public inbox for libc-hacker@sourceware.org
 help / color / mirror / Atom feed
* [alpha] math routine improvements
@ 2007-03-14 17:48 Richard Henderson
  2007-03-18 19:02 ` Jakub Jelinek
  0 siblings, 1 reply; 2+ messages in thread
From: Richard Henderson @ 2007-03-14 17:48 UTC (permalink / raw)
  To: libc-hacker

Branchless versions of some existing rounding routines.
Implement some easy functions that were missing.
Remove inline versions of fdim, because they handle inf incorrectly.

Committed.


r~


	* sysdeps/alpha/fpu/s_ceil.c: Rewrite without branches.
	* sysdeps/alpha/fpu/s_ceilf.c: Likewise.
	* sysdeps/alpha/fpu/s_floor.c: Likewise.
	* sysdeps/alpha/fpu/s_floorf.c: Likewise.
	* sysdeps/alpha/fpu/s_rint.c: Likewise.
	* sysdeps/alpha/fpu/s_rintf.c: Likewise.

	* sysdeps/alpha/fpu/s_fmax.S: New file.
	* sysdeps/alpha/fpu/s_fmaxf.S: New file.
	* sysdeps/alpha/fpu/s_fmin.S: New file.
	* sysdeps/alpha/fpu/s_fminf.S: New file.
	* sysdeps/alpha/fpu/s_isnan.c: New file.
	* sysdeps/alpha/fpu/s_isnanf.c: New file.
	* sysdeps/alpha/fpu/s_llrint.c: New file.
	* sysdeps/alpha/fpu/s_llrintf.c: New file.
	* sysdeps/alpha/fpu/s_lrint.c: New file.
	* sysdeps/alpha/fpu/s_lrintf.c: New file.
	* sysdeps/alpha/fpu/s_nearbyint.c: New file.
	* sysdeps/alpha/fpu/s_nearbyintf.c: New file.

	* sysdeps/alpha/fpu/bits/mathinline.h (__floorf, __floor): Remove.
	(__fdimf, fdimf, __fdim, fdim): Remove.
	(__signbitf, __signbit, __signbitl): Use gcc builtin if available.
	(__isnanf, __isnan, __isnanl): New.

Index: sysdeps/alpha/fpu/s_ceil.c
===================================================================
RCS file: /cvs/glibc/libc/sysdeps/alpha/fpu/s_ceil.c,v
retrieving revision 1.4
diff -u -p -r1.4 s_ceil.c
--- sysdeps/alpha/fpu/s_ceil.c	1 Feb 2006 03:13:45 -0000	1.4
+++ sysdeps/alpha/fpu/s_ceil.c	14 Mar 2007 17:37:15 -0000
@@ -1,4 +1,4 @@
-/* Copyright (C) 1998, 2000, 2006 Free Software Foundation, Inc.
+/* Copyright (C) 1998, 2000, 2006, 2007 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Richard Henderson.
 
@@ -27,25 +27,20 @@
 double
 __ceil (double x)
 {
-  if (isless (fabs (x), 9007199254740992.0))	/* 1 << DBL_MANT_DIG */
-    {
-      double tmp1, new_x;
-
-      new_x = -x;
-      __asm (
+  double two52 = copysign (0x1.0p52, x);
+  double r, tmp;
+  
+  __asm (
 #ifdef _IEEE_FP_INEXACT
-	     "cvttq/svim %2,%1\n\t"
+	 "addt/suim %2, %3, %1\n\tsubt/suim %1, %3, %0"
 #else
-	     "cvttq/svm %2,%1\n\t"
+	 "addt/sum %2, %3, %1\n\tsubt/sum %1, %3, %0"
 #endif
-	     "cvtqt/m %1,%0\n\t"
-	     : "=f"(new_x), "=&f"(tmp1)
-	     : "f"(new_x));
-
-      /* Fix up the negation we did above, as well as handling -0 properly. */
-      x = copysign(new_x, x);
-    }
-  return x;
+	 : "=&f"(r), "=&f"(tmp)
+	 : "f"(-x), "f"(-two52));
+
+  /* Fix up the negation we did above, as well as handling -0 properly. */
+  return copysign (r, x);
 }
 
 weak_alias (__ceil, ceil)
Index: sysdeps/alpha/fpu/s_ceilf.c
===================================================================
RCS file: /cvs/glibc/libc/sysdeps/alpha/fpu/s_ceilf.c,v
retrieving revision 1.3
diff -u -p -r1.3 s_ceilf.c
--- sysdeps/alpha/fpu/s_ceilf.c	6 Jul 2001 04:55:47 -0000	1.3
+++ sysdeps/alpha/fpu/s_ceilf.c	14 Mar 2007 17:37:15 -0000
@@ -1,4 +1,4 @@
-/* Copyright (C) 1998, 2000 Free Software Foundation, Inc.
+/* Copyright (C) 1998, 2000, 2007 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Richard Henderson.
 
