INSTALL: added a paragraph about -ffast-math / -fast compiler options.

git-svn-id: svn://scm.gforge.inria.fr/svn/mpfr/trunk@5351 280ebfd0-de03-0410-8827-d642c229c3f4

1 files changed, 6 insertions, 0 deletions

diff --git a/INSTALL b/INSTALL
index e73a5871e..05cd2b11d 100644
--- a/INSTALL
+++ b/INSTALL
@@ -151,6 +151,12 @@ and continue the install. On some platforms, you should provide further
 options to match those used by GMP, or set some environment variables.
 For instance, see the "Notes on AIX/PowerPC" section below.
 
+Warning! Do NOT use optimization options that can change the semantics
+of math operations, such as GCC's -ffast-math or Sun CC's -fast.
+Otherwise conversions from/to double's may be incorrect on infinities
+and NaN's. Since native FP arithmetic is used in a few places only,
+such options would not make MPFR faster anyway.
+
 On some platforms, try with "gmake" (GNU make) instead of "make".
 Problems have been reported with the Tru64 make.