79db55028c
Make data structure acurately get ALL edges, even loads of null fields of
...
nodes that are not shadow nodes
This fixes em3d to be _correct_ if not optimial
llvm-svn: 2274
2002-04-17 03:24:59 +00:00
071dea7362
* Remove the concept of a critical shadow node
...
* Make the function pointer argument explicit for a call nodes
* Eliminate unreachable global values
* Merge call nodes that are identical
llvm-svn: 2266
2002-04-16 20:39:59 +00:00
46b08cdc7c
* Eliminate ArgDSNode's completely, now rely on scalar map
...
* Fold call nodes that are indistinguishable for each other. This is a big
win for external functions like sqrt, which would multiply dramatically
before.
* Global nodes with no edges to or from them are now eliminated from the graph.
llvm-svn: 2257
2002-04-16 03:44:03 +00:00
2e9fa6d101
Move FunctionArgument out of iOther.h into Argument.h and rename class to
...
be 'Argument' instead of FunctionArgument.
llvm-svn: 2216
2002-04-09 19:48:49 +00:00
06be180225
Add explicit #includes of Function.h to make up for the removed #include
...
in iOther.h
llvm-svn: 2209
2002-04-09 19:08:28 +00:00
d92b01c385
Add #includes to make up for #includes pruned out of header files.
...
llvm-svn: 2207
2002-04-09 18:37:46 +00:00
3734fe4f35
Play around with #includes
...
llvm-svn: 2174
2002-04-08 21:58:53 +00:00
9ba54b2f31
Call nodes are never equivalent
...
Shadow nodes are never critical.
llvm-svn: 2102
2002-04-04 19:21:51 +00:00
d6c8dd1473
Add hack to get timing of analysis
...
llvm-svn: 2101
2002-04-04 19:21:27 +00:00
d7b6c331ea
* Eliminate commented out code
...
* Do not demand a shadow node when resolving
* Raise arbitrary inline limit
llvm-svn: 2100
2002-04-04 19:21:06 +00:00
4f47bb8ae6
Support resolving function arguments/return values to pointers that index
...
into other objects.
llvm-svn: 2094
2002-04-01 22:20:48 +00:00
7203e9f05a
Avoid incorrectly adding null values to the scalar map!
...
llvm-svn: 2085
2002-04-01 00:45:09 +00:00
f224b6b511
Critical shadow nodes no do not know their parent explictly.
...
llvm-svn: 2084
2002-04-01 00:15:30 +00:00
b4648bb38f
Minor cleanups (use dyn_cast instead of testing manually)
...
Shadow nodes now don't explicitly know their parent.
llvm-svn: 2083
2002-04-01 00:14:41 +00:00
6b06aa557e
Allow merging of identical call nodes. Make the shadow node pointed to
...
by the call node noncritical before the call is destroyed.
llvm-svn: 2082
2002-04-01 00:13:56 +00:00
a7b3b50381
Increase limit for perimeter
...
llvm-svn: 2081
2002-04-01 00:12:58 +00:00
d63afb418e
Unify the destruction code used for node pairs vs normal nodes. This was
...
causing a problem before because global values with incoming edges didn't
copy the incoming edges to the node they were being merged from, causing
the poolalloc pass to die.
llvm-svn: 2079
2002-03-31 19:57:44 +00:00
d45f4df2ae
* Move the isEquivalentTo implementations here. They can probably be put
...
someplace nicer in the file though.
* Add new dump method for debugging
llvm-svn: 2064
2002-03-31 07:16:08 +00:00
49c3ea8c1e
Print out the instruction instead of just the address!
...
llvm-svn: 2063
2002-03-31 07:15:20 +00:00
ed9bfd95a9
* Move isEquivalentTo implementations to NodeImpl
...
* Implement a new form of node folding to catch cases missed in Addtree
* Add removeIndistinguishableNodePairs to merge calls (todo) and globals
llvm-svn: 2062
2002-03-31 07:14:52 +00:00
a2cdaba6ed
* Convert CallMap to be a vector, because the keys can change, and the map
...
doesn't resort!
* Be more generous with the cached matches that we allow now.
* Recursive calls should all work now!
llvm-svn: 2061
2002-03-31 07:13:27 +00:00
8d4894e3fa
Implement getEscapingAllocations & getNonEscapingAllocations
...
llvm-svn: 2021
2002-03-28 19:33:00 +00:00
c3ae15cf0b
* Rename UnlinkUndistinguishableShadowNodes & RemoveUnreachableShadowNodes
...
to reflect that they can eliminate arbitrary nodes.
* Rename the ShadowNodeEliminate.cpp file to EliminateNodes.cpp for the
same reason
llvm-svn: 2020
2002-03-28 19:16:48 +00:00
8a5c470c59
Ooops, I did such a great job pruning nodes, that I accidentally deleted
...
ALL allocation nodes... hrm... bad.
llvm-svn: 2018
2002-03-28 18:38:38 +00:00
b1782d0c90
Many changes
...
* Simplify a lot of the inlining stuff. There are still problems, but not
many
* Break up the Function representation to have a vector for every different
node type so it is fast to find nodes of a particular flavor.
* Do more intelligent merging of call values
* Allow elimination of unreachable shadow and allocation nodes
* Generalize indistinguishability testing to allow merging of identical calls.
* Increase shadow node merging power
llvm-svn: 2010
2002-03-28 17:56:03 +00:00
be449bc1ef
* Add #define to enabled debug messages
...
* Move removeEdgesTo to be a member of DSNode
* Implement (but #ifdef out) the new, spiffier, method of determining
shadow node equivalence. This cannot be enabled until more is
implemented.
llvm-svn: 2004
2002-03-27 19:48:03 +00:00
d9c86c7c96
* Implement DSNode::removeAllIncomingEdges
...
* Implement Critical Shadow node handling
* Implement routines to determine whether an allocation node is a malloc
or alloca
llvm-svn: 2003
2002-03-27 19:46:05 +00:00
0ad8b8f5d0
* Add critical node support
...
* Optimize graph after building it. This should be unneccesary in the future
llvm-svn: 2002
2002-03-27 19:45:12 +00:00
b6c8c5138f
* Destroy alloca nodes when a graph gets inlined
...
* Add links to all subtrees when a shadow node gets resolved
* Add critical node handling
llvm-svn: 2001
2002-03-27 19:44:33 +00:00
b358f71d02
* Optimizers return true on change
...
* Implement indistinguishable shadow node elimination
llvm-svn: 1999
2002-03-27 00:55:13 +00:00
5c3280971c
Fix long line
...
llvm-svn: 1998
2002-03-27 00:54:31 +00:00
f8e2d060cc
* Because of optimization, the shadow nodes between arguments might get
...
removed. Check to see if they are there.
* Repeat optimizations while changing
llvm-svn: 1997
2002-03-27 00:53:57 +00:00
e604d8ec65
Initial checkin of Datastructure analysis.
...
Has bugs, but shouldn't crash in theory.
llvm-svn: 1994
2002-03-26 22:39:06 +00:00
Chris Lattner