[PATCH v3] math: Add math-use-builtins-fabs (BZ#29027)

[PATCH v3] math: Add math-use-builtins-fabs (BZ#29027)

[Adhemerval Zanella]

Both float, double, and _Float128 are assumed to be supported
(float and double already only uses builtins).  Only long double
is parametrized due GCC bug 29253 which prevents its usage on
powerpc.

It allows to remove i686, ia64, x86_64, powerpc, and sparc arch
specific implementation.

On ia64 it also fixes the sNAN handling:

  math/test-float64x-fabs
  math/test-ldouble-fabs

Checked on x86_64-linux-gnu, i686-linux-gnu, powerpc-linux-gnu,
powerpc64-linux-gnu, sparc64-linux-gnu, and ia64-linux-gnu.
---
 sysdeps/generic/math-use-builtins-fabs.h      |  3 +
 sysdeps/generic/math-use-builtins.h           |  1 +
 sysdeps/i386/fpu/s_fabsl.S                    | 10 ---
 sysdeps/ia64/fpu/s_fabsl.S                    | 82 -------------------
 sysdeps/ieee754/ldbl-128/s_fabsl.c            |  9 +-
 sysdeps/ieee754/ldbl-128ibm/s_fabsl.c         |  5 ++
 .../s_fabsl.S => ieee754/ldbl-96/s_fabsl.c}   | 18 ++--
 sysdeps/powerpc/fpu/math-use-builtins-fabs.h  |  8 ++
 sysdeps/powerpc/powerpc32/fpu/s_fabsl.S       | 52 ------------
 sysdeps/powerpc/powerpc64/fpu/s_fabsl.S       | 34 --------
 sysdeps/sparc/sparc32/fpu/s_fabsl.c           |  9 --
 sysdeps/sparc/sparc64/fpu/s_fabsl.c           |  7 --
 12 files changed, 30 insertions(+), 208 deletions(-)
 create mode 100644 sysdeps/generic/math-use-builtins-fabs.h
 delete mode 100644 sysdeps/i386/fpu/s_fabsl.S
 delete mode 100644 sysdeps/ia64/fpu/s_fabsl.S
 rename sysdeps/{x86_64/fpu/s_fabsl.S => ieee754/ldbl-96/s_fabsl.c} (78%)
 create mode 100644 sysdeps/powerpc/fpu/math-use-builtins-fabs.h
 delete mode 100644 sysdeps/powerpc/powerpc32/fpu/s_fabsl.S
 delete mode 100644 sysdeps/powerpc/powerpc64/fpu/s_fabsl.S
 delete mode 100644 sysdeps/sparc/sparc32/fpu/s_fabsl.c
 delete mode 100644 sysdeps/sparc/sparc64/fpu/s_fabsl.c

diff --git a/sysdeps/generic/math-use-builtins-fabs.h b/sysdeps/generic/math-use-builtins-fabs.h
new file mode 100644
index 0000000000..531066b3b4
--- /dev/null
+++ b/sysdeps/generic/math-use-builtins-fabs.h
@@ -0,0 +1,3 @@
+/* Generic implementations already assume builtin support for
+   float, double, and _Float128.  */
+#define USE_FABSL_BUILTIN 1
diff --git a/sysdeps/generic/math-use-builtins.h b/sysdeps/generic/math-use-builtins.h
index 3ec15e59d4..6bd424d900 100644
--- a/sysdeps/generic/math-use-builtins.h
+++ b/sysdeps/generic/math-use-builtins.h
@@ -36,5 +36,6 @@
 #include <math-use-builtins-fma.h>
 #include <math-use-builtins-fmax.h>
 #include <math-use-builtins-fmin.h>
+#include <math-use-builtins-fabs.h>
 
