summaryrefslogtreecommitdiffstats
path: root/llvm/lib/CodeGen/InstrSched
Commit message (Collapse)AuthorAgeFilesLines
* * Add a file header with some informationChris Lattner2002-04-091-35/+25
| | | | | | | | | | * Delete the DelaySlotInfo objects created by the SchedulingManager class. These leaked objects were accounting for 3/4 of the memory leaked by the backend, so this is a relatively major win. * Reorganize SchedulingManager::getDelaySlotInfoForInstr so that it has better code locality (making it easier to read). llvm-svn: 2197
* Clean up the ownership model a bit so that nodes actually get deleted moreChris Lattner2002-04-091-12/+8
| | | | | | | frequently. This still leaks edges quite a bit, but it leaks no nodes (I think). llvm-svn: 2190
* s/Method/FunctionChris Lattner2002-04-082-3/+3
| | | | llvm-svn: 2180
* Change references to the Method class to be references to the FunctionChris Lattner2002-04-071-8/+8
| | | | | | | class. The Method class is obsolete (renamed) and all references to it are being converted over to Function. llvm-svn: 2144
* Use deterministic iterator for SchedGraphs. This is actually notVikram S. Adve2002-03-241-24/+10
| | | | | | useful right now when we only do local scheduling. llvm-svn: 1975
* Bug fix: nextToTry was not being initialized in one case.Vikram S. Adve2002-03-241-3/+5
| | | | llvm-svn: 1974
* Add option to disable scheduling.Vikram S. Adve2002-03-241-0/+1
| | | | llvm-svn: 1973
* Add option to disable scheduling.Vikram S. Adve2002-03-241-39/+53
| | | | | | | | Destroy live-variable information after scheduling so it is recomputed before later phases (e.g., reg. allocation). Use deterministic iterator to enumerate sched graphs. llvm-svn: 1972
* Minor changes.Vikram S. Adve2002-03-241-7/+3
| | | | llvm-svn: 1971
* Rename Method to FunctionChris Lattner2002-03-232-6/+7
| | | | llvm-svn: 1957
* Change to make it build with GCC 2.95.3Anand Shukla2002-02-261-1/+1
| | | | llvm-svn: 1802
* * Silence signed/unsigned warningsChris Lattner2002-02-242-3/+4
| | | | | | * Make it compile with GCC 3.0.4 llvm-svn: 1787
* * Pull BasicBlock::pred_* and BasicBlock::succ_* out of BasicBlock.h and intoChris Lattner2002-02-122-14/+12
| | | | | | | | | llvm/Support/CFG.h * Make pred & succ iterators for intervals global functions * Add #includes that are now neccesary because BasicBlock.h doesn't include InstrTypes.h anymore llvm-svn: 1750
* Method.h no longer includes BasicBlock.hChris Lattner2002-02-121-0/+1
| | | | | | | Method::inst_* is now in llvm/Support/InstIterator.h GraphTraits specializations for BasicBlock and Methods are now in llvm/Support/CFG.h llvm-svn: 1746
* Convert operand iterator over to work like an STL iteratorChris Lattner2002-02-051-7/+7
| | | | llvm-svn: 1720
* Minor change: Methods that return ValueSet's that are guaranteed to be validChris Lattner2002-02-051-7/+5
| | | | | | return references instead of pointers. llvm-svn: 1719
* * Eliminate the LiveVarSet class, making applyTranferFuncForMInst a staticChris Lattner2002-02-052-10/+7
| | | | | | | | | function in the one .cpp file that uses it. Use ValueSet's instead. * Prepare to delete LiveVarSet.h & LiveVarSet.cpp * Eliminate the ValueSet class, making all old member functions into global templates that will eventually be moved to Support. llvm-svn: 1711
* * Add a #include not indirectly included any moreChris Lattner2002-02-051-0/+1
| | | | llvm-svn: 1707
* * The itf exposed by InstrScheduling is now a single function to create the ↵Chris Lattner2002-02-043-50/+62
| | | | | | | | | | | | right pass * InstructionScheduling is now a real pass * InstrSched _uses_ LiveVar analysis, instead of creating it's own copy many times through a loop. In this was LiveVarAnalysis is actually even SHARED by Register allocation. * SchedPriorities is now passed the live var information in llvm-svn: 1700
* Remove dependence on BBLiveVar from InstrSchedulingChris Lattner2002-02-042-6/+4
| | | | | | Minor cleanups llvm-svn: 1688
* Move instrIsFeasible from InstrScheduling.hChris Lattner2002-02-041-0/+15
| | | | llvm-svn: 1674
* * Move command line parsing debug stuff to InstrScheduling internal header fileChris Lattner2002-02-041-1/+10
| | | | | | SchedPriorities.h llvm-svn: 1669
* Remove extraneous #includesChris Lattner2002-02-031-2/+0
| | | | llvm-svn: 1645
* * Straighten out #includesChris Lattner2002-02-031-10/+7
| | | | | | * Switch from MachineCodeForVMInstr to MachineCodeForInstruction model llvm-svn: 1644
* * Get machine instrs from Instructin's by using MachineCodeForInstruction::getChris Lattner2002-02-031-10/+10
