Skip to content

Latest commit

 

History

History
109 lines (79 loc) · 3.57 KB

README.md

File metadata and controls

109 lines (79 loc) · 3.57 KB
Raw

SYCL Academy

Exercise 8: USM Vector Add

In this exercise you will learn how to use the USM API to write a SYCL application which performs a vector add.

1.) Allocate data on the host

Allocate memory on the host for the input and out data just as you did when using the buffer/accessor model.

2.) Create a queue

Create a queue using the USM device selector from exercise 7, remember to handle errors.

3.) Allocate device memory

When using the USM model the first thing you need to do is allocate the USM device memory.

To do this call malloc_device to allocate memory for the two inputs and the output.

4.) Copy to the device

Before you can perform any computation on the data you must copy it to the device.

To do this call the queue member function memcpy for each of the two inputs, remember to call wait on the event that is returned.

5.) Define the kernel function

Now you can define the kernel function itself, which is largely the same as in exercise 7.

This can be done differently from the buffer/accessor model, by calling the shortcut member function parallel_for on the queue rather than creating a command group.

Note that as you are accessing a pointer rather than an accessor you must retrieve an integral index, which can be done by calling the subscript operator on the id passed into the kernel function, with index 0.

Remember to name your kernel function, and to call wait on the event that is returned.

6.) Copy back the result

Once the kernel function has completed you can copy the result back to the host.

As you did when copying to the device, you can do this by calling the queue member function memcpy, again remember to call wait on the event that is returned.

7.) Free the device memory

Finally once you have copied the data back from the device you can free that memory.

To do this call free on each of the USM device allocations, note this is the SYCL API free and not the standard C free.

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_8

Alternatively from a terminal at the command line:

icpx -fsycl -o sycl-ex-8 -I../External/Catch2/single_include ../Code_Exercises/Exercise_08_USM_Vector_Add/source.cpp
./sycl-ex-8

In Intel DevCloud, to run computational applications, you will submit jobs to a queue for execution on compute nodes, especially some features like longer walltime and multi-node computation is only available through the job queue. Please refer to the guide.

So wrap the binary into a script job_submission and run:

qsub job_submission

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_8

alternatively, without CMake:

cd Code_Exercises/Exercise_08_USM_Vector_Add
/path/to/adaptivecpp/bin/acpp -o sycl-ex-8 -I../../External/Catch2/single_include --acpp-targets="<target specification>" source.cpp
./sycl-ex-8