Remove powerpc32 sqrt wrappers.

This patch removes the powerpc32-specific wrappers for sqrt and sqrtf.

These wrappers, by adding architecture-specific uses of _LIB_VERSION
and __kernel_standard, unnecessarily complicate cleanups of libm error
handling.  They also do not serve a useful optimization purpose.  GCC
knows about sqrt as a built-in function, and can generate direct calls
to a hardware square root instruction, either on its own, in the
-fno-math-errno case, or together with an inline check for the
argument being negative and a call to the out-of-line sqrt function
for error handling only in that case (and has been able to do so for a
long time).  Thus in practice the wrapper will only be called only in
the case of negative arguments, which is not a case it is useful to
optimize for.

Tested with build-many-glibcs.py for powerpc-linux-gnu-power4.

	* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/w_sqrt_compat-power5.S:
	Remove file.
	* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/w_sqrt_compat-ppc32.S:
	Likewise.
	* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/w_sqrt_compat.c:
	Likewise.
	* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/w_sqrtf_compat-power5.S:
	Likewise.
	* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/w_sqrtf_compat-ppc32.S:
	Likewise.
	* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/w_sqrtf_compat.c:
	Likewise.
	* sysdeps/powerpc/powerpc32/power4/fpu/w_sqrt_compat.S: Likewise.
	* sysdeps/powerpc/powerpc32/power4/fpu/w_sqrtf_compat.S: Likewise.
	* sysdeps/powerpc/powerpc32/power5/fpu/w_sqrt_compat.S: Likewise.
	* sysdeps/powerpc/powerpc32/power5/fpu/w_sqrtf_compat.S: Likewise.
	* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/Makefile
	(libm-sysdep-routines): Remove w_sqrt_compat-power5,
	w_sqrt_compat-ppc32, w_sqrtf_compat-power5 and
	w_sqrtf_compat-ppc32.
This commit is contained in:
Joseph Myers 2017-08-25 11:33:04 +00:00
parent 3d7b66f66c
commit bba5c93094
12 changed files with 23 additions and 604 deletions

View File

@ -1,3 +1,26 @@
2017-08-25 Joseph Myers <joseph@codesourcery.com>
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/w_sqrt_compat-power5.S:
Remove file.
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/w_sqrt_compat-ppc32.S:
Likewise.
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/w_sqrt_compat.c:
Likewise.
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/w_sqrtf_compat-power5.S:
Likewise.
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/w_sqrtf_compat-ppc32.S:
Likewise.
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/w_sqrtf_compat.c:
Likewise.
* sysdeps/powerpc/powerpc32/power4/fpu/w_sqrt_compat.S: Likewise.
* sysdeps/powerpc/powerpc32/power4/fpu/w_sqrtf_compat.S: Likewise.
* sysdeps/powerpc/powerpc32/power5/fpu/w_sqrt_compat.S: Likewise.
* sysdeps/powerpc/powerpc32/power5/fpu/w_sqrtf_compat.S: Likewise.
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/Makefile
(libm-sysdep-routines): Remove w_sqrt_compat-power5,
w_sqrt_compat-ppc32, w_sqrtf_compat-power5 and
w_sqrtf_compat-ppc32.
2017-08-24 Gabriel F. T. Gomes <gftg@linux.vnet.ibm.com> 2017-08-24 Gabriel F. T. Gomes <gftg@linux.vnet.ibm.com>
* math/math.h [defined __cplusplus] (issignaling): In the long * math/math.h [defined __cplusplus] (issignaling): In the long

View File

@ -9,8 +9,6 @@ sysdep_routines += s_isnan-power7 s_isnan-power6 s_isnan-power5 s_isnan-ppc32 \
libm-sysdep_routines += s_llrintf-power6 s_llrintf-ppc32 s_llrint-power6 \ libm-sysdep_routines += s_llrintf-power6 s_llrintf-ppc32 s_llrint-power6 \
s_llrint-ppc32 s_llround-power6 s_llround-power5+ \ s_llrint-ppc32 s_llround-power6 s_llround-power5+ \
s_llround-ppc32 s_isnan-power7 \ s_llround-ppc32 s_isnan-power7 \
w_sqrt_compat-power5 w_sqrt_compat-ppc32 \
w_sqrtf_compat-power5 w_sqrtf_compat-ppc32 \
s_isnan-power6 s_isnan-power5 s_isnan-ppc32 \ s_isnan-power6 s_isnan-power5 s_isnan-ppc32 \
s_isnanf-power6 s_isnanf-power5 s_isinf-power7 \ s_isnanf-power6 s_isnanf-power5 s_isinf-power7 \
s_isinf-ppc32 s_isinff-ppc32 s_finite-power7 \ s_isinf-ppc32 s_isinff-ppc32 s_finite-power7 \

