Rahman Lavaee 3d6841b2b1 [Propeller] Use Fixed MBB ID instead of volatile MachineBasicBlock::Number.
Let Propeller use specialized IDs for basic blocks, instead of MBB number.

This allows optimizations not just prior to asm-printer, but throughout the entire codegen.
This patch only implements the functionality under the new `LLVM_BB_ADDR_MAP` version, but the old version is still being used. A later patch will change the used version.

####Background
Today Propeller uses machine basic block (MBB) numbers, which already exist, to map native assembly to machine IR.  This is done as follows.
    - Basic block addresses are captured and dumped into the `LLVM_BB_ADDR_MAP` section just before the AsmPrinter pass which writes out object files. This ensures that we have a mapping that is close to assembly.
    - Profiling mapping works by taking a virtual address of an instruction and looking up the `LLVM_BB_ADDR_MAP` section to find the MBB number it corresponds to.
    - While this works well today, we need to do better when we scale Propeller to target other Machine IR optimizations like spill code optimization.  Register allocation happens earlier in the Machine IR pipeline and we need an annotation mechanism that is valid at that point.
    - The current scheme will not work in this scenario because the MBB number of a particular basic block is not fixed and changes over the course of codegen (via renumbering, adding, and removing the basic blocks).
    - In other words, the volatile MBB numbers do not provide a one-to-one correspondence throughout the lifetime of Machine IR.  Profile annotation using MBB numbers is restricted to a fixed point; only valid at the exact point where it was dumped.
    - Further, the object file can only be dumped before AsmPrinter and cannot be dumped at an arbitrary point in the Machine IR pass pipeline.  Hence, MBB numbers are not suitable and we need something else.
####Solution
We propose using fixed unique incremental MBB IDs for basic blocks instead of volatile MBB numbers. These IDs are assigned upon the creation of machine basic blocks. We modify `MachineFunction::CreateMachineBasicBlock` to assign the fixed ID to every newly created basic block.  It assigns `MachineFunction::NextMBBID` to the MBB ID and then increments it, which ensures having unique IDs.

 To ensure correct profile attribution, multiple equivalent compilations must generate the same Propeller IDs. This is guaranteed as long as the MachineFunction passes run in the same order. Since the `NextBBID` variable is scoped to `MachineFunction`, interleaving of codegen for different functions won't cause any inconsistencies.

The new encoding is generated under the new version number 2 and we keep backward-compatibility with older versions.

####Impact on Size of the `LLVM_BB_ADDR_MAP` Section
Emitting the Propeller ID results in a 23% increase in the size of the `LLVM_BB_ADDR_MAP` section for the clang binary.

Reviewed By: tmsriram

Differential Revision: https://reviews.llvm.org/D100808
2023-01-17 15:25:29 -08:00
..
2022-08-24 18:09:49 -07:00
2022-12-10 11:24:26 -08:00
2022-12-16 10:51:26 -08:00
2022-07-30 10:35:56 -07:00
2022-11-29 17:02:04 +01:00

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

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 LiveStacks analysis 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.