+ implement assembler optimization Str/LdrAdd/Sub2Str/Ldr Postindex done

git-svn-id: trunk@40017 -

compiler/aarch64/aoptcpu.pas

@@ -21,26 +21,44 @@
  ****************************************************************************
 }
 
-
 Unit aoptcpu;
 
 {$i fpcdefs.inc}
 
-Interface
+{ $define DEBUG_AOPTCPU}
 
-uses cpubase, aasmtai, aopt, aoptcpub;
+Interface
 
-Type
-  TCpuAsmOptimizer = class(TAsmOptimizer)
-    { uses the same constructor as TAopObj }
-    function PeepHoleOptPass1Cpu(var p: tai): boolean; override;
-    procedure PeepHoleOptPass2;override;
-  End;
+    uses
+      globtype, globals,
+      cutils,
+      cgbase, cpubase, aasmtai, aasmcpu, aopt, aoptcpub;
+
+    Type
+      TCpuAsmOptimizer = class(TAsmOptimizer)
+        { uses the same constructor as TAopObj }
+        function PeepHoleOptPass1Cpu(var p: tai): boolean; override;
+        procedure PeepHoleOptPass2;override;
+        function GetNextInstructionUsingReg(Current : tai; out Next : tai; reg : TRegister) : Boolean;
+        function LookForPostindexedPattern(p : taicpu) : boolean;
+        procedure DebugMsg(const s : string; p : tai);
+      End;
 
 Implementation
 
   uses
-    aasmbase,aasmcpu,cgbase;
+    aasmbase;
+
+{$ifdef DEBUG_AOPTCPU}
+  procedure TCpuAsmOptimizer.DebugMsg(const s: string;p : tai);
+    begin
+      asml.insertbefore(tai_comment.Create(strpnew(s)), p);
+    end;
+{$else DEBUG_AOPTCPU}
+  procedure TCpuAsmOptimizer.DebugMsg(const s: string;p : tai);inline;
+    begin
+    end;
+{$endif DEBUG_AOPTCPU}
 
   function CanBeCond(p : tai) : boolean;
     begin
@@ -48,11 +66,105 @@ Implementation
     end;
 
 
+  function MatchInstruction(const instr: tai; const op: TAsmOps; const postfix: TOpPostfixes): boolean;
+    begin
+      result :=
+        (instr.typ = ait_instruction) and
+        ((op = []) or (taicpu(instr).opcode in op)) and
+        ((postfix = []) or (taicpu(instr).oppostfix in postfix));
+    end;
+
+
+  function MatchInstruction(const instr: tai; const op: TAsmOp; const postfix: TOpPostfixes): boolean;
+    begin
+      result :=
+        (instr.typ = ait_instruction) and
+        (taicpu(instr).opcode = op) and
+        ((postfix = []) or (taicpu(instr).oppostfix in postfix));
+    end;
+
+
+  function TCpuAsmOptimizer.GetNextInstructionUsingReg(Current: tai;
+    Out Next: tai; reg: TRegister): Boolean;
+    begin
+      Next:=Current;
+      repeat
+        Result:=GetNextInstruction(Next,Next);
+      until not (Result) or
+            not(cs_opt_level3 in current_settings.optimizerswitches) or
+            (Next.typ<>ait_instruction) or
+            RegInInstruction(reg,Next) or
+            is_calljmp(taicpu(Next).opcode);
+    end;
+
+  {
+    optimize
+      ldr/str regX,[reg1]
+      ...
+      add/sub reg1,reg1,regY/const
+
+      into
+
+      ldr/str regX,[reg1], regY/const
+  }
+  function TCpuAsmOptimizer.LookForPostindexedPattern(p: taicpu) : boolean;
+    var
+      hp1 : tai;
+    begin
+      Result:=false;
+      if (p.oper[1]^.typ = top_ref) and
+        (p.oper[1]^.ref^.addressmode=AM_OFFSET) and
+        (p.oper[1]^.ref^.index=NR_NO) and
+        (p.oper[1]^.ref^.offset=0) and
+        GetNextInstructionUsingReg(p, hp1, p.oper[1]^.ref^.base) and
+        { we cannot check NR_DEFAULTFLAGS for modification yet so don't allow a condition }
+        MatchInstruction(hp1, [A_ADD, A_SUB], [PF_None]) and
+        (taicpu(hp1).oper[0]^.reg=p.oper[1]^.ref^.base) and
+        (taicpu(hp1).oper[1]^.reg=p.oper[1]^.ref^.base) and
+        (
+         { valid offset? }
+         (taicpu(hp1).oper[2]^.typ=top_const) and
+         (taicpu(hp1).oper[2]^.val>=-256) and
+         (abs(taicpu(hp1).oper[2]^.val)<256)
+        ) and
+        { don't apply the optimization if the base register is loaded }
+        (p.oper[0]^.reg<>p.oper[1]^.ref^.base) and
+        not(RegModifiedBetween(taicpu(hp1).oper[0]^.reg,p,hp1)) and
+        { don't apply the optimization if the (new) index register is loaded }
+        (p.oper[0]^.reg<>taicpu(hp1).oper[2]^.reg) and
+        not(RegModifiedBetween(taicpu(hp1).oper[2]^.reg,p,hp1)) then
+        begin
+          DebugMsg('Peephole Str/LdrAdd/Sub2Str/Ldr Postindex done', p);
+          p.oper[1]^.ref^.addressmode:=AM_POSTINDEXED;
+          if taicpu(hp1).opcode=A_ADD then
+            p.oper[1]^.ref^.offset:=taicpu(hp1).oper[2]^.val
+          else
+            p.oper[1]^.ref^.offset:=-taicpu(hp1).oper[2]^.val;
+          asml.Remove(hp1);
+          hp1.Free;
+          Result:=true;
+        end;
+    end;
+
+
   function TCpuAsmOptimizer.PeepHoleOptPass1Cpu(var p: tai): boolean;
     var
       next1: tai;
     begin
       result := false;
+      if p.typ=ait_instruction then
+        begin
+          case taicpu(p).opcode of
+            A_LDR:
+              begin
+                Result:=LookForPostindexedPattern(taicpu(p));
+              end;
+            A_STR:
+              begin
+                Result:=LookForPostindexedPattern(taicpu(p));
+              end;
+          end;
+        end;
     end;
 
 

compiler/aarch64/cpubase.pas

@@ -48,6 +48,8 @@ unit cpubase;
     type
       TAsmOp= {$i a64op.inc}
 
+      TAsmOps = set of TAsmOp;
+
       { This should define the array of instructions as string }
       op2strtable=array[tasmop] of string[11];