Thursday 19th May 2011
This month's meetings
<< < 2011 / 5 > >>
- Standup call
- Sent patches for SMS upstream as follow up to the Linaro GCC merged one
- Patches improve code size
- Looking into oppertunities in SPEC and others
- Committed the approved patches to the branches
Has installed the Linaro Android image on his PandaBoard
- Installed SDK, NDK, etc to try some of the examples
- To look into the trace utility that Jim mentioned
- Trying to measure -funwind-table overhead. Want to measure C++ and C applications
- Perhaps try Python and Firefox? xdeb?
- Just returned!
- NEON DImode constant patch - two different versions, one in CSL, one in trunk
- Should we take all of the tree of patches?
- Closing off previous cycle's blueprints, planning for this cycle
- Remaining is floating point exception flags. Easy are fixed, rest are hard
- Will send out patch set for the easy and postpone the less important/hard ones
- Fleshing out the QEMU blueprint page
- Committed reduction support in SOP upstream
- Continued looking into different benchmarks such as FFMEG, EEMBC, pixman
- Looking into implementation of reverse access
- ...and potential engineering blueprints
- Various reviews as he's on the review roster
- ...and Yao's patchset for GDB displaced stepping
- Still has issues so unfortunately not in this release. Last issue remaining...
- Ran the release process. Has a clean testsuite except for displaced stepping
- Sent latrace some patches for ARM
- Looked at ltrace as it was hanging on a particular ptrace call
- Added the stub for ARM and it seems to work for a single process
- Optimising unaligned memcpy(). Some nasty cases to investigate.
- Such as if source misaligned by 12, dest by 8: hurts
- Running the SPEC latrace through analysis tools
- Some such as tonto have many libm calls that were hurting the trace
- Checked that the binutils errata fix fixes apt/GTK error
- Looking into the vmov.i64 truncation problem
- Looks like Natty is clear as it's not built in the 'has 64 bit extra target' option
- Looked at the NEON vs non-NEON GCC testsuite results
- With NEON on there are some extra ICEs - will look into them
- Michael: doesn't think they're regressions so OK for now
- Stubbed out blueprints and delegated
- Set up to use Ubuntu as a testsuite
- Ran the release process
- Shifted the meeting
Trying to sort out 16x16->64 bit multiply patch
- Complicated as there's a bootstrap failure
- Bernd re: shrink-wrap: no change from Richard's patch. Not working there at the moment.
WorkingGroups/ToolChain/Meetings/Archive/2011-05-19 (last modified 2013-08-30 11:48:27)