This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.
| Index Nav: | [Date Index] [Subject Index] [Author Index] [Thread Index] | |
|---|---|---|
| Message Nav: | [Date Prev] [Date Next] | [Thread Prev] [Thread Next] |
| Other format: | [Raw text] | |
Hello, Attached is the patch for variable expansion in unroller. Considering the last discussions on the register pressure utility and the current status of gcc; I resubmit this patch without the register pressure utility but with a restriction to only one expansion per accumulator to avoid register pressure. This parameter can be redefined by setting --param max-variable-expansions-in-unroller=value. The testing I made on POWER4 without the restriction on the number of expansions showed 1.2% improvement on SPECFP with 5% improvement to art benchmark. When running SPECFP (only c benchmark this time) with the restriction on the expansions (the default restriction) I got 4.8% improvement on art benchmark. I hope to post some more results later this week. I'll submit this patch once all the regression tests are done. Comments are welcome, Revital
Attachment:
ChangeLog_28_9.txt
Description: Text document
Attachment:
diff_28_9
Description: Binary data
| Index Nav: | [Date Index] [Subject Index] [Author Index] [Thread Index] | |
|---|---|---|
| Message Nav: | [Date Prev] [Date Next] | [Thread Prev] [Thread Next] |