adash [Fri, 15 May 2009 00:37:00 +0000 (00:37 +0000)]
new files
jjenista [Thu, 14 May 2009 22:34:21 +0000 (22:34 +0000)]
use sese pretty id and id to generate method names, param struct and local struct that is unique--keep a mapping of seses to the bogus flat method for retrieval in order to reproduce those names at call sites. Also some bug fixes
jjenista [Thu, 14 May 2009 21:37:42 +0000 (21:37 +0000)]
Lots of changes that aren't working yet, so new methods are commented and existing code altered for new stuff is checked in. Won't affect other compiler modes, and checks in stuff to reduce conflicts while getting the sese's working.
bdemsky [Thu, 14 May 2009 21:36:53 +0000 (21:36 +0000)]
changes
bdemsky [Thu, 14 May 2009 20:18:07 +0000 (20:18 +0000)]
more efficiency hacks...we're looking fields for values we already know, and doing arithmetic just to undo it..
bdemsky [Thu, 14 May 2009 02:53:52 +0000 (02:53 +0000)]
more changes...fix soft abort case...it needs to let other transactions through...
bdemsky [Thu, 14 May 2009 02:28:02 +0000 (02:28 +0000)]
change
adash [Thu, 14 May 2009 01:35:52 +0000 (01:35 +0000)]
Add new files
bdemsky [Thu, 14 May 2009 01:06:14 +0000 (01:06 +0000)]
slightly better locks for stm...
jzhou [Wed, 13 May 2009 23:14:44 +0000 (23:14 +0000)]
Updates to support multicore version runtime of Tilera. Also fix a bug in BuildCodeMulticore.java: it faulty omitted codes which repeatedly transfer objects to the same core. In current scheduling, to send 4 objs to 2 cores, it could be specified as an array [0,0,1,1] indicating the target cores for each object. Originaly the array was generated as [0,0,1,null] which caused missing of objects.
bdemsky [Wed, 13 May 2009 08:05:01 +0000 (08:05 +0000)]
make atomic work in normal java mode...it just uses coarse locks
bdemsky [Wed, 13 May 2009 03:06:41 +0000 (03:06 +0000)]
changes
bdemsky [Wed, 13 May 2009 02:13:05 +0000 (02:13 +0000)]
various bug fixes
adash [Wed, 13 May 2009 01:50:19 +0000 (01:50 +0000)]
add new files
bdemsky [Wed, 13 May 2009 01:27:33 +0000 (01:27 +0000)]
changes to compile
bdemsky [Wed, 13 May 2009 01:22:50 +0000 (01:22 +0000)]
try to simplify garbage collector lock
bdemsky [Wed, 13 May 2009 00:11:58 +0000 (00:11 +0000)]
changes to build slightly more optimal code
bdemsky [Tue, 12 May 2009 21:47:46 +0000 (21:47 +0000)]
changes:
1) bug fix
2) support for fast memcpy...contact me for objects
bdemsky [Tue, 12 May 2009 03:28:59 +0000 (03:28 +0000)]
small changes
plus bugfixes...one race condition on riskyflag vs lock fixed
adash [Tue, 12 May 2009 00:41:54 +0000 (00:41 +0000)]
bug fix: acquire lock before copying the object into cache
adash [Mon, 11 May 2009 23:09:36 +0000 (23:09 +0000)]
bug fix: changes to make things safe for the GC
jjenista [Mon, 11 May 2009 22:41:22 +0000 (22:41 +0000)]
Good cut of variable analysis
jjenista [Mon, 11 May 2009 18:04:32 +0000 (18:04 +0000)]
When we stall for some specific SESE, make other variables that come from that SESE available as well
jjenista [Fri, 8 May 2009 22:38:35 +0000 (22:38 +0000)]
Change isAvailable to notAvailable, updated computations for that analysis, use those results to prune waitSet for stalls--looks good\!
