Ok, the updated changes look good, although you might want to add a comment in the Makefile explaining that the code in this file is sensitive to reassociation. Please go ahead and commit. Thanks,
-David Coakley / AMD Open Source Compiler Engineering On Fri, Mar 4, 2011 at 12:36 AM, Wu Yongchong <wuyongch...@gmail.com> wrote: > Hi, David > > Thanks for your advice. > > 1. x8664 has the same issue . but in x8664 -O3 comes with -OPT:ro=1 > while in IA64, -O3 comes with -OPT:ro=2, so we can not see the > problem in x8664 > > 2. I check that, in debug mode the library also build with -O3. > > Here is the new patch, remove the target specific code. > > Index: osprey/libu/Makefile.gbase > =================================================================== > --- osprey/libu/Makefile.gbase (revision 3495) > +++ osprey/libu/Makefile.gbase (working copy) > @@ -358,3 +358,10 @@ > ./massage > numbers.h > endif > > +ifeq ($(BUILD_COMPILER), OSP) > +defgu2s.o defgu2s.pic.o defgu2s.pg.o : CFLAGS += -OPT:ro=1 > +endif > + > +ifeq ($(BUILD_COMPILER), SELF) > +defgu2s.o defgu2s.pic.o defgu2s.pg.o : CFLAGS += -OPT:ro=1 > +endif > > On Fri, Mar 4, 2011 at 7:03 AM, David Coakley <dcoak...@gmail.com> wrote: >> Hi Yongchong, >> >> You should be able to use target-specific variable assignment to avoid >> rewriting the rules: >> >> defgu2s: CFLAGS += -OPT:ro=1 >> >> See: >> >> http://www.gnu.org/software/make/manual/html_node/Target_002dspecific.html#Target_002dspecific >> >> Two things I am unclear on: >> 1. The change is specific to IA64. Wouldn't this be a problem for >> other targets, too? >> 2. What if you are building the library in debug mode? Should the >> flag change also be conditional on -O3 existing in CFLAGS? >> >> -David Coakley / AMD Open Source Compiler Engineering >> >> On Thu, Mar 3, 2011 at 12:34 AM, Wu Yongchong <wuyongch...@gmail.com> wrote: >>> can a gatekeeper help review this patch >>> >>> This is a round off issue in IA64 system relative open64 library, the >>> libffio.a. >>> >>> we use opencc -O3 to build libffio.a, the round off option in -O3 is >>> -OPT:ro=2, it allow reassociation transforma\ >>> tions. >>> >>> in line 531 of file osprey/libu/numconv/mpp/defgu2s.c >>> >>> expression >>> pre_result = a*x + (a*y + x*b +(b*y >>> + (table_high*exact_low_number + >>> exact_high_number*table_low + >>> table_low*exact_low_number))); >>> >>> the compiler will re-associated it to >>> >>> pre_result = (a*x + a*y + x*b + b*y) + >>> (table_high*exact_low_number + >>> exact_high_number*table_low + >>> table_low*exact_low_number); >>> >>> so do the precision of pre_result, this difference may affect the >>> result of 200.sixtrack and them get outp\ >>> ut difference error. >>> >>> Here is my patch, I just add -OPT:ro=1 to make file defgu2s.c when >>> building library libffio.a >>> >>> =================================================================== >>> --- osprey/libu/Makefile.gbase (revision 3495) >>> +++ osprey/libu/Makefile.gbase (working copy) >>> @@ -358,3 +358,15 @@ >>> ./massage > numbers.h >>> endif >>> >>> +ifeq ($(BUILD_ARCH), IA64) >>> + ifeq ($(BUILD_COMPILER), OSP) >>> +defgu2s.o : defgu2s.c >>> + $(cc) -c $(CPPFLAGS) $(CFLAGS) -OPT:ro=1 $< -o $@ >>> + >>> +defgu2s.pic.o : defgu2s.c >>> + $(cc) $(pic_flag) -c $(CPPFLAGS) $(CFLAGS) -OPT:ro=1 $< -o $@ >>> + >>> +defgu2s.pg.o : defgu2s.c >>> + $(cc) $(prof_flag) -c $(CPPFLAGS) $(CFLAGS) -OPT:ro=1 $< -o $@ >>> + endif >>> +endif >>> \ No newline at end of file >>> >>> -- >>> yongchong >>> >>> ------------------------------------------------------------------------------ >>> Free Software Download: Index, Search & Analyze Logs and other IT data in >>> Real-Time with Splunk. Collect, index and harness all the fast moving IT >>> data >>> generated by your applications, servers and devices whether physical, >>> virtual >>> or in the cloud. Deliver compliance at lower cost and gain new business >>> insights. http://p.sf.net/sfu/splunk-dev2dev >>> _______________________________________________ >>> Open64-devel mailing list >>> Open64-devel@lists.sourceforge.net >>> https://lists.sourceforge.net/lists/listinfo/open64-devel >>> >> > > > > -- > yongchong > ------------------------------------------------------------------------------ What You Don't Know About Data Connectivity CAN Hurt You This paper provides an overview of data connectivity, details its effect on application quality, and explores various alternative solutions. http://p.sf.net/sfu/progress-d2d _______________________________________________ Open64-devel mailing list Open64-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/open64-devel
