mirror of
https://sourceware.org/git/glibc.git
synced 2024-11-22 13:00:06 +00:00
Optimize nearbyint{,f} on sparc.
* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/Makefile: Add vis3 nearbyint{,f} to libm-sysdep_routes. * sysdeps/sparc/sparc64/fpu/multiarch/Makefile: Likewise. * sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_nearbyint-vis3.S: New file. * sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_nearbyint.S: New file. * sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_nearbyintf-vis3.S: New file. * sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_nearbyintf.S: New file. * sysdeps/sparc/sparc32/sparcv9/fpu/s_nearbyint.S: New file. * sysdeps/sparc/sparc32/sparcv9/fpu/s_nearbyintf.S: New file. * sysdeps/sparc/sparc64/fpu/multiarch/s_nearbyint-vis3.S: New file. * sysdeps/sparc/sparc64/fpu/multiarch/s_nearbyint.S: New file. * sysdeps/sparc/sparc64/fpu/multiarch/s_nearbyintf-vis3.S: New file. * sysdeps/sparc/sparc64/fpu/multiarch/s_nearbyintf.S: New file. * sysdeps/sparc/sparc64/fpu/s_nearbyint.S: New file. * sysdeps/sparc/sparc64/fpu/s_nearbyintf.S: New file.
This commit is contained in:
parent
c19a9f896c
commit
c42d5e9862
22
ChangeLog
22
ChangeLog
@ -1,5 +1,27 @@
|
||||
2013-01-15 David S. Miller <davem@davemloft.net>
|
||||
|
||||
* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/Makefile: Add vis3
|
||||
nearbyint{,f} to libm-sysdep_routes.
|
||||
* sysdeps/sparc/sparc64/fpu/multiarch/Makefile: Likewise.
|
||||
* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_nearbyint-vis3.S:
|
||||
New file.
|
||||
* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_nearbyint.S: New
|
||||
file.
|
||||
* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_nearbyintf-vis3.S:
|
||||
New file.
|
||||
* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_nearbyintf.S: New
|
||||
file.
|
||||
* sysdeps/sparc/sparc32/sparcv9/fpu/s_nearbyint.S: New file.
|
||||
* sysdeps/sparc/sparc32/sparcv9/fpu/s_nearbyintf.S: New file.
|
||||
* sysdeps/sparc/sparc64/fpu/multiarch/s_nearbyint-vis3.S: New
|
||||
file.
|
||||
* sysdeps/sparc/sparc64/fpu/multiarch/s_nearbyint.S: New file.
|
||||
* sysdeps/sparc/sparc64/fpu/multiarch/s_nearbyintf-vis3.S: New
|
||||
file.
|
||||
* sysdeps/sparc/sparc64/fpu/multiarch/s_nearbyintf.S: New file.
|
||||
* sysdeps/sparc/sparc64/fpu/s_nearbyint.S: New file.
|
||||
* sysdeps/sparc/sparc64/fpu/s_nearbyintf.S: New file.
|
||||
|
||||
* sysdeps/ieee754/dbl-64/s_nearbyint.c (__nearbyint): Use
|
||||
libc_feholdexcept and libc_fesetenv.
|
||||
|
||||
|
@ -7,7 +7,8 @@ libm-sysdep_routines += m_copysignf-vis3 m_copysign-vis3 s_ceilf-vis3 \
|
||||
s_floorf-vis3 s_llrintf-vis3 s_llrint-vis3 \
|
||||
s_rintf-vis3 s_rint-vis3 w_sqrt-vis3 w_sqrtf-vis3 \
|
||||
s_fminf-vis3 s_fmin-vis3 s_fmaxf-vis3 s_fmax-vis3 \
|
||||
s_fmaf-vis3 s_fma-vis3 s_fdimf-vis3 s_fdim-vis3
|
||||
s_fmaf-vis3 s_fma-vis3 s_fdimf-vis3 s_fdim-vis3 \
|
||||
s_nearbyint-vis3 s_nearbyintf-vis3
|
||||
sysdep_routines += s_copysignf-vis3 s_copysign-vis3
|
||||
endif
|
||||
endif
|
||||
|
@ -0,0 +1,65 @@
|
||||
/* Round float to int floating-point values without generating
|
||||
an inexact exception, sparc32 v9 vis3 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>
|
||||
#include <math_ldbl_opt.h>
|
||||
|
||||
/* We pop constants into the FPU registers using the incoming
|
||||
argument stack slots, since this avoid having to use any PIC
|
||||
references. We also thus avoid having to allocate a register
|
||||
window.
