aboutsummaryrefslogtreecommitdiff
path: root/sysdeps/i386/fpu
diff options
context:
space:
mode:
authorAdhemerval Zanella <adhemerval.zanella@linaro.org>2024-11-14 11:07:11 -0300
committerAdhemerval Zanella <adhemerval.zanella@linaro.org>2024-12-18 17:24:43 -0300
commited608a40e2903351c64a3cea51ce8bd7c62996ef (patch)
treee4e517e967698aadb2ff89fef5cde47bbf70761c /sysdeps/i386/fpu
parent5fb4b566efd7785ce6fc75d16e369b6918549620 (diff)
downloadglibc-ed608a40e2903351c64a3cea51ce8bd7c62996ef.zip
glibc-ed608a40e2903351c64a3cea51ce8bd7c62996ef.tar.gz
glibc-ed608a40e2903351c64a3cea51ce8bd7c62996ef.tar.bz2
math: Use asinhf from CORE-MATH
The CORE-MATH implementation is correctly rounded (for any rounding mode) and shows slight better performance to the generic asinhf. The code was adapted to glibc style and to use the definition of math_config.h (to handle errno, overflow, and underflow). Benchtest on x64_64 (Ryzen 9 5900X, gcc 14.2.1), aarch64 (Neoverse-N1, gcc 13.3.1), and powerpc (POWER10, gcc 13.2.1): Latency master patched improvement x86_64 64.5128 56.9717 11.69% x86_64v2 63.3065 57.2666 9.54% x86_64v3 62.8719 51.4170 18.22% i686 189.1630 137.635 27.24% aarch64 (Neoverse) 25.3551 20.5757 18.85% power10 17.9712 13.3302 25.82% reciprocal-throughput master patched improvement x86_64 20.0844 15.4731 22.96% x86_64v2 19.2919 15.4000 20.17% x86_64v3 18.7226 11.9009 36.44% i686 103.7670 80.2681 22.65% aarch64 (Neoverse) 12.5005 8.68969 30.49% power10 7.2220 5.03617 30.27% Signed-off-by: Alexei Sibidanov <sibid@uvic.ca> Signed-off-by: Paul Zimmermann <Paul.Zimmermann@inria.fr> Signed-off-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>: Reviewed-by: DJ Delorie <dj@redhat.com>
Diffstat (limited to 'sysdeps/i386/fpu')
-rw-r--r--sysdeps/i386/fpu/libm-test-ulps3
-rw-r--r--sysdeps/i386/fpu/s_asinhf.S139
2 files changed, 0 insertions, 142 deletions
diff --git a/sysdeps/i386/fpu/libm-test-ulps b/sysdeps/i386/fpu/libm-test-ulps
index 11606b9..669ecbe 100644
--- a/sysdeps/i386/fpu/libm-test-ulps
+++ b/sysdeps/i386/fpu/libm-test-ulps
@@ -90,19 +90,16 @@ ldouble: 3
Function: "asinh_downward":
double: 1
-float: 1
float128: 4
ldouble: 5
Function: "asinh_towardzero":
double: 1
-float: 1
float128: 2
ldouble: 4
Function: "asinh_upward":
double: 1
-float: 1
float128: 4
ldouble: 5
diff --git a/sysdeps/i386/fpu/s_asinhf.S b/sysdeps/i386/fpu/s_asinhf.S
deleted file mode 100644
index fe52e20..0000000
--- a/sysdeps/i386/fpu/s_asinhf.S
+++ /dev/null
@@ -1,139 +0,0 @@
-/* ix87 specific implementation of arcsinh.