@@ -26,30 +26,20 @@
 float
 __ceilf (float x)
 {
-  if (isless (fabsf (x), 16777216.0f))	/* 1 << FLT_MANT_DIG */
-    {
-      /* Note that Alpha S_Floating is stored in registers in a
-	 restricted T_Floating format, so we don't even need to
-	 convert back to S_Floating in the end.  The initial
-	 conversion to T_Floating is needed to handle denormals.  */
-
-      float tmp1, tmp2, new_x;
-
-      new_x = -x;
-      __asm ("cvtst/s %3,%2\n\t"
+  float two23 = copysignf (0x1.0p23, x);
+  float r, tmp;
+  
+  __asm (
 #ifdef _IEEE_FP_INEXACT
-	     "cvttq/svim %2,%1\n\t"
+	 "adds/suim %2, %3, %1\n\tsubs/suim %1, %3, %0"
 #else
-	     "cvttq/svm %2,%1\n\t"
+	 "adds/sum %2, %3, %1\n\tsubs/sum %1, %3, %0"
 #endif
-	     "cvtqt/m %1,%0\n\t"
-	     : "=f"(new_x), "=&f"(tmp1), "=&f"(tmp2)
-	     : "f"(new_x));
-
-      /* Fix up the negation we did above, as well as handling -0 properly. */
-      x = copysignf(new_x, x);
-    }
-  return x;
+	 : "=&f"(r), "=&f"(tmp)
+	 : "f"(-x), "f"(-two23));
+
+  /* Fix up the negation we did above, as well as handling -0 properly. */
+  return copysignf (r, x);
 }
 
 weak_alias (__ceilf, ceilf)
Index: sysdeps/alpha/fpu/s_floor.c
===================================================================
RCS file: /cvs/glibc/libc/sysdeps/alpha/fpu/s_floor.c,v
retrieving revision 1.6
diff -u -p -r1.6 s_floor.c
--- sysdeps/alpha/fpu/s_floor.c	1 Feb 2006 03:13:45 -0000	1.6
+++ sysdeps/alpha/fpu/s_floor.c	14 Mar 2007 17:37:15 -0000
@@ -1,4 +1,4 @@
-/* Copyright (C) 1998, 1999, 2000, 2006 Free Software Foundation, Inc.
+/* Copyright (C) 1998, 1999, 2000, 2006, 2007 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Richard Henderson.
 
@@ -28,25 +28,21 @@
 double
 __floor (double x)
 {
-  if (isless (fabs (x), 9007199254740992.0))	/* 1 << DBL_MANT_DIG */
-    {
-      double tmp1, new_x;
-
-      __asm (
+  double two52 = copysign (0x1.0p52, x);
+  double r, tmp;
+  
+  __asm (
 #ifdef _IEEE_FP_INEXACT
-	     "cvttq/svim %2,%1\n\t"
+	 "addt/suim %2, %3, %1\n\tsubt/suim %1, %3, %0"
 #else
-	     "cvttq/svm %2,%1\n\t"
+	 "addt/sum %2, %3, %1\n\tsubt/sum %1, %3, %0"
 #endif
-	     "cvtqt/m %1,%0\n\t"
-	     : "=f"(new_x), "=&f"(tmp1)
-	     : "f"(x));
-
-      /* floor(-0) == -0, and in general we'll always have the same
-	 sign as our input.  */
-      x = copysign(new_x, x);
-    }
-  return x;
+	 : "=&f"(r), "=&f"(tmp)
+	 : "f"(x), "f"(two52));
+
+  /* floor(-0) == -0, and in general we'll always have the same
+     sign as our input.  */
+  return copysign (r, x);
 }
 
 weak_alias (__floor, floor)
Index: sysdeps/alpha/fpu/s_floorf.c
===================================================================
RCS file: /cvs/glibc/libc/sysdeps/alpha/fpu/s_floorf.c,v
retrieving revision 1.5
diff -u -p -r1.5 s_floorf.c
--- sysdeps/alpha/fpu/s_floorf.c	6 Jul 2001 04:55:47 -0000	1.5
+++ sysdeps/alpha/fpu/s_floorf.c	14 Mar 2007 17:37:15 -0000
@@ -1,4 +1,4 @@
-/* Copyright (C) 1998, 1999, 2000 Free Software Foundation, Inc.
+/* Copyright (C) 1998, 1999, 2000, 2007 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Richard Henderson.
 
@@ -27,30 +27,21 @@
 float
 __floorf (float x)
 {
-  if (isless (fabsf (x), 16777216.0f))	/* 1 << FLT_MANT_DIG */
-    {
-      /* Note that Alpha S_Floating is stored in registers in a
-	 restricted T_Floating format, so we don't even need to
-	 convert back to S_Floating in the end.  The initial
-	 conversion to T_Floating is needed to handle denormals.  */
-
-      float tmp1, tmp2, new_x;
-
-      __asm ("cvtst/s %3,%2\n\t"
+  float two23 = copysignf (0x1.0p23, x);
+  float r, tmp;
+  
+  __asm (
 #ifdef _IEEE_FP_INEXACT
-	     "cvttq/svim %2,%1\n\t"
+	 "adds/suim %2, %3, %1\n\tsubs/suim %1, %3, %0"
 #else
-	     "cvttq/svm %2,%1\n\t"
+	 "adds/sum %2, %3, %1\n\tsubs/sum %1, %3, %0"
 #endif
-	     "cvtqt/m %1,%0\n\t"
-	     : "=f"(new_x), "=&f"(tmp1), "=&f"(tmp2)
-	     : "f"(x));
-
-      /* floor(-0) == -0, and in general we'll always have the same
-	 sign as our input.  */
-      x = copysignf(new_x, x);
-    }
-  return x;
+	 : "=&f"(r), "=&f"(tmp)
+	 : "f"(x), "f"(two23));
+
+  /* floor(-0) == -0, and in general we'll always have the same
+     sign as our input.  */
+  return copysignf (r, x);
 }
 
