1), RES10); \ TESTIT_COMPLEX_R2(11, FUNC, CONJ(ARG0),-(_Complex float)(ARG1), RES11); \ TESTIT_COMPLEX_R2(12, FUNC, CONJ(ARG0),-CONJ(ARG1), RES12); \ TESTIT_COMPLEX_R2(13, FUNC, -CONJ(ARG0),(_Complex float)(ARG1), RES13); \ TESTIT_COMPLEX_R2(14, FUNC, -CONJ(ARG0),CONJ(ARG1), RES14); \ TESTIT_COMPLEX_R2(15, FUNC, -CONJ(ARG0),-(_Complex float)(ARG1), RES15); \ TESTIT_COMPLEX_R2(16, FUNC, -CONJ(ARG0),-CONJ(ARG1), RES16); \ } while (0) int main (void) { TESTIT_COMPLEX (1, cacos, 1, CONJ(0)); TESTIT_COMPLEX_R (1, cacos, -1, CONJ(3.141593F)); TESTIT_COMPLEX (1, cacos, CONJ(1), 0); TESTIT_COMPLEX_R (1, cacos, CONJ(-1), 3.141593F); TESTIT_COMPLEX_R_ALLNEG (cacos, 3.45678F + 2.34567FI, 0.60971F - 2.11780FI, 2.531875F - 2.117800FI, 0.60971F + 2.11780FI, 2.531875F + 2.117800FI); TESTIT_COMPLEX_ALLNEG (casin, 0, 0, -CONJ(0), CONJ(0), CONJ(-0.F)); TESTIT_COMPLEX_R_ALLNEG (casin, 3.45678F + 2.34567FI, 0.96107F + 2.11780FI, -0.96107F + 2.11780FI, 0.96107F - 2.11780FI, -0.96107F - 2.11780FI); TESTIT_COMPLEX_ALLNEG (catan, 0, 0, -CONJ(0), CONJ(0), CONJ(-0.F)); TESTIT_COMPLEX_R_ALLNEG (catan, 3.45678F + 2.34567FI, 1.37188F + 0.12997FI, -1.37188F + 0.12997FI, 1.37188F - 0.12997FI, -1.37188F - 0.12997FI); TESTIT_COMPLEX (1, cacosh, 1, 0); TESTIT_COMPLEX_R (1, cacosh, -1, 3.141593FI); TESTIT_COMPLEX (1, cacosh, CONJ(1), CONJ(0)); TESTIT_COMPLEX_R (1, cacosh, CONJ(-1), CONJ(3.141593FI)); TESTIT_COMPLEX_R_ALLNEG (cacosh, 3.45678F + 2.34567FI, 2.11780F + 0.60971FI, 2.11780F + 2.531875FI, 2.11780F - 0.60971FI, 2.11780F - 2.531875FI); TESTIT_COMPLEX_ALLNEG (casinh, 0, 0, -CONJ(0), CONJ(0), CONJ(-0.F)); TESTIT_COMPLEX_R_ALLNEG (casinh, 3.45678F + 2.34567FI, 2.12836F + 0.58310FI, -2.12836F + 0.58310FI, 2.12836F - 0.58310FI, -2.12836F - 0.58310FI); TESTIT_COMPLEX_ALLNEG (catanh, 0, 0, -CONJ(0), CONJ(0), CONJ(-0.F)); TESTIT_COMPLEX_R_ALLNEG (catanh, 3.45678F + 2.34567FI, 0.19693F + 1.43190FI, -0.19693F + 1.43190FI, 0.19693F - 1.43190FI, -0.19693F - 1.43190FI); TESTIT_COMPLEX_ALLNEG (csin, 0, 0, -0.F, CONJ(0), CONJ(-0.F)); TESTIT_COMPLEX_R_ALLNEG (csin, 3.45678F + 2.34567FI, -1.633059F - 4.917448FI, 1.633059F - 4.917448FI, -1.633059F + 4.917448FI, 1.633059F + 4.917448FI); TESTIT_COMPLEX_ALLNEG (ccos, 0, CONJ(1), 1, 1, CONJ(1)); TESTIT_COMPLEX_R_ALLNEG (ccos, 3.45678F + 2.34567FI, -5.008512F + 1.603367FI, -5.008512F - 1.603367FI, -5.008512F - 1.603367FI, -5.008512F + 1.603367FI); TESTIT_COMPLEX_ALLNEG (ctan, 0, 0, -0.F, CONJ(0), CONJ(-0.F)); TESTIT_COMPLEX_R_ALLNEG (ctan, 3.45678F + 2.34567FI, 0.010657F + 0.985230FI, -0.010657F + 0.985230FI, 0.010657F - 0.985230FI, -0.010657F - 0.985230FI); TESTIT_COMPLEX_ALLNEG (csinh, 0, 0, -0.F, CONJ(0), CONJ(-0.F)); TESTIT_COMPLEX_R_ALLNEG (csinh, 3.45678F + 2.34567FI, -11.083178F + 11.341487FI, 