diff options
| author | Adhemerval Zanella <azanella@linux.vnet.ibm.com> | 2014-02-27 09:45:41 -0600 |
|---|---|---|
| committer | Adhemerval Zanella <azanella@linux.vnet.ibm.com> | 2014-02-27 12:58:33 -0600 |
| commit | 4393fc119c34e97519b9b7a4fc94066b283be452 (patch) | |
| tree | dffac0629930499b0f88886fca4f1b094ef74fa5 /sysdeps/powerpc/powerpc64/power8/fpu/s_isinf.S | |
| parent | 487972aea52004f604c2878c8c9d3e77670f2c32 (diff) | |
| download | glibc-4393fc119c34e97519b9b7a4fc94066b283be452.tar.gz glibc-4393fc119c34e97519b9b7a4fc94066b283be452.tar.xz glibc-4393fc119c34e97519b9b7a4fc94066b283be452.zip | |
PowerPC: Optimized isinf/isinff for POWER8
This patch add a optimized isinf/isinff implementation for POWER8 using the new Move From VSR Doubleword instruction to gains some cycles from FP to GRP register move.
Diffstat (limited to 'sysdeps/powerpc/powerpc64/power8/fpu/s_isinf.S')
| -rw-r--r-- | sysdeps/powerpc/powerpc64/power8/fpu/s_isinf.S | 61 |
1 files changed, 61 insertions, 0 deletions
diff --git a/sysdeps/powerpc/powerpc64/power8/fpu/s_isinf.S b/sysdeps/powerpc/powerpc64/power8/fpu/s_isinf.S new file mode 100644 index 0000000000..0e92af8d9f --- /dev/null +++ b/sysdeps/powerpc/powerpc64/power8/fpu/s_isinf.S @@ -0,0 +1,61 @@ +/* isinf(). PowerPC64/POWER8 version. + Copyright (C) 2014 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include <math_ldbl_opt.h> + +#define MFVSRD_R3_V1 .byte 0x7c,0x23,0x00,0x66 /* mfvsrd r3,vs1 */ + +/* int [r3] __isinf([fp1] x) */ + +EALIGN (__isinf, 4, 0) + CALL_MCOUNT 0 + MFVSRD_R3_V1 + lis r9,0x7ff0 /* r9 = 0x7ff0 */ + rldicl r10,r3,0,1 /* r10 = r3 & (0x8000000000000000) */ + sldi r9,r9,32 /* r9 = r9 << 52 */ + cmpd cr7,r10,r9 /* fp1 & 0x7ff0000000000000 ? */ + beq cr7,L(inf) + li r3,0 /* Not inf */ + blr +L(inf): + sradi r3,r3,63 /* r3 = r3 >> 63 */ + ori r3,r3,1 /* r3 = r3 | 0x1 */ + blr +END (__isinf) + +hidden_def (__isinf) +weak_alias (__isinf, isinf) + +/* It turns out that the 'double' version will also always work for + single-precision. */ +strong_alias (__isinf, __isinff) +hidden_def (__isinff) +weak_alias (__isinff, isinff) + +#ifdef NO_LONG_DOUBLE +strong_alias (__isinf, __isinfl) +weak_alias (__isinf, isinfl) +#endif + +#ifndef IS_IN_libm +# if LONG_DOUBLE_COMPAT(libc, GLIBC_2_0) +compat_symbol (libc, __isinf, __isinfl, GLIBC_2_0); +compat_symbol (libc, isinf, isinfl, GLIBC_2_0); +# endif +#endif |