Boost workload performance and get more done optimizing native code
summary
Uses the WindowsPerf Visual Studio extension and an Windows Performance Analyzer plugin plus Arm telemetry to run top‑down hotspot analysis on a C Python native Arm64 workload that computes 10^100 (Googleplex) for ~10s. Counting shows load/store speculation dominates (~35% of speculative work); sampling maps ~79% of ld_spec samples to the Python DLL multiply routine. Tooling uses multiplexed counters now, plans RMSP hardware sampling and ETW integration; disassembly and source correlation confirm actionable hotspots.
AI-written summary · about
chapters (9)
- 0:00Optimizing native code
- 1:14WindowsPerf introduction
- 3:02Top-down analysis on CPython
- 4:49CPU front-end/back-end bottlenecks
- 6:54Load/store speculation metrics
- 9:21Hotspot sampling in source
- 11:51VS performance insights
- 12:57Windows Performance Analyzer
- 13:39Key takeaways and future
related talks
45mPerf tools updates and beyond
01h 02mProfiling with Heaptrack and Hotspot on Embedded Linux
48mProfiling data structures
45mPerformance lab. The power of the perf tool!
41mLinux Systems Performance
54mx86 Internals for Fun & Profit
51mAccelerating performance of Java applications on Arm64
28mLevel Up Your Game: OS Kernel and Game Interactions revealed with Perfetto