jjenista [Fri, 8 May 2009 17:56:23 +0000 (17:56 +0000)]
compute when to write dynamic address of variables
bdemsky [Fri, 8 May 2009 05:39:45 +0000 (05:39 +0000)]
small changes
adash [Fri, 8 May 2009 01:27:04 +0000 (01:27 +0000)]
Add new benchmark:....some prelimnary files
jzhou [Thu, 7 May 2009 21:48:55 +0000 (21:48 +0000)]
updates on Math operations for multicore version
bdemsky [Thu, 7 May 2009 21:38:37 +0000 (21:38 +0000)]
small changes
adash [Thu, 7 May 2009 19:35:10 +0000 (19:35 +0000)]
throw out commented code
adash [Thu, 7 May 2009 19:26:22 +0000 (19:26 +0000)]
race condition fixed: race due to thread rec becoming NULL before accessing its field/element
jzhou [Thu, 7 May 2009 18:35:05 +0000 (18:35 +0000)]
commented out all functions by mistake
jjenista [Thu, 7 May 2009 18:28:08 +0000 (18:28 +0000)]
Curious. Found a bug in liveness where liveIn set becomes empty erroneously. While tracing bug back the output started producing the correct liveIn set, but the code diff confirms to me that I haven't made a functional change to the system. I've been trying to understand what happened, but for now I'll check in this working, stable capture and keep an eye on liveness results in case the bug is just latent now.
jzhou [Thu, 7 May 2009 18:28:06 +0000 (18:28 +0000)]
updates for multicore version runtime, temporarily disable math operations in multicore version
jjenista [Wed, 6 May 2009 23:08:02 +0000 (23:08 +0000)]
add frame for another pass to compute whether variables are available, then use to do computation for stalls
adash [Wed, 6 May 2009 21:06:38 +0000 (21:06 +0000)]
locking technique for risky objects enabled
jjenista [Wed, 6 May 2009 18:27:25 +0000 (18:27 +0000)]
Add CodePlan object for injecting code before or after a FlatNode
jjenista [Tue, 5 May 2009 22:08:19 +0000 (22:08 +0000)]
Several bug fixes. Now that tokens have a mutable element, reference var sets, they must be deep copied when merging table to avoid altering two tables with one operation. A bug in removal by temp prevented sese-var mapping from being updated also. A change in table implementation makes a difference between public and private ops like add and remove. Any public op calls the private version and then asserts consistency. Private versions can be called in more complex operatios and don't necessarily have to leave the table consistent until the bigger op is finished. Much easier to find source of bugs this way.
jjenista [Mon, 4 May 2009 22:24:19 +0000 (22:24 +0000)]
This is BANANAS. How can a table that calls assertConsistency all right before being put in a hashtable come out and fail the call? Anyway, this is a stable compile but still a crash, but I fixed a get operation that was actually modifying the table while retrieving data.
jjenista [Mon, 4 May 2009 21:04:04 +0000 (21:04 +0000)]
Why did last checkin of VarSrcTokTable add carot-M to every line end but the whole file didn't show up in the cvs diff--anyway, looks like everything is touched but actually just taking that line character out and adding section to assertConsistency that makes sure reference var sets are consistent as well.
adash [Mon, 4 May 2009 19:21:12 +0000 (19:21 +0000)]
do a make tabbing
protect stats insde bounds
add function to stm for instrumentation ---- still working on locks
jjenista [Mon, 4 May 2009 17:29:24 +0000 (17:29 +0000)]
convert token tables into triples that sets of reference temps map to, instead of quadruples that duplicate data. Moved table consistency checks from analysis level to the exit of table-modification methods. Note: this capture compiles but fails consistency checks\!
jjenista [Thu, 30 Apr 2009 21:48:29 +0000 (21:48 +0000)]
preliminary stall computation working
jjenista [Thu, 30 Apr 2009 21:25:17 +0000 (21:25 +0000)]
fixed liveness to calculate live-out set, and a couple other bugs
jjenista [Wed, 29 Apr 2009 23:59:09 +0000 (23:59 +0000)]
stable, something still wrong with calculating out-sets
jjenista [Wed, 29 Apr 2009 21:51:12 +0000 (21:51 +0000)]
stable capture, bug fixes--a problem: using an SESEs liveIn set liberally, some cases call for using an SESEs liveOut instead, I think, which is at least part of why no stalls are reported
jjenista [Wed, 29 Apr 2009 17:47:24 +0000 (17:47 +0000)]
Reinitialize liveness for the second pass, uses virtual reads to compute new answer. The combination of liveness results and variable analysis still generates no stalls
afedward [Wed, 29 Apr 2009 00:05:01 +0000 (00:05 +0000)]
Updating CVS with partially ported benchmark. Sequencer.java's run() function is the last function to be ported, then compilation and testing.
