On 09/09/2012 11:04 PM, Richard Henderson wrote:
As promised the other week. I've cleaned up the patch set and re-based it on Blueswirl's areg0 patch set. For reference, the entire patch set is available atgit://repo.or.cz/qemu/rth.git rth/s390-reorg-3 Testing has mostly been on the gcc testsuite, where the execute test failures are reduced to FAIL: gcc.dg/attr-ifunc-1.c execution test FAIL: gcc.dg/attr-ifunc-3.c execution test FAIL: gcc.dg/attr-ifunc-4.c execution test FAIL: gcc.dg/attr-ifunc-5.c execution test FAIL: gcc.dg/cdce1.c execution test FAIL: gcc.dg/cleanup-11.c execution test FAIL: gcc.dg/cleanup-9.c execution test FAIL: gcc.dg/torture/pr41094.c -O0 execution test FAIL: gcc.dg/torture/tls/tls-test.c * execution test I'll probably start trying out the glibc testsuite next, as that's quite likely to flush out remaining problems with the fp support (both cdce1 and pr41094 are both failures in pow). I've also done some testing with -march={z10,z196}, but I don't have those results handy atm.
So I've had a look at the patch set. Overall it looks ok. I have to admit that I found the old code a lot easier to understand and read. But if you want to keep maintaining it the new way, I'm definitely ok with that as well :).
There's one thing I would really like to see first though: Performance numbers. Could you please try and compare the old code with your new code performance wise? How do small linux-user programs compare? How does a full Linux boot compare?
Alex
