mirror of
https://sourceware.org/git/glibc.git
synced 2024-11-22 21:10:07 +00:00
Optimize sparc {ceil,floor}{,f} using vis2 'siam' instruction.
* sysdeps/sparc/sparc-ifunc.h (SPARC_ASM_IFUNC2): New macro. (SPARC_ASM_VIS2_IFUNC): Likewise. (SPARC_ASM_VIS3_VIS2_IFUNC): Likewise. * sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_ceil-vis3.S: Make use of 'siam' instruction. * sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_ceilf-vis3.S: Likewise. * sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_floor-vis3.S: Likewise. * sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_floorf-vis3.S: Likewise. * sysdeps/sparc/sparc64/fpu/multiarch/s_ceil-vis3.S: Likewise. * sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf-vis3.S: Likewise. * sysdeps/sparc/sparc64/fpu/multiarch/s_floor-vis3.S: Likewise. * sysdeps/sparc/sparc64/fpu/multiarch/s_floorf-vis3.S: Likewise. * sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_ceil-vis2.S: New file. * sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_ceilf-vis2.S: New file. * sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_floor-vis2.S: New file. * sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_floorf-vis2.S: New file. * sysdeps/sparc/sparc64/fpu/multiarch/s_ceil-vis2.S: New file. * sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf-vis2.S: New file. * sysdeps/sparc/sparc64/fpu/multiarch/s_floor-vis2.S: New file. * sysdeps/sparc/sparc64/fpu/multiarch/s_floorf-vis2.S: New file. * sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_ceil.S: Hook in new VIS2 routines. * sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_ceilf.S: Likewise. * sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_floor.S: Likewise. * sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_floorf.S: Likewise. * sysdeps/sparc/sparc64/fpu/multiarch/s_ceil.S: Likewise. * sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf.S: Likewise. * sysdeps/sparc/sparc64/fpu/multiarch/s_floor.S: Likewise. * sysdeps/sparc/sparc64/fpu/multiarch/s_floorf.S: Likewise. * sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/Makefile: Add new VIS2 routines to libm-sysdep_routines. * sysdeps/sparc/sparc64/fpu/multiarch/Makefile: Likewise.
This commit is contained in:
David S. Miller
parent
65a82e3dd5
commit
8b954ab9b8
41
ChangeLog
41
ChangeLog
@ -1,5 +1,46 @@
|
|||||||
2013-01-14 David S. Miller <davem@davemloft.net>
|
2013-01-14 David S. Miller <davem@davemloft.net>
|
||||||
|
|
||||||
|
* sysdeps/sparc/sparc-ifunc.h (SPARC_ASM_IFUNC2): New macro.
|
||||||
|
(SPARC_ASM_VIS2_IFUNC): Likewise.
|
||||||
|
(SPARC_ASM_VIS3_VIS2_IFUNC): Likewise.
|
||||||
|
* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_ceil-vis3.S: Make
|
||||||
|
use of 'siam' instruction.
|
||||||
|
* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_ceilf-vis3.S:
|
||||||
|
Likewise.
|
||||||
|
* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_floor-vis3.S:
|
||||||
|
Likewise.
|
||||||
|
* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_floorf-vis3.S:
|
||||||
|
Likewise.
|
||||||
|
* sysdeps/sparc/sparc64/fpu/multiarch/s_ceil-vis3.S: Likewise.
|
||||||
|
* sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf-vis3.S: Likewise.
|
||||||
|
* sysdeps/sparc/sparc64/fpu/multiarch/s_floor-vis3.S: Likewise.
|
||||||
|
* sysdeps/sparc/sparc64/fpu/multiarch/s_floorf-vis3.S: Likewise.
|
||||||
|
* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_ceil-vis2.S: New
|
||||||
|
file.
|
||||||
|
* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_ceilf-vis2.S: New
|
||||||
|
file.
|
||||||
|
* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_floor-vis2.S: New
|
||||||
|
file.
|
||||||
|
* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_floorf-vis2.S: New
|
||||||
|
file.
|
||||||
|
* sysdeps/sparc/sparc64/fpu/multiarch/s_ceil-vis2.S: New file.
|
||||||
|
* sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf-vis2.S: New file.
|
||||||
|
* sysdeps/sparc/sparc64/fpu/multiarch/s_floor-vis2.S: New file.
|
||||||
|
* sysdeps/sparc/sparc64/fpu/multiarch/s_floorf-vis2.S: New file.
|
||||||
|
* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_ceil.S: Hook in
|
||||||
|
new VIS2 routines.
|
||||||
|
* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_ceilf.S: Likewise.
|
||||||
|
* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_floor.S: Likewise.
|
||||||
|
* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_floorf.S:
|
||||||
|
Likewise.
|
||||||
|
* sysdeps/sparc/sparc64/fpu/multiarch/s_ceil.S: Likewise.
|
||||||
|
* sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf.S: Likewise.
|
||||||
|
* sysdeps/sparc/sparc64/fpu/multiarch/s_floor.S: Likewise.
|
||||||
|
* sysdeps/sparc/sparc64/fpu/multiarch/s_floorf.S: Likewise.
|
||||||
|
* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/Makefile: Add new VIS2
|
||||||
|
routines to libm-sysdep_routines.
|
||||||
|
* sysdeps/sparc/sparc64/fpu/multiarch/Makefile: Likewise.
|
||||||
|
|
||||||
* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/Makefile: Add vis3
|
* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/Makefile: Add vis3
|
||||||
fdim/fdimf to libm-sysdep_routines.
|
fdim/fdimf to libm-sysdep_routines.
|
||||||
* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_fdim-vis3.S: New
|
* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_fdim-vis3.S: New
|
||||||
|
|||||||
@ -51,6 +51,33 @@ ENTRY (__##name) \
|
|||||||
mov %o1, %o0; \
|
mov %o1, %o0; \
|
||||||
END (__##name)
|
END (__##name)
|
||||||
|
|
||||||
|
# define SPARC_ASM_IFUNC2(name, m1, f1, m2, f2, dflt) \
|
||||||
|
ENTRY (__##name) \
|
||||||
|
.type __##name, @gnu_indirect_function; \
|
||||||
|
SETUP_PIC_REG_LEAF(o3, o5); \
|
||||||
|
set m1, %o1; \
|
||||||
|
andcc %o0, %o1, %g0; \
|
||||||
|
be 8f; \
|
||||||
|
nop; \
|
||||||
|
sethi %gdop_hix22(f1), %o1; \
|
||||||
|
xor %o1, %gdop_lox10(f1), %o1; \
|
||||||
|
ba 10f; \
|
||||||
|
nop; \
|
||||||
|
8: set m2, %o1; \
|
||||||
|
andcc %o0, %o1, %g0; \
|
||||||
|
be 9f; \
|
||||||
|
nop; \
|
||||||
|
sethi %gdop_hix22(f2), %o1; \
|
||||||
|
xor %o1, %gdop_lox10(f2), %o1; \
|
||||||
|
ba 10f; \
|
||||||
|
nop; \
|
||||||
|
9: sethi %gdop_hix22(dflt), %o1; \
|
||||||
|
xor %o1, %gdop_lox10(dflt), %o1; \
|
||||||
|
10: add %o3, %o1, %o1; \
|
||||||
|
retl; \
|
||||||
|
mov %o1, %o0; \
|
||||||
|
END (__##name)
|
||||||
|
|
||||||
# else /* SHARED */
|
# else /* SHARED */
|
||||||
|
|
||||||
# ifdef __arch64__
|
# ifdef __arch64__
|
||||||
@ -82,19 +109,54 @@ ENTRY (__##name) \
|
|||||||
mov %o1, %o0; \
|
mov %o1, %o0; \
|
||||||
END (__##name)
|
END (__##name)
|
||||||
|
|
||||||
|
# define SPARC_ASM_IFUNC2(name, m1, f1, m2, f2, dflt) \
|
||||||
|
ENTRY (__##name) \
|
||||||
|
.type __##name, @gnu_indirect_function; \
|
||||||
|
set m1, %o1; \
|
||||||
|
andcc %o0, %o1, %g0; \
|
||||||
|
be 8f; \
|
||||||
|
nop; \
|
||||||
|
SET(f1, %g1, %o1); \
|
||||||
|
ba 10f; \
|
||||||
|
nop; \
|
||||||
|
8: set m2, %o1; \
|
||||||
|
andcc %o0, %o1, %g0; \
|
||||||
|
be 9f; \
|
||||||
|
nop; \
|
||||||
|
SET(f2, %g1, %o1); \
|
||||||
|
ba 10f; \
|
||||||
|
nop; \
|
||||||
|
9: SET(dflt, %g1, %o1); \
|
||||||
|
10: retl; \
|
||||||
|
mov %o1, %o0; \
|
||||||
|
END (__##name)
|
||||||
|
|
||||||
# endif /* SHARED */
|
# endif /* SHARED */
|
||||||
|
|
||||||
|
#define SPARC_ASM_VIS2_IFUNC(name) \
|
||||||
|
SPARC_ASM_IFUNC1(name, HWCAP_SPARC_VIS2, \
|
||||||
|
