topic Intel OpenCL Compiler does not know about cl_intel_printf in OpenCL* for CPU

Intel OpenCL Compiler does not know about cl_intel_printf

kugeleck — Sun, 17 Jun 2012 21:58:08 GMT

Hello,

I'm using Intel OpenCL SDK 2012 on Linux and I get the following error message when I compile my kernel:

:5:48: warning: OpenCL extension 'cl_intel_printf' is not supported - ignoring

This does not make sense since I can use it and it's working.

In my code I first check for cl_intel_printf with

#if defined cl_intel_printf

and then I enable it with

#pragma OPENCL EXTENSION cl_intel_printf : enable

This warning is quite annoying since I want to compile my OpenCL code with

-Werror

as a build parameter during development. This obviously does not work since the false warning prevents the code from executing.

I also noticed that

-cl-opt-disable

seems to be ignored since the build log contains

Kernel  was successfully vectorized

As far as I understand it vectorization is an optimization step.

How can I get rid of the warning?

Regards
kugeleck

Intel OpenCL Compiler does not know about cl_intel_printf

EvgeniyPeshkov — Thu, 21 Jun 2012 15:52:46 GMT

Are you using CPU or Intel GPU device?

On GPU cl_intel_printf extention and -cl-opt-disable compiler option are both unsupported.

Intel OpenCL Compiler does not know about cl_intel_printf

kugeleck — Thu, 21 Jun 2012 21:51:28 GMT

I'm using a dual core Intel CPU.

Intel OpenCL Compiler does not know about cl_intel_printf

EvgeniyPeshkov — Fri, 22 Jun 2012 09:05:33 GMT

I've tried -cl-opt-disable compiler option and yes, there is still "Kernel was successfully vectorized" string in log.

As OpenCL specification says-cl-opt-disable disables all optimizations that are listed in "Optimization Options" chapter like-cl-mad-enable,-cl-unsafe-math-optimizations etc. It's optimizations of math functions commonly. So this option doesn't have to disable Intel autovectorization. And vectorization is another kind of optimization related to parallel execution of work-items by splitting them to compute using wide vector ALUs.

You only can disable autovectorization by usingvec_type_hint kernel attribute (see OpenCL specification), or using -g (debug) option that disables autovectorization and all optimizations.

In respect of cl_intel printf extention, I have the same warning. Maybe it is a bug of SDK.