[tree-ssa] RFC: Never gimplify ASM_EXPRs

[Richard Henderson]

On Thu, Jun 05, 2003 at 02:49:10PM -0700, Zack Weinberg wrote:
> What stops us from doing that for "m" parameters to asms?

Nothing, and that may be a reasonable first approximation.

Long term we should note that while "=m" does require the
original address, the contents are killed, and thus the old
value is dead.  Similarly, while "m" requires the original
address, the contents are *not* killed.

I wonder if GIMPLE needs better annotations on parameters?
In particular, I recall that Ada can explicitly set in, out
or inout in the source code.


r~