llvm-6502/lib/CodeGen
Bob Wilson f8c4cfb7cc Refactor target hook for tail duplication as requested by Chris.
Make tail duplication of indirect branches much more aggressive (for targets
that indicate that it is profitable), based on further experience with
this transformation.  I compiled 3 large applications with and without
this more aggressive tail duplication and measured minimal changes in code
size.  ("size" on Darwin seems to round the text size up to the nearest
page boundary, so I can only say that any code size increase was less than
one 4k page.) Radar 7421267.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@89814 91177308-0d34-0410-b5e6-96231b3b80d8
2009-11-24 23:35:49 +00:00
..
AsmPrinter Use StringRef instead of std::string in DIEString. 2009-11-24 19:42:17 +00:00
PBQP Mark more constants unsigned, as warned about by icc (#68). 2009-09-06 12:56:52 +00:00
SelectionDAG Remove ISD::DEBUG_LOC and ISD::DBG_LABEL, which are no longer used. 2009-11-23 23:20:51 +00:00
AggressiveAntiDepBreaker.cpp <rdar://problem/6721894>. Allow multiple registers to be renamed together (super and sub) if necessary to break an anti-dependence. 2009-11-24 00:59:08 +00:00
AggressiveAntiDepBreaker.h Remove some old experimental code that is no longer needed. Remove additional, speculative scheduling pass as its cost did not translate into significant performance improvement. Minor tweaks. 2009-11-20 19:32:48 +00:00
AntiDepBreaker.h Remove some old experimental code that is no longer needed. Remove additional, speculative scheduling pass as its cost did not translate into significant performance improvement. Minor tweaks. 2009-11-20 19:32:48 +00:00
BranchFolding.cpp Refactor target hook for tail duplication as requested by Chris. 2009-11-24 23:35:49 +00:00
BranchFolding.h Perform tail duplication only once, after tail merging is complete. 2009-11-17 17:06:18 +00:00
CMakeLists.txt Fix CMake makefiles 2009-11-04 01:32:06 +00:00
CodePlacementOpt.cpp Move the utility function UpdateTerminator() from CodePlacementOpt() into 2009-11-12 03:55:33 +00:00
CriticalAntiDepBreaker.cpp Remove some old experimental code that is no longer needed. Remove additional, speculative scheduling pass as its cost did not translate into significant performance improvement. Minor tweaks. 2009-11-20 19:32:48 +00:00
CriticalAntiDepBreaker.h Remove some old experimental code that is no longer needed. Remove additional, speculative scheduling pass as its cost did not translate into significant performance improvement. Minor tweaks. 2009-11-20 19:32:48 +00:00
DeadMachineInstructionElim.cpp Remove includes of Support/Compiler.h that are no longer needed after the 2009-10-25 06:57:41 +00:00
DwarfEHPrepare.cpp Pull LLVMContext out of PromoteMemToReg. 2009-11-23 03:50:44 +00:00
ELF.h Move DataTypes.h to include/llvm/System, update all users. This breaks the last 2009-10-26 01:35:46 +00:00
ELFCodeEmitter.cpp Rename TargetAsmInfo (and its subclasses) to MCAsmInfo. 2009-08-22 20:48:53 +00:00
ELFCodeEmitter.h
ELFWriter.cpp strength reduce a ton of type equality tests to check the typeid (Through 2009-10-05 05:54:46 +00:00
ELFWriter.h Implement the JIT side of the GDB JIT debugging interface. To enable this 2009-09-20 23:52:43 +00:00
ExactHazardRecognizer.cpp Do a scheduling pass ignoring anti-dependencies to identify candidate registers that should be renamed. 2009-11-03 20:57:50 +00:00
ExactHazardRecognizer.h Post RA scheduler changes. Introduce a hazard recognizer that uses the target schedule information to accurately model the pipeline. Update the scheduler to correctly handle multi-issue targets. 2009-08-10 15:55:25 +00:00
GCMetadata.cpp Remove includes of Support/Compiler.h that are no longer needed after the 2009-10-25 06:57:41 +00:00
GCMetadataPrinter.cpp rename TAI -> MAI, being careful not to make MAILJMP instructions :) 2009-08-22 21:43:10 +00:00
GCStrategy.cpp Remove includes of Support/Compiler.h that are no longer needed after the 2009-10-25 06:57:41 +00:00
