Re: RoarVM: The Manycore SqueakVM
Posted by johnmci on Nov 06, 2010; 9:00pm
URL: https://forum.world.st/RoarVM-The-Manycore-SqueakVM-tp3025321p3030358.html
The other cheerful thing the switch statement does is put a range check in.
This would cause a range check to be done for each bytecode looked up unless you did the gnuify.
15 years back we actually altered the 68K binary to no-op out the range check which improved performance a
measurable amount on.
On 2010-11-06, at 1:42 PM, Levente Uzonyi wrote:
> On Sat, 6 Nov 2010, Igor Stasenko wrote:
>
>> I suppose its related to fact, that with small number of cases (like up to 16)
>> its not worth using indirect jump, because with branch prediction you
>> can achieve same or even better performance.
>> That's why i think, modern compilers do not using jump tables.
>> Having more than 16 number of cases, is an edge case. Not saying about 256.
>
> I expect a compiler to generate the fastest possible code, if I ask for that. I think we agree, that gcc doesn't do this.
>
>
> Levente
>
>>
>>
>> Btw, I was also thinking that compilers generating jump tables for
>> case statements.
>>--
===========================================================================
John M. McIntosh <[hidden email]> Twitter: squeaker68882
Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com
===========================================================================
URL: https://forum.world.st/RoarVM-The-Manycore-SqueakVM-tp3025321p3030358.html
The other cheerful thing the switch statement does is put a range check in.
This would cause a range check to be done for each bytecode looked up unless you did the gnuify.
15 years back we actually altered the 68K binary to no-op out the range check which improved performance a
measurable amount on.
On 2010-11-06, at 1:42 PM, Levente Uzonyi wrote:
> On Sat, 6 Nov 2010, Igor Stasenko wrote:
>
>> I suppose its related to fact, that with small number of cases (like up to 16)
>> its not worth using indirect jump, because with branch prediction you
>> can achieve same or even better performance.
>> That's why i think, modern compilers do not using jump tables.
>> Having more than 16 number of cases, is an edge case. Not saying about 256.
>
> I expect a compiler to generate the fastest possible code, if I ask for that. I think we agree, that gcc doesn't do this.
>
>
> Levente
>
>>
>>
>> Btw, I was also thinking that compilers generating jump tables for
>> case statements.
>>
===========================================================================
John M. McIntosh <[hidden email]> Twitter: squeaker68882
Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com
===========================================================================
smime.p7s (5K) | Free forum by Nabble | Edit this page |