diff options
author | Joseph Myers <joseph@codesourcery.com> | 2012-05-06 18:23:44 +0000 |
---|---|---|
committer | Joseph Myers <joseph@codesourcery.com> | 2012-05-06 18:23:44 +0000 |
commit | d8b82cad1b525bdcbfff88d218c7c45032e4a3af (patch) | |
tree | 77e88bf7642e69d74d4df3d771ffbb4274937b31 /sysdeps/i386/fpu | |
parent | 6c23e11c4dd036e65073d4e4d6bc971a445deaea (diff) | |
download | glibc-d8b82cad1b525bdcbfff88d218c7c45032e4a3af.tar.gz glibc-d8b82cad1b525bdcbfff88d218c7c45032e4a3af.tar.xz glibc-d8b82cad1b525bdcbfff88d218c7c45032e4a3af.zip |
Fix exp10 inaccuracy and exceptions (bugs 13884, 13914).
Diffstat (limited to 'sysdeps/i386/fpu')
-rw-r--r-- | sysdeps/i386/fpu/e_exp10l.S | 41 | ||||
-rw-r--r-- | sysdeps/i386/fpu/e_expl.S | 37 |
2 files changed, 31 insertions, 47 deletions
diff --git a/sysdeps/i386/fpu/e_exp10l.S b/sysdeps/i386/fpu/e_exp10l.S index 04ec8001d9..d843e2b5e8 100644 --- a/sysdeps/i386/fpu/e_exp10l.S +++ b/sysdeps/i386/fpu/e_exp10l.S @@ -1,39 +1,2 @@ -/* - * Written by Ulrich Drepper <drepper@cygnus.com>. - */ - -#include <machine/asm.h> - -/* 10^x = 2^(x * log2l(10)) */ -ENTRY(__ieee754_exp10l) - fldt 4(%esp) -/* I added the following ugly construct because expl(+-Inf) resulted - in NaN. The ugliness results from the bright minds at Intel. - For the i686 the code can be written better. - -- drepper@cygnus.com. */ - fxam /* Is NaN or +-Inf? */ - fstsw %ax - movb $0x45, %dh - andb %ah, %dh - cmpb $0x05, %dh - je 1f /* Is +-Inf, jump. */ - fldl2t - fmulp /* x * log2(10) */ - fld %st - frndint /* int(x * log2(10)) */ - fsubr %st,%st(1) /* fract(x * log2(10)) */ - fxch - f2xm1 /* 2^(fract(x * log2(10))) - 1 */ - fld1 - faddp /* 2^(fract(x * log2(10))) */ - fscale /* e^x */ - fstp %st(1) - ret - -1: testl $0x200, %eax /* Test sign. */ - jz 2f /* If positive, jump. */ - fstp %st - fldz /* Set result to 0. */ -2: ret -END (__ieee754_exp10l) -strong_alias (__ieee754_exp10l, __exp10l_finite) +#define USE_AS_EXP10L +#include <e_expl.S> diff --git a/sysdeps/i386/fpu/e_expl.S b/sysdeps/i386/fpu/e_expl.S index 45c4d07539..9adf2a489e 100644 --- a/sysdeps/i386/fpu/e_expl.S +++ b/sysdeps/i386/fpu/e_expl.S @@ -24,9 +24,29 @@ #include <machine/asm.h> +#ifdef USE_AS_EXP10L +# define IEEE754_EXPL __ieee754_exp10l +# define EXPL_FINITE __exp10l_finite +# define FLDLOG fldl2t +#else +# define IEEE754_EXPL __ieee754_expl +# define EXPL_FINITE __expl_finite +# define FLDLOG fldl2e +#endif + .section .rodata.cst16,"aM",@progbits,16 .p2align 4 +#ifdef USE_AS_EXP10L + ASM_TYPE_DIRECTIVE(c0,@object) +c0: .byte 0, 0, 0, 0, 0, 0, 0x9a, 0xd4, 0x00, 0x40 + .byte 0, 0, 0, 0, 0, 0 + ASM_SIZE_DIRECTIVE(c0) + ASM_TYPE_DIRECTIVE(c1,@object) +c1: .byte 0x58, 0x92, 0xfc, 0x15, 0x37, 0x9a, 0x97, 0xf0, 0xef, 0x3f + .byte 0, 0, 0, 0, 0, 0 + ASM_SIZE_DIRECTIVE(c1) +#else ASM_TYPE_DIRECTIVE(c0,@object) c0: .byte 0, 0, 0, 0, 0, 0, 0xaa, 0xb8, 0xff, 0x3f .byte 0, 0, 0, 0, 0, 0 @@ -35,6 +55,7 @@ c0: .byte 0, 0, 0, 0, 0, 0, 0xaa, 0xb8, 0xff, 0x3f c1: .byte 0x20, 0xfa, 0xee, 0xc2, 0x5f, 0x70, 0xa5, 0xec, 0xed, 0x3f .byte 0, 0, 0, 0, 0, 0 ASM_SIZE_DIRECTIVE(c1) +#endif ASM_TYPE_DIRECTIVE(csat,@object) csat: .byte 0, 0, 0, 0, 0, 0, 0, 0x80, 0x0e, 0x40 .byte 0, 0, 0, 0, 0, 0 @@ -47,7 +68,7 @@ csat: .byte 0, 0, 0, 0, 0, 0, 0, 0x80, 0x0e, 0x40 #endif .text -ENTRY(__ieee754_expl) +ENTRY(IEEE754_EXPL) fldt 4(%esp) /* I added the following ugly construct because expl(+-Inf) resulted in NaN. The ugliness results from the bright minds at Intel. @@ -75,8 +96,8 @@ ENTRY(__ieee754_expl) andb $2, %ah jz 3f fchs -3: fldl2e /* 1 log2(e) */ - fmul %st(1), %st /* 1 x log2(e) */ +3: FLDLOG /* 1 log2(base) */ + fmul %st(1), %st /* 1 x log2(base) */ frndint /* 1 i */ fld %st(1) /* 2 x */ frndint /* 2 xi */ @@ -92,11 +113,11 @@ ENTRY(__ieee754_expl) fldt MO(c1) /* 4 */ fmul %st(4), %st /* 4 c1 * x */ faddp %st, %st(1) /* 3 f = f + c1 * x */ - f2xm1 /* 3 2^(fract(x * log2(e))) - 1 */ + f2xm1 /* 3 2^(fract(x * log2(base))) - 1 */ fld1 /* 4 1.0 */ - faddp /* 3 2^(fract(x * log2(e))) */ + faddp /* 3 2^(fract(x * log2(base))) */ fstp %st(1) /* 2 */ - fscale /* 2 scale factor is st(1); e^x */ + fscale /* 2 scale factor is st(1); base^x */ fstp %st(1) /* 1 */ fstp %st(1) /* 0 */ jmp 2f @@ -105,5 +126,5 @@ ENTRY(__ieee754_expl) fstp %st fldz /* Set result to 0. */ 2: ret -END(__ieee754_expl) -strong_alias (__ieee754_expl, __expl_finite) +END(IEEE754_EXPL) +strong_alias (IEEE754_EXPL, EXPL_FINITE) |