Preliminary support for EABI-hardfloat calling conventions

git-svn-id: https://yquem.inria.fr/compcert/svn/compcert/trunk@2473 fca1b0fc-160b-0410-b1d3-a4f43f01ea2e

3 files changed, 938 insertions, 48 deletions

diff --git a/arm/PrintAsm.ml b/arm/PrintAsm.ml
index 66aa908..e1d5eaa 100644
--- a/arm/PrintAsm.ml
+++ b/arm/PrintAsm.ml
@@ -272,7 +272,7 @@ let print_annot_val oc txt args res =
   | [IR src], [IR dst] ->
       if dst = src then 0 else (fprintf oc "	mov	%a, %a\n" ireg dst ireg src; 1)
   | [FR src], [FR dst] ->
-      if dst = src then 0 else (fprintf oc "	fcpy	%a, %a\n" freg dst freg src; 1)
+      if dst = src then 0 else (fprintf oc "	fcpyd	%a, %a\n" freg dst freg src; 1)
   | _, _ -> assert false
 
 (* Handling of memcpy *)
@@ -508,53 +508,141 @@ let print_builtin_inline oc name args res =
 
 type direction = Incoming | Outgoing
 
-let ireg_param = function
-  | 0 -> IR0 | 1 -> IR1 | 2 -> IR2 | 3 -> IR3 | _ -> assert false
-
-let freg_param = function
-  | 0 -> FR0 | 1 -> FR1 | 2 -> FR2 | 3 -> FR3 | _ -> assert false
-
-let fixup_double oc dir f i1 i2 =
-  match dir with
-  | Incoming ->     (* f <- (i1, i2)  *)
-      fprintf oc "	fmdrr	%a, %a, %a\n" freg f ireg i1 ireg i2
-  | Outgoing ->     (* (i1, i2) <- f *)
-      fprintf oc "	fmrrd	%a, %a, %a\n" ireg i1 ireg i2 freg f
-
-let fixup_single oc dir f i =
-  match dir with
-  | Incoming ->     (* f <- i; f <- double_of_single f *)
-      fprintf oc "	fmsr	%a, %a\n" freg_single f ireg i;
-      fprintf oc "	fcvtds	%a, %a\n" freg f freg_single f
-  | Outgoing ->     (* f <- single_of_double f; i <- f *)
-      fprintf oc "	fcvtsd	%a, %a\n" freg_single f freg f;
-      fprintf oc "	fmrs	%a, %a\n" ireg i freg_single f
-
-let fixup_conventions oc dir tyl =
-  let rec fixup i tyl =
-    if i >= 4 then 0 else
-      match tyl with
-      | [] -> 0
-      | Tint :: tyl' ->
-          fixup (i+1) tyl'
-      | Tlong :: tyl' ->
-          fixup (((i + 1) land (-2)) + 2) tyl'
-      | Tfloat :: tyl' ->
-          let i = (i + 1) land (-2) in
-          if i >= 4 then 0 else begin
-            fixup_double oc dir (freg_param i) (ireg_param i) (ireg_param (i+1));
-            1 + fixup (i+2) tyl'
-          end
-      | Tsingle :: tyl' ->
-          fixup_single oc dir (freg_param i) (ireg_param i);
-          2 + fixup (i+1) tyl'
-  in fixup 0 tyl
-
-let fixup_arguments oc dir sg =
-  fixup_conventions oc dir sg.sig_args
-
-let fixup_result oc dir sg =
-  fixup_conventions oc dir (proj_sig_res sg :: [])
+module FixupEABI = struct
+
+  let ireg_param = function
+    | 0 -> IR0 | 1 -> IR1 | 2 -> IR2 | 3 -> IR3 | _ -> assert false
+
+  let freg_param = function
+    | 0 -> FR0 | 1 -> FR1 | 2 -> FR2 | 3 -> FR3 | _ -> assert false
+
+  let fixup_double oc dir f i1 i2 =
+    match dir with
+    | Incoming ->     (* f <- (i1, i2)  *)
+        fprintf oc "	fmdrr	%a, %a, %a\n" freg f ireg i1 ireg i2
+    | Outgoing ->     (* (i1, i2) <- f *)
+        fprintf oc "	fmrrd	%a, %a, %a\n" ireg i1 ireg i2 freg f
+
+  let fixup_single oc dir f i =
+    match dir with
+    | Incoming ->     (* f <- i; f <- double_of_single f *)
+        fprintf oc "	fmsr	%a, %a\n" freg_single f ireg i;
+        fprintf oc "	fcvtds	%a, %a\n" freg f freg_single f
+    | Outgoing ->     (* f <- single_of_double f; i <- f *)
+        fprintf oc "	fcvtsd	%a, %a\n" freg_single f freg f;
+        fprintf oc "	fmrs	%a, %a\n" ireg i freg_single f
+
+  let fixup_conventions oc dir tyl =
+    let rec fixup i tyl =
+      if i >= 4 then 0 else
+        match tyl with
+        | [] -> 0
+        | Tint :: tyl' ->
+            fixup (i+1) tyl'
+        | Tlong :: tyl' ->
+            fixup (((i + 1) land (-2)) + 2) tyl'
+        | Tfloat :: tyl' ->
+            let i = (i + 1) land (-2) in
+            if i >= 4 then 0 else begin
+              fixup_double oc dir (freg_param i) (ireg_param i) (ireg_param (i+1));
+              1 + fixup (i+2) tyl'
+            end
+        | Tsingle :: tyl' ->
+            fixup_single oc dir (freg_param i) (ireg_param i);
+            2 + fixup (i+1) tyl'
+    in fixup 0 tyl
+
+  let fixup_arguments oc dir sg =
+    fixup_conventions oc dir sg.sig_args
+
+  let fixup_result oc dir sg =
+    fixup_conventions oc dir (proj_sig_res sg :: [])
+
+end
+
+module FixupHF = struct
+
+  type fsize = Single | Double
+
+  let rec find_single used pos =
+    if pos >= Array.length used then pos
+    else if used.(pos) then find_single used (pos + 1)
+    else begin used.(pos) <- true; pos end
+