 #endif /* MATH_USE_BUILTINS_H  */
diff --git a/sysdeps/i386/fpu/s_fabsl.S b/sysdeps/i386/fpu/s_fabsl.S
deleted file mode 100644
index 6642ed4151..0000000000
--- a/sysdeps/i386/fpu/s_fabsl.S
+++ /dev/null
@@ -1,10 +0,0 @@
-#include <libm-alias-ldouble.h>
-#include <sysdep.h>
-
-	.text
-ENTRY(__fabsl)
-	fldt	4(%esp)
-	fabs
-	ret
-END(__fabsl)
-libm_alias_ldouble (__fabs, fabs)
diff --git a/sysdeps/ia64/fpu/s_fabsl.S b/sysdeps/ia64/fpu/s_fabsl.S
deleted file mode 100644
index 07838b1057..0000000000
--- a/sysdeps/ia64/fpu/s_fabsl.S
+++ /dev/null
@@ -1,82 +0,0 @@
-.file "fabsl.s"
-
-
-// Copyright (c) 2000 - 2003, Intel Corporation
-// All rights reserved.
-//
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-// * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//
-// * Redistributions in binary form must reproduce the above copyright
-// notice, this list of conditions and the following disclaimer in the
-// documentation and/or other materials provided with the distribution.
-//
-// * The name of Intel Corporation may not be used to endorse or promote
-// products derived from this software without specific prior written
-// permission.
-
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL OR ITS
-// CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
-// EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
-// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
-// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
-// OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY OR TORT (INCLUDING
-// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
-// SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Intel Corporation is the author of this code, and requests that all
-// problem reports or change requests be submitted to it directly at
-// http://www.intel.com/software/products/opensource/libraries/num.htm.
-//
-// History
-//==============================================================
-// 02/02/00 Initial version
-// 02/07/02 Added __libm_fabsl entry point to test in case compiler inlines
-// 05/20/02 Cleaned up namespace and sf0 syntax
-// 02/10/03 Reordered header: .section, .global, .proc, .align
-//
-// API
-//==============================================================
-// long double fabsl  (long double x)
-//
-// Overview of operation
-//==============================================================
-// returns absolute value of x
-
-// floating-point registers used: 1
-// f8, input
-
-.section .text
-.global __libm_fabsl#
-
-.proc __libm_fabsl#
-__libm_fabsl:
-.endp __libm_fabsl#
-
-GLOBAL_IEEE754_ENTRY(fabsl)
-
-// set invalid or denormal flags and take fault if
-// necessary
-
-{ .mfi
-      nop.m 999
-      fcmp.eq.unc.s0 p6,p7 = f8,f1
-      nop.i 999 ;;
-}
-
-{ .mfb
-      nop.m 999
-      fmerge.s       f8 = f0,f8
-      br.ret.sptk    b0 ;;
-}
-
-GLOBAL_IEEE754_END(fabsl)
-libm_alias_ldouble_other (__fabs, fabs)
diff --git a/sysdeps/ieee754/ldbl-128/s_fabsl.c b/sysdeps/ieee754/ldbl-128/s_fabsl.c
index 7e08980252..d2603418c2 100644
--- a/sysdeps/ieee754/ldbl-128/s_fabsl.c
+++ b/sysdeps/ieee754/ldbl-128/s_fabsl.c
@@ -23,12 +23,11 @@ static char rcsid[] = "$NetBSD: $";
 #include <math.h>
 #include <math_private.h>
 #include <libm-alias-ldouble.h>
+#include <math-use-builtins.h>
 
-_Float128 __fabsl(_Float128 x)
+_Float128
+__fabsl (_Float128 x)
 {
-	uint64_t hx;
-	GET_LDOUBLE_MSW64(hx,x);
-	SET_LDOUBLE_MSW64(x,hx&0x7fffffffffffffffLL);
-        return x;
+  return __builtin_fabsf128 (x);
 }
 libm_alias_ldouble (__fabs, fabs)
diff --git a/sysdeps/ieee754/ldbl-128ibm/s_fabsl.c b/sysdeps/ieee754/ldbl-128ibm/s_fabsl.c
index cba7de6660..7f26315e84 100644
--- a/sysdeps/ieee754/ldbl-128ibm/s_fabsl.c
+++ b/sysdeps/ieee754/ldbl-128ibm/s_fabsl.c
@@ -24,9 +24,13 @@ static char rcsid[] = "$NetBSD: $";
 #include <math.h>
 #include <math_private.h>
 #include <math_ldbl_opt.h>
+#include <math-use-builtins.h>
 
 long double __fabsl(long double x)
 {
+#if USE_FABSL_BUILTIN
+	return __builtin_fabsl (x);
+#else
 	uint64_t hx, lx;
 	double xhi, xlo;
 
