qiurui
/
glibc
mirror of git://sourceware.org/git/glibc.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

math: Optimize fma call on asinpif 

The fma is required only for x == +/-0x1.6371e8p-4f in FE_TOWARDZERO
to provide correctly rounded results.

Checked on x86_64-linux-gnu and aarch64-linux-gnu.

Reviewed-by: Wilco Dijkstra  <Wilco.Dijkstra@arm.com>
This commit is contained in:
Adhemerval Zanella 2025-10-10 09:50:21 -03:00
parent fab32b6526
commit 82a4f50b4e
3 changed files with 59 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
math/auto-libm-test-in
View File

@ -524,6 +524,8 @@ asinpi 0x1.f1c012p-1
asinpi -0x1.8805060cb885cp-3 asinpi -0x1.8805060cb885cp-3
asinpi 0x8.14d7e32b5c44642p-4 asinpi 0x8.14d7e32b5c44642p-4
asinpi -0xa.7ca6c96caefe80b9d757de58a578p-4 asinpi -0xa.7ca6c96caefe80b9d757de58a578p-4
asinpi 0x1.6371e8p-4
asinpi -0x1.6371e8p-4
atan inf atan inf
atan -inf atan -inf

50
math/auto-libm-test-out-asinpi
View File

@ -2780,3 +2780,53 @@ asinpi -0xa.7ca6c96caefe80b9d757de58a578p-4
= asinpi tonearest ibm128 -0xa.7ca6c96caefe80b9d757de58a8p-4 : -0x3.a3e55379cf8d0f73aac00cc2e5p-4 : inexact-ok = asinpi tonearest ibm128 -0xa.7ca6c96caefe80b9d757de58a8p-4 : -0x3.a3e55379cf8d0f73aac00cc2e5p-4 : inexact-ok
= asinpi towardzero ibm128 -0xa.7ca6c96caefe80b9d757de58a8p-4 : -0x3.a3e55379cf8d0f73aac00cc2e4p-4 : inexact-ok = asinpi towardzero ibm128 -0xa.7ca6c96caefe80b9d757de58a8p-4 : -0x3.a3e55379cf8d0f73aac00cc2e4p-4 : inexact-ok
= asinpi upward ibm128 -0xa.7ca6c96caefe80b9d757de58a8p-4 : -0x3.a3e55379cf8d0f73aac00cc2e4p-4 : inexact-ok = asinpi upward ibm128 -0xa.7ca6c96caefe80b9d757de58a8p-4 : -0x3.a3e55379cf8d0f73aac00cc2e4p-4 : inexact-ok
asinpi 0x1.6371e8p-4
= asinpi downward binary32 0x1.6371e8p-4 : 0x7.148bcp-8 : inexact-ok
= asinpi tonearest binary32 0x1.6371e8p-4 : 0x7.148bc8p-8 : inexact-ok
= asinpi towardzero binary32 0x1.6371e8p-4 : 0x7.148bcp-8 : inexact-ok
= asinpi upward binary32 0x1.6371e8p-4 : 0x7.148bc8p-8 : inexact-ok
= asinpi downward binary64 0x1.6371e8p-4 : 0x7.148bc7fffff78p-8 : inexact-ok
= asinpi tonearest binary64 0x1.6371e8p-4 : 0x7.148bc7fffff7cp-8 : inexact-ok
= asinpi towardzero binary64 0x1.6371e8p-4 : 0x7.148bc7fffff78p-8 : inexact-ok
= asinpi upward binary64 0x1.6371e8p-4 : 0x7.148bc7fffff7cp-8 : inexact-ok
= asinpi downward intel96 0x1.6371e8p-4 : 0x7.148bc7fffff7af9p-8 : inexact-ok
= asinpi tonearest intel96 0x1.6371e8p-4 : 0x7.148bc7fffff7af98p-8 : inexact-ok
= asinpi towardzero intel96 0x1.6371e8p-4 : 0x7.148bc7fffff7af9p-8 : inexact-ok
= asinpi upward intel96 0x1.6371e8p-4 : 0x7.148bc7fffff7af98p-8 : inexact-ok
= asinpi downward m68k96 0x1.6371e8p-4 : 0x7.148bc7fffff7af9p-8 : inexact-ok
= asinpi tonearest m68k96 0x1.6371e8p-4 : 0x7.148bc7fffff7af98p-8 : inexact-ok
= asinpi towardzero m68k96 0x1.6371e8p-4 : 0x7.148bc7fffff7af9p-8 : inexact-ok
= asinpi upward m68k96 0x1.6371e8p-4 : 0x7.148bc7fffff7af98p-8 : inexact-ok
= asinpi downward binary128 0x1.6371e8p-4 : 0x7.148bc7fffff7af94c63520731f08p-8 : inexact-ok
= asinpi tonearest binary128 0x1.6371e8p-4 : 0x7.148bc7fffff7af94c63520731f08p-8 : inexact-ok
