skills.rest
ROCm
ROCmOFFICIAL
Explore all skills by this author →
WebsiteGitHub

port-to-layout-api

Port FlyDSL GPU kernels from raw buffer_ops to the layout API.

260|104|Updated Nov 23, 2025
Search Tags:#port#flydsl#gpu-kernel#layout-api#buffer-ops
One-click install
npx skills add https://github.com/ROCm/FlyDSL --skill port-to-layout-api
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: port-to-layout-api
Source: https://github.com/ROCm/FlyDSL/tree/main/.claude/skills/port-to-layout-api
Command: npx skills add https://github.com/ROCm/FlyDSL --skill port-to-layout-api

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Port raw buffer_ops to the higher-level layout API, enabling more readable, maintainable, and optimizer-friendly kernels.

Core Features & Use Cases

  • Porting raw buffer_ops like create_buffer_resource, buffer_load, and buffer_store to fx.rocdl.make_buffer_tensor and BufferCopy-based flows.
  • Handling multi-dimensional tensors with slices, logical divisions, and copy atoms to support 2D/3D layouts.
  • Encouraging safer, more consistent kernel porting with masking guidance and OOB guards.

Quick Start

Run the port-to-layout-api tool on a kernel file to migrate raw buffer_ops usage to the layout API.

Frequently Asked Questions about port-to-layout-api

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I migrate FlyDSL GPU kernels from raw buffer_ops to the layout API?▼

To migrate FlyDSL GPU kernels from raw buffer_ops to the layout API, port manual byte-offset arithmetic to fx.rocdl.make_buffer_tensor, slice, and BufferCopy atoms. This enables more readable, maintainable, and optimizer-friendly kernel code.

What is the best way to handle non-contiguous accesses when porting to the FlyDSL layout API?▼

Handling non-contiguous accesses during FlyDSL layout API porting involves using slices and logical divisions. The process guides migration to make_buffer_tensor and copy atoms, specifically supporting 2D and 3D tensor layouts.

How do I replace create_buffer_resource and buffer_load with the FlyDSL layout API?▼

Replace raw buffer_ops like create_buffer_resource, buffer_load, and buffer_store with fx.rocdl.make_buffer_tensor and BufferCopy-based flows. This transition shifts raw buffer operations to higher-level layout abstractions.

Does the FlyDSL layout API provide guidance for masking and OOB guards during buffer_ops porting?▼

Yes, porting buffer_ops to the FlyDSL layout API includes explicit masking guidance and OOB guards. This ensures safer, more consistent kernel porting when replacing manual byte-offset arithmetic with slice operations.

Can I use BufferCopy atoms for 2D and 3D tensor layouts in FlyDSL?▼

Yes, you can use fx.make_copy_atom and BufferCopy variants for 2D and 3D tensor layouts in FlyDSL. The layout API handles multi-dimensional tensors via slices and logical divisions to support complex layouts.

Related Skills

Explore More Software Engineering Skills→
ROCmROCm
260

debug-flydsl-kernel

Diagnose incorrect FlyDSL GPU kernel outputs, NaNs, infs, and crashes.

Official
Advanced
ROCmROCm
260

FlyDSL Tile Programming

Generate and validate FlyDSL GPU kernels from kernel specifications using tile programming.

Official
Advanced
nanorepublicananorepublica
61

Global Tech Stack

Validate code against centralized global tech stack standards.

Community
Basic
UtsavBalar1231UtsavBalar1231
1

linux-kernel-pro

Assist Linux kernel development for device drivers, modules, and subsystems.

Community
Advanced