 weak_alias (__floorf, floorf)
Index: sysdeps/alpha/fpu/s_fmax.S
===================================================================
RCS file: sysdeps/alpha/fpu/s_fmax.S
diff -N sysdeps/alpha/fpu/s_fmax.S
--- /dev/null	1 Jan 1970 00:00:00 -0000
+++ sysdeps/alpha/fpu/s_fmax.S	14 Mar 2007 17:37:15 -0000
@@ -0,0 +1,58 @@
+/* Copyright (C) 2007 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by Richard Henderson.
+
+   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, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#include <sysdep.h>
+#include <math_ldbl_opt.h>
+
+        .set noat
+	.set noreorder
+
+	.text
+ENTRY (__fmax)
+	.prologue 0
+
+	cmptun/su	$f16, $f16, $f10
+	cmptun/su	$f17, $f17, $f11
+	fmov		$f17, $f0
+	unop
+
+	trapb
+	fbne		$f10, $ret
+	fmov		$f16, $f0
+	fbne		$f11, $ret
+
+	cmptlt/su	$f16, $f17, $f11
+	trapb
+	fcmovne		$f11, $f17, $f0
+$ret:	ret
+
+END (__fmax)
+
+/* Given the in-register format of single-precision, this works there too.  */
+strong_alias (__fmax, __fmaxf)
+weak_alias (__fmaxf, fmaxf)
+
+weak_alias (__fmax, fmax)
+#ifdef NO_LONG_DOUBLE
+strong_alias (__fmax, __fmaxl)
+weak_alias (__fmaxl, fmaxl)
+#endif
+#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_0)
+compat_symbol (libm, __fmax, fmaxl, GLIBC_2_0);
+#endif
Index: sysdeps/alpha/fpu/s_fmaxf.S
===================================================================
RCS file: sysdeps/alpha/fpu/s_fmaxf.S
diff -N sysdeps/alpha/fpu/s_fmaxf.S
--- /dev/null	1 Jan 1970 00:00:00 -0000
+++ sysdeps/alpha/fpu/s_fmaxf.S	14 Mar 2007 17:37:15 -0000
@@ -0,0 +1 @@
+/* __fmaxf is in s_fmax.c  */
Index: sysdeps/alpha/fpu/s_fmin.S
===================================================================
RCS file: sysdeps/alpha/fpu/s_fmin.S
diff -N sysdeps/alpha/fpu/s_fmin.S
--- /dev/null	1 Jan 1970 00:00:00 -0000
+++ sysdeps/alpha/fpu/s_fmin.S	14 Mar 2007 17:37:15 -0000
@@ -0,0 +1,58 @@
+/* Copyright (C) 2007 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by Richard Henderson.
+
+   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, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#include <sysdep.h>
+#include <math_ldbl_opt.h>
+
+        .set noat
+	.set noreorder
+
+	.text
+ENTRY (__fmin)
+	.prologue 0
+
+	cmptun/su	$f16, $f16, $f10
+	cmptun/su	$f17, $f17, $f11
+	fmov		$f17, $f0
+	unop
+
+	trapb
+	fbne		$f10, $ret
+	fmov		$f16, $f0
+	fbne		$f11, $ret
+
+	cmptlt/su	$f17, $f16, $f11
+	trapb
+	fcmovne		$f11, $f17, $f0
+$ret:	ret
+
+END (__fmin)
+
+/* Given the in-register format of single-precision, this works there too.  */
+strong_alias (__fmin, __fminf)
+weak_alias (__fminf, fminf)
+
+weak_alias (__fmin, fmin)
+#ifdef NO_LONG_DOUBLE
+strong_alias (__fmin, __fminl)
+weak_alias (__fminl, fminl)
+#endif
+#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_0)
+compat_symbol (libm, __fmin, fminl, GLIBC_2_0);
+#endif
Index: sysdeps/alpha/fpu/s_fminf.S
===================================================================
RCS file: sysdeps/alpha/fpu/s_fminf.S
diff -N sysdeps/alpha/fpu/s_fminf.S
--- /dev/null	1 Jan 1970 00:00:00 -0000
+++ sysdeps/alpha/fpu/s_fminf.S	14 Mar 2007 17:37:15 -0000
@@ -0,0 +1 @@
+/* __fminf is in s_fmin.c  */
Index: sysdeps/alpha/fpu/s_isnan.c
===================================================================
RCS file: sysdeps/alpha/fpu/s_isnan.c
diff -N sysdeps/alpha/fpu/s_isnan.c
--- /dev/null	1 Jan 1970 00:00:00 -0000
+++ sysdeps/alpha/fpu/s_isnan.c	14 Mar 2007 17:37:15 -0000
@@ -0,0 +1,58 @@
+/* Return 1 if argument is a NaN, else 0.
+   Copyright (C) 2007 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, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+/* Ugly kludge to avoid declarations.  */
+#define __isnanf	not___isnanf
+#define isnanf		not_isnanf
+#define __GI___isnanf	not__GI___isnanf
+
+#include <math.h>
+#include <math_ldbl_opt.h>
+
+#undef __isnanf
+#undef isnanf
+#undef __GI___isnanf
+
+/* The hidden_proto in include/math.h was obscured by the macro hackery.  */
+__typeof (__isnan) __isnanf;
+hidden_proto (__isnanf)
+
+
+int
+__isnan (double x)
+{
+  return isunordered (x, x);
+}
+
+hidden_def (__isnan)
+weak_alias (__isnan, isnan)
+
+/* It turns out that the 'double' version will also always work for
+   single-precision.  */
+strong_alias (__isnan, __isnanf)
+hidden_def (__isnanf)
+weak_alias (__isnanf, isnanf)
+
+#ifdef NO_LONG_DOUBLE
+strong_alias (__isnan, __isnanl)
+weak_alias (__isnan, isnanl)
+#endif
+#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_0)
+compat_symbol (libm, __isnan, isnanl, GLIBC_2_0);
+#endif
Index: sysdeps/alpha/fpu/s_isnanf.c
===================================================================
RCS file: sysdeps/alpha/fpu/s_isnanf.c
diff -N sysdeps/alpha/fpu/s_isnanf.c
--- /dev/null	1 Jan 1970 00:00:00 -0000
+++ sysdeps/alpha/fpu/s_isnanf.c	14 Mar 2007 17:37:15 -0000
@@ -0,0 +1 @@
+/* In s_isnan.c */
Index: sysdeps/alpha/fpu/s_llrint.c
===================================================================
RCS file: sysdeps/alpha/fpu/s_llrint.c
diff -N sysdeps/alpha/fpu/s_llrint.c
