Library RelationAlgebra.compiler_opts


compiler_opts: certifying compiler optimisations

To illustrate some usage of the kat and hkat tactics, we formalise most of the compiler optimisations studied in the following paper:
Dexter Kozen and Maria-Cristina Patron. Certification of compiler optimizations using Kleene algebra with tests. In Proc. 1st Int. Conf. Computational Logic (CL2000), Vol. 1861 of LNAI, pages 568-582, July 2000. Springer-Verlag.
Most goals are solved with one single call to kat or hkat.
The remaining cases correspond to situations where one has to exploit permutations of some Kleene variables (the Horn theory of KA with such commutation hypotheses is undecidable).

Require Import kat normalisation rewriting kat_tac.
Set Implicit Arguments.

in this module, we prefer the ";" notation for composition
Infix " ;" := (dot _ _ _) (left associativity, at level 40): ra_terms.

preliminary lemmas


Lemma lemma_1 `{L: monoid.laws} `{Hl: BKA l} n (x y: X n n):
  x;y x;y;x x;y^* x;(y;x)^*.
Proof.
  intro H. apply antisym. apply str_ind_r'. ka.
  rewrite str_dot, <-dotA. rewrite H at 2. ka.
  rewrite str_dot. apply str_ind_l'. ka.
  rewrite str_unfold_l. ra_normalise. rewrite <-H. ka.
Qed.

Lemma lemma_1' `{L: monoid.laws} `{Hl: BKA l} n (x y: X n n):
  y;x x;(y;x) y^*;x (x;y)^*;x.
Proof. monoid.dual @lemma_1. Qed.

Lemma lemma_1'' `{L: monoid.laws} `{Hl: BKA l} n (p q r: X n n):
  p;q q;p p;r r (p;q)^*;r q^*;r.
Proof.
  intros Hpq Hpr. apply antisym.
  rewrite Hpq. apply str_ind_l'. ka. apply str_move in Hpq. mrewrite Hpq. mrewrite Hpr. ka.
  apply str_ind_l'. ka. rewrite <-str_snoc at 2. rewrite Hpq at 2 3. mrewrite Hpr. ka.
Qed.

Lemma lemma_2 `{L: laws} n b (q: X n n):
  [b];q q;[b] [b];q^* [b];(q;[b])^*.
Proof. hkat. Qed.

3.1 Deadcode elimination


Lemma opti_3_1_a `{L: laws} n (a: tst n) (p q: X n n):
  p p;[!a] p;([a];q+[!a]) p.
Proof. hkat. Qed.

Lemma opti_3_1_b `{L: laws} n (a: tst n) (p q: X n n):
  p p;[!a] p;([a];q)^*;[!a] p.
Proof. hkat. Qed.

3.2 Common sub-expression elimination


Lemma opti_3_2 `{L: laws} n (a b: tst n) (p q r w: X n n):
  p p;[a]
  [a];q [a];q;[b]
  [b];r [b]
  r w;r
  q;w w
  p;q p;r.
Proof.
  intros Hpa Haq Hbr Hr Hw.
  rewrite Hr, <-Hw. mrewrite <-Hr.
  hkat.
Qed.

3.3 Copy propagation


Lemma opti_3_3 `{L: laws} n (a b: tst n) (p q r s v w: X n n):
  q q;[a]
  [a];r [a];r;[b]
  [b];s [b]
  s w;s
  r;w w
  s;v v;s
  q;v v
  p;q;r;v p;s;v.
Proof.
  intros Hqa Har Hbs Hs Hw Hsv Hv.
  mrewrite Hsv. rewrite <-Hv at 2. mrewrite <-Hsv.
  rewrite Hs, <-Hw. mrewrite <-Hs.
  hkat.
Qed.

3.4 Loop Hoisting


Lemma opti_3_4i `{L: laws} n (a b: tst n) (p q r s u w: X n n):
  u;[b] u
  [b];u [b]
  [b];q q;[b]
  [b];s s;[b]
  [b];r r;[b]
  [a];w w;[a]
  u;r q
  u;w w
  q;s;w w;q;s
  p;u;([a];r;s)^*;[!a];w p;([a];q;s)^*;[!a];w.
Proof.
  intros ? ? ? ? ? ? Hur Huw Hqsw.
  transitivity (p;u;[b];([a];[b];(u;r);s)^*;[!a];w). hkat. rewrite Hur.
  transitivity (p;u;([a];q;s)^*;w;[!a]). hkat.
  assert (E: w;([a];q;s)^* ([a];q;s)^*;w) by (apply str_move; mrewrite Hqsw; hkat).
  mrewrite <-E. mrewrite Huw. mrewrite E. hkat.
Qed.

Lemma opti_3_4ii `{L: laws} n (a: tst n) (p q u w: X n n):
  u w;u
  u;w w
  w;p;q p;q;w
  w;[a] [a];w
  ([a];u;p;q)^*;[!a];u ([a];p;q)^*;[!a];u.
Proof.
  intros Hwu Huw Hpq Hw. rewrite Hwu at 1 2. transitivity (w;([a];u;(p;q;w))^*;[!a];u). hkat.
  rewrite <-Hpq. mrewrite Huw. mrewrite Hpq. rewrite <-lemma_1.
  rewrite (str_move (z:=[a];p;q)). rewrite Hwu at 2. hkat. mrewrite <-Hpq. hkat.
  mrewrite <-Hpq. rewrite <-Huw at 1. rewrite Hwu. mrewrite Huw. hkat.
Qed.

3.5 Induction variable elimination


Lemma opti_3_5 `{L: laws} n (a b c: tst n) (p q r: X n n):
  q q;[b]
  [b] [b];q
  [c];r [c];r;[b]
  [b];p [b];p;[c]
  [c];q [c];r
  q;([a];p;q)^* q;([a];p;r)^*.
Proof.
  intros Hq Hb Hr Hbp Hcq.
  assert (E: [b];p;q [b];p;r) by (rewrite Hbp; mrewrite Hcq; hkat).
  transitivity (q;([a];([b];p;q))^*;[b]). hkat. rewrite E. hkat.
Qed.

(3.6 and 3.7 are void)

3.8 Loop unrolling


Lemma lemma_3 `{L: monoid.laws} `{Hl: BKA l} n (u: X n n): u^* (1+u);(u;u)^*.
Proof. ka. Qed.

Lemma opti_3_8 `{L: laws} n a (p: X n n):
  ([a];p)^*;[!a] ([a];p;([a];p+[!a]))^*;[!a].
Proof. kat. Qed.

3.9 Redundant loads and stores


Lemma opti_3_9 `{L: laws} n a (p q: X n n):
  p p;[a] [a];q [a] p;q p.
Proof. intros Hp Hq. hkat. Qed.

3.10 Array bounds check elimination

3.11 Introduction of sentinels