diff options
Diffstat (limited to 'REORG.TODO/soft-fp/testit.c')
-rw-r--r-- | REORG.TODO/soft-fp/testit.c | 788 |
1 files changed, 788 insertions, 0 deletions
diff --git a/REORG.TODO/soft-fp/testit.c b/REORG.TODO/soft-fp/testit.c new file mode 100644 index 0000000000..1aaf8b45c8 --- /dev/null +++ b/REORG.TODO/soft-fp/testit.c @@ -0,0 +1,788 @@ +#include <stdio.h> +#include <stdlib.h> +#include <math.h> + +#include "soft-fp.h" +#include "single.h" +#include "double.h" + +#include <fpu_control.h> + +/*======================================================================*/ +/* declarations for the functions we are testing */ + +double __adddf3(double, double); +double __subdf3(double, double); +double __muldf3(double, double); +double __divdf3(double, double); +double __negdf2(double); +double __sqrtdf2(double); +double __negdf3(double a, double dummy) { return __negdf2(a); } +double __sqrtdf3(double a, double dummy) { return __sqrtdf2(a); } + +float __addsf3(float, float); +float __subsf3(float, float); +float __mulsf3(float, float); +float __divsf3(float, float); +float __negsf2(float); +float __sqrtsf2(float); +float __negsf3(float a, float dummy) { return __negsf2(a); } +float __sqrtsf3(float a, float dummy) { return __sqrtsf2(a); } + +int __fixdfsi(double); +int __fixsfsi(float); +double __floatsidf(int); +float __floatsisf(int); +double __extendsfdf2(float); +float __truncdfsf2(double); + +int __eqdf2(double, double); +int __nedf2(double, double); +int __gtdf2(double, double); +int __gedf2(double, double); +int __ltdf2(double, double); +int __ledf2(double, double); + +int __eqsf2(float, float); +int __nesf2(float, float); +int __gtsf2(float, float); +int __gesf2(float, float); +int __ltsf2(float, float); +int __lesf2(float, float); + +/*======================================================================*/ +/* definitions for functions we are checking against */ + +double r_adddf3(double a, double b) { return a + b; } +double r_subdf3(double a, double b) { return a - b; } +double r_muldf3(double a, double b) { return a * b; } +double r_divdf3(double a, double b) { return a / b; } +double r_negdf3(double a, double b) { return -a; } +double sqrt(double x); +double r_sqrtdf3(double a, double b) { return sqrt(a); } + +float r_addsf3(float a, float b) { return a + b; } +float r_subsf3(float a, float b) { return a - b; } +float r_mulsf3(float a, float b) { return a * b; } +float r_divsf3(float a, float b) { return a / b; } +float r_negsf3(float a, float b) { return -a; } +float sqrtf(float x); +float r_sqrtsf3(float a, float b) { return sqrtf(a); } + +int r_fixdfsi(double a) { return (int)a; } +int r_fixsfsi(float a) { return (int)a; } +double r_floatsidf(int a) { return (double)a; } +float r_floatsisf(int a) { return (float)a; } +double r_extendsfdf2(float a) { return (double)a; } +float r_truncdfsf2(double a) { return (float)a; } + +int r_eqdf2(double a, double b) { return !(a == b); } +int r_nedf2(double a, double b) { return a != b; } +int r_gtdf2(double a, double b) { return a > b; } +int r_gedf2(double a, double b) { return (a >= b) - 1; } +int r_ltdf2(double a, double b) { return -(a < b); } +int r_ledf2(double a, double b) { return 1 - (a <= b); } + +int r_eqsf2(float a, float b) { return !