jihoonl [Thu, 1 Apr 2010 05:38:34 +0000 (05:38 +0000)]
still fixing
jihoonl [Thu, 1 Apr 2010 01:55:31 +0000 (01:55 +0000)]
still have racing condition..
yeom [Wed, 31 Mar 2010 23:45:11 +0000 (23:45 +0000)]
fixes for memory effects analysis.
jjenista [Wed, 31 Mar 2010 17:21:08 +0000 (17:21 +0000)]
enforce strict monotonicity for initial method contexts and back edges as defined by the PointerMethod, which allow the nonterminating test cases to converge now. Still gotta test on the wider benchmark suite...
jjenista [Tue, 30 Mar 2010 22:13:20 +0000 (22:13 +0000)]
turn strong updates back on from yesterday testing, add a non-terminating example using just strong updates
jjenista [Tue, 30 Mar 2010 21:39:53 +0000 (21:39 +0000)]
a test that shows our analysis doesn't always terminate
jjenista [Tue, 30 Mar 2010 21:39:33 +0000 (21:39 +0000)]
why was this missing--latent bug
jjenista [Tue, 30 Mar 2010 21:31:09 +0000 (21:31 +0000)]
bug fixes, make stack/Q method-visiting a cmd line arg, some code for debugging monotonicity problem, commented out in this commit
jjenista [Fri, 26 Mar 2010 20:09:16 +0000 (20:09 +0000)]
all changes for official testing
jjenista [Fri, 26 Mar 2010 15:56:27 +0000 (15:56 +0000)]
collect last night's changes
bdemsky [Fri, 26 Mar 2010 05:40:53 +0000 (05:40 +0000)]
stack implementation...may expose bugs
bdemsky [Fri, 26 Mar 2010 05:34:56 +0000 (05:34 +0000)]
bug fix
bdemsky [Fri, 26 Mar 2010 05:15:28 +0000 (05:15 +0000)]
bug fix
jjenista [Fri, 26 Mar 2010 01:08:14 +0000 (01:08 +0000)]
current version of benchmark lists and run modes
yeom [Thu, 25 Mar 2010 23:15:35 +0000 (23:15 +0000)]
add power bench.
yeom [Thu, 25 Mar 2010 22:55:53 +0000 (22:55 +0000)]
add annoated kmeans
yeom [Thu, 25 Mar 2010 22:55:15 +0000 (22:55 +0000)]
changes.
jjenista [Thu, 25 Mar 2010 22:54:43 +0000 (22:54 +0000)]
a helpful makefile to run for the record
yeom [Thu, 25 Mar 2010 22:42:03 +0000 (22:42 +0000)]
add annotated raytracer
bdemsky [Thu, 25 Mar 2010 22:15:29 +0000 (22:15 +0000)]
bug fixes and optimizations
jjenista [Thu, 25 Mar 2010 21:56:13 +0000 (21:56 +0000)]
new Canonical hash and equals, also running with assertions disabled, barely improved
jjenista [Thu, 25 Mar 2010 21:03:06 +0000 (21:03 +0000)]
fixing the heap region descriptions not a pressing issue...
jjenista [Thu, 25 Mar 2010 20:15:40 +0000 (20:15 +0000)]
makefile for checking final results
jjenista [Thu, 25 Mar 2010 20:14:08 +0000 (20:14 +0000)]
bug fixes for task initial heaps and making all new edges have true predicates attached to reach states on the edge
jjenista [Thu, 25 Mar 2010 20:12:18 +0000 (20:12 +0000)]
print preds stacked for DOT
yeom [Thu, 25 Mar 2010 19:46:59 +0000 (19:46 +0000)]
bug fixes
jzhou [Thu, 25 Mar 2010 17:17:38 +0000 (17:17 +0000)]
Add BAMBOOCOMPILETIME flag into the state class. When this flag is set, only excute Bamboo schedule analysis and report the time
bdemsky [Thu, 25 Mar 2010 03:50:17 +0000 (03:50 +0000)]
little optimization to allow it to quickly recompute methods it already has information for
bdemsky [Thu, 25 Mar 2010 03:49:18 +0000 (03:49 +0000)]
code for Jim to use once he gets everything working...