+  let rec find_double used pos =
+    if pos + 1 >= Array.length used then pos
+    else if used.(pos) || used.(pos + 1) then find_double used (pos + 2)
+    else begin used.(pos) <- true; used.(pos + 1) <- true; pos / 2 end
+
+  let rec fixup_actions used fr tyl =
+    match tyl with
+    | [] -> []
+    | (Tint | Tlong) :: tyl' -> fixup_actions used fr tyl'
+    | Tfloat :: tyl' ->
+        if fr >= 8 then [] else begin
+          let dr = find_double used 0 in
+          assert (dr < 8);
+          (fr, Double, dr) :: fixup_actions used (fr + 1) tyl'
+        end
+    | Tsingle :: tyl' ->
+        if fr >= 8 then [] else begin
+          let sr = find_single used 0 in
+          assert (sr < 16);
+          (fr, Single, sr) :: fixup_actions used (fr + 1) tyl'
+        end
+
+  let rec fixup_outgoing oc = function
+  | [] -> 0
+  | (fr, Double, dr) :: act ->
+      if fr = dr then fixup_outgoing oc act else begin
+        fprintf oc "	fcpyd	d%d, d%d\n" dr fr;
+        1 + fixup_outgoing oc act
+      end
+  | (fr, Single, sr) :: act ->
+        fprintf oc "	fcvtsd	s%d, d%d\n" sr fr;
+        1 + fixup_outgoing oc act
+
+  let rec fixup_incoming oc = function
+  | [] -> 0
+  | (fr, Double, dr) :: act ->
+      let n = fixup_incoming oc act in
+      if fr = dr then n else begin
+        fprintf oc "	fcpyd	d%d, d%d\n" fr dr;
+        1 + n
+      end
+  | (fr, Single, sr) :: act ->
+      let n = fixup_incoming oc act in
+      fprintf oc "	fcvtds	d%d, s%d\n" fr sr;
+        1 + n
+
+  let fixup_arguments oc dir sg =
+    if sg.sig_cc.cc_vararg then
+      FixupEABI.fixup_arguments oc dir sg
+    else begin
+      let act = fixup_actions (Array.make 16 false) 0 sg.sig_args in
+      match dir with
+      | Outgoing -> fixup_outgoing oc act
+      | Incoming -> fixup_incoming oc act
+    end
+
+  let fixup_result oc dir sg =
+    if sg.sig_cc.cc_vararg then
+      FixupEABI.fixup_result oc dir sg
+    else begin
+      match proj_sig_res sg, dir with
+      | Tsingle, Outgoing ->
+          fprintf oc "	fcvtsd	s0, d0\n"; 1
+      | Tsingle, Incoming ->
+          fprintf oc "	fcvtds	d0, s0\n"; 1
+      | _ -> 0
+    end
+end
+
+let (fixup_arguments, fixup_result) =
+  match Configuration.variant with
+  | "eabi"      -> (FixupEABI.fixup_arguments, FixupEABI.fixup_result)
+  | "hardfloat" -> (FixupHF.fixup_arguments, FixupHF.fixup_result)
+  | _ -> assert false
 
 (* Printing of instructions *)
 
diff --git a/arm/hardfloat/Conventions1.v b/arm/hardfloat/Conventions1.v
new file mode 100644
index 0000000..e3875e7
--- /dev/null
+++ b/arm/hardfloat/Conventions1.v
@@ -0,0 +1,670 @@
+(* *********************************************************************)
+(*                                                                     *)
+(*              The Compcert verified compiler                         *)
+(*                                                                     *)
+(*          Xavier Leroy, INRIA Paris-Rocquencourt                     *)
+(*                                                                     *)
+(*  Copyright Institut National de Recherche en Informatique et en     *)
+(*  Automatique.  All rights reserved.  This file is distributed       *)
+(*  under the terms of the INRIA Non-Commercial License Agreement.     *)
+(*                                                                     *)
+(* *********************************************************************)
+
+(** Function calling conventions and other conventions regarding the use of 
+    machine registers and stack slots. *)
+
+Require Import Coqlib.
+Require Import AST.
+Require Import Locations.
+
+(** * Classification of machine registers *)
+
+(** Machine registers (type [mreg] in module [Locations]) are divided in
+  the following groups:
+- Temporaries used for spilling, reloading, and parallel move operations.
+- Allocatable registers, that can be assigned to RTL pseudo-registers.
+  These are further divided into:
+-- Callee-save registers, whose value is preserved across a function call.
+-- Caller-save registers that can be modified during a function call.
+
+  We follow the PowerPC application binary interface (ABI) in our choice
+  of callee- and caller-save registers.
+*)
+
+Definition int_caller_save_regs :=
+  R0 :: R1 :: R2 :: R3 :: R12 :: nil.
+
+Definition float_caller_save_regs :=
+  F0 :: F1 :: F2 :: F3 :: F4 :: F5 :: F6 :: F7 :: nil.
+
+Definition int_callee_save_regs :=
+  R4 :: R5 :: R6 :: R7 :: R8 :: R9 :: R10 :: R11 :: nil.
