Christopher Collins created MYNEWT-746:
------------------------------------------
Summary: Sim - Floating point calculations sometimes incorrect
Key: MYNEWT-746
URL: https://issues.apache.org/jira/browse/MYNEWT-746
Project: Mynewt
Issue Type: Bug
Reporter: Christopher Collins
Fix For: v1_1_0_rel
*Note: this problem does not occur in OS X; it has only been seen in Linux.*
In sim, some floating point calculations yield wildly incorrect results
(usually 0 or NaN). The cause appears to be that the floating point registers
are not restored when longjmp() is called. If subsequent code tries to use an
intermediate value stored in a floating point register, the result is
indeterminate.
There is a gcc / clang compiler option which mostly solves this problem:
*-ffloat-store*:
{quote}
-ffloat-store
Do not store floating point variables in registers, and inhibit other
options that might change whether a floating point value is taken from a
register or memory.
This option prevents undesirable excess precision on machines such as the
68000 where the floating registers (of the 68881) keep more precision than a
double is supposed to have. Similarly for the x86 architecture. For most
programs, the excess precision does only good, but a few programs rely on the
precise definition of IEEE floating point. Use -ffloat-store for such programs,
after modifying them to store all pertinent intermediate computations into
variables.
{quote}
The problems disappeared when I started using this option. However, it seems
there is still the occasional miscalculation, as a test failure just occurred
due to an incorrect floating point value. In this case, assigning {{18.0}} to
an int yielded the value {{-2147483648}}.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