IfConversion.cpp Allow target to disable if-converting predicable instructions. e.g. NEON instructions under ARM mode. 2009-11-21 06:20:26 +00:00
IntrinsicLowering.cpp Codegen support for the llvm.invariant/lifetime.start/end intrinsics: 2009-11-10 09:08:09 +00:00
LatencyPriorityQueue.cpp Remove some old experimental code that is no longer needed. Remove additional, speculative scheduling pass as its cost did not translate into significant performance improvement. Minor tweaks. 2009-11-20 19:32:48 +00:00
LiveInterval.cpp The Indexes Patch. 2009-11-03 23:52:08 +00:00
LiveIntervalAnalysis.cpp More consistent labelling of basic blocks in debug output 2009-11-20 18:54:59 +00:00
LiveStackAnalysis.cpp The Indexes Patch. 2009-11-03 23:52:08 +00:00
LiveVariables.cpp Be more clever about calculating live variables through new basic blocks. 2009-11-21 02:05:21 +00:00
LLVMTargetMachine.cpp Remove DebugLabelFolder pass. It is not used by dwarf writer anymore. 2009-11-24 19:37:07 +00:00
LowerSubregs.cpp Code clean up. 2009-10-25 07:49:57 +00:00
MachineBasicBlock.cpp Teach MachineBasicBlock::updateTerminator() to handle a failing TII->ReverseBranchCondition(Cond) call. 2009-11-22 18:28:04 +00:00
MachineDominators.cpp Change Pass::print to take a raw ostream instead of std::ostream, 2009-08-23 06:03:38 +00:00
MachineFunction.cpp Do not merge jump tables this early. Branch folding will do any necessary 2009-11-14 20:15:03 +00:00
MachineFunctionAnalysis.cpp Constify MachineFunctionAnalysis' TargetMachine reference. 2009-11-09 18:18:49 +00:00
MachineFunctionPass.cpp Add a form of addPreserved which takes a string argument, to allow passes 2009-10-08 17:00:02 +00:00
MachineInstr.cpp Print the debug info line and column in MachineInstr::print even when there's 2009-11-23 21:29:08 +00:00
MachineLICM.cpp Enable hoisting load from constant memories. 2009-11-20 23:31:34 +00:00
MachineLoopInfo.cpp Add getTopBlock and getBottomBlock member functions to MachineLoopInfo. 2009-10-20 04:16:37 +00:00
MachineModuleInfo.cpp Remove DebugLabelFolder pass. It is not used by dwarf writer anymore. 2009-11-24 19:37:07 +00:00
MachineModuleInfoImpls.cpp Don't sort the vector when it is empty. This should fix some expensive checking 2009-09-16 11:43:12 +00:00
MachinePassRegistry.cpp
MachineRegisterInfo.cpp Simplify a few more uses of reg_iterator. 2009-09-25 22:26:13 +00:00
MachineSink.cpp Remove includes of Support/Compiler.h that are no longer needed after the 2009-10-25 06:57:41 +00:00
MachineVerifier.cpp Add MachineBasicBlock::getName, and use it in place of getBasicBlock()->getName. 2009-11-20 01:17:03 +00:00
MachO.h rename TAI -> MAI, being careful not to make MAILJMP instructions :) 2009-08-22 21:43:10 +00:00
MachOCodeEmitter.cpp rename TAI -> MAI, being careful not to make MAILJMP instructions :) 2009-08-22 21:43:10 +00:00
MachOCodeEmitter.h rename TAI -> MAI, being careful not to make MAILJMP instructions :) 2009-08-22 21:43:10 +00:00
MachOWriter.cpp eliminate the "Value" printing methods that print to a std::ostream. 2009-08-23 04:37:46 +00:00
MachOWriter.h rename TAI -> MAI, being careful not to make MAILJMP instructions :) 2009-08-22 21:43:10 +00:00
Makefile
ObjectCodeEmitter.cpp Remove accidental commited comment 2009-08-05 07:00:43 +00:00
OcamlGC.cpp Remove includes of Support/Compiler.h that are no longer needed after the 2009-10-25 06:57:41 +00:00
Passes.cpp
PHIElimination.cpp Be more clever about calculating live variables through new basic blocks. 2009-11-21 02:05:21 +00:00
PHIElimination.h Be more clever about calculating live variables through new basic blocks. 2009-11-21 02:05:21 +00:00
PostRASchedulerList.cpp Remove some old experimental code that is no longer needed. Remove additional, speculative scheduling pass as its cost did not translate into significant performance improvement. Minor tweaks. 2009-11-20 19:32:48 +00:00
PreAllocSplitting.cpp - Change TargetInstrInfo::reMaterialize to pass in TargetRegisterInfo. 2009-11-14 02:55:43 +00:00
