Hi Wilco, Wilco Dijkstra <wilco.dijks...@arm.com> wrote:
>> Testsuite fails are order “a few hundred” mostly seem to be related to >> tree-prof >> and vector tests (plus the anticipated scan-asm stuff, where code-gen will >> have >> changed). I don’t have cycles to analyse the causes right now - but that >> gives >> an idea. > > Are those tests specific to Power? I got 14 failures in total across the full > C and C++ test suites. Note it's easy to update the default options for a > specific test directory if needed. No, the test was on x86_64-darwin16 (m32/m64) and a very rough estimate of the number of fails (but definitely in the hundreds, not tens) - I haven’t tried the same on PPC; the h/w is tied up with gcc-7 backport testing. the change for *-*-darwin* will be that the ABI mandates common items to be indirected, thus defaulting to no-common will have the effect of changing the generated code - and it’s a fairly common pattern in the testsuite to have uninit globals. * I’d expect broken scan-asms (probably not difficult to fix, just tedious). * not sure why it seems to have a particularly adverse effect on the tree-prof stuff. short on cycles in the run up to stage #1 end + gcc-7 release .. but will try to poke at it over the next few days. cheers Iain