Release 5.5.0.0 of FDPR-Pro:
Release 5.5.0.0 provides the following changes:
Support for SLES11 and GCC4.4 code.
SuSE SLES11 and GCC4.4 were added to the list of supported OS/compiler versions:
- Redhat EL3 to EL5
- SuSe SLES9 to SLES11
- GCC 3.3 to GCC4.4
- XLC 7.0 to XLC9.0
Call value profiling (-ocvp).
The new optimization performs code specialization according to function call values.
To enable this optimization the flag -icvp shoule be added in the instrumentation phase. Note: this is an initial version of the optimization.
DERAT optimization (-oderat).
The new optimization fixes possible improper order of the index and based register in indexed memory access. In some processors, e.g., Power6, a wrong order can cause severe performance penalty. This occurs in particular in GCC 4.1 code. The optimization is performed automatically when -O3 is selected.
Optimizing FDPRPRO processing
Various algorithms were imroved inside FDPRPRO to speed up its analysis and internal representation build-up phases.
Other changes:
- Support for XLC IPA-@glue functions
- Enablement of inlining of functions that contains branch tables.
- Branch prediction optimization: unset all existing prediction bits (if any).
