add Intel syntax alternatives to all gcc inline assembly

[ecm-pushbx]

Based on #225 I developed the patch in https://pushbx.org/ecm/test/20260422-generate-listings.diff (AT&T syntax) and https://pushbx.org/ecm/test/20260422-generate-listings-intel.diff (Intel syntax) to generate assembler listing files from the .c files.

This commit includes only the changes to inline assembly, as these shouldn't break anything for other users or other compilers than ia16-elf-gcc. As it turns out, gcc extended inline assembly allows to specify two variants for any snippet, one for AT&T syntax and the other for Intel syntax.

Previously discussed in https://codeberg.org/tkchia/gcc-ia16/issues/9#issuecomment-13551167

I verified that both sys.com and kgc8632.sys are byte by byte exactly identicalised matches before and after applying this commit.

[andrewbird]

So this is interesting, why would GCC allow this and what if the two sides aren't equivalent?

[ecm-pushbx]

In my case I make use of it so I can use -masm=intel despite the whole kernel not being updated to use (only) this variant. As for if they're not equivalent I'd hazard a guess that's user error.

[andrewbird]

The use of ATT syntax is only in the GCC inline, or did I miss some other example in the kernel? It seems to me that if we switched to Intel syntax here it would be more in keeping with the rest of the assembly source in the kernel. Would anyone object to that?

[ecm-pushbx]

The use of ATT syntax is only in the GCC inline, or did I miss some other example in the kernel?

If you pass -masm=intel then this switches the entire kernel (the C code) to be built as Intel syntax, whereas otherwise it uses AT&T syntax. It is only visible with -S or the listing file options.

It seems to me that if we switched to Intel syntax here it would be more in keeping with the rest of the assembly source in the kernel. Would anyone object to that?

That would require editing all makefiles to always pass -masm=intel

[andrewbird]

I thought that it might be nice to keep just one version of the inline assembly code and also support your use case.

[ecm-pushbx]

I thought that it might be nice to keep just one version of the inline assembly code and also support your use case.

I don't know how that'd be possible unfortunately. If you use -masm=intel you need either Intel inline code, or (maybe) switch explicitly to AT&T and back after the inline code.

[andrewbird]

Okay, sorry for the noise.