my build time impact of clang 5.0

[Matt Smith]

On Oct 03 09:46, Andriy Gapon wrote:
>On 02/10/2017 21:34, Dan Mack wrote:
>>
>> Another significant change in build times this week - not complaining,
>> just my observations on build times; same server doing buildworld during
>> the various phases of compiler changes over the last year or so FWIW:
>>
>> |--------------+--------------+---------------+----------+-----------|
>> | Ver (svn-id) | World (mins) | Kernel (mins) | Relative | Comment   |
>> |--------------+--------------+---------------+----------+-----------|
>> |       292733 |           90 |            16 |      0.5 |           |
>> |       299948 |           89 |            16 |      0.5 |           |
>> |       322724 |          174 |            21 |      1.0 | clang 4.x |
>> |       323310 |          175 |            21 |      1.0 | clang 4.x |
>> |       323984 |          175 |            21 |      1.0 | clang 4.x |
>> |       324130 |          285 |            21 |      1.6 | clang 5.x |
>> |       324204 |          280 |            21 |      1.6 | clang 5.x |
>> |--------------+--------------+---------------+----------+-----------|
>
>It shocked me to a realize that I can build several platforms that do not have
>clang support yet (like powerpc) one after another in a fraction of time
>required to build just amd64.
>

It is insane now how long it takes on lesser powered hardware. When we 
had gcc as the system compiler I could do a full buildworld/kernel in 
around 3 hours. With the first version of clang that we had imported I 
think it increased to around 6 hours. Clang4 made it around 8 hours. And 
now clang5 it took 12 hours!

This is on an Intel Atom D525 with amd64. And I have a few things 
disabled too like lib32 and profiled libraries.

-- 
Matt