(a == b); } +int r_nesf2(float a, float b) { return a != b; } +int r_gtsf2(float a, float b) { return a > b; } +int r_gesf2(float a, float b) { return (a >= b) - 1; } +int r_ltsf2(float a, float b) { return -(a < b); } +int r_lesf2(float a, float b) { return 1 - (a <= b); } + +/*======================================================================*/ + +void print_float(float x) +{ + union _FP_UNION_S ux; + ux.flt = x; + printf("%-20.8e %X %02X %06lX", + x, ux.bits.sign, ux.bits.exp, (unsigned long)ux.bits.frac); +} + +void print_double(double x) +{ + union _FP_UNION_D ux; + ux.flt = x; +#if _FP_W_TYPE_SIZE < _FP_FRACBITS_D + printf("%-30.18e %X %04X %06lX%08lX", + x, ux.bits.sign, ux.bits.exp, + (unsigned long)ux.bits.frac1, (unsigned long)ux.bits.frac0); +#else + printf("%-30.18e %X %04X %014lX", + x, ux.bits.sign, ux.bits.exp, + (unsigned long)ux.bits.frac); +#endif +} + +float rand_float(void) +{ + union { + union _FP_UNION_S u; + int i; + } u; + + u.i = lrand48() << 1; + + if (u.u.bits.exp == _FP_EXPMAX_S) + u.u.bits.exp--; + else if (u.u.bits.exp == 0 && u.u.bits.frac != 0) + u.u.bits.exp++; + + return u.u.flt; +} + + +double rand_double(void) +{ + union { + union _FP_UNION_D u; + int i[2]; + } u; + + u.i[0] = lrand48() << 1; + u.i[1] = lrand48() << 1; + + if (u.u.bits.exp == _FP_EXPMAX_D) + u.u.bits.exp--; +#if _FP_W_TYPE_SIZE < _FP_FRACBITS_D + else if (u.u.bits.exp == 0 && !(u.u.bits.frac0 == 0 && u.u.bits.frac1 == 0)) + u.u.bits.exp++; +#else + else if (u.u.bits.exp == 0 && u.u.bits.frac != 0) + u.u.bits.exp++; +#endif + + return u.u.flt; +} + +#define NSPECIALS 10 + +float gen_special_float(int i) +{ + FP_DECL_EX; + FP_DECL_S(X); + float x; + + switch (i & ~1) + { + case 0: + X_c = FP_CLS_NAN; X_f = 0x1234; + break; + case 2: + X_c = FP_CLS_NAN; X_f = 0x1; + break; + case 4: + X_c = FP_CLS_INF; + break; + case 6: + X_c = FP_CLS_ZERO; + break; + case 8: + X_c = FP_CLS_NORMAL; X_e = 0; + X_f = 0x4321; + break; + } + X_s = (i & 1); + + FP_PACK_S(x, X); + return x; +} + +double gen_special_double(int i) +{ + FP_DECL_EX; + FP_DECL_D(X); + double x; + + switch (i & ~1) + { + case 0: + X_c = FP_CLS_NAN; +#if _FP_W_TYPE_SIZE < _FP_FRACBITS_D + __FP_FRAC_SET_2(X, _FP_QNANNEGATEDP ? 0 : _FP_QNANBIT_D, 0x1234); +#else + _FP_FRAC_SET_1(X, (_FP_QNANNEGATEDP ? 0 : _FP_QNANBIT_D) | 0x1234); +#endif + break; + case 2: + X_c = FP_CLS_NAN; +#if _FP_W_TYPE_SIZE < _FP_FRACBITS_D + __FP_FRAC_SET_2(X, _FP_QNANNEGATEDP ? 0 : _FP_QNANBIT_D, 0x1); +#else + _FP_FRAC_SET_1(X, (_FP_QNANNEGATEDP ? 0 : _FP_QNANBIT_D) | 0x1); +#endif + break; + case 4: + X_c = FP_CLS_INF; + break; + case 6: + X_c = FP_CLS_ZERO; + break; + case 8: + X_c = FP_CLS_NORMAL; X_e = 0; +#if _FP_W_TYPE_SIZE < _FP_FRACBITS_D + __FP_FRAC_SET_2(X, 0, 0x87654321); +#else + _FP_FRAC_SET_1(X, 0x87654321); +#endif + break; + } + X_s = (i & 1); + + FP_PACK_D(x, X); + return x; +} + +float build_float(const char *s, const char *e, const char *f) +{ + union _FP_UNION_S u; + + u.bits.sign = strtoul(s, 0, 16); + u.bits.exp = strtoul(e, 0, 16); + u.bits.frac = strtoul(f, 0, 16); + + return u.flt; +} + +double build_double(const char *s, const char *e, const char *f) +{ + union _FP_UNION_D u; + + u.bits.sign = strtoul(s, 0, 16); + u.bits.exp = strtoul(e, 0, 16); +#if _FP_W_TYPE_SIZE < _FP_FRACBITS_D + { + size_t len = strlen(f)+1; + char *dup = memcpy(alloca(len), f, len); + char *low = dup + len - _FP_W_TYPE_SIZE/4 - 1; + + u.bits.frac0 = strtoul(low, 0, 16); + *low = 0; + u.bits.frac1 = strtoul(dup, 0, 16); + } +#else + u.bits.frac = strtoul(f, 0, 16); +#endif + + return