__##name##_vis2, __##name##_generic)
|
||||||
|
|
||||||
# ifdef HAVE_AS_VIS3_SUPPORT
|
# ifdef HAVE_AS_VIS3_SUPPORT
|
||||||
|
|
||||||
#define SPARC_ASM_VIS3_IFUNC(name) \
|
#define SPARC_ASM_VIS3_IFUNC(name) \
|
||||||
SPARC_ASM_IFUNC1(name, HWCAP_SPARC_VIS3, \
|
SPARC_ASM_IFUNC1(name, HWCAP_SPARC_VIS3, \
|
||||||
__##name##_vis3, __##name##_generic)
|
__##name##_vis3, __##name##_generic)
|
||||||
|
|
||||||
|
#define SPARC_ASM_VIS3_VIS2_IFUNC(name) \
|
||||||
|
SPARC_ASM_IFUNC2(name, HWCAP_SPARC_VIS3, \
|
||||||
|
__##name##_vis3, \
|
||||||
|
HWCAP_SPARC_VIS2, \
|
||||||
|
__##name##_vis2, __##name##_generic)
|
||||||
|
|
||||||
# else /* HAVE_AS_VIS3_SUPPORT */
|
# else /* HAVE_AS_VIS3_SUPPORT */
|
||||||
|
|
||||||
#define SPARC_ASM_VIS3_IFUNC(name) \
|
#define SPARC_ASM_VIS3_IFUNC(name) \
|
||||||
SPARC_ASM_IFUNC_DFLT(name, __##name##_generic)
|
SPARC_ASM_IFUNC_DFLT(name, __##name##_generic)
|
||||||
|
|
||||||
|
#define SPARC_ASM_VIS3_VIS2_IFUNC(name) \
|
||||||
|
SPARC_ASM_VIS2_IFUNC(name)
|
||||||
|
|
||||||
# endif /* HAVE_AS_VIS3_SUPPORT */
|
# endif /* HAVE_AS_VIS3_SUPPORT */
|
||||||
|
|
||||||
|
|
||||||
|
|||||||
@ -1,4 +1,6 @@
|
|||||||
ifeq ($(subdir),math)
|
ifeq ($(subdir),math)
|
||||||
|
libm-sysdep_routines += s_ceil-vis2 s_ceilf-vis2 \
|
||||||
|
s_floor-vis2 s_floorf-vis2
|
||||||
ifeq ($(have-as-vis3),yes)
|
ifeq ($(have-as-vis3),yes)
|
||||||
libm-sysdep_routines += m_copysignf-vis3 m_copysign-vis3 s_ceilf-vis3 \
|
libm-sysdep_routines += m_copysignf-vis3 m_copysign-vis3 s_ceilf-vis3 \
|
||||||
s_ceil-vis3 s_fabs-vis3 s_fabsf-vis3 s_floor-vis3 \
|
s_ceil-vis3 s_fabs-vis3 s_fabsf-vis3 s_floor-vis3 \
|
||||||
|
|||||||
61
sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_ceil-vis2.S
Normal file
61
sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_ceil-vis2.S
Normal file
@ -0,0 +1,61 @@
|
|||||||
|
/* ceil function, sparc32 v9 vis2 version.
|
||||||
|
Copyright (C) 2013 Free Software Foundation, Inc.
|
||||||
|
This file is part of the GNU C Library.
|
||||||
|
Contributed by David S. Miller <davem@davemloft.net>, 2013.
|
||||||
|
|
||||||
|
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>
|
||||||
|
|
||||||
|
/* 'siam' (Set Interval Arithmetic Mode) is used to quickly override
|
||||||
|
the rounding mode during this routine.
|
||||||
|
|
||||||
|
We add then subtract (or subtract than add if the initial
|
||||||
|
value was negative) 2**23 to the value, then subtract it
|
||||||
|
back out.
|
||||||
|
|
||||||
|
This will clear out the fractional portion of the value and,
|
||||||
|
with suitable 'siam' initiated rouding mode settings, round
|
||||||
|
the final result in the proper direction. */
|
||||||
|
|
||||||
|
#define TWO_FIFTYTWO 0x43300000 /* 2**52 */
|
||||||
|
|
||||||
|
#define ZERO %f10 /* 0.0 */
|
||||||
|
#define SIGN_BIT %f12 /* -0.0 */
|
||||||
|
|
||||||
|
ENTRY (__ceil_vis2)
|
||||||
|
sethi %hi(TWO_FIFTYTWO), %o2
|
||||||
|
sllx %o0, 32, %o0
|
||||||
|
or %o0, %o1, %o0
|
||||||
|
stx %o0, [%sp + 72]
|
||||||
|
sllx %o2, 32, %o2
|
||||||
|
fzero ZERO
|
||||||
|
ldd [%sp + 72], %f0
|
||||||
|
fnegd ZERO, SIGN_BIT
|
||||||
|
stx %o2, [%sp + 72]
|
||||||
|
fabsd %f0, %f14
|
||||||
|
ldd [%sp + 72], %f16
|
||||||
|
fcmpd %fcc3, %f14, %f16
|
||||||
|
fmovduge %fcc3, ZERO, %f16
|
||||||
|
fand %f0, SIGN_BIT, SIGN_BIT
|
||||||
|
for %f16, SIGN_BIT, %f16
|
||||||
|
siam (1 << 2) | 2
|
||||||
|
faddd %f0, %f16, %f18
|
||||||
|
siam (1 << 2) | 0
|
||||||
|
fsubd %f18, %f16, %f18
|
||||||
|
siam (0 << 2)
|
||||||
|
retl
|
||||||
|
for %f18, SIGN_BIT, %f0
|
||||||
|
END (__ceil_vis2)
|
||||||
@ -19,27 +19,21 @@
|
|||||||
|
|
||||||
#include <sysdep.h>
|
#include <sysdep.h>
|
||||||
|
|
||||||
/* Since changing the rounding mode is extremely expensive, we
|
/* 'siam' (Set Interval Arithmetic Mode) is used to quickly override
|
||||||
try to round up using a method that is rounding mode
|
the rounding mode during this routine.
|
||||||
agnostic.
|
|
||||||
|
|
||||||
We add then subtract (or subtract than add if the initial
|
We add then subtract (or subtract than add if the initial
|
||||||
value was negative) 2**23 to the value, then subtract it
|
value was negative) 2**23 to the value, then subtract it
|
||||||
back out.
|
back out.
|
||||||
|
|
||||||
This will clear out the fractional portion of the value.
|
This will clear out the fractional portion of the value and,
|
||||||
One of two things will happen for non-whole initial values.
|
with suitable 'siam' initiated rouding mode settings, round
|
||||||
Either the rounding mode will round it up, or it will be
|
the final result in the proper direction.
|
||||||
rounded down. If the value started out whole, it will be
|
|
||||||
equal after the addition and subtraction. This means we
|
|
||||||
can accurately detect with one test whether we need to add
|
|
||||||
another 1.0 to round it up properly.
|
|
||||||
|
|
||||||
VIS instructions are used to facilitate the formation of
|
We also use VIS3 moves to avoid using the stack to transfer
|
||||||
easier constants, and the propagation of the sign bit. */
|
values between float and integer registers. */
|
||||||
|
|
||||||
#define TWO_FIFTYTWO 0x43300000 /* 2**52 */
|
#define TWO_FIFTYTWO 0x43300000 /* 2**52 */
|
||||||
#define ONE_DOT_ZERO 0x3ff00000 /* 1.0 */
|
|
||||||
|
|
||||||
#define ZERO %f10 /* 0.0 */
|
#define ZERO %f10 /* 0.0 */
|
||||||
#define SIGN_BIT %f12 /* -0.0 */
|
#define SIGN_BIT %f12 /* -0.0 */
|
||||||
@ -47,32 +41,22 @@
|
|||||||
ENTRY (__ceil_vis3)
|
ENTRY (__ceil_vis3)
|
||||||
sethi %hi(TWO_FIFTYTWO), %o2
|
sethi %hi(TWO_FIFTYTWO), %o2
|
||||||
sllx %o0, 32, %o0
|
sllx %o0, 32, %o0
|
||||||
sethi %hi(ONE_DOT_ZERO), %o3
|
sllx %o2, 32, %o2
|
||||||
or %o0, %o1, %o0
|
or %o0, %o1, %o0
|
||||||
movxtod %o0, %f0
|
movxtod %o0, %f0
|
||||||
sllx %o2, 32, %o2
|
|
||||||
fzero ZERO
|
fzero ZERO
|
||||||
sllx %o3, 32, %o3
|
|
||||||
|
|
||||||
fnegd ZERO, SIGN_BIT
|
fnegd ZERO, SIGN_BIT
|
||||||
|
|
||||||
movxtod %o2, %f16
|
movxtod %o2, %f16
|
||||||
fabsd %f0, %f14
|
fabsd %f0, %f14
|
||||||
|
|
||||||
fcmpd %fcc3, %f14, %f16
|
fcmpd %fcc3, %f14, %f16
|
||||||
|
|
||||||
fmovduge %fcc3, ZERO, %f16
|
fmovduge %fcc3, ZERO, %f16
|
||||||
fand %f0, SIGN_BIT, SIGN_BIT
|
fand %f0, SIGN_BIT, SIGN_BIT
|
||||||
|
|
||||||
for %f16, SIGN_BIT, %f16
|
for %f16, SIGN_BIT, %f16
|
||||||
|
siam (1 << 2) | 2
|
||||||
faddd %f0, %f16, %f18
|
faddd %f0, %f16, %f18
|
||||||
|
siam (1 << 2) | 0
|
||||||
fsubd %f18, %f16, %f18
|
fsubd %f18, %f16, %f18
|
||||||
fcmpd %fcc2, %f18, %f0
|
siam (0 << 2)
|
||||||
movxtod %o3, %f20
|
|
||||||
|
|
||||||
fmovduge %fcc2, ZERO, %f20
|
|
||||||
faddd %f18, %f20, %f0
|
|
||||||
fabsd %f0, %f0
|
|
||||||
retl
|
retl
|
||||||
for %f0, SIGN_BIT, %f0
|
for %f18, SIGN_BIT, %f0
|
||||||
END (__ceil_vis3)
|
END (__ceil_vis3)
|
||||||
|
|||||||
@ -1,7 +1,7 @@
|
|||||||
#include <sparc-ifunc.h>
|
#include <sparc-ifunc.h>
|
||||||
#include <math_ldbl_opt.h>
|
#include <math_ldbl_opt.h>
|
||||||
|
|
||||||
SPARC_ASM_VIS3_IFUNC(ceil)
|
SPARC_ASM_VIS3_VIS2_IFUNC(ceil)
|
||||||
|
|
||||||
weak_alias (__ceil, ceil)
|
weak_alias (__ceil, ceil)
|
||||||
|
|
||||||
|
|||||||
58
sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_ceilf-vis2.S
Normal file
58
sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_ceilf-vis2.S
Normal file
@ -0,0 +1,58 @@
|
|||||||
|
/* Float ceil function, sparc32 v9 vis2 version.
