Commit Graph

243 Commits

Author SHA1 Message Date
Jakob Stoklund Olesen 53e2e48de7 VirtRegMap is counting spill slots, not register spills.
Fix the stats counters to reflect that.

llvm-svn: 139819
2011-09-15 18:31:13 +00:00
Jakob Stoklund Olesen 6cc4e4ddd0 Also count identity copies.
llvm-svn: 130996
2011-05-06 17:59:57 +00:00
Jakob Stoklund Olesen 71d3b895ba Also add <imp-def> operands for defined and dead super-registers when rewriting.
We cannot rely on the <imp-def> operands added by LiveIntervals in all cases as
demonstrated by the test case.

llvm-svn: 130313
2011-04-27 17:42:31 +00:00
Jakob Stoklund Olesen e6e6750670 Don't completely eliminate identity copies that also modify super register liveness.
Turn them into noop KILL instructions instead. This lets the scavenger know when
super-registers are killed and defined.

llvm-svn: 128645
2011-03-31 17:55:25 +00:00
Jakob Stoklund Olesen 710656d7b3 Dump the register map before rewriting.
llvm-svn: 128143
2011-03-23 04:32:49 +00:00
Jakob Stoklund Olesen 5bfec69b1d Add VirtRegMap::rewrite() and use it in the new register allocators.
The rewriter works almost identically to -rewriter=trivial, except it also
eliminates any identity copies.

This makes the new register allocators independent of VirtRegRewriter.cpp which
will be going away at the same time as RegAllocLinearScan.

llvm-svn: 125967
2011-02-18 22:03:18 +00:00
Jakob Stoklund Olesen 2fb5b31578 Simplify a bunch of isVirtualRegister() and isPhysicalRegister() logic.
These functions not longer assert when passed 0, but simply return false instead.

No functional change intended.

llvm-svn: 123155
2011-01-10 02:58:51 +00:00
Jakob Stoklund Olesen d82ac37594 Remove MachineRegisterInfo::getLastVirtReg(), it was giving wrong results
when no virtual registers have been allocated.

It was only used to resize IndexedMaps, so provide an IndexedMap::resize()
method such that

 Map.grow(MRI.getLastVirtReg());

can be replaced with the simpler

 Map.resize(MRI.getNumVirtRegs());

This works correctly when no virtuals are allocated, and it bypasses the to/from
index conversions.

llvm-svn: 123130
2011-01-09 21:58:20 +00:00
Jakob Stoklund Olesen 1331a15b0c Replace TargetRegisterInfo::printReg with a PrintReg class that also works without a TRI instance.
Print virtual registers numbered from 0 instead of the arbitrary
FirstVirtualRegister. The first virtual register is printed as %vreg0.
TRI::NoRegister is printed as %noreg.

llvm-svn: 123107
2011-01-09 03:05:53 +00:00
Jakob Stoklund Olesen cf4d5ced0f Fix VirtRegMap to use TRI::index2VirtReg and TRI::virtReg2Index instead of
depending on TRI::FirstVirtualRegister.

Also use TRI::printReg instead of printing virtual registers directly.

llvm-svn: 123101
2011-01-08 23:11:07 +00:00
Jakob Stoklund Olesen 39aed737a6 Remember to resize SpillSlotToUsesMap when allocating an emergency spill slot.
Use amazing new function call technology instead of writing identical code in
multiple places.

This fixes PR8604.

llvm-svn: 119306
2010-11-16 00:41:01 +00:00
Owen Anderson df7a4f2515 Now with fewer extraneous semicolons!
llvm-svn: 115996
2010-10-07 22:25:06 +00:00
Owen Anderson a57b97e7e7 Fix batch of converting RegisterPass<> to INTIALIZE_PASS().
llvm-svn: 109045
2010-07-21 22:09:45 +00:00
Jakob Stoklund Olesen ddbf7a858e Use the right floating point load/store instructions in PPCInstrInfo::foldMemoryOperandImpl().
The PowerPC floating point registers can represent both f32 and f64 via the
two register classes F4RC and F8RC. F8RC is considered a subclass of F4RC to
allow cross-class coalescing. This coalescing only affects whether registers
are spilled as f32 or f64.

Spill slots must be accessed with load/store instructions corresponding to the
class of the spilled register. PPCInstrInfo::foldMemoryOperandImpl was looking
at the instruction opcode which is wrong.

