This is the mail archive of the libc-hacker@sourceware.org mailing list for the glibc project.

Note that libc-hacker is a closed list. You may look at the archives of this list, but subscription and posting are not open.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

[PATCH] PPC fenv fixes.


Steven Munroe wrote:
> Jakub Jelinek wrote:
>   
>> Hi!
>>
>> feholdexcept is supposed to clear the sticky exception bits:
>> "The feholdexcept function saves the current floating-point environment in the object
>> pointed to by envp, clears the floating-point status flags, and then installs a non-stop
>> (continue on floating-point exceptions) mode, if available, for all floating-point
>> exceptions."
>> and
>> "A floating-point status flag is a system variable whose value is set
>> (but never cleared) when a floating-point exception is raised, which occurs as a side effect
>> of exceptional floating-point arithmetic to provide auxiliary information."
>>
>> Several feholdexcept implementations didn't handle this right.
>> Also, e.g. feupdateenv was a stub on x86_64 eventhough it is IMHO easily
>> implementable and other arches had bugs in it too (e.g. ia64 would rise
>> wrong exceptions and always return a failure).
>> I added a new testcase for feholdexcept/feupdateenv (hopefully I haven't
>> misinterpreted the standard) and I believe other arches (ppc*, s390*, ...)
>> will need some care too.
>>   
>>     
>
> powerpc32 and powerpc64 both show two failures with the new test-fenv:
>
> Test: feholdexcept_tests FE_DIVBYZERO|FE_INVALID test
>   Fail: Exception "DIVBYZERO" is not set
>   Pass: Exception "INVALID" is set
>   Pass: Exception "INEXACT" is not set
>   Pass: Exception "UNDERFLOW" is not set
>   Pass: Exception "OVERFLOW" is not set
>
> and
>
> Test: feholdexcept_tests FE_INEXACT|FE_INVALID test
>   Pass: Exception "DIVBYZERO" is not set
>   Fail: Exception "INVALID" is not set
>   Pass: Exception "INEXACT" is set
>   Pass: Exception "UNDERFLOW" is not set
>   Pass: Exception "OVERFLOW" is not set
>
> We will look into this.
>   
feupdateenv was not merging pending and current exceptions correctly and
caused 2 failures with the new feholdexcept tests. The attached patch
fixes feupdateenv for ppc32/ppc64.


2007-03-26  Steven Munroe  <sjmunroe@us.ibm.com>

	* sysdeps/powerpc/fpu/feupdateenv.c: Change mask to merge exceptions
	from env.


diff -urN libc25-cvstip-20070320/sysdeps/powerpc/fpu/feupdateenv.c libc25/sysdeps/powerpc/fpu/feupdateenv.c
--- libc25-cvstip-20070320/sysdeps/powerpc/fpu/feupdateenv.c	2001-07-05 23:56:02.000000000 -0500
+++ libc25/sysdeps/powerpc/fpu/feupdateenv.c	2007-03-27 09:13:54.875363776 -0500
@@ -1,5 +1,5 @@
 /* Install given floating-point environment and raise exceptions.
-   Copyright (C) 1997,99,2000,01 Free Software Foundation, Inc.
+   Copyright (C) 1997,99,2000,01,07 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997.
 
@@ -30,8 +30,10 @@
   new.fenv = *envp;
   old.fenv = fegetenv_register ();
 
-  /* Copy the set exceptions from `old' to `new'.  */
-  new.l[1] = (new.l[1] & 0xE00000FF) | (old.l[1] & 0x1FFFFF00);
+  /* Restore rounding mode and exception enable from *envp and merge
+     exceptions.  Leave fraction rounded/inexact and FP result/CC bits
+     unchanged.  */
+  new.l[1] = (old.l[1] & 0x1FFFFF00) | (new.l[1] & 0x1FF80FFF);
 
   /* Atomically enable and raise (if appropriate) exceptions set in `new'. */
   fesetenv_register (new.fenv);

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]