* min/max optimization on x86 can be applied now without fastmath as NaN is handled properly now

git-svn-id: trunk@47769 -
This commit is contained in:
florian 2020-12-13 09:52:45 +00:00
parent b90ec52a0e
commit 6d7d567a25

View File

@ -1597,8 +1597,6 @@ implementation
not(might_have_sideeffects(left)) and
((t1=nil) or tassignmentnode(thenstmnt).left.isequal(tassignmentnode(elsestmnt).left)) and
{$if defined(i386) or defined(x86_64)}
{ for now, limit it to fastmath mode as NaN handling is not implemented properly yet }
(cs_opt_fastmath in current_settings.optimizerswitches) and
{$ifdef i386}
(((current_settings.fputype>=fpu_sse) and is_single(tassignmentnode(thenstmnt).left.resultdef)) or
((current_settings.fputype>=fpu_sse2) and is_double(tassignmentnode(thenstmnt).left.resultdef))