spring runtime assembly cleaning

[dustanddreams]

This mechanical diff factors multiple occurrences of the stack gymkhana done in non-leaf functions, and standardizes on the ENTER_FUNCTION and LEAVE_FUNCTION found on amd64.

This helps reading the code a bit, and also comparing platforms against each other.

No testsuite regression observed on the four affected platforms.

Not sure this is worth a Changes entry as this is quite mechanical and almost doesn't change anything (the astute reader will notice that this fixes a missing stack adjustment in the raising exception path of caml_call_realloc_stack on s390x, which was harmless since an exception is raised anyway.)

[shindere]

Not able to review but on the topic of the Changes entry it's also a good way to credit both authors and reviewers of PRs and to convey an idea of what is worked on so I would be of the opinion to add an entry if that's okay with you.

[tmcgilchrist]

I am more familiar with power and arm64 assembly, which look correct to me, and the other backends look like a straight forward replacement. General question about the correct use of CFI directives, this PR follows the existing patterns so not specific to this change.

[tmcgilchrist]

On the ordering of CFI directives, should they be before or after the instruction they relate to? When I compile C code to get the assembly, I get the instruction first and then the CFI directive. For compiled OCaml code and runtime assembly there is a mix of before and after. Does it matter? For example on entry to a main function:

_main:                                  ; @main
	.cfi_startproc
; %bb.0:
	sub	sp, sp, #64
	.cfi_def_cfa_offset 64
	stp	x29, x30, [sp, #48]             ; 16-byte Folded Spill
	add	x29, sp, #48
	.cfi_def_cfa w29, 16
	.cfi_offset w30, -8
	.cfi_offset w29, -16
   ...

[dustanddreams]

My understanding is that CFI annotations must appear after the instruction which causes the changes they refer to.

[tmcgilchrist]

Looks good to me. 👍🏻

[gasche]

Approved on behalf of @tmcgilchrist. Thanks everyone!

[gasche]

Hmm, it looks like the CI is failing on macos-arm64. Many lines that look like this:

<instantiation>:1:14: error: expected ')'
CFI_OFFSET(29, -16)
             ^
runtime/arm64.S:438:9: note: while in macro instantiation
        ENTER_FUNCTION
        ^
<instantiation>:1:11: error: invalid operand
CFI_OFFSET(29, -16)
          ^
runtime/arm64.S:438:9: note: while in macro instantiation
        ENTER_FUNCTION
        ^

[dustanddreams]

Hmm, it looks like the CI is failing on macos-arm64. Many lines that look like this:

<instantiation>:1:14: error: expected ')'
CFI_OFFSET(29, -16)
             ^
runtime/arm64.S:438:9: note: while in macro instantiation
        ENTER_FUNCTION
        ^
<instantiation>:1:11: error: invalid operand
CFI_OFFSET(29, -16)
          ^
runtime/arm64.S:438:9: note: while in macro instantiation
        ENTER_FUNCTION
        ^

This should be fixed now. I moved the inclusion of ../runtime/caml/asm.h above the declaration of the macros which use CFI_OFFSET.

[dustanddreams]

This should be fixed now. I moved the inclusion of ../runtime/caml/asm.h above the declaration of the macros which use CFI_OFFSET.

...and riscv64.S needed the same change.

The other backends do not because they use cpp macros rather than as macros.