mirror of
https://gitlab.com/freepascal.org/fpc/source.git
synced 2025-08-17 17:49:13 +02:00
* fixed optimiser so it compiles again
* fixed several bugs which were in there already for a long time, but
which only popped up now :) -O2/-O3 will now optimise less than in
the past (and correctly so), but -O2u/-O3u will optimise a bit more
* some more small improvements for -O3 are still possible
This commit is contained in:
Jonas Maebe
parent
52980d5488
commit
aad37ce53c
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
@ -21,23 +21,23 @@
|
||||
|
||||
****************************************************************************
|
||||
}
|
||||
Unit rrOpt386;
|
||||
unit rropt386;
|
||||
|
||||
{$i fpcdefs.inc}
|
||||
|
||||
Interface
|
||||
interface
|
||||
|
||||
Uses aasmbase,aasmtai,aasmcpu;
|
||||
uses aasmbase,aasmtai,aasmcpu;
|
||||
|
||||
procedure doRenaming(asml: TAAsmoutput; first, last: Tai);
|
||||
procedure doRenaming(asml: taasmoutput; first, last: tai);
|
||||
|
||||
Implementation
|
||||
implementation
|
||||
|
||||
Uses
|
||||
uses
|
||||
{$ifdef replaceregdebug}cutils,{$endif}
|
||||
verbose,globals,cpubase,daopt386,csopt386,cginfo,rgobj;
|
||||
verbose,globals,cpubase,daopt386,csopt386,rgobj, cgbase, cgobj;
|
||||
|
||||
function canBeFirstSwitch(p: Taicpu; reg: tregister): boolean;
|
||||
function canBeFirstSwitch(p: taicpu; supreg: tsuperregister): boolean;
|
||||
{ checks whether an operation on reg can be switched to another reg without an }
|
||||
{ additional mov, e.g. "addl $4,%reg1" can be changed to "leal 4(%reg1),%reg2" }
|
||||
begin
|
||||
@ -45,64 +45,56 @@ begin
|
||||
case p.opcode of
|
||||
A_MOV,A_MOVZX,A_MOVSX,A_LEA:
|
||||
canBeFirstSwitch :=
|
||||
(p.oper[1].typ = top_reg) and
|
||||
(reg32(p.oper[1].reg).enum = reg.enum);
|
||||
(p.oper[1]^.typ = top_reg) and
|
||||
(getsupreg(p.oper[1]^.reg) = supreg);
|
||||
A_IMUL:
|
||||
canBeFirstSwitch :=
|
||||
(p.ops >= 2) and
|
||||
(reg32(p.oper[p.ops-1].reg).enum = reg.enum) and
|
||||
(p.oper[0].typ <> top_ref) and
|
||||
(getsupreg(p.oper[p.ops-1]^.reg) = supreg) and
|
||||
(p.oper[0]^.typ <> top_ref) and
|
||||
(not pTaiprop(p.optinfo)^.FlagsUsed);
|
||||
A_INC,A_DEC,A_SUB,A_ADD:
|
||||
A_INC,A_DEC:
|
||||
canBeFirstSwitch :=
|
||||
(p.oper[1].typ = top_reg) and
|
||||
(p.oper[0]^.typ = top_reg) and
|
||||
(p.opsize = S_L) and
|
||||
(reg32(p.oper[1].reg).enum = reg.enum) and
|
||||
(p.oper[0].typ <> top_ref) and
|
||||
(not pTaiprop(p.optinfo)^.FlagsUsed);
|
||||
A_SUB,A_ADD:
|
||||
canBeFirstSwitch :=
|
||||
(p.oper[1]^.typ = top_reg) and
|
||||
(p.opsize = S_L) and
|
||||
(getsupreg(p.oper[1]^.reg) = supreg) and