+
+Definition float_callee_save_regs :=
+  F8 :: F9 :: F10 :: F11 :: F12 :: F13 :: F14 :: F15 :: nil.
+
+Definition destroyed_at_call :=
+  int_caller_save_regs ++ float_caller_save_regs.
+
+Definition dummy_int_reg := R0.     (**r Used in [Coloring]. *)
+Definition dummy_float_reg := F0.   (**r Used in [Coloring]. *)
+
+(** The [index_int_callee_save] and [index_float_callee_save] associate
+  a unique positive integer to callee-save registers.  This integer is
+  used in [Stacking] to determine where to save these registers in
+  the activation record if they are used by the current function. *)
+
+Definition index_int_callee_save (r: mreg) :=
+  match r with
+  | R4 => 0  | R5 => 1  | R6 => 2  | R7 => 3
+  | R8 => 4  | R9 => 5  | R10 => 6 | R11 => 7
+  | _ => -1
+  end.
+
+Definition index_float_callee_save (r: mreg) :=
+  match r with
+  | F8 => 0  | F9 => 1  | F10 => 2  | F11 => 3
+  | F12 => 4  | F13 => 5  | F14 => 6  | F15 => 7
+  | _ => -1
+  end.
+
+Ltac ElimOrEq :=
+  match goal with
+  |  |- (?x = ?y) \/ _ -> _ =>
+       let H := fresh in
+       (intro H; elim H; clear H;
+        [intro H; rewrite <- H; clear H | ElimOrEq])
+  |  |- False -> _ =>
+       let H := fresh in (intro H; contradiction)
+  end.
+
+Ltac OrEq :=
+  match goal with
+  | |- (?x = ?x) \/ _ => left; reflexivity
+  | |- (?x = ?y) \/ _ => right; OrEq
+  | |- False => fail
+  end.
+
+Ltac NotOrEq :=
+  match goal with
+  | |- (?x = ?y) \/ _ -> False =>
+       let H := fresh in (
+       intro H; elim H; clear H; [intro; discriminate | NotOrEq])
+  | |- False -> False =>
+       contradiction
+  end.
+
+Lemma index_int_callee_save_pos:
+  forall r, In r int_callee_save_regs -> index_int_callee_save r >= 0.
+Proof.
+  intro r. simpl; ElimOrEq; unfold index_int_callee_save; omega.
+Qed.
+
+Lemma index_float_callee_save_pos:
+  forall r, In r float_callee_save_regs -> index_float_callee_save r >= 0.
+Proof.
+  intro r. simpl; ElimOrEq; unfold index_float_callee_save; omega.
+Qed.
+
+Lemma index_int_callee_save_pos2:
+  forall r, index_int_callee_save r >= 0 -> In r int_callee_save_regs.
+Proof.
+  destruct r; simpl; intro; omegaContradiction || OrEq.
+Qed.
+
+Lemma index_float_callee_save_pos2:
+  forall r, index_float_callee_save r >= 0 -> In r float_callee_save_regs.
+Proof.
+  destruct r; simpl; intro; omegaContradiction || OrEq.
+Qed.
+
+Lemma index_int_callee_save_inj:
+  forall r1 r2, 
+  In r1 int_callee_save_regs ->
+  In r2 int_callee_save_regs ->
+  r1 <> r2 ->
+  index_int_callee_save r1 <> index_int_callee_save r2.
+Proof.
+  intros r1 r2. 
+  simpl; ElimOrEq; ElimOrEq; unfold index_int_callee_save;
+  intros; congruence.
+Qed.
+
+Lemma index_float_callee_save_inj:
+  forall r1 r2, 
+  In r1 float_callee_save_regs ->
+  In r2 float_callee_save_regs ->
+  r1 <> r2 ->
+  index_float_callee_save r1 <> index_float_callee_save r2.
+Proof.
+  intros r1 r2. 
+  simpl; ElimOrEq; ElimOrEq; unfold index_float_callee_save;
+  intros; congruence.
+Qed.
+
+(** The following lemmas show that
+    (temporaries, destroyed at call, integer callee-save, float callee-save)
+    is a partition of the set of machine registers. *)
+
+Lemma int_float_callee_save_disjoint:
+  list_disjoint int_callee_save_regs float_callee_save_regs.
+Proof.
+  red; intros r1 r2. simpl; ElimOrEq; ElimOrEq; discriminate.
+Qed.
+
+Lemma register_classification:
+  forall r, 
+  In r destroyed_at_call \/ In r int_callee_save_regs \/ In r float_callee_save_regs.
+Proof.
+  destruct r; 
+  try (left; simpl; OrEq);
+  try (right; left; simpl; OrEq);
+  try (right; right; simpl; OrEq).
+Qed.
+
+
+Lemma int_callee_save_not_destroyed:
+  forall r, 
+    In r destroyed_at_call -> In r int_callee_save_regs -> False.
+Proof.
+  intros. revert H0 H. simpl. ElimOrEq; NotOrEq.
+Qed.
+
+Lemma float_callee_save_not_destroyed:
+  forall r, 
+    In r destroyed_at_call -> In r float_callee_save_regs -> False.
+Proof.
+  intros. revert H0 H. simpl. ElimOrEq; NotOrEq.
+Qed.
+
+Lemma int_callee_save_type:
+  forall r, In r int_callee_save_regs -> mreg_type r = Tint.
+Proof.
+  intro. simpl; ElimOrEq; reflexivity.
+Qed.
+
+Lemma float_callee_save_type:
+  forall r, In r float_callee_save_regs -> mreg_type r = Tfloat.
+Proof.
+  intro. simpl; ElimOrEq; reflexivity.
+Qed.
