1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-30 07:22:55 +01:00
Commit Graph

160 Commits

Author SHA1 Message Date
Chris Lattner
9fd0c48f80 Two changes:
1. Functions do not make things incomplete, only variables
 2. Constant global variables no longer need to be marked incomplete, because
    we are guaranteed that the initializer for the global will be in the
    graph we are hacking on now.  This makes resolution of indirect calls happen
    a lot more in the bu pass, supports things like vtables and the C counterparts
    (giant constant arrays of function pointers), etc...

Testcase here: test/Regression/Analysis/DSGraph/constant_globals.ll

llvm-svn: 11852
2004-02-25 23:36:08 +00:00
Chris Lattner
9fe3bf296d Simplify the dead node elimination stuff
Make the incompleteness marker faster by looping directly over the globals
instead of over the scalars to find the globals

Fix a bug where we didn't mark a global incomplete if it didn't have any
outgoing edges.  This wouldn't break any current clients but is still wrong.

llvm-svn: 11848
2004-02-25 23:08:00 +00:00
Chris Lattner
573441bfbd Use isNull instead of getNode() to test for existence of a node, this is cheaper.
FIX MAJOR BUG, whereby we didn't merge null edges correctly. Correcting this
fixes poolallocation on 175.vpr, and possibly others.

llvm-svn: 11695
2004-02-22 00:53:54 +00:00
Chris Lattner
7bff00313e Fix an iterator invalidation problem which was causing some nodes to not be
correctly merged over!

llvm-svn: 11693
2004-02-21 22:28:26 +00:00
Chris Lattner
16690fad3d Adjust to the changed StructType interface. In particular, getElementTypes() is gone.
llvm-svn: 11228
2004-02-09 04:37:31 +00:00
Chris Lattner
cc1376078c Instead of callign removeTriviallyDeadNodes on the global graph every time
removeDeadNodes is called, only call it at the end of the pass being run.
This saves 1.3 seconds running DSA on 177.mesa (5.3->4.0s), which is
pretty big.  This is only possible because of the automatic garbage
collection done on forwarding nodes.

llvm-svn: 11178
2004-02-08 01:51:48 +00:00
Chris Lattner
32a3eb0b88 Substantially improve the DSA code by removing 'forwarding' nodes from
DSGraphs while they are forwarding.  When the last reference to the forwarding
node is dropped, the forwarding node is autodeleted.  This should simplify
removeTriviallyDead nodes, and is only (efficiently) possible because we are
using an ilist of dsnodes now.

llvm-svn: 11175
2004-02-08 01:27:18 +00:00
Chris Lattner
84ff796305 Bugfix for ilist conversion. The ilist wants to make an 'end' node which has
G == 0

llvm-svn: 11174
2004-02-08 01:05:37 +00:00
Chris Lattner
29067016a4 Switch the Nodes list from being an std::vector<DSNode*> to an ilist<DSNode>
llvm-svn: 11173
2004-02-08 00:53:26 +00:00
Chris Lattner
2c836fc933 Change to use node_iterators instead of direct access to Nodes
llvm-svn: 11171
2004-02-08 00:23:16 +00:00
Chris Lattner
6af67c7eb4 getNodes() is gone, use node_begin/end instead
Rename stats from dsnode -> dsa
Add a new stat

llvm-svn: 11167
2004-02-07 23:58:05 +00:00
Chris Lattner
ce838bfae6 There is no need to clone over nodes that are going to be dead anyway
llvm-svn: 11157
2004-02-07 22:00:03 +00:00
Chris Lattner
dfa4f14e34 Fix a bug aflicting 265.gap
llvm-svn: 11006
2004-01-29 08:36:22 +00:00
Chris Lattner
caecd769c4 Minor bugfixes
llvm-svn: 11005
2004-01-29 03:32:15 +00:00
Chris Lattner
7ab2349541 Rename DSGraph::ScalarMapTy -> DSScalarMap
llvm-svn: 11001
2004-01-28 09:15:42 +00:00
Chris Lattner
9a79b1690c Fix a bug
llvm-svn: 11000
2004-01-28 03:31:34 +00:00
Chris Lattner
8324e48b84 Eliminate the call to removeTriviallyDeadNodes from updateFromGlobals graph,
moving it to the start of removeDeadNodes.  This speeds up DSA by 2s on perlbmk
from 41s

llvm-svn: 10999
2004-01-28 03:24:41 +00:00
Chris Lattner
394c9343b4 In updateFromGlobalsGraph, instead of iterating over all of the scalars in the
function to find the globals, iterate over all of the globals directly.  This
speeds the function up from 14s to 6.3s on perlbmk, reducing DSA time from
53->46s.

