05Performance

Performance

AdvancedModule 055 topics~4h 40m

Topics, in order

01 Benchmarking and ProfilingA benchmark answers "how fast is this function?" by running it enough times to get a stable number. A profile answers "where does the time (or memory) go in the whole program?" by sampling … Advanced 1h 02 The CompilerThe Go compiler favours fast builds over aggressive optimization, but it does a specific set of things well, and knowing them tells you which code shapes are fast. Inlining replaces a call … Advanced 55 min 03 Cache-Friendly DataA CPU can execute several instructions per nanosecond, but fetching a value from main memory takes around a hundred nanoseconds. Between the two sit caches: small, fast copies of recently … Advanced 55 min 04 cgo, SIMD and AssemblyPure Go compiles to scalar code: one number per instruction. Numerical work — and AI is mostly numerical work — wants SIMD: instructions that process 4, 8 or 16 numbers at once. Go gives you … Advanced 55 min 05 Networking and the NetpollerIn Go you write network code that looks blocking — conn.Read(buf) simply waits — and it is cheap anyway. Under the surface every socket is non-blocking and registered with the operating … Advanced 55 min

About this module

Modules III and IV explained what the runtime does. This module is about measuring a program, understanding what the compiler and the CPU did with it, and making it faster on purpose.

#	Lesson	The question it answers
01	Benchmarking and Profiling	How do I measure correctly, and find where the time and memory go?
02	The Compiler	What do inlining, bounds-check elimination and PGO do to my code?
03	Cache-Friendly Data	Why does data layout matter more than instruction count?
04	cgo, SIMD and Assembly	How do I reach native code and vector instructions, and what does it cost?
05	Networking and the Netpoller	How does a Go server handle connections, and how do I stream a response?

When you finish you can write a trustworthy benchmark, read a CPU and heap profile, explain a speed-up in terms of cache lines and inlining, and build a streaming HTTP server that behaves under load. Module VI uses every one of these.