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/x86_64/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/x86_64/fpu')
| -rw-r--r-- | sysdeps/x86_64/fpu/e_exp10l.S | 2 | ||||
| -rw-r--r-- | sysdeps/x86_64/fpu/e_expl.S | 37 | ||||
| -rw-r--r-- | sysdeps/x86_64/fpu/libm-test-ulps | 9 |
3 files changed, 40 insertions, 8 deletions
diff --git a/sysdeps/x86_64/fpu/e_exp10l.S b/sysdeps/x86_64/fpu/e_exp10l.S new file mode 100644 index 0000000000..d843e2b5e8 --- /dev/null +++ b/sysdeps/x86_64/fpu/e_exp10l.S @@ -0,0 +1,2 @@ +#define USE_AS_EXP10L +#include <e_expl.S> diff --git a/sysdeps/x86_64/fpu/e_expl.S b/sysdeps/x86_64/fpu/e_expl.S index d497b28973..fd613f91d3 100644 --- a/sysdeps/x86_64/fpu/e_expl.S +++ b/sysdeps/x86_64/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 8(%rsp) /* I added the following ugly construct because expl(+-Inf) resulted in NaN. The ugliness results from the bright minds at Intel. @@ -72,8 +93,8 @@ 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 */ @@ -89,11 +110,11 @@ 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 @@ -102,5 +123,5 @@ 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) diff --git a/sysdeps/x86_64/fpu/libm-test-ulps b/sysdeps/x86_64/fpu/libm-test-ulps index 70b091f547..f33c07f326 100644 --- a/sysdeps/x86_64/fpu/libm-test-ulps +++ b/sysdeps/x86_64/fpu/libm-test-ulps @@ -1287,6 +1287,12 @@ idouble: 2 ifloat: 1 ildouble: 1 ldouble: 1 +Test "exp10 (-305) == 1.0e-305": +double: 1 +idouble: 1 +Test "exp10 (-36) == 1.0e-36": +double: 1 +idouble: 1 Test "exp10 (0.75) == 5.62341325190349080394951039776481231": double: 1 float: 1 @@ -1301,6 +1307,9 @@ idouble: 6 ifloat: 2 ildouble: 8 ldouble: 8 +Test "exp10 (36) == 1.0e36": +double: 1 +idouble: 1 # exp_downward Test "exp_downward (1) == e": |