Don't add .intel_syntax on non-Intel machines

[kornelski]

--mca doesn't work on aarch64 macOS:

<stdin>:1:1: error: unknown directive
.intel_syntax
^

This PR disables the directive on other targets.

Perhaps a better solution would be to add another variant to OutputStyle instead, but the command-line parser is too clever to modify.

[pacak]

Change looks reasonable. I'd like to poke around a bit to see if it makes sense to detect target triple and make it available to more code - there's at least one ticket that depends on this information (#284), if it's too much effort then I'll merge this. Appreciate the report ❤️

Perhaps a better solution would be to add another variant to OutputStyle instead, but the command-line parser is too clever to modify.

If adding a variant is all you want then just adding it with a doc comment and fixing complaints from rustc should do the trick... Not sure if it's a better solution though - I'd expect stuff like that just to work...

[kornelski]

I assumed you'd want OutputStyle to default to Intel on x86, and some Other value on other platforms, but I don't see how to do that in the parser given that it may depend on another target option. Getters on OutputStyle also don't have the context.

[pacak]

After some poking around - I think I found a better approach. If original file contains .intel_syntax - pass it though, otherwise do nothing. This way rustc/llvm deals with the difference. Will try to implement it in a day or two.

[pacak]

I implemented my approach in #378, fixes mca for aarch for me, at least using crosscompilation:

cargo run --release -- --manifest-path sample/Cargo.toml  --target aarch64-unknown-linux-gnu -M=-mcpu=cyclone --lib 8 --mca

I still have to specify -mcpu because otherwise it defaults to my host cpu and fails. Don't see anything obvious in the .s file to pass though.