Problems with Intel Compiler 11.1 (Linux) and NVIDIA OpenCL

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

I’m having troubles with the OpenCL Header in CUDA Toolkit 3.2 RC and the Intel Compiler (ICC) 11.1 64bit. Just including cl.h leeds to following error during compilation of that source file:

— begin —
/opt/cuda/include/CL/cl_platform.h(358): error: invalid attribute for “__cl_float2={float}”
typedef cl_float __cl_float2 attribute((vector_size(8)));
— end —

If I use the g++ there are no problems but in the release notes is written that ICC 11.1 is supported by CUDA toolkit 3.2.

Any ideas? Thanks.

2

Hi,

I am having the same problem, did you find a solution?

thanks

3

Hi,

I am having the same problem, did you find a solution?

thanks

4

Same here, any suggestion?

5

Same here, any suggestion?

6

This is really annoying, would somebody please give some advice? TIA!

7

This is really annoying, would somebody please give some advice? TIA!

8

Looks like you are really desperate. The issue is Intel related. You can take cl_platform.h distributed by AMD in their streaming SDK and replace the one provided by NVIDIA - they included define to handle explicitly Intel compiler.

  • rzolau
9

Looks like you are really desperate. The issue is Intel related. You can take cl_platform.h distributed by AMD in their streaming SDK and replace the one provided by NVIDIA - they included define to handle explicitly Intel compiler.

  • rzolau
10

A quick-fix is to add the compile flag:

-no-gcc

e.g.:

icc -no-gcc ...

or:

CFLAGS += -no-gcc

etc…

The caveats to the option shouldn’t be an issue, but check the help just in case.

hth

David

11

A quick-fix is to add the compile flag:

-no-gcc

e.g.:

icc -no-gcc ...

or:

CFLAGS += -no-gcc

etc…

The caveats to the option shouldn’t be an issue, but check the help just in case.

hth

David