Topics / Specialization, templates and JIT

Performance topic

Specialization, templates and JIT

Generate code for the case at hand: monomorphized templates, const generics, specialized fast paths and run-time code generation.

1,716proven techniques
77pitfalls: tried, then slower
683repositories
275from papers

Examples from real commits

tinygo-org/tinygo 25.7× Generate type-specific hash and equal functions as LLVM IR that walk struct fields and array elements, replacing the interface plus reflection path. proof: commit → servo/webrender 9.4× Use a single 2D signed distance function for uniform-radius rounded rectangles to reduce shader ALU ops. proof: commit → DynamoRIO/dynamorio 2.3× Use DC ZVA instruction to zero memory in hardware-aligned blocks instead of byte stores. proof: commit →

Languages

  • C++674
  • C249
  • Python117
  • CUDA110
  • Rust106
  • Go31

Hardware

  • CPU423
  • x86287
  • NVIDIA GPU281
  • GPU160
  • ARM120

Fields

  • compilers285
  • GPU280
  • general278
  • data & compression269
  • graphics245

Where these come from

Repositories with the most techniques on this topic in the database:

WebKit/WebKit · 12LuaJIT/LuaJIT · 12postgres/postgres · 9NVIDIA/cccl · 8NVIDIA/TensorRT-LLM · 8CliMA/Oceananigans.jl · 8oscar-system/Oscar.jl · 8flashinfer-ai/flashinfer · 8NVIDIA/warp · 8xtensor-stack/xsimd · 8

Ask Claude for the full answer

Each technique comes with its bottleneck, measured speedup, trade-offs, conditions and the commit or paper behind it, served to Claude through MCP.

claude mcp add --transport http adtomic https://www.adtomic.co/mcp

Connect Claude Pricing

Related topics

Reducing synchronizationPipelining and async overlapBit packing and smaller typesLazy evaluationZero-copy