Ctrl+K
This page contains proposed changes for a future release of ROCm. Read the latest Linux release of ROCm documentation for your production environments.

Data type support

Data type support#

hipCUB supports the following data types on both ROCm and CUDA:

  • int8

  • int16

  • int32

  • float32

  • float64

float8, bfloat8, and tensorfloat32 are not supported by hipCUB on neither ROCm nor CUDA.

The NVIDIA back end does not support float16 nor bfloat16 with the following API calls: block_adjacent_difference, device_adjacent_difference, device_reduce, device_scan, device_segmented_reduce and device_select.

The NVIDIA backend also does not support bfloat16 with device_histogram.