Commit message (Collapse) | Author | Age | Files | Lines | |
---|---|---|---|---|---|
* | Avoid extra copy | Chris Lattner | 2002-10-20 | 1 | -2/+1 |
| | | | | llvm-svn: 4239 | ||||
* | Simplify code a bit, add comment flyer | Chris Lattner | 2002-10-20 | 1 | -10/+11 |
| | | | | llvm-svn: 4238 | ||||
* | Split some long lines | Chris Lattner | 2002-10-20 | 1 | -3/+4 |
| | | | | llvm-svn: 4237 | ||||
* | Remove spurious caller pointer in DSCallSite. | Vikram S. Adve | 2002-10-20 | 5 | -23/+32 |
| | | | | | | Also add functions to access pointer argument nodes cleanly. llvm-svn: 4235 | ||||
* | This function can be static | Chris Lattner | 2002-10-20 | 1 | -2/+2 |
| | | | | llvm-svn: 4234 | ||||
* | Fix previous checkin :( | Chris Lattner | 2002-10-20 | 1 | -1/+1 |
| | | | | llvm-svn: 4233 | ||||
* | Avoid extra callSite copy | Chris Lattner | 2002-10-20 | 1 | -1/+1 |
| | | | | llvm-svn: 4232 | ||||
* | Print the array flag | Chris Lattner | 2002-10-20 | 1 | -0/+2 |
| | | | | llvm-svn: 4231 | ||||
* | Added a first-class representation for each call site that can be | Vikram S. Adve | 2002-10-20 | 6 | -60/+76 |
| | | | | | | | | | used in the DS graphs. Essentially, what was vector<DSNodeHandle> before is now a DSCallSite with the same vector, plus pointers to the CallInst and the caller Function. The special-purpose class BUDataStructure::CallSite is no longer needed. llvm-svn: 4228 | ||||
* | Convert typerec to be a structure instead of a pair | Chris Lattner | 2002-10-18 | 3 | -8/+7 |
| | | | | llvm-svn: 4226 | ||||
* | Print Mod/ref info | Chris Lattner | 2002-10-17 | 1 | -0/+2 |
| | | | | llvm-svn: 4224 | ||||
* | Calculate mod/ref info | Chris Lattner | 2002-10-17 | 1 | -0/+3 |
| | | | | llvm-svn: 4223 | ||||
* | Remove more obsolete code | Chris Lattner | 2002-10-17 | 1 | -31/+0 |
| | | | | llvm-svn: 4221 | ||||
* | * Make the DSGraph cloner automatically merge global nodes | Chris Lattner | 2002-10-17 | 3 | -47/+15 |
| | | | | | | | | * BUClosure doesn't have to worry about global nodes * TDClosure now works with global nodes * Reenable DNE on TD pass, now that globals work right llvm-svn: 4220 | ||||
* | Remove obsolete code | Chris Lattner | 2002-10-17 | 2 | -35/+0 |
| | | | | llvm-svn: 4218 | ||||
* | Enable incompleteness marking | Chris Lattner | 2002-10-17 | 1 | -2/+2 |
| | | | | llvm-svn: 4217 | ||||
* | * First try at implementing TD pass this does not merge global nodes yet, | Chris Lattner | 2002-10-17 | 1 | -117/+82 |
| | | | | | | | | among other things. * Significant rewrite of TD pass to avoid potentially N^2 algorithms if possible. It is still not complete, but at least it's checked in now. llvm-svn: 4215 | ||||
* | Reenable printing of TD analysis | Chris Lattner | 2002-10-17 | 1 | -3/+2 |
| | | | | llvm-svn: 4214 | ||||
* | * Add data structures and code to track the call sites for each function | Chris Lattner | 2002-10-17 | 1 | -2/+8 |
| | | | | llvm-svn: 4213 | ||||
* | Cleanup data structure graph printer, eliminate hard coded printing in | Chris Lattner | 2002-10-17 | 1 | -120/+18 |
| | | | | | | favor of generic printer. llvm-svn: 4209 | ||||
* | Make sure to print out the call nodes as well | Chris Lattner | 2002-10-16 | 1 | -1/+14 |
| | | | | llvm-svn: 4203 | ||||
* | * Print the "return" node in the graphs | Chris Lattner | 2002-10-16 | 1 | -1/+18 |
| | | | | llvm-svn: 4199 | ||||
* | The second element of the iterator is really an offset, not a link | Chris Lattner | 2002-10-16 | 1 | -1/+1 |
| | | | | llvm-svn: 4196 | ||||
* | Specify the graph name | Chris Lattner | 2002-10-16 | 1 | -1/+1 |
| | | | | llvm-svn: 4195 | ||||
* | - DSGraph Printing Improvements: | Chris Lattner | 2002-10-16 | 1 | -7/+7 |
| | | | | | | | * Print edge source labels again * Override node shape to be Mrecord again, instead of just record. llvm-svn: 4193 | ||||
* | Halfway conversion from custom printing to GraphWriter printing | Chris Lattner | 2002-10-13 | 1 | -0/+33 |
| | | | | llvm-svn: 4146 | ||||
* | Use PARALLEL_DIRS. | Vikram S. Adve | 2002-10-11 | 1 | -1/+1 |
| | | | | llvm-svn: 4114 | ||||
* | Added capability to get execution count of a loop if it is a predictable | Misha Brukman | 2002-10-11 | 1 | -11/+137 |
| | | | | | | number of iterations. llvm-svn: 4113 | ||||
