llvm/lib/CodeGen
Chandler Carruth 51901d85f7 Prevent rotating the blocks of a loop (and thus getting a backedge to be
fallthrough) in cases where we might fail to rotate an exit to an outer
loop onto the end of the loop chain.

Having *some* rotation, but not performing this rotation, is the primary
fix of thep performance regression with -enable-block-placement for
Olden/em3d (a whopping 30% regression). Still working on reducing the
test case that actually exercises this and the new rotation strategy out
of this code, but I want to check if this regresses other test cases
first as that may indicate it isn't the correct fix.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145195 91177308-0d34-0410-b5e6-96231b3b80d8
2011-11-27 20:18:00 +00:00
..
AsmPrinter DISubrange supports unsigned lower/upper array bounds, so let's not fake it in the end while emitting DWARF. If a FE needs to encode signed lower/upper array bounds then we need to extend DISubrange or ad DISignedSubrange. 2011-11-17 23:43:15 +00:00
SelectionDAG Fix an obvious omission in the SelectionDAGBuilder where we were 2011-11-22 11:37:46 +00:00
AggressiveAntiDepBreaker.cpp
AggressiveAntiDepBreaker.h
AllocationOrder.cpp
AllocationOrder.h
Analysis.cpp land David Blaikie's patch to de-constify Type, with a few tweaks. 2011-07-18 04:54:35 +00:00
AntiDepBreaker.h
BranchFolding.cpp Reapply r142920 with fix: 2011-10-26 01:10:25 +00:00
BranchFolding.h
CalcSpillWeights.cpp Move CalculateRegClass to MRI::recomputeRegClass. 2011-08-09 16:46:27 +00:00
CallingConvLower.cpp
CMakeLists.txt Delete VirtRegRewriter. 2011-11-13 00:16:01 +00:00
CodeGen.cpp Prune more RALinScan. RALinScan was also here! 2011-11-13 01:33:10 +00:00
CodePlacementOpt.cpp
CriticalAntiDepBreaker.cpp
CriticalAntiDepBreaker.h
DeadMachineInstructionElim.cpp
DwarfEHPrepare.cpp This code is dead, what with the new EH model and the auto-upgraders in place. 2011-11-07 23:36:48 +00:00
EdgeBundles.cpp Twinify GraphWriter a little bit. 2011-11-15 16:26:38 +00:00
ELF.h
ELFCodeEmitter.cpp Fix asserts in CodeGen from: 2011-09-10 01:07:54 +00:00
ELFCodeEmitter.h Fix asserts in CodeGen from: 2011-09-10 01:07:54 +00:00
ELFWriter.cpp Goodbye TargetAsmInfo. This eliminate last bit of CodeGen and Target in llvm-mc. 2011-07-20 19:50:42 +00:00
ELFWriter.h
ExecutionDepsFix.cpp Fix PR11422. 2011-11-23 04:03:08 +00:00
ExpandISelPseudos.cpp
ExpandPostRAPseudos.cpp Give targets a chance to expand even standard pseudos. 2011-10-10 20:34:28 +00:00
GCMetadata.cpp Remove all remaining uses of Value::getNameStr(). 2011-11-15 16:27:03 +00:00
GCMetadataPrinter.cpp
GCStrategy.cpp Add a custom safepoint method, in order for language implementers to decide which machine instruction gets to be a safepoint. 2011-11-11 18:32:52 +00:00
IfConversion.cpp Added missing &. Fixes <rdar://problem/10393723> 2011-11-04 23:49:14 +00:00
InlineSpiller.cpp Use getVNInfoBefore() when it makes sense. 2011-11-14 01:39:36 +00:00
InterferenceCache.cpp Allow null interference cursors to be queried. 2011-07-23 03:10:17 +00:00
InterferenceCache.h Allow null interference cursors to be queried. 2011-07-23 03:10:17 +00:00
IntrinsicLowering.cpp land David Blaikie's patch to de-constify Type, with a few tweaks. 2011-07-18 04:54:35 +00:00
LatencyPriorityQueue.cpp
LexicalScopes.cpp Don't forget to reconstruct D after changing the scope that we're 2011-10-13 21:43:44 +00:00
LiveDebugVariables.cpp Rename SlotIndexes to match how they are used. 2011-11-13 20:45:27 +00:00
LiveDebugVariables.h
LiveInterval.cpp Use getVNInfoBefore() when it makes sense. 2011-11-14 01:39:36 +00:00
