That is indeed the sort of horror we used to implement out DontOptimise(something.cpp) and DontOptimiseMuch(somethingelse.cpp) macros at a now-famously-defunct investment bank. The code outlived the bank... B^>
Rgds Damon On 29 May 2012, at 18:12, Kelly O'Hair wrote: > > You should be able to do something like: > > # Turn off aliasing with GCC for ExtensionSubtables.cpp > ifeq ($(PLATFORM), linux) > CXXFLAGS += $(CXXFLAGS_$(@F)) > CC_NEWER_THAN_43 := \ > $(shell $(EXPR) $(CC_MAJORVER) \> 4 \| \ > \( $(CC_MAJORVER) = 4 \& $(CC_MINORVER) \> 3 \) ) > ifeq ($(CC_NEWER_THAN_43),1) > CXXFLAGS_ExtensionSubtables.o = -fno-strict-aliasing > endif > endif > > -kto > > On May 29, 2012, at 6:50 AM, Phil Race wrote: > >> I have one concern >> >> It sounds like the option is new in gcc 4.4 : >> http://gcc.gnu.org/gcc-4.4/porting_to.html >> >> I am not sure you can assume that since .. >> >> http://hg.openjdk.java.net/jdk8/jdk8/raw-file/tip/README-builds.html#gcc >> " The GNU gcc compiler version should be 4.3 or newer." >> >> And I suggest a push to 2d-dev since its in the 2d code we get >> a warning and the "build" change is in 2D build files, specifically to work >> around that warning .. its really precious little to do with the build >> system. >> >> -phil. >> >> On 5/29/12 6:13 AM, Andrew Hughes wrote: >>> ----- Original Message ----- >>> >>>> In anticipation of the updated fix I created >>>> 7171223 : Building ExtensionSubtables.cpp should use >>>> -fno-strict-aliasing >>>> >>>> -phil. >>>> >>>> >>>>> Ok for jdk8? If so, can I have a bug ID for this please? >>>>> >>>>> Thanks, >>>>> >>> Thanks for the bug ID, Phil. >>> >>> Here's the revised webrev: >>> >>> >>> http://cr.openjdk.java.net/~andrew/no-strict-aliasing/webrev.02/ >>> >>> >>> Sorry for the delay. I found when trying to build the build tree >>> that HotSpot had regressed and become unbuildable: >>> >>> >>> http://mail.openjdk.java.net/pipermail/hotspot-dev/2012-May/005945.html >>> >>> >>> All patched now. Is build the preferred tree to push this too, >>> or would you prefer somewhere else? >>> >>> Thanks, >>> >> >
