Files
llvm/lib/CodeGen
Lang Hames d54d4f6b2e New PBQP solver, and updates to the PBQP graph.
The previous PBQP solver was very robust but consumed a lot of memory,
performed a lot of redundant computation, and contained some unnecessarily tight
coupling that prevented experimentation with novel solution techniques. This new
solver is an attempt to address these shortcomings.

Important/interesting changes:

1) The domain-independent PBQP solver class, HeuristicSolverImpl, is gone.
It is replaced by a register allocation specific solver, PBQP::RegAlloc::Solver
(see RegAllocSolver.h).

The optimal reduction rules and the backpropagation algorithm have been extracted
into stand-alone functions (see ReductionRules.h), which can be used to build
domain specific PBQP solvers. This provides many more opportunities for
domain-specific knowledge to inform the PBQP solvers' decisions. In theory this
should allow us to generate better solutions. In practice, we can at least test
out ideas now.

As a side benefit, I believe the new solver is more readable than the old one.

2) The solver type is now a template parameter of the PBQP graph.

This allows the graph to notify the solver of any modifications made (e.g. by
domain independent rules) without the overhead of a virtual call. It also allows
the solver to supply policy information to the graph (see below).

3) Significantly reduced memory overhead.

Memory management policy is now an explicit property of the PBQP graph (via
the CostAllocator typedef on the graph's solver template argument). Because PBQP
graphs for register allocation tend to contain many redundant instances of
single values (E.g. the value representing an interference constraint between
GPRs), the new RASolver class uses a uniquing scheme. This massively reduces
memory consumption for large register allocation problems. For example, looking
at the largest interference graph in each of the SPEC2006 benchmarks (the
largest graph will always set the memory consumption high-water mark for PBQP),
the average memory reduction for the PBQP costs was 400x. That's times, not
percent. The highest was 1400x. Yikes. So - this is fixed.

"PBQP: No longer feasting upon every last byte of your RAM".

Minor details:

- Fully C++11'd. Never copy-construct another vector/matrix!

- Cute tricks with cost metadata: Metadata that is derived solely from cost
matrices/vectors is attached directly to the cost instances themselves. That way
if you unique the costs you never have to recompute the metadata. 400x less
memory means 400x less cost metadata (re)computation.

Special thanks to Arnaud de Grandmaison, who has been the source of much
encouragement, and of many very useful test cases.

This new solver forms the basis for future work, of which there's plenty to do.
I will be adding TODO notes shortly.

- Lang.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202551 91177308-0d34-0410-b5e6-96231b3b80d8
2014-02-28 22:25:24 +00:00
..
2014-01-24 17:20:08 +00:00
2013-03-25 13:47:46 +00:00
2013-10-11 15:40:14 +00:00
2014-01-24 17:20:08 +00:00
2014-01-22 02:38:23 +00:00
2013-10-04 16:52:58 +00:00
2014-02-22 19:31:28 +00:00
2014-01-24 17:20:08 +00:00
2013-08-23 17:48:53 +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 r4, [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.