Speed up genattrtab

[Paolo Bonzini]

On 06/17/2010 01:39 PM, Michael Matz wrote:
> On Wed, 16 Jun 2010, Paolo Bonzini wrote:
>> On 06/16/2010 05:51 PM, Jan Hubicka wrote:
>>> With a lot of cores, we might actually execute getnattrtab and compilation
>>> of stage1 insn-attrtab.c at same time and then just sanity check that
>>> genattrtab
>>> did not change results :)
>>
>> That's a different story.  It doesn't break the idea of bootstrapping,
>> and can be applied to all insn-* files.
>
> But it also doesn't give any gains.  You still would have to run
> genattrtab (to get something to compare with), and you still would have to
> compile insn-attrtab.c (to check that cc1 produced the same code as last
> stage).

You can parallelize those two, so you would still get a 50% gain on a 
beefy-enough machine (pre Jakub's patch).  However, the insn_dfa_code 
split makes the serialization of genattrtab and insn-attrtab.o 
compilation much less troublesome.

Paolo