jjenista [Tue, 28 Apr 2009 23:01:35 +0000 (23:01 +0000)]
Something is wrong with liveness results that want to prune away everything during stall computations, this stable capture shows all vars needed without liveness
jjenista [Tue, 28 Apr 2009 18:55:45 +0000 (18:55 +0000)]
added another liveness pass that picks up on virtual reads from stall pass, nasty bug fix in token table merge that left tables inconsistent
adash [Mon, 27 Apr 2009 19:58:00 +0000 (19:58 +0000)]
change makefile and add instructions to compile and run LeeRouter
jjenista [Mon, 27 Apr 2009 18:50:02 +0000 (18:50 +0000)]
Lots of bug fixes with regard to token table--table is consistent after every operation now, and an assertion method to verify consistency has been added for debugging
adash [Sat, 25 Apr 2009 01:51:45 +0000 (01:51 +0000)]
generate stats in runtime. Use -stmstats option for compiling
jjenista [Fri, 24 Apr 2009 16:46:21 +0000 (16:46 +0000)]
variable analysis reports no stalls, stable
jzhou [Fri, 24 Apr 2009 00:05:40 +0000 (00:05 +0000)]
add Tilera version stuff in buildscript and removes some redundant codes to avoid compilation warnings
jjenista [Thu, 23 Apr 2009 22:26:14 +0000 (22:26 +0000)]
bug fixes and implementation adjustments: matches analysis up through IV
jjenista [Thu, 23 Apr 2009 18:41:44 +0000 (18:41 +0000)]
quick capture
bdemsky [Thu, 23 Apr 2009 00:02:22 +0000 (00:02 +0000)]
change
bdemsky [Thu, 23 Apr 2009 00:00:26 +0000 (00:00 +0000)]
change parsing
jjenista [Wed, 22 Apr 2009 19:57:50 +0000 (19:57 +0000)]
variable analysis stable, reports stalls all over--buggy
jjenista [Wed, 22 Apr 2009 19:56:48 +0000 (19:56 +0000)]
add a convenient method to retreive FlatMethod from task or method
adash [Wed, 22 Apr 2009 15:39:00 +0000 (15:39 +0000)]
optimizations for readLine(), this makes executions considerably faster
bdemsky [Wed, 22 Apr 2009 09:18:23 +0000 (09:18 +0000)]
typos
bdemsky [Wed, 22 Apr 2009 09:14:43 +0000 (09:14 +0000)]
optimization (or bug fix) depending on how you view it
adash [Wed, 22 Apr 2009 00:43:26 +0000 (00:43 +0000)]
Tests read from a file that uses buffering directly, thereby eliminating the read method calls
to read number of rows from an input txt file
jzhou [Wed, 22 Apr 2009 00:16:08 +0000 (00:16 +0000)]
reorganize multicore version runtime codes for easy support of new platforms
jjenista [Tue, 21 Apr 2009 23:27:33 +0000 (23:27 +0000)]
variable analysis added
jjenista [Tue, 21 Apr 2009 22:35:11 +0000 (22:35 +0000)]
Root SESE ironed out, some reorganization for variable analysis. Fixed-point variable analysis modeled after other fixed point stuff, but strangely exits early. Build is stable though.