|
||||||
|
Copyright (C) 2013 Free Software Foundation, Inc.
|
||||||
|
This file is part of the GNU C Library.
|
||||||
|
Contributed by David S. Miller <davem@davemloft.net>, 2013.
|
||||||
|
|
||||||
|
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>
|
||||||
|
|
||||||
|
/* 'siam' (Set Interval Arithmetic Mode) is used to quickly override
|
||||||
|
the rounding mode during this routine.
|
||||||
|
|
||||||
|
We add then subtract (or subtract than add if the initial
|
||||||
|
value was negative) 2**23 to the value, then subtract it
|
||||||
|
back out.
|
||||||
|
|
||||||
|
This will clear out the fractional portion of the value and,
|
||||||
|
with suitable 'siam' initiated rouding mode settings, round
|
||||||
|
the final result in the proper direction. */
|
||||||
|
|
||||||
|
#define TWO_TWENTYTHREE 0x4b000000 /* 2**23 */
|
||||||
|
|
||||||
|
#define ZERO %f10 /* 0.0 */
|
||||||
|
#define SIGN_BIT %f12 /* -0.0 */
|
||||||
|
|
||||||
|
ENTRY (__ceilf_vis2)
|
||||||
|
st %o0, [%sp + 68]
|
||||||
|
sethi %hi(TWO_TWENTYTHREE), %o2
|
||||||
|
fzeros ZERO
|
||||||
|
ld [%sp + 68], %f0
|
||||||
|
fnegs ZERO, SIGN_BIT
|
||||||
|
st %o2, [%sp + 68]
|
||||||
|
fabss %f0, %f14
|
||||||
|
ld [%sp + 68], %f16
|
||||||
|
fcmps %fcc3, %f14, %f16
|
||||||
|
fmovsuge %fcc3, ZERO, %f16
|
||||||
|
fands %f0, SIGN_BIT, SIGN_BIT
|
||||||
|
fors %f16, SIGN_BIT, %f16
|
||||||
|
siam (1 << 2) | 2
|
||||||
|
fadds %f0, %f16, %f1
|
||||||
|
siam (1 << 2) | 0
|
||||||
|
fsubs %f1, %f16, %f1
|
||||||
|
siam (0 << 2)
|
||||||
|
retl
|
||||||
|
fors %f1, SIGN_BIT, %f0
|
||||||
|
END (__ceilf_vis2)
|
||||||
@ -19,27 +19,21 @@
|
|||||||
|
|
||||||
#include <sysdep.h>
|
#include <sysdep.h>
|
||||||
|
|
||||||
/* Since changing the rounding mode is extremely expensive, we
|
/* 'siam' (Set Interval Arithmetic Mode) is used to quickly override
|
||||||
try to round up using a method that is rounding mode
|
the rounding mode during this routine.
|
||||||
agnostic.
|
|
||||||
|
|
||||||
We add then subtract (or subtract than add if the initial
|
We add then subtract (or subtract than add if the initial
|
||||||
value was negative) 2**23 to the value, then subtract it
|
value was negative) 2**23 to the value, then subtract it
|
||||||
back out.
|
back out.
|
||||||
|
|
||||||
This will clear out the fractional portion of the value.
|
This will clear out the fractional portion of the value and,
|
||||||
One of two things will happen for non-whole initial values.
|
with suitable 'siam' initiated rouding mode settings, round
|
||||||
Either the rounding mode will round it up, or it will be
|
the final result in the proper direction.
|
||||||
rounded down. If the value started out whole, it will be
|
|
||||||
equal after the addition and subtraction. This means we
|
|
||||||
can accurately detect with one test whether we need to add
|
|
||||||
another 1.0 to round it up properly.
|
|
||||||
|
|
||||||
VIS instructions are used to facilitate the formation of
|
We also use VIS3 moves to avoid using the stack to transfer
|
||||||
easier constants, and the propagation of the sign bit. */
|
values between float and integer registers. */
|
||||||
|
|
||||||
#define TWO_TWENTYTHREE 0x4b000000 /* 2**23 */
|
#define TWO_TWENTYTHREE 0x4b000000 /* 2**23 */
|
||||||
#define ONE_DOT_ZERO 0x3f800000 /* 1.0 */
|
|
||||||
|
|
||||||
#define ZERO %f10 /* 0.0 */
|
#define ZERO %f10 /* 0.0 */
|
||||||
#define SIGN_BIT %f12 /* -0.0 */
|
#define SIGN_BIT %f12 /* -0.0 */
|
||||||
@ -47,28 +41,19 @@
|
|||||||
ENTRY (__ceilf_vis3)
|
ENTRY (__ceilf_vis3)
|
||||||
movwtos %o0, %f0
|
movwtos %o0, %f0
|
||||||
sethi %hi(TWO_TWENTYTHREE), %o2
|
sethi %hi(TWO_TWENTYTHREE), %o2
|
||||||
sethi %hi(ONE_DOT_ZERO), %o3
|
|
||||||
fzeros ZERO
|
fzeros ZERO
|
||||||
|
|
||||||
fnegs ZERO, SIGN_BIT
|
fnegs ZERO, SIGN_BIT
|
||||||
|
|
||||||
movwtos %o2, %f16
|
movwtos %o2, %f16
|
||||||
fabss %f0, %f14
|
fabss %f0, %f14
|
||||||
|
|
||||||
fcmps %fcc3, %f14, %f16
|
fcmps %fcc3, %f14, %f16
|
||||||
|
|
||||||
fmovsuge %fcc3, ZERO, %f16
|
fmovsuge %fcc3, ZERO, %f16
|
||||||
fands %f0, SIGN_BIT, SIGN_BIT
|
fands %f0, SIGN_BIT, SIGN_BIT
|
||||||
|
|
||||||
fors %f16, SIGN_BIT, %f16
|
fors %f16, SIGN_BIT, %f16
|
||||||
|
siam (1 << 2) | 2
|
||||||
fadds %f0, %f16, %f1
|
fadds %f0, %f16, %f1
|
||||||
|
siam (1 << 2) | 0
|
||||||
fsubs %f1, %f16, %f1
|
fsubs %f1, %f16, %f1
|
||||||
fcmps %fcc2, %f1, %f0
|
siam (0 << 2)
|
||||||
movwtos %o3, %f9
|
|
||||||
|
|
||||||
fmovsuge %fcc2, ZERO, %f9
|
|
||||||
fadds %f1, %f9, %f0
|
|
||||||
fabss %f0, %f0
|
|
||||||
retl
|
retl
|
||||||
fors %f0, SIGN_BIT, %f0
|
fors %f1, SIGN_BIT, %f0
|
||||||
END (__ceilf_vis3)
|
END (__ceilf_vis3)
|
||||||
|
|||||||
@ -1,6 +1,6 @@
|
|||||||
#include <sparc-ifunc.h>
|
#include <sparc-ifunc.h>
|
||||||
|
|
||||||
SPARC_ASM_VIS3_IFUNC(ceilf)
|
SPARC_ASM_VIS3_VIS2_IFUNC(ceilf)
|
||||||
|
|
||||||
weak_alias (__ceilf, ceilf)
|
weak_alias (__ceilf, ceilf)
|
||||||
|
|
||||||
|
|||||||
61
sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_floor-vis2.S
Normal file
61
sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_floor-vis2.S
Normal file
@ -0,0 +1,61 @@
|
|||||||
|
/* floor function, sparc32 v9 vis2 version.
