Get the latest tech news

Integer Overflow Checking Cost

How much overhead should we expect from enabling integer overflow checks? Using a compiler flag or built-in intrinsics, we should be able to do the check with a conditional branch that branches based on the overflow flag that add and sub set. Code that looks like add %esi, %edi should turn into something like add %esi, %edi jo <handle_overflow> Assuming that branch is always correctly predicted (which should be the case for most code), the costs of the branch are the cost of executing that correctly predicted not-taken branch, the pollution the branch causes in the branch history table, and the cost of decoding the branch (on x86, jo and jno don't fuse with add or sub, which means that on the fast path, the branch will take up one of the 4 opcodes that can come from the decoded instruction cache per cycle).

None

Get the Android app