= asinpi towardzero binary128 0x1.6371e8p-4 : 0x7.148bc7fffff7af94c63520731f08p-8 : inexact-ok
= asinpi upward binary128 0x1.6371e8p-4 : 0x7.148bc7fffff7af94c63520731f0cp-8 : inexact-ok
= asinpi downward ibm128 0x1.6371e8p-4 : 0x7.148bc7fffff7af94c63520731ep-8 : inexact-ok
= asinpi tonearest ibm128 0x1.6371e8p-4 : 0x7.148bc7fffff7af94c63520732p-8 : inexact-ok
= asinpi towardzero ibm128 0x1.6371e8p-4 : 0x7.148bc7fffff7af94c63520731ep-8 : inexact-ok
= asinpi upward ibm128 0x1.6371e8p-4 : 0x7.148bc7fffff7af94c63520732p-8 : inexact-ok
asinpi -0x1.6371e8p-4
= asinpi downward binary32 -0x1.6371e8p-4 : -0x7.148bc8p-8 : inexact-ok
= asinpi tonearest binary32 -0x1.6371e8p-4 : -0x7.148bc8p-8 : inexact-ok
= asinpi towardzero binary32 -0x1.6371e8p-4 : -0x7.148bcp-8 : inexact-ok
= asinpi upward binary32 -0x1.6371e8p-4 : -0x7.148bcp-8 : inexact-ok
= asinpi downward binary64 -0x1.6371e8p-4 : -0x7.148bc7fffff7cp-8 : inexact-ok
= asinpi tonearest binary64 -0x1.6371e8p-4 : -0x7.148bc7fffff7cp-8 : inexact-ok
= asinpi towardzero binary64 -0x1.6371e8p-4 : -0x7.148bc7fffff78p-8 : inexact-ok
= asinpi upward binary64 -0x1.6371e8p-4 : -0x7.148bc7fffff78p-8 : inexact-ok
= asinpi downward intel96 -0x1.6371e8p-4 : -0x7.148bc7fffff7af98p-8 : inexact-ok
= asinpi tonearest intel96 -0x1.6371e8p-4 : -0x7.148bc7fffff7af98p-8 : inexact-ok
= asinpi towardzero intel96 -0x1.6371e8p-4 : -0x7.148bc7fffff7af9p-8 : inexact-ok
= asinpi upward intel96 -0x1.6371e8p-4 : -0x7.148bc7fffff7af9p-8 : inexact-ok
= asinpi downward m68k96 -0x1.6371e8p-4 : -0x7.148bc7fffff7af98p-8 : inexact-ok
= asinpi tonearest m68k96 -0x1.6371e8p-4 : -0x7.148bc7fffff7af98p-8 : inexact-ok
= asinpi towardzero m68k96 -0x1.6371e8p-4 : -0x7.148bc7fffff7af9p-8 : inexact-ok
= asinpi upward m68k96 -0x1.6371e8p-4 : -0x7.148bc7fffff7af9p-8 : inexact-ok
= asinpi downward binary128 -0x1.6371e8p-4 : -0x7.148bc7fffff7af94c63520731f0cp-8 : inexact-ok
= asinpi tonearest binary128 -0x1.6371e8p-4 : -0x7.148bc7fffff7af94c63520731f08p-8 : inexact-ok
= asinpi towardzero binary128 -0x1.6371e8p-4 : -0x7.148bc7fffff7af94c63520731f08p-8 : inexact-ok
= asinpi upward binary128 -0x1.6371e8p-4 : -0x7.148bc7fffff7af94c63520731f08p-8 : inexact-ok
= asinpi downward ibm128 -0x1.6371e8p-4 : -0x7.148bc7fffff7af94c63520732p-8 : inexact-ok
= asinpi tonearest ibm128 -0x1.6371e8p-4 : -0x7.148bc7fffff7af94c63520732p-8 : inexact-ok
= asinpi towardzero ibm128 -0x1.6371e8p-4 : -0x7.148bc7fffff7af94c63520731ep-8 : inexact-ok
= asinpi upward ibm128 -0x1.6371e8p-4 : -0x7.148bc7fffff7af94c63520731ep-8 : inexact-ok

9
sysdeps/ieee754/flt-32/s_asinpif.c
View File

@ -79,8 +79,13 @@ __asinpif (float x)
c0 += c2 * z2; c0 += c2 * z2;
c4 += c6 * z2; c4 += c6 * z2;
c0 += c4 * z4; c0 += c4 * z4;
double r = fma (-c0, copysign (f, x), copysign (0.5, x)); #ifndef __FP_FAST_FMA
return r; /* The fma is required only for x == 0x1.6371e8p-4f in FE_TOWARDZERO
to provide correctly rounded results. */
if (__glibc_likely (ax != 0x1.6371e8p-4f))
return copysign (0.5, x) - c0 * copysign (f, x);
#endif
return fma (-c0, copysign (f, x), copysign (0.5, x));
} }
} }
libm_alias_float (__asinpi, asinpi) libm_alias_float (__asinpi, asinpi)