Atiksh Sharma·3d agoThe Architecture of a Mistake We Just Learned to Live WithIn August 1173, work started on constructing a bell tower in Tuscany. It was expected to stand tall for many centuries. However, within 5…
ENERZAi·Oct 27Porting a Ternary STT Model to the Synaptics Torq NPUHello from ENERZAi! In our previous post, Journey to Torq Compiler…
ENERZAi·Oct 27Ternary STT 모델을 Synaptics Torq NPU에서 온디바이스 구동에너자이가 250M 크기의 ternary STT 모델을 Synaptics SL2619의 Torq NPU에서 실시간 구동한 과정을 소개합니다. 2-bit packed 가중치를 컴파일러 끝까지 유지한 방법을 다룹니
ENERZAi·Oct 20Optimizing the MLIR-Based Torq NPU CompilerHi, this is ENERZAi. We recently contributed to the backend implementation and model optimization of the Torq compiler, an open-source…
ENERZAi·Oct 20MLIR 기반 Torq NPU 컴파일러 개발 여정안녕하세요, 에너자이입니다. 에너자이는 최근 글로벌 반도체사인 Synaptics의 Astra SL2610 제품군에 탑재된 Torq NPU를 위해, 오픈소스 IREE/MLIR 기반 Torq 컴파일러의 백엔드 구현과 모델 최적화에 참여했는데요. 오늘은…
Harry, Hsiao·Sep 12Learning Rust #1 — Why Did --release Make My Program 50,000× Faster?I recently learning Rust with Google’s Comprehensive Rust course.A response icon3A response icon3
InGoPenAIbyshashank Jain·Sep 3AI as a Compiler Optimizer: How LLM Guided Search and MCTS Could Change Java CompilationCompilers have traditionally worked in a fairly deterministic way. You give them a program, they run a sequence of optimization passes, and…
Zahran Yahia Khan·Aug 22Why More SIMD and More Threads Made My C++ Engine SlowerWhat I learned optimizing a 500-million-user bitmap query engine to the DDR4 memory-bandwidth ceiling
Nader Mahbub Khan Nabil·Aug 12Designing Anastasia, a Bare-Metal, Zero-CRT JIT/AOT Compiler Engine in C++20How Anastasia Engine achieves >10 Billion op/s SIMD throughput with custom AVX-512 codegen, freestanding ELF relocations, and zero…
Sahil Bheke·Jul 30Inside the Go Compiler: Optimizations That Make Your Code FasterMost Go developers know how to write efficient Go code. But very few know what happens to that code after go build.
