> > Please rearrange this like: > > > > for (i = 0; i < le32_to_cpu(gpt->num_partition_entries) && > > i < state->limit-1; i++) > > Better still cache le32_to_cpu(gpt->num_partition_entries) to a local > so that it only has to be calculated once (even if it gets spilled > to the stack). > It can be bounded by state->limit-1 at the same time. This is block I/O, slow path. No need for such micro optimizations here. GCC is capable to do such optimizations automatically, and it does as I can see from disassembly. Anyway this would be a subject for another patch. -- Eugene
