diff options
| author | David Cournapeau <cournape@gmail.com> | 2008-09-20 16:26:30 +0000 |
|---|---|---|
| committer | David Cournapeau <cournape@gmail.com> | 2008-09-20 16:26:30 +0000 |
| commit | c372fc1bad92634ae05224263fa56c86fae9b380 (patch) | |
| tree | 144ab6d7f9d53058d075953932660cedb6c5527c /numpy | |
| parent | bc45f6bf99ae12105f19bc7ca00469ffb9bd60f3 (diff) | |
| download | numpy-c372fc1bad92634ae05224263fa56c86fae9b380.tar.gz | |
Fix for isfinite on VS 2003.
VS 2003 seems to think it is ok to simplify x-x to 0 for float, but this is
wrong for NaN and Inf. To alleviate, we force the operation to occur with (x)
+ (-x).
Diffstat (limited to 'numpy')
| -rw-r--r-- | numpy/core/src/math_c99.inc.src | 4 |
1 files changed, 3 insertions, 1 deletions
diff --git a/numpy/core/src/math_c99.inc.src b/numpy/core/src/math_c99.inc.src index 542500d49..14399f699 100644 --- a/numpy/core/src/math_c99.inc.src +++ b/numpy/core/src/math_c99.inc.src @@ -139,8 +139,10 @@ double trunc(double x) # define isnan(x) ((x) != (x)) #endif +/* VS 2003 with /Ox optimizes (x)-(x) to 0, which is not IEEE compliant. So we + * force (x) + (-x), which seems to work. */ #if !defined(HAVE_DECL_ISFINITE) - # define isfinite(x) !isnan((x) - (x)) + # define isfinite(x) !isnan((x) + (-x)) #endif #if !defined(HAVE_DECL_ISINF) |