README.md

SYCL Academy

Exercise 17: Vectors

---

In this exercise you will learn how to use vec to explicitly vectorized your kernel function.

---

1.) Use vectors

Now that global memory access is coalesced another optimization you can do is to use the vec class to present the pixels in the image.

To do this first you want to reinterpret the buffer objects as buffers with the element type float4. You can do this by calling reinterpret on the buffers and specifying the type float4.

Note you will have to provide a new ranges which will be the original ranges divided by 4 to account for the channels which are now represented by the four elements of the vector type.

2.) Refactor the kernel function

Now that you are using vector types, you want to replace the representation of pixels in the kernel function with a float4 and remove any multiplication or offsets to account for the number of channels.

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_17

Alternatively from a terminal at the command line:

icpx -fsycl -o sycl-ex-17 -I../External/Catch2/single_include ../Code_Exercises/Exercise_17_Vectors/source.cpp
./sycl-ex-17

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_17

alternatively, without CMake:

cd Code_Exercises/Exercise_17_Vectors
/path/to/adaptivecpp/bin/acpp -o sycl-ex-17 -I../../External/Catch2/single_include --acpp-targets="<target specification>" source.cpp
./sycl-ex-17