X86 has similar floating point register classes, but doesn't try to fold
memory operands, so there is no problem there.

llvm-svn: 97262
2010-02-26 21:09:24 +00:00
Dan Gohman 4a618827de Fix "the the" and similar typos.
llvm-svn: 95781
2010-02-10 16:03:48 +00:00
David Greene 91c42f851b Change errs() to dbgs().
llvm-svn: 92587
2010-01-05 01:25:45 +00:00
David Greene 1fbe054450 Add a bool flag to StackObjects telling whether they reference spill
slots.  The AsmPrinter will use this information to determine whether to
print a spill/reload comment.

Remove default argument values.  It's too easy to pass a wrong argument
value when multiple arguments have default values.  Make everything
explicit to trap bugs early.

Update all targets to adhere to the new interfaces..

llvm-svn: 87022
2009-11-12 20:49:22 +00:00
Lang Hames 05fb9637f6 The Indexes Patch.
This introduces a new pass, SlotIndexes, which is responsible for numbering
instructions for register allocation (and other clients). SlotIndexes numbering
is designed to match the existing scheme, so this patch should not cause any
changes in the generated code.

For consistency, and to avoid naming confusion, LiveIndex has been renamed
SlotIndex.

The processImplicitDefs method of the LiveIntervals analysis has been moved
into its own pass so that it can be run prior to SlotIndexes. This was
necessary to match the existing numbering scheme.

llvm-svn: 85979
2009-11-03 23:52:08 +00:00
Evan Cheng 4729191bb2 Distinquish stack slots from other stack objects. They (and fixed objects) get FixedStack PseudoSourceValues.
llvm-svn: 84326
2009-10-17 09:20:14 +00:00
Chris Lattner 471ba48cb9 remove some uses of llvm/Support/Streams.h
llvm-svn: 79842
2009-08-23 08:43:55 +00:00
Daniel Dunbar 796e43eede Move more to raw_ostream, provide support for writing MachineBasicBlock,
LiveInterval, etc to raw_ostream.

llvm-svn: 76965
2009-07-24 10:36:58 +00:00
Evan Cheng 1283c6a066 Part 1.
- Change register allocation hint to a pair of unsigned integers. The hint type is zero (which means prefer the register specified as second part of the pair) or entirely target dependent.
- Allow targets to specify alternative register allocation orders based on allocation hint.

Part 2.
- Use the register allocation hint system to implement more aggressive load / store multiple formation.
- Aggressively form LDRD / STRD. These are formed *before* register allocation. It has to be done this way to shorten live interval of base and offset registers. e.g.
v1025 = LDR v1024, 0
v1026 = LDR v1024, 0
=>
v1025,v1026 = LDRD v1024, 0

If this transformation isn't done before allocation, v1024 will overlap v1025 which means it more difficult to allocate a register pair.

- Even with the register allocation hint, it may not be possible to get the desired allocation. In that case, the post-allocation load / store multiple pass must fix the ldrd / strd instructions. They can either become ldm / stm instructions or back to a pair of ldr / str instructions.

This is work in progress, not yet enabled.

llvm-svn: 73381
2009-06-15 08:28:29 +00:00
Evan Cheng 085caf10be Move register allocation preference (or hint) from LiveInterval to MachineRegisterInfo. This allows more passes to set them.
llvm-svn: 73346
2009-06-14 20:22:55 +00:00
Evan Cheng 3f77805642 Make sure to color with only allocatable registers for the specific register class.
llvm-svn: 70821
2009-05-04 03:30:11 +00:00
Evan Cheng 210fc62a91 In some rare cases, the register allocator can spill registers but end up not utilizing registers at all. The fundamental problem is linearscan's backtracking can end up freeing more than one allocated registers. However, reloads and restores might be folded into uses / defs and freed registers might not be used at all.
VirtRegMap keeps track of allocations so it knows what's not used. As a horrible hack, the stack coloring can color spill slots with *free* registers. That is, it replace reload and spills with copies from and to the free register. It unfold instructions that load and store the spill slot and replace them with register using variants.

Not yet enabled. This is part 1. More coming.

