[C2M] Float constants from math.h give errors and warnings about mir HOT 6 CLOSED

Thank you for reporting this issue. GLIBC uses GCC builtins for getting inifinity and NaN. For non GCC conforming compilers, math.h uses big values and 0.0 / 0.0 for NAN.

I can not remove warnings because these warnings are useful in other cases but I can fix the error. The patch will be ready soon.

from mir.

Here is the patch
https://github.com/vnmakarov/mir/compare/84b277a..5f91154

from mir.

Is there a way to get Nan, -Inf and Inf values in MIR.
I am using following in my C output but not sure if this works:

#define nan (0./0.)
#define inf (1./0.)

Basically in a JIT mode, it is better to not have to rely on any external headers; in Ravi, no headers are used, and I rely only upon MIR.

from mir.

Is there a way to get Nan, -Inf and Inf values in MIR.
I am using following in my C output but not sure if this works:

#define nan (0./0.)
#define inf (1./0.)

It might be target-dependent. But for x86-64 1.0/0.0 is infinitity. 0.0/0.0 is actually -nan. It is better to use function nanl for getting NAN or the following code

static unsigned long nan_long = 0x7fffffffffffffff;
#define nan (*(double *)&nan_long)

Another problem is that NAN and INFITITY is defined in math.h. If I define builtin functions for such values and define macro __GNUC__, there would be no problem. But defining __GNUC__ would create much more other problems because most of GCC extensions are not implemented in c2m and a lot of programs check this to use these extensions.

Basically in a JIT mode, it is better to not have to rely on any external headers; in Ravi, no headers are used, and I rely only upon MIR.

from mir.

@vnmakarov Okay thank you. I suppose I can use just a macro - or would the static be optimized away?

I agree it is not a good idea to define __GNUC__ unless you want years of pain! :-)

Regards

from mir.

As <math.h> can be included without issues since the patch and the warnings can be silenced with -w, I am closing this.

from mir.