View File

@ -1,31 +0,0 @@
/* sqrt function. PowerPC32/POWER5 version.
Copyright (C) 2013-2017 Free Software Foundation, Inc.
This file is part of the GNU C Library.
The GNU C Library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
The GNU C Library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with the GNU C Library; if not, see
<http://www.gnu.org/licenses/>. */
#include <sysdep.h>
#include <math_ldbl_opt.h>
#undef weak_alias
#define weak_alias(a,b)
#undef strong_alias
#define strong_alias(a,b)
#undef compat_symbol
#define compat_symbol(a,b,c,d)
#define __sqrt __sqrt_power5
#include <sysdeps/powerpc/powerpc32/power5/fpu/w_sqrt_compat.S>

View File

@ -1,31 +0,0 @@
/* sqrt function. PowerPC32 default version.
Copyright (C) 2013-2017 Free Software Foundation, Inc.
This file is part of the GNU C Library.
The GNU C Library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
The GNU C Library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with the GNU C Library; if not, see
<http://www.gnu.org/licenses/>. */
#include <sysdep.h>
#include <math_ldbl_opt.h>
#undef weak_alias
#define weak_alias(a,b)
#undef strong_alias
#define strong_alias(a,b)
#undef compat_symbol
#define compat_symbol(a,b,c,d)
#define __sqrt __sqrt_ppc32
#include <sysdeps/powerpc/powerpc32/power4/fpu/w_sqrt_compat.S>

View File

@ -1,40 +0,0 @@
/* Multiple versions of sqrt.
Copyright (C) 2013-2017 Free Software Foundation, Inc.
This file is part of the GNU C Library.
The GNU C Library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
The GNU C Library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with the GNU C Library; if not, see
<http://www.gnu.org/licenses/>. */
#include <math.h>
#include <math_ldbl_opt.h>
#include <shlib-compat.h>
#include "init-arch.h"
extern __typeof (__sqrt) __sqrt_ppc32 attribute_hidden;
extern __typeof (__sqrt) __sqrt_power5 attribute_hidden;
libc_ifunc (__sqrt,
(hwcap & PPC_FEATURE_POWER5)
? __sqrt_power5
: __sqrt_ppc32);
weak_alias (__sqrt, sqrt)
#ifdef NO_LONG_DOUBLE
strong_alias (__sqrt, __sqrtl)
weak_alias (__sqrt, sqrtl)
#endif
#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_0)
compat_symbol (libm, __sqrt, sqrtl, GLIBC_2_0);
#endif

View File

@ -1,26 +0,0 @@
/* sqrtf function. PowerPC32/POWER5 version.
Copyright (C) 2013-2017 Free Software Foundation, Inc.
This file is part of the GNU C Library.
The GNU C Library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
The GNU C Library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with the GNU C Library; if not, see
<http://www.gnu.org/licenses/>. */
#include <sysdep.h>
#undef weak_alias
#define weak_alias(a,b)
#define __sqrtf __sqrtf_power5
#include <sysdeps/powerpc/powerpc32/power5/fpu/w_sqrtf_compat.S>

View File

@ -1,26 +0,0 @@
/* sqrtf function. PowerPC32 default version.
Copyright (C) 2013-2017 Free Software Foundation, Inc.
This file is part of the GNU C Library.
The GNU C Library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
The GNU C Library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with the GNU C Library; if not, see
<http://www.gnu.org/licenses/>. */
#include <sysdep.h>
#undef weak_alias
#define weak_alias(a,b)
#define __sqrtf __sqrtf_ppc32
#include <sysdeps/powerpc/powerpc32/power4/fpu/w_sqrtf_compat.S>

View File

@ -1,32 +0,0 @@
/* Multiple versions of sqrtf.
Copyright (C) 2013-2017 Free Software Foundation, Inc.
This file is part of the GNU C Library.
The GNU C Library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
The GNU C Library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with the GNU C Library; if not, see
<http://www.gnu.org/licenses/>. */
#include <math.h>
#include <math_ldbl_opt.h>
#include <shlib-compat.h>
#include "init-arch.h"
extern __typeof (__sqrtf) __sqrtf_ppc32 attribute_hidden;
extern __typeof (__sqrtf) __sqrtf_power5 attribute_hidden;
libc_ifunc (__sqrtf,
(hwcap & PPC_FEATURE_POWER5)
? __sqrtf_power5
: __sqrtf_ppc32);
weak_alias (__sqrtf, sqrtf)

