+ ARM/Aarch64: *xt*Op2Op optimizations

compiler/armgen/aoptarm.pas

@@ -622,6 +622,7 @@ Implementation
   function TARMAsmOptimizer.OptPass1UXTB(var p : tai) : Boolean;
     var
       hp1, hp2: tai;
+      so: tshifterop;
     begin
       Result:=false;
       if GetNextInstructionUsingReg(p,hp1,taicpu(p).oper[0]^.reg) then
@@ -720,6 +721,25 @@ Implementation
               taicpu(hp1).loadReg(1,taicpu(p).oper[1]^.reg);
               result:=RemoveCurrentP(p);
             end
+          else if MatchInstruction(hp1, [A_ADD,A_SUB], [C_None], [PF_None,PF_S]) and
+            (taicpu(p).ops=2) and
+            (taicpu(hp1).ops=3) and
+            MatchOperand(taicpu(hp1).oper[2]^, taicpu(p).oper[0]^.reg) and
+            not(MatchOperand(taicpu(hp1).oper[1]^, taicpu(p).oper[0]^.reg)) and
+            RegEndofLife(taicpu(p).oper[0]^.reg,taicpu(hp1)) and
+            { reg1 might not be modified inbetween }
+            not(RegModifiedBetween(taicpu(p).oper[1]^.reg,p,hp1)) then
+            begin
+              DebugMsg('Peephole UxtbOp2Op done', p);
+              AllocRegBetween(taicpu(p).oper[1]^.reg,p,hp1,UsedRegs);
+              taicpu(hp1).loadReg(2,taicpu(p).oper[1]^.reg);
+              taicpu(hp1).ops:=4;
+              shifterop_reset(so);
+              so.shiftmode:=SM_UXTB;
+              so.shiftimm:=0;
+              taicpu(hp1).loadshifterop(3,so);
+              result:=RemoveCurrentP(p);
+            end
           else if RemoveSuperfluousMove(p, hp1, 'UxtbMov2Uxtb') then
             Result:=true;
         end;
@@ -729,6 +749,7 @@ Implementation
   function TARMAsmOptimizer.OptPass1UXTH(var p : tai) : Boolean;
     var
       hp1: tai;
+      so: tshifterop;
     begin
       Result:=false;
       if GetNextInstructionUsingReg(p,hp1,taicpu(p).oper[0]^.reg) then
@@ -802,6 +823,25 @@ Implementation
               taicpu(hp1).loadReg(1,taicpu(p).oper[1]^.reg);
               result:=RemoveCurrentP(p);
             end
+          else if MatchInstruction(hp1, [A_ADD,A_SUB], [C_None], [PF_None,PF_S]) and
+            (taicpu(p).ops=2) and
+            (taicpu(hp1).ops=3) and
+            MatchOperand(taicpu(hp1).oper[2]^, taicpu(p).oper[0]^.reg) and
+            not(MatchOperand(taicpu(hp1).oper[1]^, taicpu(p).oper[0]^.reg)) and
+            RegEndofLife(taicpu(p).oper[0]^.reg,taicpu(hp1)) and
+            { reg1 might not be modified inbetween }
+            not(RegModifiedBetween(taicpu(p).oper[1]^.reg,p,hp1)) then
+            begin
+              DebugMsg('Peephole UxthOp2Op done', p);
+              AllocRegBetween(taicpu(p).oper[1]^.reg,p,hp1,UsedRegs);
+              taicpu(hp1).loadReg(2,taicpu(p).oper[1]^.reg);
+              taicpu(hp1).ops:=4;
+              shifterop_reset(so);
+              so.shiftmode:=SM_UXTH;
+              so.shiftimm:=0;
+              taicpu(hp1).loadshifterop(3,so);
+              result:=RemoveCurrentP(p);
+            end
           else if RemoveSuperfluousMove(p, hp1, 'UxthMov2Data') then
             Result:=true;
         end;
@@ -811,6 +851,7 @@ Implementation
