Message ID | 20101112040945.GA23026@bromo.med.uc.edu |
---|---|
State | New |
Headers | show |
On 11/12/2010 05:09 AM, Jack Howarth wrote: >> If the configure changes have the effect of deprecating something or >> adjusting the required versions of something, there is a serious problem >> and they should not have been approved: the documentation of prerequisites >> in install.texi does not appear to have been changed and it still >> specifies CLooG-PPL. It is not acceptable to change the prerequisites >> without including the documentation changes in the same commit. That was not mentioned in the discussion of the patch. In fact I suppose Cloog-PPL was simply not tested. Andreas, did you try all possible configurations? Also, if I understood correctly ISL and PPL are different ways to "do the same thing", and they should cause no differences in code generation. I assumed this because the patch didn't require any testsuite adjustment. Is this the case? If so, we only need to document the new possible choice of Cloog libraries. If not, however, as Joseph said we need: 1) documentation of the new prerequisites; 2) a --enable-cloog-ppl option to enable Cloog-PPL tests _and_ disable Cloog-ISL at the same time. In this case, I suggest reverting the patch on trunk. Paolo
On 11/12/2010 09:43 AM, Sven Verdoolaege wrote: > On Fri, Nov 12, 2010 at 09:26:00AM +0100, Paolo Bonzini wrote: >> Also, if I understood correctly ISL and PPL are different ways to >> "do the same thing", and they should cause no differences in code >> generation. I assumed this because the patch didn't require any >> testsuite adjustment. Is this the case? > > Semantically, the results should be the same, but there may > be syntactic differences. Perhaps no such syntactic differences > occur for the gcc testsuite. What does "semantic" and "syntactic" mean? I suppose you mean that the produced code should be correct in any case (of course) but the GCC assembly language output may change? This is what we care about. Paolo
On 11/12/2010 10:08 AM, Sven Verdoolaege wrote: > On Fri, Nov 12, 2010 at 09:59:45AM +0100, Paolo Bonzini wrote: >> On 11/12/2010 09:43 AM, Sven Verdoolaege wrote: >>> On Fri, Nov 12, 2010 at 09:26:00AM +0100, Paolo Bonzini wrote: >>>> Also, if I understood correctly ISL and PPL are different ways to >>>> "do the same thing", and they should cause no differences in code >>>> generation. I assumed this because the patch didn't require any >>>> testsuite adjustment. Is this the case? >>> >>> Semantically, the results should be the same, but there may >>> be syntactic differences. Perhaps no such syntactic differences >>> occur for the gcc testsuite. >> >> What does "semantic" and "syntactic" mean? I suppose you mean that >> the produced code should be correct in any case (of course) but the >> GCC assembly language output may change? > > Exactly. That's bad. Sebastian, please revert the patch. It would also be appreciated to compile SPEC with both backends, and see how many different decisions are taken. Paolo
Index: configure =================================================================== --- configure (revision 166643) +++ configure (working copy) @@ -5730,7 +5731,7 @@ CFLAGS="${CFLAGS} ${clooginc} ${pplinc} ${gmpinc}" CPPFLAGS="${CPPFLAGS} ${_clooglegacyinc} ${_cloogorginc}" - LDFLAGS="${LDFLAGS} ${clooglibs}" + LDFLAGS="${LDFLAGS} ${clooglibs} ${ppllibs}" { $as_echo "$as_me:${as_lineno-$LINENO}: checking for installed CLooG" >&5 $as_echo_n "checking for installed CLooG... " >&6; }