|
From: Johan B. <jb...@gm...> - 2009-06-25 07:20:13
|
Hi Julian, I am currently working on my masters degree and no longer at Palm, so I can't comment on what the latest status is. Before I left, I implemented VFP and we used it on VFP enabled builds with no problems. At one point I was talking to Julian about getting the work merged into the mainline, and sent him the updated patch that included VFP, but I didn't hear anything back after lawyers got involved. :( Julian S, do you have any update what happened? Is the best thing right now that a third party rewrites all the ARM vex code? -Johan On Wed, Jun 24, 2009 at 7:26 PM, Julian Brown<ju...@co...> wrote: > Hi, > > I'm planning to spend the next month or so working on the ARM port of > Valgrind (by Evan Geller, Johan Björk, etc.). So far I have managed to > update the latest patch from here: > > http://bugs.kde.org/show_bug.cgi?id=126449 > > to (almost) the latest SVN revision, although the merge from the > BUILD_TWEAKS branch has broken things for the time being. > > My aim is to get Valgrind's internal tests running smoothly, and then > (probably) to try to get GCC's testsuite to run smoothly also (with > cross-compilation, but running test binaries on real hardware under > Valgrind/Nulgrind). I might have a go at VFP support too, and/or > adding any missing instructions I find. > > I have a couple of questions: firstly, how actively are folks working > on the ARM patch at the moment? (I don't want to duplicate work > unnecessarily). Secondly, is the ARM patch likely to be merged to > Valgrind trunk (or otherwise) any time soon? (I'm kind of guessing the > recent configuration system cleanups are part of a wider plan to ease > cross-platform support). Thirdly, are there any more major > restructuring plans on the horizon? > > Are Johan's notes about his last patch drop: > >> * No VFP/Neon support (VEX will panic if it hits a VFP DPI >> instruction. >> * Stacktraces are not always complete, there is probably more >> improvement possible in this area >> * There are probably a few more lingering issues in instruction >> decoding, vcp_fnfns fail on the last decimal at a few of the tests >> (with soft-floating point, so it should not happen), as well as I've >> seen some apps that fail even when used with --tool=none. Pretty rare >> though. > > All still true? > > Thanks, > > Julian > |