On Monday 02 October 2006 12:57, Joseph S. Myers wrote:
> On Mon, 2 Oct 2006, Jan van Dijk wrote:
> > * the C99 and C++ standards say *nothing* about the details of compex
> > multiplication
>
> The C99 standard says that real operands aren't converted to complex, but
> as I note in bug 24581, the compiler doesn't expect PLUS_EXPR and
> MULT_EXPR to have arguments of different types, so the front ends might
> need adapting to handle real * complex and real + complex specially.
Dear Joseph,
My question was a slightly different one. To me it is not clear whether the
standard allows the treatment of (r,0) as r in complex operations. For
example: is it allowed to handle (r,0)*(x,y) as r*(x,y)?
If this *is* the case, the special treatment of real*complex etc. that you
suggested seems unnecessary: we merely have to teach the builtin
multiplication routines to gracefully handle something like r*(Inf,0) by
special handling of multiplications in which 0 is involved. lower_complex
already has this effect (for -O1 and higher).
If this is *not* the case, r*z may need to be handled separately by the front
ends. But note that this implies that lower_complex and friends are
off-specs. The 0*whatever-terms cannot be discarded, as is presently done.
I repeat my previous mail: there does not seem to be consensus about what
standard(s) to implement (Annex G? LIA-x,...), and how to interpret the free
space in those documents. Could you clarify?
> Patches welcome, see <http://gcc.gnu.org/contribute.html>.
Triggered by 1*(Inf,0) = (Inf,NaN), I looked inside the compiler for the first
time, yesterday. I am still leaning what a tree is. So that patch will not be
rolled by me anytime soon, although I am eagerly learning gcc's internals.
Thank you for your encouraging words, nevertheless :-)
Regards, Jan.