--- /dev/null	1 Jan 1970 00:00:00 -0000
+++ sysdeps/alpha/fpu/s_llrint.c	14 Mar 2007 17:37:15 -0000
@@ -0,0 +1 @@
+/* In s_lrint.c */
Index: sysdeps/alpha/fpu/s_llrintf.c
===================================================================
RCS file: sysdeps/alpha/fpu/s_llrintf.c
diff -N sysdeps/alpha/fpu/s_llrintf.c
--- /dev/null	1 Jan 1970 00:00:00 -0000
+++ sysdeps/alpha/fpu/s_llrintf.c	14 Mar 2007 17:37:15 -0000
@@ -0,0 +1 @@
+/* In s_lrintf.c */
Index: sysdeps/alpha/fpu/s_lrint.c
===================================================================
RCS file: sysdeps/alpha/fpu/s_lrint.c
diff -N sysdeps/alpha/fpu/s_lrint.c
--- /dev/null	1 Jan 1970 00:00:00 -0000
+++ sysdeps/alpha/fpu/s_lrint.c	14 Mar 2007 17:37:15 -0000
@@ -0,0 +1,48 @@
+/* Copyright (C) 2007 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, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#define __llrint	not___llrint
+#define llrint		not_llrint
+#include <math.h>
+#include <math_ldbl_opt.h>
+#undef __llrint
+#undef llrint
+
+long int
+__lrint (double x)
+{
+  long ret;
+
+  __asm ("cvttq/svd %1,%0" : "=&f"(ret) : "f"(x));
+
+  return ret;
+}
+
+strong_alias (__lrint, __llrint)
+weak_alias (__lrint, lrint)
+weak_alias (__llrint, llrint)
+#ifdef NO_LONG_DOUBLE
+strong_alias (__lrint, __lrintl)
+strong_alias (__lrint, __llrintl)
+weak_alias (__lrintl, lrintl)
+weak_alias (__llrintl, llrintl)
+#endif
+#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_0)
+compat_symbol (libm, __lrint, lrintl, GLIBC_2_0);
+compat_symbol (libm, __llrint, llrintl, GLIBC_2_0);
+#endif
Index: sysdeps/alpha/fpu/s_lrintf.c
===================================================================
RCS file: sysdeps/alpha/fpu/s_lrintf.c
diff -N sysdeps/alpha/fpu/s_lrintf.c
--- /dev/null	1 Jan 1970 00:00:00 -0000
+++ sysdeps/alpha/fpu/s_lrintf.c	14 Mar 2007 17:37:15 -0000
@@ -0,0 +1,39 @@
+/* Copyright (C) 2007 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, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#define __llrintf	not___llrintf
+#define llrintf		not_llrintf
+#include <math.h>
+#undef __llrintf
+#undef llrintf
+
+long int
+__lrintf (float x)
+{
+  double tmp;
+  long ret;
+
+  __asm ("cvtst/s %2,%1\n\tcvttq/svd %1,%0"
+	 : "=&f"(ret), "=&f"(tmp) : "f"(x));
+
+  return ret;
+}
+
+strong_alias (__lrintf, __llrintf)
+weak_alias (__lrintf, lrintf)
+weak_alias (__llrintf, llrintf)
Index: sysdeps/alpha/fpu/s_nearbyint.c
===================================================================
RCS file: sysdeps/alpha/fpu/s_nearbyint.c
diff -N sysdeps/alpha/fpu/s_nearbyint.c
--- /dev/null	1 Jan 1970 00:00:00 -0000
+++ sysdeps/alpha/fpu/s_nearbyint.c	14 Mar 2007 17:37:15 -0000
@@ -0,0 +1,48 @@
+/* Copyright (C) 2007 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by Richard Henderson.
+
+   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, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#include <math.h>
+#include <math_ldbl_opt.h>
+
+#ifdef _IEEE_FP_INEXACT
+#error "Don't compile with -mieee-with-inexact"
+#endif
+
+double
+__nearbyint (double x)
+{
+  double two52 = copysign (0x1.0p52, x);
+  double r;
+  
+  r = x + two52;
+  r = r - two52;
+
+  /* nearbyint(-0.1) == -0, and in general we'll always have the same sign
+     as our input.  */
+  return copysign (r, x);
+}
+
+weak_alias (__nearbyint, nearbyint)
+#ifdef NO_LONG_DOUBLE
+strong_alias (__nearbyint, __nearbyintl)
+weak_alias (__nearbyint, nearbyintl)
+#endif
+#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_0)
+compat_symbol (libm, __nearbyint, nearbyintl, GLIBC_2_0);
+#endif
Index: sysdeps/alpha/fpu/s_nearbyintf.c
===================================================================
RCS file: sysdeps/alpha/fpu/s_nearbyintf.c
diff -N sysdeps/alpha/fpu/s_nearbyintf.c
--- /dev/null	1 Jan 1970 00:00:00 -0000
+++ sysdeps/alpha/fpu/s_nearbyintf.c	14 Mar 2007 17:37:15 -0000
@@ -0,0 +1,40 @@
+/* Copyright (C) 2007 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by Richard Henderson.
+
+   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, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#include <math.h>
+
+#ifdef _IEEE_FP_INEXACT
+#error "Don't compile with -mieee-with-inexact"
+#endif
+
+float
+__nearbyintf (float x)
+{
+  float two23 = copysignf (0x1.0p23, x);
+  float r;
+
+  r = x + two23;
+  r = r - two23;
+
+  /* nearbyint(-0.1) == -0, and in general we'll always have the same sign
+     as our input.  */
+  return copysign (r, x);
+}
+
+weak_alias (__nearbyintf, nearbyintf)
Index: sysdeps/alpha/fpu/s_rint.c
===================================================================
RCS file: /cvs/glibc/libc/sysdeps/alpha/fpu/s_rint.c,v
retrieving revision 1.3
diff -u -p -r1.3 s_rint.c
--- sysdeps/alpha/fpu/s_rint.c	1 Feb 2006 03:13:45 -0000	1.3
+++ sysdeps/alpha/fpu/s_rint.c	14 Mar 2007 17:37:15 -0000
@@ -1,4 +1,4 @@
-/* Copyright (C) 2000, 2006 Free Software Foundation, Inc.
+/* Copyright (C) 2000, 2006, 2007 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Richard Henderson.
 