adash [Tue, 21 Apr 2009 19:29:40 +0000 (19:29 +0000)]
verified functionality and correctness with C versions
bdemsky [Tue, 21 Apr 2009 04:08:29 +0000 (04:08 +0000)]
remove shortcircuits
bdemsky [Tue, 21 Apr 2009 01:51:55 +0000 (01:51 +0000)]
make method static
adash [Tue, 21 Apr 2009 00:27:09 +0000 (00:27 +0000)]
print better error messages
bdemsky [Tue, 21 Apr 2009 00:05:58 +0000 (00:05 +0000)]
barrier bug
adash [Mon, 20 Apr 2009 23:37:28 +0000 (23:37 +0000)]
small changes and support for printing doubles in String class
adash [Mon, 20 Apr 2009 23:30:35 +0000 (23:30 +0000)]
Add runtime support for STM Barrier class
adash [Mon, 20 Apr 2009 23:27:00 +0000 (23:27 +0000)]
new Barrier Class
adash [Mon, 20 Apr 2009 23:21:47 +0000 (23:21 +0000)]
add new Barrier...the code still breaks on compiling.
enable printing double values
bdemsky [Sat, 18 Apr 2009 01:17:08 +0000 (01:17 +0000)]
integrate typeanalysis
bdemsky [Sat, 18 Apr 2009 00:58:57 +0000 (00:58 +0000)]
computes which types can actually refer to the same objects as other types
bdemsky [Fri, 17 Apr 2009 22:53:48 +0000 (22:53 +0000)]
New locking strategy... We no longer read lock objects... Instead:
We first acquire all write locks.
We next for read object, we check that (1) it isn't locked and the version number matches.
The serialization point for the transaction occurs between the two checks...
hkhang [Fri, 17 Apr 2009 22:20:12 +0000 (22:20 +0000)]
initial checkin
jjenista [Fri, 17 Apr 2009 21:54:39 +0000 (21:54 +0000)]
variable source tokens have a temp for the live variable name and a temp for the source variable
bdemsky [Fri, 17 Apr 2009 19:50:38 +0000 (19:50 +0000)]
reduce frequency of polling about GC
bdemsky [Fri, 17 Apr 2009 19:17:08 +0000 (19:17 +0000)]
updates
adash [Fri, 17 Apr 2009 19:04:15 +0000 (19:04 +0000)]
test file for Double to String change
jjenista [Fri, 17 Apr 2009 16:51:08 +0000 (16:51 +0000)]
liveness analysis simplified to ignore SESE's, analyzes each SESE in isolation. Results look correct, even for loops
jjenista [Fri, 17 Apr 2009 15:52:53 +0000 (15:52 +0000)]
reorganizing mlp analysis passes
bdemsky [Fri, 17 Apr 2009 00:47:01 +0000 (00:47 +0000)]
changes
bdemsky [Fri, 17 Apr 2009 00:46:32 +0000 (00:46 +0000)]
optimizations
jjenista [Thu, 16 Apr 2009 17:28:32 +0000 (17:28 +0000)]
another forward pass for mlp analysis
adash [Thu, 16 Apr 2009 16:42:59 +0000 (16:42 +0000)]
bug fixes for compilation...study functioning
bdemsky [Thu, 16 Apr 2009 08:41:47 +0000 (08:41 +0000)]
changes to make compiler work without STM flag with atomic statements/etc
bdemsky [Thu, 16 Apr 2009 07:45:21 +0000 (07:45 +0000)]
bug fix
bdemsky [Thu, 16 Apr 2009 07:41:46 +0000 (07:41 +0000)]
changes
bdemsky [Thu, 16 Apr 2009 07:40:55 +0000 (07:40 +0000)]
bug fixes
bdemsky [Thu, 16 Apr 2009 07:31:22 +0000 (07:31 +0000)]
bug fixes.
bdemsky [Thu, 16 Apr 2009 06:46:30 +0000 (06:46 +0000)]
debug support
bdemsky [Thu, 16 Apr 2009 05:58:59 +0000 (05:58 +0000)]
finally fix bug...after a full day on this one.
adash [Wed, 15 Apr 2009 23:48:28 +0000 (23:48 +0000)]
Add new Barrier ...will need to change it to a better implementation
jjenista [Wed, 15 Apr 2009 21:27:31 +0000 (21:27 +0000)]
fixed liveness analysis