1. 12 Mar, 2019 1 commit
  2. 04 Feb, 2019 2 commits
    • Frédéric Bonnard's avatar
      Update changelog for 1.0.13-3 release · 91e15a30
      Frédéric Bonnard authored
      91e15a30
    • Frédéric Bonnard's avatar
      Fix 2 FTBFS · f41bc20e
      Frédéric Bonnard authored
      The tests relies on python3 which is not installed by default currently.
      Also, on Ubuntu remove use -Wl,-Bsymbolic-functions in LDFLAGS which
      breaks what's expected in libdfp.so.
      
      Closes: #921119
      Thanks: Steve Langasek for the bug report, and test on Ubuntu
      f41bc20e
  3. 31 Jan, 2019 22 commits
  4. 22 Jan, 2019 1 commit
  5. 22 Oct, 2018 8 commits
  6. 01 Jul, 2016 1 commit
  7. 01 Jun, 2016 1 commit
    • Paul E. Murphy's avatar
      libdecnumber: Fix broken magic constant · 4ca2dc60
      Paul E. Murphy authored
      A bogus bit was being set in the exponent leading to strange,
      incorrect exponents.
      
      Note, BID encoded binary mantissas' always contain an
      implicit prefix.  For small values, it is always 0b0, for
      large values it is is 0b100.
      
      To signify the 0b100 prefix, the value 0b11 is appended
      after the sign bit, and the locations of the exponent is
      shifted two bits.
      
      This was noted in, and resolves Issue #49.
      Signed-off-by: 's avatarPaul E. Murphy <murphyp@linux.vnet.ibm.com>
      4ca2dc60
  8. 22 Mar, 2016 1 commit
  9. 03 Mar, 2016 3 commits
    • Paul E. Murphy's avatar
      s390: Fix soft-dfp TFmode conversions · 450e8416
      Paul E. Murphy authored
      s390 soft-dfp now passes make check. The conversions
      are implemented through strings. They are not efficient,
      but they should be correct.
      
      These override the default long double implementations in
      base-math/.
      
      A later patch should move the handling of the long double
      functions into a separate directory, much like how glibc
      handles it today.
      Signed-off-by: 's avatarPaul E. Murphy <murphyp@linux.vnet.ibm.com>
      450e8416
    • Paul E. Murphy's avatar
      s390: Properly override long double functions · fbf8f498
      Paul E. Murphy authored
      s390 hard-dfp has builtins for long double conversions,
      use them. The variants in base-math assume ibm long
      double 128, which will cause errors if ever used.
      Signed-off-by: 's avatarPaul E. Murphy <murphyp@linux.vnet.ibm.com>
      fbf8f498
    • Paul E. Murphy's avatar
      s390: Fix left_justify · 687f7a7e
      Paul E. Murphy authored
      This was very broken. After digging up the z/Architecture
      Principles of Operation, I was able to determine what was
      going on.
      
      z support is very similar to POWER, but it has a lot more
      flexibility in how things are done.
      
      Note ???x??? instructions operate on _Decimal128, and
           ???d??? operate on _Decimal64/32. Similar to POWER,
      _Decimal32 is emulated through _Decimal64 instructions.
      
      Finally, the "Round toward zero" is the constant 5 or
      9, not 1 like POWER.
      Signed-off-by: 's avatarPaul E. Murphy <murphyp@linux.vnet.ibm.com>
      687f7a7e