Profile-guided optimization

Profile-guided optimization (PGO), also known as profile-directed feedback (PDF), and feedback-directed optimization (FDO) is a compiler optimization technique in computer programming that uses profiling to improve program runtime performance.

GCC PGO example

Build instrumented binary:

gcc foo.c -o foo_instrumented -fprofile-generate

Run instrumented binary with a representative workload:

./foo_instrumented

Then build optimized binary using generated profile:

gcc -O3 foo.c -o foo -fprofile-use=foo.gcda

Clang PGO example

Build instrumented binary:

clang foo.c -o foo_instrumented -fprofile-generate

or

clang foo.c -o foo_instrumented -fprofile-instr-generate

Run instrumented binary with a representative workload:

./foo_instrumented

Then build optimized binary using generated profile:

llvm-profdata merge -ouput foo.profdata default.profraw
clang -O3 foo.c -o foo -fprofile-use=foo.profdata

Links

• https://clang.llvm.org/docs/UsersManual.html#profile-guided-optimization
• https://gcc.gnu.org/wiki/AutoFDO/Tutorial
• https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html
• https://github.com/google/llvm-propeller
• https://github.com/facebookincubator/BOLT
• https://github.com/google/autofdo