Archives
All the articles I've archived.
2026 21
October 1
September 1
August 2
June 5
-
How to Write Layout Correctly: A Deep Dive into TileLang Layout Inference
-
[TIL] Collected: RoPE in TileLang
-
[TIL] Understanding IM2COL
-
[TIL] GPU Shared Memory Swizzle
-
[TIL] Tilelang Element-wise Op & Vectorization
Notes on three TileLang vector-add kernels, how vectorization appears in IR/CUDA/PTX, and why one-element-per-thread leaves bandwidth on the table.
May 11
-
[Book Club] FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness
Paper reading notes on FlashAttention, online softmax, IO complexity, and block-sparse attention.
-
[TIL] Multihead Attention
-
[Book Club] TileLang: Bridge Programmability And Performance In Modern Neural Kernels
-
[TIL] Collected: Convolution
-
[TIL] TileLang EP-01: Initial glance
Very brief glimpse through how the two system express GPU programming differently by simple(if not silly) examples
-
[TIL] Collected: Softmax Attention
-
[TIL] Collected: Softmax
-
[TIL] Collected: Transpose, Reduction
-
[TIL] Triton Kernel to LLVM IR Walkthrough
-
[TIL] A small note on layout computation
-
[TIL] Block exchange
2025 7
June 7
-
[TIL] MLIR Toy Explained Ch 7
Adding a Composite Type
-
[TIL] MLIR Toy Explained Ch 6
Lowering
-
[TIL] MLIR Toy Explained Ch 5
Progressive lowering
-
[TIL] MLIR Toy Explained Ch 4
Interfaces
-
[TIL] MLIR Toy Explained Ch 3
Analysis & Transformation
-
[TIL] MLIR Toy Explained Ch 2
Emitting MLIR
-
[TIL] MLIR Toy Explained Ch 1
Lexer, parser and AST.