RFC: PowerPC floating point features

[Michael Eager]

Daniel Jacobowitz wrote:
> On Sun, Apr 06, 2008 at 10:25:38AM -0700, Michael Eager wrote:
>> For an instruction supported on all variants (both BookE and E500)
>> with a double precision FPU.
> 
> I think you have your terminology switched.  E500 is (very
> approximately) an implementation of Book E; the FPR-based FPU is
> usually called "classic FP" and dates back long before Book E,
> at least to the 601.

Perhaps.

BookE, like some other PPC versions, has FP registers.

gcc uses E500 to mean specifically the PPC 8540:
The 8540 does FP operations in GPRs, which the docs
describe as defined in the SPE APU.

e500.h:
#define TARGET_E500 (rs6000_cpu == PROCESSOR_PPC8540)
#define TARGET_E500_SINGLE (TARGET_HARD_FLOAT && rs6000_float_gprs == 1)
#define TARGET_E500_DOUBLE (TARGET_HARD_FLOAT && rs6000_float_gprs == 2)

I'm just going with the flow:  gcc describes doing FP ops in
GPRS as a E500 feature.  Maybe there should be some better terminology.

I think that the ppc feature options are a bit confusing and indirect.
For example, rather than defining TARGET_FPRS directly, it's
defined backwards, by reference to doing float in GPRS:
#define TARGET_FPRS (rs6000_float_gprs == 0)

Perhaps there should be "TARGET_FPS" and "TARGET_FGPRS".

I'm happy to adopt whatever terminology is reasonable to make the
code more understandable.



-- 
Michael Eager	 eager@eagercon.com
1960 Park Blvd., Palo Alto, CA 94306  650-325-8077