<div dir="ltr">On 7 January 2013 22:15, Tim Northover <span dir="ltr"><<a href="mailto:t.p.northover@gmail.com" target="_blank">t.p.northover@gmail.com</a>></span> wrote:<br><div class="gmail_extra"><div class="gmail_quote">
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">I personally favour: build an AArch64-native Clang on x86; dispatch an<br>
llvm-regression run and a clang-regression run to models (each takes<br>
~5 hours).<br></blockquote><div><br></div><div>This is more or less what we have for ARM 32-bits anyway. A nightly check-all would already be of great help, though it'd be hard to count it as "don't break" target with that turnaround.</div>
<div><br></div><div><br></div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Alternatives might involve the llvm testsuite (> 1 day, probably not<br>
useful). Or SPEC runs (similar times, but even more difficult for<br>
others to investigate when it breaks; possible licensing issues<br>
anyway).<br></blockquote><div><br></div><div>With LNT you can run the whole simple tests or individual tests, so if you need a more fine grained control over what gets tested, you can.</div><div><br></div><div>
I'd think that time is more important than coverage at this stage, so if you can set up a number of tests with enough coverage that would only take a few hours, you could go for that. When you get a faster model, or an FPGA materialises you increase the number of tests, and so on, up to the full test-suite.</div>
<div><br></div><div style>At least that would keep the "don't break" status because you could fix and test quickly afterwards...</div><div style><br></div><div style>cheers,</div><div style>--renato</div><div>
<br></div></div></div></div>