How to use decimal floating point in Gnu C++ (g++)?

Question

GCC 4.5 added support for decimal floating points in the runtime library (http://gcc.gnu.org/gcc-4.5/changes.html). I'm able to compile code including , using namespace std::decimal, then using decimal64 and so on in the code.

Unfortunately, I'm missing some library for linking the code. I've not been able to find out which libraries should be required. DFP support is enabled in gcc (--enable-decimal-float=dpd)

Additionally, should there be some fast way of providing decimal literals in the code? By fast, I mean user defined literals that are handled by templates and translated at compile time. I do not want to provide doubles that are converted at runtime (despite performance and the fact that I really can't stand moving evaluations from compile time to runtime there are still rounding issues...). I've already found the suffix "df" but that does not seem to be recognized by the compiler.

I'm using gcc version 4.7.1 on target powerpc-ibm-aix7.1.0.0.

Linker error messages:

ld: 0711-317 ERROR: Undefined symbol: .__dpd_floatsisd
ld: 0711-317 ERROR: Undefined symbol: .__dpd_floatsidd
ld: 0711-317 ERROR: Undefined symbol: .__dpd_floatsitd
ld: 0711-317 ERROR: Undefined symbol: .__dpd_mulsd3
ld: 0711-317 ERROR: Undefined symbol: .__dpd_muldd3
ld: 0711-317 ERROR: Undefined symbol: .__dpd_multd3
ld: 0711-317 ERROR: Undefined symbol: .__dpd_floatdisd
ld: 0711-317 ERROR: Undefined symbol: .__dpd_floatunsdisd
ld: 0711-317 ERROR: Undefined symbol: .__dpd_floatdidd
ld: 0711-317 ERROR: Undefined symbol: .__dpd_floatunsdidd
ld: 0711-317 ERROR: Undefined symbol: .__dpd_floatditd
ld: 0711-317 ERROR: Undefined symbol: .__dpd_floatunsditd

TIA.

The literal suffix for `_Decimal64` is `dd` (`df` is for `_Decimal32`), but that doesn't explain the compiler not recognizing the latter (not to mention the DPD routines). Did you try passing `--enable-version-specific-runtime-libs` to the compiler? — Frédéric Hamidi, Aug 17 '12 at 08:35
Did you really mean passing that option to the compiler? It's not recognized. "g++ -v" shows that the compiler itself is configured with the option mentioned. — user1034081, Aug 17 '12 at 11:04

score 1 · Answer 1 · edited Sep 01 '12 at 11:26

I've recently compiled gcc4.7.1 from source and had trouble with missing symbol __floatunsidf. This turned out to be a symbol required by libstdc++.so and defined in libgcc_s.so. Given that both Libraries are provided by gcc they should be compatible. In my case i still had a linker include to an old version of libgcc_s that i'd lifted from my target platform. Its probably worth doing

# to find out where libstdc++.so is:
gcc -print-file-name=libstdc++.so

# see if it references the symbol you're missing
readelf -a "path to libstdc++.so" | grep "symbol name"

and then doing the same for libgcc_s to see if they both define the symbol you are missing. Nb one will mark the symbol as undefined - thats how the linker knows its required.

How to use decimal floating point in Gnu C++ (g++)?

1 Answers1