u.flt; +} + +/*======================================================================*/ + +fpu_control_t fcw0, fcw1; + +void test_float_arith(float (*tf)(float, float), + float (*rf)(float, float), + float x, float y) +{ + float tr, rr; + rr = (*rf)(x, y); + tr = (*tf)(x, y); + if (memcmp(&tr, &rr, sizeof(float)) != 0) + { + fputs("error:\n\tx = ", stdout); print_float(x); + fputs("\n\ty = ", stdout); print_float(y); + fputs("\n\ttrue = ", stdout); print_float(rr); + fputs("\n\tfalse = ", stdout); print_float(tr); + putchar('\n'); + } +} + +void test_double_arith(double (*tf)(double, double), + double (*rf)(double, double), + double x, double y) +{ + double tr, rr; +#ifdef __i386__ + /* Don't worry. Even this does not make it error free + on ia32. If the result is denormal, it will not + honour the double precision and generate bad results + anyway. On the other side, who wants to use ia32 + for IEEE math? I don't. */ + _FPU_GETCW(fcw0); + fcw1 = ((fcw0 & ~_FPU_EXTENDED) | _FPU_DOUBLE); + _FPU_SETCW(fcw1); +#endif + rr = (*rf)(x, y); +#ifdef __i386__ + _FPU_SETCW(fcw0); +#endif + tr = (*tf)(x, y); + if (memcmp(&tr, &rr, sizeof(double)) != 0) + { + fputs("error:\n\tx = ", stdout); print_double(x); + fputs("\n\ty = ", stdout); print_double(y); + fputs("\n\ttrue = ", stdout); print_double(rr); + fputs("\n\tfalse = ", stdout); print_double(tr); + putchar('\n'); + } +} + +void test_float_double_conv(float x) +{ + double tr, rr; + rr = r_extendsfdf2(x); + tr = __extendsfdf2(x); + if (memcmp(&tr, &rr, sizeof(double)) != 0) + { + fputs("error:\n\tx = ", stdout); print_float(x); + fputs("\n\ttrue = ", stdout); print_double(rr); + fputs("\n\tfalse = ", stdout); print_double(tr); + putchar('\n'); + } +} + +void test_double_float_conv(double x) +{ + float tr, rr; + rr = r_truncdfsf2(x); + tr = __truncdfsf2(x); + if (memcmp(&tr, &rr, sizeof(float)) != 0) + { + fputs("error:\n\tx = ", stdout); print_double(x); + fputs("\n\ttrue = ", stdout); print_float(rr); + fputs("\n\tfalse = ", stdout); print_float(tr); + putchar('\n'); + } +} + +void test_int_float_conv(int x) +{ + float tr, rr; + rr = r_floatsisf(x); + tr = __floatsisf(x); + if (memcmp(&tr, &rr, sizeof(float)) != 0) + { + printf("error\n\tx = %d", x); + fputs("\n\ttrue = ", stdout); print_float(rr); + fputs("\n\tfalse = ", stdout); print_float(tr); + putchar('\n'); + } +} + +void test_int_double_conv(int x) +{ + double tr, rr; + rr = r_floatsidf(x); + tr = __floatsidf(x); + if (memcmp(&tr, &rr, sizeof(double)) != 0) + { + printf("error\n\tx = %d", x); + fputs("\n\ttrue = ", stdout); print_double(rr); + fputs("\n\tfalse = ", stdout); print_double(tr); + putchar('\n'); + } +} + +void test_float_int_conv(float x) +{ + int tr, rr; + rr = r_fixsfsi(x); + tr = __fixsfsi(x); + if (rr != tr) + { + fputs("error:\n\tx = ", stdout); print_float(x); + printf("\n\ttrue = %d\n\tfalse = %d\n", rr, tr); + } +} + +void test_double_int_conv(double x) +{ + int tr, rr; + rr = r_fixsfsi(x); + tr = __fixsfsi(x); + if (rr != tr) + { + fputs("error:\n\tx = ", stdout); print_double(x); + printf("\n\ttrue = %d\n\tfalse = %d\n", rr, tr); + } +} + +int eq0(int x) { return x == 0; } +int ne0(int x) { return x != 0; } +int le0(int x) { return x <= 0; } +int lt0(int x) { return x < 0; } +int ge0(int x) { return x >= 0; } +int gt0(int x) { return x > 0; } + +void test_float_cmp(int (*tf)(float, float), + int (*rf)(float, float), + int (*cmp0)(int), + float x, float y) +{ + int tr, rr; + rr = (*rf)(x, y); + tr = (*tf)(x, y); + if (cmp0(rr) != cmp0(tr)) + { + fputs("error:\n\tx = ", stdout); print_float(x); + fputs("\n\ty = ", stdout); print_float(y); + printf("\n\ttrue = %d\n\tfalse = %d\n", rr, tr); + } +} + +void test_double_cmp(int (*tf)(double, double), + int (*rf)(double, double), + int (*cmp0)(int), + double x, double y) +{ + int tr, rr; + rr = (*rf)(x, y); + tr = (*tf)(x, y); + if (cmp0(rr) != cmp0(tr)) + { + fputs("error:\n\tx = ", stdout); print_double(x); + fputs("\n\ty = ", stdout); print_double(y); + printf("\n\ttrue = %d\n\tfalse = %d\n", rr, tr); + } +} + + +/*======================================================================*/ + + +int main(int ac, char **av) +{ +#ifdef __alpha__ + __ieee_set_fp_control(0); +#endif + av++, ac--; + switch (*(*av)++) + { + { + float (*r)(float, float); + float (*t)(float, float); + + do { + case 'a': r = r_addsf3; t = __addsf3; break; + case 's': r = r_subsf3; t = __subsf3; break; + case 'm': r = r_mulsf3; t = __mulsf3; break; + case 'd': r = r_divsf3; t = __divsf3; break; + case 'r': r = r_sqrtsf3; t = __sqrtsf3; break; + case 'j': r = r_negsf3; t = __negsf3; break; + } while (0); + + switch (*(*av)++) + { + case 'n': + { + int count = (ac > 1 ? atoi(av[1]) : 100); + while (count--) + test_float_arith(t, r, rand_float(), rand_float()); + } + break; + + case 's': + { + int i, j; + for (i = 0; i < NSPECIALS; i++) + for (j = 0; j < NSPECIALS; j++) + test_float_arith(t, r, gen_special_float(i), + gen_special_float(j)); + } + break; + + case 0: + if (ac < 7) abort(); + test_float_arith(t, r, build_float(av[1], av[2], av[3]), + build_float(av[4], av[5], av[6])); + break; + } + } + break; + + { + double (*r)(double, double); + double (*t)(double, double); + + do { + case 'A': r = r_adddf3; t = __adddf3; break; + case 'S': r = r_subdf3; t = __subdf3; break; + case 'M': r = r_muldf3; t = __muldf3; break; + case 'D': r = r_divdf3; t = __divdf3; break; + case 'R': r = r_sqrtdf3; t = __sqrtdf3; break; + case 'J': r = r_negdf3; t = __negdf3; break; + } while (0); + + switch (*(*av)++) + { + case 'n': + { + int count = (ac > 1 ? atoi(av[1]) : 100); + while (count--) + test_double_arith(t, r, rand_double(), rand_double()); + } + break; + + case 's': + { + int i, j; + for (i = 0; i < NSPECIALS; i++) + for (j = 0; j < NSPECIALS; j++) + test_double_arith(t, r, gen_special_double(i), + gen_special_double(j)); + } + break; + + case 0: + if (ac < 7) abort(); + test_double_arith(t, r, build_double(av[1], av[2], av[3]), + build_double(av[4], av[5], av[6])); + break; + } + } + break; + + case 'c': + switch (*(*av)++) + { + case 'n': + { + int count = (ac > 1 ? atoi(av[1]) : 100); + while (count--) + test_float_double_conv(rand_float()); + } + break; + + case 's': + { + int i; + for (i = 0; i < NSPECIALS; i++) + test_float_double_conv(gen_special_float(i)); + } + break; + + case 0: + if (ac < 4) abort(); + test_float_double_conv(build_float(av[1], av[2], av[3])); + break; + } + break; + + case 'C': + switch (*(*av)++) + { + case 'n': + { + int count = (ac > 1 ? atoi(av[1]) : 100); + while (count--) + test_double_float_conv(rand_double()); + } + break; + + case 's': + { + int i; + for (i = 0; i < NSPECIALS; i++) + test_double_float_conv(gen_special_double(i)); + } + break; + + case 0: + if (ac < 4) abort(); + test_double_float_conv(build_double(av[1], av[2], av[3])); + break; + } + break; + + case 'i': + switch (*(*av)++) + { + case 