|
||||||
|
Copyright (C) 2013 Free Software Foundation, Inc.
|
||||||
|
This file is part of the GNU C Library.
|
||||||
|
Contributed by David S. Miller <davem@davemloft.net>, 2013.
|
||||||
|
|
||||||
|
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>
|
||||||
|
|
||||||
|
/* 'siam' (Set Interval Arithmetic Mode) is used to quickly override
|
||||||
|
the rounding mode during this routine.
|
||||||
|
|
||||||
|
We add then subtract (or subtract than add if the initial
|
||||||
|
value was negative) 2**23 to the value, then subtract it
|
||||||
|
back out.
|
||||||
|
|
||||||
|
This will clear out the fractional portion of the value and,
|
||||||
|
with suitable 'siam' initiated rouding mode settings, round
|
||||||
|
the final result in the proper direction. */
|
||||||
|
|
||||||
|
#define TWO_FIFTYTWO 0x43300000 /* 2**52 */
|
||||||
|
|
||||||
|
#define ZERO %f10 /* 0.0 */
|
||||||
|
#define SIGN_BIT %f12 /* -0.0 */
|
||||||
|
|
||||||
|
ENTRY (__floor_vis2)
|
||||||
|
sethi %hi(TWO_FIFTYTWO), %o2
|
||||||
|
sllx %o0, 32, %o0
|
||||||
|
or %o0, %o1, %o0
|
||||||
|
stx %o0, [%sp + 72]
|
||||||
|
sllx %o2, 32, %o2
|
||||||
|
fzero ZERO
|
||||||
|
ldd [%sp + 72], %f0
|
||||||
|
fnegd ZERO, SIGN_BIT
|
||||||
|
stx %o2, [%sp + 72]
|
||||||
|
fabsd %f0, %f14
|
||||||
|
ldd [%sp + 72], %f16
|
||||||
|
fcmpd %fcc3, %f14, %f16
|
||||||
|
fmovduge %fcc3, ZERO, %f16
|
||||||
|
fand %f0, SIGN_BIT, SIGN_BIT
|
||||||
|
for %f16, SIGN_BIT, %f16
|
||||||
|
siam (1 << 2) | 3
|
||||||
|
faddd %f0, %f16, %f18
|
||||||
|
siam (1 << 2) | 0
|
||||||
|
fsubd %f18, %f16, %f18
|
||||||
|
siam (0 << 2)
|
||||||
|
retl
|
||||||
|
for %f18, SIGN_BIT, %f0
|
||||||
|
END (__floor_vis2)
|
||||||
@ -19,27 +19,21 @@
|
|||||||
|
|
||||||
#include <sysdep.h>
|
#include <sysdep.h>
|
||||||
|
|
||||||
/* Since changing the rounding mode is extremely expensive, we
|
/* 'siam' (Set Interval Arithmetic Mode) is used to quickly override
|
||||||
try to round up using a method that is rounding mode
|
the rounding mode during this routine.
|
||||||
agnostic.
|
|
||||||
|
|
||||||
We add then subtract (or subtract than add if the initial
|
We add then subtract (or subtract than add if the initial
|
||||||
value was negative) 2**23 to the value, then subtract it
|
value was negative) 2**23 to the value, then subtract it
|
||||||
back out.
|
back out.
|
||||||
|
|
||||||
This will clear out the fractional portion of the value.
|
This will clear out the fractional portion of the value and,
|
||||||
One of two things will happen for non-whole initial values.
|
with suitable 'siam' initiated rouding mode settings, round
|
||||||
Either the rounding mode will round it up, or it will be
|
the final result in the proper direction.
|
||||||
rounded down. If the value started out whole, it will be
|
|
||||||
equal after the addition and subtraction. This means we
|
|
||||||
can accurately detect with one test whether we need to add
|
|
||||||
another 1.0 to round it up properly.
|
|
||||||
|
|
||||||
VIS instructions are used to facilitate the formation of
|
We also use VIS3 moves to avoid using the stack to transfer
|
||||||
easier constants, and the propagation of the sign bit. */
|
values between float and integer registers. */
|
||||||
|
|
||||||
#define TWO_FIFTYTWO 0x43300000 /* 2**52 */
|
#define TWO_FIFTYTWO 0x43300000 /* 2**52 */
|
||||||
#define ONE_DOT_ZERO 0x3ff00000 /* 1.0 */
|
|
||||||
|
|
||||||
#define ZERO %f10 /* 0.0 */
|
#define ZERO %f10 /* 0.0 */
|
||||||
#define SIGN_BIT %f12 /* -0.0 */
|
#define SIGN_BIT %f12 /* -0.0 */
|
||||||
@ -47,32 +41,22 @@
|
|||||||
ENTRY (__floor_vis3)
|
ENTRY (__floor_vis3)
|
||||||
sethi %hi(TWO_FIFTYTWO), %o2
|
sethi %hi(TWO_FIFTYTWO), %o2
|
||||||
sllx %o0, 32, %o0
|
sllx %o0, 32, %o0
|
||||||
sethi %hi(ONE_DOT_ZERO), %o3
|
sllx %o2, 32, %o2
|
||||||
or %o0, %o1, %o0
|
or %o0, %o1, %o0
|
||||||
movxtod %o0, %f0
|
movxtod %o0, %f0
|
||||||
sllx %o2, 32, %o2
|
|
||||||
fzero ZERO
|
fzero ZERO
|
||||||
sllx %o3, 32, %o3
|
|
||||||
|
|
||||||
fnegd ZERO, SIGN_BIT
|
fnegd ZERO, SIGN_BIT
|
||||||
|
|
||||||
movxtod %o2, %f16
|
movxtod %o2, %f16
|
||||||
fabsd %f0, %f14
|
fabsd %f0, %f14
|
||||||
|
|
||||||
fcmpd %fcc3, %f14, %f16
|
fcmpd %fcc3, %f14, %f16
|
||||||
|
|
||||||
fmovduge %fcc3, ZERO, %f16
|
fmovduge %fcc3, ZERO, %f16
|
||||||
fand %f0, SIGN_BIT, SIGN_BIT
|
fand %f0, SIGN_BIT, SIGN_BIT
|
||||||
|
|
||||||
for %f16, SIGN_BIT, %f16
|
for %f16, SIGN_BIT, %f16
|
||||||
|
siam (1 << 2) | 3
|
||||||
faddd %f0, %f16, %f18
|
faddd %f0, %f16, %f18
|
||||||
|
siam (1 << 2) | 0
|
||||||
fsubd %f18, %f16, %f18
|
fsubd %f18, %f16, %f18
|
||||||
fcmpd %fcc2, %f18, %f0
|
siam (0 << 2)
|
||||||
movxtod %o3, %f20
|
|
||||||
|
|
||||||
fmovdule %fcc2, ZERO, %f20
|
|
||||||
fsubd %f18, %f20, %f0
|
|
||||||
fabsd %f0, %f0
|
|
||||||
retl
|
retl
|
||||||
for %f0, SIGN_BIT, %f0
|
for %f18, SIGN_BIT, %f0
|
||||||
END (__floor_vis3)
|
END (__floor_vis3)
|
||||||
|
|||||||
@ -1,7 +1,7 @@
|
|||||||
#include <sparc-ifunc.h>
|
#include <sparc-ifunc.h>
|
||||||
#include <math_ldbl_opt.h>
|
#include <math_ldbl_opt.h>
|
||||||
|
|
||||||
SPARC_ASM_VIS3_IFUNC(floor)
|
SPARC_ASM_VIS3_VIS2_IFUNC(floor)
|
||||||
|
|
||||||
weak_alias (__floor, floor)
|
weak_alias (__floor, floor)
|
||||||
|
|
||||||
|
|||||||
58
sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_floorf-vis2.S
Normal file
58
sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_floorf-vis2.S
Normal file
@ -0,0 +1,58 @@
|
|||||||
|
/* Float floor function, sparc32 v9 vis2 version.
|
||||||
|
Copyright (C) 2013 Free Software Foundation, Inc.
|
||||||
|
This file is part of the GNU C Library.
|
||||||
|
Contributed by David S. Miller <davem@davemloft.net>, 2013.
|
||||||
|
|
||||||
|
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>
|
||||||
|
|
||||||
|
/* 'siam' (Set Interval Arithmetic Mode) is used to quickly override
|
||||||
|
the rounding mode during this routine.
|
||||||
|
|
||||||
|
We add then subtract (or subtract than add if the initial
|
||||||
|
value was negative) 2**23 to the value, then subtract it
|
||||||
|
back out.
