Commit Graph

53 Commits

Author SHA1 Message Date
Jakob Stoklund Olesen 2d60075590 Be more strict when detecting critical edges before loop splitting.
An exit block with a critical edge must only have predecessors in the loop, or
just before the loop. This guarantees that the inserted copies in the loop
predecessors dominate the exit block.

llvm-svn: 117144
2010-10-22 20:28:23 +00:00
Jakob Stoklund Olesen 9a74301621 Add print methods
llvm-svn: 117143
2010-10-22 20:28:21 +00:00
Jakob Stoklund Olesen f4bbe50fc3 Don't include the destination interval in the union when computing
Parent - union(Y, ...). Doh.

llvm-svn: 117042
2010-10-21 18:47:08 +00:00
Jakob Stoklund Olesen 3f1f7b67e3 Eliminate curli from SplitEditor. Use the LiveRangeEdit reference instead.
llvm-svn: 116547
2010-10-15 00:34:01 +00:00
Jakob Stoklund Olesen 0f3e98ce2e Move stack slot assignments into LiveRangeEdit.
All registers created during splitting or spilling are assigned to the same
stack slot as the parent register.

When splitting or rematting, we may not spill at all. In that case the stack
slot is still assigned, but it will be dead.

llvm-svn: 116546
2010-10-15 00:16:55 +00:00
Jakob Stoklund Olesen 72911e49fa Create a new LiveRangeEdit class to keep track of the new registers created when
splitting or spillling, and to help with rematerialization.

Use LiveRangeEdit in InlineSpiller and SplitKit. This will eventually make it
possible to share remat code between InlineSpiller and SplitKit.

llvm-svn: 116543
2010-10-14 23:49:52 +00:00
Jakob Stoklund Olesen f11318018a Only split around a loop if the live range has uses outside the loop periphery.
Before we would also split around a loop if any peripheral block had multiple
uses. This could cause repeated splitting when splitting a different live range
would insert uses into the periphery.

Now -spiller=inline passes the nightly test suite again.

llvm-svn: 116494
2010-10-14 18:26:45 +00:00
Jakob Stoklund Olesen 959fcc6c63 Rename SplitEditor::rewrite to finish() and break it out into a couple of new
functions: computeRemainder and rewrite.

When the remainder breaks up into multiple components, remember to rewrite those
uses as well.

llvm-svn: 116121
2010-10-08 23:42:21 +00:00
Jakob Stoklund Olesen 0f1677e190 After splitting, the remaining LiveInterval may be fragmented into multiple
connected components. These components should be allocated different virtual
registers because there is no reason for them to be allocated together.

Add the ConnectedVNInfoEqClasses class to calculate the connected components,
and move values to new LiveIntervals.

Use it from SplitKit::rewrite by creating new virtual registers for the
components.

llvm-svn: 116006
2010-10-07 23:34:34 +00:00
Jakob Stoklund Olesen 9bfd9679f9 Print more loop info.
llvm-svn: 115951
2010-10-07 18:47:07 +00:00
Jakob Stoklund Olesen 7c31730053 Print out MBB number when rewriting.
llvm-svn: 115950
2010-10-07 18:47:05 +00:00
Jakob Stoklund Olesen 49715fd494 Cache interval iterators in SplitEditor::addTruncSimpleRange so we only have to
do one find().

llvm-svn: 115929
2010-10-07 17:56:39 +00:00
Jakob Stoklund Olesen 9575af4b06 Clean up debug printing.
llvm-svn: 115928
2010-10-07 17:56:35 +00:00
Jakob Stoklund Olesen 4d5156c7d0 Count uses in all nested loops, not just the deepest.
llvm-svn: 115710
2010-10-05 23:10:12 +00:00
Jakob Stoklund Olesen 56e2925e6c Remove SplitAnalysis::removeUse. It was only used to make SplitAnalysis
reusable, but that is no longer relevant since a split will always replace the
original.

llvm-svn: 115709
2010-10-05 23:10:09 +00:00
Jakob Stoklund Olesen 0445e2a053 dupli always has an interval now.
llvm-svn: 115708
2010-10-05 23:10:04 +00:00
Jakob Stoklund Olesen 2dfa8be26a We can split around loops with multiple exits now.
llvm-svn: 115696
2010-10-05 22:19:35 +00:00
Jakob Stoklund Olesen 89d276aa48 Update SplitEditor API to reflect the fact that the original live interval is
never kept after splitting.

Keeping the original interval made sense when the split region doesn't modify
the register, and the original is spilled. We can get the same effect by
detecting reloaded values when spilling around copies.

llvm-svn: 115695
2010-10-05 22:19:33 +00:00
Jakob Stoklund Olesen b46d32367f Intervals are half-open.
llvm-svn: 115694
2010-10-05 22:19:29 +00:00
Jakob Stoklund Olesen 671bab1c7d When we find a reaching definition, make sure it is visited from all paths by
erasing it from the visited set. That ensures we create the right phi defs.

llvm-svn: 115666
2010-10-05 20:36:28 +00:00
Jakob Stoklund Olesen b0cedd5f96 Don't use nextIndex to check for live out of instruction.
Insert copy after defining instruction.