+
+Ltac NoRepet :=
+  match goal with
+  | |- list_norepet nil =>
+      apply list_norepet_nil
+  | |- list_norepet (?a :: ?b) =>
+      apply list_norepet_cons; [simpl; intuition discriminate | NoRepet]
+  end.
+
+Lemma int_callee_save_norepet:
+  list_norepet int_callee_save_regs.
+Proof.
+  unfold int_callee_save_regs; NoRepet.
+Qed.
+
+Lemma float_callee_save_norepet:
+  list_norepet float_callee_save_regs.
+Proof.
+  unfold float_callee_save_regs; NoRepet.
+Qed.
+
+(** * Function calling conventions *)
+
+(** The functions in this section determine the locations (machine registers
+  and stack slots) used to communicate arguments and results between the
+  caller and the callee during function calls.  These locations are functions
+  of the signature of the function and of the call instruction.  
+  Agreement between the caller and the callee on the locations to use
+  is guaranteed by our dynamic semantics for Cminor and RTL, which demand 
+  that the signature of the call instruction is identical to that of the
+  called function.
+
+  Calling conventions are largely arbitrary: they must respect the properties
+  proved in this section (such as no overlapping between the locations
+  of function arguments), but this leaves much liberty in choosing actual
+  locations.  *)
+
+(** ** Location of function result *)
+
+(** The result value of a function is passed back to the caller in
+  registers [R0] or [F0] or [R0,R1], depending on the type of the
+  returned value.  We treat a function without result as a function
+  with one integer result. *)
+
+Definition loc_result (s: signature) : list mreg :=
+  match s.(sig_res) with
+  | None => R0 :: nil
+  | Some Tint => R0 :: nil
+  | Some (Tfloat | Tsingle) => F0 :: nil
+  | Some Tlong => R1 :: R0 :: nil
+  end.
+
+(** The result location is a caller-save register or a temporary *)
+
+Lemma loc_result_caller_save:
+  forall (s: signature) (r: mreg), 
+  In r (loc_result s) -> In r destroyed_at_call.
+Proof.
+  intros.
+  assert (r = R0 \/ r = R1 \/ r = F0).
+    unfold loc_result in H. destruct (sig_res s); [destruct t|idtac]; simpl in H; intuition.
+  destruct H0 as [A | [A | A]]; subst r; simpl; OrEq.
+Qed.
+
+(** ** Location of function arguments *)
+
+(** We use the following calling conventions, adapted from the ARM EABI-HF:
+- The first 4 integer arguments are passed in registers [R0] to [R3].
+- The first 2 long integer arguments are passed in an aligned pair of
+  two integer registers.
+- The first 8 single- and double-precision float arguments are passed
+  in registers [F0...F7]
+- Extra arguments are passed on the stack, in [Outgoing] slots, consecutively
+  assigned (1 word for an integer or single argument, 2 words for a float
+  or a long), starting at word offset 0.
+
+This convention is not quite that of the ARM EABI-HF, whereas single float
+arguments are passed in 32-bit float registers.  Unfortunately,
+this does not fit the data model of CompCert.  In [PrintAsm.ml]
+we insert additional code around function calls that moves
+data appropriately. *)
+
+Definition int_param_regs :=
+  R0 :: R1 :: R2 :: R3 :: nil.
+
+Definition float_param_regs :=
+  F0 :: F1 :: F2 :: F3 :: F4 :: F5 :: F6 :: F7 :: nil.
+
+Definition ireg_param (n: Z) : mreg :=
+  match list_nth_z int_param_regs n with Some r => r | None => R0 end.
+
+Definition freg_param (n: Z) : mreg :=
+  match list_nth_z float_param_regs n with Some r => r | None => F0 end.
+
+Fixpoint loc_arguments_rec
+     (tyl: list typ) (ir fr ofs: Z) {struct tyl} : list loc :=
+  match tyl with
+  | nil => nil
+  | Tint :: tys =>
+      if zlt ir 4
+      then R (ireg_param ir) :: loc_arguments_rec tys (ir + 1) fr ofs
+      else S Outgoing ofs Tint :: loc_arguments_rec tys ir fr (ofs + 1)
+  | Tfloat :: tys =>
+      if zlt fr 8
+      then R (freg_param fr) :: loc_arguments_rec tys ir (fr + 1) ofs
+      else let ofs := align ofs 2 in
+           S Outgoing ofs Tfloat :: loc_arguments_rec tys ir fr (ofs + 2)
+  | Tsingle :: tys =>
+      if zlt fr 8
+      then R (freg_param fr) :: loc_arguments_rec tys ir (fr + 1) ofs
+      else S Outgoing ofs Tsingle :: loc_arguments_rec tys ir fr (ofs + 1)
+  | Tlong :: tys =>
+      let ir := align ir 2 in
+      if zlt ir 4
+      then R (ireg_param (ir + 1)) :: R (ireg_param ir) :: loc_arguments_rec tys (ir + 2) fr ofs
+      else let ofs := align ofs 2 in
+          S Outgoing (ofs + 1) Tint :: S Outgoing ofs Tint :: loc_arguments_rec tys ir fr (ofs + 2)
+  end.
+
+(** For variable-argument functions, we use the default ARM EABI (not HF)
+  calling conventions:
+- The first 4 integer arguments are passed in registers [R0] to [R3].
+- The first 2 long integer arguments are passed in an aligned pair of
+  two integer registers.
+- The first 2 double-precision float arguments are passed in [F0] or [F2]
+- The first 4 single-precision float arguments are passed in [F0...F3]
+- Integer arguments and float arguments are kept in sync so that
+  they can all be mapped back to [R0...R3] in [PrintAsm.ml].