|
||||||
|
|
||||||
|
This will clear out the fractional portion of the value and,
|
||||||
|
with suitable 'siam' initiated rouding mode settings, round
|
||||||
|
the final result in the proper direction. */
|
||||||
|
|
||||||
|
#define TWO_TWENTYTHREE 0x4b000000 /* 2**23 */
|
||||||
|
|
||||||
|
#define ZERO %f10 /* 0.0 */
|
||||||
|
#define SIGN_BIT %f12 /* -0.0 */
|
||||||
|
|
||||||
|
ENTRY (__floorf_vis2)
|
||||||
|
st %o0, [%sp + 68]
|
||||||
|
sethi %hi(TWO_TWENTYTHREE), %o2
|
||||||
|
fzeros ZERO
|
||||||
|
ld [%sp + 68], %f0
|
||||||
|
fnegs ZERO, SIGN_BIT
|
||||||
|
st %o2, [%sp + 68]
|
||||||
|
fabss %f0, %f14
|
||||||
|
ld [%sp + 68], %f16
|
||||||
|
fcmps %fcc3, %f14, %f16
|
||||||
|
fmovsuge %fcc3, ZERO, %f16
|
||||||
|
fands %f0, SIGN_BIT, SIGN_BIT
|
||||||
|
fors %f16, SIGN_BIT, %f16
|
||||||
|
siam (1 << 2) | 3
|
||||||
|
fadds %f0, %f16, %f1
|
||||||
|
siam (1 << 2) | 0
|
||||||
|
fsubs %f1, %f16, %f1
|
||||||
|
siam (0 << 2)
|
||||||
|
retl
|
||||||
|
fors %f1, SIGN_BIT, %f0
|
||||||
|
END (__floorf_vis2)
|
||||||
@ -19,27 +19,21 @@
|
|||||||
|
|
||||||
#include <sysdep.h>
|
#include <sysdep.h>
|
||||||
|
|
||||||
/* Since changing the rounding mode is extremely expensive, we
|
/* 'siam' (Set Interval Arithmetic Mode) is used to quickly override
|
||||||
try to round up using a method that is rounding mode
|
the rounding mode during this routine.
|
||||||
agnostic.
|
|
||||||
|
|
||||||
We add then subtract (or subtract than add if the initial
|
We add then subtract (or subtract than add if the initial
|
||||||
value was negative) 2**23 to the value, then subtract it
|
value was negative) 2**23 to the value, then subtract it
|
||||||
back out.
|
back out.
|
||||||
|
|
||||||
This will clear out the fractional portion of the value.
|
This will clear out the fractional portion of the value and,
|
||||||
One of two things will happen for non-whole initial values.
|
with suitable 'siam' initiated rouding mode settings, round
|
||||||
Either the rounding mode will round it up, or it will be
|
the final result in the proper direction.
|
||||||
rounded down. If the value started out whole, it will be
|
|
||||||
equal after the addition and subtraction. This means we
|
|
||||||
can accurately detect with one test whether we need to add
|
|
||||||
another 1.0 to round it up properly.
|
|
||||||
|
|
||||||
VIS instructions are used to facilitate the formation of
|
We also use VIS3 moves to avoid using the stack to transfer
|
||||||
easier constants, and the propagation of the sign bit. */
|
values between float and integer registers. */
|
||||||
|
|
||||||
#define TWO_TWENTYTHREE 0x4b000000 /* 2**23 */
|
#define TWO_TWENTYTHREE 0x4b000000 /* 2**23 */
|
||||||
#define ONE_DOT_ZERO 0x3f800000 /* 1.0 */
|
|
||||||
|
|
||||||
#define ZERO %f10 /* 0.0 */
|
#define ZERO %f10 /* 0.0 */
|
||||||
#define SIGN_BIT %f12 /* -0.0 */
|
#define SIGN_BIT %f12 /* -0.0 */
|
||||||
@ -47,28 +41,19 @@
|
|||||||
ENTRY (__floorf_vis3)
|
ENTRY (__floorf_vis3)
|
||||||
movwtos %o0, %f0
|
movwtos %o0, %f0
|
||||||
sethi %hi(TWO_TWENTYTHREE), %o2
|
sethi %hi(TWO_TWENTYTHREE), %o2
|
||||||
sethi %hi(ONE_DOT_ZERO), %o3
|
|
||||||
fzeros ZERO
|
fzeros ZERO
|
||||||
|
|
||||||
fnegs ZERO, SIGN_BIT
|
fnegs ZERO, SIGN_BIT
|
||||||
|
|
||||||
movwtos %o2, %f16
|
movwtos %o2, %f16
|
||||||
fabss %f0, %f14
|
fabss %f0, %f14
|
||||||
|
|
||||||
fcmps %fcc3, %f14, %f16
|
fcmps %fcc3, %f14, %f16
|
||||||
|
|
||||||
fmovsuge %fcc3, ZERO, %f16
|
fmovsuge %fcc3, ZERO, %f16
|
||||||
fands %f0, SIGN_BIT, SIGN_BIT
|
fands %f0, SIGN_BIT, SIGN_BIT
|
||||||
|
|
||||||
fors %f16, SIGN_BIT, %f16
|
fors %f16, SIGN_BIT, %f16
|
||||||
|
siam (1 << 2) | 3
|
||||||
fadds %f0, %f16, %f1
|
fadds %f0, %f16, %f1
|
||||||
|
siam (1 << 2) | 0
|
||||||
fsubs %f1, %f16, %f1
|
fsubs %f1, %f16, %f1
|
||||||
fcmps %fcc2, %f1, %f0
|
siam (0 << 2)
|
||||||
movwtos %o3, %f9
|
|
||||||
|
|
||||||
fmovsule %fcc2, ZERO, %f9
|
|
||||||
fsubs %f1, %f9, %f0
|
|
||||||
fabss %f0, %f0
|
|
||||||
retl
|
retl
|
||||||
fors %f0, SIGN_BIT, %f0
|
fors %f1, SIGN_BIT, %f0
|
||||||
END (__floorf_vis3)
|
END (__floorf_vis3)
|
||||||
|
|||||||
@ -1,6 +1,6 @@
|
|||||||
#include <sparc-ifunc.h>
|
#include <sparc-ifunc.h>
|
||||||
|
|
||||||
SPARC_ASM_VIS3_IFUNC(floorf)
|
SPARC_ASM_VIS3_VIS2_IFUNC(floorf)
|
||||||
|
|
||||||
weak_alias (__floorf, floorf)
|
weak_alias (__floorf, floorf)
|
||||||
|
|
||||||
|
|||||||
@ -1,4 +1,6 @@
|
|||||||
ifeq ($(subdir),math)
|
ifeq ($(subdir),math)
|
||||||
|
libm-sysdep_routines += s_ceil-vis2 s_ceilf-vis2 \
|
||||||
|
s_floor-vis2 s_floorf-vis2
|
||||||
ifeq ($(have-as-vis3),yes)
|
ifeq ($(have-as-vis3),yes)
|
||||||
libm-sysdep_routines += m_signbitf-vis3 m_signbit-vis3 s_ceilf-vis3 \
|
libm-sysdep_routines += m_signbitf-vis3 m_signbit-vis3 s_ceilf-vis3 \
|
||||||
s_ceil-vis3 m_finitef-vis3 m_finite-vis3 \
|
s_ceil-vis3 m_finitef-vis3 m_finite-vis3 \
|
||||||
|
|||||||
57
sysdeps/sparc/sparc64/fpu/multiarch/s_ceil-vis2.S
Normal file
57
sysdeps/sparc/sparc64/fpu/multiarch/s_ceil-vis2.S
Normal file
@ -0,0 +1,57 @@
|
|||||||
|
/* ceil function, sparc64 vis2 version.
|
||||||
|
Copyright (C) 2012-2013 Free Software Foundation, Inc.
|
||||||
|
This file is part of the GNU C Library.
|
||||||
|
Contributed by David S. Miller <davem@davemloft.net>, 2012.
|
||||||
|
|
||||||
|
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>
|
||||||
|
|
||||||
|
/* 'siam' (Set Interval Arithmetic Mode) is used to quickly override
|
||||||
|
the rounding mode during this routine.
|
||||||
|
|
||||||
|
We add then subtract (or subtract than add if the initial
|
||||||
|
value was negative) 2**23 to the value, then subtract it
|
||||||
|
back out.
|
||||||
|
|
||||||
|
This will clear out the fractional portion of the value and,
|
||||||
|
with suitable 'siam' initiated rouding mode settings, round
|
||||||
|
the final result in the proper direction. */
|
||||||
|
|
||||||
|
#define TWO_FIFTYTWO 0x43300000 /* 2**52 */
|
||||||
|
|
||||||
|
#define ZERO %f10 /* 0.0 */
|
||||||
|
#define SIGN_BIT %f12 /* -0.0 */
|
||||||
|
|
||||||
|
ENTRY (__ceil_vis2)
|
||||||
|
sethi %hi(TWO_FIFTYTWO), %o2
|
||||||
|
fzero ZERO
|
||||||
|
sllx %o2, 32, %o2
|
||||||
|
fnegd ZERO, SIGN_BIT
|
||||||
|
stx %o2, [%sp + STACK_BIAS + 128]
|
||||||
|
fabsd %f0, %f14
|
||||||
|
ldd [%sp + STACK_BIAS + 128], %f16
|
||||||
|
fcmpd %fcc3, %f14, %f16
|
||||||
|
fmovduge %fcc3, ZERO, %f16
|
||||||
|
fand %f0, SIGN_BIT, SIGN_BIT
|
||||||
|
for %f16, SIGN_BIT, %f16
|
||||||
|
siam (1 << 2) | 2
|
||||||
|
faddd %f0, %f16, %f18
|
||||||
|
siam (1 << 2) | 0
|
||||||
|
fsubd %f18, %f16, %f18
|
||||||
|
siam (0 << 2)
|
||||||
|
retl
|
||||||
|
for %f18, SIGN_BIT, %f0
|
||||||
|
END (__ceil_vis2)
|
||||||
@ -19,57 +19,41 @@
|
|||||||
|
|
||||||
#include <sysdep.h>
|
#include <sysdep.h>
|
||||||
|
|
||||||
/* Since changing the rounding mode is extremely expensive, we
|
/* 'siam' (Set Interval Arithmetic Mode) is used to quickly override
|
||||||
try to round up using a method that is rounding mode
|
the rounding mode during this routine.
