Module Tunneling



Branch tunneling (optimization of branches to branches).

Require Import Coqlib.
Require Import Maps.
Require Import UnionFind.
Require Import Ast.
Require Import Values.
Require Import Globalenvs.
Require Import Op.
Require Import Locations.
Require Import LTL.

Branch tunneling shortens sequences of branches (with no intervening computations) by rewriting the branch and conditional branch instructions so that they jump directly to the end of the branch sequence. For example:
     L1: goto L2;                          L1: goto L3;
     L2; goto L3;               becomes    L2: goto L3;
     L3: instr;                            L3: instr;
     L4: if (cond) goto L1;                L4: if (cond) goto L3;
This optimization can be applied to several of our intermediate languages. We choose to perform it on the LTL language, after register allocation but before code linearization. Register allocation can delete instructions (such as dead computations or useless moves), therefore there are more opportunities for tunneling after allocation than before. Symmetrically, prior tunneling helps linearization to produce better code, e.g. by revealing that some goto instructions are dead code (as the "goto L3" in the example above).

branch_target f pc returns the node of the CFG that is at the end of the branch sequence starting at pc. If the instruction at pc is not a goto, pc itself is returned. The naive definition of branch_target is:
  branch_target f pc = branch_target f pc'  if f(pc) = goto pc'
                     = pc                   otherwise
However, this definition can fail to terminate if the program can contain loops consisting only of branches, as in
     L1: goto L1;
or << L1: goto L2; L2: goto L1; >> Coq warns us of this fact by not accepting the definition of branch_target above. The proper way to handle this problem is to detect goto cycles in the control-flow graph. For simplicity, we just bound arbitrarily the number of iterations performed by branch_target, never chasing more than 10 goto instructions. (This many consecutive branches is rarely, if even, encountered.) For a sequence of more than 10 goto instructions, we can return (as branch target) any of the labels of the goto instructions. This is semantically correct in any case. However, the proof is simpler if we return the label of the first goto in the sequence.

Module U := UnionFind.UF(PTree).

Definition record_goto (uf: U.t) (pc: node) (i: instruction) : U.t :=
  match i with
  | Lnop s => U.union uf pc s
  | _ => uf
  end.

Definition record_gotos (f: LTL.function) : U.t :=
  PTree.fold record_goto f.(fn_code) U.empty.

The tunneling optimization simply rewrites all LTL basic blocks, replacing the destinations of the Bgoto and Bcond instructions with their final target, as computed by branch_target.

Definition tunnel_instr (uf: U.t) (b: instruction) : instruction :=
  match b with
  | Lnop s =>
      Lnop (U.repr uf s)
  | Lop op args res s =>
      Lop op args res (U.repr uf s)
  | Lload chunk addr args dst s =>
      Lload chunk addr args dst (U.repr uf s)
  | Lstore chunk addr args src s =>
      Lstore chunk addr args src (U.repr uf s)
  | Lcall sig ros args res s =>
      Lcall sig ros args res (U.repr uf s)
  | Lcond cond args s1 s2 =>
      Lcond cond args (U.repr uf s1) (U.repr uf s2)
  | Lreturn or =>
      Lreturn or
  | Latomic aop args res s =>
      Latomic aop args res (U.repr uf s)
  | Lfence s =>
      Lfence (U.repr uf s)
  | Lthreadcreate fn arg s =>
      Lthreadcreate fn arg (U.repr uf s)
  end.

Definition tunnel_function (f: LTL.function) : LTL.function :=
  let uf := record_gotos f in
  mkfunction
    (fn_sig f)
    (fn_params f)
    (fn_stacksize f)
    (PTree.map (fun pc b => tunnel_instr uf b) (fn_code f))
    (U.repr uf (fn_entrypoint f)).

Definition tunnel_fundef (f: LTL.fundef) : LTL.fundef :=
  transf_fundef tunnel_function f.

Definition tunnel_program (p: LTL.program) : LTL.program :=
  transform_program tunnel_fundef p.