View File

@ -1,109 +0,0 @@
/* sqrt function. PowerPC32 version.
Copyright (C) 2007-2017 Free Software Foundation, Inc.
This file is part of the GNU C Library.
The GNU C Library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
The GNU C Library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with the GNU C Library; if not, see
<http://www.gnu.org/licenses/>. */
#include <sysdep.h>
#include <math_ldbl_opt.h>
#include <math-svid-compat.h>
/* double [fp1] sqrt (double x [fp1])
Power4 (ISA V2.0) and above implement sqrt in hardware (not optional).
The fsqrt instruction generates the correct value for all inputs and
sets the appropriate floating point exceptions. Extended checking is
only needed to set errno (via __kernel_standard) if the input value
is negative.
The fsqrt will set FPCC and FU (Floating Point Unordered or NaN
to indicated that the input value was negative or NaN. Use Move to
Condition Register from FPSCR to copy the FPCC field to cr1. The
branch on summary overflow transfers control to w_sqrt to process
any error conditions. Otherwise we can return the result directly.
This part of the function is a leaf routine, so no need to stack a
frame or execute prologue/epilogue code. This means it is safe to
transfer directly to w_sqrt as long as the input value (f1) is
preserved. Putting the sqrt result into f2 (double parameter 2)
allows passing both the input value and sqrt result into the extended
wrapper so there is no need to recompute.
This tactic avoids the overhead of stacking a frame for the normal
(non-error) case. Until gcc supports prologue shrink-wrapping
this is the best we can do. */
.section ".text"
.machine power4
EALIGN (__sqrt, 5, 0)
fsqrt fp2,fp1
mcrfs cr1,4
bso- cr1,.Lw_sqrt
fmr fp1,fp2
blr
.align 4
.Lw_sqrt:
mflr r0
stwu r1,-16(r1)
cfi_adjust_cfa_offset(16)
fmr fp12,fp2
stw r0,20(r1)
stw r30,8(r1)
cfi_offset(lr,20-16)
cfi_offset(r30,8-16)
#ifdef SHARED
SETUP_GOT_ACCESS(r30,got_label)
addis r30,r30,_GLOBAL_OFFSET_TABLE_-got_label@ha
addi r30,r30,_GLOBAL_OFFSET_TABLE_-got_label@l
lwz r9,_LIB_VERSION@got(30)
lwz r0,0(r9)
#else
lis r9,_LIB_VERSION@ha
lwz r0,_LIB_VERSION@l(r9)
#endif
/* if (_LIB_VERSION == _IEEE_) return z; */
cmpwi cr7,r0,-1
beq- cr7,.L4
/* if (x != x) return z; !isnan*/
fcmpu cr7,fp1,fp1
bne- cr7,.L4
/* if (x < 0.0)
return __kernel_standard (x, x, 26) */
fmr fp2,fp1
fabs fp0,fp1
li r3,26
fcmpu cr7,fp1,fp0
bne- cr7,.L11
.L4:
lwz r0,20(r1)
fmr fp1,fp12
lwz r30,8(r1)
addi r1,r1,16
mtlr r0
blr
.L11:
bl __kernel_standard@plt
fmr fp12,fp1
b .L4
END (__sqrt)
weak_alias (__sqrt, sqrt)
#ifdef NO_LONG_DOUBLE
weak_alias (__sqrt, sqrtl)
strong_alias (__sqrt, __sqrtl)
#endif
#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_0)
compat_symbol (libm, __sqrt, sqrtl, GLIBC_2_0)
#endif

