<div dir="ltr"><br><div class="gmail_extra"><br><br><div class="gmail_quote">2014-07-02 13:42 GMT+02:00 David T. Lewis <span dir="ltr"><<a href="mailto:lewis@mail.msen.com" target="_blank">lewis@mail.msen.com</a>></span>:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div class="HOEnZb"><div class="h5"><br>
On Wed, Jul 02, 2014 at 08:01:35AM +0200, Nicolas Cellier wrote:<br>
><br>
> 2014-07-02 7:44 GMT+02:00 Nicolas Cellier <<br>
> <a href="mailto:nicolas.cellier.aka.nice@gmail.com">nicolas.cellier.aka.nice@gmail.com</a>>:<br>
><br>
> ><br>
> ><br>
> ><br>
> > 2014-07-02 0:28 GMT+02:00 David T. Lewis <<a href="mailto:lewis@mail.msen.com">lewis@mail.msen.com</a>>:<br>
> ><br>
> ><br>
> >> On Tue, Jul 01, 2014 at 09:21:00PM +0200, Nicolas Cellier wrote:<br>
> >> ><br>
> >> > 2014-07-01 4:22 GMT+02:00 <<a href="mailto:commits@source.squeak.org">commits@source.squeak.org</a>>:<br>
> >> ><br>
> >> > ><br>
> >> > > Item was changed:<br>
> >> > > ----- Method: LargeIntegersPlugin>>cDigitSub:len:with:len:into: (in<br>
> >> > > category 'C core') -----<br>
> >> > > + cDigitSub: pByteSmall len: smallLen with: pByteLarge len: largeLen<br>
> >> into:<br>
> >> > > pByteRes<br>
> >> > > + | z |<br>
> >> > > - cDigitSub: pByteSmall<br>
> >> > > - len: smallLen<br>
> >> > > - with: pByteLarge<br>
> >> > > - len: largeLen<br>
> >> > > - into: pByteRes<br>
> >> > > - | z limit |<br>
> >> > > <var: #pByteSmall type: 'unsigned char * '><br>
> >> > > <var: #pByteLarge type: 'unsigned char * '><br>
> >> > > <var: #pByteRes type: 'unsigned char * '><br>
> >> > ><br>
> >> > > + z := 0. "Loop invariant is -1<=z<=1"<br>
> >> > > + 0 to: smallLen - 1 do:<br>
> >> > > - z := 0.<br>
> >> > > - "Loop invariant is -1<=z<=1"<br>
> >> > > - limit := smallLen - 1.<br>
> >> > > - 0 to: limit do:<br>
> >> > > [:i |<br>
> >> > > z := z + (pByteLarge at: i) - (pByteSmall at: i).<br>
> >> > > + pByteRes at: i put: z - (z // 256 * 256).<br>
> >> "sign-tolerant<br>
> >> > > form of (z bitAnd: 255)"<br>
> >> > ><br>
> >> ><br>
> >> > Frankly, having z declared unsigned int and just doing pByteRes at: i<br>
> >> put:<br>
> >> > (z bitAnd: 16rFF) as I suggested would be way way simpler and will<br>
> >> ALWAYS<br>
> >> > work.<br>
> >> > Why the hell invoke the complications of signed arithmetic when the<br>
> >> content<br>
> >> > pByteRes is unsigned???<br>
> >><br>
> >> Nicolas,<br>
> >><br>
> >> This sounds right to me.<br>
> >><br>
> >> Do you have a final version of #cDigitSub:len:with:len:into: that you can<br>
> >> recommend? The version in VMMaker-nice.342 produces failures in the<br>
> >> KernelTests-Numbers tests when compiled in 64-bit mode, but I think that<br>
> >> you have since suggested some other improvements in this email thread.<br>
> >><br>
> >> Dave<br>
> >><br>
> >><br>
> > Hi David,<br>
> > that's interesting. VMMaker-nice.342 contains all the 32bits LargeInteger<br>
> > modifications, not just the #cDigitSub:len:with:len:into:.<br>
> > I tried to declare everything as <unsigned int> rather than <usqInt> with<br>
> > the assumption that sizeof(unsigned int)*2==sizeof(unsigned long long).<br>
> > But I never tried to compile a 64 bits VM, I think I lost the recipe for<br>
> > brewing such a flavour.<br>
> > I'm on MacOSX, could you provide a link to the shortest way to do it?<br>
> > Once I can check it, I'll publish a corrected version.<br>
> > Thanks<br>
> ><br>
> > Nicolas<br>
> ><br>
><br>
> One thing I remember is that my modifications are full of hardcoded<br>
> constants like 16rFFFFFFFF that are making the code hardly future proof.<br>
> Normally, I would never code like that in C, but use standard macros like<br>
> UINT_MAX.<br>
> It's just that I don't know how to do it in SLANG...<br>
> Any hint?<br>
><br>
> As for our own constants, like SmallInteger minVal and maxVal, we should<br>
> really have our own MACRO for this, there are way too many repetitions of<br>
> these constants spreaded in the code.<br>
<br>
</div></div>In VMM trunk, see CCodeGenerator>>emitDefineBytesPerWordOn: as well as the<br>
methods in category 'constants' in ObjectMemory. The constants defined in this<br>
manner form the basis for generating code that can be compiled for either<br>
32-bit or 64-bit images.<br>
<br>
BTW, this is what makes it is easy to test your #cDigitSub:len:with:len:into:<br>
change on various VM and image word sizes. The combination of 64-bit VM with<br>
32-bit image is by far the best for finding bugs, but a 64-bit image sometimes<br>
produces interesting results too.<br>
<br>
Dave<br>
<br>
</blockquote></div><br></div><div class="gmail_extra">Thanks both,<br></div><div class="gmail_extra">I had forgotten that we MUST have a way to simulate the VM, so C-Header only is a no go.<br></div><div class="gmail_extra">
I'll inquire as soon as I find time.<br><br></div><div class="gmail_extra">Nicolas<br></div></div>