|
|
@@ -42,7 +42,8 @@ unit aoptx86;
|
|
|
TOptsToCheck = (
|
|
|
aoc_MovAnd2Mov_3,
|
|
|
aoc_ForceNewIteration,
|
|
|
- aoc_DoPass2JccOpts
|
|
|
+ aoc_DoPass2JccOpts,
|
|
|
+ aoc_MovlMovq2MovlMovl
|
|
|
);
|
|
|
|
|
|
TX86AsmOptimizer = class(TAsmOptimizer)
|
|
|
@@ -11453,59 +11454,74 @@ unit aoptx86;
|
|
|
if MatchOpType(taicpu(p), top_reg, top_reg) and
|
|
|
(taicpu(p).opsize = S_L) then
|
|
|
begin
|
|
|
- TransferUsedRegs(TmpUsedRegs);
|
|
|
- { Mark the start point for sequential calls to
|
|
|
- GetNextInstructionUsingReg, RegModifiedBetween and
|
|
|
- UpdateUsedRegsBetween in case this optimisation is run multiple
|
|
|
- times }
|
|
|
- hp2 := p;
|
|
|
- repeat
|
|
|
- if (
|
|
|
- not(cs_opt_level3 in current_settings.optimizerswitches) or
|
|
|
- { Look further ahead for this one }
|
|
|
- GetNextInstructionUsingReg(hp2, hp1, taicpu(p).oper[1]^.reg)
|
|
|
- ) and
|
|
|
- MatchInstruction(hp1,A_MOV,[S_Q]) and
|
|
|
- not RegModifiedBetween(taicpu(p).oper[0]^.reg, hp2, hp1) and
|
|
|
- MatchOpType(taicpu(hp1), top_reg, top_reg) and
|
|
|
- SuperRegistersEqual(taicpu(p).oper[1]^.reg, taicpu(hp1).oper[0]^.reg) then
|
|
|
- begin
|
|
|
- UpdateUsedRegsBetween(TmpUsedRegs, tai(hp2.Next), hp1);
|
|
|
-
|
|
|
- taicpu(hp1).opsize := S_L;
|
|
|
- taicpu(hp1).loadreg(0, taicpu(p).oper[0]^.reg);
|
|
|
- setsubreg(taicpu(hp1).oper[1]^.reg, R_SUBD);
|
|
|
+ { If the movq instruction is followed by addq or subq, it
|
|
|
+ might be possible to convert them to a leaq instruction
|
|
|
+ whose opportunity might be lost if it's changed to a movl
|
|
|
+ first, so we can't do this optimisation on a first iteration }
|
|
|
+ if not (aoc_MovlMovq2MovlMovl in OptsToCheck) and
|
|
|
+ not NotFirstIteration and
|
|
|
+ { If -O2 and under, do the optimisation anyway because Pass 2
|
|
|
+ won't run more than once }
|
|
|
+ (cs_opt_level3 in current_settings.optimizerswitches) then
|
|
|
+ begin
|
|
|
+ { Flag that we need to run Pass 2 again }
|
|
|
+ Include(OptsToCheck, aoc_ForceNewIteration);
|
|
|
+ end
|
|
|
+ else
|
|
|
+ begin
|
|
|
+ TransferUsedRegs(TmpUsedRegs);
|
|
|
+ { Mark the start point for sequential calls to
|
|
|
+ GetNextInstructionUsingReg, RegModifiedBetween and
|
|
|
+ UpdateUsedRegsBetween in case this optimisation is run multiple
|
|
|
+ times }
|
|
|
+ hp2 := p;
|
|
|
+ repeat
|
|
|
+ if (
|
|
|
+ not(cs_opt_level3 in current_settings.optimizerswitches) or
|
|
|
+ { Look further ahead for this one }
|
|
|
+ GetNextInstructionUsingReg(hp2, hp1, taicpu(p).oper[1]^.reg)
|
|
|
+ ) and
|
|
|
+ MatchInstruction(hp1,A_MOV,[S_Q]) and
|
|
|
+ not RegModifiedBetween(taicpu(p).oper[0]^.reg, hp2, hp1) and
|
|
|
+ MatchOpType(taicpu(hp1), top_reg, top_reg) and
|
|
|
+ SuperRegistersEqual(taicpu(p).oper[1]^.reg, taicpu(hp1).oper[0]^.reg) then
|
|
|
+ begin
|
|
|
+ UpdateUsedRegsBetween(TmpUsedRegs, tai(hp2.Next), hp1);
|
|
|
|
|
|
- AllocRegBetween(taicpu(p).oper[0]^.reg, p, hp1, UsedRegs);
|
|
|
+ taicpu(hp1).opsize := S_L;
|
|
|