@@ -24,24 +24,15 @@
 double
 __rint (double x)
 {
-  if (isless (fabs (x), 9007199254740992.0))	/* 1 << DBL_MANT_DIG */
-    {
-      double tmp1, new_x;
-      __asm (
-#ifdef _IEEE_FP_INEXACT
-	     "cvttq/svid %2,%1\n\t"
-#else
-	     "cvttq/svd %2,%1\n\t"
-#endif
-	     "cvtqt/d %1,%0\n\t"
-	     : "=f"(new_x), "=&f"(tmp1)
-	     : "f"(x));
-
-      /* rint(-0.1) == -0, and in general we'll always have the same
-	 sign as our input.  */
-      x = copysign(new_x, x);
-    }
-  return x;
+  double two52 = copysign (0x1.0p52, x);
+  double r;
+  
+  r = x + two52;
+  r = r - two52;
+
+  /* rint(-0.1) == -0, and in general we'll always have the same sign
+     as our input.  */
+  return copysign (r, x);
 }
 
 weak_alias (__rint, rint)
Index: sysdeps/alpha/fpu/s_rintf.c
===================================================================
RCS file: /cvs/glibc/libc/sysdeps/alpha/fpu/s_rintf.c,v
retrieving revision 1.2
diff -u -p -r1.2 s_rintf.c
--- sysdeps/alpha/fpu/s_rintf.c	6 Jul 2001 04:55:47 -0000	1.2
+++ sysdeps/alpha/fpu/s_rintf.c	14 Mar 2007 17:37:15 -0000
@@ -1,4 +1,4 @@
-/* Copyright (C) 2000 Free Software Foundation, Inc.
+/* Copyright (C) 2000, 2007 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Richard Henderson.
 
