Tuesday, June 19, 2007
Random thoughts about trip day 1/2
Day 1/2 Monday/Tuesday June 18/19, 2007
Today was a long plane ride, 9.5 hours. The time difference is 16 hours so we arrived in Nartia at 2:02 or there abouts. I flew next to new parents (10 month old), he was cute and good except for once right before falling asleep where he throw a small fit. Once he was asleep he was really quiet for the rest of the journey. Got out of customs and immigration around 2:40, waited a while to get my bag. Going through customs and immigration was a breeze. Took the limousine bus to the new otani hotel. It is around a 1.5 hour ride on the highway. Nartia airport is really outside the city of Tokyo (though technically Tokyo is at least 5 different cities with their own services and such).
Tomorrow's meeting starts at 10 am and ends around lunch time.
On the pointer plus front, I merged it into the FSF mainline last friday. I have iv-opts to fix really still. I sent a patch to richi but it did not bootstrap because of problems with respect of other parts of iv-opts/address need to be fixed up. I had a full patch in the office which actually implements that but I forgot to sent it to myself to send it to richi. Anyways when I get back state side, I will be able to finish it up and test it and add a testcase or two to make sure we optimize the cases again. The problem with iv-opts right now, is that it does not rewrite pointer plus. So if we fix that, we need to also fix a couple of other parts trying to do the rewrite. Not doing the rewrite causes gzip and a couple other benchmarks to be slower. I think right now, not rewriting pointer plus, causes us to produce too many IVs.
Yes, IV-OPTs itself needs a rewrite to deal with auto-increment (load/store with update, etc.) but I am not doing to do that rewrite.
This week is going to be a busy one with meetings all day.
If I have time tonight, I will go see camel (LLVM guy studying in Tokyo). Maybe I will bring Trevor and Russ along, I will talk to them once I talk with camel.
I also hope richi and others are able to fix up the bug reports while I am in Tokyo. Most of the problems are obvious issues (except when it comes to scev but then again that was a latent bug and pointer plus just put in obvious checks to make sure we were not doing anything stupid). I still have data reference to fix up. One problem with it currently is that I need to figure out what is the real definition of affine multivariate. I would have liked the data-ref people to fix this bug as I don't understand this area of GCC that well. I did file the bug reports about these data-ref issues before doing the merge. I guess that pointer plus improves forwprop too much :).
Wow I did not think I would write this much for this blog entry but I guess I did.
PS Cassandra, the girl who I was going after, now has a boyfriend (and it is not me).
Today was a long plane ride, 9.5 hours. The time difference is 16 hours so we arrived in Nartia at 2:02 or there abouts. I flew next to new parents (10 month old), he was cute and good except for once right before falling asleep where he throw a small fit. Once he was asleep he was really quiet for the rest of the journey. Got out of customs and immigration around 2:40, waited a while to get my bag. Going through customs and immigration was a breeze. Took the limousine bus to the new otani hotel. It is around a 1.5 hour ride on the highway. Nartia airport is really outside the city of Tokyo (though technically Tokyo is at least 5 different cities with their own services and such).
Tomorrow's meeting starts at 10 am and ends around lunch time.
On the pointer plus front, I merged it into the FSF mainline last friday. I have iv-opts to fix really still. I sent a patch to richi but it did not bootstrap because of problems with respect of other parts of iv-opts/address need to be fixed up. I had a full patch in the office which actually implements that but I forgot to sent it to myself to send it to richi. Anyways when I get back state side, I will be able to finish it up and test it and add a testcase or two to make sure we optimize the cases again. The problem with iv-opts right now, is that it does not rewrite pointer plus. So if we fix that, we need to also fix a couple of other parts trying to do the rewrite. Not doing the rewrite causes gzip and a couple other benchmarks to be slower. I think right now, not rewriting pointer plus, causes us to produce too many IVs.
Yes, IV-OPTs itself needs a rewrite to deal with auto-increment (load/store with update, etc.) but I am not doing to do that rewrite.
This week is going to be a busy one with meetings all day.
If I have time tonight, I will go see camel (LLVM guy studying in Tokyo). Maybe I will bring Trevor and Russ along, I will talk to them once I talk with camel.
I also hope richi and others are able to fix up the bug reports while I am in Tokyo. Most of the problems are obvious issues (except when it comes to scev but then again that was a latent bug and pointer plus just put in obvious checks to make sure we were not doing anything stupid). I still have data reference to fix up. One problem with it currently is that I need to figure out what is the real definition of affine multivariate. I would have liked the data-ref people to fix this bug as I don't understand this area of GCC that well. I did file the bug reports about these data-ref issues before doing the merge. I guess that pointer plus improves forwprop too much :).
Wow I did not think I would write this much for this blog entry but I guess I did.
PS Cassandra, the girl who I was going after, now has a boyfriend (and it is not me).
