- Revised non-overflow constraints on memory injections so that

  injections compose (Values, Memdata, Memory)
- Memory chunks: Mfloat64 now has alignment 8; introduced Mfloat64al32
  that works like old Mfloat64 (i.e. has alignment 4); simplified
  handling of memcpy builtin accordingly.


git-svn-id: https://yquem.inria.fr/compcert/svn/compcert/trunk@1983 fca1b0fc-160b-0410-b1d3-a4f43f01ea2e

5 files changed, 20 insertions, 13 deletions

diff --git a/ia32/Asm.v b/ia32/Asm.v
index 6210286..a78c8bf 100644
--- a/ia32/Asm.v
+++ b/ia32/Asm.v
@@ -474,17 +474,17 @@ Definition exec_instr (c: code) (i: instruction) (rs: regset) (m: mem) : outcome
   | Pmovsd_fi rd n =>
       Next (nextinstr (rs#rd <- (Vfloat n))) m
   | Pmovsd_fm rd a =>
-      exec_load Mfloat64 m a rs rd
+      exec_load Mfloat64al32 m a rs rd
   | Pmovsd_mf a r1 =>
-      exec_store Mfloat64 m a rs r1
+      exec_store Mfloat64al32 m a rs r1
   | Pfld_f r1 =>
       Next (nextinstr (rs#ST0 <- (rs r1))) m
   | Pfld_m a =>
-      exec_load Mfloat64 m a rs ST0
+      exec_load Mfloat64al32 m a rs ST0
   | Pfstp_f rd =>
       Next (nextinstr (rs#rd <- (rs ST0) #ST0 <- Vundef)) m
   | Pfstp_m a =>
-      exec_store Mfloat64 m a rs ST0
+      exec_store Mfloat64al32 m a rs ST0
   | Pxchg_rr r1 r2 =>
       Next (nextinstr (rs#r1 <- (rs r2) #r2 <- (rs r1))) m
   (** Moves with conversion *)
@@ -709,7 +709,7 @@ Inductive extcall_arg (rs: regset) (m: mem): loc -> val -> Prop :=
       extcall_arg rs m (S (Outgoing ofs Tint)) v
   | extcall_arg_float_stack: forall ofs bofs v,
       bofs = Stacklayout.fe_ofs_arg + 4 * ofs ->
-      Mem.loadv Mfloat64 m (Val.add (rs (IR ESP)) (Vint (Int.repr bofs))) = Some v ->
+      Mem.loadv Mfloat64al32 m (Val.add (rs (IR ESP)) (Vint (Int.repr bofs))) = Some v ->
       extcall_arg rs m (S (Outgoing ofs Tfloat)) v.
 
 Definition extcall_arguments
diff --git a/ia32/Asmgen.v b/ia32/Asmgen.v
index 3fc3efb..5a6c1ab 100644
--- a/ia32/Asmgen.v
+++ b/ia32/Asmgen.v
@@ -452,7 +452,7 @@ Definition transl_load (chunk: memory_chunk)
       do r <- ireg_of dest; OK(Pmov_rm r am :: k)
   | Mfloat32 =>
       do r <- freg_of dest; OK(Pcvtss2sd_fm r am :: k)
-  | Mfloat64 =>
+  | Mfloat64 | Mfloat64al32 =>
       do r <- freg_of dest; OK(Pmovsd_fm r am :: k)
   end.
 
@@ -469,7 +469,7 @@ Definition transl_store (chunk: memory_chunk)
       do r <- ireg_of src; OK(Pmov_mr am r :: k)
   | Mfloat32 =>
       do r <- freg_of src; OK(Pcvtsd2ss_mf am r :: k)
-  | Mfloat64 =>
+  | Mfloat64 | Mfloat64al32 =>
       do r <- freg_of src; OK(Pmovsd_mf am r :: k)
   end.
 
