diff options
| author | H.J. Lu <hjl.tools@gmail.com> | 2017-10-22 08:09:08 -0700 |
|---|---|---|
| committer | H.J. Lu <hjl.tools@gmail.com> | 2017-10-22 08:10:18 -0700 |
| commit | 6089a3ee24cede17e9443aef0aa72fa1a0ba1548 (patch) | |
| tree | c160e567478cc34bf93e845362968562317a41f2 /sysdeps/i386/fpu/e_log2f.S | |
| parent | 80bb5935632faf8a8e94daf976340e549be9d4c5 (diff) | |
| download | glibc-6089a3ee24cede17e9443aef0aa72fa1a0ba1548.tar.gz glibc-6089a3ee24cede17e9443aef0aa72fa1a0ba1548.tar.xz glibc-6089a3ee24cede17e9443aef0aa72fa1a0ba1548.zip | |
i386: Replace assembly versions of e_log2f with generic e_log2f.c
This patch replaces i386 assembly versions of e_log2f with generic
e_log2f.c. For workload-spec2017.wrf, on Nehalem, it improves
performance by:
Before After Improvement
reciprocal-throughput 92.3845 30.8752 199%
latency 112.855 54.8645 105%
On Skylake, it improves performance by:
Before After Improvement
reciprocal-throughput 98.7488 22.7507 334%
latency 118.01 51.6083 128%
On IvyBridge with --disable-multi-arch, it improves performance by:
Before After Improvement
reciprocal-throughput 106.635 28.8596 269%
latency 129.888 56.9187 128%
* sysdeps/i386/fpu/e_log2f.S: Removed.
* sysdeps/i386/fpu/e_log2f_data.c: Likewise.
* sysdeps/i386/fpu/w_log2f.c: Likewise.
* sysdeps/i386/fpu/libm-test-ulps: Updated for generic e_log2f.c.
* sysdeps/i386/i686/fpu/multiarch/libm-test-ulps: Likewise.
* sysdeps/i386/i686/fpu/multiarch/Makefile (libm-sysdep_routines):
Add e_log2f-sse2.
(CFLAGS-e_log2f-sse2.c): New.
* sysdeps/i386/i686/fpu/multiarch/e_log2f-sse2.c: New file.
* sysdeps/i386/i686/fpu/multiarch/e_log2f.c: Likewise.
Diffstat (limited to 'sysdeps/i386/fpu/e_log2f.S')
| -rw-r--r-- | sysdeps/i386/fpu/e_log2f.S | 69 |
1 files changed, 0 insertions, 69 deletions
diff --git a/sysdeps/i386/fpu/e_log2f.S b/sysdeps/i386/fpu/e_log2f.S deleted file mode 100644 index 344eeb495e..0000000000 --- a/sysdeps/i386/fpu/e_log2f.S +++ /dev/null @@ -1,69 +0,0 @@ -/* - * Written by J.T. Conklin <jtc@netbsd.org>. - * Adapted for use as log2 by Ulrich Drepper <drepper@cygnus.com>. - * Public domain. - * - * Changed to use fyl2xp1 for values near 1, <drepper@cygnus.com>. - */ - -#include <machine/asm.h> - - .section .rodata.cst8,"aM",@progbits,8 - - .p2align 3 - .type one,@object -one: .double 1.0 - ASM_SIZE_DIRECTIVE(one) - /* It is not important that this constant is precise. It is only - a value which is known to be on the safe side for using the - fyl2xp1 instruction. */ - .type limit,@object -limit: .double 0.29 - ASM_SIZE_DIRECTIVE(limit) - - -#ifdef PIC -# define MO(op) op##@GOTOFF(%edx) -#else -# define MO(op) op -#endif - - .text -ENTRY(__ieee754_log2f) -#ifdef PIC - LOAD_PIC_REG (dx) -#endif - fldl MO(one) - flds 4(%esp) // x : 1 - fxam - fnstsw - fld %st // x : x : 1 - sahf - jc 3f // in case x is NaN or ħInf -4: fsub %st(2), %st // x-1 : x : 1 - fld %st // x-1 : x-1 : x : 1 - fabs // |x-1| : x-1 : x : 1 - fcompl MO(limit) // x-1 : x : 1 - fnstsw // x-1 : x : 1 - andb $0x45, %ah - jz 2f - fxam - fnstsw - andb $0x45, %ah - cmpb $0x40, %ah - jne 5f - fabs // log2(1) is +0 in all rounding modes. -5: fstp %st(1) // x-1 : 1 - fyl2xp1 // log(x) - ret - -2: fstp %st(0) // x : 1 - fyl2x // log(x) - ret - -3: jp 4b // in case x is ħInf - fstp %st(1) - fstp %st(1) - ret -END (__ieee754_log2f) -strong_alias (__ieee754_log2f, __log2f_finite) |