|
||||
|
||||
VIS instructions are used to facilitate the formation of
|
||||
easier constants, and the propagation of the sign bit. */
|
||||
|
||||
#define TWO_FIFTYTWO 0x43300000 /* 2**52 */
|
||||
|
||||
#define ZERO %f10 /* 0.0 */
|
||||
#define SIGN_BIT %f12 /* -0.0 */
|
||||
|
||||
ENTRY (__nearbyint_vis3)
|
||||
st %fsr, [%sp + 88]
|
||||
sethi %hi(TWO_FIFTYTWO), %o2
|
||||
sethi %hi(0xf8003e0), %o5
|
||||
ld [%sp + 88], %o4
|
||||
sllx %o0, 32, %o0
|
||||
or %o5, %lo(0xf8003e0), %o5
|
||||
or %o0, %o1, %o0
|
||||
movxtod %o0, %f0
|
||||
andn %o4, %o5, %o4
|
||||
fzero ZERO
|
||||
st %o4, [%sp + 80]
|
||||
sllx %o2, 32, %o2
|
||||
fnegd ZERO, SIGN_BIT
|
||||
ld [%sp + 80], %fsr
|
||||
movxtod %o2, %f16
|
||||
fabsd %f0, %f14
|
||||
fcmpd %fcc3, %f14, %f16
|
||||
fmovduge %fcc3, ZERO, %f16
|
||||
fand %f0, SIGN_BIT, SIGN_BIT
|
||||
for %f16, SIGN_BIT, %f16
|
||||
faddd %f0, %f16, %f6
|
||||
fsubd %f6, %f16, %f0
|
||||
fabsd %f0, %f0
|
||||
for %f0, SIGN_BIT, %f0
|
||||
retl
|
||||
ld [%sp + 88], %fsr
|
||||
END (__nearbyint_vis3)
|
19
sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_nearbyint.S
Normal file
19
sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_nearbyint.S
Normal file
@ -0,0 +1,19 @@
|
||||
#include <sparc-ifunc.h>
|
||||
#include <math_ldbl_opt.h>
|
||||
|
||||
SPARC_ASM_VIS3_IFUNC(nearbyint)
|
||||
|
||||
weak_alias (__nearbyint, nearbyint)
|
||||
|
||||
#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_1)
|
||||
compat_symbol (libm, __nearbyint, nearbyintl, GLIBC_2_1)
|
||||
#endif
|
||||
|
||||
# undef weak_alias
|
||||
# define weak_alias(a, b)
|
||||
# undef compat_symbol
|
||||
# define compat_symbol(a, b, c, d)
|
||||
|
||||
#define __nearbyint __nearbyint_generic
|
||||
|
||||
#include "../s_nearbyint.S"
|
@ -0,0 +1,61 @@
|
||||
/* Round float to int floating-point values without generating
|
||||
an inexact exception, sparc32 v9 vis3 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>
|
||||
|
||||
/* We pop constants into the FPU registers using the incoming
|
||||
argument stack slots, since this avoid having to use any PIC
|
||||
references. We also thus avoid having to allocate a register
|
||||
window.
