Some time ago, Iain Buclaw kindly added non-asm implementations of math functions in std.math. However, CTFE still seems unable to correctly evaluate certain functions like atan2:

	import std.math;
	enum x = atan2(1.0, 2.0);

Compiler output:

	/usr/src/d/phobos/std/math.d(856): Error: asm statements cannot be interpreted at compile time
	/usr/src/d/phobos/std/math.d(917):        called from here: atan2(cast(real)y, cast(real)x)
	test.d(2):        called from here: atan2(1.00000, 2.00000)

Looking at std/math.d, it appears that version=InlineAsm_X86_Any is defined, which leads to the asm implementation of atan2, which CTFE can't interpret.

However, right below that block is an else block containing a D implementation of atan2 that AFAICT *can* be interpreted in CTFE.

So it would appear that version=InlineAsm_X86_Any should be suppressed during CTFE? But I'm not sure how this could be done in the compiler.

In any case, it's an annoying limitation. :-(


T

-- 
Never step over a puddle, always step around it. Chances are that whatever made it is still dripping.

On Thursday, 27 March 2014 at 18:18:25 UTC, H. S. Teoh wrote:
> Some time ago, Iain Buclaw kindly added non-asm implementations of math
> functions in std.math. However, CTFE still seems unable to correctly
> evaluate certain functions like atan2:
>
> 	import std.math;
> 	enum x = atan2(1.0, 2.0);
>
> Compiler output:
>
> 	/usr/src/d/phobos/std/math.d(856): Error: asm statements cannot be interpreted at compile time
> 	/usr/src/d/phobos/std/math.d(917):        called from here: atan2(cast(real)y, cast(real)x)
> 	test.d(2):        called from here: atan2(1.00000, 2.00000)
>
> Looking at std/math.d, it appears that version=InlineAsm_X86_Any is
> defined, which leads to the asm implementation of atan2, which CTFE
> can't interpret.
>
> However, right below that block is an else block containing a D
> implementation of atan2 that AFAICT *can* be interpreted in CTFE.
>
> So it would appear that version=InlineAsm_X86_Any should be suppressed
> during CTFE? But I'm not sure how this could be done in the compiler.
>
> In any case, it's an annoying limitation. :-(
>
>
> T

//----
auto atan(double x)
{
    version(InlineAsm_X86_Any)
    if (!__ctfe)
    {
        asm
        {
            ...
        }
        return;
    }

    Fall back non-InlineAsm_X86_Any and ctfe
}
//----

?