View File

@ -1,101 +0,0 @@
/* sqrtf function. PowerPC32 version.
Copyright (C) 2007-2017 Free Software Foundation, Inc.
This file is part of the GNU C Library.
The GNU C Library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
The GNU C Library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with the GNU C Library; if not, see
<http://www.gnu.org/licenses/>. */
#include <sysdep.h>
#include <math_ldbl_opt.h>
#include <math-svid-compat.h>
/* float [fp1] sqrts (float x [fp1])
Power4 (ISA V2.0) and above implement sqrt in hardware (not optional).
The fsqrts instruction generates the correct value for all inputs and
sets the appropriate floating point exceptions. Extended checking is
only needed to set errno (via __kernel_standard) if the input value
is negative.
The fsqrts will set FPCC and FU (Floating Point Unordered or NaN
to indicated that the input value was negative or NaN. Use Move to
Condition Register from FPSCR to copy the FPCC field to cr1. The
branch on summary overflow transfers control to w_sqrt to process
any error conditions. Otherwise we can return the result directly.
This part of the function is a leaf routine, so no need to stack a
frame or execute prologue/epilogue code. This means it is safe to
transfer directly to w_sqrt as long as the input value (f1) is
preserved. Putting the sqrt result into f2 (float parameter 2)
allows passing both the input value and sqrt result into the extended
wrapper so there is no need to recompute.
This tactic avoids the overhead of stacking a frame for the normal
(non-error) case. Until gcc supports prologue shrink-wrapping
this is the best we can do. */
.section ".text"
.machine power4
EALIGN (__sqrtf, 5, 0)
fsqrts fp2,fp1
mcrfs cr1,4
bso- cr1,.Lw_sqrtf
fmr fp1,fp2
blr
.align 4
.Lw_sqrtf:
mflr r0
stwu r1,-16(r1)
cfi_adjust_cfa_offset(16)
fmr fp12,fp2
stw r0,20(r1)
stw r30,8(r1)
cfi_offset(lr,20-16)
cfi_offset(r30,8-16)
#ifdef SHARED
SETUP_GOT_ACCESS(r30,got_label)
addis r30,r30,_GLOBAL_OFFSET_TABLE_-got_label@ha
addi r30,r30,_GLOBAL_OFFSET_TABLE_-got_label@l
lwz r9,_LIB_VERSION@got(30)
lwz r0,0(r9)
#else
lis r9,_LIB_VERSION@ha
lwz r0,_LIB_VERSION@l(r9)
#endif
/* if (_LIB_VERSION == _IEEE_) return z; */
cmpwi cr7,r0,-1
beq- cr7,.L4
/* if (x != x, 0) return z; !isnan */
fcmpu cr7,fp1,fp1
bne- cr7,.L4
/* if (x < 0.0)
return __kernel_standard (x, x, 126) */
fmr fp2,fp1
fabs fp0,fp1
li r3,126
fcmpu cr7,1,0
bne- cr7,.L11
.L4:
lwz r0,20(r1)
fmr fp1,fp12
lwz r30,8(r1)
addi r1,r1,16
mtlr r0
blr
.L11:
bl __kernel_standard@plt
fmr fp12,fp1
b .L4
END (__sqrtf)
weak_alias (__sqrtf, sqrtf)

View File

@ -1,107 +0,0 @@
/* sqrt function. PowerPC32 version.
Copyright (C) 2007-2017 Free Software Foundation, Inc.
This file is part of the GNU C Library.
The GNU C Library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
The GNU C Library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with the GNU C Library; if not, see
<http://www.gnu.org/licenses/>. */
#include <sysdep.h>
#include <math_ldbl_opt.h>
#include <math-svid-compat.h>
/* double [fp1] sqrt (double x [fp1])
Power4 (ISA V2.0) and above implement sqrt in hardware (not optional).
The fsqrt instruction generates the correct value for all inputs and
sets the appropriate floating point exceptions. Extended checking is
only needed to set errno (via __kernel_standard) if the input value
is negative.
So compare the input value against the absolute value of itself.
This will compare equal unless the value is negative (EDOM) or a NAN,
in which case we branch to the extend wrapper. If equal we can return
the result directly.
This part of the function looks like a leaf routine, so no need to
stack a frame or execute prologue/epilogue code. It is safe to
branch directly to w_sqrt as long as the input value (f1) is
preserved. Putting the sqrt result into f2 (float parameter 2)
allows passing both the input value and sqrt result into the extended
wrapper so there is no need to recompute.
This tactic avoids the overhead of stacking a frame for the normal
(non-error) case. Until gcc supports prologue shrink-wrapping
this is the best we can do. */
.section ".text"
.machine power4
EALIGN (__sqrt, 5, 0)
fabs fp0,fp1
fsqrt fp2,fp1
fcmpu cr1,fp0,fp1
bne- cr1,.Lw_sqrt
fmr fp1,fp2
blr
.align 4
.Lw_sqrt:
mflr r0
stwu r1,-16(r1)
cfi_adjust_cfa_offset(16)
fmr fp12,fp2
stw r0,20(r1)
stw r30,8(r1)
cfi_offset(lr,20-16)
cfi_offset(r30,8-16)
#ifdef SHARED
SETUP_GOT_ACCESS(r30,got_label)
addis r30,r30,_GLOBAL_OFFSET_TABLE_-got_label@ha
addi r30,r30,_GLOBAL_OFFSET_TABLE_-got_label@l
lwz r9,_LIB_VERSION@got(30)
lwz r0,0(r9)
#else
lis r9,_LIB_VERSION@ha
lwz r0,_LIB_VERSION@l(r9)
#endif
/* if (_LIB_VERSION == _IEEE_) return z; */
cmpwi cr7,r0,-1
beq- cr7,.L4
/* if (x != x) return z; !isnan*/
fcmpu cr7,fp1,fp1
bne- cr7,.L4
/* if (x < 0.0)
return __kernel_standard (x, x, 26) */
fmr fp2,fp1
li r3,26
bne- cr1,.L11
.L4:
lwz r0,20(r1)
fmr fp1,fp12
lwz r30,8(r1)
addi r1,r1,16
mtlr r0
blr
.L11:
bl __kernel_standard@plt
fmr fp12,fp1
b .L4
END (__sqrt)
weak_alias (__sqrt, sqrt)
#ifdef NO_LONG_DOUBLE
weak_alias (__sqrt, sqrtl)
strong_alias (__sqrt, __sqrtl)
#endif
#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_0)
compat_symbol (libm, __sqrt, sqrtl, GLIBC_2_0)
#endif