   function TARMAsmOptimizer.OptPass1SXTB(var p : tai) : Boolean;
     var
       hp1, hp2: tai;
+      so: tshifterop;
     begin
       Result:=false;
       if GetNextInstructionUsingReg(p,hp1,taicpu(p).oper[0]^.reg) then
@@ -909,6 +950,25 @@ Implementation
               taicpu(hp1).loadReg(1,taicpu(p).oper[1]^.reg);
               result:=RemoveCurrentP(p);
             end
+          else if MatchInstruction(hp1, [A_ADD,A_SUB], [C_None], [PF_None,PF_S]) and
+            (taicpu(p).ops=2) and
+            (taicpu(hp1).ops=3) and
+            MatchOperand(taicpu(hp1).oper[2]^, taicpu(p).oper[0]^.reg) and
+            not(MatchOperand(taicpu(hp1).oper[1]^, taicpu(p).oper[0]^.reg)) and
+            RegEndofLife(taicpu(p).oper[0]^.reg,taicpu(hp1)) and
+            { reg1 might not be modified inbetween }
+            not(RegModifiedBetween(taicpu(p).oper[1]^.reg,p,hp1)) then
+            begin
+              DebugMsg('Peephole SxtbOp2Op done', p);
+              AllocRegBetween(taicpu(p).oper[1]^.reg,p,hp1,UsedRegs);
+              taicpu(hp1).loadReg(2,taicpu(p).oper[1]^.reg);
+              taicpu(hp1).ops:=4;
+              shifterop_reset(so);
+              so.shiftmode:=SM_SXTB;
+              so.shiftimm:=0;
+              taicpu(hp1).loadshifterop(3,so);
+              result:=RemoveCurrentP(p);
+            end
           else if GetNextInstructionUsingReg(p, hp1, taicpu(p).oper[0]^.reg) and
             RemoveSuperfluousMove(p, hp1, 'SxtbMov2Sxtb') then
             Result:=true;
@@ -919,6 +979,7 @@ Implementation
   function TARMAsmOptimizer.OptPass1SXTH(var p : tai) : Boolean;
     var
       hp1: tai;
+      so: tshifterop;
     begin
       Result:=false;
       if GetNextInstructionUsingReg(p,hp1,taicpu(p).oper[0]^.reg) then
@@ -1017,6 +1078,25 @@ Implementation
               taicpu(hp1).loadReg(1,taicpu(p).oper[1]^.reg);
               result:=RemoveCurrentP(p);
             end
+          else if MatchInstruction(hp1, [A_ADD,A_SUB], [C_None], [PF_None,PF_S]) and
+            (taicpu(p).ops=2) and
+            (taicpu(hp1).ops=3) and
+            MatchOperand(taicpu(hp1).oper[2]^, taicpu(p).oper[0]^.reg) and
+            not(MatchOperand(taicpu(hp1).oper[1]^, taicpu(p).oper[0]^.reg)) and
+            RegEndofLife(taicpu(p).oper[0]^.reg,taicpu(hp1)) and
+            { reg1 might not be modified inbetween }
+            not(RegModifiedBetween(taicpu(p).oper[1]^.reg,p,hp1)) then
+            begin
+              DebugMsg('Peephole SxthOp2Op done', p);
+              AllocRegBetween(taicpu(p).oper[1]^.reg,p,hp1,UsedRegs);
+              taicpu(hp1).loadReg(2,taicpu(p).oper[1]^.reg);
+              taicpu(hp1).ops:=4;
+              shifterop_reset(so);
+              so.shiftmode:=SM_SXTH;
+              so.shiftimm:=0;
+              taicpu(hp1).loadshifterop(3,so);
+              result:=RemoveCurrentP(p);
+            end
           else if GetNextInstructionUsingReg(p, hp1, taicpu(p).oper[0]^.reg) and
             RemoveSuperfluousMove(p, hp1, 'SxthMov2Sxth') then
             Result:=true;