LiveIntervalAnalysis.cpp Fix early-clobber handling in shrinkToUses. 2011-11-14 18:45:38 +00:00
LiveIntervalUnion.cpp Simplify the interference checking code a bit. 2011-08-12 00:22:04 +00:00
LiveIntervalUnion.h Simplify the interference checking code a bit. 2011-08-12 00:22:04 +00:00
LiveRangeCalc.cpp Switch extendInBlock() to take a kill slot instead of the last use slot. 2011-09-13 16:47:56 +00:00
LiveRangeCalc.h Unbreak msvc. 2011-09-13 03:58:34 +00:00
LiveRangeEdit.cpp Rename SlotIndexes to match how they are used. 2011-11-13 20:45:27 +00:00
LiveRangeEdit.h Migrate LLVM and Clang to use the new makeArrayRef(...) functions where previously explicit non-default constructors were used. 2011-07-18 12:00:32 +00:00
LiveStackAnalysis.cpp Move getCommonSubClass() into TRI. 2011-09-30 22:18:51 +00:00
LiveVariables.cpp Silence a bunch (but not all) "variable written but not read" warnings 2011-08-12 14:54:45 +00:00
LLVMBuild.txt build: Add initial cut at LLVMBuild.txt files. 2011-11-03 18:53:17 +00:00
LLVMTargetMachine.cpp Sink codegen optimization level into MCCodeGenInfo along side relocation model 2011-11-16 08:38:26 +00:00
LocalStackSlotAllocation.cpp
MachineBasicBlock.cpp Handle the case of a no-return invoke correctly. It actually still has 2011-11-23 08:23:54 +00:00
MachineBlockFrequencyInfo.cpp Add more constantness in BlockFrequencyInfo. 2011-08-03 21:30:57 +00:00
MachineBlockPlacement.cpp Prevent rotating the blocks of a loop (and thus getting a backedge to be 2011-11-27 20:18:00 +00:00
MachineBranchProbabilityInfo.cpp Reuse the logic in getEdgeProbability within getHotSucc in order to 2011-11-14 08:55:59 +00:00
MachineCSE.cpp We need to verify that the machine instruction we're using as a replacement for 2011-10-12 23:03:40 +00:00
MachineDominators.cpp
MachineFunction.cpp Remove all remaining uses of Value::getNameStr(). 2011-11-15 16:27:03 +00:00
MachineFunctionAnalysis.cpp Sink codegen optimization level into MCCodeGenInfo along side relocation model 2011-11-16 08:38:26 +00:00
MachineFunctionPass.cpp
MachineFunctionPrinterPass.cpp
MachineInstr.cpp Added invariant field to the DAG.getLoad method and changed all calls. 2011-11-08 18:42:53 +00:00
MachineLICM.cpp Rename MVT::untyped to MVT::Untyped to match similar nomenclature. 2011-11-16 01:02:57 +00:00
MachineLoopInfo.cpp
MachineLoopRanges.cpp
MachineModuleInfo.cpp Add an ivar that maps a landing pad's EH symbol to the call sites that may jump 2011-10-05 22:20:38 +00:00
MachineModuleInfoImpls.cpp
MachinePassRegistry.cpp
MachineRegisterInfo.cpp Also inflate register classes around inline asm. 2011-10-12 23:37:40 +00:00
MachineSink.cpp While sinking machine instructions, sink matching DBG_VALUEs also otherwise live debug variable pass will drop DBG_VALUEs on the floor. 2011-09-07 00:07:58 +00:00
MachineSSAUpdater.cpp
MachineVerifier.cpp Remove all remaining uses of Value::getNameStr(). 2011-11-15 16:27:03 +00:00
Makefile
ObjectCodeEmitter.cpp
OcamlGC.cpp
OptimizePHIs.cpp When deleting a phi cycle after looking through copies, constrain the register 2011-10-17 21:54:46 +00:00
Passes.cpp Delete the linear scan register allocator. 2011-11-12 22:39:45 +00:00
PeepholeOptimizer.cpp If MI is deleted then remove it from the set. If a new MI is created, it could 2011-10-13 02:16:18 +00:00
PHIElimination.cpp Add an isSSA() flag to MachineRegisterInfo. 2011-07-29 22:51:22 +00:00
PHIEliminationUtils.cpp
PHIEliminationUtils.h
PostRASchedulerList.cpp Remove all remaining uses of Value::getNameStr(). 2011-11-15 16:27:03 +00:00
ProcessImplicitDefs.cpp Handle REG_SEQUENCE with implicitly defined operands. 2011-07-28 21:38:51 +00:00
