paweld/fpc
1
0
Fork 0
mirror of https://gitlab.com/freepascal.org/fpc/source.git synced 2025-04-21 18:09:30 +02:00
Code Issues Packages Projects Releases Wiki Activity

* x86: LeaLea2Lea now handles cases where the second LEA instruction

Browse Source 
has a second register in the reference.
This commit is contained in:
J. Gareth "Curious Kit" Moreton 2022-11-09 02:45:51 +00:00 committed by FPK
parent 66ff7a928c
commit 931bda5633
1 changed files with 62 additions and 44 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

106
compiler/x86/aoptx86.pas
View File

@ -5646,6 +5646,8 @@ unit aoptx86;
hp1: tai;
ref: Integer;
saveref: treference;
offsetcalc: Int64;
TempReg: TRegister;
Multiple: TCGInt;
Adjacent: Boolean;
begin
@ -5826,49 +5828,47 @@ unit aoptx86;
) then
begin
{ changes
lea (regX,scale), reg1
lea offset(reg1,reg1), reg1
lea offset1(regX,scale), reg1
lea offset2(reg1,reg1), reg2
to
lea offset(regX,scale*2), reg1
lea (offset1*scale*2)+offset2(regX,scale*2), reg2
and
lea (regX,scale1), reg1
lea offset(reg1,scale2), reg1
lea offset1(regX,scale1), reg1
lea offset2(reg1,scale2), reg2
to
lea offset(regX,scale1*scale2), reg1
lea (offset1*scale1*2)+offset2(regX,scale1*scale2), reg2
and
lea offset1(regX,scale1), reg1
lea offset2(reg3,reg1,scale2), reg2
to
lea (offset1*scale*2)+offset2(reg3,regX,scale1*scale2), reg2
... so long as the final scale does not exceed 8
(Similarly, allow the first instruction to be "lea (regX,regX),reg1")
}
if (taicpu(p).oper[0]^.ref^.base<>NR_STACK_POINTER_REG) and { lea (%rsp,scale),reg is not a valid encoding }
(taicpu(p).oper[0]^.ref^.offset = 0) and
(taicpu(hp1).oper[0]^.ref^.index = taicpu(p).oper[1]^.reg) and
(
(
(taicpu(p).oper[0]^.ref^.base = NR_NO)
) or (
(taicpu(p).oper[0]^.ref^.scalefactor <= 1) and
(
(taicpu(p).oper[0]^.ref^.base = taicpu(p).oper[0]^.ref^.index) and
(
{ RegUsedBetween always returns False if p and hp1 are adjacent }
Adjacent or
not(RegUsedBetween(taicpu(p).oper[0]^.ref^.index, p, hp1))
)
)
)
(taicpu(p).oper[0]^.ref^.base <> taicpu(p).oper[0]^.ref^.index) or
(taicpu(p).oper[0]^.ref^.scalefactor <= 1)
) and (
(
{ lea (reg1,scale2), reg1 variant }
(taicpu(hp1).oper[0]^.ref^.base = NR_NO) and
{ lea (reg1,scale2), reg2 variant }
(taicpu(hp1).oper[0]^.ref^.base <> taicpu(p).oper[1]^.reg) and
(
Adjacent or
not RegModifiedBetween(taicpu(hp1).oper[0]^.ref^.base, p, hp1)
) and
(
(
(taicpu(p).oper[0]^.ref^.base = NR_NO) and
(taicpu(hp1).oper[0]^.ref^.scalefactor * taicpu(p).oper[0]^.ref^.scalefactor <= 8)
) or (
{ lea (regX,regX), reg1 variant }
(taicpu(p).oper[0]^.ref^.base <> NR_NO) and
(taicpu(p).oper[0]^.ref^.base = taicpu(p).oper[0]^.ref^.index) and
