Re: Efficiency of reading cyclic structures (SRFI-38)

Hi Manuel.

MS, 2017-09-23 09:41:
> Hi Sven,
> 
>> The attached patch speeds up reading large expressions that contain
>> many instances of sharing (via the notation of SRFI-38).
>> Read time for a 40 MB expression (180 MB without sharing) with
>> around 500.000 sharings:
>> before: 5.20 sec
>> after:  0.37 sec
>> (Times are averages measured on an Intel i7-4790K; data on an SSD;
>> IO caches cleared befor each run: echo 3 > /proc/sys/vm/drop_caches )
>> 
>> To use the patch, build as normal, but after the "make install" step
>> add the following steps:
>> 
>>   patch -p0 < bigloo-cycles.patch
>>   ( cd runtime && make lib && make lib_u )
>>   make install
>> 
>> Manuel, please consider for inclusion.
> I have included a slightly modified version of your patch (mostly because
> I was annoyed by always allocating a 32 words vector even if not used).

Yes, that was a good idea.

> Thanks for the patch. I'm preparing a new version that I will upload
> by an hour or two.

Thanks. The new version works as expected.

Ciao
Sven