On 17/10/17 13:29, Segher Boessenkool wrote: > On Tue, Oct 17, 2017 at 12:49:57PM +0200, David Brown wrote: >> On 17/10/17 10:12, Segher Boessenkool wrote: >> I don't have a build of trunk gcc conveniently available (I am building >> a gcc 7 at the moment). If this is a bug, then perhaps it is fixed >> since the latest version I tried (6.3.0). It is quite easy to test >> there using the marvellous godbolt online compiler: > > While godbolt is a very nice tool for exploring code generation by > various compilers, it is not handy at all for debugging GCC bugs (if > you manage to do that at all with it). Agreed. But I am still not entirely sure if this is a bug or part of the design of asm volatile, and godbolt is a convenient way to test the code on a variety of compiler versions. I have not been involved in gcc development or debugging (maybe I should be). > >> The gcc documentation specifically says that asm volatile statements >> /can/ be moved. It just does not make it clear what kind of movements >> are allowed. > > It doesn't have to. It is just warning you not to assume it will not > move code here. Anything that is not disallowed is allowed. > gcc would be more useful for the kind of code in discussion here if there were a clearer distinction. I am afraid I am getting conflicting messages from you about this - or I am misunderstanding something. On the one hand, you have stated that movement of "asm volatile" statements with respect to each other is a bug - on the other hand, you don't want the documentation to say that this is disallowed. >> Is the patch below something you added to the trunk to fix this, or is >> it something the trunk already had? > > Neither. I tested it on GCC 5; it should still apply to trunk as well, > since the code there hasn't changed as far as I see. > That is great, because it means there is an easy back-port to any relevant version of gcc (many embedded development toolkits are a number of major versions behind trunk). > > Segher >
