[EXT] [PATCH 6/7] powerpc: Remove modf optimization

[Adhemerval Zanella Netto]

On 28/05/25 17:31, Peter Bergner wrote:
> On 5/28/25 7:59 AM, Adhemerval Zanella wrote:
>> The generic implementation is slight more optimized than the powerpc
>> one, where it has a more optimized inf/nan check (by not using FP
>> unit checks, along with branch prediction hints), and removed one
>> branch by issuing trunc instead of a combination of floor/ceil (which
>> also generated less code).
>>
>> On power10 with gcc 14.2.1:
>>
>> reciprocal-throughput        master         patch        difference
>> workload-0_1                 1.1351        0.9067            20.12%
>> workload-1_maxint            1.4230        0.9040            36.47%
>> workload-maxint_maxfloat     1.5038        0.9076            39.65%
>> workload-integral            1.1280        0.9111            19.23%
>>
>> latency                      master         patch        difference
>> workload-0_1                 1.1440        2.7117          -137.03%
>> workload-1_maxint            4.0556        2.7070            33.25%
>> workload-maxint_maxfloat     3.2122        2.7164            15.43%
>> workload-integral            3.2381        2.7281            15.75%
> 
> I like the idea of using an optimized generic routine over an arch
> specific routine, but I'm confused by your data.  For the data above,
> it looks like the throughput is less with the patch (ie, smaller number)
> and the latency is longer with the patch (ie, bigger number).  ...and not
> just on Power, but x86 as well.  Is that just my confusion?  If so, what
> are these data values actually specifying?

You are right and I think I messed up the number with a wrong assumption
about the reciprocal-throughput. I think the powerpc might the best
implementation, the final copysign is indeed hurting performance even
on aarch64. Let me recheck everything.