Comments for Andreas Gal http://andreasgal.com Project Scientist, University of California, Irvine Thu, 21 Aug 2008 00:11:57 +0000 http://wordpress.org/?v=MU Comment on Tree Folding by David Mandelin’s blog » Blog Archive » SquirrelFish http://andreasgal.com/2008/02/28/tree-folding/#comment-31 David Mandelin’s blog » Blog Archive » SquirrelFish Wed, 04 Jun 2008 01:47:03 +0000 http://andreasgal.wordpress.com/?p=53#comment-31 [...] and then reload the state on the new trace, and no cross-trace optimization is possible. Ideas like trace folding have potential for a big improvement [...] [...] and then reload the state on the new trace, and no cross-trace optimization is possible. Ideas like trace folding have potential for a big improvement [...]

]]> Comment on Tree Folding by Trace-Trees FAQ « Andreas Gal http://andreasgal.com/2008/02/28/tree-folding/#comment-30 Trace-Trees FAQ « Andreas Gal Mon, 02 Jun 2008 09:01:32 +0000 http://andreasgal.wordpress.com/?p=53#comment-30 [...] Tree Folding [...] [...] Tree Folding [...]

]]> Comment on Cell VM by Andreas http://andreasgal.com/2008/05/22/cell-vm/#comment-18 Andreas Fri, 23 May 2008 05:50:28 +0000 http://andreasgal.wordpress.com/?p=64#comment-18 Thanks. The formating of the TR on Michael's website seems to be totally screwed up. I will try to re-latex it and post a better version. Thanks. The formating of the TR on Michael’s website seems to be totally screwed up. I will try to re-latex it and post a better version.

]]> Comment on Cell VM by Taras http://andreasgal.com/2008/05/22/cell-vm/#comment-17 Taras Fri, 23 May 2008 02:31:49 +0000 http://andreasgal.wordpress.com/?p=64#comment-17 Figure 8 is upside down. Figure 8 is upside down.

]]> Comment on Recording Traces in Spidermonkey by Andreas http://andreasgal.com/2008/05/21/recording-traces-in-spidermonkey/#comment-16 Andreas Thu, 22 May 2008 17:24:45 +0000 http://andreasgal.wordpress.com/?p=56#comment-16 Yes, we definitively want to have two separate interpreter loops, one non-recording and one recording, both derived from the same source with some macro magic. There will be a small overhead in branch instructions to monitor for potential trace start points, but that kind of check is usually shadowed by the general dispatch overhead (in the JamVM tracer it was around 1%). Yes, we definitively want to have two separate interpreter loops, one non-recording and one recording, both derived from the same source with some macro magic. There will be a small overhead in branch instructions to monitor for potential trace start points, but that kind of check is usually shadowed by the general dispatch overhead (in the JamVM tracer it was around 1%).

]]> Comment on Recording Traces in Spidermonkey by shaver http://andreasgal.com/2008/05/21/recording-traces-in-spidermonkey/#comment-15 shaver Thu, 22 May 2008 12:55:59 +0000 http://andreasgal.wordpress.com/?p=56#comment-15 I think we'll want the macros to have RECORD() and maybe INTERP() wrappers around the relevant portions of their bodies, such that we can avoid are-we-tracing? check overhead in the interpreter case. With RECORD()/INTERP() wrapping, we can include the jstrace.h file twice to create two switches or jump tables. We live for major jsinterp shakeups. :) I think we’ll want the macros to have RECORD() and maybe INTERP() wrappers around the relevant portions of their bodies, such that we can avoid are-we-tracing? check overhead in the interpreter case. With RECORD()/INTERP() wrapping, we can include the jstrace.h file twice to create two switches or jump tables.

We live for major jsinterp shakeups. :)

]]> Comment on Tree Filtering & Ordering of Filters by Tree Folding « Andreas Gal http://andreasgal.com/2007/06/20/tree-filtering-ordering-of-filters/#comment-6 Tree Folding « Andreas Gal Fri, 29 Feb 2008 00:54:28 +0000 http://andreasgal.wordpress.com/2007/06/20/tree-filtering-ordering-of-filters/#comment-6 [...] to the tree. The tree is serialized into a sequence of instructions (read about tree serialization here), which is then passed through the IdentifySpliceSites and TreeFolding filters. At the end of this [...] [...] to the tree. The tree is serialized into a sequence of instructions (read about tree serialization here), which is then passed through the IdentifySpliceSites and TreeFolding filters. At the end of this [...]

]]> Comment on Linear Time Algorithm to Transform Out of SSA Form by Register Swapping vs Temporary Registers « Dr. Andreas Gal http://andreasgal.com/2007/06/16/linear-time-algorithm-to-transform-out-of-ssa-form/#comment-4 Register Swapping vs Temporary Registers « Dr. Andreas Gal Sun, 17 Jun 2007 18:46:20 +0000 http://andreasgal.wordpress.com/2007/06/16/linear-time-algorithm-to-transform-out-of-ssa-form/#comment-4 [...] Linear Time Algorithm to Transform Out of SSA Form [...] [...] Linear Time Algorithm to Transform Out of SSA Form [...]

]]>