'n': + { + int count = (ac > 1 ? atoi(av[1]) : 100); + while (count--) + test_int_float_conv(lrand48() << 1); + } + break; + + case 0: + if (ac < 2) abort(); + test_int_float_conv(strtol(av[1], 0, 0)); + break; + } + break; + + case 'I': + switch (*(*av)++) + { + case 'n': + { + int count = (ac > 1 ? atoi(av[1]) : 100); + while (count--) + test_int_double_conv(lrand48() << 1); + } + break; + + case 0: + if (ac < 2) abort(); + test_int_double_conv(strtol(av[1], 0, 0)); + break; + } + break; + + case 'f': + switch (*(*av)++) + { + case 'n': + { + int count = (ac > 1 ? atoi(av[1]) : 100); + while (count--) + test_float_int_conv(rand_float()); + } + break; + + case 's': + { + int i; + for (i = 0; i < NSPECIALS; i++) + test_float_int_conv(gen_special_float(i)); + } + break; + + case 0: + if (ac < 4) abort(); + test_float_int_conv(build_float(av[1], av[2], av[3])); + break; + } + break; + + case 'F': + switch (*(*av)++) + { + case 'n': + { + int count = (ac > 1 ? atoi(av[1]) : 100); + while (count--) + test_double_int_conv(rand_double()); + } + break; + + case 's': + { + int i; + for (i = 0; i < NSPECIALS; i++) + test_double_int_conv(gen_special_double(i)); + } + break; + + case 0: + if (ac < 4) abort(); + test_double_int_conv(build_double(av[1], av[2], av[3])); + break; + } + break; + + { + int (*r)(float, float); + int (*t)(float, float); + int (*c)(int); + + do { + case 'e': r = r_eqsf2; t = __eqsf2; c = eq0; break; + case 'n': r = r_nesf2; t = __nesf2; c = ne0; break; + case 'l': + switch (*(*av)++) + { + case 'e': r = r_lesf2; t = __lesf2; c = le0; break; + case 't': r = r_ltsf2; t = __ltsf2; c = lt0; break; + } + break; + case 'g': + switch (*(*av)++) + { + case 'e': r = r_gesf2; t = __gesf2; c = ge0; break; + case 't': r = r_gtsf2; t = __gtsf2; c = gt0; break; + } + break; + } while (0); + + switch (*(*av)++) + { + case 'n': + { + int count = (ac > 1 ? atoi(av[1]) : 100); + while (count--) + test_float_cmp(t, r, c, rand_float(), rand_float()); + } + break; + + case 's': + { + int i, j; + for (i = 0; i < NSPECIALS; i++) + for (j = 0; j < NSPECIALS; j++) + test_float_cmp(t, r, c, gen_special_float(i), + gen_special_float(j)); + } + break; + + case 0: + if (ac < 7) abort(); + test_float_cmp(t, r, c, build_float(av[1], av[2], av[3]), + build_float(av[4], av[5], av[6])); + break; + } + } + break; + + { + int (*r)(double, double); + int (*t)(double, double); + int (*c)(int); + + do { + case 'E': r = r_eqdf2; t = __eqdf2; c = eq0; break; + case 'N': r = r_nedf2; t = __nedf2; c = ne0; break; + case 'L': + switch (*(*av)++) + { + case 'E': r = r_ledf2; t = __ledf2; c = le0; break; + case 'T': r = r_ltdf2; t = __ltdf2; c = lt0; break; + } + break; + case 'G': + switch (*(*av)++) + { + case 'E': r = r_gedf2; t = __gedf2; c = ge0; break; + case 'T': r = r_gtdf2; t = __gtdf2; c = gt0; break; + } + break; + } while (0); + + switch (*(*av)++) + { + case 'n': + { + int count = (ac > 1 ? atoi(av[1]) : 100); + while (count--) + test_double_cmp(t, r, c, rand_double(), rand_double()); + } + break; + + case 's': + { + int i, j; + for (i = 0; i < NSPECIALS; i++) + for (j = 0; j < NSPECIALS; j++) + test_double_cmp(t, r, c, gen_special_double(i), + gen_special_double(j)); + } + break; + + case 0: + if (ac < 7) abort(); + test_double_cmp(t, r, c, build_double(av[1], av[2], av[3]), + build_double(av[4], av[5], av[6])); + break; + } + } + break; + + default: + abort(); + } + + return 0; +} |