PrologEpilogInserter.cpp A few 80-col violations. 2011-10-14 20:36:23 +00:00
PrologEpilogInserter.h
PseudoSourceValue.cpp
README.txt
RegAllocBase.h
RegAllocBasic.cpp Privatize an unused part of the LiveIntervalUnion::Query interface. 2011-08-11 21:00:42 +00:00
RegAllocFast.cpp If a register is both an early clobber and part of a tied use, handle the use 2011-11-22 06:27:18 +00:00
RegAllocGreedy.cpp Stop tracking spill slot uses in VirtRegMap. 2011-11-13 01:23:30 +00:00
RegAllocPBQP.cpp Switch PBQP to VRM's trivial rewriter. 2011-11-13 00:02:24 +00:00
RegisterClassInfo.cpp Detect proper register sub-classes. 2011-08-05 21:28:14 +00:00
RegisterClassInfo.h Detect proper register sub-classes. 2011-08-05 21:28:14 +00:00
RegisterCoalescer.cpp Rename SlotIndexes to match how they are used. 2011-11-13 20:45:27 +00:00
RegisterCoalescer.h Rename member variables to follow coding standards. 2011-08-09 01:01:27 +00:00
RegisterScavenging.cpp Silence a bunch (but not all) "variable written but not read" warnings 2011-08-12 14:54:45 +00:00
RenderMachineFunction.cpp Rename SlotIndexes to match how they are used. 2011-11-13 20:45:27 +00:00
RenderMachineFunction.h
ScheduleDAG.cpp Make a bunch of symbols private. 2011-08-19 01:42:18 +00:00
ScheduleDAGEmit.cpp
ScheduleDAGInstrs.cpp PostRA scheduler fix. Clear stale loop dependencies. 2011-10-07 06:33:09 +00:00
ScheduleDAGInstrs.h PostRA scheduler fix. Clear stale loop dependencies. 2011-10-07 06:33:09 +00:00
ScheduleDAGPrinter.cpp Twinify GraphWriter a little bit. 2011-11-15 16:26:38 +00:00
ScoreboardHazardRecognizer.cpp Remove an invalid assert that is really just asserting when the scheduler emits 2011-09-27 21:59:16 +00:00
ShadowStackGC.cpp Use the C personality function instead of the C++ personality function. 2011-09-22 17:56:40 +00:00
ShrinkWrapping.cpp Remove all remaining uses of Value::getNameStr(). 2011-11-15 16:27:03 +00:00
SjLjEHPrepare.cpp Record landing pads with a SmallSetVector to avoid multiple entries. 2011-11-16 07:57:21 +00:00
SlotIndexes.cpp Rename SlotIndexes to match how they are used. 2011-11-13 20:45:27 +00:00
Spiller.cpp Rename SlotIndexes to match how they are used. 2011-11-13 20:45:27 +00:00
Spiller.h
SpillPlacement.cpp Be more conservative when forming compact regions. 2011-08-03 23:09:38 +00:00
SpillPlacement.h Be more conservative when forming compact regions. 2011-08-03 23:09:38 +00:00
SplitKit.cpp Use getVNInfoBefore() when it makes sense. 2011-11-14 01:39:36 +00:00
SplitKit.h Hoist back-copies to the least busy dominator. 2011-09-14 16:45:39 +00:00
Splitter.cpp Move code into anonymous namespaces. 2011-11-26 23:01:57 +00:00
Splitter.h
StackProtector.cpp Enable stack protectors for all arrays, not just char arrays. rdar://5875909 2011-11-23 07:13:56 +00:00
StackSlotColoring.cpp Stop tracking unused registers in VirtRegMap. 2011-11-13 00:39:45 +00:00
StrongPHIElimination.cpp Rename SlotIndexes to match how they are used. 2011-11-13 20:45:27 +00:00
TailDuplication.cpp Trim an unneeded header. 2011-08-09 23:49:21 +00:00
TargetInstrInfoImpl.cpp Make use of MachinePointerInfo::getFixedStack. 2011-11-15 07:51:13 +00:00
TargetLoweringObjectFileImpl.cpp Remove all remaining uses of Value::getNameStr(). 2011-11-15 16:27:03 +00:00
TwoAddressInstructionPass.cpp Disable expensive two-address optimizations at -O0. rdar://10453055 2011-11-16 18:44:48 +00:00
UnreachableBlockElim.cpp
VirtRegMap.cpp More dead code elimination in VirtRegMap. 2011-11-13 01:23:34 +00:00
VirtRegMap.h More dead code elimination in VirtRegMap. 2011-11-13 01:23:34 +00:00