+- Extra arguments are passed on the stack, in [Outgoing] slots, consecutively
+  assigned (1 word for an integer or single argument, 2 words for a float
+  or a long), starting at word offset 0.
+*)
+
+Fixpoint loc_arguments_vararg
+     (tyl: list typ) (ofs: Z) {struct tyl} : list loc :=
+  match tyl with
+  | nil => nil
+  | Tint :: tys =>
+      (if zlt ofs 0 then R (ireg_param (ofs + 4)) else S Outgoing ofs Tint)
+      :: loc_arguments_vararg tys (ofs + 1)
+  | Tfloat :: tys =>
+      let ofs := align ofs 2 in
+      (if zlt ofs 0 then R (freg_param (ofs + 4)) else S Outgoing ofs Tfloat)
+      :: loc_arguments_vararg tys (ofs + 2)
+  | Tsingle :: tys =>
+      (if zlt ofs 0 then R (freg_param (ofs + 4)) else S Outgoing ofs Tsingle)
+      :: loc_arguments_vararg tys (ofs + 1)
+  | Tlong :: tys =>
+      let ofs := align ofs 2 in
+      (if zlt ofs 0 then R (ireg_param (ofs+1+4)) else S Outgoing (ofs+1) Tint)
+      :: (if zlt ofs 0 then R (ireg_param (ofs+4)) else S Outgoing ofs Tint)
+      :: loc_arguments_vararg tys (ofs + 2)
+  end.
+
+(** [loc_arguments s] returns the list of locations where to store arguments
+  when calling a function with signature [s].  *)
+
+Definition loc_arguments (s: signature) : list loc :=
+  if s.(sig_cc).(cc_vararg)
+  then loc_arguments_vararg s.(sig_args) (-4)
+  else loc_arguments_rec s.(sig_args) 0 0 0.
+
+(** [size_arguments s] returns the number of [Outgoing] slots used
+  to call a function with signature [s]. *)
+
+Fixpoint size_arguments_rec (tyl: list typ) (ir fr ofs: Z) {struct tyl} : Z :=
+  match tyl with
+  | nil => ofs
+  | Tint :: tys =>
+      if zlt ir 4
+      then size_arguments_rec tys (ir + 1) fr ofs
+      else size_arguments_rec tys ir fr (ofs + 1)
+  | Tfloat :: tys =>
+      if zlt fr 8
+      then size_arguments_rec tys ir (fr + 1) ofs
+      else size_arguments_rec tys ir fr (align ofs 2 + 2)
+  | Tsingle :: tys =>
+      if zlt fr 8
+      then size_arguments_rec tys ir (fr + 1) ofs
+      else size_arguments_rec tys ir fr (ofs + 1)
+  | Tlong :: tys =>
+      let ir := align ir 2 in
+      if zlt ir 4
+      then size_arguments_rec tys (ir + 2) fr ofs
+      else size_arguments_rec tys ir fr (align ofs 2 + 2)
+  end.
+
+Fixpoint size_arguments_vararg (tyl: list typ) (ofs: Z) {struct tyl} : Z :=
+  match tyl with
+  | nil => Zmax 0 ofs
+  | (Tint | Tsingle) :: tys => size_arguments_vararg tys (ofs + 1)
+  | (Tfloat | Tlong) :: tys => size_arguments_vararg tys (align ofs 2 + 2)
+  end.
+
+Definition size_arguments (s: signature) : Z :=
+  if s.(sig_cc).(cc_vararg)
+  then size_arguments_vararg s.(sig_args) (-4)
+  else size_arguments_rec s.(sig_args) 0 0 0.
+
+(** Argument locations are either non-temporary registers or [Outgoing] 
+  stack slots at nonnegative offsets. *)
+
+Definition loc_argument_acceptable (l: loc) : Prop :=
+  match l with
+  | R r => In r destroyed_at_call
+  | S Outgoing ofs ty => ofs >= 0 /\ ty <> Tlong
+  | _ => False
+  end.
+
+Remark ireg_param_in_params: forall n, In (ireg_param n) int_param_regs.
+Proof.
+  unfold ireg_param; intros. 
+  destruct (list_nth_z int_param_regs n) as [r|] eqn:NTH.
+  eapply list_nth_z_in; eauto. 
+  simpl; auto.
+Qed.
+
+Remark freg_param_in_params: forall n, In (freg_param n) float_param_regs.
+Proof.
+  unfold freg_param; intros. 
+  destruct (list_nth_z float_param_regs n) as [r|] eqn:NTH.
+  eapply list_nth_z_in; eauto. 
+  simpl; auto.
+Qed.
+
+Remark loc_arguments_rec_charact:
+  forall tyl ir fr ofs l,
+  In l (loc_arguments_rec tyl ir fr ofs) ->
+  match l with
+  | R r => In r int_param_regs \/ In r float_param_regs
+  | S Outgoing ofs' ty => ofs' >= ofs /\ ty <> Tlong
+  | S _ _ _ => False
+  end.
+Proof.
+  assert (INCR: forall l ofs1 ofs2,
+            match l with
+            | R r => In r int_param_regs \/ In r float_param_regs
+            | S Outgoing ofs' ty => ofs' >= ofs2 /\ ty <> Tlong
+            | S _ _ _ => False
+            end ->
+            ofs1 <= ofs2 ->
+            match l with
+            | R r => In r int_param_regs \/ In r float_param_regs
+            | S Outgoing ofs' ty => ofs' >= ofs1 /\ ty <> Tlong
+            | S _ _ _ => False
+            end).
+  {
+    intros. destruct l; auto. destruct sl; auto. intuition omega.
