There are a few optimizations to be made:

• user shaderc or spirv_tools optimizations
• use for loop instead of unrolled for loop
• store 4 bytes in each int instead of 1 and unpack at runtime