@@ -23,30 +23,15 @@
 float
 __rintf (float x)
 {
-  if (isless (fabsf (x), 16777216.0f))	/* 1 << FLT_MANT_DIG */
-    {
-      /* Note that Alpha S_Floating is stored in registers in a
-	 restricted T_Floating format, so we don't even need to
-	 convert back to S_Floating in the end.  The initial
-	 conversion to T_Floating is needed to handle denormals.  */
-
-      float tmp1, tmp2, new_x;
-
-      __asm ("cvtst/s %3,%2\n\t"
-#ifdef _IEEE_FP_INEXACT
-	     "cvttq/svid %2,%1\n\t"
-#else
-	     "cvttq/svd %2,%1\n\t"
-#endif
-	     "cvtqt/d %1,%0\n\t"
-	     : "=f"(new_x), "=&f"(tmp1), "=&f"(tmp2)
-	     : "f"(x));
-
-      /* rint(-0.1) == -0, and in general we'll always have the same
-	 sign as our input.  */
-      x = copysignf(new_x, x);
-    }
-  return x;
+  float two23 = copysignf (0x1.0p23, x);
+  float r;
+
+  r = x + two23;
+  r = r - two23;
+
+  /* rint(-0.1) == -0, and in general we'll always have the same sign
+     as our input.  */
+  return copysign (r, x);
 }
 
 weak_alias (__rintf, rintf)
Index: sysdeps/alpha/fpu/bits/mathinline.h
===================================================================
RCS file: /cvs/glibc/libc/sysdeps/alpha/fpu/bits/mathinline.h,v
retrieving revision 1.16
diff -u -p -r1.16 mathinline.h
--- sysdeps/alpha/fpu/bits/mathinline.h	14 Mar 2007 00:40:50 -0000	1.16
+++ sysdeps/alpha/fpu/bits/mathinline.h	14 Mar 2007 17:37:15 -0000
@@ -1,5 +1,6 @@
 /* Inline math functions for Alpha.
-   Copyright (C) 1996, 1997, 1999-2001, 2004 Free Software Foundation, Inc.
+   Copyright (C) 1996, 1997, 1999-2001, 2004, 2007
+   Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by David Mosberger-Tang.
 
@@ -83,111 +84,64 @@ __inline_fabs (fabs, double)
 # undef __inline_fabs
 #endif
 
-
-/* Use the -inf rounding mode conversion instructions to implement
-   floor.  We note when the exponent is large enough that the value
-   must be integral, as this avoids unpleasant integer overflows.  */
-
-__MATH_INLINE float
-__NTH (__floorf (float __x))
-{
-  /* Check not zero since floor(-0) == -0.  */
-  if (__x != 0 && fabsf (__x) < 16777216.0f)  /* 1 << FLT_MANT_DIG */
-    {
-      /* Note that Alpha S_Floating is stored in registers in a
-	 restricted T_Floating format, so we don't even need to
-	 convert back to S_Floating in the end.  The initial
-	 conversion to T_Floating is needed to handle denormals.  */
-
-      float __tmp1, __tmp2;
-
-      __asm ("cvtst/s %3,%2\n\t"
-#ifdef _IEEE_FP_INEXACT
-	     "cvttq/svim %2,%1\n\t"
-#else
-	     "cvttq/svm %2,%1\n\t"
-#endif
-	     "cvtqt/m %1,%0\n\t"
-	     : "=f"(__x), "=&f"(__tmp1), "=&f"(__tmp2)
-	     : "f"(__x));
-    }
-  return __x;
-}
-
-__MATH_INLINE double
-__NTH (__floor (double __x))
-{
-  if (__x != 0 && fabs (__x) < 9007199254740992.0)  /* 1 << DBL_MANT_DIG */
-    {
-      double __tmp1;
-      __asm (
-#ifdef _IEEE_FP_INEXACT
-	     "cvttq/svim %2,%1\n\t"
-#else
-	     "cvttq/svm %2,%1\n\t"
-#endif
-	     "cvtqt/m %1,%0\n\t"
-	     : "=f"(__x), "=&f"(__tmp1)
-	     : "f"(__x));
-    }
-  return __x;
-}
-
-__MATH_INLINE float __NTH (floorf (float __x)) { return __floorf(__x); }
-__MATH_INLINE double __NTH (floor (double __x)) { return __floor(__x); }
-
-
 #ifdef __USE_ISOC99
 
-__MATH_INLINE float
-__NTH (__fdimf (float __x, float __y))
-{
-  return __x <= __y ? 0.0f : __x - __y;
-}
-
-__MATH_INLINE float
-__NTH (fdimf (float __x, float __y))
-{
-  return __x <= __y ? 0.0f : __x - __y;
-}
-
-__MATH_INLINE double
-__NTH (__fdim (double __x, double __y))
-{
-  return __x <= __y ? 0.0 : __x - __y;
-}
-
-__MATH_INLINE double
-__NTH (fdim (double __x, double __y))
-{
-  return __x <= __y ? 0.0 : __x - __y;
-}
-
 /* Test for negative number.  Used in the signbit() macro.  */
 __MATH_INLINE int
 __NTH (__signbitf (float __x))
 {
+#if !__GNUC_PREREQ (4, 0)
   __extension__ union { float __f; int __i; } __u = { __f: __x };
   return __u.__i < 0;
+#else
+  return __builtin_signbitf (__x);
+#endif
 }
 
 __MATH_INLINE int
 __NTH (__signbit (double __x))
 {
+#if !__GNUC_PREREQ (4, 0)
   __extension__ union { double __d; long __i; } __u = { __d: __x };
   return __u.__i < 0;
+#else
+  return __builtin_signbit (__x);
+#endif
 }
 
 __MATH_INLINE int
 __NTH (__signbitl (long double __x))
 {
+#if !__GNUC_PREREQ (4, 0)
   __extension__ union {
     long double __d;
     long __i[sizeof(long double)/sizeof(long)];
   } __u = { __d: __x };
   return __u.__i[sizeof(long double)/sizeof(long) - 1] < 0;
+#else
+  return __builtin_signbitl (__x);
+#endif
 }
 
