[Libreoffice-commits] core.git: sal/rtl

[Eike Rathke (via logerrit)]

 sal/rtl/math.cxx |    7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

New commits:
commit a10c33fdbe980effc3a14e773d1b94a14be7d428
Author:     Eike Rathke <erack at redhat.com>
AuthorDate: Thu Dec 17 22:02:48 2020 +0100
Commit:     Eike Rathke <erack at redhat.com>
CommitDate: Fri Dec 18 03:17:03 2020 +0100

    Replace log2() call with parts.exponent-1023, tdf#138360 follow-up
    
    ... to save some cycles as we anyway need only the integer value
    of the exponent and even exactly this value for the number of
    possible decimals.
    
    Change-Id: I8d462f53cadde6a95d57d1342d8487fbfa001ae9
    Reviewed-on: https://gerrit.libreoffice.org/c/core/+/107928
    Tested-by: Jenkins
    Reviewed-by: Eike Rathke <erack at redhat.com>

diff --git a/sal/rtl/math.cxx b/sal/rtl/math.cxx
index 46a3e925b95b..527b508e848c 100644
--- a/sal/rtl/math.cxx
+++ b/sal/rtl/math.cxx
@@ -1167,9 +1167,10 @@ double SAL_CALL rtl_math_round(double fValue, int nDecPlaces,
             // Determine how many decimals are representable in the precision.
             // Anything greater 2^52 and 0.0 was already ruled out above.
             // Theoretically 0.5, 0.25, 0.125, 0.0625, 0.03125, ...
-            const double fDec = 52 - log2(fValue) + 1;
-            if (fDec < nDecPlaces)
-                nDecPlaces = static_cast<sal_Int32>(fDec);
+            const sal_math_Double* pd = reinterpret_cast<const sal_math_Double*>(&fValue);
+            const sal_Int32 nDec = 52 - (pd->parts.exponent - 1023);
+            if (nDec < nDecPlaces)
+                nDecPlaces = nDec;
         }
 
         /* TODO: this was without the inverse factor and determining max