Smart scheduler build problem and flatten solution OK

1 files changed, 121 insertions, 20 deletions

diff --git a/aarch64/OpWeightsAsm.ml b/aarch64/OpWeightsAsm.ml
index 17869a99..c59d4695 100644
--- a/aarch64/OpWeightsAsm.ml
+++ b/aarch64/OpWeightsAsm.ml
@@ -1,19 +1,120 @@
+(* *************************************************************)
+(*                                                             *)
+(*             The Compcert verified compiler                  *)
+(*                                                             *)
+(*           David Monniaux     CNRS, VERIMAG                  *)
+(*           Léo Gourdin        UGA, VERIMAG                   *)
+(*                                                             *)
+(*  Copyright Kalray. Copyright VERIMAG. All rights reserved.  *)
+(*  This file is distributed under the terms of the INRIA      *)
+(*  Non-Commercial License Agreement.                          *)
+(*                                                             *)
+(* *************************************************************)
+
 open Asmblock;;
-(* TODO remove? open PrepassSchedulingOracleDeps;;*)
+
+(*type called_function = (Registers.reg, AST.ident) Datatypes.sum*)
+  
+type opweights =
+  {
+    pipelined_resource_bounds : int array;
+    nr_non_pipelined_units : int;
+    latency_of_op : instruction -> int -> int;
+    resources_of_op : instruction -> int -> int array;
+    (*non_pipelined_resources_of_op : Op.operation -> int -> int array;*)
+    (*latency_of_load : AST.trapping_mode -> AST.memory_chunk -> Op.addressing -> int -> int;*)
+    (*resources_of_load : AST.trapping_mode -> AST.memory_chunk -> Op.addressing -> int -> int array;*)
+    (*resources_of_store : AST.memory_chunk -> Op.addressing -> int -> int array;*)
+    (*resources_of_cond : Op.condition -> int -> int array;*)
+    (*latency_of_call : AST.signature -> called_function -> int;*)
+    (*resources_of_call : AST.signature -> called_function -> int array;*)
+    (*resources_of_builtin : AST.external_function -> int array*)
+  };;
 
 module Cortex_A53=
   struct
     let resource_bounds = [| 2; 2; 1; 1 |];; (* instr ; ALU ; MAC; LSU *)
     let nr_non_pipelined_units = 1;;
 
-    let latency_of_op (i : basic) (nargs : int) =
-      match i with
+    let latency_of_op (i : instruction) (nargs : int) = 1;;
+      (*match i with
       | PArith _
-        | PArithP _ _ -> 1 (* XXX *)
-        | PArithPP
+        | PArithP _ _ _
+          | Padrp _ _ -> 1 (* XXX not sure *)
+          | Pmovz _ _ _ -> 1
+          | Pmovn _ _ _ -> 1
+          | Pfmovimms _ -> 1
+          | Pfmovimmd _ -> 1
+        | PArithPP _ _ _
+          | Pmov -> 1 (* XXX not sure *)
+          | Pmovk _ _ _ -> 1
+          | Paddadr _ _ -> 1
+          | Psbfiz _ _ _ -> 2 (* XXX Maybe 1 *)
+          | Psbfx _ _ _ -> 2
+          | Pubfiz _ _ _ -> 2 (* XXX Maybe 1 *)
+          | Pubfx _ _ _ -> 2
+          | Pfmov -> 1
+          | Pfcvtds -> 6
+          | Pfcvtsd -> 6
+          | Pfabs _ -> 6
+          | Pfneg _ -> 6
+          | Pscvtf _ _ -> 6
+          | Pucvtf _ _ -> 6
+          | Pfcvtzs _ _ -> 6
+          | Pfcvtzu _ _ -> 6
+          | Paddimm _ _ -> 1
+          | Psubimm _ _ -> 1
+        | PArithPPP _ _ _ _
+          | Pasrv _ -> 2
+          | Plslv _ -> 2
+          | Plsrv _ -> 2
+          | Prorv _ -> 2 (* XXX Does not exits in Asmblockgen?? *)
+          | Psmulh -> 4
+          | Pumulh -> 4
+          | Psdiv _ -> 50
+          | Pudiv _ -> 50
+          | Paddext _ -> 2
+          | Psubext _ -> 2
+          | Pfadd _ -> 6
+          | Pfdiv _ -> 50
+          | Pfmul _ -> 6
+          | Pfsub _ -> 6
+        | PArithRR0R _ _ _ _
+          | Padd _ _ -> 2 (* XXX Maybe 6 as used in shifts *)
+          | Psub _ _ -> 2 (* XXX Maybe 6 as used in shifts *)
+          | Pand _ _ -> 1 (* XXX Maybe 2 as used in shifts *)
+          | Pbic _ _ -> 1 (* XXX Maybe 2 as used in shifts *)
+          | Peon _ _ -> 1 (* XXX Maybe 2 as used in shifts *)
+          | Peor _ _ -> 1 (* XXX Maybe 2 as used in shifts *)
+          | Porr _ _ -> 1 (* XXX Maybe 2 as used in shifts *)
+          | Porn _ _ -> 1 (* XXX Maybe 2 as used in shifts *)
+        | PArithRR0 _ _ _
+          | Pandimm _ _ -> 1
+          | Peorimm _ _ -> 1
+          | Porrimm _ _ -> 1
+        | PArithARRRR0 _ _ _ _ _
+          | Pmadd _ -> 4
+          | Pmsub _ -> 4
+        | PArithComparisonPP _ _ _ (* XXX not sure *)
+          | Pcmpext _ -> 1
+          | Pcmnext _ -> 1
+          | Pfcmp _ -> 1
+        | PArithComparisonR0R _ _ _ (* XXX not sure *)
+          | Pcmp _ _ -> 1
+          | Pcmn _ _ -> 1
+          | Ptst _ _ -> 1
+        | PArithComparisonP _ _ _ (* XXX not sure *)
+          | Pfcmp0 _ -> 1
+          | Pcmpimm _ _ -> 1
+          | Pcmnimm _ _ -> 1
+          | Ptstimm _ _ -> 1
+        | Printf.eprintf "Error NYI OpWeightsAsm"*)
+
+
+
     