|
||||
(p.oper[0]^.typ <> top_ref) and
|
||||
((p.opcode <> A_SUB) or
|
||||
(p.oper[0].typ = top_const)) and
|
||||
(p.oper[0]^.typ = top_const)) and
|
||||
(not pTaiprop(p.optinfo)^.FlagsUsed);
|
||||
A_SHL:
|
||||
canBeFirstSwitch :=
|
||||
(p.opsize = S_L) and
|
||||
(p.oper[1].typ = top_reg) and
|
||||
(p.oper[1].reg.enum = reg.enum) and
|
||||
(p.oper[0].typ = top_const) and
|
||||
(p.oper[0].val in [1,2,3]) and
|
||||
(p.oper[1]^.typ = top_reg) and
|
||||
(getsupreg(p.oper[1]^.reg) = supreg) and
|
||||
(p.oper[0]^.typ = top_const) and
|
||||
(p.oper[0]^.val in [1,2,3]) and
|
||||
(not pTaiprop(p.optinfo)^.FlagsUsed);
|
||||
end;
|
||||
end;
|
||||
|
||||
procedure switchReg(var reg: tregister; reg1, reg2: tregister);
|
||||
|
||||
procedure switchReg(var reg: tregister; reg1, reg2: tsuperregister);
|
||||
var
|
||||
supreg: tsuperregister;
|
||||
begin
|
||||
if reg1.enum>lastreg then
|
||||
internalerror(2003010801);
|
||||
if reg2.enum>lastreg then
|
||||
internalerror(2003010801);
|
||||
if reg.enum>lastreg then
|
||||
internalerror(2003010801);
|
||||
if reg.enum = reg1.enum then
|
||||
reg := reg2
|
||||
else if reg.enum = reg2.enum then
|
||||
reg := reg1
|
||||
else if (reg.enum in regset8bit) then
|
||||
begin
|
||||
if (reg.enum = changeregsize(reg1,S_B).enum) then
|
||||
reg := changeregsize(reg2,S_B)
|
||||
else if reg.enum = changeregsize(reg2,S_B).enum then
|
||||
reg := changeregsize(reg1,S_B);
|
||||
end
|
||||
else if (reg.enum in regset16bit) then
|
||||
begin
|
||||
if reg.enum = changeregsize(reg1,S_W).enum then
|
||||
reg := changeregsize(reg2,S_W)
|
||||
else if reg.enum = changeregsize(reg2,S_W).enum then
|
||||
reg := changeregsize(reg1,S_W);
|
||||
end;
|
||||
if (reg = NR_NO) or
|
||||
(getregtype(reg) <> R_INTREGISTER) then
|
||||
exit;
|
||||
supreg := getsupreg(reg);
|
||||
if (supreg = reg1) then
|
||||
setsupreg(reg,reg2)
|
||||
else if (supreg = reg2) then
|
||||
setsupreg(reg,reg1);
|
||||
end;
|
||||
|
||||
|
||||
procedure switchOp(var op: toper; reg1, reg2: tregister);
|
||||
procedure switchOp(var op: toper; reg1, reg2: tsuperregister);
|
||||
begin
|
||||
case op.typ of
|
||||
top_reg:
|
||||
@ -115,35 +107,36 @@ begin
|
||||
end;
|
||||
end;
|
||||
|
||||
procedure doSwitchReg(hp: Taicpu; reg1,reg2: tregister);
|
||||
|
||||
procedure doSwitchReg(hp: taicpu; reg1,reg2: tsuperregister);
|
||||
var
|
||||
opCount: longint;
|
||||
begin
|
||||
for opCount := 0 to hp.ops-1 do
|
||||
switchOp(hp.oper[opCount],reg1,reg2);
|
||||
switchOp(hp.oper[opCount]^,reg1,reg2);
|
||||
end;
|
||||
|
||||
|
||||
procedure doFirstSwitch(p: Taicpu; reg1, reg2: tregister);
