search

StanfordPL / stoke

STOKE: A stochastic superoptimizer and program synthesizer
http://stoke.stanford.edu
Apache License 2.0
735 stars 75 forks source link

divb %spl and idivb %spl consistenly fail fuzzer #764

Open stefanheule opened 8 years ago

stefanheule commented 8 years ago

I wonder if this is because it's a div-by-0? It happens with no other register, and %spl happens to be zero. The output has status 0, which also seems wrong? Is the sandbox not handling div-by-0?

High because our tests fail.

bchurchill commented 8 years ago

Do you think the circuit is wrong or the sandbox? Because it's spl I suspect sandbox bug.

bchurchill commented 8 years ago

Also, is there memory involved in the division, or just a register-register or immediate-register divide?

bchurchill commented 8 years ago

OMG the sandbox code special cases handing division when rsp is an argument... this is insane.

bchurchill commented 8 years ago

So, we decided to disable this on the fuzz tester. Putting it as low priority but we should circle around on this eventually. Even if it's not important, it seems like an interesting thing to debug.

stefanheule commented 8 years ago

I fully agree.