README.md

SYCL Academy

Exercise 16: Coalesced Global Memory

---

In this exercise you will learn how to apply row-major and column-major when linearizing the global id in order to compare the performance difference due to coalesced global memory access.

---

1.) Evaluate global memory access

Now that you have a working image convolution kernel you should evaluate whether the global memory access patterns in your kernel are coalesced.

Consider two alternative ways to linearize the global id:

auto rowMajorLinearId    = (idx[1] * width) + idx[0];  // row-major
auto columnMajorLinearId = (idx[0] * height) + idx[1];  // column-major

Try using both of these and compare the execution time of each.

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_16

Alternatively from a terminal at the command line:

icpx -fsycl -o sycl-ex-16 -I../External/Catch2/single_include ../Code_Exercises/Exercise_16_Coalesced_Global_Memory/source.cpp
./sycl-ex-16

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_16

alternatively, without CMake:

cd Code_Exercises/Exercise_16_Coalesced_Global_Memory
/path/to/adaptivecpp/bin/acpp -o sycl-ex-16 -I../../External/Catch2/single_include --acpp-targets="<target specification>" source.cpp
./sycl-ex-16