Why careless struct member ordering turns a 16-byte payload into 32 bytes due to word alignment, doubling L1 cache line misses

Why careless struct member ordering turns a 16-byte payload into 32 bytes due to word alignment, doubling L1 cache line misses

In my codebase, I noticed that struct padding was causing a 16-byte payload to expand to 32 bytes, leading to L1 cache line misses and performance degradation.

This issue arises from the way C/C++ compilers align data to word boundaries, which can lead to padding bytes being inserted between struct members. To mitigate this, I used a combination of struct packing and bit-field manipulation to minimize padding and optimize memory usage.

For example, consider the following struct:

The compiler will align the int member b to a 4-byte boundary, resulting in 4 bytes of padding between a and b. This will cause the struct to occupy 12 bytes instead of 6.

To fix this, I used the __attribute__((packed)) compiler directive to pack the struct tightly:

This will eliminate the padding and reduce the struct size to 6 bytes.

Another approach is to use bit-field manipulation to reduce the size of the int member:

By packing the int member, we can reduce its size to 2 bytes, resulting in a total struct size of 6 bytes.

By understanding the intricacies of struct alignment and padding, developers can write more efficient and effective C/C++ code.

Drishtant Ghosh
Follow for daily systems engineering & code teardowns.

🔗 Reference & Source Breakdown

• Source Material: Tech Infrastructure: 🤖 AI Infrastructure - Cloud Cost Reality Check ↗
• Recommended Visual Asset: Real-world visual artifact: Clean dark-mode terminal screenshot of code from Drix10/Grind running or compiling.
• First Comment: Check out the code & architecture on GitHub → https://github.com/Drix10/Grind ↗
  Personal blog & deep-dives: https://blogs.drix10.com
• Syndicated Channel: LinkedIn & Personal Blog Hub