[Caml-list] flambda optimizations depends on unrelated things

[Caml-list] flambda optimizations depends on unrelated things

[Alexey Egorov]

Hello,

I'm trying hard to make efficient pure functional streams
implementation - and while playing with flambda, I'm discovered that
optimizations is VERY fragile and sometimes depends on completely
unrelated things.

Sometimes flambda goes mad if you change filename or just add some
empty lines to your source.

Here is example, where renaming source file increases code size by a
factor of 2 because of some unfortunate inliner decisions:

$ ls
fragile.ml
$ cp fragile.ml fragile2.ml
$ md5sum *
6a13d560736c81b3e5b5dfe7f13c2f56  fragile2.ml
6a13d560736c81b3e5b5dfe7f13c2f56  fragile.ml
$ ocamlopt -version
4.05.0
$ ocamlopt -O3 fragile.ml -dflambda 2>&1 | wc -l
1774
$ ocamlopt -O3 fragile2.ml -dflambda 2>&1 | wc -l
3835

Here is the code - https://pastebin.com/Y1DDidyN

Re: [Caml-list] flambda optimizations depends on unrelated things

[David Allsopp]

Alexey Egorov wrote:
> Hello,
> 
> I'm trying hard to make efficient pure functional streams
> implementation - and while playing with flambda, I'm discovered that
> optimizations is VERY fragile and sometimes depends on completely
> unrelated things.
> 
> Sometimes flambda goes mad if you change filename or just add some
> empty lines to your source.
> 
> Here is example, where renaming source file increases code size by a
> factor of 2 because of some unfortunate inliner decisions:
> 
> $ ls
> fragile.ml
> $ cp fragile.ml fragile2.ml
> $ md5sum *
> 6a13d560736c81b3e5b5dfe7f13c2f56 fragile2.ml
> 6a13d560736c81b3e5b5dfe7f13c2f56 fragile.ml
> $ ocamlopt -version
> 4.05.0
> $ ocamlopt -O3 fragile.ml -dflambda 2>&1 | wc -l
> 1774
> $ ocamlopt -O3 fragile2.ml -dflambda 2>&1 | wc -l
> 3835
> 
> Here is the code - https://pastebin.com/Y1DDidyN

Please could you file a report on Mantis? https://caml.inria.fr/mantis


David

Re: [Caml-list] flambda optimizations depends on unrelated things

[Alexey Egorov]

Sure https://caml.inria.fr/mantis/view.php?id=7614

2017-08-27 13:33 GMT+05:00 David Allsopp <dra-news@metastack.com>:
> Alexey Egorov wrote:
>> Hello,
>>
>> I'm trying hard to make efficient pure functional streams
>> implementation - and while playing with flambda, I'm discovered that
>> optimizations is VERY fragile and sometimes depends on completely
>> unrelated things.
>>
>> Sometimes flambda goes mad if you change filename or just add some
>> empty lines to your source.
>>
>> Here is example, where renaming source file increases code size by a
>> factor of 2 because of some unfortunate inliner decisions:
>>
>> $ ls
>> fragile.ml
>> $ cp fragile.ml fragile2.ml
>> $ md5sum *
>> 6a13d560736c81b3e5b5dfe7f13c2f56 fragile2.ml
>> 6a13d560736c81b3e5b5dfe7f13c2f56 fragile.ml
>> $ ocamlopt -version
>> 4.05.0
>> $ ocamlopt -O3 fragile.ml -dflambda 2>&1 | wc -l
>> 1774
>> $ ocamlopt -O3 fragile2.ml -dflambda 2>&1 | wc -l
>> 3835
>>
>> Here is the code - https://pastebin.com/Y1DDidyN
>
> Please could you file a report on Mantis? https://caml.inria.fr/mantis
>
>
> David
>
>