My MOVW patch from last week left an unused variable that killed -Werror builds, so I posted a new patch to fix it. http://firstname.lastname@example.org/msg04141.html Ramana approved that in record time, so I've committed it.
- Tom has committed Mark Shinwell's BRANCH_COST patch upstream, so I have backported it to Linaro GCC 4.6.
- Maxim has committed his compound conditionals patch upstream, so I've backported that to Linaro GCC 4.6 also.
- Reviewed what other patches are queued for forward porting to 4.6/7, hoping that other's might have picked some of them off, and found no other progress just yet.
- Reduced the test case for GCC PR48783. For some reason it is retaining ".global" directives for symbols that have been optimized away, which leads to link errors in the kernel build.
- Tried to find out why the "discourage neon in A8" patch has caused test failures. I'm still not sure - I've asked Michael Hope to rerun the tests in case it's just random.
Reposted the EABI half-precision patch with the extra bits Joseph says it needed in the version scripts. http://email@example.com/msg04251.html
- Looked at the linking problem reported by Barry Song on the linaro-toolchain list.
- Public holidays on Monday and Friday.
- I will be attending UDS in Budapest from 8th - 14th May. I shall continue to read my email, but will not be attending any calls.
CoreMark ARMv6/v7 regression patch set for combine: exchanged some comments on upstream list with Jeff Law. Started some testing onpowerpc64, hit some issues with native bootstrap that I don't think I completely overcome, but at least completed a 32-bit bootstrap successfully. Unfortunately hit two regressions after running tests, will work on fixing this week, and update the upstream submission.
PR48783: unused reference to aeabi_uldivmod. Investigated and found this to be a case where during RTL expand, .global directives for the aeabi_* runtime symbols are directly inserted after some div optab didn't have the insns, and a libfunc was to be used. However later optimizations that removed the divisions left the .global symbols dangling with no uses. We could probably just avoid adding the directives completely for the EABI functions.
- PR42017: ARM LR register not used in leaf functions. This is a case where DF takes EPILOGUE_USES as part of the initial live registers at end of functionn, and liveness computations never get to kill it, since there are no function calls to clobber in a leaf function. Submitted a patch to remove LR from EPILOGUE_USES before reload, awaiting review.
- LP #689887: investigated more; not much progress, but found that the bootstrap fail happens only under --with-mode=thumb (Thumb-2). Under ARM mode the native bootstrap succeeds.
Try to fix the problems with the CoreMark combine patches.
- Investigate more optimizations/bugs.
- Prepare for travel to Budapest on Saturday.
- if-conversion improvement (needed to vectorize Telecom/viterbi) - merged
- auto-detection of vector size - proposed for merging
- fixed PR 48765 on trunk
- completed testig of vzip fix (PR 48252) on gcc 4.5 and gcc 4.6
- the initial support for resuming at a certain stack frame went upstream
- learned about the various signal frame layouts on ARM
- RT frames, non-RT frames for present and pre 2.6.18 kernels
- implemented support for RT signal frame detection on ARM
- started to implement support for unw_resume if signal frames are involved
- attended a class on Friday
- Continued analysing DENbench benchmarks.
- Discussed the SMS patched with Ayal Zaks (SMS maintainer).
- Booked the flights for Budapest summit.
- Committed series of patches to fix bug #615978 (Failure to software single-step into signal handler) to mainline and Linaro GDB.
- Drafted blueprint linaro-toolchain-o-cross-debug
- Split bug #771900 (Linaro GCC 4.5 switch optimization breaks profiled bootstrap) off bug #759409 (Profiled bootstrap fails in GCC 4.5). Tracked down root cause of bug #759409 / PR middle-end/43085, and posted fix to gcc-patches.
WorkingGroups/ToolChain/ActivityReports/2011-04-29 (last modified 2011-05-05 14:42:28)