diff --git a/ia32/Asmgenproof1.v b/ia32/Asmgenproof1.v
index 7b3427b..7ad8a02 100644
--- a/ia32/Asmgenproof1.v
+++ b/ia32/Asmgenproof1.v
@@ -1679,7 +1679,11 @@ Proof.
     transitivity (Val.add ((rs#(preg_of dest) <- v) PC) Vone).
     auto. decEq. apply Pregmap.gso; auto with ppcgen.
   exists rs2. split. 
-  destruct chunk; ArgsInv; apply exec_straight_one; simpl; auto.
+  destruct chunk; ArgsInv; apply exec_straight_one; auto.
+  (* Mfloat64 -> Mfloat64al32 *)
+  rewrite <- H. simpl. unfold exec_load. rewrite H1.
+  destruct (eval_addrmode ge x rs); simpl in *; try discriminate.
+  erewrite Mem.load_float64al32; eauto. 
   split. unfold rs2. rewrite nextinstr_nf_inv1. SRes. apply preg_of_important.
   intros. unfold rs2. repeat SOther. 
 Qed.
@@ -1726,6 +1730,10 @@ Proof.
   intros. SOther.
 (* float64 *)
   econstructor; split.
+  apply exec_straight_one. simpl. unfold exec_store. erewrite Mem.storev_float64al32; eauto. auto.
+  intros. SOther.
+(* float64al32 *)
+  econstructor; split.
   apply exec_straight_one. simpl. unfold exec_store. rewrite H1. eauto. auto.
   intros. SOther.
 Qed.
diff --git a/ia32/PrintAsm.ml b/ia32/PrintAsm.ml
index 8b795ee..b473464 100644
--- a/ia32/PrintAsm.ml
+++ b/ia32/PrintAsm.ml
@@ -363,7 +363,7 @@ let print_builtin_vload_common oc chunk addr res =
       fprintf oc "	movl	%a, %a\n" addressing addr ireg res
   | Mfloat32, FR res ->
       fprintf oc "	cvtss2sd %a, %a\n" addressing addr freg res
-  | Mfloat64, FR res ->
+  | (Mfloat64 | Mfloat64al32), FR res ->
       fprintf oc "	movsd	%a, %a\n" addressing addr freg res
   | _ ->
       assert false
@@ -406,7 +406,7 @@ let print_builtin_vstore_common oc chunk addr src =
   | Mfloat32, FR src ->
       fprintf oc "	cvtsd2ss %a, %%xmm7\n" freg src;
       fprintf oc "	movss	%%xmm7, %a\n" addressing addr
-  | Mfloat64, FR src ->
+  | (Mfloat64 | Mfloat64al32), FR src ->
       fprintf oc "	movsd	%a, %a\n" freg src addressing addr
   | _ ->
       assert false
diff --git a/ia32/standard/Stacklayout.v b/ia32/standard/Stacklayout.v
index 063fb4f..be854fd 100644
--- a/ia32/standard/Stacklayout.v
+++ b/ia32/standard/Stacklayout.v
@@ -107,7 +107,7 @@ Remark frame_env_aligned:
   /\ (8 | fe.(fe_ofs_float_local))
   /\ (8 | fe.(fe_ofs_float_callee_save))
   /\ (4 | fe.(fe_ofs_retaddr))
-  /\ (4 | fe.(fe_stack_data))
+  /\ (8 | fe.(fe_stack_data))
   /\ (4 | fe.(fe_size)).
 Proof.
   intros.
@@ -128,8 +128,7 @@ Proof.
   set (x6 := x5 + 8 * bound_float_local b).
   assert (8 | x6). unfold x6. apply Zdivide_plus_r; auto. exists (bound_float_local b); ring.
   set (x7 := x6 + 8 * bound_float_callee_save b).
-  assert (4 | x7).
-    apply Zdivides_trans with 8. exists 2; auto.
+  assert (8 | x7).
     unfold x7. apply Zdivide_plus_r; auto. exists (bound_float_callee_save b); ring.
   set (x8 := align (x7 + bound_stack_data b) 4).
   assert (4 | x8). apply align_divides. omega.