|
||||
|
||||
VIS instructions are used to facilitate the formation of
|
||||
easier constants, and the propagation of the sign bit. */
|
||||
|
||||
#define TWO_TWENTYTHREE 0x4b000000 /* 2**23 */
|
||||
|
||||
#define ZERO %f10 /* 0.0 */
|
||||
#define SIGN_BIT %f12 /* -0.0 */
|
||||
|
||||
ENTRY (__nearbyintf_vis3)
|
||||
st %fsr, [%sp + 88]
|
||||
movwtos %o0, %f1
|
||||
sethi %hi(TWO_TWENTYTHREE), %o2
|
||||
sethi %hi(0xf8003e0), %o5
|
||||
ld [%sp + 88], %o4
|
||||
fzeros ZERO
|
||||
or %o5, %lo(0xf8003e0), %o5
|
||||
fnegs ZERO, SIGN_BIT
|
||||
andn %o4, %o5, %o4
|
||||
st %o4, [%sp + 80]
|
||||
ld [%sp + 80], %fsr
|
||||
movwtos %o2, %f16
|
||||
fabss %f1, %f14
|
||||
fcmps %fcc3, %f14, %f16
|
||||
fmovsuge %fcc3, ZERO, %f16
|
||||
fands %f1, SIGN_BIT, SIGN_BIT
|
||||
fors %f16, SIGN_BIT, %f16
|
||||
fadds %f1, %f16, %f5
|
||||
fsubs %f5, %f16, %f0
|
||||
fabss %f0, %f0
|
||||
fors %f0, SIGN_BIT, %f0
|
||||
retl
|
||||
ld [%sp + 88], %fsr
|
||||
END (__nearbyintf_vis3)
|
12
sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_nearbyintf.S
Normal file
12
sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_nearbyintf.S
Normal file
@ -0,0 +1,12 @@
|
||||
#include <sparc-ifunc.h>
|
||||
|
||||
SPARC_ASM_VIS3_IFUNC(nearbyintf)
|
||||
|
||||
weak_alias (__nearbyintf, nearbyintf)
|
||||
|
||||
# undef weak_alias
|
||||
# define weak_alias(a, b)
|
||||
|
||||
#define __nearbyintf __nearbyintf_generic
|
||||
|
||||
#include "../s_nearbyintf.S"
|
72
sysdeps/sparc/sparc32/sparcv9/fpu/s_nearbyint.S
Normal file
72
sysdeps/sparc/sparc32/sparcv9/fpu/s_nearbyint.S
Normal file
@ -0,0 +1,72 @@
|
||||
/* Round float to int floating-point values without generating
|
||||
an inexact exception, sparc32 v9 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>
|
||||
#include <math_ldbl_opt.h>
|
||||
|
||||
/* We pop constants into the FPU registers using the incoming
|
||||
argument stack slots, since this avoid having to use any PIC
|
||||
references. We also thus avoid having to allocate a register
|
||||
window.
|
||||
|
||||
VIS instructions are used to facilitate the formation of
|
||||
easier constants, and the propagation of the sign bit. */
|
||||
|
||||
#define TWO_FIFTYTWO 0x43300000 /* 2**52 */
|
||||
|
||||
#define ZERO %f10 /* 0.0 */
|
||||
#define SIGN_BIT %f12 /* -0.0 */
|
||||
|
||||
ENTRY (__nearbyint)
|
||||
st %fsr, [%sp + 88]
|
||||
sethi %hi(TWO_FIFTYTWO), %o2
|
||||
sethi %hi(0xf8003e0), %o5
|
||||
ld [%sp + 88], %o4
|
||||
sllx %o0, 32, %o0
|
||||
or %o5, %lo(0xf8003e0), %o5
|
||||
or %o0, %o1, %o0
|
||||
andn %o4, %o5, %o4
|
||||
fzero ZERO
|
||||
st %o4, [%sp + 80]
|
||||
stx %o0, [%sp + 72]
|
||||
sllx %o2, 32, %o2
|
||||
fnegd ZERO, SIGN_BIT
|
||||
ldd [%sp + 72], %f0
|
||||
ld [%sp + 80], %fsr
|
||||
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
|
||||
faddd %f0, %f16, %f6
|
||||
fsubd %f6, %f16, %f0
|
||||
fabsd %f0, %f0
|
||||
for %f0, SIGN_BIT, %f0
|
||||
retl
|
||||
ld [%sp + 88], %fsr
|
||||
END (__nearbyint)
|
||||
weak_alias (__nearbyint, nearbyint)
|
||||
|
||||
#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_1)
|
||||
compat_symbol (libm, __nearbyint, nearbyintl, GLIBC_2_1)
|
||||
#endif
|
64
sysdeps/sparc/sparc32/sparcv9/fpu/s_nearbyintf.S
Normal file
64
sysdeps/sparc/sparc32/sparcv9/fpu/s_nearbyintf.S
Normal file
@ -0,0 +1,64 @@
|
||||
/* Round float to int floating-point values without generating
|
||||
an inexact exception, sparc32 v9 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>
|
||||
|
||||
/* We pop constants into the FPU registers using the incoming
|
||||
argument stack slots, since this avoid having to use any PIC
|
||||
references. We also thus avoid having to allocate a register
|
||||
window.
