Getting FPGA OpenCL platform on OneAPI devcloud

[Elliott]

Hi,

I'm trying to run an OpenCL program on the iGPUs on Intel OneAPI devcloud. The complete source code and instructions to reproduce this are here:

https://github.com/elliottslaughter/Summer20/tree/4034e0141e8f401944f635699950e02586187f9d/OpenCLEx

Basically, I'm running

ssh devcloud
git clone https://github.com/elliottslaughter/Summer20.git
qsub -I -l nodes=1:gpu:ppn=2 -d .
cd Summer20/OpenCLEx
clang -c -cl-std=CL1.2 -target spir64 -emit-llvm -Xclang -finclude-default-header -flto sample_kernel.cl -o sample_kernel64.ll
llvm-spirv sample_kernel64.ll -o sample_kernel64.spv
clang++ -O0 -g -std=c++0x -o main main.cpp -lOpenCL

When I run the program, I get:

$ ./main 
Running on platform: Intel(R) FPGA Emulation Platform for OpenCL(TM)
Running on device: Intel(R) FPGA Emulation Device
CL_DEVICE_ADDRESS_BITS is 64 for this device.
Reading SPIR-V from file: sample_kernel64.spv
Building program with build options: (none)
getPlatformVersion OpenCL 1.2 Intel(R) FPGA SDK for OpenCL(TM), Version 20.3
main: main.cpp:108: cl::Program createProgramWithIL(const cl::Context &, const std::vector<cl_uchar> &): Assertion `false' failed.
Aborted

This would seem to indicate that I'm running on the OpenCL platform for FPGAs. This is confusing since I'm on a GPU node and am using the default ~/.bash_profile, and so on. Furthermore, when I run ldd on the binary it seems to be pointing to the OpenCL for OneAPI:

$ ldd main
	linux-vdso.so.1 (0x00007fff9d1ca000)
	libOpenCL.so.1 => /glob/development-tools/versions/oneapi/beta07/inteloneapi/compiler/latest/linux/lib/libOpenCL.so.1 (0x00007f2f2446f000)
	libstdc++.so.6 => /usr/lib/x86_64-linux-gnu/libstdc++.so.6 (0x00007f2f23ecc000)
	libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007f2f23b2e000)
	libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007f2f23916000)
	libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007f2f23525000)
	libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007f2f23306000)
	libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007f2f23102000)
	/lib64/ld-linux-x86-64.so.2 (0x00007f2f24255000)

 This is in the same tree as my copy of Clang:

$ which clang
/glob/development-tools/versions/oneapi/beta07/inteloneapi/compiler/latest/linux/bin/clang

 Am I doing something wrong? Is there something I need to do to get access to the OpenCL platform for iGPUs?

(As an aside, I also note that the OpenCL seems to be old, version 1.2. I believe I need version 2.1 to get access to clCreateProgramWithIL, which I need for my use case. Is there a newer version available?)

Thanks.

[Elliott · Accepted solution]

A colleague pointed out to me that my code was selecting the wrong OpenCL platform, and in fact the code already has a flag to choose platform. Running my example with

./main -p 2

Causes it to run on the GPU. (And it is using OpenCL 2.1 now.)

[ArunJ_Intel]

Hi Elliott,

Glad to know you found the solution. Is there any further assistance you would like regarding this issue from Intel?

Thanks

Arun

[Elliott]

Thanks for asking. No further assistance is needed at this time.

[ArunJ_Intel]

Thanks for the confirmation. We would not be monitoring this conversation further. Please feel free to raise a new thread in case of any further issues.