Name		Name	Last commit message	Last commit date
parent directory ..
.gitignore		.gitignore
CMakeLists.txt		CMakeLists.txt
Makefile		Makefile
README.md		README.md
device_radix_sort_vs2017.sln		device_radix_sort_vs2017.sln
device_radix_sort_vs2017.vcxproj		device_radix_sort_vs2017.vcxproj
device_radix_sort_vs2017.vcxproj.filters		device_radix_sort_vs2017.vcxproj.filters
device_radix_sort_vs2019.sln		device_radix_sort_vs2019.sln
device_radix_sort_vs2019.vcxproj		device_radix_sort_vs2019.vcxproj
device_radix_sort_vs2019.vcxproj.filters		device_radix_sort_vs2019.vcxproj.filters
device_radix_sort_vs2022.sln		device_radix_sort_vs2022.sln
device_radix_sort_vs2022.vcxproj		device_radix_sort_vs2022.vcxproj
device_radix_sort_vs2022.vcxproj.filters		device_radix_sort_vs2022.vcxproj.filters
main.hip		main.hip

README.md

hipCUB Device Radix Sort Example

This simple program showcases the usage of the hipcub::DeviceRadixSort::SortPairs function.

Host side data is instantiated in std::vector<float> and std::vector<int> as key-value pairs.
Device side storage is allocated using hipMalloc.
Data is copied from host to device using hipMemcpy.
hipCUB::DeviceRadixSort::SortPairs makes use of temporary memory on the device that needs to be allocated manually. A first call to hipcub::DeviceRadixSort::SortPairs (with d_temp_storage set to null) calculates the size in bytes needed for the temporary storage and stores it in the variable temp_storage_bytes.
temp_storage_bytes is used to allocate device memory in d_temp_storage using hipMalloc.
Finally a second call to hipcub::DeviceRadixSort::SortPairs is invoked that sorts the pairs on the basis of keys.
Result is transferred from device to host.
Free all device side memory using hipFree

The device-level API provided by hipCUB is used in this example. It performs global device level operations (in this case pair sorting using hipcub::DeviceRadixSort::SortPairs) on the GPU.