This is the mail archive of the newlib@sourceware.org mailing list for the newlib 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]

Re: [PATCH] libm round - fix for 16-bit CPU


On 06/20/2018 12:04 PM, Jon Beniston wrote:
Hi,

The following is a patch for round on 16-bit CPUs to avoid a shift being out
of range:

libm/common/s_round.c (round): Make constant long so shift isn't out of
range on 16-bit CPUs

---
  newlib/libm/common/s_round.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/newlib/libm/common/s_round.c b/newlib/libm/common/s_round.c
index 047574a84..f0926d85f 100644
--- a/newlib/libm/common/s_round.c
+++ b/newlib/libm/common/s_round.c
@@ -68,7 +68,7 @@ SEEALSO
            msw &= 0x80000000;
            if (exponent_less_1023 == -1)
              /* Result is +1.0 or -1.0. */
-            msw |= (1023 << 20);
+            msw |= (1023L << 20);
            lsw = 0;
          }
        else
Might it be slightly better to use a cast with the type of msw? While it would probably make no difference on any 16-bit target, it could prevent an intermediate change to 64 bits instead of 32 on some targets.  So

+            msw |= (((__int32_t)1023 << 20);
(Or, better for maintenance, but not as portable with a GCC extension:

+            msw |= (((typeof(msw))1023 << 20);
(There do seem to be multiple uses of typeof already, although the first few I checked are all gated with validity checks.  So probably not appropriate as shown here.))

Craig


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