16 июля 2011, 20:25 от Andreas Kloeckner <[email protected]>:
> On Mon, 11 Jul 2011 10:37:45 +0400, Алексей Гурин <[email protected]> wrote:
> >
> > 11 июля 2011, 08:33 от Andreas Kloeckner <[email protected]>:
> > > Dear Алексей,
> > >
> > > On Fri, 08 Jul 2011 10:33:04 +0400, Алексей Гурин wrote:
> > >
> > > What kind of system are you on? (OS, Bitness, ...) What seems to be the
> > > case is that Python's struct module and nvcc disagree about alignment on
> > > your system. That's nasty. Unfortunately, I can't seem to provoke a
> > > similar bug on my 64-bit Linux system.
> > >
> > > Andreas
> >
> > i'm on Ubuntu linux 11.04 32-bit using CUDA 4.0.17.
>
> Can someone on 32-bit system confirm? What compiler are you using? Can
> you try and figure out the string passed to struct.pack() to pack the
> arguments?
>
> Andreas
I use gcc-4.4. Same error occurs when i use pycuda prepare with string "Pdi"
and then prepared call of function.
##########################
someFunc.prepare('Pdi')
someFunc.prepared_call(grid,block,arrayD,1,N)
##########################
In this case N=10000
Hex representation of arguments i send to GPU is
int N = 10000 = 0x00002710
double 1 = 0x3FF0000000000000
and what i actually receive on GPU function is
double 2.1220487890535086e-310 = 0x000027103FF00000
which is a mix of int and double values where four bytes of int goes first.
If I put double before other arguments like 'dPi' it works correctly.
_______________________________________________
PyCUDA mailing list
[email protected]
http://lists.tiker.net/listinfo/pycuda