the idea is that this takes in a flatmethod and generates next/prev nodes for each node that the pointer analysis actually cares about
yeom [Thu, 25 Mar 2010 03:30:48 +0000 (03:30 +0000)]
changes for proper handling multi-dimensional array of task parameter.
jjenista [Thu, 25 Mar 2010 01:26:34 +0000 (01:26 +0000)]
bug fixes
jjenista [Wed, 24 Mar 2010 22:18:52 +0000 (22:18 +0000)]
bug fix in evaluating the various classes of out-of-context predicates
jjenista [Wed, 24 Mar 2010 22:18:13 +0000 (22:18 +0000)]
improved debug graph capture
jjenista [Wed, 24 Mar 2010 18:17:51 +0000 (18:17 +0000)]
just giving the task a name different from the Lambda class's method
jjenista [Wed, 24 Mar 2010 17:03:02 +0000 (17:03 +0000)]
bug fixes, display improvements, sharing query changes, still losing preds during call site transfer func...
yeom [Tue, 23 Mar 2010 23:21:20 +0000 (23:21 +0000)]
changes for a better strategy of clearing memory.
jjenista [Tue, 23 Mar 2010 18:52:42 +0000 (18:52 +0000)]
bug fix for primitive parameters, made a simple method for encapsulating the primitive/immutable test, leaving a hook for debugging the graph equals method that was useful, easy to turn back on
jjenista [Tue, 23 Mar 2010 00:46:28 +0000 (00:46 +0000)]
little bug fixes, adjusted code for detecting sharing, use the reverse topological sort for new analysis
jjenista [Mon, 22 Mar 2010 23:02:03 +0000 (23:02 +0000)]
fix bug that a hrn ID integer and a boolean out-of-context --identify-- an out-of-context node
jjenista [Mon, 22 Mar 2010 18:30:08 +0000 (18:30 +0000)]
little bug in final reporting
jjenista [Mon, 22 Mar 2010 18:23:44 +0000 (18:23 +0000)]
bug fixes: if an out-of-context edge in callee has no matching desitination, obviously don't bring it, and don't bother adding shadow into shadow node descriptions, not worth the hassle of changing string descriptions when making shadow nodes on demand
jjenista [Mon, 22 Mar 2010 17:41:36 +0000 (17:41 +0000)]
wow, this bug has been latent for a long time, easy fix
jjenista [Mon, 22 Mar 2010 17:13:08 +0000 (17:13 +0000)]
add two out-of-context properties to equals and hashcode
jjenista [Mon, 22 Mar 2010 03:56:52 +0000 (03:56 +0000)]
makefile for new analysis version for benchmarks
jjenista [Mon, 22 Mar 2010 03:33:17 +0000 (03:33 +0000)]
couple fixes to make sure out-of-context nodes get all the states they need, and make sure global sweep never removes states from an out-of-context node
jjenista [Mon, 22 Mar 2010 02:56:57 +0000 (02:56 +0000)]
fixed problem by differentiating between an element that is out of the callee context, and an element that is also out of the caller context as well, results looking better but reachability states aren't propagating over more than one context
jjenista [Sat, 20 Mar 2010 00:38:06 +0000 (00:38 +0000)]
trying to fix call site transfer bug, fixed caused other problems, still working on it
yeom [Fri, 19 Mar 2010 23:46:23 +0000 (23:46 +0000)]
changes.
jjenista [Fri, 19 Mar 2010 22:25:44 +0000 (22:25 +0000)]
making debugging easier
jjenista [Fri, 19 Mar 2010 21:28:39 +0000 (21:28 +0000)]
make sure change sets ignore predicates hanging off reach states and alter reach set intersection to ignore predicates, and have a preference of taking preds from the first reach set argument so client algorithms can indicate preference
jjenista [Fri, 19 Mar 2010 19:21:10 +0000 (19:21 +0000)]
make sure straight union of reach states or reach sets is never done, reach tuples in states with same id should merge arity values, and reach states with same state but different preds should become one state with preds ORed
yeom [Fri, 19 Mar 2010 18:50:45 +0000 (18:50 +0000)]
changes.