+/* Test for NaN.  Used in the isnan() macro.  */
+
+__MATH_INLINE int
+__NTH (__isnanf (float __x))
+{
+  return isunordered (__x, __x);
+}
+
+__MATH_INLINE int
+__NTH (__isnan (double __x))
+{
+  return isunordered (__x, __x);
+}
+
+__MATH_INLINE int
+__NTH (__isnanl (long double __x))
+{
+  return isunordered (__x, __x);
+}
 #endif /* C99 */
 
 #endif /* __NO_MATH_INLINES */
	* sysdeps/alpha/fpu/s_ceilf.c: Likewise.
	* sysdeps/alpha/fpu/s_floor.c: Likewise.
	* sysdeps/alpha/fpu/s_floorf.c: Likewise.
	* sysdeps/alpha/fpu/s_rint.c: Likewise.
	* sysdeps/alpha/fpu/s_rintf.c: Likewise.

	* sysdeps/alpha/fpu/s_fmax.S: New file.
	* sysdeps/alpha/fpu/s_fmaxf.S: New file.
	* sysdeps/alpha/fpu/s_fmin.S: New file.
	* sysdeps/alpha/fpu/s_fminf.S: New file.
	* sysdeps/alpha/fpu/s_isnan.c: New file.
	* sysdeps/alpha/fpu/s_isnanf.c: New file.
	* sysdeps/alpha/fpu/s_llrint.c: New file.
	* sysdeps/alpha/fpu/s_llrintf.c: New file.
	* sysdeps/alpha/fpu/s_lrint.c: New file.
	* sysdeps/alpha/fpu/s_lrintf.c: New file.
	* sysdeps/alpha/fpu/s_nearbyint.c: New file.
	* sysdeps/alpha/fpu/s_nearbyintf.c: New file.

	* sysdeps/alpha/fpu/bits/mathinline.h (__floorf, __floor): Remove.
	(__fdimf, fdimf, __fdim, fdim): Remove.
	(__signbitf, __signbit, __signbitl): Use gcc builtin if available.
	(__isnanf, __isnan, __isnanl): New.

^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: [alpha] math routine improvements
  2007-03-14 17:48 [alpha] math routine improvements Richard Henderson
@ 2007-03-18 19:02 ` Jakub Jelinek
  0 siblings, 0 replies; 2+ messages in thread
From: Jakub Jelinek @ 2007-03-18 19:02 UTC (permalink / raw)
  To: Richard Henderson; +Cc: Glibc hackers

On Wed, Mar 14, 2007 at 10:48:06AM -0700, Richard Henderson wrote:
> Branchless versions of some existing rounding routines.
> Implement some easy functions that were missing.
> Remove inline versions of fdim, because they handle inf incorrectly.

Have you checked libc.so/libm.so ABI before/after the patch
(I'm using:
    for i in `find . -name '*.so*' -a -type f`; do
      readelf -Ws $i \
          | sed -n '/\.symtab/,$d;/ UND /d;/@GLIBC_PRIVATE/d;/\(GLOBAL\|WEAK\)/p' \
          | awk '{ if ($4 == "OBJECT") { printf "%s %s %s %s %s\n", $8, $4, $5, $6, $3 } else { printf "%s %s %s %s\n", $8, $4, $5, $6 }}' \
          | LC_ALL=C sort -u > ~/xxabilist/$1-$2-$3/$b/`echo $i | sed 's|^\./||;s|/|.|g'`.abi
    done
and then diff the files)?

From a quick glance at the new files and math/Versions and/or
sysdeps/ldbl-opt/ compat_symbol versions I'd say following is needed,
but it is untested.

2007-03-18  Jakub Jelinek  <jakub@redhat.com>

	* sysdeps/alpha/fpu/s_nearbyint.c (nearbyintl): Fix version on
	compat_symbol to GLIBC_2_1.
	* sysdeps/alpha/fpu/s_fmin.S (fminl): Likewise.
	* sysdeps/alpha/fpu/s_trunc.c (truncl): Likewise.
	* sysdeps/alpha/fpu/s_fmax.S (fmaxl): Likewise.
	* sysdeps/alpha/fpu/s_lrint.c (lrintl, llrintl): Likewise.
	* sysdeps/alpha/fpu/s_lround.c (lroundl, llroundl): Likewise.
	* sysdeps/alpha/fpu/s_round.c (roundl): Likewise.
	* sysdeps/alpha/fpu/s_isnan.c (isnanl): Provide compat_symbol in
	libc, not libm.
	(__isnanl): New compat_symbol.

