Introdução à Programação em Triton: Um Tutorial Prático
Um tutorial científico abrangente projetado para oferecer um caminho completo de aprendizado sobre o Triton, uma linguagem baseada em Python e um compilador para escrever kernels GPU personalizados. O curso aborda modelos de programação, semântica da linguagem, comportamento numérico e otimização de desempenho, passando do cálculo básico de adição vetorial até operadores fundidos e segmentados usados em sistemas modernos de aprendizado profundo.
Visão Geral do Curso
📚 Resumo do Conteúdo
Um tutorial científico abrangente projetado para fornecer um caminho completo de aprendizado sobre Triton, uma linguagem baseada em Python e compilador para escrever kernels personalizados para GPU. O curso aborda modelos de programação, semântica da linguagem, comportamento numérico e otimização de desempenho, avançando desde a adição vetorial básica até operadores fundidos e segmentados usados em sistemas modernos de aprendizado profundo.
Domine a arte da engenharia de kernels de GPU de alto desempenho a partir dos princípios fundamentais.
Autor: EvoClass
Agradecimentos: Documentação do Triton e repositório GitHub do Triton.
🎯 Objetivos de Aprendizagem
- Definir o Triton e seu papel na pilha de software de aprendizado profundo.
- Distinguir o Triton do CUDA, código PyTorch em modo eager e assembly de baixo nível para GPU.
- Identificar quais cargas de trabalho são candidatas adequadas ao Triton e compreender a relevância da fusão de kernels e dos gargalos.
- Realizar uma instalação limpa do ambiente Triton e verificar a pilha de software.
- Implementar um kernel básico de cópia vetorial para validar a lógica do ambiente versus a lógica do kernel.
- Identificar e categorizar gargalos de GPU para justificar o uso da fusão de operadores do PyTorch.
- Definir uma instância de programa e calcular as dimensões de uma grade de lançamento 1D usando
cdiv. - Realizar aritmética de ponteiros para mapear IDs de programa específicos (
pid) para deslocamentos de memória. - Distinguir entre tensores do PyTorch (metadados do lado host) e tensores do Triton (blocos no nível do compilador).
- Calcular o mapeamento entre um ID de Programa (
pid) e deslocamentos de memória específicos usandotl.arange.
Aulas 共 10 课时 · 预计 30.0h
Aulas
Lesson
This lesson introduces Triton as a bridge between high-level Python productivity and low-level CUDA performance, focusing on its tile-centric programming model. Students will learn how Triton automates complex hardware tasks like memory management and synchronization to enable the development of high-performance custom kernels.
This lesson introduces Triton as a high-performance, block-based programming model that bridges the gap between high-level PyTorch and low-level CUDA. Students will learn to configure a GPU development environment and utilize Triton to overcome memory bottlenecks by fusing operations and optimizing data movement within the GPU's SRAM.
AI023: The Triton Programming Model: Grids and Pointers (Lesson 3) introduces the block-based parallel paradigm, contrasting Triton’s efficient tile-level processing with the overhead of PyTorch’s eager execution. Students will learn to manage memory through pointer arithmetic and coordinate systems, enabling them to optimize GPU performance by minimizing global memory round-trips.
This lesson introduces the Triton programming model, focusing on the transition from scalar CUDA threads to vectorized program instances that operate on data blocks. Students will learn how to utilize program IDs (pid) for SPMD execution, manage memory offsets, and implement masking to handle data boundaries effectively.
This lesson introduces the parallel execution model for GPU programming, focusing on how to implement a vector addition kernel using block-based execution. Students will learn to identify performance bottlenecks—specifically memory-bound versus compute-bound operations—and optimize hardware utilization by managing occupancy and block size.
This lesson explores the Performance Paradox in Triton programming, explaining how fixed GPU launch overheads can make functionally correct code inefficient for small workloads. Students will learn to distinguish between latency-bound and throughput-bound operations, identify the importance of asynchronous execution in benchmarking, and apply strategies like workload batching to minimize the impact of the launch tax.
AI023: Introduction to Triton Programming — Beyond 1D: Why 2D Layout Awareness Matters (Lesson 7) This lesson explores how transitioning from 1D elementwise processing to 2D tiled grids allows Triton kernels to maximize spatial locality and hardware efficiency. Students learn to implement layout-aware kernels by utilizing strides and broadcasting to process data blocks, which is essential for high-performance operations like matrix multiplication.
This lesson explores reduction operations in Triton, focusing on how to collapse multi-dimensional tensors while managing memory layouts and hardware-level data dependencies. Students will also learn to implement numerically stable Softmax functions by addressing common floating-point issues like overflow and underflow.
This lesson explores the transition from memory-bound elementwise operations to compute-bound tiled matrix multiplication (GEMM) in Triton. Students will learn to optimize LLM performance by implementing 2D tiling, managing tensor strides to avoid memory access errors, and applying operator fusion to reduce global memory overhead.
This lesson explores the systematic optimization lifecycle for Triton kernels, focusing on the transition from functional correctness to hardware-aware performance. Students will learn to utilize debugging tools like the Triton interpreter, establish strong performance baselines, and implement autotuning strategies to maximize hardware utilization.