View File

@ -1,99 +0,0 @@
/* sqrtf function. PowerPC32 version.
Copyright (C) 2007-2017 Free Software Foundation, Inc.
This file is part of the GNU C Library.
The GNU C Library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
The GNU C Library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with the GNU C Library; if not, see
<http://www.gnu.org/licenses/>. */
#include <sysdep.h>
#include <math_ldbl_opt.h>
#include <math-svid-compat.h>
/* float [fp1] sqrts (float x [fp1])
Power4 (ISA V2.0) and above implement sqrt in hardware (not optional).
The fsqrts instruction generates the correct value for all inputs and
sets the appropriate floating point exceptions. Extended checking is
only needed to set errno (via __kernel_standard) if the input value
is negative.
So compare the input value against the absolute value of itself.
This will compare equal unless the value is negative (EDOM) or a NAN,
in which case we branch to the extend wrapper. If equal we can return
the result directly.
This part of the function looks like a leaf routine, so no need to
stack a frame or execute prologue/epilogue code. It is safe to
branch directly to w_sqrt as long as the input value (f1) is
preserved. Putting the sqrt result into f2 (float parameter 2)
allows passing both the input value and sqrt result into the extended
wrapper so there is no need to recompute.
This tactic avoids the overhead of stacking a frame for the normal
(non-error) case. Until gcc supports prologue shrink-wrapping
this is the best we can do. */
.section ".text"
.machine power4
EALIGN (__sqrtf, 5, 0)
fabs fp0,fp1
fsqrts fp2,fp1
fcmpu cr1,fp0,fp1
bne- cr1,.Lw_sqrtf
fmr fp1,fp2
blr
.align 4
.Lw_sqrtf:
mflr r0
stwu r1,-16(r1)
cfi_adjust_cfa_offset(16)
fmr fp12,fp2
stw r0,20(r1)
stw r30,8(r1)
cfi_offset(lr,20-16)
cfi_offset(r30,8-16)
#ifdef SHARED
SETUP_GOT_ACCESS(r30,got_label)
addis r30,r30,_GLOBAL_OFFSET_TABLE_-got_label@ha
addi r30,r30,_GLOBAL_OFFSET_TABLE_-got_label@l
lwz r9,_LIB_VERSION@got(30)
lwz r0,0(r9)
#else
lis r9,_LIB_VERSION@ha
lwz r0,_LIB_VERSION@l(r9)
#endif
/* if (_LIB_VERSION == _IEEE_) return z; */
cmpwi cr7,r0,-1
beq- cr7,.L4
/* if (x != x, 0) return z; !isnan */
fcmpu cr7,fp1,fp1
bne- cr7,.L4
/* if (x < 0.0)
return __kernel_standard (x, x, 126) */
fmr fp2,fp1
li r3,126
bne- cr1,.L11
.L4:
lwz r0,20(r1)
fmr fp1,fp12
lwz r30,8(r1)
addi r1,r1,16
mtlr r0
blr
.L11:
bl __kernel_standard@plt
fmr fp12,fp1
b .L4
END (__sqrtf)
weak_alias (__sqrtf, sqrtf)