Re: optimization/9651: [Alpha] FPE with NAN in spite of isnan protection

Re: optimization/9651: [Alpha] FPE with NAN in spite of isnan protection

[Richard Henderson]

The following reply was made to PR optimization/9651; it has been noted by GNATS.

From: Richard Henderson <rth@redhat.com>
To: Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
Cc: gcc-gnats@gcc.gnu.org, gcc-patches@gcc.gnu.org,
   Will Woods <will.woods@hp.com>
Subject: Re: optimization/9651: [Alpha] FPE with NAN in spite of isnan protection
Date: Tue, 11 Feb 2003 12:49:34 -0800

 On Tue, Feb 11, 2003 at 03:53:46AM +0100, Falk Hueffner wrote:
 > 	* rtlanal.c (may_trap_p): Handle FIX.
 
 Applied.
 
 
 r~

Re: optimization/9651: [Alpha] FPE with NAN in spite of isnan protection

[rth]

Synopsis: [Alpha] FPE with NAN in spite of isnan protection

State-Changed-From-To: open->closed
State-Changed-By: rth
State-Changed-When: Tue Feb 11 20:59:18 2003
State-Changed-Why:
    Applied 3.3 and mainline.

http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=9651

Re: optimization/9651: [Alpha] FPE with NAN in spite of isnan protection

[Falk Hueffner]

The following reply was made to PR optimization/9651; it has been noted by GNATS.

From: Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
To: gcc-gnats@gcc.gnu.org,gcc-patches@gcc.gnu.org
Cc: Will Woods <will.woods@hp.com>
Subject: Re: optimization/9651: [Alpha] FPE with NAN in spite of isnan protection
Date: 11 Feb 2003 03:53:46 +0100

 --=-=-=
 
 Hi,
 
 This patch marks fix expressions as trappable, preventing them from
 being scheduled out of their block.
 
 2003-02-11  Falk Hueffner  <falk.hueffner@student.uni-tuebingen.de>
 
         PR optimization/9651
 	* rtlanal.c (may_trap_p): Handle FIX.
 
 -- 
 	Falk
 
 
 --=-=-=
 Content-Type: text/x-patch
 Content-Disposition: attachment; filename=fix-trap.patch
 
 Index: gcc/rtlanal.c
 ===================================================================
 RCS file: /cvs/gcc/gcc/gcc/rtlanal.c,v
 retrieving revision 1.145
 diff -u -r1.145 rtlanal.c
 --- gcc/rtlanal.c	17 Jan 2003 03:28:09 -0000	1.145
 +++ gcc/rtlanal.c	11 Feb 2003 02:50:01 -0000
 @@ -2517,6 +2517,12 @@
  	return 1;
        break;
  
 +    case FIX:
 +      /* Conversion of floating point might trap.  */
 +      if (flag_trapping_math && HONOR_NANS (GET_MODE (XEXP (x, 0))))
 +	return 1;
 +      break;
 +
      case NEG:
      case ABS:
        /* These operations don't trap even with floating point.  */
 
 --=-=-=--

optimization/9651: [Alpha] FPE with NAN in spite of isnan protection

[Falk Hueffner]

>Number:         9651
>Category:       optimization
>Synopsis:       [Alpha] FPE with NAN in spite of isnan protection
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          wrong-code
>Submitter-Id:   net
>Arrival-Date:   Tue Feb 11 00:36:00 UTC 2003
>Closed-Date:
>Last-Modified:
>Originator:     Falk Hueffner
>Release:        3.4 20030203 (experimental)
>Organization:
>Environment:
System: Linux juist 2.5.59 #4 Sat Jan 18 12:46:41 CET 2003 alpha unknown unknown GNU/Linux
Architecture: alpha

	
host: alphaev68-unknown-linux-gnu
build: alphaev68-unknown-linux-gnu
target: alphaev68-unknown-linux-gnu
configured with: ../configure --disable-nls --enable-languags=c++ --with-gcc-version-trigger=/src/gcc-2003.02.03/gcc/version.c --no-create --no-recursion : (reconfigured) ../configure --disable-nls --enable-languages=c++
>Description:
% cat test.cc
#include <math.h>
double f(void);
int foo() {
    double d = f();

    if (isnan(d) || (d != (int) d))
        d = 0;
    return (int) d;
}
double f(void) { return NAN; }
int main() { return foo(); }
% g++ -O2 -mcpu=ev4 test.cc && ./a.out
zsh: floating point exception (core dumped)  ./a.out

It looks like a cvttq/c was dragged out of the second condition:

[...]
        .prologue 1
        jsr $26,($27),_Z1fv             !lituse_jsr!4
        ldah $29,0($26)         !gpdisp!5
        lda $29,0($29)          !gpdisp!5
        cpys $f0,$f0,$f2
        cpys $f0,$f0,$f16
        ldq $27,__isnan($29)            !literal!2
        jsr $26,($27),__isnan           !lituse_jsr!2
        ldah $29,0($26)         !gpdisp!3
        lda $29,0($29)          !gpdisp!3
->      cvttq/c $f2,$f10
        stt $f10,16($30)
        cpys $f31,$f31,$f31
        ldq $1,16($30)
        bne $0,$L3


>How-To-Repeat:
	
>Fix:
-fno-sched-interblock works around it.
>Release-Note:
>Audit-Trail:
>Unformatted: