[EXT] Re: [PATCH 0/7] Simplify and optimize modf/modff

[Peter Bergner]

On 6/10/25 8:07 AM, Adhemerval Zanella Netto wrote:
> Ping.
> 
> On 28/05/25 14:59, Adhemerval Zanella wrote:
>> The patchset uses the powerpc implementation idea to make it generic and
>> leverage rounding instruction on newer chips. The PowerPC implementation
>> is also suboptimal, where a trunc operation can replace the floor/ceil
>> operation.
>>
>> The optimized implementation is gated through USE_TRUNCF_BUILTIN, since
>> libc exports modf/modff, and making it generic would add a lot of
>> complexity (since trunc will need to be pulled) and not all ABIs have
>> rounding instruction in the base ABI.
>>
>> Adhemerval Zanella (7):
>>   benchtests: Add modff benchtest
>>   benchtests: Improve modf benchtest
>>   math: Simplify and optimize modff implementation
>>   math: Simplify and optimize modf implementation
>>   powerpc: Remove modff optimization
>>   powerpc: Remove modf optimization
>>   x86_64: Optimize modf/modff for x86_64-v2

Given we resolved the performance data initially shown was just
confusion on my part, the powerpc changes are ok.

Peter