mirror of
https://gitlab.com/freepascal.org/fpc/source.git
synced 2025-04-16 05:59:28 +02:00
* x86: New OR->BTS peephole optimisation
This commit is contained in:
parent
314f632377
commit
827fe2bd83
@ -14027,6 +14027,45 @@ unit aoptx86;
|
||||
hp1,hp2 : tai;
|
||||
begin
|
||||
Result:=false;
|
||||
{ If x is a power of 2 (popcnt = 1), change:
|
||||
or $x, %reg/ref
|
||||
|
||||
To:
|
||||
bts lb(x), %reg/ref
|
||||
}
|
||||
if (taicpu(p).opcode = A_OR) and
|
||||
{$ifndef x86_64}
|
||||
(
|
||||
(cs_opt_size in current_settings.optimizerswitches) or
|
||||
{ BTS takes more than 1 cycle on earlier processors }
|
||||
(current_settings.optimizecputype >= cpu_Pentium2)
|
||||
) and
|
||||
{$endif not x86_64}
|
||||
MatchOpType(taicpu(p), top_const, top_reg) and { "btx $x,mem" is unacceptably slow }
|
||||
(PopCnt(QWord(taicpu(p).oper[0]^.val)) = 1) and
|
||||
{ For sizes less than S_L, the byte size is equal or larger with BT,
|
||||
so don't bother optimising }
|
||||
(taicpu(p).opsize >= S_L) and
|
||||
(
|
||||
{ If the value can bit into an 8-bit signed integer, a smaller
|
||||
instruction can be encded with OR, so don't optimise if it falls
|
||||
within this range }
|
||||
(taicpu(p).oper[0]^.val < -128) or
|
||||
(taicpu(p).oper[0]^.val >= 127)
|
||||
) and
|
||||
(
|
||||
{ Don't optimise if a test instruction follows }
|
||||
not GetNextInstruction(p, hp1) or
|
||||
not MatchInstruction(hp1, A_TEST, [taicpu(p).opsize])
|
||||
) then
|
||||
begin
|
||||
DebugMsg(SPeepholeOptimization + 'Changed OR $0x' + hexstr(taicpu(p).oper[0]^.val, 2) + ' to BTS ' + debug_tostr(BsrQWord(taicpu(p).oper[0]^.val)) + ' to shrink instruction size (Or2Bts)', p);
|
||||
taicpu(p).opcode := A_BTS;
|
||||
taicpu(p).oper[0]^.val := BsrQWord(taicpu(p).oper[0]^.val); { Essentially the base 2 logarithm }
|
||||
Result := True;
|
||||
Exit;
|
||||
end;
|
||||
|
||||
{ If x is a power of 2 (popcnt = 1), change:
|
||||
test $x, %reg/ref
|
||||
je / sete / cmove (or jne / setne)
|
||||
|
Loading…
Reference in New Issue
Block a user