11.083178F +11.341487FI, -11.083178F - 11.341487FI, 11.083178F -11.341487FI); TESTIT_COMPLEX_ALLNEG (ccosh, 0, 1, CONJ(1), CONJ(1), 1); TESTIT_COMPLEX_R_ALLNEG (ccosh, 3.45678F + 2.34567FI, -11.105238F + 11.318958FI,-11.105238F -11.318958FI, -11.105238F - 11.318958FI,-11.105238F +11.318958FI); TESTIT_COMPLEX_ALLNEG (ctanh, 0, 0, -0.F, CONJ(0), CONJ(-0.F)); TESTIT_COMPLEX_R_ALLNEG (ctanh, 3.45678F + 2.34567FI, 1.000040F - 0.001988FI, -1.000040F - 0.001988FI, 1.000040F + 0.001988FI, -1.000040F + 0.001988FI); TESTIT_COMPLEX (1, clog, 1, 0); TESTIT_COMPLEX_R (1, clog, -1, 3.141593FI); TESTIT_COMPLEX (1, clog, CONJ(1), CONJ(0)); TESTIT_COMPLEX_R (1, clog, CONJ(-1), CONJ(3.141593FI)); TESTIT_COMPLEX_R_ALLNEG (clog, 3.45678F + 2.34567FI, 1.429713F + 0.596199FI, 1.429713F + 2.545394FI, 1.429713F - 0.596199FI, 1.429713F - 2.545394FI); TESTIT_COMPLEX_ALLNEG (csqrt, 0, 0, 0, CONJ(0), CONJ(0)); TESTIT_COMPLEX_R_ALLNEG (csqrt, 3.45678F + 2.34567FI, 1.953750F + 0.600299FI, 0.600299F + 1.953750FI, 1.953750F - 0.600299FI, 0.600299F - 1.953750FI); TESTIT_COMPLEX2_ALLNEG (cpow, 1, 0, 1, 1, CONJ(1), CONJ(1), CONJ(1), CONJ(1), 1, 1, CONJ(1), CONJ(1), 1, 1, 1, 1, CONJ(1), CONJ(1)); TESTIT_COMPLEX2_ALLNEG (cpow, 1.FI, 0, 1, 1, CONJ(1), CONJ(1), CONJ(1), CONJ(1), 1, 1, CONJ(1), CONJ(1), 1, 1, 1, 1, CONJ(1), CONJ(1)); TESTIT_COMPLEX_R2_ALLNEG (cpow, 2, 3, 8, 8, CONJ(1/8.F), CONJ(1/8.F), CONJ(-8), CONJ(-8), -1/8.F, -1/8.F, CONJ(8), CONJ(8), 1/8.F, 1/8.F, -8, -8, CONJ(-1/8.F), CONJ(-1/8.F)); TESTIT_COMPLEX_R2_ALLNEG (cpow, 3, 4, 81, 81, CONJ(1/81.F), CONJ(1/81.F), CONJ(81), CONJ(81), 1/81.F, 1/81.F, CONJ(81), CONJ(81), 1/81.F, 1/81.F, 81, 81, CONJ(1/81.F), CONJ(1/81.F)); TESTIT_COMPLEX_R2_ALLNEG (cpow, 3, 5, 243, 243, CONJ(1/243.F), CONJ(1/243.F), CONJ(-243), CONJ(-243), -1/243.F, -1/243.F, CONJ(243), CONJ(243), 1/243.F, 1/243.F, -243, -243, CONJ(-1/243.F), CONJ(-1/243.F)); TESTIT_COMPLEX_R2_ALLNEG (cpow, 4, 2, 16, 16, CONJ(1/16.F), CONJ(1/16.F), CONJ(16), CONJ(16), 1/16.F, 1/16.F, CONJ(16), CONJ(16), 1/16.F, 1/16.F, 16, 16, CONJ(1/16.F), CONJ(1/16.F)); TESTIT_COMPLEX_R2_ALLNEG (cpow, 1.5, 3, 3.375F, 3.375F, CONJ(1/3.375F), CONJ(1/3.375F), CONJ(-3.375F), CONJ(-3.375F), -1/3.375F, -1/3.375F, CONJ(3.375F), CONJ(3.375F), 1/3.375F, 1/3.375F, -3.375F, -3.375F, CONJ(-1/3.375F), CONJ(-1/3.375F)); TESTIT_COMPLEX2 (1, cpow, 16, 0.25F, 2); TESTIT_COMPLEX_R2 (1, cpow, 3.45678F + 2.34567FI, 1.23456 + 4.56789FI, 0.212485F + 0.319304FI); TESTIT_COMPLEX_R2 (1, cpow, 3.45678F - 2.34567FI, 1.23456 + 4.56789FI, 78.576402F + -41.756208FI); TESTIT_COMPLEX_R2 (1, cpow, -1.23456F + 2.34567FI, 2.34567 - 1.23456FI, -110.629847F + -57.021655FI); TESTIT_COMPLEX_R2 (1, cpow, -1.23456F - 2.34567FI, 2.34567 - 1.23456FI, 0.752336F + 0.199095FI); return 0; }