Making Most Out of Compiler

Points

• Don’t blame compilers for being slow. They will note generate the best possible code for you.
• There is a chance your code will grow a lot and you won’t notice
• More visibility to your compiler opens more optimizations.
• Use -O0 -g for debugging purposes, use -O3 for optimization purposes.
• Compilers assume really old hardware (e.g. core2 for x86, 2006), use better options.
• -march=native is extremely dangerous because of illegal instructions and non-homogeneous hardware

LTO

• Link time optimizations
• Thin LTO provides summaries of modules
• -flto , -flto=thin (LLVM)
• Use thin LTO for release builds, usual without

Faster than O0

• -O1
• -fno-omit-frame-pointer
• -fno-optimize-sibling-calls
• -Og -> for GCC only

X86 Westmere era (2010)

• -msse2, -msse3, -mssse3, -msse4.1, -msse4.1
• -mpclmul
• -maes
• -mcx16
• -mpopcnt

X86 Haswell era (2013)

• -mavx, -mavx2
• -mbmi2
• -mfma

Why not avx/avx2/avx512 by default?

• Share AVX with non AVX -> CPU downclocking
• Heavy AVX (floating point) -> CPU downclocking

Inline threshold for LLVM

• -mllvm -inline-threshold=N (default 225)
• Inclininig increases your build size and compile time.
• Idea: run -mllvm -inline-threshold=1000 , compare benchmarks, set always_inline attributes to a dozen of calls

Reference

• Making the Most Out of Your Compiler - Danila Kutenin - CppCon 2021
• LLVM Vectorizing debug
• GCC Options That Control Optimization
• ThinLTO
• GCC x86 Options