CompilerSutra
CompilerSutra @CompilerSutra ยท
Most people think performance = instruction count. Not true. Same instructions. Same code. GCC is ~25% faster than Clang here. Reason? AGU pressure + scheduling. Compilers donโ€™t just generate code they decide how the CPU gets fed. godbolt.org/z/E4havxWcW #clang #gcc #llvm
1
27
v.sys
v.sys @v_sysbuild ยท
Replying to @v_sysbuild
Reading eth.c was useful. Seeing how Linux handles Ethernet at the production level made the spec feel less abstract. The gap between "I understand this" and "I can implement this" is where most of the time actually goes. #CLang #LowLevel
114
v.sys
v.sys @v_sysbuild ยท
Replying to @v_sysbuild
Also wrote recvip() - inverse of eval. Strip IP, move to ICMP, verify checksum, rebuild structs inward. eval() builds outward, recv() peels back. Same pattern. Next layer goes deeper than IP โ€” still uncovering what the OS was hiding. #LowLevel #LearningInPublic #CLang
297