|
||||
procedure doFirstSwitch(p: taicpu; reg1, reg2: tsuperregister);
|
||||
var
|
||||
tmpRef: treference;
|
||||
begin
|
||||
case p.opcode of
|
||||
A_MOV,A_MOVZX,A_MOVSX,A_LEA:
|
||||
begin
|
||||
changeOp(p.oper[1],reg1,reg2);
|
||||
changeOp(p.oper[0],reg2,reg1);
|
||||
changeOp(p.oper[1]^,reg1,reg2);
|
||||
changeOp(p.oper[0]^,reg2,reg1);
|
||||
end;
|
||||
A_IMUL:
|
||||
begin
|
||||
p.ops := 3;
|
||||
p.loadreg(2,p.oper[1].reg);
|
||||
changeOp(p.oper[2],reg1,reg2);
|
||||
p.loadreg(2,p.oper[1]^.reg);
|
||||
changeOp(p.oper[2]^,reg1,reg2);
|
||||
end;
|
||||
A_INC,A_DEC:
|
||||
begin
|
||||
reference_reset(tmpref);
|
||||
tmpref.base := reg1;
|
||||
tmpref.base := newreg(R_INTREGISTER,reg1,R_SUBWHOLE);
|
||||
case p.opcode of
|
||||
A_INC:
|
||||
tmpref.offset := 1;
|
||||
@ -152,51 +145,51 @@ begin
|
||||
end;
|
||||
p.ops := 2;
|
||||
p.opcode := A_LEA;
|
||||
p.loadreg(1,reg2);
|
||||
p.loadreg(1,newreg(R_INTREGISTER,reg2,R_SUBWHOLE));
|
||||
p.loadref(0,tmpref);
|
||||
end;
|
||||
A_SUB,A_ADD:
|
||||
begin
|
||||
reference_reset(tmpref);
|
||||
tmpref.base := reg1;
|
||||
case p.oper[0].typ of
|
||||
tmpref.base := newreg(R_INTREGISTER,reg1,R_SUBWHOLE);
|
||||
case p.oper[0]^.typ of
|
||||
top_const:
|
||||
begin
|
||||
tmpref.offset := p.oper[0].val;
|
||||
tmpref.offset := longint(p.oper[0]^.val);
|
||||
if p.opcode = A_SUB then
|
||||
tmpref.offset := - tmpRef.offset;
|
||||
end;
|
||||
top_symbol:
|
||||
tmpref.symbol := p.oper[0].sym;
|
||||
tmpref.symbol := p.oper[0]^.sym;
|
||||
top_reg:
|
||||
begin
|
||||
tmpref.index := p.oper[0].reg;
|
||||
tmpref.index := p.oper[0]^.reg;
|
||||
tmpref.scalefactor := 1;
|
||||
end;
|
||||
else internalerror(200010031);
|
||||
end;
|
||||
p.opcode := A_LEA;
|
||||
p.loadref(0,tmpref);
|
||||
p.loadreg(1,reg2);
|
||||
p.loadreg(1,newreg(R_INTREGISTER,reg2,R_SUBWHOLE));
|
||||
end;
|
||||
A_SHL:
|
||||
begin
|
||||
reference_reset(tmpref);
|
||||
tmpref.index := reg1;
|
||||
tmpref.scalefactor := 1 shl p.oper[0].val;
|
||||
tmpref.base := newreg(R_INTREGISTER,reg1,R_SUBWHOLE);
|
||||
tmpref.scalefactor := 1 shl p.oper[0]^.val;
|
||||
p.opcode := A_LEA;
|
||||
p.loadref(0,tmpref);
|
||||
p.loadreg(1,reg2);
|
||||
p.loadreg(1,newreg(R_INTREGISTER,reg2,R_SUBWHOLE));
|
||||
end;
|
||||
else internalerror(200010032);
|
||||
end;
|
||||
end;
|
||||
|
||||
|
||||
function switchRegs(asml: TAAsmoutput; reg1, reg2: tregister; start: Tai): Boolean;
|
||||
function switchRegs(asml: taasmoutput; reg1, reg2: tsuperregister; start: tai): Boolean;
|
||||
{ change movl %reg1,%reg2 ... bla ... to ... bla with reg1 and reg2 switched }