|
||||||
agnostic.
|
|
||||||
|
|
||||||
We add then subtract (or subtract than add if the initial
|
We add then subtract (or subtract than add if the initial
|
||||||
value was negative) 2**23 to the value, then subtract it
|
value was negative) 2**23 to the value, then subtract it
|
||||||
back out.
|
back out.
|
||||||
|
|
||||||
This will clear out the fractional portion of the value.
|
This will clear out the fractional portion of the value and,
|
||||||
One of two things will happen for non-whole initial values.
|
with suitable 'siam' initiated rouding mode settings, round
|
||||||
Either the rounding mode will round it up, or it will be
|
the final result in the proper direction.
|
||||||
rounded down. If the value started out whole, it will be
|
|
||||||
equal after the addition and subtraction. This means we
|
|
||||||
can accurately detect with one test whether we need to add
|
|
||||||
another 1.0 to round it up properly.
|
|
||||||
|
|
||||||
VIS instructions are used to facilitate the formation of
|
We also use VIS3 moves to avoid using the stack to transfer
|
||||||
easier constants, and the propagation of the sign bit. */
|
values between float and integer registers. */
|
||||||
|
|
||||||
#define TWO_FIFTYTWO 0x43300000 /* 2**52 */
|
#define TWO_FIFTYTWO 0x43300000 /* 2**52 */
|
||||||
#define ONE_DOT_ZERO 0x3ff00000 /* 1.0 */
|
|
||||||
|
|
||||||
#define ZERO %f10 /* 0.0 */
|
#define ZERO %f10 /* 0.0 */
|
||||||
#define SIGN_BIT %f12 /* -0.0 */
|
#define SIGN_BIT %f12 /* -0.0 */
|
||||||
|
|
||||||
ENTRY (__ceil_vis3)
|
ENTRY (__ceil_vis3)
|
||||||
sethi %hi(TWO_FIFTYTWO), %o2
|
sethi %hi(TWO_FIFTYTWO), %o2
|
||||||
sethi %hi(ONE_DOT_ZERO), %o3
|
|
||||||
fzero ZERO
|
fzero ZERO
|
||||||
|
|
||||||
sllx %o2, 32, %o2
|
sllx %o2, 32, %o2
|
||||||
fnegd ZERO, SIGN_BIT
|
fnegd ZERO, SIGN_BIT
|
||||||
|
|
||||||
sllx %o3, 32, %o3
|
|
||||||
movxtod %o2, %f16
|
movxtod %o2, %f16
|
||||||
fabsd %f0, %f14
|
fabsd %f0, %f14
|
||||||
|
|
||||||
fcmpd %fcc3, %f14, %f16
|
fcmpd %fcc3, %f14, %f16
|
||||||
|
|
||||||
fmovduge %fcc3, ZERO, %f16
|
fmovduge %fcc3, ZERO, %f16
|
||||||
fand %f0, SIGN_BIT, SIGN_BIT
|
fand %f0, SIGN_BIT, SIGN_BIT
|
||||||
|
|
||||||
for %f16, SIGN_BIT, %f16
|
for %f16, SIGN_BIT, %f16
|
||||||
|
siam (1 << 2) | 2
|
||||||
faddd %f0, %f16, %f18
|
faddd %f0, %f16, %f18
|
||||||
|
siam (1 << 2) | 0
|
||||||
fsubd %f18, %f16, %f18
|
fsubd %f18, %f16, %f18
|
||||||
fcmpd %fcc2, %f18, %f0
|
siam (0 << 2)
|
||||||
movxtod %o3, %f20
|
|
||||||
|
|
||||||
fmovduge %fcc2, ZERO, %f20
|
|
||||||
faddd %f18, %f20, %f0
|
|
||||||
fabsd %f0, %f0
|
|
||||||
retl
|
retl
|
||||||
for %f0, SIGN_BIT, %f0
|
for %f18, SIGN_BIT, %f0
|
||||||
END (__ceil_vis3)
|
END (__ceil_vis3)
|
||||||
|
|||||||
@ -1,6 +1,6 @@
|
|||||||
#include <sparc-ifunc.h>
|
#include <sparc-ifunc.h>
|
||||||
|
|
||||||
SPARC_ASM_VIS3_IFUNC(ceil)
|
SPARC_ASM_VIS3_VIS2_IFUNC(ceil)
|
||||||
|
|
||||||
weak_alias (__ceil, ceil)
|
weak_alias (__ceil, ceil)
|
||||||
|
|
||||||
|
|||||||
56
sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf-vis2.S
Normal file
56
sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf-vis2.S
Normal file
@ -0,0 +1,56 @@
|
|||||||
|
/* Float ceil function, sparc64 vis2 version.
|
||||||
|
Copyright (C) 2012-2013 Free Software Foundation, Inc.
|
||||||
|
This file is part of the GNU C Library.
|
||||||
|
Contributed by David S. Miller <davem@davemloft.net>, 2012.
|
||||||
|
|
||||||
|
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>
|
||||||
|
|
||||||
|
/* 'siam' (Set Interval Arithmetic Mode) is used to quickly override
|
||||||
|
the rounding mode during this routine.
|
||||||
|
|
||||||
|
We add then subtract (or subtract than add if the initial
|
||||||
|
value was negative) 2**23 to the value, then subtract it
|
||||||
|
back out.
|
||||||
|
|
||||||
|
This will clear out the fractional portion of the value and,
|
||||||
|
with suitable 'siam' initiated rouding mode settings, round
|
||||||
|
the final result in the proper direction. */
|
||||||
|
|
||||||
|
#define TWO_TWENTYTHREE 0x4b000000 /* 2**23 */
|
||||||
|
|
||||||
|
#define ZERO %f10 /* 0.0 */
|
||||||
|
#define SIGN_BIT %f12 /* -0.0 */
|
||||||
|
|
||||||
|
ENTRY (__ceilf_vis2)
|
||||||
|
sethi %hi(TWO_TWENTYTHREE), %o2
|
||||||
|
fzeros ZERO
|
||||||
|
fnegs ZERO, SIGN_BIT
|
||||||
|
st %o2, [%sp + STACK_BIAS + 128]
|
||||||
|
fabss %f1, %f14
|
||||||
|
ld [%sp + STACK_BIAS + 128], %f16
|
||||||
|
fcmps %fcc3, %f14, %f16
|
||||||
|
fmovsuge %fcc3, ZERO, %f16
|
||||||
|
fands %f1, SIGN_BIT, SIGN_BIT
|
||||||
|
fors %f16, SIGN_BIT, %f16
|
||||||
|
siam (1 << 2) | 2
|
||||||
|
fadds %f1, %f16, %f5
|
||||||
|
siam (1 << 2) | 0
|
||||||
|
fsubs %f5, %f16, %f5
|
||||||
|
siam (0 << 2)
|
||||||
|
retl
|
||||||
|
fors %f5, SIGN_BIT, %f0
|
||||||
|
END (__ceilf_vis2)
|
||||||
@ -19,55 +19,40 @@
|
|||||||
|
|
||||||
#include <sysdep.h>
|
#include <sysdep.h>
|
||||||
|
|
||||||
/* Since changing the rounding mode is extremely expensive, we
|
/* 'siam' (Set Interval Arithmetic Mode) is used to quickly override
|
||||||
try to round up using a method that is rounding mode
|
the rounding mode during this routine.
|
||||||
agnostic.
|
|
||||||
|
|
||||||
We add then subtract (or subtract than add if the initial
|
We add then subtract (or subtract than add if the initial
|
||||||
value was negative) 2**23 to the value, then subtract it
|
value was negative) 2**23 to the value, then subtract it
|
||||||
back out.
|
back out.
|
||||||
|
|
||||||
This will clear out the fractional portion of the value.
|
This will clear out the fractional portion of the value and,
|
||||||
One of two things will happen for non-whole initial values.
|
with suitable 'siam' initiated rouding mode settings, round
|
||||||
Either the rounding mode will round it up, or it will be
|
the final result in the proper direction.
|
||||||
rounded down. If the value started out whole, it will be
|
|
||||||
equal after the addition and subtraction. This means we
|
|
||||||
can accurately detect with one test whether we need to add
|
|
||||||
another 1.0 to round it up properly.
