Generate tile-centric GPU kernels for NVIDIA tensor core execution.
npx skills add https://github.com/jstzwj/ai-infra-plugins --skill tile-ir
Please help me install this Agent Skill. Skill: tile-ir Source: https://github.com/jstzwj/ai-infra-plugins/tree/main/plugins/tile-ir/skills/tile-ir Command: npx skills add https://github.com/jstzwj/ai-infra-plugins --skill tile-ir
This Skill streamlines GPU development by providing a dedicated instruction set tailored to tile-based processing, reducing complexity and enabling high-performance tensor computations.
Utilize the tile-ir skill to generate a GEMM kernel optimized for tensor core execution on an NVIDIA GPU.
High-intent search queries and answers about installing and using this skill.
To optimize matrix multiplication for NVIDIA tensor cores, you can use a tile-centric instruction set to directly program MMA operations and tensor view manipulations. This approach abstracts low-level architecture details for portable execution.
Tile-based GPU programming provides a dedicated instruction set tailored to tile processing, reducing complexity for high-performance tensor computations. You need it to accelerate deep learning workloads like convolutions and attention mechanisms.
You generate an optimized GEMM kernel by utilizing a tile-focused instruction set compatible with NVIDIA hardware. This streamlines GPU development by enabling efficient implementation of matrix operations across different GPU architectures.
Yes, tile-ir supports complex control flows and memory views across different GPU architectures. It abstracts low-level architectural details, allowing portable and future-proof GPU programming for deep learning workloads.
Yes, you need the tile-ir libraries and Python bindings to integrate tile operations with compiler and runtime systems. This environment setup enables efficient GPU programming and domain-specific optimizations.
The best way to abstract low-level GPU architecture details is using a tile-focused instruction set. It abstracts hardware specifics, allowing you to program portable and future-proof tensor operations without managing low-level code.
jwu1980-testAdd CUDA kernels to FlashInfer with JIT and TVM-FFI bindings.
tile-aiOptimize GEMV, GEMM, and grouped GEMM kernels on NVIDIA Ampere and Hopper GPUs.
tondevrelDevelop and train deep learning models using PyTorch with tensor operations and automatic differentiation.
ROCmGenerate and validate FlyDSL GPU kernels from kernel specifications using tile programming.