(taicpu(hp1).oper[0]^.ref^.scalefactor <= 4)
)
)
@ -5879,8 +5879,6 @@ unit aoptx86;
)
) then
begin
DebugMsg(SPeepholeOptimization + 'LeaLea2Lea 2 done',p);
{ Make everything homogeneous to make calculations easier }
if (taicpu(p).oper[0]^.ref^.base <> NR_NO) then
begin
@ -5893,32 +5891,52 @@ unit aoptx86;
taicpu(p).oper[0]^.ref^.base := NR_NO;
end;
if (taicpu(hp1).oper[0]^.ref^.base = NR_NO) then
begin
{ Just to prevent miscalculations }
if (taicpu(hp1).oper[0]^.ref^.scalefactor = 0) then
taicpu(hp1).oper[0]^.ref^.scalefactor := taicpu(p).oper[0]^.ref^.scalefactor
else
taicpu(hp1).oper[0]^.ref^.scalefactor := taicpu(hp1).oper[0]^.ref^.scalefactor * taicpu(p).oper[0]^.ref^.scalefactor;
end
else
begin
taicpu(hp1).oper[0]^.ref^.base := NR_NO;
taicpu(hp1).oper[0]^.ref^.scalefactor := taicpu(p).oper[0]^.ref^.scalefactor * 2;
end;
{ Make sure the offset doesn't go out of range (use 64-bit arithmetic)}
offsetcalc := taicpu(hp1).oper[0]^.ref^.offset;
Inc(offsetcalc, Int64(taicpu(p).oper[0]^.ref^.offset) * max(taicpu(hp1).oper[0]^.ref^.scalefactor, 1));
taicpu(hp1).oper[0]^.ref^.index := taicpu(p).oper[0]^.ref^.index;
RemoveCurrentP(p);
result:=true;
exit;
end
if (offsetcalc <= $7FFFFFFF) and (offsetcalc >= -2147483648) then
begin
DebugMsg(SPeepholeOptimization + 'LeaLea2Lea 2 done',p);
if (taicpu(hp1).oper[0]^.ref^.base = taicpu(p).oper[1]^.reg) and
(taicpu(hp1).oper[0]^.ref^.index <> taicpu(p).oper[1]^.reg) then
begin
{ Put the register to change in the index register }
TempReg := taicpu(hp1).oper[0]^.ref^.index;
taicpu(hp1).oper[0]^.ref^.index := taicpu(hp1).oper[0]^.ref^.base;
taicpu(hp1).oper[0]^.ref^.base := TempReg;
end;
if (taicpu(hp1).oper[0]^.ref^.base <> taicpu(p).oper[1]^.reg) then
begin
{ Just to prevent miscalculations }
if (taicpu(hp1).oper[0]^.ref^.scalefactor = 0) then
taicpu(hp1).oper[0]^.ref^.scalefactor := taicpu(p).oper[0]^.ref^.scalefactor
else
taicpu(hp1).oper[0]^.ref^.scalefactor := taicpu(hp1).oper[0]^.ref^.scalefactor * max(taicpu(p).oper[0]^.ref^.scalefactor, 1);
end
else
begin
taicpu(hp1).oper[0]^.ref^.base := NR_NO;
taicpu(hp1).oper[0]^.ref^.scalefactor := taicpu(p).oper[0]^.ref^.scalefactor * 2;
end;
if (taicpu(p).oper[0]^.ref^.offset <> 0) then
Inc(taicpu(hp1).oper[0]^.ref^.offset, taicpu(p).oper[0]^.ref^.offset * max(taicpu(p).oper[0]^.ref^.scalefactor, 1));
taicpu(hp1).oper[0]^.ref^.index := taicpu(p).oper[0]^.ref^.index;
RemoveCurrentP(p);
result:=true;
exit;
end;
end;
{ changes
lea offset1(regX), reg1
lea offset2(reg1), reg1
to
lea offset1+offset2(regX), reg1 }
else if
if
(
(taicpu(hp1).oper[0]^.ref^.index = taicpu(p).oper[1]^.reg) and
(taicpu(p).oper[0]^.ref^.index = NR_NO)