paweld/fpc
1
0
Fork 0
mirror of https://gitlab.com/freepascal.org/fpc/source.git synced 2025-08-22 10:09:22 +02:00
Code Issues Packages Projects Releases Wiki Activity

* Strength reduction optimization: - Use a temp for complex loop start values to prevent double evaluation.

Browse Source 
- For slow CPUs perform the optimization for all sizes of array elements.
This commit is contained in:
Yuriy Sydorov 2021-10-08 18:04:03 +03:00
parent 6bf4c7a68d
commit 7388735b11
1 changed files with 10 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
compiler/optloop.pas
View File

@ -350,7 +350,7 @@ unit optloop;
var var
tempnode : ttempcreatenode; tempnode,startvaltemp : ttempcreatenode;
dummy : longint; dummy : longint;
nn : tnode; nn : tnode;
nt : tnodetype; nt : tnodetype;
@ -453,10 +453,13 @@ unit optloop;
{ direct array access? } { direct array access? }
((tvecnode(n).left.nodetype=loadn) or ((tvecnode(n).left.nodetype=loadn) or
{ ... or loop invariant expression? } { ... or loop invariant expression? }
is_loop_invariant(tfornode(arg),tvecnode(n).right)) and is_loop_invariant(tfornode(arg),tvecnode(n).right))
{$if not (defined(cpu16bitalu) or defined(cpu8bitalu))}
{ removing the multiplication is only worth the { removing the multiplication is only worth the
effort if it's not a simple shift } effort if it's not a simple shift }
not(ispowerof2(tcgvecnode(n).get_mul_size,dummy)) then and not(ispowerof2(tcgvecnode(n).get_mul_size,dummy))
{$endif}
then
begin begin
changedforloop:=true; changedforloop:=true;
{ did we use the same expression before already? } { did we use the same expression before already? }
@ -487,6 +490,8 @@ unit optloop;
cordconstnode.create(tcgvecnode(n).get_mul_size,sizeuinttype,false),nil)))); cordconstnode.create(tcgvecnode(n).get_mul_size,sizeuinttype,false),nil))));
addstatement(initcodestatements,tempnode); addstatement(initcodestatements,tempnode);
startvaltemp:=maybereplacewithtemp(tfornode(arg).right,initcode,initcodestatements,tfornode(arg).right.resultdef.size,true);
nn:=caddrnode.create( nn:=caddrnode.create(
cvecnode.create(tvecnode(n).left.getcopy,tfornode(arg).right.getcopy) cvecnode.create(tvecnode(n).left.getcopy,tfornode(arg).right.getcopy)
); );
@ -508,6 +513,8 @@ unit optloop;
n:=ctypeconvnode.create_internal(cderefnode.create(ctemprefnode.create(tempnode)),n.resultdef); n:=ctypeconvnode.create_internal(cderefnode.create(ctemprefnode.create(tempnode)),n.resultdef);
{ ... and add a temp. release node } { ... and add a temp. release node }
if startvaltemp<>nil then
addstatement(deletecodestatements,ctempdeletenode.create(startvaltemp));
addstatement(deletecodestatements,ctempdeletenode.create(tempnode)); addstatement(deletecodestatements,ctempdeletenode.create(tempnode));
end; end;
{ Copy the nf_write,nf_modify flags to the new deref node of the temp. { Copy the nf_write,nf_modify flags to the new deref node of the temp.