I don't know what is the right design in this case. Intrinsics are useful because they sometimes give more performance, but normal functions are sometimes more handy because they allow more flexibility, like taking their address ("first class functions"):


import std.math;
void main() {
    auto a = [&sin, &cos];
}


DMD 2.055:

OPTLINK (R) for Win32  Release 8.00.12
Copyright (C) Digital Mars 1989-2010  All rights reserved.
http://www.digitalmars.com/ctg/optlink.html
test.obj(test)
 Error 42: Symbol Undefined _D3std4math3sinFNaNbNfeZe
test.obj(test)
 Error 42: Symbol Undefined _D3std4math3cosFNaNbNfeZe

Isn't it possible to find some middle way that allows me to use std.math.sin as true functions, while keeping the advantages of intrinsics?

(Currently I define wrapper functions like mysin, mycos, etc).

Bye,
bearophile

On 09/18/2011 08:57 PM, bearophile wrote:
> I don't know what is the right design in this case. Intrinsics are useful because they sometimes give more performance, but normal functions are sometimes more handy because they allow more flexibility, like taking their address ("first class functions"):
>
>
> import std.math;
> void main() {
>      auto a = [&sin,&cos];
> }
>
>
> DMD 2.055:
>
> OPTLINK (R) for Win32  Release 8.00.12
> Copyright (C) Digital Mars 1989-2010  All rights reserved.
> http://www.digitalmars.com/ctg/optlink.html
> test.obj(test)
>   Error 42: Symbol Undefined _D3std4math3sinFNaNbNfeZe
> test.obj(test)
>   Error 42: Symbol Undefined _D3std4math3cosFNaNbNfeZe
>
> Isn't it possible to find some middle way that allows me to use std.math.sin as true functions, while keeping the advantages of intrinsics?
>
> (Currently I define wrapper functions like mysin, mycos, etc).
>
> Bye,
> bearophile

+1, the compiler should just rewrite your example so that it 'just works'. An alternative to your fix is to create an object file that contains the appropriate symbols.

On 18-09-2011 21:47, Timon Gehr wrote:
> On 09/18/2011 08:57 PM, bearophile wrote:
>> I don't know what is the right design in this case. Intrinsics are
>> useful because they sometimes give more performance, but normal
>> functions are sometimes more handy because they allow more
>> flexibility, like taking their address ("first class functions"):
>>
>>
>> import std.math;
>> void main() {
>> auto a = [&sin,&cos];
>> }
>>
>>
>> DMD 2.055:
>>
>> OPTLINK (R) for Win32 Release 8.00.12
>> Copyright (C) Digital Mars 1989-2010 All rights reserved.
>> http://www.digitalmars.com/ctg/optlink.html
>> test.obj(test)
>> Error 42: Symbol Undefined _D3std4math3sinFNaNbNfeZe
>> test.obj(test)
>> Error 42: Symbol Undefined _D3std4math3cosFNaNbNfeZe
>>
>> Isn't it possible to find some middle way that allows me to use
>> std.math.sin as true functions, while keeping the advantages of
>> intrinsics?
>>
>> (Currently I define wrapper functions like mysin, mycos, etc).
>>
>> Bye,
>> bearophile
>
> +1, the compiler should just rewrite your example so that it 'just
> works'. An alternative to your fix is to create an object file that
> contains the appropriate symbols.

You could write a wrapper function that calls the intrinsic, but I suppose this defeats the purpose...

- Alex

On 9/19/11 3:32 PM, Alex Rønne Petersen wrote:
> On 18-09-2011 21:47, Timon Gehr wrote:
>> On 09/18/2011 08:57 PM, bearophile wrote:
>>> I don't know what is the right design in this case. Intrinsics are
>>> useful because they sometimes give more performance, but normal
>>> functions are sometimes more handy because they allow more
>>> flexibility, like taking their address ("first class functions"):
>>>
>>>
>>> import std.math;
>>> void main() {
>>> auto a = [&sin,&cos];
>>> }
>>>
>>>
>>> DMD 2.055:
>>>
>>> OPTLINK (R) for Win32 Release 8.00.12
>>> Copyright (C) Digital Mars 1989-2010 All rights reserved.
>>> http://www.digitalmars.com/ctg/optlink.html
>>> test.obj(test)
>>> Error 42: Symbol Undefined _D3std4math3sinFNaNbNfeZe
>>> test.obj(test)
>>> Error 42: Symbol Undefined _D3std4math3cosFNaNbNfeZe
>>>
>>> Isn't it possible to find some middle way that allows me to use
>>> std.math.sin as true functions, while keeping the advantages of
>>> intrinsics?
>>>
>>> (Currently I define wrapper functions like mysin, mycos, etc).
>>>
>>> Bye,
>>> bearophile
>>
>> +1, the compiler should just rewrite your example so that it 'just
>> works'. An alternative to your fix is to create an object file that
>> contains the appropriate symbols.
>
> You could write a wrapper function that calls the intrinsic, but I
> suppose this defeats the purpose...
>
> - Alex

Or maybe you could define a wrapper function that calls the intrinsic...