[ros-dev] revision 26772, Mon May 14 03:18:42 2007 UTC

[Timo Kreuzer]

jimtabor schrieb:
> No, this would break compatibility. The way to verify this. Write a
> driver, install it
> and send a packet to it, so to activate the test. Have the driver dump 
> to WinDbg.
Ok, I take my time to go into this, because I *am* interested in 
compatibility.

Let's analyze how the conversion between FLOAT and EFLOAT works on XP. 
We could simply use some documentation floating around, but let's do it 
on our own, just for the fun. We don't need no 
itchy-bitchy-kmtest-driver-callback-winpooch-hooky-thingy, because 
Windows has already all we need to do this completely in user mode.
DC_ATTR has some EFLOATs and what we need is a function that takes a 
FLOAT and stores it as EFLOAT. SetWorldTransform() seems suitable (In 
fact the values will be stored in the DC_ATTR structure only after a 
following call to GetWorldTransform()). And we need a windows compatible 
structure definition for DC_ATTR, wich we can get from a well known 
book. For the code see attachment.

Now let's see what we get with different parameters for the float:

f: 1.000000 (0x3f800000) -> lMant: 0x40000000, lExp: 0x000002
f: -1.000000 (0xbf800000) -> lMant: 0xc0000000, lExp: 0x000002
f: 2.000000 (0x40000000) -> lMant: 0x40000000, lExp: 0x000003
f: -2.000000 (0xc0000000) -> lMant: 0xc0000000, lExp: 0x000003
f: 1.500000 (0x3fc00000) -> lMant: 0x60000000, lExp: 0x000002
f: -1.500000 (0xbfc00000) -> lMant: 0xa0000000, lExp: 0x000002
f: 1235.000000 (0x449a6000) -> lMant: 0x4d300000, lExp: 0x00000c
f: -1235.000000 (0xc49a6000) -> lMant: 0xb2d00000, lExp: 0x00000c

Looks like Feng Yuan was right.
Now let's see what your code would do with f=1.

FtoEF( EFLOAT_S * efp, FLOATL f)
{

...
 Exp = EXP(worker.l);            // Exp = (((0x3f800000) >> 23L) & 0xFF) = 0x7F
 Mant = MANT(worker.l);  // Mant = ((0x3f800000) & 0x7FFFFFL) = 0x00000

 Mant = ((Mant << 7) | 0x40000000 | SIGN(worker.l)); 
        // Mant = (0 << 7) | 0x40000000 | (0x3f800000 & 0x80000000 )= 0 | 0x40000000 | 0 = 0x40000000
 Exp -= (1 - EXCESS); // Exp = 0x7F - (1-126) = 0x7F + 0x7D = 0xFC

 efp->lMant = Mant;
 efp->lExp = Exp;
...
}

Replace the - by a + and you get

 Exp += (1 - EXCESS); // Exp = 0x7F + (1-126L) = 0x7F - 0x7D = 0x2

Wich is what we have discovered windows does.

And it looks like there's another thing, you didn't take into account. 
Ok, I give you a hint: It has something todo with the sign bit.
You don't get it? Well, it's all there!

Thanks,
Timo

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.reactos.org/pipermail/ros-dev/attachments/20070521/18212580/attachment.html 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: main.c
Type: image/x-xbitmap
Size: 1453 bytes
Desc: not available
Url : http://www.reactos.org/pipermail/ros-dev/attachments/20070521/18212580/attachment.xbm 
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: gdi.h
Url: http://www.reactos.org/pipermail/ros-dev/attachments/20070521/18212580/attachment.h