------- Comment #9 from wilson at codesourcery dot com 2010-03-17 23:25 ------- Subject: Re: [ia64] Inappropriate address spills
On Wed, 2010-03-17 at 22:09 +0000, sje at cup dot hp dot com wrote: > I tried the patch and didn't have any problem bootstrapping and I didn't see > any regressions. It also fixed my small test case, but when I went back and > tried some of the other tests from PR 28490 I still saw some of the bad gprel > usages. Here is a slightly more cutdown testcase from 28490 that still fails > with the patch applied and when compiling with -O2. I already mentioned in my previous message that the patch does not eliminate all instances of the gprel usages (constant-pool references). It just eliminates most of them. If you want to eliminate all of them, then you need to get LEGITIMATE_PIC_OPERAND_P working, which in turn will require setting flag_pic by default. which in turn will require verifying that this doesn't break anything. That is probably more work than I have time for. And by "fails", you mean that the compiler is still emitting undesirable code in some cases. The code isn't incorrect, just non-optimal. And we get less non-optimal code with the patch, so it seems to be better than nothing. Unless you want to try to write the better LEGITIMATE_PIC_OPERAND_P solution. Jim -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42040
