From: Maynard J. <may...@us...> - 2008-08-05 13:31:17
|
Maynard Johnson wrote: > This patch fixes a couple different problems found by a Cell-OProfile user > relating to overlay symbols for Cell SPE applications. First, the oprofile code > has a mechanism to filter out symbols thought to be bogus, including symbols > with the same VMA. This mechanism was incorrectly filtering out many of the > overlay symbols in the user's Cell application. (Background info: The Cell SPEs > have a small main memory, so programmers use the technique of defining functions > that can reside in the same memory space at different times -- aka "overlay > symbols".) > > Second, the mechanism in oprofile for determining the size of a symbol was not > always making the correct calculation for the overlay symbols. This problem > results in mis-attribution of samples to wrong symbols AND the failure of > 'opreport -lXd' to dump the symbol contents needed by a visualization tool that > displays the disassembly view. > > These problems had not been detected in previous testing since our SPE overlay > test application was too simplistic. > > Hearing no comments to contrary, I'll commit it later this week. > Whoops! Wrong version of the patch. Correct one is attached. -Maynard > Regards, > -Maynard > |