This is the mail archive of the gdb-patches@sources.redhat.com mailing list for the GDB project.


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

Re: [RFA] Patch to sim/common/cgen-ops.h


Ben Elliston wrote:
> 
> fche wrote:
> 
>    :      * the way to handle out-of-range arguments to the SUBWORD
>    :        functions.  Perhaps it's okay to use assert(), since we're
>    :        wasting a few cycles already by doing the test.
> 
>    Sure.
> 
> I'll make that change, then.
> 
>    :      * whether it makes sense to join two SFs into a DF.
> 
>    i.e., make a double out of two floats?  I don't think so. Maybe you
>    mean to build a DF out of two SI's -- go through a DI first then.
> 
> No, I want to join two SFs together to form a DF -- one architecture that I
> am thinking of allows you to store pairs of single precision fp registers in
> one of its double precision fp registers.  If you can think of a simpler way
> (perhaps using bitwise operators?), then we can eliminate the need for this.
> I just thought that using (join ..) was clearer.
> 
> Ben

The operation itself is not meaningful (you are subsuming the 2nd SF's exponent as a mantissa
and you are stealing mantissa bits from the 1st SF to form the new DF exponent).  Even if
this is valid behavior on the chip, IMO it would be more straightforward to have two different
sets of virtual registers (SF vs DF).  They would both share the same storage and using pointers
you would get your desired result.

-- Jeff J.

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