|
||||
var
|
||||
endP, hp, lastreg1,lastreg2: Tai;
|
||||
endP, hp, lastreg1,lastreg2: tai;
|
||||
switchDone, switchLast, tmpResult, sequenceEnd, reg1Modified, reg2Modified: boolean;
|
||||
reg1StillUsed, reg2StillUsed, isInstruction: boolean;
|
||||
begin
|
||||
@ -217,8 +210,8 @@ begin
|
||||
{ "mov %oldReg,%newReg; <operations on %newReg>; mov %newReg, }
|
||||
{ %oldReg" to "<operations on %oldReg>" }
|
||||
switchLast := storeBack(endP,reg1,reg2);
|
||||
reg1StillUsed := reg1.enum in pTaiprop(endp.optinfo)^.usedregs;
|
||||
reg2StillUsed := reg2.enum in pTaiprop(endp.optinfo)^.usedregs;
|
||||
reg1StillUsed := reg1 in pTaiprop(endp.optinfo)^.usedregs;
|
||||
reg2StillUsed := reg2 in pTaiprop(endp.optinfo)^.usedregs;
|
||||
isInstruction := endp.typ = ait_instruction;
|
||||
sequenceEnd :=
|
||||
switchLast or
|
||||
@ -227,13 +220,13 @@ begin
|
||||
(not reg1StillUsed and not reg2StillUsed) or
|
||||
{ no support for (i)div, mul and imul with hardcoded operands }
|
||||
(((not isInstruction) or
|
||||
noHardCodedRegs(Taicpu(endP),reg1,reg2)) and
|
||||
noHardCodedRegs(taicpu(endP),reg1,reg2)) and
|
||||
(not reg1StillUsed or
|
||||
(isInstruction and findRegDealloc(reg1,endP) and
|
||||
regLoadedWithNewValue(reg1,false,Taicpu(endP)))) and
|
||||
regLoadedWithNewValue(reg1,false,taicpu(endP)))) and
|
||||
(not reg2StillUsed or
|
||||
(isInstruction and findRegDealloc(reg2,endP) and
|
||||
regLoadedWithNewValue(reg2,false,Taicpu(endP)))));
|
||||
regLoadedWithNewValue(reg2,false,taicpu(endP)))));
|
||||
|
||||
{ we can't switch reg1 and reg2 in something like }
|
||||
{ movl %reg1,%reg2 }
|
||||
@ -249,7 +242,7 @@ begin
|
||||
if not reg1Modified then
|
||||
begin
|
||||
reg1Modified := regModifiedByInstruction(reg1,endP);
|
||||
if reg1Modified and not canBeFirstSwitch(Taicpu(endP),reg1) then
|
||||
if reg1Modified and not canBeFirstSwitch(taicpu(endP),reg1) then
|
||||
begin
|
||||
tmpResult := false;
|
||||
break;
|
||||
@ -264,9 +257,9 @@ begin
|
||||
tmpResult :=
|
||||
(endp.typ <> ait_label) and
|
||||
((not isInstruction) or
|
||||
(NoHardCodedRegs(Taicpu(endP),reg1,reg2) and
|
||||
RegSizesOk(reg1,reg2,Taicpu(endP)) and
|
||||
(Taicpu(endp).opcode <> A_JMP)));
|
||||
(NoHardCodedRegs(taicpu(endP),reg1,reg2) and
|
||||
RegSizesOk(reg1,reg2,taicpu(endP)) and
|
||||
(taicpu(endp).opcode <> A_JMP)));
|
||||
end;
|
||||
end;
|
||||
|
||||
@ -289,33 +282,35 @@ begin
|
||||
reg1Modified := regModifiedByInstruction(reg1,hp);
|
||||
if reg1Modified then
|
||||
begin
|
||||
doFirstSwitch(Taicpu(hp),reg1,reg2);
|
||||
doFirstSwitch(taicpu(hp),reg1,reg2);
|
||||
switchDone := true;