yeom [Fri, 19 Mar 2010 00:51:39 +0000 (00:51 +0000)]
add interface.
jzhou [Thu, 18 Mar 2010 22:15:37 +0000 (22:15 +0000)]
bug fix in multicore Bamboo, TILERZ_LINUX mode now works for 56 cores
yeom [Thu, 18 Mar 2010 20:32:57 +0000 (20:32 +0000)]
a test.
jjenista [Thu, 18 Mar 2010 18:48:56 +0000 (18:48 +0000)]
couple of little bugs working in tandem to wreck reachability, looking much better now
adash [Thu, 18 Mar 2010 01:22:52 +0000 (01:22 +0000)]
added the new looup table for prefetch cache
updated gCollect as per the new prefetch cache
send trans disagree quickly and unlock the objects quickly
some bugs fixes that resulted from the dsm caching
adash [Thu, 18 Mar 2010 00:35:31 +0000 (00:35 +0000)]
get rid of "size" field, causes unnecessary aborts for FileSystem
jzhou [Thu, 18 Mar 2010 00:02:15 +0000 (00:02 +0000)]
BAMBOO TILERA_ZLINUX mode works with single-core version but still has problem with 56-core version
jjenista [Wed, 17 Mar 2010 22:31:11 +0000 (22:31 +0000)]
use different graph writing options to better find reachability problems
jzhou [Wed, 17 Mar 2010 22:24:23 +0000 (22:24 +0000)]
TILERA_ZLINUX mode for BAMBOO can be compiled now, but still has segmentation fault
jjenista [Wed, 17 Mar 2010 21:31:29 +0000 (21:31 +0000)]
fixed issues with dot graph writing that make our debugging lives harder
jjenista [Wed, 17 Mar 2010 20:47:34 +0000 (20:47 +0000)]
using allocation sites in new ways exposed a latent bug
yeom [Wed, 17 Mar 2010 18:51:07 +0000 (18:51 +0000)]
a test
yeom [Wed, 17 Mar 2010 18:29:36 +0000 (18:29 +0000)]
changes for proper handling field.
jjenista [Wed, 17 Mar 2010 18:27:58 +0000 (18:27 +0000)]
was accidentally dropping param var to node edges when constructing callee initial heaps
jjenista [Wed, 17 Mar 2010 15:01:14 +0000 (15:01 +0000)]
working on reach prop, pred bugs, basic tests still do not work, gotta squash more bugs
jjenista [Wed, 17 Mar 2010 01:00:55 +0000 (01:00 +0000)]
reach propagation not working, getting null states...
jzhou [Wed, 17 Mar 2010 00:03:17 +0000 (00:03 +0000)]
code clean: define TILERA_BME mode and TILERA_ZLINUX mode. TILERA_ZLINUX mode is not tested yet
jjenista [Tue, 16 Mar 2010 22:44:07 +0000 (22:44 +0000)]
working hard on reachability, bunch of changes, still isnt working right though
yeom [Tue, 16 Mar 2010 21:17:15 +0000 (21:17 +0000)]
bug fix.
yeom [Tue, 16 Mar 2010 17:58:28 +0000 (17:58 +0000)]
changes for handling tasks.
