OpenEuphoria

1. Phix: float32 return value

• Posted by andreasWagner 6 days ago
• 143 views

Hello,

It's me again, even at the risk of annoying you all. After a year, I've started working on a wrapper for Raylib again. (Or rather, I've started programming again in general)

So far, I've been able to find a workaround/solution for everything. But now I'm stuck on something that is (in my opinion) relatively simple. More specifically, this:

constant xGetMouseWheelMove = define_c_func(ray,GetMouseWheelMove,{},C_FLOAT) 

Under Phix64bit, calling the function returns incorrect values, but under Phix32bit it works. My goal is 64bit.(also because my workarounds/solutions for passing structures and receiving them as return values only work with 64-bit).
And, to be perfectly honest, I don't feel like making my code even more unreadable than it already is with if 32bit then do this else do that.

To make it easier to investigate and visualize the whole thing, I have built a test system.

A DLL:

// File: test_ffi.c 
// compile with:- m32 for 32bit dll 
//gcc -shared -m64 -o test_ffi64.dll test_ffi.c   
//or 
//gcc -shared -m64 -s -o test_ffi64.dll test_ffi.c -Wl,--out-implib,libtest_ffi.a 
 
#include <windows.h> 
 
// Funktioniert in GCC/MinGW 
__attribute__((dllexport)) float GetTestFloat() { 
    return 123.456f; 
} 
 
__attribute__((dllexport)) float MultiplyFloats(float a, float b) { 
    return a * b; 
} 
 

And a test program:

sequence dll={"test_ffi64.dll","test_ffi32.dll"} 
atom result 
atom x=2.5,y=3.5 
atom lib=open_dll(dll) 
 
constant GetTestFloat= define_c_func(lib,"GetTestFloat",{},C_FLOAT) 
constant MultiplyFloats=define_c_func(lib,"MultiplyFloats",{C_FLOAT,C_FLOAT},C_FLOAT) 
 
 
printf(1,"%d Bit %s\n",{machine_bits(),"System"}) 
 
result=c_func(GetTestFloat,{}) 
printf(1,"%f should be 123.456\n",result) 
 
result=c_func(MultiplyFloats,{x,y}) 
printf(1,"%f should be %f\n",{result,x*y}) 
 
wait_key() 

As I said, it works with phix32 bit but not with phix64:

My working system is Windows11 64-bit Phix1.0.5

3. Re: Phix: float32 return value

• Posted by andreasWagner 6 days ago
• 123 views

First of all, thank you for your quick reply.

okay, i end here:
D:\testpool\ray_workplace\Phix>p p
swi should be 5956(not 6060)

and its more like line 1271 not 1289 in pttree.e where the constants end.

Okay, I'll take another look tomorrow.

5. Re: Phix: float32 return value

• Posted by petelomax 5 days ago
• 111 views

Right, the improvements list for 1.0.6 is currently at 156 lines long, so my pttree.e is probably a bit mangled vs yours.
The intention was not to change T_swi (which should be the very last call to tt_stringF() in that file) but to "add 8" or
whatever it actually needs, so I'd take a (blind) stab at 5964 next. No biggie. If you really get stuck you'll have to wait
until 1.0.6 ships, but that's at least a few months off yet.

7. Re: Phix: float32 return value

• Posted by petelomax 4 days ago
• 77 views

My bad, I shd've gotten line nos from 1.0.5 srcs... maybe next time.

D:\testpool\ray_workplace\Phix\pilasm.e:4338 
            elsif ttidx=T_movsd or ttidx=T_movss then op = iff(ttidx=T_movsd?.. 
^ duplicate case value [3364, il offset:2791] 

That error means it has converted the if to a switch (for performance reasons) but hit a snag in that
there are two (or more) branches for 3364 (T_movsd) - there is one on the-1.0.5-line 4291 (ie where I should have pointed you).
I'll have a think about making the error say "duplicate case value [3364 on lines 4338 and 4291]".

C_DOUBLE instead of C_FLOAT return types would also be hassle-free (as long as the C code is returning a 64-bit float and not a 32-bit float), if that helps or you need the fractions.
Should you want to leave it as-is for now, I'm happy if you`re happy.
(undo if you can, but having non-compilable sources for the compiler itself lying around is not really going to hurt you, unless/until some other snag has no similar workaround)