|
||||
|
||||
VIS instructions are used to facilitate the formation of
|
||||
easier constants, and the propagation of the sign bit. */
|
||||
|
||||
#define TWO_TWENTYTHREE 0x4b000000 /* 2**23 */
|
||||
|
||||
#define ZERO %f10 /* 0.0 */
|
||||
#define SIGN_BIT %f12 /* -0.0 */
|
||||
|
||||
ENTRY (__nearbyintf)
|
||||
st %fsr, [%sp + 88]
|
||||
st %o0, [%sp + 68]
|
||||
sethi %hi(TWO_TWENTYTHREE), %o2
|
||||
sethi %hi(0xf8003e0), %o5
|
||||
ld [%sp + 88], %o4
|
||||
fzeros ZERO
|
||||
or %o5, %lo(0xf8003e0), %o5
|
||||
fnegs ZERO, SIGN_BIT
|
||||
andn %o4, %o5, %o4
|
||||
st %o4, [%sp + 80]
|
||||
ld [%sp + 68], %f1
|
||||
ld [%sp + 80], %fsr
|
||||
st %o2, [%sp + 68]
|
||||
fabss %f1, %f14
|
||||
ld [%sp + 68], %f16
|
||||
fcmps %fcc3, %f14, %f16
|
||||
fmovsuge %fcc3, ZERO, %f16
|
||||
fands %f1, SIGN_BIT, SIGN_BIT
|
||||
fors %f16, SIGN_BIT, %f16
|
||||
fadds %f1, %f16, %f5
|
||||
fsubs %f5, %f16, %f0
|
||||
fabss %f0, %f0
|
||||
fors %f0, SIGN_BIT, %f0
|
||||
retl
|
||||
ld [%sp + 88], %fsr
|
||||
END (__nearbyintf)
|
||||
weak_alias (__nearbyintf, nearbyintf)
|
@ -9,7 +9,7 @@ libm-sysdep_routines += m_signbitf-vis3 m_signbit-vis3 s_ceilf-vis3 \
|
||||
s_lrintf-vis3 s_lrint-vis3 s_rintf-vis3 \
|
||||
s_rint-vis3 s_fminf-vis3 s_fmin-vis3 \
|
||||
s_fmaxf-vis3 s_fmax-vis3 s_fmaf-vis3 \
|
||||
s_fma-vis3
|
||||
s_fma-vis3 s_nearbyint-vis3 s_nearbyintf-vis3
|
||||
sysdep_routines += s_signbitf-vis3 s_signbit-vis3 s_finitef-vis3 \
|
||||
s_finite-vis3 s_isinff-vis3 s_isinf-vis3 \
|
||||
s_isnanf-vis3 s_isnan-vis3
|
||||
|
61
sysdeps/sparc/sparc64/fpu/multiarch/s_nearbyint-vis3.S
Normal file
61
sysdeps/sparc/sparc64/fpu/multiarch/s_nearbyint-vis3.S
Normal file
@ -0,0 +1,61 @@
|
||||
/* Round float to int floating-point values without generating
|
||||
an inexact exception, sparc64 vis3 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>
|
||||
|
||||
/* We pop constants into the FPU registers using the incoming
|
||||
argument stack slots, since this avoid having to use any PIC
|
||||
references. We also thus avoid having to allocate a register
|
||||
window.