+ taicpu(hp1).loadreg(0, taicpu(p).oper[0]^.reg);
|
|
|
+ setsubreg(taicpu(hp1).oper[1]^.reg, R_SUBD);
|
|
|
|
|
|
- DebugMsg(SPeepholeOptimization + 'Made 32-to-64-bit zero extension more efficient (MovlMovq2MovlMovl 1)', hp1);
|
|
|
+ AllocRegBetween(taicpu(p).oper[0]^.reg, p, hp1, TmpUsedRegs);
|
|
|
|
|
|
- if not RegUsedAfterInstruction(taicpu(p).oper[1]^.reg, hp1, TmpUsedRegs) then
|
|
|
- begin
|
|
|
- DebugMsg(SPeepholeOptimization + 'Mov2Nop 8 done', p);
|
|
|
- RemoveCurrentP(p);
|
|
|
- Result := True;
|
|
|
- Exit;
|
|
|
- end;
|
|
|
+ DebugMsg(SPeepholeOptimization + 'Made 32-to-64-bit zero extension more efficient (MovlMovq2MovlMovl 1)', hp1);
|
|
|
|
|
|
- { Initial instruction wasn't actually changed }
|
|
|
- Include(OptsToCheck, aoc_ForceNewIteration);
|
|
|
+ if not RegUsedAfterInstruction(taicpu(p).oper[1]^.reg, hp1, TmpUsedRegs) then
|
|
|
+ begin
|
|
|
+ DebugMsg(SPeepholeOptimization + 'Mov2Nop 8 done', p);
|
|
|
+ RemoveCurrentP(p);
|
|
|
+ Result := True;
|
|
|
+ Exit;
|
|
|
+ end;
|
|
|
|
|
|
- if (cs_opt_level3 in current_settings.optimizerswitches) then
|
|
|
- begin
|
|
|
- { GetNextInstructionUsingReg will return a different
|
|
|
- instruction, so check this optimisation again }
|
|
|
+ { Initial instruction wasn't actually changed }
|
|
|
+ Include(OptsToCheck, aoc_ForceNewIteration);
|
|
|
|
|
|
- { Update the start point for the next calls to
|
|
|
- GetNextInstructionUsingReg, RegModifiedBetween and
|
|
|
- UpdateUsedRegsBetween to grant a speed boost }
|
|
|
- hp2 := hp1;
|
|
|
- Continue; { Jump back to "repeat" }
|
|
|
+ if (cs_opt_level3 in current_settings.optimizerswitches) then
|
|
|
+ begin
|
|
|
+ { GetNextInstructionUsingReg will return a different
|
|
|
+ instruction, so check this optimisation again }
|
|
|
+
|
|
|
+ { Update the start point for the next calls to
|
|
|
+ GetNextInstructionUsingReg, RegModifiedBetween and
|
|
|
+ UpdateUsedRegsBetween to grant a speed boost }
|
|
|
+ hp2 := hp1;
|
|
|
+ Continue; { Jump back to "repeat" }
|
|
|
+ end;
|
|
|
end;
|
|
|
- end;
|
|
|
-
|
|
|
- Break;
|
|
|
- until False;
|
|
|
+ Break;
|
|
|
+ until False;
|
|
|
+ end;
|
|
|
end;
|
|
|
{$endif x86_64}
|
|
|
|
|
|
@@ -16306,8 +16322,10 @@ unit aoptx86;
|
|
|
UpdateUsedRegs(UsedRegs, tai(p.Next));
|
|
|
|
|
|
hp2 := hp1;
|
|
|
+ Include(OptsToCheck, aoc_MovlMovq2MovlMovl);
|
|
|
if OptPass2MOV(hp1) then
|
|
|
Include(OptsToCheck, aoc_ForceNewIteration);
|
|
|
+ Exclude(OptsToCheck, aoc_MovlMovq2MovlMovl);
|
|
|
|
|
|
{ Reset the tracking to the current instruction }
|
|
|
RestoreUsedRegs(TempTracking);
|
|
|
@@ -16655,8 +16673,10 @@ unit aoptx86;
|
|
|
UpdateUsedRegs(UsedRegs, tai(p.Next));
|
|
|
|
|
|
hp2 := hp1;
|
|
|
+ Include(OptsToCheck, aoc_MovlMovq2MovlMovl);
|
|
|
if OptPass2MOV(hp1) then
|
|
|
Include(OptsToCheck, aoc_ForceNewIteration);
|
|
|
+ Exclude(OptsToCheck, aoc_MovlMovq2MovlMovl);
|
|
|
|
|
|
{ Reset the tracking to the current instruction }
|
|
|
RestoreUsedRegs(TempTracking);
|
J. Gareth "Curious Kit" Moreton