Appendix A – Library performance

As a result, the module has significant overheads compared to the dedicated decFloats modules which work directly on the fixed-size encodings. This appendix compares the performance of the decNumber module with the decDouble and decQuad implementations of the same operations. As the tables below show, there is a significant performance advantage in using the decFloats modules when arbitrary-precision operations are not required.

Description of the tables

Throughout the tables, worst-case cycle times are shown for the main operations in the decDouble and decQuad modules, compared with the same operations using the decNumber module (which requires conversion of operands and results).

Worst-case timings are quoted because best-case timings are generally trivial special cases (such as NaN arguments) and ‘typical’ instruction mixes are very application-dependent.

For each operation, the name of the operation is given, along with a brief description of the worst-case form of the operation. This is the worst case for the decFloats module (in some cases the worst case is different for the decNumber module).

decDouble performance tables

decQuad performance tables

Notes

1. All timings were made on an IBM X41T Tablet PC (Pentium M, 1.5GHz, 1.5GB RAM) under Windows XP Tablet Edition with SP2; the modules were compiled using GCC version 3.4.4 with optimization settings -O3 -march=i686.
2. The default tuning parameters were used (DECUSE64=1, DECDPUN=3, etc.); some of these only affect decNumber.
3. Timings include call/return overhead, and for the decNumber module also include the costs of converting operand(s) to decNumbers and results back to the appropriate format using the decimal64 or decimal128 module.
4. ‘BCD’ for decNumber is Packed BCD, using the decPacked module; for decFloats it is 8-bit BCD.
5. The worst case for each operation is not always obvious from the code and is implementation-dependent (for example, in the decFloats modules, an unaligned add is sometimes faster than an aligned add). It is possible that there may be unusual cases which are slower than the decFloats counts listed above, although a wide variety of micro-benchmarks have been tried.
6. A string-to-number conversion can theoretically have an arbitrarily large worst case as the string could contain any number of leading, trailing, or embedded zeros; the timings above measured cases where the input string’s coefficient had up to eight more digits than the precision of the destination format.