@@ -39,5 +43,6 @@ long double __fabsl(long double x)
 	INSERT_WORDS64 (xlo, lx);
 	x = ldbl_pack (xhi, xlo);
 	return x;
+#endif
 }
 long_double_symbol (libm, __fabsl, fabsl);
diff --git a/sysdeps/x86_64/fpu/s_fabsl.S b/sysdeps/ieee754/ldbl-96/s_fabsl.c
similarity index 78%
rename from sysdeps/x86_64/fpu/s_fabsl.S
rename to sysdeps/ieee754/ldbl-96/s_fabsl.c
index cfcf01e027..0eceb0c997 100644
--- a/sysdeps/x86_64/fpu/s_fabsl.S
+++ b/sysdeps/ieee754/ldbl-96/s_fabsl.c
@@ -1,5 +1,5 @@
-/* Absolute value of floating point number.
-   Copyright (C) 2002-2022 Free Software Foundation, Inc.
+/* Absolute value of floating-point number.
+   Copyright (C) 2022 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
@@ -16,13 +16,13 @@
    License along with the GNU C Library; if not, see
    <https://www.gnu.org/licenses/>.  */
 
-#include <sysdep.h>
+#include <math.h>
 #include <libm-alias-ldouble.h>
+#include <math-use-builtins.h>
 
-       .text
-ENTRY(__fabsl)
-       fldt    8(%rsp)
-       fabs
-       ret
-END(__fabsl)
+long double
+__fabsl (long double x)
+{
+  return __builtin_fabsl (x);
+}
 libm_alias_ldouble (__fabs, fabs)
diff --git a/sysdeps/powerpc/fpu/math-use-builtins-fabs.h b/sysdeps/powerpc/fpu/math-use-builtins-fabs.h
new file mode 100644
index 0000000000..8976d0b956
--- /dev/null
+++ b/sysdeps/powerpc/fpu/math-use-builtins-fabs.h
@@ -0,0 +1,8 @@
+/* Generic implementations already assume builtin support for
+   float, double, and _Float128 .  */
+
+/* We can't enable it for long double due GCC Bugzilla Bug 29253
+   "expand_abs wrong default code for floating point" where it can
+   cause spurious "invalid" exceptions from comparisons with NaN in
+   the code sequence generated.  */
+#define USE_FABSL_BUILTIN 0
diff --git a/sysdeps/powerpc/powerpc32/fpu/s_fabsl.S b/sysdeps/powerpc/powerpc32/fpu/s_fabsl.S
deleted file mode 100644
index 8bdcdc8cfb..0000000000
--- a/sysdeps/powerpc/powerpc32/fpu/s_fabsl.S
+++ /dev/null
@@ -1,52 +0,0 @@
-/* Copy a sign bit between floating-point values.
-   IBM extended format long double version.
-   Copyright (C) 2004-2022 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <https://www.gnu.org/licenses/>.  */
-
-#include <sysdep.h>
-#include <math_ldbl_opt.h>
-
-ENTRY(__fabsl)
-/* long double [f1,f2] fabs (long double [f1,f2] x);
-   fabs(x,y) returns a value with the magnitude of x and
-   with the sign bit of y.  */
-#ifdef _ARCH_PPCGR
-	/* fsel available.  */
-	fmr	fp0,fp1
-#else
-	/* Use integer operations to test sign of high part to avoid
-	   exceptions on sNaNs.  */
-	stwu	r1,-16(r1)
-	cfi_adjust_cfa_offset (16)
-	stfd	fp1,8(r1)
-#endif
-	fabs	fp1,fp1
-#ifdef _ARCH_PPCGR
-	fneg	fp3,fp2
-	fsel	fp2,fp0,fp2,fp3
-#else
-	lwz	r3,8+HIWORD(r1)
-	cmpwi	cr6,r3,0
-	addi	r1,r1,16
-	cfi_adjust_cfa_offset (-16)
-	bgelr	cr6
-	fneg	fp2,fp2
-#endif
-	blr
-END (__fabsl)
-
-long_double_symbol (libm, __fabsl, fabsl)
diff --git a/sysdeps/powerpc/powerpc64/fpu/s_fabsl.S b/sysdeps/powerpc/powerpc64/fpu/s_fabsl.S
deleted file mode 100644
index ec796a1cd3..0000000000
--- a/sysdeps/powerpc/powerpc64/fpu/s_fabsl.S
+++ /dev/null
@@ -1,34 +0,0 @@
-/* Copy a sign bit between floating-point values.
-   IBM extended format long double version.
-   Copyright (C) 2004-2022 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <https://www.gnu.org/licenses/>.  */
-
-#include <sysdep.h>
-#include <math_ldbl_opt.h>
-
-ENTRY_TOCLESS (__fabsl)
-/* long double [f1,f2] fabs (long double [f1,f2] x);
-   fabs(x,y) returns a value with the magnitude of x and
-   with the sign bit of y.  */
-	fmr	fp0,fp1
-	fabs	fp1,fp1
-	fneg	fp3,fp2
-	fsel	fp2,fp0,fp2,fp3
-	blr
-END (__fabsl)
-
-long_double_symbol (libm, __fabsl, fabsl)
diff --git a/sysdeps/sparc/sparc32/fpu/s_fabsl.c b/sysdeps/sparc/sparc32/fpu/s_fabsl.c
deleted file mode 100644
index 411a2e1511..0000000000
--- a/sysdeps/sparc/sparc32/fpu/s_fabsl.c
+++ /dev/null
@@ -1,9 +0,0 @@
-#include <math.h>
-#include <math_ldbl_opt.h>
-#include <libm-alias-ldouble.h>
-
-long double __fabsl (long double x)
-{
-  return __builtin_fabsl (x);
-}
-libm_alias_ldouble (__fabs, fabs)
diff --git a/sysdeps/sparc/sparc64/fpu/s_fabsl.c b/sysdeps/sparc/sparc64/fpu/s_fabsl.c
deleted file mode 100644
index 72e1c51340..0000000000
--- a/sysdeps/sparc/sparc64/fpu/s_fabsl.c
+++ /dev/null
@@ -1,7 +0,0 @@
-#include <libm-alias-ldouble.h>
-
-long double __fabsl (long double x)
-{
-  return __builtin_fabsl (x);
-}
-libm_alias_ldouble (__fabs, fabs)
-- 
2.34.1