| | | | | | * Convert a few (*X). to X-> llvm-svn: 1643
* Make it build with 2.95.3Chris Lattner2002-01-231-1/+1
| | | | llvm-svn: 1554
* Changes to build successfully with GCC 3.02Chris Lattner2002-01-205-187/+153
| | | | llvm-svn: 1503
* Renamed inst_const_iterator -> const_inst_iteratorChris Lattner2001-12-041-1/+1
| | | | | | | Renamed op_const_iterator -> const_op_iterator Renamed PointerType::getValueType() -> PointerType::getElementType() llvm-svn: 1408
* Create a new #include "Support/..." directory structure to move thingsChris Lattner2001-11-274-29/+17
| | | | | | | | from "llvm/Support/..." that are not llvm dependant. Move files and fix #includes llvm-svn: 1400
* Stuff after a #endif is not valid CPP input and breaks some compilersChris Lattner2001-11-261-3/+3
| | | | llvm-svn: 1375
* Eliminate most uses of the machine instruction vector for each LLVM instr,Vikram S. Adve2001-11-122-63/+75
| | | | | | | | since some m. instr. may be generated by LLVM instrs. in other blocks. Handle non-SSA (anti and output) edges and true edges uniformly by working with machine instructions alone. llvm-svn: 1269
* Major improvement to how nodes are built for a BB.Vikram S. Adve2001-11-124-37/+75
| | | | | | LLVM instruction is no longer recorded in each node, but BB is. llvm-svn: 1262
* Only add true dep. edges from an earlier to a later instruction.Vikram S. Adve2001-11-112-31/+40
| | | | | | | This wasn't a problem until we started putting copies for Phi values that produced cycles in the SchedGraph! llvm-svn: 1254
* Use node->getOpCode() instead of node->getMachineInstr()->getOpCode().Vikram S. Adve2001-11-091-36/+38
| | | | | | Much nicer. llvm-svn: 1223
* Major change to how defs are found when adding dependences (theyVikram S. Adve2001-11-082-171/+69
| | | | | | | | are now found as part of the initial walk of the machine code). Also memory load/store instructions can be generated for non-memory LLVM instructions, which wasn't handled before. It is now. llvm-svn: 1199
* MachineCodeForMethod is now an annotation on class Method.Vikram S. Adve2001-11-081-1/+1
| | | | llvm-svn: 1198
* Modified graph construction to use one pass to find all defs.Vikram S. Adve2001-11-052-56/+144
| | | | | | | Avoids having to handle some special cases that cause complex interactions with instr. selection. llvm-svn: 1138
* Allow combinations of True/Anti/Output flags for each edge toVikram S. Adve2001-10-281-7/+16
| | | | | | support, e.g., dependences on Call instructions. llvm-svn: 1009
* Add edges between call instructions and (a) load/store instructions, andVikram S. Adve2001-10-281-34/+111
| | | | | | | (b) any instructions that use or set CC registers. Whether or not the latter are needed really should be machine-dependent. llvm-svn: 1008
* Modify code that processes delay slots so that it preserves anyVikram S. Adve2001-10-221-11/+36
| | | | | | useful instructions already inserted into delay slots. llvm-svn: 945
* Separate VM instruction that generates the instructions that compute a valueVikram S. Adve2001-10-171-1/+2
| | | | | | | from the value itself (the one causing an edge) because the latter may be a temporary used within the instruction sequence for the VM instruction. llvm-svn: 858
* *** empty log message ***Vikram S. Adve2001-10-171-15/+24
| | | | llvm-svn: 857
* * Fix privacy issues on RegToRefVecMapChris Lattner2001-10-131-5/+5
| | | | | | * Fix initialization order problems... llvm-svn: 762
* Add graph edges due to implicit refs in each machine instruction.Vikram S. Adve2001-10-111-27/+52
| | | | llvm-svn: 724
* Don't insert useful instructions in delay slot of a RETURN.Vikram S. Adve2001-10-102-27/+32
| | | | llvm-svn: 721
* Commit more code over to new cast styleChris Lattner2001-10-021-2/+3
| | | | llvm-svn: 697
* Convert more code to use new style castsChris Lattner2001-10-011-2/+1
| | | | | | Eliminate old style casts from value.h llvm-svn: 696
* Add support for new style castsChris Lattner2001-10-011-1/+1
| | | | llvm-svn: 694
* Bug fixes:Vikram S. Adve2001-09-301-378/+429
| | | | | | | | | | | (1) Ensure that delay slot instructions are not moved out of place (this was happening for some CALL instructions). Basically, we need to move all delay slot instructions out of the graph and handle them along with the delayed control transfer instruction. (2) Mark scheduled instructions correctly when instructions are scheduled in more than one cycle in a single step (due to delay slots). llvm-svn: 678
* Minor changes for bug fixes in SchedGraph.cpp.Vikram S. Adve2001-09-301-9/+15
| | | | llvm-svn: 677
OpenPOWER on IntegriCloud