+  }
+  induction tyl; simpl loc_arguments_rec; intros.
+  elim H.
+  destruct a.
+- (* int *)
+  destruct (zlt ir 4); destruct H.
+  subst. left; apply ireg_param_in_params.
+  eapply IHtyl; eauto.
+  subst. split; [omega | congruence].
+  eapply INCR. eapply IHtyl; eauto. omega.
+- (* float *)
+  destruct (zlt fr 8); destruct H.
+  subst. right; apply freg_param_in_params.
+  eapply IHtyl; eauto.
+  subst. split. apply Zle_ge. apply align_le. omega. congruence.
+  eapply INCR. eapply IHtyl; eauto.
+  apply Zle_trans with (align ofs 2). apply align_le; omega. omega.
+- (* long *)
+  set (ir' := align ir 2) in *.
+  assert (ofs <= align ofs 2) by (apply align_le; omega). 
+  destruct (zlt ir' 4).
+  destruct H. subst l; left; apply ireg_param_in_params.
+  destruct H. subst l; left; apply ireg_param_in_params.
+  eapply IHtyl; eauto.
+  destruct H. subst l; split; [ omega | congruence ].
+  destruct H. subst l; split; [ omega | congruence ].
+  eapply INCR. eapply IHtyl; eauto. omega.
+- (* single *)
+  destruct (zlt fr 8); destruct H.
+  subst. right; apply freg_param_in_params.
+  eapply IHtyl; eauto.
+  subst. split; [omega | congruence].
+  eapply INCR. eapply IHtyl; eauto. omega.
+Qed.
+
+Remark loc_arguments_vararg_charact:
+  forall tyl ofs l,
+  In l (loc_arguments_vararg tyl ofs) ->
+  match l with
+  | R r => In r int_param_regs \/ In r float_param_regs
+  | S Outgoing ofs' ty => ofs' >= Zmax 0 ofs /\ ty <> Tlong
+  | S _ _ _ => False
+  end.
+Proof.
+  assert (INCR: forall l ofs1 ofs2,
+            match l with
+            | R r => In r int_param_regs \/ In r float_param_regs
+            | S Outgoing ofs' ty => ofs' >= Zmax 0 ofs2 /\ ty <> Tlong
+            | S _ _ _ => False
+            end ->
+            ofs1 <= ofs2 ->
+            match l with
+            | R r => In r int_param_regs \/ In r float_param_regs
+            | S Outgoing ofs' ty => ofs' >= Zmax 0 ofs1 /\ ty <> Tlong
+            | S _ _ _ => False
+            end).
+  {
+    intros. destruct l; auto. destruct sl; auto. intuition xomega.
+  }
+  induction tyl; simpl loc_arguments_vararg; intros.
+  elim H.
+  destruct a.
+- (* int *)
+  destruct H. 
+  destruct (zlt ofs 0); subst l. 
+  left; apply ireg_param_in_params.
+  split. xomega. congruence.
+  eapply INCR. eapply IHtyl; eauto. omega.
+- (* float *)
+  set (ofs' := align ofs 2) in *.
+  assert (ofs <= ofs') by (apply align_le; omega).
+  destruct H.
+  destruct (zlt ofs' 0); subst l.
+  right; apply freg_param_in_params.
+  split. xomega. congruence.
+  eapply INCR. eapply IHtyl; eauto. omega.
+- (* long *)
+  set (ofs' := align ofs 2) in *.
+  assert (ofs <= ofs') by (apply align_le; omega).
+  destruct H.
+  destruct (zlt ofs' 0); subst l.
+  left; apply ireg_param_in_params.
+  split. xomega. congruence.
+  destruct H.
+  destruct (zlt ofs' 0); subst l.
+  left; apply ireg_param_in_params.
+  split. xomega. congruence.
+  eapply INCR. eapply IHtyl; eauto. omega.
+- (* single *)
+  destruct H. 
+  destruct (zlt ofs 0); subst l. 
+  right; apply freg_param_in_params.
+  split. xomega. congruence.
+  eapply INCR. eapply IHtyl; eauto. omega.
+Qed.
+
+Lemma loc_arguments_acceptable:
+  forall (s: signature) (l: loc),
+  In l (loc_arguments s) -> loc_argument_acceptable l.
+Proof.
+  unfold loc_arguments; intros.
+  assert (forall r, In r int_param_regs \/ In r float_param_regs -> In r destroyed_at_call).
+  {
+    intros. elim H0; simpl; ElimOrEq; OrEq.
+  }
+  red. destruct (cc_vararg (sig_cc s)). 
+  exploit loc_arguments_vararg_charact; eauto.
+  destruct l; auto.
+  exploit loc_arguments_rec_charact; eauto.
+  destruct l; auto.
+Qed.
+
+Hint Resolve loc_arguments_acceptable: locs.
+
+(** The offsets of [Outgoing] arguments are below [size_arguments s]. *)
+
+Remark size_arguments_rec_above:
+  forall tyl ir fr ofs0,
+  ofs0 <= size_arguments_rec tyl ir fr ofs0.
+Proof.
+  induction tyl; simpl; intros.
+  omega.
+  destruct a.
+  destruct (zlt ir 4); eauto. apply Zle_trans with (ofs0 + 1); auto; omega.
+  destruct (zlt fr 8); eauto.
+  apply Zle_trans with (align ofs0 2). apply align_le; omega.
+  apply Zle_trans with (align ofs0 2 + 2); auto; omega.
+  set (ir' := align ir 2).
+  destruct (zlt ir' 4); eauto. 
+  apply Zle_trans with (align ofs0 2). apply align_le; omega.
+  apply Zle_trans with (align ofs0 2 + 2); auto; omega.