Re: [PATCH v3] math: Add math-use-builtins-fabs (BZ#29027)

[Andreas Schwab]

On Mai 05 2022, Adhemerval Zanella via Libc-alpha wrote:

> diff --git a/sysdeps/ieee754/ldbl-128/s_fabsl.c b/sysdeps/ieee754/ldbl-128/s_fabsl.c
> index 7e08980252..d2603418c2 100644
> --- a/sysdeps/ieee754/ldbl-128/s_fabsl.c
> +++ b/sysdeps/ieee754/ldbl-128/s_fabsl.c
> @@ -23,12 +23,11 @@ static char rcsid[] = "$NetBSD: $";
>  #include <math.h>
>  #include <math_private.h>
>  #include <libm-alias-ldouble.h>
> +#include <math-use-builtins.h>

I don't think you need that.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 7578 EB47 D4E5 4D69 2510  2552 DF73 E780 A9DA AEC1
"And now for something completely different."

Re: [PATCH v3] math: Add math-use-builtins-fabs (BZ#29027)

[Adhemerval Zanella]

On 10/05/2022 06:04, Andreas Schwab wrote:
> On Mai 05 2022, Adhemerval Zanella via Libc-alpha wrote:
> 
>> diff --git a/sysdeps/ieee754/ldbl-128/s_fabsl.c b/sysdeps/ieee754/ldbl-128/s_fabsl.c
>> index 7e08980252..d2603418c2 100644
>> --- a/sysdeps/ieee754/ldbl-128/s_fabsl.c
>> +++ b/sysdeps/ieee754/ldbl-128/s_fabsl.c
>> @@ -23,12 +23,11 @@ static char rcsid[] = "$NetBSD: $";
>>  #include <math.h>
>>  #include <math_private.h>
>>  #include <libm-alias-ldouble.h>
>> +#include <math-use-builtins.h>
> 
> I don't think you need that.
> 

Ack, I will remove it.