paweld/fpc
1
0
Fork 0
mirror of https://gitlab.com/freepascal.org/fpc/source.git synced 2025-10-14 08:46:08 +02:00
Code Issues Packages Projects Releases Wiki Activity

* prefer a loadsize of 4 over 8 bytes for packed loads on 64 bit,

Browse Source 
because unaligned 8 byte loads are much slower (at least on ppc64)
    than unaligned 4 byte loads

git-svn-id: trunk@6744 -
This commit is contained in:
Jonas Maebe 2007-03-06 15:51:41 +00:00
parent 6a7321c318
commit eae0ba86e7
1 changed files with 6 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
compiler/cutils.pas
View File

@ -280,7 +280,12 @@ implementation
3,5,7,9,10,12,16:
result := 2;
{$ifdef cpu64bit}
11,13,14,15,17..26,28,32:
{ performance penalty for unaligned 8 byte access is much }
{ higher than for unaligned 4 byte access, at least on ppc, }
{ so use 4 bytes even in some cases where a value could }
{ always loaded using a single 8 byte load (e.g. in case of }
{ 28 bit values) }
11,13,14,15,17..32:
result := 4;
else
result := 8;