This is the mail archive of the 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: Ada bootstrap failure on ia64

> Robert Dewar claimed at the time that the Ada standard required all FP 
> operations to be performed in the widest available FP format, and if the 
> widest format was a soft-fp format, then that was obviously very bad for 
> Ada performance.

Well the deal actually is that Ada requires all hardware formats to be
supported, and it is not the Ada standard, but the Ada runtime, that
uses the longest format for some operations. This is fixable, and is on
our list of things to improve sometime, but it is significant work that
will not happen in the near future.

Generally Ada programmers will tend to use Long_Long_Float assuming that it is
the highest precision efficient fpt format, but that usage is not guaranteed.

Whether or not the software types should be in standard is an interesting
Ada discussion, but not necessarily relevant to the discussion here.

> It makes sense that WIDEST_HARDWARE_FP_SIZE should be the FP precision 
> not the FP format size, but since we haven't made this distinction 
> before, it isn't surprising that there is confusion here.

I think either would be fine for GNAT

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