added section "Problems due to compiler bugs"

git-svn-id: svn://scm.gforge.inria.fr/svn/mpfr/trunk@2489 280ebfd0-de03-0410-8827-d642c229c3f4

1 files changed, 23 insertions, 0 deletions

diff --git a/BUGS b/BUGS
index 94bdfaf68..5295f380a 100644
--- a/BUGS
+++ b/BUGS
@@ -52,3 +52,26 @@ Potential bugs:
 
 * mpfr_hypot may fail for x very large, y very small and a very large
   target precision. Other functions may be affected by similar problems.
+
+Problems due to compiler bugs:
+
+* on some architectures (for example alpha-dec-osf), gcc 3.3 wrongly
+  compares "long double" floating-point numbers, with optimization level
+  1 or higher. This bug can be detected by the following program:
+
+  #include <stdio.h>
+  #include <stdlib.h>
+  #include <float.h>
+  int
+  main ()
+  {
+    long double d;
+    d = 1.0; while (d < LDBL_MAX / 2.0) d += d;
+    if (d == (long double) 0.0)
+      printf ("d equals 0.0\n");
+  }
+
+  This results in a problem in the mpfr_set_ld function. A workaround is
+  to compile set_ld.c with -O0 (no optimization).
+
+