|
||||
end;
|
||||
end;
|
||||
if not switchDone then
|
||||
if reg1Modified then
|
||||
doSwitchReg(Taicpu(hp),reg1,reg2)
|
||||
doSwitchReg(taicpu(hp),reg1,reg2)
|
||||
else
|
||||
doReplaceReg(Taicpu(hp),reg2,reg1);
|
||||
doReplaceReg(taicpu(hp),reg2,reg1);
|
||||
end;
|
||||
if regininstruction(reg1.enum,hp) then
|
||||
if regininstruction(reg1,hp) then
|
||||
lastreg1 := hp;
|
||||
if regininstruction(reg2.enum,hp) then
|
||||
if regininstruction(reg2,hp) then
|
||||
lastreg2 := hp;
|
||||
getNextInstruction(hp,hp);
|
||||
end;
|
||||
if switchLast then
|
||||
doSwitchReg(Taicpu(hp),reg1,reg2)
|
||||
else getLastInstruction(hp,hp);
|
||||
allocRegBetween(asmL,reg1,start,lastreg1);
|
||||
allocRegBetween(asmL,reg2,start,lastreg2);
|
||||
doSwitchReg(taicpu(hp),reg1,reg2)
|
||||
else
|
||||
getLastInstruction(hp,hp);
|
||||
allocRegBetween(asmL,newreg(R_INTREGISTER,reg1,R_SUBWHOLE),start,lastreg1);
|
||||
allocRegBetween(asmL,newreg(R_INTREGISTER,reg2,R_SUBWHOLE),start,lastreg2);
|
||||
end;
|
||||
end;
|
||||
|
||||
procedure doRenaming(asml: TAAsmoutput; first, last: Tai);
|
||||
|
||||
procedure doRenaming(asml: taasmoutput; first, last: tai);
|
||||
var
|
||||
p: Tai;
|
||||
p: tai;
|
||||
begin
|
||||
p := First;
|
||||
SkipHead(p);
|
||||
@ -324,25 +319,18 @@ begin
|
||||
case p.typ of
|
||||
ait_instruction:
|
||||
begin
|
||||
case Taicpu(p).opcode of
|
||||
case taicpu(p).opcode of
|
||||
A_MOV:
|
||||
begin
|
||||
if not(pTaiprop(p.optinfo)^.canBeRemoved) and
|
||||
(Taicpu(p).oper[0].typ = top_reg) and
|
||||
(Taicpu(p).oper[1].typ = top_reg) and
|
||||
(Taicpu(p).opsize = S_L) and
|
||||
{ (Taicpu(p).oper[0].reg.enum in (rg.usableregsint+[R_EDI])) and
|
||||
(Taicpu(p).oper[1].reg.enum in (rg.usableregsint+[R_EDI])) then}
|
||||
(Taicpu(p).oper[0].reg.enum in ([R_EDI])) and
|
||||
(Taicpu(p).oper[1].reg.enum in ([R_EDI])) then
|
||||
if switchRegs(asml,Taicpu(p).oper[0].reg,
|
||||
Taicpu(p).oper[1].reg,p) then
|
||||
(taicpu(p).oper[0]^.typ = top_reg) and
|
||||
(taicpu(p).oper[1]^.typ = top_reg) and
|
||||
(taicpu(p).opsize = S_L) and
|
||||
(getsupreg(taicpu(p).oper[0]^.reg) in ([RS_EAX,RS_EBX,RS_ECX,RS_EDX,RS_ESI,RS_EDI])) and
|
||||
(getsupreg(taicpu(p).oper[1]^.reg) in ([RS_EAX,RS_EBX,RS_ECX,RS_EDX,RS_ESI,RS_EDI])) then
|
||||
if switchRegs(asml,getsupreg(taicpu(p).oper[0]^.reg),
|
||||
getsupreg(taicpu(p).oper[1]^.reg),p) then
|
||||
begin
|
||||
{ getnextinstruction(p,hp);
|
||||
asmL^.remove(p);
|
||||
dispose(p,done);
|
||||
p := hp;
|
||||
continue }
|
||||
pTaiprop(p.optinfo)^.canBeRemoved := true;
|
||||
end;
|
||||
end;
|
||||
@ -358,7 +346,14 @@ End.
