diff options
| author | Joseph Myers <joseph@codesourcery.com> | 2012-05-07 19:13:08 +0000 |
|---|---|---|
| committer | Joseph Myers <joseph@codesourcery.com> | 2012-05-07 19:13:08 +0000 |
| commit | 495fd99f3a119e5c0c542ccc6cf9c93b1fb9e892 (patch) | |
| tree | 946a0ea8ff908e0770368c1d85e8d4355b5da415 /sysdeps/i386/fpu/e_expl.S | |
| parent | 6693d69429d92682bdb85d36ae8e4335c393c467 (diff) | |
| download | glibc-495fd99f3a119e5c0c542ccc6cf9c93b1fb9e892.zip glibc-495fd99f3a119e5c0c542ccc6cf9c93b1fb9e892.tar.gz glibc-495fd99f3a119e5c0c542ccc6cf9c93b1fb9e892.tar.bz2 | |
Fix x86/x86_64 expm1l inaccuracy and exceptions (bugs 13885, 13923).
Diffstat (limited to 'sysdeps/i386/fpu/e_expl.S')
| -rw-r--r-- | sysdeps/i386/fpu/e_expl.S | 45 |
1 files changed, 44 insertions, 1 deletions
diff --git a/sysdeps/i386/fpu/e_expl.S b/sysdeps/i386/fpu/e_expl.S index 9adf2a4..bab0a08 100644 --- a/sysdeps/i386/fpu/e_expl.S +++ b/sysdeps/i386/fpu/e_expl.S @@ -28,6 +28,10 @@ # define IEEE754_EXPL __ieee754_exp10l # define EXPL_FINITE __exp10l_finite # define FLDLOG fldl2t +#elif defined USE_AS_EXPM1L +# define IEEE754_EXPL __expm1l +# undef EXPL_FINITE +# define FLDLOG fldl2e #else # define IEEE754_EXPL __ieee754_expl # define EXPL_FINITE __expl_finite @@ -69,6 +73,12 @@ csat: .byte 0, 0, 0, 0, 0, 0, 0, 0x80, 0x0e, 0x40 .text ENTRY(IEEE754_EXPL) +#ifdef USE_AS_EXPM1L + movzwl 4+8(%esp), %eax + xorb $0x80, %ah // invert sign bit (now 1 is "positive") + cmpl $0xc006, %eax // is num positive and exp >= 6 (number is >= 128.0)? + jae HIDDEN_JUMPTARGET (__expl) // (if num is denormal, it is at least >= 64.0) +#endif fldt 4(%esp) /* I added the following ugly construct because expl(+-Inf) resulted in NaN. The ugliness results from the bright minds at Intel. @@ -78,7 +88,9 @@ ENTRY(IEEE754_EXPL) #ifdef PIC LOAD_PIC_REG (cx) #endif +#ifndef USE_AS_EXPM1L movzwl 4+8(%esp), %eax +#endif andl $0x7fff, %eax cmpl $0x400d, %eax jle 3f @@ -96,7 +108,16 @@ ENTRY(IEEE754_EXPL) andb $2, %ah jz 3f fchs -3: FLDLOG /* 1 log2(base) */ +3: +#ifdef USE_AS_EXPM1L + /* Test for +-0 as argument. */ + fstsw %ax + movb $0x45, %dh + andb %ah, %dh + cmpb $0x40, %dh + je 2f +#endif + FLDLOG /* 1 log2(base) */ fmul %st(1), %st /* 1 x log2(base) */ frndint /* 1 i */ fld %st(1) /* 2 x */ @@ -114,17 +135,39 @@ ENTRY(IEEE754_EXPL) fmul %st(4), %st /* 4 c1 * x */ faddp %st, %st(1) /* 3 f = f + c1 * x */ f2xm1 /* 3 2^(fract(x * log2(base))) - 1 */ +#ifdef USE_AS_EXPM1L + fstp %st(1) /* 2 */ + fscale /* 2 scale factor is st(1); base^x - 2^i */ + fxch /* 2 i */ + fld1 /* 3 1.0 */ + fscale /* 3 2^i */ + fld1 /* 4 1.0 */ + fsubrp %st, %st(1) /* 3 2^i - 1.0 */ + fstp %st(1) /* 2 */ + faddp %st, %st(1) /* 1 base^x - 1.0 */ +#else fld1 /* 4 1.0 */ faddp /* 3 2^(fract(x * log2(base))) */ fstp %st(1) /* 2 */ fscale /* 2 scale factor is st(1); base^x */ fstp %st(1) /* 1 */ +#endif fstp %st(1) /* 0 */ jmp 2f 1: testl $0x200, %eax /* Test sign. */ jz 2f /* If positive, jump. */ fstp %st +#ifdef USE_AS_EXPM1L + fld1 + fchs +#else fldz /* Set result to 0. */ +#endif 2: ret END(IEEE754_EXPL) +#ifdef USE_AS_EXPM1L +libm_hidden_def (__expm1l) +weak_alias (__expm1l, expm1l) +#else strong_alias (IEEE754_EXPL, EXPL_FINITE) +#endif |