Fix LiveIntervalMap::extendTo to properly handle live segments starting before
the current basic block.

Make sure the open live range is extended to the inserted copy's use slot.

llvm-svn: 115665
2010-10-05 20:36:25 +00:00
Lang Hames 564956867e Removed VNInfo::isDefAccurate(). Def "accuracy" can be checked by testing whether LiveIntervals::getInstructionFromIndex(def) returns NULL.
llvm-svn: 114791
2010-09-25 12:04:16 +00:00
Jakob Stoklund Olesen 6f8bd42ec7 Build the complement interval dupli after the split intervals instead of
creating it before and subtracting split ranges.

This way, the SSA update code in LiveIntervalMap can properly create and use new
phi values in dupli. Now it is possible to create split regions where a value
escapes along two different CFG edges, creating phi values outside the split
region.

This is a work in progress and probably quite broken.

llvm-svn: 114492
2010-09-21 22:32:21 +00:00
Jakob Stoklund Olesen 9855109b65 Use the value mapping provided by LiveIntervalMap. This simplifies the code a
great deal because we don't have to worry about maintaining SSA form.

Unconditionally copy back to dupli when the register is live out of the split
range, even if the live-out value was defined outside the range. Skipping the
back-copy only makes sense when the live range is going to spill outside the
split range, and we don't know that it will. Besides, this was a hack to avoid
SSA update issues.

Clear up some confusion about the end point of a half-open LiveRange. Methinks
LiveRanges need to be closed so both start and end are included in the range.
The low bits of a SlotIndex are symbolic, so a half-open range doesn't really
make sense. This would be a pervasive change, though.

llvm-svn: 114043
2010-09-16 00:01:36 +00:00
Jakob Stoklund Olesen 614e13936a Mechanically replace LiveInterval* with LiveIntervalMap for intervals being
edited without actually using LiveIntervalMap functionality.

llvm-svn: 113816
2010-09-13 23:29:11 +00:00
Jakob Stoklund Olesen 36dad6db7c Allow LiveIntervalMap to be reused by resetting the current live interval.
llvm-svn: 113815
2010-09-13 23:29:09 +00:00
Jakob Stoklund Olesen 535e8e5f60 Let's just declare that it is impossible to construct a std::pair from a null
pointer and work around that.

llvm-svn: 113788
2010-09-13 21:29:45 +00:00
Michael J. Spencer abca173494 Fix the msvc 2010 build.
The Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 16.00.30319.01
implements parts of C++0x based on the draft standard. An old version of
the draft had a bug that makes std::pair<T1*, T2*>(something, 0) fail to
compile. This is because the template<class U, class V> pair(U&& x, V&& y)
constructor is selected, even though it later fails to implicitly convert
U and V to frist_type and second_type.

This has been fixed in n3090, but it seems that Microsoft is not going to
update msvc.

llvm-svn: 111535
2010-08-19 18:16:39 +00:00
Jakob Stoklund Olesen e98030ad58 Thinking about it, we don't need MachineDominatorTree after all. The DomValue
map discovers the iterated dominance frontier for free.

llvm-svn: 111400
2010-08-18 20:29:53 +00:00
Jakob Stoklund Olesen f4088b022a Revert r111394. It was too aggressive.
We must complete the DFS, otherwise we might miss needed phi-defs, and
prematurely color live ranges with a non-dominating value.

This is not a big deal since we get to color more of the CFG and the next
mapValue call will be faster.

llvm-svn: 111397
2010-08-18 20:06:05 +00:00
Jakob Stoklund Olesen 5b4cb08471 Aggressively prune the DFS when inserting phi-defs.
llvm-svn: 111394
2010-08-18 19:00:11 +00:00
Jakob Stoklund Olesen ce6f055b4d Add the LiveIntervalMap class. Don't hook it up yet.
LiveIntervalMap maps values from a parent LiveInterval to a child interval that
is a strict subset. It will create phi-def values as needed to preserve the
VNInfo SSA form in the child interval.

This leads to an algorithm very similar to the one in SSAUpdaterImpl.h, but with
enough differences that the code can't be reused:

- We don't need to manipulate PHI instructions.
- LiveIntervals have kills.
- We have MachineDominatorTree.
- We can use df_iterator.

llvm-svn: 111393
2010-08-18 19:00:08 +00:00
Jakob Stoklund Olesen 27e1f26534 Clean up the Spiller.h interface.
The earliestStart argument is entirely specific to linear scan allocation, and
can be easily calculated by RegAllocLinearScan.

