README.md

SYCL Academy

Exercise 19: Work-Group Sizes

---

In this exercise you will learn how to use different work-group sizes in order to compare the performance difference.

---

1.) Try different work-group sizes

In order to optimize for occupancy try different work-group sizes, by specifying a different local range for the nd_range.

Note that if the work-group size exceeds the maximum work-group size the target device supports then it will fail to execute.

Compare the performance of the various work-group sizes you try.

Build and execution hints

For DPC++: Using CMake to configure then build the exercise:

mkdir build
cd build
cmake .. "-GUnix Makefiles" -DSYCL_ACADEMY_USE_DPCPP=ON -DSYCL_ACADEMY_ENABLE_SOLUTIONS=OFF -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx
make exercise_19

Alternatively from a terminal at the command line:

icpx -fsycl -o sycl-ex-19 -I../External/Catch2/single_include ../Code_Exercises/Exercise_19_Work_Group_Sizes/source.cpp
./sycl-ex-19

For AdaptiveCpp:

# <target specification> is a list of backends and devices to target, for example
# "omp;generic" compiles for CPUs with the OpenMP backend and GPUs using the generic single-pass compiler.
# The simplest target specification is "omp" which compiles for CPUs using the OpenMP backend.
cmake -DSYCL_ACADEMY_USE_ADAPTIVECPP=ON -DSYCL_ACADEMY_INSTALL_ROOT=/insert/path/to/AdaptiveCpp -DACPP_TARGETS="<target specification>" ..
make exercise_19

alternatively, without CMake:

cd Code_Exercises/Exercise_19_Work_Group_Sizes
/path/to/adaptivecpp/bin/acpp -o sycl-ex-19 -I../../External/Catch2/single_include --acpp-targets="<target specification>" source.cpp
./sycl-ex-19