paweld/fpc
1
0
Fork 0
mirror of https://gitlab.com/freepascal.org/fpc/source.git synced 2025-08-14 16:49:07 +02:00
Code Issues Packages Projects Releases Wiki Activity

+ better high level optimization for 64-bit signed division by 2 on 32-bit and

Browse Source 
lower CPUs. Instead of the (sar(temp,sizeof(temp)*8-1) and 1) expression, use
  the equivalent, but simpler (temp shr (sizeof(temp)*8-1))

git-svn-id: trunk@36839 -
This commit is contained in:
nickysn 2017-08-04 13:51:51 +00:00
parent 1aab7e7ee2
commit b9fdca6f49
1 changed files with 21 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
compiler/nmat.pas
View File

@ -493,6 +493,7 @@ implementation
power,shiftval : longint; power,shiftval : longint;
statements : tstatementnode; statements : tstatementnode;
temp,resulttemp : ttempcreatenode; temp,resulttemp : ttempcreatenode;
masknode : tnode;
begin begin
result := nil; result := nil;
{ divide/mod a number by a constant which is a power of 2? } { divide/mod a number by a constant which is a power of 2? }
@ -530,10 +531,16 @@ implementation
left)); left));
left:=nil; left:=nil;
addstatement(statements,cassignmentnode.create(ctemprefnode.create(resulttemp), { masknode is (sar(temp,shiftval) and (tordconstnode(right).value-1))
cinlinenode.create(in_sar_x_y,false, for tordconstnode(right).value=2, masknode is simply (temp shr shiftval)}
ccallparanode.create(cordconstnode.create(power,u8inttype,false), if power=1 then
ccallparanode.create(caddnode.create(addn,ctemprefnode.create(temp), masknode:=
cshlshrnode.create(shrn,
ctemprefnode.create(temp),
cordconstnode.create(shiftval,u8inttype,false)
)
else
masknode:=
caddnode.create(andn, caddnode.create(andn,
cinlinenode.create(in_sar_x_y,false, cinlinenode.create(in_sar_x_y,false,
ccallparanode.create(cordconstnode.create(shiftval,u8inttype,false), ccallparanode.create(cordconstnode.create(shiftval,u8inttype,false),
@ -541,7 +548,13 @@ implementation
), ),
cordconstnode.create(tordconstnode(right).value-1, cordconstnode.create(tordconstnode(right).value-1,
right.resultdef,false) right.resultdef,false)
)),nil );
addstatement(statements,cassignmentnode.create(ctemprefnode.create(resulttemp),
cinlinenode.create(in_sar_x_y,false,
ccallparanode.create(cordconstnode.create(power,u8inttype,false),
ccallparanode.create(caddnode.create(addn,ctemprefnode.create(temp),
masknode),nil
)))) ))))
); );
addstatement(statements,ctempdeletenode.create(temp)); addstatement(statements,ctempdeletenode.create(temp));