Tuning by the GCC PGO

Purpose

Profile-guided optimization (PGO), also known as feedback-directed optimization, is a compiler optimization technology. PGO is a dynamic optimization process where a pre-test is required. The pre-test samples a program, obtains features such as hotspot functions and function distribution of the program, and feeds sample data to the next compilation. Based on the sample data, the compiler re-compiles the program to generate an optimized one.

Procedure

1. Compile Redis.

   make CFLAGS="-fprofile-generate=/opt/gcc_profile" LDFLAGS="-fprofile-generate=/opt/gcc_profile" -j

2. Start Redis. For details, see the Starting Redis section in the Redis Deployment Guide.
3. Test Redis on the Redis client and obtain the sampling result.
4. Stop Redis.

   redis-cli -h 192.168.222.124 -p 6379 shutdown

   

   192.168.222.124 and 6379 are the IP address and port number of the Redis server. Change them according to the actual configurations.

5. Clean up build results for recompilation.

   make distclean

   

   Before recompiling with the collected data, clean up the previous build results to avoid interference.

6. Recompile Redis.

   make CFLAGS="-fprofile-use=/opt/gcc_profile -fprofile-correction -Wno-error=coverage-mismatch" LDFLAGS="-fprofile-use=/opt/gcc_profile -fprofile-correction -Wno-error=coverage-mismatch" -j

   

   /opt/gcc_profile is the directory for storing sampling files, which can be customized. If GCC PGO needs to be performed again, delete the data from the /opt/gcc_profile directory to avoid confusion.