Hacker News Reader

Comment by lomnakkus | Hacker News Reader

> A good compiler should be able to set all the values with one load operations.

Hm? I believe that might actually be entirely wrong for many pieces of actual hardware that I've worked with. Setting a single bit may change the interpretation of a subsequently set bit, so you can't really "batch" the updates.

(I'm happy for you if this isn't an issue you've faced. We did. Feel sorry for us.)

View on HN

That's what the keyword volatile is for. If something is not marked volatile the compiler is allowed to change the order and collapse any memory accesses it deems unnecessary.

Indeed, but I don't think that was the poster I was responding to was talking about?

I mean, he/she was talking about bitfields and similar things and how writes to those could be coalesced? In genreal, they cannot. (I dunno, it seemed appropriate to mention the caveats.)

Not coalescing writes on non-volatile bitfields probably isn't guaranteed by the standard. If you depend on it I hope you have a test catching that.

Why use bit fields at all instead of straightforward bit math then? They're not really convenient and have weird alignment rules already.

This is totally right and shouldn't be downvoted, but the reply is also right that volatile should prevent optimization.

On some hardware, "writing" to "memory" is really sending a message to a device, and writing 0x80 followed by 0x08 is not the same thing as writing 0x88.

What we did was load the hardware register word to a local variable, set the appropriate bits then store it back.

Reply on news.ycombinator.com