alpha: Remove s_ceil{f} and s_floor{f} implementation (BZ#22665)

Message ID 1515002624-26820-1-git-send-email-adhemerval.zanella@linaro.org
State New
Headers show
Series
  • alpha: Remove s_ceil{f} and s_floor{f} implementation (BZ#22665)
Related show

Commit Message

Adhemerval Zanella Jan. 3, 2018, 6:03 p.m.
As discussed in libc-alpha [1], alpha ceil{f} and floor{f}
implementation uses cvttq/svm and although the Alpha Architecture
Handbook version 3 states that that CVTfi OUTPUT Exceptions
(B.3 Mapping to IEEE Standard) should not generate Inexact if INE
bit is set on fpcr, the Alpha 21264 [1] chip manual (A.8 IEEE
Floating-Point Conformance) states that CVTfi and CVTif OUTPUT
does generate inexact exception for inexact result regardless.

As Joseph noted [2] to correctly fix it on alpha we need to either
avoid the instruction or avoid any inexact bit from it being set
on return from the function (while preserving the inexact bit that
might be set on the entry to the function).  The later will result
mf_fpcr followed by a mt_fpcr to get and set the fpcr which will
defeat the optimization itself.

So the patch just remove the alpha optimized and rely on generic
implementation.  It fixes the math/test-*-{ceil,floor} on alpha.

	[BZ #15479]
	[BZ #22665]
	* sysdeps/alpha/fpu/s_ceil.c: Remove file.
	* sysdeps/alpha/fpu/s_ceilf.c: Likewise.
	* sysdeps/alpha/fpu/s_floor.c: Likewise.
	* sysdeps/alpha/fpu/s_floorf.c: Likewise.

[1] https://www.star.bnl.gov/public/daq/HARDWARE/21264_data_sheet.pdf
[2] https://sourceware.org/ml/libc-alpha/2018-01/msg00086.html
---
 ChangeLog                    |  9 ++++++++
 sysdeps/alpha/fpu/s_ceil.c   | 50 ----------------------------------------
 sysdeps/alpha/fpu/s_ceilf.c  | 54 -------------------------------------------
 sysdeps/alpha/fpu/s_floor.c  | 51 ----------------------------------------
 sysdeps/alpha/fpu/s_floorf.c | 55 --------------------------------------------
 5 files changed, 9 insertions(+), 210 deletions(-)
 delete mode 100644 sysdeps/alpha/fpu/s_ceil.c
 delete mode 100644 sysdeps/alpha/fpu/s_ceilf.c
 delete mode 100644 sysdeps/alpha/fpu/s_floor.c
 delete mode 100644 sysdeps/alpha/fpu/s_floorf.c

Comments

Richard Henderson Jan. 4, 2018, 6:57 p.m. | #1
On 01/03/2018 10:03 AM, Adhemerval Zanella wrote:
> As discussed in libc-alpha [1], alpha ceil{f} and floor{f}
> implementation uses cvttq/svm and although the Alpha Architecture
> Handbook version 3 states that that CVTfi OUTPUT Exceptions
> (B.3 Mapping to IEEE Standard) should not generate Inexact if INE
> bit is set on fpcr, the Alpha 21264 [1] chip manual (A.8 IEEE
> Floating-Point Conformance) states that CVTfi and CVTif OUTPUT
> does generate inexact exception for inexact result regardless.
> 
> As Joseph noted [2] to correctly fix it on alpha we need to either
> avoid the instruction or avoid any inexact bit from it being set
> on return from the function (while preserving the inexact bit that
> might be set on the entry to the function).  The later will result
> mf_fpcr followed by a mt_fpcr to get and set the fpcr which will
> defeat the optimization itself.
> 
> So the patch just remove the alpha optimized and rely on generic
> implementation.  It fixes the math/test-*-{ceil,floor} on alpha.
> 
> 	[BZ #15479]
> 	[BZ #22665]
> 	* sysdeps/alpha/fpu/s_ceil.c: Remove file.
> 	* sysdeps/alpha/fpu/s_ceilf.c: Likewise.
> 	* sysdeps/alpha/fpu/s_floor.c: Likewise.
> 	* sysdeps/alpha/fpu/s_floorf.c: Likewise.