|
||||
|
||||
{
|
||||
$Log$
|
||||
Revision 1.22 2003-06-03 21:09:05 peter
|
||||
Revision 1.23 2003-11-22 00:40:19 jonas
|
||||
* fixed optimiser so it compiles again
|
||||
* fixed several bugs which were in there already for a long time, but
|
||||
which only popped up now :) -O2/-O3 will now optimise less than in
|
||||
the past (and correctly so), but -O2u/-O3u will optimise a bit more
|
||||
* some more small improvements for -O3 are still possible
|
||||
|
||||
Revision 1.22 2003/06/03 21:09:05 peter
|
||||
* internal changeregsize for optimizer
|
||||
* fix with a hack to not remove the first instruction of a block
|
||||
which will leave blockstart pointing to invalid memory
|
||||
|
||||
@ -24,7 +24,6 @@ unit psub;
|
||||
|
||||
{$i fpcdefs.inc}
|
||||
|
||||
{$define NOOPT}
|
||||
|
||||
interface
|
||||
|
||||
@ -752,14 +751,6 @@ implementation
|
||||
if not(cs_no_regalloc in aktglobalswitches) then
|
||||
begin
|
||||
cg.do_register_allocation(aktproccode,headertai);
|
||||
(*
|
||||
{$ifndef NoOpt}
|
||||
if (cs_optimize in aktglobalswitches) and
|
||||
{ do not optimize pure assembler procedures }
|
||||
not(pi_is_assembler in current_procinfo.flags) then
|
||||
optimize(aktproccode);
|
||||
{$endif NoOpt}
|
||||
*)
|
||||
end;
|
||||
|
||||
{ Add save and restore of used registers }
|
||||
@ -776,6 +767,17 @@ implementation
|
||||
aktfilepos:=exitpos;
|
||||
gen_stackfree_code(templist,usesacc,usesacchi);
|
||||
aktproccode.concatlist(templist);
|
||||
|
||||
{$ifndef NoOpt}
|
||||
if not(cs_no_regalloc in aktglobalswitches) then
|
||||
begin
|
||||
if (cs_optimize in aktglobalswitches) and
|
||||
{ do not optimize pure assembler procedures }
|
||||
not(pi_is_assembler in current_procinfo.flags) then
|
||||
optimize(aktproccode);
|
||||
end;
|
||||
{$endif NoOpt}
|
||||
|
||||
{ Add end symbol and debug info }
|
||||
aktfilepos:=exitpos;
|
||||
gen_proc_symbol_end(templist);
|
||||
@ -1312,7 +1314,14 @@ implementation
|
||||
end.
|
||||
{
|
||||
$Log$
|
||||
Revision 1.171 2003-11-10 22:02:52 peter
|
||||
Revision 1.172 2003-11-22 00:40:19 jonas
|
||||
* fixed optimiser so it compiles again
|
||||
* fixed several bugs which were in there already for a long time, but
|
||||
which only popped up now :) -O2/-O3 will now optimise less than in
|
||||
the past (and correctly so), but -O2u/-O3u will optimise a bit more
|
||||
* some more small improvements for -O3 are still possible
|
||||
|
||||
Revision 1.171 2003/11/10 22:02:52 peter
|
||||
* cross unit inlining fixed
|
||||
|
||||
Revision 1.170 2003/11/07 15:58:32 florian
|
||||
|
||||
Loading…
Reference in New Issue
Block a user