On 28/11/18 11:35 +0000, Jonathan Wakely wrote:
On 28/11/18 10:54 +0100, Christophe Lyon wrote:
The reason I'm still using cortex-a9 at QEMU level is that many tests
override -mcpu/-march, and I used that as a compromise.
That means you're not fully testing armv5t, because you're still
linking to a libstdc++.so (and libstd++fs.a) that use the cortex-a9
instruction set. Usually that doesn't matter, because most libstdc++
code doesn't care about which specific instructions it gets compiled
to. That
Oops, sorry, unfinished thought. I was going to say:
That is not true for the shared_ptr code, which depends on whether
atomic instructions are supported. There are also some parts of
<random> that depend on specific instructions being supported (but
some of them do runtime checks for the CPU).