jjenista [Tue, 16 Mar 2010 17:12:43 +0000 (17:12 +0000)]
checkin a crash fix so yonghun can run on benchmarks while I keep wokring on reachability
jjenista [Tue, 16 Mar 2010 00:32:31 +0000 (00:32 +0000)]
updating the global sweep and some related code--this currently bombs, only update if you must, it will compile though
jjenista [Mon, 15 Mar 2010 22:15:27 +0000 (22:15 +0000)]
have to test predicates of callee states before admitting to caller, and calculate what predicate caller version will have
jjenista [Mon, 15 Mar 2010 21:28:49 +0000 (21:28 +0000)]
lots of untested code, but compiles, for attaching preds to reach states
jjenista [Mon, 15 Mar 2010 19:55:17 +0000 (19:55 +0000)]
most of the missing context-rewriting code
jjenista [Sat, 13 Mar 2010 18:16:13 +0000 (18:16 +0000)]
a start on reachability, not fully functioning yet
adash [Sat, 13 Mar 2010 02:08:34 +0000 (02:08 +0000)]
more changes
adash [Sat, 13 Mar 2010 01:59:42 +0000 (01:59 +0000)]
changes for optimizations
adash [Sat, 13 Mar 2010 01:58:40 +0000 (01:58 +0000)]
changes
jjenista [Fri, 12 Mar 2010 20:11:01 +0000 (20:11 +0000)]
if an operation on canonical objects wants to use other canonical ops, it must supply canonical arguments. that said, if the operation is straightforward and the result isn't hashed, its okay to build the object from scratch and then canonicalize it at the end
jjenista [Fri, 12 Mar 2010 19:59:57 +0000 (19:59 +0000)]
a test that shows reachabiltiy states are broken
jjenista [Fri, 12 Mar 2010 19:16:06 +0000 (19:16 +0000)]
bunch of bug fixes, graphs appear to be working mechanically over call chains, still needs attention to reachability as symbols get rewritten over context boundaries
jjenista [Fri, 12 Mar 2010 01:16:17 +0000 (01:16 +0000)]
making a big change to call site transform to calculate out-of-context edges correctly, but in current form BLEEDS ELEMENTS BETWEEN GRAPHS again, checking in this stable version though so I don't mess up the good stuff while trying to fix it
jjenista [Thu, 11 Mar 2010 23:38:53 +0000 (23:38 +0000)]
debugging something else, but commiting this bug fix--mapping out-of-context nodes to a unique heap region id requires that the callee method be part of the identifier string otherwise an ooc region can appear in more than one context of a call chain, which is bad
jjenista [Thu, 11 Mar 2010 22:18:14 +0000 (22:18 +0000)]
bug fix yonghun found, return value's region may be a new node for caller context, might have to a matching object for caller context
jjenista [Thu, 11 Mar 2010 21:54:35 +0000 (21:54 +0000)]
there was a simple fix for variable problem, in new model we want to cut out all region to region AND var to region edges in caller before we map callee effects back in, things look good for one call but chains any longer get messed up--investigating...
jjenista [Thu, 11 Mar 2010 20:08:50 +0000 (20:08 +0000)]
caller local variables to in-context regions are modeled as out-of-context edges in callee to preserve info, then matched in call site transform. This approach is running and stable, but the answer doesn't look right with respect to the age of nodes in caller
jzhou [Thu, 11 Mar 2010 19:52:53 +0000 (19:52 +0000)]
remove some redundant outputs
jzhou [Thu, 11 Mar 2010 19:23:14 +0000 (19:23 +0000)]
some missing check in
jjenista [Thu, 11 Mar 2010 00:13:06 +0000 (00:13 +0000)]
Not the most elegant fix, but out-of-context nodes can be uniquely identified by a string, which is mapped to a particular hrn id to get convergence
jjenista [Wed, 10 Mar 2010 23:02:58 +0000 (23:02 +0000)]
out-of-context edges prevent convergence...
jjenista [Wed, 10 Mar 2010 21:59:43 +0000 (21:59 +0000)]
lots of bug fixes, system cannot compute even simple programs without variables as out-of-context edges
jzhou [Wed, 10 Mar 2010 18:49:38 +0000 (18:49 +0000)]
mark before adding zero-overhead linux mode
jjenista [Wed, 10 Mar 2010 00:28:14 +0000 (00:28 +0000)]
big checkin, lots of call site transfer bug fixes, analysis runs for a tiny example correctly
jjenista [Tue, 9 Mar 2010 01:05:31 +0000 (01:05 +0000)]
callee elements brought into caller get predicates from caller element that satisfied them, properly, also bug fix for edges from nodes that aren't satisifed from throwing a null pointer exception
jjenista [Mon, 8 Mar 2010 21:14:09 +0000 (21:14 +0000)]
edge merge
yeom [Mon, 8 Mar 2010 18:19:25 +0000 (18:19 +0000)]
changes.
adash [Sat, 6 Mar 2010 22:55:04 +0000 (22:55 +0000)]
changes to trans.c for performance improvement