result mismatch in BFS

[jinz2014]

Running the BFS example in OpenCL-U/D shows:

Computed node 2 cost (-2147483647 != 45) does not match the expected value

Thanks

[el1goluj]

Hi jinz2014,
Sorry for the delay. Could you please give us more details?

1. What input graph are you using?
2. What GPU are you testing on?
3. Does the error happens for both U and D version? In every run?

Thanks,
Juan

[jinz2014]

Sorry for the late reply. A few things need to be checked:

Some gcc version compiles the application with the following error:
g++ -std=c++11 -I/soft/libraries/intel/system_studio_2019/opencl/SDK/include -c -MMD -o main.o main.cpp
In file included from /soft/compilers/gcc/7.4.0/linux-rhel7-x86_64/include/c++/7.4.0/bits/move.h:54:0,
from /soft/compilers/gcc/7.4.0/linux-rhel7-x86_64/include/c++/7.4.0/atomic:42,
from kernel.h:36,
from main.cpp:36:
/soft/compilers/gcc/7.4.0/linux-rhel7-x86_64/include/c++/7.4.0/type_traits:680:65: error: there are no arguments to ‘__is_trivially_copyable’ that depend on a template parameter, so a declaration of ‘__is_trivially_copyable’ must be available [-fpermissive]
: public integral_constant<bool, __is_trivially_copyable(_Tp)>
....
...

So I switch to intel's compiler, but in Makefile, the following doesn't add the option for Intel C++ compilers.

51 is_pthreads=$(shell ${CXX} -v 2&gt; /dev/stdout | grep "Thread model" | awk {'print $$3'})
52 ifeq ($(is_pthreads),posix)
53 LIB+=-pthread
54 endif

When I run BFS, I got the following error for OpenCL-D BFS:

$./bfs

Intel(R) Xeon(R) CPU Number of nodes = 264346 Number of edges = 733846 Initialization Time (ms): 241.688000
Allocation Time (ms): 0.118000
Copy To Device Time (ms): 2.828000
Kernel Time (ms): 1.654000
Copy Back and Merge Time (ms): 0.532000
Computed node 2 cost (-2147483647 != 45) does not match the expected value

When I run BFS, I got the following error for OpenCL-U BFS:
$./bfs
Intel(R) Xeon(R) CPU Allocation Time (ms): 0.043000
Number of nodes = 264346 Number of edges = 733846 Initialization Time (ms): 123.644000
Segmentation fault

[el1goluj]

Hi jinz2014,
The Chai benchmarks have not been extensively tested on Intel Xeon CPU as the OpenCL device.

• For OpenCL-D, the error might be related to different consistency models on CPU and GPU or different thread scheduling.
• For OpenCL-U, please double check that your OpenCL stack for CPU supports OpenCL 2.0 (or higher) features.
  I will try to check and give more updates later.
  Thanks,
  Juan