-    let resources_of_op (op : operation) (nargs : int) =
-      match op with
+    let resources_of_op (i : instruction) (nargs : int) = [| 0; 0; 0; 0 |];;
+      (*match op with
       | Omove 
         | Ointconst _
         | Olongconst _
@@ -179,9 +280,9 @@ module Cortex_A53=
          | Ccompfszero _                         (* r equal to 0.0 *)
          | Cnotcompfszero _                     (* r not equal to 0.0 *) ->
           [| 1; 1; 1; 0 |]
-       | _ -> [| 1; 1; 0; 0 |] );;
+       | _ -> [| 1; 1; 0; 0 |] )*);;
     
-    let latency_of_load trap chunk (addr : addressing) (nargs : int) = 3;;
+    let latency_of_load trap chunk _ _ = 3;;
     let latency_of_call _ _ = 6;;
     
     let resources_of_load trap chunk addressing nargs = [| 1; 0; 0; 1 |];;
@@ -193,20 +294,20 @@ module Cortex_A53=
   end;;
 
 let get_opweights () : opweights =
-  match !Clflags.option_mtune with
-  | "cortex-a53" | "cortex-a35" | "" ->
+  (*match !Clflags.option_mtune with*)
+  (*| "cortex-a53" | "cortex-a35" | "" ->*)
      {
        pipelined_resource_bounds = Cortex_A53.resource_bounds;
        nr_non_pipelined_units = Cortex_A53.nr_non_pipelined_units;
        latency_of_op = Cortex_A53.latency_of_op;
        resources_of_op = Cortex_A53.resources_of_op;
-       non_pipelined_resources_of_op = Cortex_A53.non_pipelined_resources_of_op;
-       latency_of_load = Cortex_A53.latency_of_load;
-       resources_of_load = Cortex_A53.resources_of_load;
-       resources_of_store = Cortex_A53.resources_of_store;
-       resources_of_cond = Cortex_A53.resources_of_cond;
-       latency_of_call = Cortex_A53.latency_of_call;
-       resources_of_call = Cortex_A53.resources_of_call;
-       resources_of_builtin = Cortex_A53.resources_of_builtin
+       (*non_pipelined_resources_of_op = Cortex_A53.non_pipelined_resources_of_op;*)
+       (*latency_of_load = Cortex_A53.latency_of_load;*)
+       (*resources_of_load = Cortex_A53.resources_of_load;*)
+       (*resources_of_store = Cortex_A53.resources_of_store;*)
+       (*resources_of_cond = Cortex_A53.resources_of_cond;*)
+       (*latency_of_call = Cortex_A53.latency_of_call;*)
+       (*resources_of_call = Cortex_A53.resources_of_call;*)
+       (*resources_of_builtin = Cortex_A53.resources_of_builtin*)
      }
-  | xxx -> failwith (Printf.sprintf "unknown -mtune: %s" xxx);;
+  (*| xxx -> failwith (Printf.sprintf "unknown -mtune: %s" xxx);;*)