- Copyright (C) 1996-2024 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
- <https://www.gnu.org/licenses/>. */
-
-#include <machine/asm.h>
-#include <libm-alias-float.h>
-
- .section .rodata
-
- .align ALIGNARG(4)
- .type huge,@object
-huge: .double 1e+36
- ASM_SIZE_DIRECTIVE(huge)
- .type one,@object
-one: .double 1.0
- ASM_SIZE_DIRECTIVE(one)
- .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(__asinhf)
- movl 4(%esp), %ecx
- movl $0x7fffffff, %eax
- andl %ecx, %eax
- andl $0x80000000, %ecx
- movl %eax, %edx
- orl $0x807fffff, %edx
- incl %edx
- jz 7f // x in ħInf or NaN
- xorl %ecx, 4(%esp)
- flds 4(%esp) // |x|
- cmpl $0x38000000, %eax
- jb 2f // |x| < 2^-14
- fldln2 // log(2) : |x|
- cmpl $0x47000000, %eax
- fxch // |x| : log(2)
- ja 3f // |x| > 2^14
-#ifdef PIC
- LOAD_PIC_REG (dx)
-#endif
- cmpl $0x40000000, %eax
- ja 5f // |x| > 2
-
- // 2^-14 <= |x| <= 2 => y = sign(x)*log1p(|x|+|x|^2/(1+sqrt(1+|x|^2)))
- fld %st // |x| : |x| : log(2)
- fmul %st(1) // |x|^2 : |x| : log(2)
- fld %st // |x|^2 : |x|^2 : |x| : log(2)
- faddl MO(one) // 1+|x|^2 : |x|^2 : |x| : log(2)
- fsqrt // sqrt(1+|x|^2) : |x|^2 : |x| : log(2)
- faddl MO(one) // 1+sqrt(1+|x|^2) : |x|^2 : |x| : log(2)
- fdivrp // |x|^2/(1+sqrt(1+|x|^2)) : |x| : log(2)
- faddp // |x|+|x|^2/(1+sqrt(1+|x|^2)) : log(2)
- fcoml MO(limit)
- fnstsw
- sahf
- ja 6f
- fyl2xp1
- jecxz 4f
- fchs
-4: ret
-
-7: flds 4(%esp)
- ret
-
-6: faddl MO(one)
- fyl2x
- jecxz 4f
- fchs
-4: ret
-
- // |x| < 2^-14 => y = x (inexact iff |x| != 0.0)
- .align ALIGNARG(4)
-2:
-#ifdef PIC
- LOAD_PIC_REG (dx)
-#endif
- jecxz 4f
- fchs // x
-4: fld %st // x : x
- faddl MO(huge) // huge+x : x
- fstp %st(0) // x
- cmpl $0x00800000, %eax
- jae 8f
- subl $4, %esp
- cfi_adjust_cfa_offset (4)
- fld %st(0)
- fmul %st(0)
- fstps (%esp)
- addl $4, %esp
- cfi_adjust_cfa_offset (-4)
-8: ret
-
- // |x| > 2^14 => y = sign(x) * (log(|x|) + log(2))
- .align ALIGNARG(4)
-3: fyl2x // log(|x|)
- fldln2 // log(2) : log(|x|)
- faddp // log(|x|)+log(2)
- jecxz 4f
- fchs
-4: ret
-
- // |x| > 2 => y = sign(x) * log(2*|x| + 1/(|x|+sqrt(x*x+1)))
- .align ALIGNARG(4)
-5: fld %st // |x| : |x| : log(2)
- fadd %st, %st(1) // |x| : 2*|x| : log(2)
- fld %st // |x| : |x| : 2*|x| : log(2)
- fmul %st(1) // |x|^2 : |x| : 2*|x| : log(2)
- faddl MO(one) // 1+|x|^2 : |x| : 2*|x| : log(2)
- fsqrt // sqrt(1+|x|^2) : |x| : 2*|x| : log(2)
- faddp // |x|+sqrt(1+|x|^2) : 2*|x| : log(2)
- fdivrl MO(one) // 1/(|x|+sqrt(1+|x|^2)) : 2*|x| : log(2)
- faddp // 2*|x|+1/(|x|+sqrt(1+|x|^2)) : log(2)
- fyl2x // log(2*|x|+1/(|x|+sqrt(1+|x|^2)))
- jecxz 4f
- fchs
-4: ret
-END(__asinhf)
-libm_alias_float (__asinh, asinh)