Introducción a la programación con Triton: Una guía práctica
Una tutorial científico completo diseñado para proporcionar una ruta completa de aprendizaje para Triton, un lenguaje basado en Python y un compilador para escribir kernels personalizados para GPU. El curso cubre modelos de programación, semántica del lenguaje, comportamiento numérico y optimización del rendimiento, avanzando desde la suma vectorial básica hasta operadores fusionados y segmentados utilizados en sistemas modernos de aprendizaje profundo.
Descripción del curso
📚 Resumen del Contenido
Una tutorial científico completo diseñado para proporcionar una ruta de aprendizaje completa para Triton, un lenguaje y compilador basado en Python para escribir kernels personalizados para GPU. El curso cubre modelos de programación, semántica del lenguaje, comportamiento numérico y optimización de rendimiento, avanzando desde la suma básica de vectores hasta operadores fusionados y segmentados utilizados en sistemas modernos de aprendizaje profundo.
Domina el arte de la ingeniería de kernels de GPU de alto rendimiento desde los principios fundamentales.
Autor: EvoClass
Agradecimientos: Documentación de Triton y repositorio de GitHub de Triton.
🎯 Objetivos de Aprendizaje
- Definir Triton y su papel en la pila de software de aprendizaje profundo.
- Distinguir Triton de CUDA, código eager de PyTorch y ensamblaje de bajo nivel de GPU.
- Identificar qué cargas de trabajo son candidatas adecuadas para Triton y comprender la relevancia de la fusión de kernels y cuellos de botella.
- Realizar una instalación limpia del entorno Triton y verificar la pila de software.
- Implementar un kernel básico de copia de vector para validar la lógica del entorno frente a la lógica del kernel.
- Identificar y categorizar cuellos de botella de GPU para justificar el uso de fusión de operadores de PyTorch.
- Definir una instancia de programa y calcular las dimensiones de una rejilla de lanzamiento 1D usando
cdiv. - Realizar aritmética de punteros para mapear IDs de programa específicos (
pid) a desplazamientos de memoria. - Distinguir entre tensores de PyTorch (metadatos del lado host) y tensores de Triton (bloques a nivel de compilador).
- Calcular el mapeo entre un ID de programa (
pid) y desplazamientos de memoria específicos usandotl.arange.
Lecciones 共 10 课时 · 预计 30.0h
Lecciones
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.