Replace std::vector with SmallVector.

llvm-svn: 111055
2010-08-13 22:56:53 +00:00
Jakob Stoklund Olesen d1191ee43c Implement splitting inside a single block.
When a live range is contained a single block, we can split it around
instruction clusters. The current approach is very primitive, splitting before
and after the largest gap between uses.

llvm-svn: 111043
2010-08-13 21:18:48 +00:00
Jakob Stoklund Olesen 3d1027e7a1 Let LiveInterval::addRange extend existing ranges, it will verify that value
numbers match. The old check could accidentally leave holes in openli.

Also let useIntv add all ranges for the phi-def value inserted by
enterIntvAtEnd. This works as long at the value mapping is established in
enterIntvAtEnd.

llvm-svn: 110995
2010-08-13 01:05:26 +00:00
Jakob Stoklund Olesen 840b81a19e Remember to actually update SplitAnalysis statistics now that we have a fancy
function to do it.

llvm-svn: 110994
2010-08-13 01:05:23 +00:00
Jakob Stoklund Olesen 991e4ee860 Handle an empty dupli.
This can happen if the original interval has been broken into two disconnected
parts. Ideally, we should be able to detect when the graph is disconnected and
create separate intervals, but that code is not implemented yet.

Example:

Two basic blocks are both branching to a loop header. Our interval is defined in
both basic blocks, and live into the loop along both edges.

We decide to split the interval around the loop. The interval is split into an
inside part and an outside part. The outside part now has two disconnected
segments, one in each basic block.

If we later decide to split the outside interval into single blocks, we get one
interval per basic block and an empty dupli for the remainder.

llvm-svn: 110976
2010-08-12 23:02:57 +00:00
Jakob Stoklund Olesen 32c181c444 Update the SplitAnalysis statistics as uses are moved from curli to the new
split intervals. THis means the analysis can be used for multiple splits as long
as curli doesn't shrink.

llvm-svn: 110975
2010-08-12 23:02:55 +00:00
Jakob Stoklund Olesen 0910689353 Also recompute HasPHIKill flags in LiveInterval::RenumberValues.
If a phi-def value were removed from the interval, the phi-kill flags are no
longer valid.

llvm-svn: 110949
2010-08-12 20:38:03 +00:00
Jakob Stoklund Olesen fa3ea11ae6 Clean up debug output.
llvm-svn: 110940
2010-08-12 18:50:55 +00:00
Jakob Stoklund Olesen 622848b262 Implement single block splitting.
Before spilling a live range, we split it into a separate range for each basic
block where it is used. That way we only get one reload per basic block if the
new smaller ranges can allocate to a register.

This type of splitting is already present in the standard spiller.

llvm-svn: 110934
2010-08-12 17:07:14 +00:00
Jakob Stoklund Olesen 62e721478b More debug spew
llvm-svn: 110720
2010-08-10 20:45:01 +00:00
Jakob Stoklund Olesen 53c5022040 Implement register class inflation.
When splitting a live range, the new registers have fewer uses and the
permissible register class may be less constrained. Recompute the register class
constraint from the uses of new registers created for a split. This may let them
be allocated from a larger set, possibly avoiding a spill.

llvm-svn: 110703
2010-08-10 18:37:40 +00:00
Jakob Stoklund Olesen 284c2dbfd7 Recalculate the spill weight and allocation hint for virtual registers created
during live range splitting.

llvm-svn: 110686
2010-08-10 17:07:22 +00:00
Jakob Stoklund Olesen 45e07c8fc5 Lazily defer duplicating the live interval we are splitting until we know it is
necessary.

Sometimes, live range splitting doesn't shrink the current interval, but simply
changes some instructions to use a new interval. That makes the original more
suitable for spilling. In this case, we don't need to duplicate the original.

llvm-svn: 110481
2010-08-06 22:17:33 +00:00
Jakob Stoklund Olesen 8c0f693150 Add LiveInterval::RenumberValues - Garbage collection for VNInfos.
After heavy editing of a live interval, it is much easier to simply renumber the
live values instead of trying to keep track of the unused ones.

llvm-svn: 110463
2010-08-06 18:46:59 +00:00
Jakob Stoklund Olesen 7e0de5ef8e Fix swapped COPY operands.
llvm-svn: 110453
2010-08-06 18:04:17 +00:00
Jakob Stoklund Olesen dc96e28d70 Checkpoint SplitKit progress.
We are now at a point where we can split around simple single-entry, single-exit
loops, although still with some bugs.

llvm-svn: 110257
2010-08-04 22:08:39 +00:00
Jakob Stoklund Olesen c698417e52 Add SplitEditor to SplitKit. This class will be used to edit live intervals and
rewrite instructions for live range splitting.

Still work in progress.

llvm-svn: 109469
2010-07-26 23:44:11 +00:00
Jakob Stoklund Olesen 0fef9dda8e Change the createSpiller interface to take a MachineFunctionPass argument.
The spillers can pluck the analyses they need from the pass reference.

Switch some never-null pointers to references.

llvm-svn: 108969
2010-07-20 23:50:15 +00:00