|
||||
|
||||
VIS instructions are used to facilitate the formation of
|
||||
easier constants, and the propagation of the sign bit. */
|
||||
|
||||
#define TWO_FIFTYTWO 0x43300000 /* 2**52 */
|
||||
|
||||
#define ZERO %f10 /* 0.0 */
|
||||
#define SIGN_BIT %f12 /* -0.0 */
|
||||
|
||||
ENTRY (__nearbyint_vis3)
|
||||
stx %fsr, [%sp + STACK_BIAS + 144]
|
||||
sethi %hi(TWO_FIFTYTWO), %o2
|
||||
sllx %o2, 32, %o2
|
||||
ldx [%sp + STACK_BIAS + 144], %o4
|
||||
sethi %hi(0xf8003e0), %o5
|
||||
fzero ZERO
|
||||
or %o5, %lo(0xf8003e0), %o5
|
||||
fnegd ZERO, SIGN_BIT
|
||||
andn %o4, %o5, %o4
|
||||
movxtod %o2, %f16
|
||||
stx %o4, [%sp + STACK_BIAS + 136]
|
||||
ldx [%sp + STACK_BIAS + 136], %fsr
|
||||
fabsd %f0, %f14
|
||||
fcmpd %fcc3, %f14, %f16
|
||||
fmovduge %fcc3, ZERO, %f16
|
||||
fand %f0, SIGN_BIT, SIGN_BIT
|
||||
for %f16, SIGN_BIT, %f16
|
||||
faddd %f0, %f16, %f6
|
||||
fsubd %f6, %f16, %f0
|
||||
fabsd %f0, %f0
|
||||
for %f0, SIGN_BIT, %f0
|
||||
retl
|
||||
ldx [%sp + STACK_BIAS + 144], %fsr
|
||||
END (__nearbyint_vis3)
|
12
sysdeps/sparc/sparc64/fpu/multiarch/s_nearbyint.S
Normal file
12
sysdeps/sparc/sparc64/fpu/multiarch/s_nearbyint.S
Normal file
@ -0,0 +1,12 @@
|
||||
#include <sparc-ifunc.h>
|
||||
|
||||
SPARC_ASM_VIS3_IFUNC(nearbyint)
|
||||
|
||||
weak_alias (__nearbyint, nearbyint)
|
||||
|
||||
# undef weak_alias
|
||||
# define weak_alias(a, b)
|
||||
|
||||
#define __nearbyint __nearbyint_generic
|
||||
|
||||
#include "../s_nearbyint.S"
|
60
sysdeps/sparc/sparc64/fpu/multiarch/s_nearbyintf-vis3.S
Normal file
60
sysdeps/sparc/sparc64/fpu/multiarch/s_nearbyintf-vis3.S
Normal file
@ -0,0 +1,60 @@
|
||||
/* Round float to int floating-point values without generating
|
||||
an inexact exception, sparc64 vis3 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>
|
||||
|
||||
/* We pop constants into the FPU registers using the incoming
|
||||
argument stack slots, since this avoid having to use any PIC
|
||||
references. We also thus avoid having to allocate a register
|
||||
window.
|
||||
|
||||
VIS instructions are used to facilitate the formation of
|
||||
easier constants, and the propagation of the sign bit. */
|
||||
|
||||
#define TWO_TWENTYTHREE 0x4b000000 /* 2**23 */
|
||||
|
||||
#define ZERO %f10 /* 0.0 */
|
||||
#define SIGN_BIT %f12 /* -0.0 */
|
||||
|
||||
ENTRY (__nearbyintf_vis3)
|
||||
stx %fsr, [%sp + STACK_BIAS + 144]
|
||||
sethi %hi(0xf8003e0), %o5
|
||||
sethi %hi(TWO_TWENTYTHREE), %o2
|
||||
ldx [%sp + STACK_BIAS + 144], %o4
|
||||
or %o5, %lo(0xf8003e0), %o5
|
||||
fzeros ZERO
|
||||
andn %o4, %o5, %o4
|
||||
fnegs ZERO, SIGN_BIT
|
||||
movwtos %o2, %f16
|
||||
stx %o4, [%sp + STACK_BIAS + 136]
|
||||
ldx [%sp + STACK_BIAS + 136], %fsr
|
||||
fabss %f1, %f14
|
||||
fcmps %fcc3, %f14, %f16
|
||||
fmovsuge %fcc3, ZERO, %f16
|
||||
fands %f1, SIGN_BIT, SIGN_BIT
|
||||
fors %f16, SIGN_BIT, %f16
|
||||
fadds %f1, %f16, %f5
|
||||
fsubs %f5, %f16, %f0
|
||||
fabss %f0, %f0
|
||||
fors %f0, SIGN_BIT, %f0
|
||||
retl
|
||||
ldx [%sp + STACK_BIAS + 144], %fsr
|
||||
END (__nearbyintf_vis3)
|
12
sysdeps/sparc/sparc64/fpu/multiarch/s_nearbyintf.S
Normal file
12
sysdeps/sparc/sparc64/fpu/multiarch/s_nearbyintf.S
Normal file
@ -0,0 +1,12 @@
|
||||
#include <sparc-ifunc.h>
|
||||
|
||||
SPARC_ASM_VIS3_IFUNC(nearbyintf)
|
||||
|
||||
weak_alias (__nearbyintf, nearbyintf)
|
||||
|
||||
# undef weak_alias
|
||||
# define weak_alias(a, b)
|
||||
|
||||
#define __nearbyintf __nearbyintf_generic
|
||||
|
||||
#include "../s_nearbyintf.S"
|
63
sysdeps/sparc/sparc64/fpu/s_nearbyint.S
Normal file
63
sysdeps/sparc/sparc64/fpu/s_nearbyint.S
Normal file
@ -0,0 +1,63 @@
|
||||
/* Round float to int floating-point values without generating
|
||||
an inexact exception, sparc64 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>
|
||||
|
||||
/* We pop constants into the FPU registers using the incoming
|
||||
argument stack slots, since this avoid having to use any PIC
|
||||
references. We also thus avoid having to allocate a register
|
||||
window.