+  destruct (zlt fr 8); eauto.
+  apply Zle_trans with (ofs0 + 1); eauto. omega.
+Qed.
+
+Remark size_arguments_vararg_above:
+  forall tyl ofs0,
+  Zmax 0 ofs0 <= size_arguments_vararg tyl ofs0.
+Proof.
+  induction tyl; simpl; intros.
+  omega.
+  destruct a; (eapply Zle_trans; [idtac|eauto]). 
+  xomega.
+  assert (ofs0 <= align ofs0 2) by (apply align_le; omega). xomega.
+  assert (ofs0 <= align ofs0 2) by (apply align_le; omega). xomega.
+  xomega.
+Qed.
+
+Lemma size_arguments_above:
+  forall s, size_arguments s >= 0.
+Proof.
+  intros; unfold size_arguments. destruct (cc_vararg (sig_cc s)).
+  apply Zle_ge. change 0 with (Zmax 0 (-4)). apply size_arguments_vararg_above.  
+  apply Zle_ge. apply size_arguments_rec_above.
+Qed.
+
+Lemma loc_arguments_rec_bounded:
+  forall ofs ty tyl ir fr ofs0,
+  In (S Outgoing ofs ty) (loc_arguments_rec tyl ir fr ofs0) ->
+  ofs + typesize ty <= size_arguments_rec tyl ir fr ofs0.
+Proof.
+  induction tyl; simpl; intros.
+  elim H.
+  destruct a.
+- (* int *)
+  destruct (zlt ir 4); destruct H.
+  discriminate.
+  eauto.
+  inv H. apply size_arguments_rec_above. 
+  eauto.
+- (* float *)
+  destruct (zlt fr 8); destruct H.
+  discriminate.
+  eauto. 
+  inv H. apply size_arguments_rec_above. 
+  eauto.
+- (* long *)
+  destruct (zlt (align ir 2) 4).
+  destruct H. discriminate. destruct H. discriminate. eauto. 
+  destruct H. inv H.
+  rewrite <- Zplus_assoc. simpl. apply size_arguments_rec_above.
+  destruct H. inv H.
+  eapply Zle_trans. 2: apply size_arguments_rec_above. simpl; omega.
+  eauto.
+- (* float *)
+  destruct (zlt fr 8); destruct H.
+  discriminate.
+  eauto.
+  inv H. apply size_arguments_rec_above.
+  eauto.
+Qed.
+
+Lemma loc_arguments_vararg_bounded:
+  forall ofs ty tyl ofs0,
+  In (S Outgoing ofs ty) (loc_arguments_vararg tyl ofs0) ->
+  Zmax 0 (ofs + typesize ty) <= size_arguments_vararg tyl ofs0.
+Proof.
+  induction tyl; simpl; intros.
+  elim H.
+  destruct a.
+- (* int *)
+  destruct H. 
+  destruct (zlt ofs0 0); inv H. apply size_arguments_vararg_above. 
+  eauto.
+- (* float *)
+  destruct H.
+  destruct (zlt (align ofs0 2) 0); inv H. apply size_arguments_vararg_above.
+  eauto.
+- (* long *)
+  destruct H. 
+  destruct (zlt (align ofs0 2) 0); inv H.
+  rewrite <- Zplus_assoc. simpl. apply size_arguments_vararg_above.
+  destruct H. 
+  destruct (zlt (align ofs0 2) 0); inv H.
+  eapply Zle_trans. 2: apply size_arguments_vararg_above. simpl; xomega.
+  eauto.
+- (* float *)
+  destruct H. 
+  destruct (zlt ofs0 0); inv H. apply size_arguments_vararg_above. 
+  eauto.
+Qed.
+
+Lemma loc_arguments_bounded:
+  forall (s: signature) (ofs: Z) (ty: typ),
+  In (S Outgoing ofs ty) (loc_arguments s) ->
+  ofs + typesize ty <= size_arguments s.
+Proof.
+  unfold loc_arguments, size_arguments; intros.
+  destruct (cc_vararg (sig_cc s)).
+  eapply Zle_trans. 2: eapply loc_arguments_vararg_bounded; eauto. xomega. 
+  eapply loc_arguments_rec_bounded; eauto.
+Qed.
diff --git a/arm/hardfloat/Stacklayout.v b/arm/hardfloat/Stacklayout.v
new file mode 100644
index 0000000..7694dcf
--- /dev/null
+++ b/arm/hardfloat/Stacklayout.v
@@ -0,0 +1,132 @@
+(* *********************************************************************)
+(*                                                                     *)
+(*              The Compcert verified compiler                         *)
+(*                                                                     *)
+(*          Xavier Leroy, INRIA Paris-Rocquencourt                     *)
+(*                                                                     *)
+(*  Copyright Institut National de Recherche en Informatique et en     *)
+(*  Automatique.  All rights reserved.  This file is distributed       *)
+(*  under the terms of the INRIA Non-Commercial License Agreement.     *)
+(*                                                                     *)
+(* *********************************************************************)
+
+(** Machine- and ABI-dependent layout information for activation records. *)
+
+Require Import Coqlib.
+Require Import Bounds.
+
+(** The general shape of activation records is as follows,
+  from bottom (lowest offsets) to top:
+- Space for outgoing arguments to function calls.
+- Local stack slots.
+- Saved values of integer callee-save registers used by the function.
+- Saved values of float callee-save registers used by the function.
+- Saved return address into caller.
+- Pointer to activation record of the caller.
+- Space for the stack-allocated data declared in Cminor.
+
+The [frame_env] compilation environment records the positions of
+the boundaries between areas in the frame part.