ProcessImplicitDefs.cpp For some targets, a copy can use a register multiple times, e.g. ppc. 2009-11-16 05:52:06 +00:00
PrologEpilogInserter.cpp Add a bool flag to StackObjects telling whether they reference spill 2009-11-12 20:49:22 +00:00
PrologEpilogInserter.h Re-enable register scavenging in Thumb1 by default. 2009-10-08 01:46:59 +00:00
PseudoSourceValue.cpp Do some cleanups suggested by Chris. 2009-11-12 21:49:55 +00:00
README.txt This remat entry is basically done. There are hooks to allow targets 2009-10-14 00:02:01 +00:00
RegAllocLinearScan.cpp Cleanups. 2009-11-20 21:13:27 +00:00
RegAllocLocal.cpp Add a bool flag to StackObjects telling whether they reference spill 2009-11-12 20:49:22 +00:00
RegAllocPBQP.cpp Added an assert to the PBQP allocator to catch infinite cost solutions which might otherwise lead to miscompilations. 2009-11-15 04:39:51 +00:00
RegisterCoalescer.cpp
RegisterScavenging.cpp RegScavenger::enterBasicBlock should always reset register state. 2009-11-12 07:49:10 +00:00
ScheduleDAG.cpp Remove some old experimental code that is no longer needed. Remove additional, speculative scheduling pass as its cost did not translate into significant performance improvement. Minor tweaks. 2009-11-20 19:32:48 +00:00
ScheduleDAGEmit.cpp Add assertion checks here to turn silent miscompiles into aborts. 2009-10-30 23:59:06 +00:00
ScheduleDAGInstrs.cpp Fix dependencies added to model memory aliasing for post-RA scheduling. The dependencies were overly conservative for memory access that are known not to alias. 2009-11-09 19:22:17 +00:00
ScheduleDAGInstrs.h Spill slots cannot alias. 2009-10-18 19:58:47 +00:00
ScheduleDAGPrinter.cpp Trim unnecessary includes. 2009-10-20 04:23:20 +00:00
ShadowStackGC.cpp Remove includes of Support/Compiler.h that are no longer needed after the 2009-10-25 06:57:41 +00:00
ShrinkWrapping.cpp Convert DOUT to DEBUG(errs()...). 2009-08-22 20:46:59 +00:00
SimpleHazardRecognizer.h Post RA scheduler changes. Introduce a hazard recognizer that uses the target schedule information to accurately model the pipeline. Update the scheduler to correctly handle multi-issue targets. 2009-08-10 15:55:25 +00:00
SimpleRegisterCoalescing.cpp Add MachineBasicBlock::getName, and use it in place of getBasicBlock()->getName. 2009-11-20 01:17:03 +00:00
SimpleRegisterCoalescing.h The Indexes Patch. 2009-11-03 23:52:08 +00:00
SjLjEHPrepare.cpp Remove includes of Support/Compiler.h that are no longer needed after the 2009-10-25 06:57:41 +00:00
SlotIndexes.cpp Added an API to the SlotIndexes pass to allow new instructions to be inserted into the numbering. 2009-11-14 00:02:51 +00:00
Spiller.cpp Removed references to LiveStacks from Spiller.* . They're no longer needed. 2009-11-20 00:53:30 +00:00
Spiller.h Removed references to LiveStacks from Spiller.* . They're no longer needed. 2009-11-20 00:53:30 +00:00
StackProtector.cpp Remove VISIBILITY_HIDDEN from class/struct found inside anonymous namespaces. 2009-10-25 06:33:48 +00:00
StackSlotColoring.cpp The Indexes Patch. 2009-11-03 23:52:08 +00:00
StrongPHIElimination.cpp The Indexes Patch. 2009-11-03 23:52:08 +00:00
TargetInstrInfoImpl.cpp Check if subreg index is zero. 2009-11-16 06:31:49 +00:00
TwoAddressInstructionPass.cpp Fix PR5300. 2009-11-18 21:33:35 +00:00
UnreachableBlockElim.cpp Remove includes of Support/Compiler.h that are no longer needed after the 2009-10-25 06:57:41 +00:00
VirtRegMap.cpp Add a bool flag to StackObjects telling whether they reference spill 2009-11-12 20:49:22 +00:00
VirtRegMap.h The Indexes Patch. 2009-11-03 23:52:08 +00:00
VirtRegRewriter.cpp Add MachineBasicBlock::getName, and use it in place of getBasicBlock()->getName. 2009-11-20 01:17:03 +00:00
VirtRegRewriter.h Kill off more cerr/cout uses and prune includes a bit. 2009-08-23 11:37:21 +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 likelyhood 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 improvments:

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.