* | Added helper functions in LoopInfo: isLoopExit and numBackEdges. | Misha Brukman | 2002-10-11 | 1 | -0/+24 |
| | | | | llvm-svn: 4112 | ||||
* | Handle post dominance correctly in the case where blocks do not have a path to | Chris Lattner | 2002-10-04 | 1 | -0/+12 |
| | | | | | | the exit node. llvm-svn: 4038 | ||||
* | Prune function nodes that are no longer referenced due to inlining | Chris Lattner | 2002-10-03 | 2 | -2/+3 |
| | | | | llvm-svn: 4036 | ||||
* | sgefa uses truely huge data structures nodes. Only print part of them if they | Chris Lattner | 2002-10-03 | 1 | -2/+7 |
| | | | | | | are so big llvm-svn: 4035 | ||||
* | Handle bug exposed by power benchmark | Chris Lattner | 2002-10-03 | 1 | -1/+1 |
| | | | | llvm-svn: 4033 | ||||
* | Reimplement/port the Bottom Up Closure pass | Chris Lattner | 2002-10-03 | 2 | -38/+61 |
| | | | | llvm-svn: 4031 | ||||
* | DataStructure.h doesn't include DSGraph.h | Chris Lattner | 2002-10-02 | 3 | -0/+3 |
| | | | | llvm-svn: 4029 | ||||
* | * Implement fully general merging of array subscripts on demand! This | Chris Lattner | 2002-10-02 | 1 | -3/+27 |
| | | | | | | does not handle the initial pointer index case yet though. llvm-svn: 4012 | ||||
* | When printing DS nodes, print the mergemap index as well to allow easier | Chris Lattner | 2002-10-02 | 1 | -1/+1 |
| | | | | | | debugging of merging process. llvm-svn: 4010 | ||||
* | * Significant rework of DSNode to support arbitrary aliasing due to merging | Chris Lattner | 2002-10-02 | 1 | -20/+146 |
| | | | | | | | | * Now all and any bytes of a DSNode can be merged together individually. This is neccesary to support the full generality of C and support aliasing correctly. llvm-svn: 4008 | ||||
* | Initial checkin of Steensgaards context insensitive flow insensitive | Chris Lattner | 2002-10-01 | 1 | -0/+224 |
| | | | | | | alias analysis llvm-svn: 3997 | ||||
* | Checkin some major reworks of data structure analysis. This is not done, | Chris Lattner | 2002-10-01 | 5 | -353/+522 |
| | | | | | | | nor does it work very well, but I need to get it checked in before I break the tree unintentionally. llvm-svn: 3996 | ||||
* | Minor tweak | Chris Lattner | 2002-09-29 | 1 | -2/+3 |
| | | | | llvm-svn: 3985 | ||||
* | Fix printing of loop information | Chris Lattner | 2002-09-29 | 1 | -2/+2 |
| | | | | llvm-svn: 3977 | ||||
* | First try at implementing the AliasSetTracker class. I'm sure it will need | Chris Lattner | 2002-09-26 | 1 | -0/+172 |
| | | | | | | revision as I start to use it though. llvm-svn: 3954 | ||||
* | Fix printing of loop information | Chris Lattner | 2002-09-26 | 1 | -2/+2 |
| | | | | llvm-svn: 3941 | ||||
* | - Add new methods to LoopInfo: getLoopPreheader, addBasicBlockToLoop. | Chris Lattner | 2002-09-26 | 1 | -1/+51 |
| | | | | | | | These allow extra information to be easily gathered, and loopinfo to be updated. llvm-svn: 3936 | ||||
* | Convert BasicVN to be an ImmutablePass | Chris Lattner | 2002-09-25 | 1 | -12/+1 |
| | | | | llvm-svn: 3924 | ||||
* | Make users of FindUsedTypes not have problems with linkage. This fixes | Chris Lattner | 2002-09-24 | 1 | -0/+3 |
| | | | | | | Cwriter. llvm-svn: 3900 | ||||
* | Clean up indvar printing | Chris Lattner | 2002-09-10 | 1 | -3/+3 |
| | | | | llvm-svn: 3650 | ||||
* | * Add capability to recognize alias properties of the following common cases: | Chris Lattner | 2002-09-08 | 1 | -12/+79 |
| | | | | | | | | | | | | | | | | | - A[c1] cannot alias A[c2] where constants c1 != c2 - A[i] cannot alias B[j] if A & B are provably different arrays This should help out array based codes. For example, from bzip2 from spec, 3 additional loads can be GCSE'd, and _21_ additional loads can be LICMd due to this change. In a test example from the Spec GAP benchmark (vecffe.c), this change allows _52_ additional loads to be GCSE'd and _224_ additional LICM'd loads. Not bad for such a simple change. Other testcases show no change at all because they just don't use arrays. Not too suprising there. llvm-svn: 3616 | ||||
* | - Renamed Type::isIntegral() to Type::isInteger() | Chris Lattner | 2002-09-03 | 1 | -3/+3 |
| | | | | | | | | - Added new method Type::isIntegral() that is the same as isInteger, but also accepts bool. SCVS: ---------------------------------------------------------------------- llvm-svn: 3572 |