//===---------------------------------------------------------------------===//

Common register allocation / spilling problem:

        mul lr, r4, lr
        str lr, [sp, #+52]
        ldr lr, [r1, #+32]
        sxth r3, r3
        ldr r4, [sp, #+52]
        mla r4, r3, lr, r4

can be:

        mul lr, r4, lr
        mov r4, lr
        str lr, [sp, #+52]
        ldr lr, [r1, #+32]
        sxth r3, r3
        mla r4, r3, lr, r4

and then "merge" mul and mov:

        mul r4, r4, lr
        str lr, [sp, #+52]
        ldr lr, [r1, #+32]
        sxth r3, r3
        mla r4, r3, lr, r4

It also increase the likelihood the store may become dead.

//===---------------------------------------------------------------------===//

bb27 ...
        ...
        %reg1037 = ADDri %reg1039, 1
        %reg1038 = ADDrs %reg1032, %reg1039, %NOREG, 10
    Successors according to CFG: 0x8b03bf0 (#5)

bb76 (0x8b03bf0, LLVM BB @0x8b032d0, ID#5):
    Predecessors according to CFG: 0x8b0c5f0 (#3) 0x8b0a7c0 (#4)
        %reg1039 = PHI %reg1070, mbb<bb76.outer,0x8b0c5f0>, %reg1037, mbb<bb27,0x8b0a7c0>

Note ADDri is not a two-address instruction. However, its result %reg1037 is an
operand of the PHI node in bb76 and its operand %reg1039 is the result of the
PHI node. We should treat it as a two-address code and make sure the ADDri is
scheduled after any node that reads %reg1039.

//===---------------------------------------------------------------------===//

Use local info (i.e. register scavenger) to assign it a free register to allow
reuse:
        ldr r3, [sp, #+4]
        add r3, r3, #3
        ldr r2, [sp, #+8]
        add r2, r2, #2
        ldr r1, [sp, #+4]  <==
        add r1, r1, #1
        ldr r0, [sp, #+4]
        add r0, r0, #2

//===---------------------------------------------------------------------===//

LLVM aggressively lift CSE out of loop. Sometimes this can be negative side-
effects:

R1 = X + 4
R2 = X + 7
R3 = X + 15

loop:
load [i + R1]
...
load [i + R2]
...
load [i + R3]

Suppose there is high register pressure, R1, R2, R3, can be spilled. We need
to implement proper re-materialization to handle this:

R1 = X + 4
R2 = X + 7
R3 = X + 15

loop:
R1 = X + 4  @ re-materialized
load [i + R1]
...
R2 = X + 7 @ re-materialized
load [i + R2]
...
R3 = X + 15 @ re-materialized
load [i + R3]

Furthermore, with re-association, we can enable sharing:

R1 = X + 4
R2 = X + 7
R3 = X + 15

loop:
T = i + X
load [T + 4]
...
load [T + 7]
...
load [T + 15]
//===---------------------------------------------------------------------===//

It's not always a good idea to choose rematerialization over spilling. If all
the load / store instructions would be folded then spilling is cheaper because
it won't require new live intervals / registers. See 2003-05-31-LongShifts for
an example.

//===---------------------------------------------------------------------===//

With a copying garbage collector, derived pointers must not be retained across
collector safe points; the collector could move the objects and invalidate the
derived pointer. This is bad enough in the first place, but safe points can
crop up unpredictably. Consider:

        %array = load { i32, [0 x %obj] }** %array_addr
        %nth_el = getelementptr { i32, [0 x %obj] }* %array, i32 0, i32 %n
        %old = load %obj** %nth_el
        %z = div i64 %x, %y
        store %obj* %new, %obj** %nth_el

If the i64 division is lowered to a libcall, then a safe point will (must)
appear for the call site. If a collection occurs, %array and %nth_el no longer
point into the correct object.

The fix for this is to copy address calculations so that dependent pointers
are never live across safe point boundaries. But the loads cannot be copied
like this if there was an intervening store, so may be hard to get right.

Only a concurrent mutator can trigger a collection at the libcall safe point.
So single-threaded programs do not have this requirement, even with a copying
collector. Still, LLVM optimizations would probably undo a front-end's careful
work.

//===---------------------------------------------------------------------===//

The ocaml frametable structure supports liveness information. It would be good
to support it.

//===---------------------------------------------------------------------===//

The FIXME in ComputeCommonTailLength in BranchFolding.cpp needs to be
revisited. The check is there to work around a misuse of directives in inline
assembly.

//===---------------------------------------------------------------------===//

It would be good to detect collector/target compatibility instead of silently
doing the wrong thing.

//===---------------------------------------------------------------------===//

It would be really nice to be able to write patterns in .td files for copies,
which would eliminate a bunch of explicit predicates on them (e.g. no side 
effects).  Once this is in place, it would be even better to have tblgen 
synthesize the various copy insertion/inspection methods in TargetInstrInfo.

//===---------------------------------------------------------------------===//

Stack coloring improvements:

1. Do proper LiveStackAnalysis on all stack objects including those which are
   not spill slots.
2. Reorder objects to fill in gaps between objects.
   e.g. 4, 1, <gap>, 4, 1, 1, 1, <gap>, 4 => 4, 1, 1, 1, 1, 4, 4

//===---------------------------------------------------------------------===//

The scheduler should be able to sort nearby instructions by their address. For
example, in an expanded memset sequence it's not uncommon to see code like this:

  movl $0, 4(%rdi)
  movl $0, 8(%rdi)
  movl $0, 12(%rdi)
  movl $0, 0(%rdi)

Each of the stores is independent, and the scheduler is currently making an
arbitrary decision about the order.

//===---------------------------------------------------------------------===//

Another opportunitiy in this code is that the $0 could be moved to a register:

  movl $0, 4(%rdi)
  movl $0, 8(%rdi)
  movl $0, 12(%rdi)
  movl $0, 0(%rdi)

This would save substantial code size, especially for longer sequences like
this. It would be easy to have a rule telling isel to avoid matching MOV32mi
if the immediate has more than some fixed number of uses. It's more involved
to teach the register allocator how to do late folding to recover from
excessive register pressure.