|
|
||||||
|
|
||||||
VIS instructions are used to facilitate the formation of
|
We also use VIS3 moves to avoid using the stack to transfer
|
||||||
easier constants, and the propagation of the sign bit. */
|
values between float and integer registers. */
|
||||||
|
|
||||||
#define TWO_TWENTYTHREE 0x4b000000 /* 2**23 */
|
#define TWO_TWENTYTHREE 0x4b000000 /* 2**23 */
|
||||||
#define ONE_DOT_ZERO 0x3f800000 /* 1.0 */
|
|
||||||
|
|
||||||
#define ZERO %f10 /* 0.0 */
|
#define ZERO %f10 /* 0.0 */
|
||||||
#define SIGN_BIT %f12 /* -0.0 */
|
#define SIGN_BIT %f12 /* -0.0 */
|
||||||
|
|
||||||
ENTRY (__ceilf_vis3)
|
ENTRY (__ceilf_vis3)
|
||||||
sethi %hi(TWO_TWENTYTHREE), %o2
|
sethi %hi(TWO_TWENTYTHREE), %o2
|
||||||
sethi %hi(ONE_DOT_ZERO), %o3
|
|
||||||
fzeros ZERO
|
fzeros ZERO
|
||||||
|
|
||||||
fnegs ZERO, SIGN_BIT
|
fnegs ZERO, SIGN_BIT
|
||||||
|
|
||||||
movwtos %o2, %f16
|
movwtos %o2, %f16
|
||||||
fabss %f1, %f14
|
fabss %f1, %f14
|
||||||
|
|
||||||
fcmps %fcc3, %f14, %f16
|
fcmps %fcc3, %f14, %f16
|
||||||
|
|
||||||
fmovsuge %fcc3, ZERO, %f16
|
fmovsuge %fcc3, ZERO, %f16
|
||||||
fands %f1, SIGN_BIT, SIGN_BIT
|
fands %f1, SIGN_BIT, SIGN_BIT
|
||||||
|
|
||||||
fors %f16, SIGN_BIT, %f16
|
fors %f16, SIGN_BIT, %f16
|
||||||
|
siam (1 << 2) | 2
|
||||||
fadds %f1, %f16, %f5
|
fadds %f1, %f16, %f5
|
||||||
|
siam (1 << 2) | 0
|
||||||
fsubs %f5, %f16, %f5
|
fsubs %f5, %f16, %f5
|
||||||
fcmps %fcc2, %f5, %f1
|
siam (0 << 2)
|
||||||
movwtos %o3, %f9
|
|
||||||
|
|
||||||
fmovsuge %fcc2, ZERO, %f9
|
|
||||||
fadds %f5, %f9, %f0
|
|
||||||
fabss %f0, %f0
|
|
||||||
retl
|
retl
|
||||||
fors %f0, SIGN_BIT, %f0
|
fors %f5, SIGN_BIT, %f0
|
||||||
END (__ceilf_vis3)
|
END (__ceilf_vis3)
|
||||||
|
|||||||
@ -1,6 +1,6 @@
|
|||||||
#include <sparc-ifunc.h>
|
#include <sparc-ifunc.h>
|
||||||
|
|
||||||
SPARC_ASM_VIS3_IFUNC(ceilf)
|
SPARC_ASM_VIS3_VIS2_IFUNC(ceilf)
|
||||||
|
|
||||||
weak_alias (__ceilf, ceilf)
|
weak_alias (__ceilf, ceilf)
|
||||||
|
|
||||||
|
|||||||
57
sysdeps/sparc/sparc64/fpu/multiarch/s_floor-vis2.S
Normal file
57
sysdeps/sparc/sparc64/fpu/multiarch/s_floor-vis2.S
Normal file
@ -0,0 +1,57 @@
|
|||||||
|
/* floor function, sparc64 vis2 version.
|
||||||
|
Copyright (C) 2012-2013 Free Software Foundation, Inc.
|
||||||
|
This file is part of the GNU C Library.
|
||||||
|
Contributed by David S. Miller <davem@davemloft.net>, 2012.
|
||||||
|
|
||||||
|
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>
|
||||||
|
|
||||||
|
/* 'siam' (Set Interval Arithmetic Mode) is used to quickly override
|
||||||
|
the rounding mode during this routine.
|
||||||
|
|
||||||
|
We add then subtract (or subtract than add if the initial
|
||||||
|
value was negative) 2**23 to the value, then subtract it
|
||||||
|
back out.
|
||||||
|
|
||||||
|
This will clear out the fractional portion of the value and,
|
||||||
|
with suitable 'siam' initiated rouding mode settings, round
|
||||||
|
the final result in the proper direction. */
|
||||||
|
|
||||||
|
#define TWO_FIFTYTWO 0x43300000 /* 2**52 */
|
||||||
|
|
||||||
|
#define ZERO %f10 /* 0.0 */
|
||||||
|
#define SIGN_BIT %f12 /* -0.0 */
|
||||||
|
|
||||||
|
ENTRY (__floor_vis2)
|
||||||
|
sethi %hi(TWO_FIFTYTWO), %o2
|
||||||
|
fzero ZERO
|
||||||
|
sllx %o2, 32, %o2
|
||||||
|
fnegd ZERO, SIGN_BIT
|
||||||
|
stx %o2, [%sp + STACK_BIAS + 128]
|
||||||
|
fabsd %f0, %f14
|
||||||
|
ldd [%sp + STACK_BIAS + 128], %f16
|
||||||
|
fcmpd %fcc3, %f14, %f16
|
||||||
|
fmovduge %fcc3, ZERO, %f16
|
||||||
|
fand %f0, SIGN_BIT, SIGN_BIT
|
||||||
|
for %f16, SIGN_BIT, %f16
|
||||||
|
siam (1 << 2) | 3
|
||||||
|
faddd %f0, %f16, %f18
|
||||||
|
siam (1 << 2) | 0
|
||||||
|
fsubd %f18, %f16, %f18
|
||||||
|
siam (0 << 2)
|
||||||
|
retl
|
||||||
|
for %f18, SIGN_BIT, %f0
|
||||||
|
END (__floor_vis2)
|
||||||
@ -19,57 +19,41 @@
|
|||||||
|
|
||||||
#include <sysdep.h>
|
#include <sysdep.h>
|
||||||
|
|
||||||
/* Since changing the rounding mode is extremely expensive, we
|
/* 'siam' (Set Interval Arithmetic Mode) is used to quickly override
|
||||||
try to round up using a method that is rounding mode
|
the rounding mode during this routine.
|
||||||
agnostic.
|
|
||||||
|
|
||||||
We add then subtract (or subtract than add if the initial
|
We add then subtract (or subtract than add if the initial
|
||||||
value was negative) 2**23 to the value, then subtract it
|
value was negative) 2**23 to the value, then subtract it
|
||||||
back out.
|
back out.
|
||||||
|
|
||||||
This will clear out the fractional portion of the value.
|
This will clear out the fractional portion of the value and,
|
||||||
One of two things will happen for non-whole initial values.
|
with suitable 'siam' initiated rouding mode settings, round
|
||||||
Either the rounding mode will round it up, or it will be
|
the final result in the proper direction.
|
||||||
rounded down. If the value started out whole, it will be
|
|
||||||
equal after the addition and subtraction. This means we
|
|
||||||
can accurately detect with one test whether we need to add
|
|
||||||
another 1.0 to round it up properly.
