LGTM
I'm not comfortable with having two implementations that are effectively the
same, but I can't think of a better solution. Note that your Clang path is
marginally better than the existing GCC one, since the compiler can decide
how
best to initialise scno.
The GCC problem only applies with -mthumb, so we could use your new version
with
-marm targets if you decide that the complexity is worthwhile. The system
call
will dominate the performance cost here, so it doesn't make a lot of
difference.
What I wanted to do was implement the whole thing in pure assembly. The
problem
is trivial — it just forwards arguments and executes `svc` — but it's
awkward to
express without using compiler-specific tricks (like
`__attribute__((naked))`).
I'm curious about the constant-pool problem, though. Clang will use constant
pools itself, so it must have some way of handling them.
https://codereview.chromium.org/986643004/
--
--
v8-dev mailing list
[email protected]
http://groups.google.com/group/v8-dev
---
You received this message because you are subscribed to the Google Groups "v8-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