|
||||
|
||||
VIS instructions are used to facilitate the formation of
|
||||
easier constants, and the propagation of the sign bit. */
|
||||
|
||||
#define TWO_FIFTYTWO 0x43300000 /* 2**52 */
|
||||
|
||||
#define ZERO %f10 /* 0.0 */
|
||||
#define SIGN_BIT %f12 /* -0.0 */
|
||||
|
||||
ENTRY (__nearbyint)
|
||||
stx %fsr, [%sp + STACK_BIAS + 144]
|
||||
sethi %hi(TWO_FIFTYTWO), %o2
|
||||
sllx %o2, 32, %o2
|
||||
ldx [%sp + STACK_BIAS + 144], %o4
|
||||
sethi %hi(0xf8003e0), %o5
|
||||
fzero ZERO
|
||||
or %o5, %lo(0xf8003e0), %o5
|
||||
fnegd ZERO, SIGN_BIT
|
||||
andn %o4, %o5, %o4
|
||||
stx %o2, [%sp + STACK_BIAS + 128]
|
||||
stx %o4, [%sp + STACK_BIAS + 136]
|
||||
ldx [%sp + STACK_BIAS + 136], %fsr
|
||||
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
|
||||
faddd %f0, %f16, %f6
|
||||
fsubd %f6, %f16, %f0
|
||||
fabsd %f0, %f0
|
||||
for %f0, SIGN_BIT, %f0
|
||||
retl
|
||||
ldx [%sp + STACK_BIAS + 144], %fsr
|
||||
END (__nearbyint)
|
||||
weak_alias (__nearbyint, nearbyint)
|
62
sysdeps/sparc/sparc64/fpu/s_nearbyintf.S
Normal file
62
sysdeps/sparc/sparc64/fpu/s_nearbyintf.S
Normal file
@ -0,0 +1,62 @@
|
||||
/* Round float to int floating-point values without generating
|
||||
an inexact exception, sparc64 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>
|
||||
|
||||
/* We pop constants into the FPU registers using the incoming
|
||||
argument stack slots, since this avoid having to use any PIC
|
||||
references. We also thus avoid having to allocate a register
|
||||
window.
|
||||
|
||||
VIS instructions are used to facilitate the formation of
|
||||
easier constants, and the propagation of the sign bit. */
|
||||
|
||||
#define TWO_TWENTYTHREE 0x4b000000 /* 2**23 */
|
||||
|
||||
#define ZERO %f10 /* 0.0 */
|
||||
#define SIGN_BIT %f12 /* -0.0 */
|
||||
|
||||
ENTRY (__nearbyintf)
|
||||
stx %fsr, [%sp + STACK_BIAS + 144]
|
||||
sethi %hi(0xf8003e0), %o5
|
||||
sethi %hi(TWO_TWENTYTHREE), %o2
|
||||
ldx [%sp + STACK_BIAS + 144], %o4
|
||||
or %o5, %lo(0xf8003e0), %o5
|
||||
fzeros ZERO
|
||||
andn %o4, %o5, %o4
|
||||
fnegs ZERO, SIGN_BIT
|
||||
st %o2, [%sp + STACK_BIAS + 128]
|
||||
stx %o4, [%sp + STACK_BIAS + 136]
|
||||
ldx [%sp + STACK_BIAS + 136], %fsr
|
||||
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
|
||||
fadds %f1, %f16, %f5
|
||||
fsubs %f5, %f16, %f0
|
||||
fabss %f0, %f0
|
||||
fors %f0, SIGN_BIT, %f0
|
||||
retl
|
||||
ldx [%sp + STACK_BIAS + 144], %fsr
|
||||
END (__nearbyintf)
|
||||
weak_alias (__nearbyintf, nearbyintf)
|
Loading…
Reference in New Issue
Block a user