--- libc/sysdeps/alpha/fpu/s_nearbyint.c.jj	2007-03-14 18:44:14.000000000 +0100
+++ libc/sysdeps/alpha/fpu/s_nearbyint.c	2007-03-18 19:25:04.000000000 +0100
@@ -43,6 +43,6 @@ weak_alias (__nearbyint, nearbyint)
 strong_alias (__nearbyint, __nearbyintl)
 weak_alias (__nearbyint, nearbyintl)
 #endif
-#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_0)
-compat_symbol (libm, __nearbyint, nearbyintl, GLIBC_2_0);
+#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_1)
+compat_symbol (libm, __nearbyint, nearbyintl, GLIBC_2_1);
 #endif
--- libc/sysdeps/alpha/fpu/s_isnan.c.jj	2007-03-14 18:44:14.000000000 +0100
+++ libc/sysdeps/alpha/fpu/s_isnan.c	2007-03-18 19:27:46.000000000 +0100
@@ -53,6 +53,9 @@ weak_alias (__isnanf, isnanf)
 strong_alias (__isnan, __isnanl)
 weak_alias (__isnan, isnanl)
 #endif
-#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_0)
-compat_symbol (libm, __isnan, isnanl, GLIBC_2_0);
+#ifndef IS_IN_libm
+# if LONG_DOUBLE_COMPAT(libc, GLIBC_2_0)
+compat_symbol (libc, __isnan, __isnanl, GLIBC_2_0);
+compat_symbol (libc, isnan, isnanl, GLIBC_2_0);
+# endif
 #endif
--- libc/sysdeps/alpha/fpu/s_fmin.S.jj	2007-03-14 18:44:14.000000000 +0100
+++ libc/sysdeps/alpha/fpu/s_fmin.S	2007-03-18 19:28:09.000000000 +0100
@@ -53,6 +53,6 @@ weak_alias (__fmin, fmin)
 strong_alias (__fmin, __fminl)
 weak_alias (__fminl, fminl)
 #endif
-#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_0)
-compat_symbol (libm, __fmin, fminl, GLIBC_2_0);
+#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_1)
+compat_symbol (libm, __fmin, fminl, GLIBC_2_1);
 #endif
--- libc/sysdeps/alpha/fpu/s_trunc.c.jj	2007-03-14 21:01:05.000000000 +0100
+++ libc/sysdeps/alpha/fpu/s_trunc.c	2007-03-18 19:24:22.000000000 +0100
@@ -48,6 +48,6 @@ weak_alias (__trunc, trunc)
 strong_alias (__trunc, __truncl)
 weak_alias (__trunc, truncl)
 #endif
-#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_0)
-compat_symbol (libm, __trunc, truncl, GLIBC_2_0);
+#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_1)
+compat_symbol (libm, __trunc, truncl, GLIBC_2_1);
 #endif
--- libc/sysdeps/alpha/fpu/s_fmax.S.jj	2007-03-14 18:44:14.000000000 +0100
+++ libc/sysdeps/alpha/fpu/s_fmax.S	2007-03-18 19:28:20.000000000 +0100
@@ -53,6 +53,6 @@ weak_alias (__fmax, fmax)
 strong_alias (__fmax, __fmaxl)
 weak_alias (__fmaxl, fmaxl)
 #endif
-#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_0)
-compat_symbol (libm, __fmax, fmaxl, GLIBC_2_0);
+#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_1)
+compat_symbol (libm, __fmax, fmaxl, GLIBC_2_1);
 #endif
--- libc/sysdeps/alpha/fpu/s_lrint.c.jj	2007-03-14 18:44:14.000000000 +0100
+++ libc/sysdeps/alpha/fpu/s_lrint.c	2007-03-18 19:26:07.000000000 +0100
@@ -42,7 +42,7 @@ strong_alias (__lrint, __llrintl)
 weak_alias (__lrintl, lrintl)
 weak_alias (__llrintl, llrintl)
 #endif
-#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_0)
-compat_symbol (libm, __lrint, lrintl, GLIBC_2_0);
-compat_symbol (libm, __llrint, llrintl, GLIBC_2_0);
+#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_1)
+compat_symbol (libm, __lrint, lrintl, GLIBC_2_1);
+compat_symbol (libm, __llrint, llrintl, GLIBC_2_1);
 #endif
--- libc/sysdeps/alpha/fpu/s_lround.c.jj	2007-03-14 21:01:05.000000000 +0100
+++ libc/sysdeps/alpha/fpu/s_lround.c	2007-03-18 19:25:43.000000000 +0100
@@ -42,7 +42,7 @@ strong_alias (__lround, __llroundl)
 weak_alias (__lroundl, lroundl)
 weak_alias (__llroundl, llroundl)
 #endif
-#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_0)
-compat_symbol (libm, __lround, lroundl, GLIBC_2_0);
-compat_symbol (libm, __llround, llroundl, GLIBC_2_0);
+#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_1)
+compat_symbol (libm, __lround, lroundl, GLIBC_2_1);
+compat_symbol (libm, __llround, llroundl, GLIBC_2_1);
 #endif
--- libc/sysdeps/alpha/fpu/s_round.c.jj	2007-03-14 21:01:05.000000000 +0100
+++ libc/sysdeps/alpha/fpu/s_round.c	2007-03-18 19:24:40.000000000 +0100
@@ -44,6 +44,6 @@ weak_alias (__round, round)
 strong_alias (__round, __roundl)
 weak_alias (__roundl, roundl)
 #endif
-#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_0)
-compat_symbol (libm, __round, roundl, GLIBC_2_0);
+#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_1)
+compat_symbol (libm, __round, roundl, GLIBC_2_1);
 #endif


	Jakub

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2007-03-18 19:02 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2007-03-14 17:48 [alpha] math routine improvements Richard Henderson
2007-03-18 19:02 ` Jakub Jelinek

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).