|
|
||||||
|
|
||||||
VIS instructions are used to facilitate the formation of
|
We also use VIS3 moves to avoid using the stack to transfer
|
||||||
easier constants, and the propagation of the sign bit. */
|
values between float and integer registers. */
|
||||||
|
|
||||||
#define TWO_FIFTYTWO 0x43300000 /* 2**52 */
|
#define TWO_FIFTYTWO 0x43300000 /* 2**52 */
|
||||||
#define ONE_DOT_ZERO 0x3ff00000 /* 1.0 */
|
|
||||||
|
|
||||||
#define ZERO %f10 /* 0.0 */
|
#define ZERO %f10 /* 0.0 */
|
||||||
#define SIGN_BIT %f12 /* -0.0 */
|
#define SIGN_BIT %f12 /* -0.0 */
|
||||||
|
|
||||||
ENTRY (__floor_vis3)
|
ENTRY (__floor_vis3)
|
||||||
sethi %hi(TWO_FIFTYTWO), %o2
|
sethi %hi(TWO_FIFTYTWO), %o2
|
||||||
sethi %hi(ONE_DOT_ZERO), %o3
|
|
||||||
fzero ZERO
|
fzero ZERO
|
||||||
|
|
||||||
sllx %o2, 32, %o2
|
sllx %o2, 32, %o2
|
||||||
fnegd ZERO, SIGN_BIT
|
fnegd ZERO, SIGN_BIT
|
||||||
|
|
||||||
sllx %o3, 32, %o3
|
|
||||||
movxtod %o2, %f16
|
movxtod %o2, %f16
|
||||||
fabsd %f0, %f14
|
fabsd %f0, %f14
|
||||||
|
|
||||||
fcmpd %fcc3, %f14, %f16
|
fcmpd %fcc3, %f14, %f16
|
||||||
|
|
||||||
fmovduge %fcc3, ZERO, %f16
|
fmovduge %fcc3, ZERO, %f16
|
||||||
fand %f0, SIGN_BIT, SIGN_BIT
|
fand %f0, SIGN_BIT, SIGN_BIT
|
||||||
|
|
||||||
for %f16, SIGN_BIT, %f16
|
for %f16, SIGN_BIT, %f16
|
||||||
|
siam (1 << 2) | 3
|
||||||
faddd %f0, %f16, %f18
|
faddd %f0, %f16, %f18
|
||||||
|
siam (1 << 2) | 0
|
||||||
fsubd %f18, %f16, %f18
|
fsubd %f18, %f16, %f18
|
||||||
fcmpd %fcc2, %f18, %f0
|
siam (0 << 2)
|
||||||
movxtod %o3, %f20
|
|
||||||
|
|
||||||
fmovdule %fcc2, ZERO, %f20
|
|
||||||
fsubd %f18, %f20, %f0
|
|
||||||
fabsd %f0, %f0
|
|
||||||
retl
|
retl
|
||||||
for %f0, SIGN_BIT, %f0
|
for %f18, SIGN_BIT, %f0
|
||||||
END (__floor_vis3)
|
END (__floor_vis3)
|
||||||
|
|||||||
@ -1,6 +1,6 @@
|
|||||||
#include <sparc-ifunc.h>
|
#include <sparc-ifunc.h>
|
||||||
|
|
||||||
SPARC_ASM_VIS3_IFUNC(floor)
|
SPARC_ASM_VIS3_VIS2_IFUNC(floor)
|
||||||
|
|
||||||
weak_alias (__floor, floor)
|
weak_alias (__floor, floor)
|
||||||
|
|
||||||
|
|||||||
56
sysdeps/sparc/sparc64/fpu/multiarch/s_floorf-vis2.S
Normal file
56
sysdeps/sparc/sparc64/fpu/multiarch/s_floorf-vis2.S
Normal file
@ -0,0 +1,56 @@
|
|||||||
|
/* Float floor function, sparc64 vis2 version.
|
||||||
|
Copyright (C) 2012-2013 Free Software Foundation, Inc.
|
||||||
|
This file is part of the GNU C Library.
|
||||||
|
Contributed by David S. Miller <davem@davemloft.net>, 2012.
|
||||||
|
|
||||||
|
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>
|
||||||
|
|
||||||
|
/* 'siam' (Set Interval Arithmetic Mode) is used to quickly override
|
||||||
|
the rounding mode during this routine.
|
||||||
|
|
||||||
|
We add then subtract (or subtract than add if the initial
|
||||||
|
value was negative) 2**23 to the value, then subtract it
|
||||||
|
back out.
|
||||||
|
|
||||||
|
This will clear out the fractional portion of the value and,
|
||||||
|
with suitable 'siam' initiated rouding mode settings, round
|
||||||
|
the final result in the proper direction. */
|
||||||
|
|
||||||
|
#define TWO_TWENTYTHREE 0x4b000000 /* 2**23 */
|
||||||
|
|
||||||
|
#define ZERO %f10 /* 0.0 */
|
||||||
|
#define SIGN_BIT %f12 /* -0.0 */
|
||||||
|
|
||||||
|
ENTRY (__floorf_vis2)
|
||||||
|
sethi %hi(TWO_TWENTYTHREE), %o2
|
||||||
|
fzeros ZERO
|
||||||
|
fnegs ZERO, SIGN_BIT
|
||||||
|
st %o2, [%sp + STACK_BIAS + 128]
|
||||||
|
fabss %f1, %f14
|
||||||
|
ld [%sp + STACK_BIAS + 128], %f16
|
||||||
|
fcmps %fcc3, %f14, %f16
|
||||||
|
fmovsuge %fcc3, ZERO, %f16
|
||||||
|
fands %f1, SIGN_BIT, SIGN_BIT
|
||||||
|
fors %f16, SIGN_BIT, %f16
|
||||||
|
siam (1 << 2) | 3
|
||||||
|
fadds %f1, %f16, %f5
|
||||||
|
siam (1 << 2) | 0
|
||||||
|
fsubs %f5, %f16, %f5
|
||||||
|
siam (0 << 2)
|
||||||
|
retl
|
||||||
|
fors %f5, SIGN_BIT, %f0
|
||||||
|
END (__floorf_vis2)
|
||||||
@ -19,55 +19,40 @@
|
|||||||
|
|
||||||
#include <sysdep.h>
|
#include <sysdep.h>
|
||||||
|
|
||||||
/* Since changing the rounding mode is extremely expensive, we
|
/* 'siam' (Set Interval Arithmetic Mode) is used to quickly override
|
||||||
try to round up using a method that is rounding mode
|
the rounding mode during this routine.
|
||||||
agnostic.
|
|
||||||
|
|
||||||
We add then subtract (or subtract than add if the initial
|
We add then subtract (or subtract than add if the initial
|
||||||
value was negative) 2**23 to the value, then subtract it
|
value was negative) 2**23 to the value, then subtract it
|
||||||
back out.
|
back out.
|
||||||
|
|
||||||
This will clear out the fractional portion of the value.
|
This will clear out the fractional portion of the value and,
|
||||||
One of two things will happen for non-whole initial values.
|
with suitable 'siam' initiated rouding mode settings, round
|
||||||
Either the rounding mode will round it up, or it will be
|
the final result in the proper direction.
|
||||||
rounded down. If the value started out whole, it will be
|
|
||||||
equal after the addition and subtraction. This means we
|
|
||||||
can accurately detect with one test whether we need to add
|
|
||||||
another 1.0 to round it up properly.
|
|
||||||
|
|
||||||
VIS instructions are used to facilitate the formation of
|
We also use VIS3 moves to avoid using the stack to transfer
|
||||||
easier constants, and the propagation of the sign bit. */
|
values between float and integer registers. */
|
||||||
|
|
||||||
#define TWO_TWENTYTHREE 0x4b000000 /* 2**23 */
|
#define TWO_TWENTYTHREE 0x4b000000 /* 2**23 */
|
||||||
#define ONE_DOT_ZERO 0x3f800000 /* 1.0 */
|
|
||||||
|
|
||||||
#define ZERO %f10 /* 0.0 */
|
#define ZERO %f10 /* 0.0 */
|
||||||
#define SIGN_BIT %f12 /* -0.0 */
|
#define SIGN_BIT %f12 /* -0.0 */
|
||||||
|
|
||||||
ENTRY (__floorf_vis3)
|
ENTRY (__floorf_vis3)
|
||||||
sethi %hi(TWO_TWENTYTHREE), %o2
|
sethi %hi(TWO_TWENTYTHREE), %o2
|
||||||
sethi %hi(ONE_DOT_ZERO), %o3
|
|
||||||
fzeros ZERO
|
fzeros ZERO
|
||||||
|
|
||||||
fnegs ZERO, SIGN_BIT
|
fnegs ZERO, SIGN_BIT
|
||||||
|
movwtos %o2, %f16
|
||||||
movwtos %o2, %f16
|
|
||||||
fabss %f1, %f14
|
fabss %f1, %f14
|
||||||
|
|
||||||
fcmps %fcc3, %f14, %f16
|
fcmps %fcc3, %f14, %f16
|
||||||
|
|
||||||
fmovsuge %fcc3, ZERO, %f16
|
fmovsuge %fcc3, ZERO, %f16
|
||||||
fands %f1, SIGN_BIT, SIGN_BIT
|
fands %f1, SIGN_BIT, SIGN_BIT
|
||||||
|
|
||||||
fors %f16, SIGN_BIT, %f16
|
fors %f16, SIGN_BIT, %f16
|
||||||
|
siam (1 << 2) | 3
|
||||||
fadds %f1, %f16, %f5
|
fadds %f1, %f16, %f5
|
||||||
|
siam (1 << 2) | 0
|
||||||
fsubs %f5, %f16, %f5
|
fsubs %f5, %f16, %f5
|
||||||
fcmps %fcc2, %f5, %f1
|
siam (0 << 2)
|
||||||
movwtos %o3, %f9
|
|
||||||
|
|
||||||
fmovsule %fcc2, ZERO, %f9
|
|
||||||
fsubs %f5, %f9, %f0
|
|
||||||
fabss %f0, %f0
|
|
||||||
retl
|
retl
|
||||||
fors %f0, SIGN_BIT, %f0
|
fors %f5, SIGN_BIT, %f0
|
||||||
END (__floorf_vis3)
|
END (__floorf_vis3)
|
||||||
|
|||||||
@ -1,6 +1,6 @@
|
|||||||
#include <sparc-ifunc.h>
|
#include <sparc-ifunc.h>
|
||||||
|
|
||||||
SPARC_ASM_VIS3_IFUNC(floorf)
|
SPARC_ASM_VIS3_VIS2_IFUNC(floorf)
|
||||||
|
|
||||||
weak_alias (__floorf, floorf)
|
weak_alias (__floorf, floorf)
|
||||||
|
|
||||||
|
|||||||
Loading…
Reference in New Issue
Block a user