tech.compute.tensor — techascent/tech.compute 4.25-1

Functions for dealing with tensors with the compute system

->array clj

(->array tens & [datatype])

->double-array clj

(->double-array tens)

->float-array clj

(->float-array tens)

->tensor clj

(->tensor data & {:keys [datatype device stream sync?] :as options})

clone-to-device clj

(clone-to-device input-tens)

(clone-to-device input-tens options)

Clone a host tensor to a device. Tensor must have relatively straighforward dimensions (transpose OK but arbitrary reorder or offset not OK) or :force? must be specified. options: :force? Copy tensor to another buffer if necessary. :sync? Sync with stream to ensure copy operation is finished before moving forward.

clone-to-host clj

(clone-to-host device-tens)

(clone-to-host device-tens options)

Copy this tensor to the host. Synchronized by default.

columns clj

(columns tens)

ensure-device clj

(ensure-device input-tens)

(ensure-device input-tens options)

Ensure a tensor can be used on a device. Some devices can use CPU tensors.

ensure-host clj

(ensure-host device-tens)

(ensure-host device-tens options)

Ensure this tensor is a 'host' tensor. Synchronized by default.

new-host-tensor clj

(new-host-tensor shape)

(new-host-tensor shape options)

new-tensor clj

(new-tensor shape)

(new-tensor shape options)

rows clj

(rows tens)