+*)
+
+Definition fe_ofs_arg := 0.
+
+Record frame_env : Type := mk_frame_env {
+  fe_size: Z;
+  fe_ofs_link: Z;
+  fe_ofs_retaddr: Z;
+  fe_ofs_local: Z;
+  fe_ofs_int_callee_save: Z;
+  fe_num_int_callee_save: Z;
+  fe_ofs_float_callee_save: Z;
+  fe_num_float_callee_save: Z;
+  fe_stack_data: Z
+}.
+
+(** Computation of the frame environment from the bounds of the current
+  function. *)
+
+Definition make_env (b: bounds) :=
+  let ol := align (4 * b.(bound_outgoing)) 8 in    (* locals *)
+  let oics := ol + 4 * b.(bound_local) in (* integer callee-saves *)
+  let oendi := oics + 4 * b.(bound_int_callee_save) in
+  let ofcs := align oendi 8 in       (* float callee-saves *)
+  let ora := ofcs + 8 * b.(bound_float_callee_save) in (* retaddr *)
+  let olink := ora + 4 in (* back link *)
+  let ostkdata := olink + 4 in (* stack data *)
+  let sz := align (ostkdata + b.(bound_stack_data)) 8 in
+  mk_frame_env sz olink ora ol
+                  oics b.(bound_int_callee_save)
+                  ofcs b.(bound_float_callee_save)
+                  ostkdata.
+
+(** Separation property *)
+
+Remark frame_env_separated:
+  forall b,
+  let fe := make_env b in
+  0 <= fe_ofs_arg
+  /\ fe_ofs_arg + 4 * b.(bound_outgoing) <= fe.(fe_ofs_local)
+  /\ fe.(fe_ofs_local) + 4 * b.(bound_local) <= fe.(fe_ofs_int_callee_save)
+  /\ fe.(fe_ofs_int_callee_save) + 4 * b.(bound_int_callee_save) <= fe.(fe_ofs_float_callee_save)
+  /\ fe.(fe_ofs_float_callee_save) + 8 * b.(bound_float_callee_save) <= fe.(fe_ofs_retaddr)
+  /\ fe.(fe_ofs_retaddr) + 4 <= fe.(fe_ofs_link)
+  /\ fe.(fe_ofs_link) + 4 <= fe.(fe_stack_data)
+  /\ fe.(fe_stack_data) + b.(bound_stack_data) <= fe.(fe_size).
+Proof.
+  intros.
+  generalize (align_le (4 * bound_outgoing b) 8 (refl_equal)).
+  generalize (align_le (fe_ofs_int_callee_save fe + 4 * b.(bound_int_callee_save)) 8 (refl_equal _)).
+  generalize (align_le (fe_stack_data fe + b.(bound_stack_data)) 8 (refl_equal)).
+  unfold fe, make_env, fe_size, fe_ofs_link, fe_ofs_retaddr,
+    fe_ofs_local, fe_ofs_int_callee_save, fe_num_int_callee_save,
+    fe_ofs_float_callee_save, fe_num_float_callee_save,
+    fe_stack_data, fe_ofs_arg.
+  intros.
+  generalize (bound_local_pos b); intro;
+  generalize (bound_int_callee_save_pos b); intro;
+  generalize (bound_float_callee_save_pos b); intro;
+  generalize (bound_outgoing_pos b); intro;
+  generalize (bound_stack_data_pos b); intro.
+  omega.
+Qed.
+
+(** Alignment property *)
+
+Remark frame_env_aligned:
+  forall b,
+  let fe := make_env b in
+  (4 | fe.(fe_ofs_link))
+  /\ (8 | fe.(fe_ofs_local))
+  /\ (4 | fe.(fe_ofs_int_callee_save))
+  /\ (8 | fe.(fe_ofs_float_callee_save))
+  /\ (4 | fe.(fe_ofs_retaddr))
+  /\ (8 | fe.(fe_stack_data))
+  /\ (8 | fe.(fe_size)).
+Proof.
+  intros.
+  unfold fe, make_env, fe_size, fe_ofs_link, fe_ofs_retaddr,
+    fe_ofs_local, fe_ofs_int_callee_save, fe_num_int_callee_save,
+    fe_ofs_float_callee_save, fe_num_float_callee_save,
+    fe_stack_data.
+  set (x1 := 4 * bound_outgoing b).
+  assert (4 | x1). unfold x1; exists (bound_outgoing b); ring.
+  set (x2 := align x1 8).
+  assert (8 | x2). apply align_divides. omega. 
+  set (x3 := x2 + 4 * bound_local b).
+  assert (4 | x3). apply Zdivide_plus_r. apply Zdivides_trans with 8; auto. exists 2; auto.
+  exists (bound_local b); ring.
+  set (x4 := align (x3 + 4 * bound_int_callee_save b) 8).
+  assert (8 | x4). apply align_divides. omega.
+  set (x5 := x4 + 8 * bound_float_callee_save b).
+  assert (8 | x5). apply Zdivide_plus_r; auto. exists (bound_float_callee_save b); ring.
+  assert (4 | x5). apply Zdivides_trans with 8; auto. exists 2; auto.
+  set (x6 := x5 + 4). 
+  assert (4 | x6). apply Zdivide_plus_r; auto. exists 1; auto.
+  set (x7 := x6 + 4).
+  assert (8 | x7). unfold x7, x6. replace (x5 + 4 + 4) with (x5 + 8) by omega. 
+  apply Zdivide_plus_r; auto. exists 1; auto.
+  set (x8 := align (x7 + bound_stack_data b) 8).
+  assert (8 | x8). apply align_divides. omega.
+  tauto.
+Qed.