www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - Automatically using stack allocations instead of GC

reply Per =?UTF-8?B?Tm9yZGzDtnc=?= <per.nordlow gmail.com> writes:
Are there any plans (or is it already happening) to make 
D-compilers automatically use stack allocations when possible in 
cases like

int foo()
{
     auto x = [1, 2]; // should be allocated on the stack
     return y = x[0] + x[1];
}

where allocations are "small enough" and cannot escape the 
current scope?

And how does/should/will this interact with ` nogc`?
Oct 23 2017
next sibling parent Daniel Kozak <kozzi11 gmail.com> writes:
On Monday, 23 October 2017 at 09:06:21 UTC, Per Nordlöw wrote:

 Are there any plans (or is it already happening) to make 
 D-compilers automatically use stack allocations when possible 
 in cases like

 int foo()
 {
     auto x = [1, 2]; // should be allocated on the stack
     return y = x[0] + x[1];
 }

 where allocations are "small enough" and cannot escape the 
 current scope?

 And how does/should/will this interact with ` nogc`?


AFAIK does not work right now.
Oct 23 2017
prev sibling next sibling parent reply Dmitry Olshansky <dmitry.olsh gmail.com> writes:
On Monday, 23 October 2017 at 09:06:21 UTC, Per Nordlöw wrote:

 Are there any plans (or is it already happening) to make 
 D-compilers automatically use stack allocations when possible 
 in cases like

 int foo()
 {
     auto x = [1, 2]; // should be allocated on the stack
     return y = x[0] + x[1];
 }

 where allocations are "small enough" and cannot escape the 
 current scope?

 And how does/should/will this interact with ` nogc`?


LDC does something like that IIRC.
Oct 23 2017
parent reply Nicholas Wilson <iamthewilsonator hotmail.com> writes:
On Monday, 23 October 2017 at 10:42:44 UTC, Dmitry Olshansky 
wrote:

 On Monday, 23 October 2017 at 09:06:21 UTC, Per Nordlöw wrote:

 Are there any plans (or is it already happening) to make 
 D-compilers automatically use stack allocations when possible 
 in cases like

 int foo()
 {
     auto x = [1, 2]; // should be allocated on the stack
     return y = x[0] + x[1];
 }

 where allocations are "small enough" and cannot escape the 
 current scope?

 And how does/should/will this interact with ` nogc`?


 LDC does something like that IIRC.


It does. 
https://github.com/ldc-developers/ldc/blob/master/gen/passes/GarbageCollect2Stack.cpp
It is enabled by default at -O2 or higher (but not -Os or -Oz).
Oct 23 2017
parent Johan Engelen <j j.nl> writes:
On Monday, 23 October 2017 at 12:49:18 UTC, Nicholas Wilson wrote:

 On Monday, 23 October 2017 at 10:42:44 UTC, Dmitry Olshansky 
 wrote:

 On Monday, 23 October 2017 at 09:06:21 UTC, Per Nordlöw wrote:

 int foo()
 {
     auto x = [1, 2]; // should be allocated on the stack
     return y = x[0] + x[1];
 }


 LDC does something like that IIRC.


 It does.


It does not.
https://godbolt.org/g/i21t8G


 https://github.com/ldc-developers/ldc/blob/master/gen/passes/GarbageCollect2Stack.cpp


This broke somehow quite a while ago :(
I haven't found time to look at it. Would be great if someone did 
(most important: add tests for it!!!!!).

-Johan
Oct 23 2017
prev sibling parent reply Walter Bright <newshound2 digitalmars.com> writes:
On 10/23/2017 2:06 AM, Per Nordlöw wrote:

 Are there any plans (or is it already happening) to make D-compilers 
 automatically use stack allocations when possible in cases like
 
 int foo()
 {
      auto x = [1, 2]; // should be allocated on the stack
      return y = x[0] + x[1];
 }
 
 where allocations are "small enough" and cannot escape the current scope?


There are no plans at the moment, but it's a good idea that `scope` can make 
possible.


 And how does/should/will this interact with ` nogc`?


If it gets allocated on the stack, then it should be compatible with  nogc.
Oct 23 2017
parent Per =?UTF-8?B?Tm9yZGzDtnc=?= <per.nordlow gmail.com> writes:
On Monday, 23 October 2017 at 10:48:37 UTC, Walter Bright wrote:

 There are no plans at the moment, but it's a good idea that 
 `scope` can make possible.


I'm glad your open for such automatic optimizations, Walter. 
Making D compilers automate these things which are cumbersome 
manual labour in languages such as Rust is, IMHO, the competitive 
way forward for D.


 And how does/should/will this interact with ` nogc`?


 If it gets allocated on the stack, then it should be compatible 
 with  nogc.


Great. I believe good diagnostics (for, in this case, mismatches 
between allocations and qualifiers) will play a key role in this 
regard.
Oct 23 2017