llvm-svn: 10996
2004-01-28 03:03:06 +00:00
Chris Lattner
528f5842dd Minor tweaks, eliminate useless integer pruning optimziation, turn on
timers by default

llvm-svn: 10993
2004-01-28 02:41:32 +00:00
Chris Lattner
18e0615a69 Further reduce the number of nodes cloned with getClonedNH, using merge instead.
This reduces the number of nodes allocated, then immediately merged and DNE'd
from 2193852 to 1298049.  unfortunately this only speeds DSA up by ~1.5s (of
53s), because it's spending most of its time waddling through the scalar map :(

llvm-svn: 10992
2004-01-28 02:11:49 +00:00
Chris Lattner
6d38674f7b Add a timer, fix a minor bug.
Also, use RC::merge when possible, reducing the number of nodes allocated, then immediately merged away from 2985444 to 2193852 on perlbmk.

llvm-svn: 10991
2004-01-28 02:05:05 +00:00
Chris Lattner
10881df6e2 Another bugfix, disable "spurious" output.
You gotta love spurious

llvm-svn: 10990
2004-01-28 01:19:52 +00:00
Chris Lattner
b0d7b564dd fix bug in previous checkin
llvm-svn: 10989
2004-01-27 22:54:56 +00:00
Chris Lattner
3bd20c5114 * Add a new commandline argument to control the "global roots hack". Default
it to be off.  If it looks like it's completely unnecessary after testing, I
  will remove it completely (which is the hope).
* Callers of the DSNode "copy ctor" can not choose to not copy links.
* Make node collapsing not create a garbage node in some cases, avoiding a
  memory allocation, and a subsequent DNE.
* When merging types, allow two functions of different types to be merged
  without collapsing.
* Use DSNodeHandle::isNull more often instead of DSNodeHandle::getNode() == 0,
  as it is much more efficient.
*** Implement the new, more efficient reachability cloner class
    In addition to only cloning nodes that are reachable from interesting
    roots, this also fixes the huge inefficiency we had where we cloned lots
    of nodes, only to merge them away immediately after they were cloned.
    Now we only actually allocate a node if there isn't one to merge it into.
* Eliminate the now-obsolete cloneReachable* and clonePartiallyInto methods
* Rewrite updateFromGlobalsGraph to use the reachability cloner
* Rewrite mergeInGraph to use the reachability cloner
* Disable the scalar map scanning code in removeTriviallyDeadNodes.  In large
  SCC's, this is extremely expensive.  We need a better data structure for the
  scalar map, because we really want to scan the unique node handles, not ALL
  of the scalars.
* Remove the incorrect SANER_CODE_FOR_CHECKING_IF_ALL_REFERRERS_ARE_FROM_SCALARMAP code.
* Move the code for eliminating integer nodes from the trivially dead
  eliminator to the dead node eliminator.
* removeDeadNodes no longer uses removeTriviallyDeadNodes, as it contains a
  superset of the node removal power.
* Only futz around with the globals graph in removeDeadNodes if it is modified

llvm-svn: 10987
2004-01-27 22:03:40 +00:00
Chris Lattner
6d2f67adb3 Initial support for implementing clonePartiallyInto in terms of cloneReachableSubgraph, though this support is currently disabled.
llvm-svn: 10970
2004-01-23 01:44:53 +00:00
Chris Lattner
a72f3a387a Eliminated the CompletedNodes argument to the cloneReachable* methods. This
map was only used to implement a marginal GlobalsGraph optimization, and it
actually slows the analysis down (due to the overhead of keeping it), so just
eliminate it entirely.

llvm-svn: 10955
2004-01-22 16:56:13 +00:00
Chris Lattner
50e765d1a2 Ok, I'm tired of pulling out all my timers to check stuff in, just do it.
llvm-svn: 10954
2004-01-22 16:36:28 +00:00
Chris Lattner
da04e4d383 Bug fix: X.mergeWith(Y) was not updating Y if Y was a null node handle!
llvm-svn: 10953
2004-01-22 16:31:08 +00:00
Chris Lattner
07d45564d0 Start implementing DSGraph::clonePartiallyInto and implement mergeInGraph
in terms of it.

