Skip to content

Memcpy Sync Task

The Memcpy sync task performs a raw byte-for-byte copy of its single input tensor into the memory space selected by the target setting. It preserves the tensor's shape and dtype and does not apply any numerical transformation.

Depending on the source and target locations, the task chooses an appropriate transfer path:

  • Host -> Host copies use a small OpenMP-based worker pool to split the buffer and accelerate CPU transfers.
  • Host <-> Device and Device -> Device copies run through cudaMemcpyAsync on the task's CUDA stream, allowing the copy to overlap with other GPU work scheduled on the same stream.

Inputs

This task has a single input tensor of arbitrary shape, dtype, and memory location.

Outputs

This task has a single output tensor whose shape and dtype exactly match the input tensor. The output memory location is determined by the target setting:

  • Host: the tensor is produced in host memory.
  • Device: the tensor is produced in device memory.

Inplace

This task does not support inplace operation.

Ownership

This task does not own any inputs or outputs.

Settings

Example:

1
2
3
{
    "target": "Device"
}
Property Pattern Type Deprecated Definition Title/Description
+ target No enum (of string) No - Destination memory space for the copy. Accepted values:
- Host: Copy into CPU memory
- Device: Copy into GPU memory