llvm dev - Nov 2012 - [LLVMdev] Extended Inline asm with double data type crashes clang

Thanks Rafael,

Hello All,

Could anyone please comment, which part in selectiondag need to be
understood/modified to fix this.

Regards,
Rajesh




On Wed, Nov 21, 2012 at 2:38 AM, Rafael Espíndola <
rafael.espindola at gmail.com> wrote:
> I reported http://llvm.org/pr14393 to track it.
>
> On 20 November 2012 05:18, rajesh viswabramana
>  <viswabramana.rajesh at gmail.com> wrote:
> > Hi,
> >
> > Clang crashes when below snippet of code is compiled (used latest svn
> > version)
> >
> > double func1()
> > {
> >   double x ;
> >   asm ( "" : "=r"(x) : "0"(x) );
> >   return x;
> > }
> >
> >> clang -S test1.c
> >> Assertion failed: (PartVT.isInteger() || PartVT == MVT::x86mmx)
&&
> >> ValueVT.isInteger() && "Unknown mismatch!", file
> >> ..\..\..\..\lib\CodeGen\SelectionDAG\SelectionDAGBuilder.cpp, line
378
> >
> >
> > Compilation sucessful, If data type passed to inline asm is of float
> >   float x ;
> >   asm ( "" : "=r"(x) : "0"(x) );
> >
> > (or) If double data type passed to float type of registers
> > double x;
> > asm ( "" : "=t"(x) : "0"(x) );
> >
> > It seems currently LLVM (selectionDAG) doesn't handle double to
general
> > purpose register storing.
> > Could anyone please comment on this.
> >
> > Regards,
> > Rajesh
> >
> >
> >
> > _______________________________________________
> > LLVM Developers mailing list
> > LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
> >
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20121121/4ebb7b1e/attachment.html>

Hi all,

I tried same code on gcc for arm(hard float),

double f2 ()
{
double x = 10.0;
asm ("" : "=r" (x) : "0" (x));
return x;
}
> arm-linux-gnueabi-gcc -S -mhard-float pr39058.c -O2
Generates proper code,
    mov    r3, #0
    mov    r2, #0
    movt    r3, 16420
    fmdrr    d0, r2, r3
    bx    lr

But llvm crashes, If data type is "double", register type is
"int".
Any comments why does llvm currently handle this case ?



Regards,
Rajesh

On Wed, Nov 21, 2012 at 3:19 PM, rajesh viswabramana <
viswabramana.rajesh at gmail.com> wrote:
> Thanks Rafael,
>
> Hello All,
>
> Could anyone please comment, which part in selectiondag need to be
> understood/modified to fix this.
>
> Regards,
> Rajesh
>
>
>
>
> On Wed, Nov 21, 2012 at 2:38 AM, Rafael Espíndola <
> rafael.espindola at gmail.com> wrote:
>
>> I reported http://llvm.org/pr14393 to track it.
>>
>> On 20 November 2012 05:18, rajesh viswabramana
>>  <viswabramana.rajesh at gmail.com> wrote:
>> > Hi,
>> >
>> > Clang crashes when below snippet of code is compiled (used latest
svn
>> > version)
>> >
>> > double func1()
>> > {
>> >   double x ;
>> >   asm ( "" : "=r"(x) : "0"(x) );
>> >   return x;
>> > }
>> >
>> >> clang -S test1.c
>> >> Assertion failed: (PartVT.isInteger() || PartVT ==
MVT::x86mmx) &&
>> >> ValueVT.isInteger() && "Unknown mismatch!",
file
>> >> ..\..\..\..\lib\CodeGen\SelectionDAG\SelectionDAGBuilder.cpp,
line 378
>> >
>> >
>> > Compilation sucessful, If data type passed to inline asm is of
float
>> >   float x ;
>> >   asm ( "" : "=r"(x) : "0"(x) );
>> >
>> > (or) If double data type passed to float type of registers
>> > double x;
>> > asm ( "" : "=t"(x) : "0"(x) );
>> >
>> > It seems currently LLVM (selectionDAG) doesn't handle double
to general
>> > purpose register storing.
>> > Could anyone please comment on this.
>> >
>> > Regards,
>> > Rajesh
>> >
>> >
>> >
>> > _______________________________________________
>> > LLVM Developers mailing list
>> > LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
>> > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>> >
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20121122/d8086c67/attachment.html>

On Thu, Nov 22, 2012 at 4:14 AM, rajesh viswabramana
<viswabramana.rajesh at gmail.com> wrote:
> Hi all,
>
> I tried same code on gcc for arm(hard float),
>
> double f2 ()
> {
> double x = 10.0;
>
> asm ("" : "=r" (x) : "0" (x));
> return x;
> }
>
>> arm-linux-gnueabi-gcc -S -mhard-float pr39058.c -O2
>
> Generates proper code,
>     mov    r3, #0
>     mov    r2, #0
>     movt    r3, 16420
>     fmdrr    d0, r2, r3
>     bx    lr
>
> But llvm crashes, If data type is "double", register type is
"int".
> Any comments why does llvm currently handle this case ?
"r" is supposed to be a single register, not a register pair; the fact
that gcc accepts this is probably an accident.  clang should reject
this code (without crashing, of course).

-Eli