On 5/28/07, Mark Mitchell <[EMAIL PROTECTED]> wrote:
Andrew -- Would you please give me a summary of the status? Are there regressions on major platforms?
The summary is that powerpc-darwin, powerpc64-linux-gnu, spu-elf, i686-linux-gnu bootstraps and tests with two regression (explained below). sh-linux-gnu and x86_64-linux-gnu are known to bootstrap (but I have not seen the testresults to know if they have any other regressions on those targets), There are two regressions on all targets which support vectorizer of integers. These two regressions in the vectorizer testsuite which are caused by forwprop finally doing a better job and transforming &a->b[0] p+ (i + 1)*4 and data-ref not understanding this change; I am testing the fix for one of the two regressions currently for the mainline (I was able to make a testcase which shows the problem for the mainline for both issues, see PR 32075 and PR 31996). For the second one I am thinking about just adding an extra forwprop to fix up the issue and then let my current patch for data-ref fix that issue. Richard Guenther tested it on SPEC and the other benchmark for me and found that gzip was regressing a bit but I have not looked into that issue yet as I don't have access to x86_64 to figure out what is going on. I cleaned up the code today so it basically ready to be merged, some (most?) of the target headers still need to be fixed for the change. The list of targets which need to be changed is: alpha ia64 mips pa s390 sparc stormy16 xtensa I don't have access to any of those targets (and I have not built a sim based compiler yet). I would like help converting those last 8, the patch should mirror what was done for rs6000, spu or sh. Thanks, Andrew Pinski