ACK.


r~

Patch

diff --git a/sysdeps/alpha/fpu/s_ceil.c b/sysdeps/alpha/fpu/s_ceil.c
deleted file mode 100644
index 16c8f99..0000000
--- a/sysdeps/alpha/fpu/s_ceil.c
+++ /dev/null
@@ -1,50 +0,0 @@ 
-/* Copyright (C) 1998-2018 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, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <math.h>
-#include <math_ldbl_opt.h>
-#include <libm-alias-double.h>
-
-/* Use the -inf rounding mode conversion instructions to implement
-   ceil, via something akin to -floor(-x).  This is much faster than
-   playing with the fpcr to achieve +inf rounding mode.  */
-
-double
-__ceil (double x)
-{
-  if (isnan (x))
-    return x + x;
-
-  if (isless (fabs (x), 9007199254740992.0))	/* 1 << DBL_MANT_DIG */
-    {
-      double tmp1, new_x;
-
-      new_x = -x;
-      __asm (
-	     "cvttq/svm %2,%1\n\t"
-	     "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;
-}
-
-libm_alias_double (__ceil, ceil)
diff --git a/sysdeps/alpha/fpu/s_ceilf.c b/sysdeps/alpha/fpu/s_ceilf.c
deleted file mode 100644
index 8561232..0000000
--- a/sysdeps/alpha/fpu/s_ceilf.c
+++ /dev/null
@@ -1,54 +0,0 @@ 
-/* Copyright (C) 1998-2018 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, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <math.h>
-#include <libm-alias-float.h>
-
-/* Use the -inf rounding mode conversion instructions to implement
-   ceil, via something akin to -floor(-x).  This is much faster than
-   playing with the fpcr to achieve +inf rounding mode.  */
-
-float
-__ceilf (float x)
-{
-  if (isnanf (x))
-    return x + 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"
-	     "cvttq/svm %2,%1\n\t"
-	     "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;
-}
-
-libm_alias_float (__ceil, ceil)
diff --git a/sysdeps/alpha/fpu/s_floor.c b/sysdeps/alpha/fpu/s_floor.c
deleted file mode 100644
index 285ad36..0000000
--- a/sysdeps/alpha/fpu/s_floor.c
+++ /dev/null
@@ -1,51 +0,0 @@ 
-/* Copyright (C) 1998-2018 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, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <math.h>
-#include <math_ldbl_opt.h>
-#include <libm-alias-double.h>
-
-
-/* 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.  */
-
-double
-__floor (double x)
-{
-  if (isnan (x))
-    return x + x;
-
-  if (isless (fabs (x), 9007199254740992.0))	/* 1 << DBL_MANT_DIG */
-    {
-      double tmp1, new_x;
-
-      __asm (
-	     "cvttq/svm %2,%1\n\t"
-	     "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;
-}
-
-libm_alias_double (__floor, floor)
diff --git a/sysdeps/alpha/fpu/s_floorf.c b/sysdeps/alpha/fpu/s_floorf.c
deleted file mode 100644
index a73b3fd..0000000
--- a/sysdeps/alpha/fpu/s_floorf.c
+++ /dev/null
@@ -1,55 +0,0 @@ 
-/* Copyright (C) 1998-2018 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, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <math.h>
-#include <libm-alias-float.h>
-
-
-/* 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.  */
-
-float
-__floorf (float x)
-{
-  if (isnanf (x))
-    return x + 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"
-	     "cvttq/svm %2,%1\n\t"
-	     "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;
-}
-
-libm_alias_float (__floor, floor)