Older blog entries for hubicka (starting at number 8)

forgot to mention. I got accepted as postgradual student today too.

Lots of GCC fun recent days. SOme interesting reload bugs are gone, other appeared, x86-64 port for 3.1.1 is getting stabilized (under controll?). Most interesting is mainline development. Today I got approval for dominance changes, so next cfg-branch merger will be nightmare again. RIchard appears to like the unroller bits as well. In case we do it next 2 days, real majority of cfg branch would get into the mainline that is great luck.

I've also found new proof of universality of H in my paper and scheduled holidays.

Lots of fun, not enought time to write diarry this week. Still I am fixing important bugs in the SSE/MMX/x86-64 support for gcc 3.1, but had some time for mainline development too. I finally have the new datastructure for dominators, so I believe everything is in the place for mainline merigng of the new loop optimizer, but I am not quite sure whether we can do it, as I am leaving in one week, but lets have fun.

Second day spend with GCC, good. I was missing it :) Richard has OKayed Zdenek's loop patch and reviewed mines tracer, so I have something to play with in the mainline.

I've also merged mainline to the branch and it went relatively well, so now I can post the double test pass, as Robert seems to be touching same problems but from frontend side.

Also I should probably attempt to push in webizer once tracer is OK, what I hope to happen today.

OK, I've passed the finals, so now it is time for GCC hacking. Lets start with the boring stuff I was delaying for a while. Hope to have ready patch for scope regions and BB numbering today

Last day of studying and not studying, hope everything will go well tomorrow. Zdenek has finished the merging of renumbering patches, so it is time to synchronize cfg branch and start merging of other patches to the mainline. I've sent another version of bb-reordering patch to Andreas, hope this time it will bring improvements. For CFG branch it certainly does. Otherwise not much time for fun because of lack of time today.

OK, some results from Andreas came. The 3rd liveness pass costs about 1% of compilation time and brings noticeable speedups to C++ code only (eon, stephanov) about 1%. We definitly need to replace heuiristics in GCC by proper analyzers, so perhaps we can make people to wait 1% longer, or we can speed up liveness. I've posted question about it. If converison adds about 3 points to specs, this is good comparisons to 2-3 points brought by double test pass. Lets re-measure it, it can be interesting to see whtehr bootstrap time changes. GCC has plenty of oppurtunities for double test converison. So for now I will study and see whether someone repsponds to the posts. Would be cool to measure behaviour on non-x86 platform. Need to set up something...

26 May 2002 (updated 26 May 2002 at 11:52 UTC) »

OK, lets see how I like the idea of keeping diary. Yesterday has passed by studying for exams, today I should as well, but again I am not :) I've looked at RTH's thread local implemetnation. It looks usefull to get lower overhead for Zdenek's thread safe profiling patch, but still the patch has large overhead (for mozilla that can be about 8MB per thread, nasty). Perhaps I can copy the data on demand, but that does increase the overhead. I will send email about that. RTH didn't reviewed the tracer patch yet, so I probably won't prepare new ones. Andreas will benchmark the third liveness pass. It is badly needed, but it will hurt performance, since our liveness pass is lame. Perhaps time to rewrite it properly using bitmaps and daniels dataflow routines.

Hmm, I should be preparing for finals, but I am not. Instead I am browsing web after a while and decided to create account...

New Advogato Features

New HTML Parser: The long-awaited libxml2 based HTML parser code is live. It needs further work but already handles most markup better than the original parser.

Keep up with the latest Advogato features by reading the Advogato status blog.

If you're a C programmer with some spare time, take a look at the mod_virgule project page and help us with one of the tasks on the ToDo list!