Topics / Batching

Performance topic

Batching

Group many small operations into one: batched kernel launches, bulk inserts, vectored syscalls, amortized locking.

1,931proven techniques
69pitfalls: tried, then slower
809repositories
179from papers

Examples from real commits

v8/v8 4× Accumulate per-chunk live bytes in a fixed 128-entry direct-mapped local array; flush the evicted entry to the atomic page counter on collision and the rest at finalize. proof: commit → cfregly/ai-performance-engineering 1.5× Widen the TMEM load atom from 32dp32b1x to 32dp32b32x so each tcgen05.ld moves 32 columns, cutting loads per thread from 256 to 8. proof: commit →

Languages

  • C++705
  • C329
  • Python242
  • Rust197
  • CUDA85
  • Go74

Hardware

  • CPU325
  • NVIDIA GPU267
  • GPU177
  • x86152
  • ARM55

Fields

  • general523
  • data & compression500
  • GPU283
  • graphics223
  • networking167

Where these come from

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

FalkorDB/FalkorDB · 10apache/datafusion · 10timescale/timescaledb · 9torvalds/linux · 9thu-pacman/chitu · 8facebookincubator/velox · 8audacity/audacity · 8blender/blender · 8tenstorrent/tt-metal · 8vllm-project/vllm-omni · 7

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

Specialization, templates and JITReducing synchronizationPipelining and async overlapBit packing and smaller typesLazy evaluation