This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [patch,fortran] Support type real for count_rate in system_clock (PR28484)


Tobias Burnus <burnus@net-b.de> wrote on Sat, 28 Oct 2006:
Brooks Moses wrote:
Steve Kargl wrote:
On Sat, Oct 28, 2006 at 12:38:42AM +0200, Tobias Burnus wrote:
You'll need to implement the real(10), real(16), and integer(16)
versions.  If gfortran (on some target) provides any or all of
these types, then we need intrinsic procedures.

Indeed. To clarify that, the Fortran Standard says that the intrinsics can be called with any real or integer kind, and so not supplying all intrinsics for all supported kinds would mean that the compiler isn't standard-conforming.

Should one also implement system_clock(int4, int8, int16) that is: Mixing all different integer kinds for system_clock? Currently, the integer have to be all of the same kind. Doing so gives then in total 3*3*3 = 27 integer and 4*3*3 = 36 real/integer combinations, i.e. 63 combinations!

One could emit the necessary type conversions, i.e. only implement
system_clock (int4, int4, int4)
in the library and then convert the parameters appropriately for all other combinations, e.g. translating
system_clock (int4, int8, int16)
as
int4 t1 = int8
int4 t2 = int16
system_clock (int4, t1, t2)
int8 = t1
int16 = t2
(I don't know the semantics of system_clock off the top of my head, the sensible sequence and choice of types depends on the INTENTs and the possible values of the arguments)


Cheers,
- Tobi


---------------------------------------------------------------- This message was sent using IMP, the Internet Messaging Program.



Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]