Though clonePartiallyInto is not cloning partial graphs yet, this change
dramatically speeds up inlining of graphs with many scalars.  For example,
this change speeds up the BU pass on 253.perlbmk from 69s to 36s, because
it avoids iteration over the scalar map, which can get pretty large.

llvm-svn: 10951
2004-01-22 15:30:58 +00:00
Chris Lattner
b60823472a Fine-grainify namespaces for this library
llvm-svn: 9948
2003-11-12 23:11:14 +00:00
Chris Lattner
95d0b541c1 Add new argument to disable checking
llvm-svn: 9922
2003-11-12 17:58:22 +00:00
Chris Lattner
87ad20251a Don't crash if no gg
llvm-svn: 9913
2003-11-12 04:57:58 +00:00
Brian Gaeke
d25f86d683 Put all LLVM code into the llvm namespace, as per bug 109.
llvm-svn: 9903
2003-11-11 22:41:34 +00:00
Chris Lattner
6deb20046e Compute node offsets correctly
llvm-svn: 9895
2003-11-11 20:12:32 +00:00
Chris Lattner
365bda897c Add new method for computing node mappings. This is used by the pool allocator
llvm-svn: 9880
2003-11-11 05:08:59 +00:00
Chris Lattner
a71b07eeca All DSGraphs keep a reference to the targetdata they are created with. This is
used to eliminate the hard coded, hacked in, sparc specific, global TargetData.
Changing the TargetData used to actually match the code fixes problems, and
eliminates a crash.

llvm-svn: 9659
2003-11-02 22:27:28 +00:00
Chris Lattner
63fb5bcaac Fix bug: test/Regression/Analysis/DSGraph/2003-11-02-NodeCollapsing.ll
llvm-svn: 9658
2003-11-02 21:02:20 +00:00
John Criswell
b402729b30 Added LLVM project notice to the top of every C++ source file.
Header files will be on the way.

llvm-svn: 9298
2003-10-20 19:43:21 +00:00
Chris Lattner
f203e908e3 Remove trivially dead nodes was not removing nodes that were dead due to forwarding!
llvm-svn: 8634
2003-09-20 21:34:07 +00:00
Chris Lattner
b12a8a6177 Switch from using CallInst's to represent call sites to using the LLVM
CallSite class.  Now we can represent function calls by invoke instructions
too!

llvm-svn: 8629
2003-09-20 16:34:13 +00:00
Misha Brukman
b9a4ddbfc5 Fixed spelling and grammar.
llvm-svn: 8478
2003-09-11 18:14:24 +00:00
Chris Lattner
e2ee62c91c Add more verbose comment
llvm-svn: 7610
2003-08-05 18:38:37 +00:00
Sumant Kowshik
1a4fce3511 Added function mergeInGlobalsGraph which merges in the entire globals graph with the graph of a function
llvm-svn: 7606
2003-08-05 17:04:41 +00:00
Chris Lattner
686e94e760 DEBUG got moved to Support/Debug.h
llvm-svn: 7492
2003-08-01 22:15:03 +00:00
Vikram S. Adve
420ec0bb0f Fix comment.
llvm-svn: 7227
2003-07-22 12:08:58 +00:00
Vikram S. Adve
d1a85b7860 Implement 2 important changes: (1) rematerialization from the globals graph,
and (2) faster inlining by cloning only reachable nodes.  In particular:
(1) Added DSGraph::cloneReachableSubgraph and DSGraph::cloneReachableNodes
    to clone the subgraph reachable from a set of root nodes, into the
    current graph, merging the global nodes into thos in the current graph.
    The TD pass now uses this for faster inlining, and so does the
    next function.
(2) Added DSGraph::updateFromGlobalGraph() to rematerialize nodes from the
    globals graph into the current graph in both BU and TD passes.
(3) `I' flags are removed from all nodes in the globals graph, because they
    are difficult to maintain correctly and are not needed anyway.
(4) Aux. function calls are only removed to the globals graph if they
    will never be resovled.  (This is what fixed gap.)  The immediate
    reason is that if we took these out of a function (and moved them to
    the globals graph) we would need to rematerialize these nodes into the
    function graph for every function in the BU pass.  The longer term
    problem is that we would need to find a way to remove them from the
    globals graph iff they have been resolved on all paths through the
    call graph.

llvm-svn: 7187
2003-07-16 21:36:31 +00:00
Chris Lattner
2a6504c799 Remove globals more aggressively from graphs.
Fix a bug where we removed nodes that were marked U.

llvm-svn: 7090
2003-07-03 02:03:53 +00:00
Chris Lattner
492b7eeb06 INCLUDE_PARENT_GRAPH is required!
llvm-svn: 7088
2003-07-02 23:57:05 +00:00
Chris Lattner
f4a221c87c Add support for ParentGraph only when building in debug mode
Minor cleanups, reenable folding of call nodes to external functions

llvm-svn: 7061
2003-07-02 04:38:49 +00:00
Chris Lattner
5ddec0f02d Rework TD pass to work with the precise call graph constructed by the BU phase
llvm-svn: 7031
2003-07-01 16:28:11 +00:00