Fix inline assembly #12798
Fix inline assembly #12798
Conversation
|
Can you rebase out the changes from #12797? |
|
@alexcrichton it looks like the asm changes use the changes from #12797. |
|
Done. I don't have more ideas other than refactoring. I.e. the body of state = match (p.token, next_state(state), next_state(next_state(state))) {
(token::COLON, StateNone, _) |
(token::MOD_SEP, _, StateNone) => {
p.bump();
break 'statement;
}
(token::COLON, st, _) |
(token::MOD_SEP, _, st) => {
p.bump();
st
}
(token::EOF, _) => break 'statement,
_ => state
} |
|
While you're at it, do you think you could write some documentation for this feature? There's not really a great place that I know of for the documentation, but docs in Also, could you edit the title of the PR to not have "WIP" in the name, and also squash the commits together? |
|
There is a chunk of docs https://github.com/mozilla/rust/wiki/Doc-detailed-release-notes#wiki-inline-assembly but not in a very useful/obvious place... and no code examples are tested. Maybe we could have a That said, I'd personally be happy with this happening in a different PR. |
| @@ -135,6 +142,10 @@ pub fn expand_asm(cx: &mut ExtCtxt, sp: Span, tts: &[ast::TokenTree]) | |||
| let (s, _str_style) = p.parse_str(); | |||
| let clob = format!("~\\{{}\\}", s); | |||
| clobs.push(clob); | |||
|
|
|||
| if s.equiv(&("volatile")) || s.equiv(&("alignstack")) || s.equiv(&("intel")) { | |||
There was a problem hiding this comment.
This can also be "volatile" == s || "alignstack" == s || "intel" == s, although maybe it would be even better (aka more extensible) as
static OPTIONS: &'static [&'static str] = &["volatile", "alignstack", "intel"];
// ...
if OPTIONS.iter().any(|opt| *opt == *s) { .. .}There was a problem hiding this comment.
@huonw Can I write "volatile" == option below?
There was a problem hiding this comment.
it's an InternedString
error: mismatched types: expected `&str` but found `parse::token::InternedString`
There was a problem hiding this comment.
Oh, my apologies, thought they were just ~strs. In that case, .equiv is good.
|
The problem is these examples can only be tested for x86 and x86_64. I will try to document this feature. GCC inline asm is extensively described here: http://locklessinc.com/articles/gcc_asm/ Should I squash all commits? |
|
I'm ok if you want to defer documentation for later, but yes, it would be nice to have the commits all squashed. |
Read+write modifier Some documentation in asm.rs rpass and cfail tests
|
Done |
## read+write modifier '+' This small sugar was left out in the original implementation (#5359). When an output operand with the '+' modifier is encountered, we store the index of that operand alongside the expression to create and append an input operand later. The following lines are equivalent: ``` asm!("" : "+m"(expr)); asm!("" : "=m"(expr) : "0"(expr)); ``` ## misplaced options and clobbers give a warning It's really annoying when a small typo might change behavior without any warning. ``` asm!("mov $1, $0" : "=r"(x) : "r"(8u) : "cc" , "volatile"); //~^ WARNING expected a clobber, but found an option ``` ## liveness Fixed incorrect order of propagation. Sometimes it caused spurious warnings in code: `warning: value assigned to `i` is never read, #[warn(dead_assignment)] on by default` ~~Note: Rebased on top of another PR. (uses other changes)~~ * [x] Implement read+write * [x] Warn about misplaced options * [x] Fix liveness (`dead_assignment` lint) * [x] Add all tests
… r=jonas-schievink internal: Remove outdated proc macro ABIs Drops support for 1.48 to 1.57. We still support 1.58+, which is 4 versions out of date, so that should be plenty.
…r=y21 Don't lint path statements in no_effect The rustc lint `path_statements` covers this case Fixes rust-lang#11547 changelog: none
read+write modifier '+'
This small sugar was left out in the original implementation (#5359).
When an output operand with the '+' modifier is encountered, we store the index of that operand alongside the expression to create and append an input operand later. The following lines are equivalent:
misplaced options and clobbers give a warning
It's really annoying when a small typo might change behavior without any warning.
liveness
Fixed incorrect order of propagation.
Sometimes it caused spurious warnings in code:
warning: value assigned toiis never read, #[warn(dead_assignment)] on by defaultNote: Rebased on top of another PR. (uses other changes)dead_assignmentlint)