problem compiling code under xp 32 bits, cuda 4.1 vs 2010

I’m trying to compile some code into ptx under windows XP and then load it into matlab with the parallel computing toolbox. Matlan complains that: “No entries were found in the PTX file”, so I did some more tests. Compiling to PTX seems to work:

D:\analysis>nvcc -arch=sm_20 -ptx -Xptxas -V distsKernelSmallD.cu
distsKernelSmallD.cu
distsKernelSmallD.cu
tmpxft_00000ac8_00000000-3_distsKernelSmallD.cudafe1.gpu
tmpxft_00000ac8_00000000-10_distsKernelSmallD.cudafe2.gpu

Compiling to object fails saying that it couldn’t find the cubin file:

D:\analysis>nvcc -arch=sm_20 -c -Xptxas -V distsKernelSmallD.cu
distsKernelSmallD.cu
tmpxft_0000025c_00000000-3_distsKernelSmallD.cudafe1.gpu
tmpxft_0000025c_00000000-8_distsKernelSmallD.cudafe2.gpu
distsKernelSmallD.cu
ptxas: NVIDIA (R) Ptx optimizing assembler
Copyright (c) 2005-2011 NVIDIA Corporation
Built on Fri_Jan_13_02:49:02_PST_2012
Cuda compilation tools, release 4.1, V0.2.1221
fatbinary fatal : Could not open input file ‘C:/DOCUME~1/Micha/LOCALS~1/Temp/tmpxft_0000025c_00000000-13_distsKernelSmallD.sm_20.cubin’

If I remote the -Xptxas -V it supposedly works (I can create an executable that runs), but it doesn’t solve the matlab problem. Under windows 7 64bit the exact same process works great (making sure that the binary is 64bit instead of 32bit). Any ideas?

System is windows xp 32bit + vs 2010 + matlab 2011b.

Thanks