llvm-svn: 70787
2009-05-03 18:32:42 +00:00
Bill Wendling b8017e02ca Oy! When reverting r68073, I added in experimental code. Sorry...
llvm-svn: 68099
2009-03-31 08:41:31 +00:00
Bill Wendling c4b08e5eb0 Revert r68073. It's causing a failure in the Apple-style builds.
llvm-svn: 68092
2009-03-31 08:26:26 +00:00
Daniel Dunbar f24f26c75a Add newlines at end of file (this can annoy gcov)
llvm-svn: 67000
2009-03-14 01:53:05 +00:00
Owen Anderson d37ddf5b5b Convert VirtRegMap to a MachineFunctionPass.
llvm-svn: 66870
2009-03-13 05:55:11 +00:00
Owen Anderson aabe06d92a Reorganization: Move the Spiller out of VirtRegMap.cpp into its own files. No (intended) functionality change.
llvm-svn: 66720
2009-03-11 22:31:21 +00:00
Evan Cheng fb8ded911e Yet another case where the spiller marked two uses of the same register on the same instruction as kill. This fixes PR3706.
llvm-svn: 66428
2009-03-09 19:00:05 +00:00
Evan Cheng de22116f39 If a MI uses the same register more than once, only mark one of them as 'kill'.
llvm-svn: 66363
2009-03-08 03:58:35 +00:00
Evan Cheng 398dee1c4a Last commit accidentially deleted this code.
llvm-svn: 65679
2009-02-28 06:02:14 +00:00
Evan Cheng ca2d65467b The last commit was overly conservative. It's ok to reuse value that's already marked livein.
llvm-svn: 65498
2009-02-26 03:02:21 +00:00
Evan Cheng ee5fd035e2 If an available register falls through to a succ block, unset the last kill. Sorry, it's impossible to reduce a sensible test case. It basically requires the moon and stars to align in order to cause a failure.
llvm-svn: 65497
2009-02-26 02:30:42 +00:00
Evan Cheng f505cd5ebb A couple of places where reused use operands should be marked kill. This is exposed by recent availability fallthrough changes.
llvm-svn: 64745
2009-02-17 06:41:03 +00:00
Bill Wendling 65c0fd4c44 Revert this. It was breaking stuff.
llvm-svn: 64428
2009-02-13 02:16:35 +00:00
Bill Wendling 1c21ac3066 Turn off the old way of handling debug information in the code generator. Use
the new way, where all of the information is passed on SDNodes and machine
instructions.

llvm-svn: 64427
2009-02-13 02:01:04 +00:00
Dan Gohman e366c29fa0 Adjust the sizes for a few SmallVectors to reflect their usage.
llvm-svn: 64381
2009-02-12 17:29:01 +00:00
Evan Cheng cf5cd6ecfe It's (currently) not safe to keep certain physical registers live across basic blocks, e.g. x86 fp stack registers.
llvm-svn: 64374
2009-02-12 10:32:17 +00:00
Evan Cheng 8b39b545a9 If availability info is kept when fallthrough into a bb, add the available registers to live-in set.
llvm-svn: 64372
2009-02-12 09:43:23 +00:00
Evan Cheng eb5ec4a0db Remove a bogus assertion. It's possible a live-in available value is used by a previous instruction.
llvm-svn: 64339
2009-02-11 23:41:57 +00:00
Evan Cheng a1968b0fc7 Implement PR3495: local spiller optimization. The local spiller can now keep availability information over BB boundaries. It visits BB's in depth first order. After visiting a BB if it find a successor which has a single predecessor it visits the successor next without clearing the availability information. This allows the successor to omit reloads or change them into copies.
llvm-svn: 64298
2009-02-11 08:24:21 +00:00
Bill Wendling 67cd395d45 Explicitly pass in debug location information to BuildMI.
llvm-svn: 63599
2009-02-03 02:29:34 +00:00
Evan Cheng c544cb0eca Change TargetInstrInfo::isMoveInstr to return source and destination sub-register indices as well.
llvm-svn: 62600
2009-01-20 19:12:24 +00:00
Evan Cheng 1718fd4375 Fix PR3124: overly strict assert.
llvm-svn: 60392
2008-12-02 02:15:36 +00:00
Evan Cheng 94169f1021 Fix PR2898. Spiller delete a store for reuse before it knows for sure the reuse happened.
Patch by Lang Hames!

llvm-svn: 57720
2008-10-17 20:56:41 +00:00
Evan Cheng 08acb24225 Fix a very subtle spiller bug: UpdateKills should not forget to track defs of aliases.
llvm-svn: 57673
2008-10-17 06:16:07 +00:00
Dan Gohman 9c4b7d5c4f Fix command-line option printing to print two spaces where needed,
instead of requiring all "short description" strings to begin with
two spaces. This makes these strings less mysterious, and it fixes
some cases where short description strings mistakenly did not
begin with two spaces.

llvm-svn: 57521
2008-10-14 20:25:08 +00:00
Dan Gohman dfe979bcb0 Don't dereference the end() iterator. Thanks to
ENABLE_EXPENSIVE_CHECKS for finding this.

llvm-svn: 57181
2008-10-06 18:00:07 +00:00