This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


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] RFA C4X just unconditionally sets -ffast-math (nearly)


c4x_override_options is just setting set_fast_math_flags(1), which is supposed
to have the same effect as unconditional use of -ffast-math. This is:


1. Quite surprising to the actual user and thus not polite.
2. Missing the twinkle-winkle dance between -fcx-limited-range and flag_complex_method.
3. Asking for obvious regressions.
4. Adding to the already sufficient "behavioral indeterminism" of the math options.


Is it OK to stop doing this? It's the only architecture doing this after all.

Marcin Dalecki

2006-12-29 Marcin Dalecki <martin@dalecki.de>
* config/c4x/c4x.c (c4x_override_options): Don't override optoins to set
-ffast-math.


Index: gcc/config/c4x/c4x.c
===================================================================
--- gcc/config/c4x/c4x.c        (revision 120252)
+++ gcc/config/c4x/c4x.c        (working copy)
@@ -296,9 +296,6 @@
void
c4x_override_options (void)
{
-  /* Convert foo / 8.0 into foo * 0.125, etc.  */
-  set_fast_math_flags (1);
-
   /* We should phase out the following at some stage.
      This provides compatibility with the old -mno-aliases option.  */
   if (! TARGET_ALIASES && ! flag_argument_noalias)


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