Doca Dma

Guidance for tasks where the user is doing hands-on DOCA DMA programming — bringing up a docadma context, configuring the single docadmataskmemcpy task type, sizing buffers via the docadmacaptaskmemcpy queries, setting LOCALREADONLY /…

About this skill

Guidance for tasks where the user is doing hands-on DOCA DMA programming — bringing up a docadma context, configuring the single docadmataskmemcpy task type, sizing buffers via the docadmacaptaskmemcpy queries, setting LOCALREADONLY / LOCALREADWRITE permissions on source / destination docammap regions (plus docammapexport for cross-peer copies), driving the progress engine, or debugging DOCAERROR returns. Trigger even when the user does not explicitly mention "DOCA DMA" or "docammap" — typical implicit phrasings include "memcpy host buffer to BlueField without using the CPU", "offload a bulk…

Maintained by NVIDIA. The source includes the instructions and any supporting files needed to use this skill.

Inside the instructions

  • 01Example questions this skill answers well
  • 02Audience
  • 03When to load this skill
  • 04What this skill provides
  • 05What this skill deliberately does not ship
  • 06Loading order

Before you start

  1. Read the instructions and check tool or account requirements.
  2. Install the complete folder when the skill references scripts or other files.
  3. Provide your task context, then review the agent's output.

Source

nvidia/skills / doca-dma

Source reviewed October 2, 2026 · Apache-2.0 / CC-BY-4.0; see source notices