Ra NEWS ------- *** ra_1.3.1 Aug 26, 2011 Fixed a NAMED handling bug when assigning non-jitted symbols to jitted symbols. See eval1.c:assignToSymbol(). We now trace jit machine execution with jit(jit=3, trace=4). We used to do this when trace=3. The files printarray.c and printvector.c are now modified slightly for the Ra build (to minimize noise when tracing jit machine execution. Added Ross Ihaka's Otago talk example code to time-jit-lib.R. *** ra_1.3.0 Aug 09, 2011 Ported to R 2.13.1. Please use the jit package version 1.3-0 (or greater) with this version of Ra. For speed, the jitter now runs with many of the previous debugging sanity checks disabled. To run with the sanity checks enabled (mostly for my use during development) use jit(jit=3). Also, you can now use jit(jit=3, trace=3) to trace jitted code execution. However, even with the above change, jitted code is about as fast as it used to be (not faster). Possibly because of non-Ra related changes in memory.c and envir.c (maybe in CHECK_OLD_TO_NEW?). Win 64-bit Ra builds have now been fully tested. Certain code changes made between R 2.9.2 and R 2.13.1 were not fully incorporated (viz. changes to do_for and source references): The current implementation of "if", "while", and "for" are closer to R 2.9.2 (which was the base code for the previous version of Ra) rather than R 2.13.1. Changes were made to these functions in the R sources, and it I haven't figured out yet exactly why these changes were made. Source references are not fully incorporated into the current version of Ra (but also not yet properly implemented in standard R?). The previous Ra modified files in utils: example.R, help.R, and help.search.R. (These changes were kindly submitted by Ben Goodrich.) These files were not modified for this version of Ra, because the code in these files was changed (between R 2.9.2 and 2.13.1) in a way that I don't (yet) understand. This means that help for utils and "make check-all" may run into problems here under certain versions of unix (they work fine under Windows). The R Win 64 bit build gives incorrect results when adding NAs to NaNs (in the standard build, as well as the Ra build). Reported as bug number TODO. As a consequence, test-jit.bat lists some NA/NaN differences for f261. These should be inconsequential (unless the difference between a real NaN and NA is important in your code). The standard Win R build hangs in no-segfault.R (usually, not always). Therefore to allow easy testing of the Ra build, I modified no-segfault.Rin and created no-segfault.R. All code changes to existing files are now marked with "RAJIT" to make integrating into new versions of R easier. Ugly but effective. This does not apply to eval*.c, which must be merged completely by hand. *** ra_1.2.8 Sep 18, 2009 Ported to R 2.9.2 Had to hack tests/primitives.R to work around effect of changes to S4 in R 2.9.2 (I need to look into a better way to do fix that, perhaps by modifying methods/R/makeBasicFuncsList.R). *** ra_1.2.7 May 11, 2009 Removed duplicate filenames in main/Makefile.in. Thanks to Dirk Eddelbuettel for picking this up. help() and example() now work correctly for Ra under linux and friends. Thanks to Ben Goodrich for this fix. *** ra_1.2.5 May 05, 2009 Ported to R 2.9.0 *** ra_1.2.4 Jan 20, 2009 Fixed tests and docs to deal with non-propagation of NaNs (they get converted to NAs in certain arithmetic expressions on some architectures). *** ra_1.2.3 Jan 5, 2009 Incorporated the new version of the jit library. *** ra_1.2.2 Jan 5, 2009 Incorporated the new version of the jit library. Clarified warning text issued by nestedJitWarning, this seems to be a common user mistake. *** ra_1.2.1 Jan 3, 2009 We now correctly don't compile symbols with a dim attribute, so matrix arithmetic now works correctly (expressions like x <- y where y was a matrix used to result in an x without a dim attribute). We now run on 64 bit machines (but haven't yet run the full Ra test suite). Ported to R 2.8.1 *** ra_1.2.0 Oct 24, 2008 Added support for rhs matrix subscripts x[i,j]. Jitted loop body x[i,1] takes 0.13 the time. Ported to R 2.8.0 *** ra_1.1.2 Sep 3, 2008 Ported to R 2.7.2 *** ra_1.1.1 Jul 15, 2008 Ported to R 2.7.1 *** ra_1.1.0 May 29, 2008 Fixed incorrect "istack=0" in jitContextChangedAux (bug reported by Ben Goodrich) Removed code causing warnings in test-jit.R and friends (thanks to Dirk Eddelbuettel) Improved windows installer for non english locales (thanks to Søren Højsgaard) Fixed string overflow in deparseAsShortString A few other tiny tweaks to code and documentation *** ra_1.0.9 Apr 24, 2008 Added base/R/zzz.R inadvertently left out of previous build *** ra_1.0.8 Apr 23, 2008 Ported to R 2.7.0 *** ra_1.0.7 Apr 03, 2008 Code and doc touchups *** ra_1.0.6 Mar 10, 2008 Code and doc touchups *** ra_1.0.5 Mar 6, 2008 Optimized standard R loops. Jitted code is now much faster. Added jit(2) support, jitProlog can now interrupt another jitProlog. Added basic NAMED handling to jitted asignments. Added Windows installer. Ported to R 2.6.2 *** ra-04 Jan 26, 2008 We now jit unary minus, basic subscripted assigns, and %% and %/% ops. The jitter state is now handled with a state machine instead of several flags. It's a much cleaner approach. Added three sections to the "implementation notes" doc: . Wouldn't it be simpler to add declarations to R instead of a jitter? . Speed versus C . About compiling code that is not in loops Numerous doc and code fixes. There will continue to be quite a lot of code churn for some time. Added some timing results to the web page. *** ra-03 Jan 15, 2008 Initial release