attempt to support `BinaryFormat::Xcoff` in `naked_asm!` #137816

folkertdev · 2025-02-28T19:15:25Z

So, the inline assembly support for xcoff is extremely limited. The LLVM XCOFFAsmParser does not support many of the attributes that LLVM itself emits, and that should exist based on the assembler docs. It also does accept some that should not exist based on those docs.

So, I've tried to do the best I can given those limitations. At least it's better than emitting the directives for elf and having that fail somewhere deep in LLVM. Given that inline assembly for this target is incomplete (under asm_experimental_arch), I think that's OK (and again I don't see how we can do better given the limitations in LLVM).

r? @Noratrieb (given that you reviewed #136637)

It seems reasonable to ping the powerpc64-ibm-aix target maintainers, hopefully they have thoughts too: @daltenty @gilamn5tr

daltenty

Thanks for attempting this. Yeah, the XCOFF assembly parser is indeed has a bunch of problems, we never implemented a lot of the directives you find in normal AIX assembly.

The general approach looks reasonable enough to me, some minor comments

daltenty · 2025-03-05T05:30:39Z

compiler/rustc_codegen_ssa/src/mir/naked_asm.rs

+            //
+            // Consequently, we try our best here but cannot do as good a job as for other targets.
+
+            // FIXME: start a section. `.csect` is not currently implemented in LLVM


Yeah, we modelled the CSect as MCSections in the LLVM backend (XCOFF doesn't really have arbitrary sections, so that works well enought). So for the assembly writing the .csect directive is written when we switch to section.

With that in mind I wonder if .section works for the parser? Otherwise, it's not clear to me which .csect this ends up in? (Without -ffunction-sections we typical have a catch all ..text..[PR], if it's that one seems fine)

doesn't look like it https://godbolt.org/z/4zfvahexs, neither .section or .pushsection is recognized.

daltenty · 2025-03-05T05:38:01Z

compiler/rustc_codegen_ssa/src/mir/naked_asm.rs

+            // fun fact: according to the assembler documentation, .align takes an exponent,
+            // but LLVM only accepts powers of 2 (but does emit the exponent)
+            // so when we hand `.align 32` to LLVM, the assembly output will contain `.align 5`
+            writeln!(begin, ".align {}", align_bytes).unwrap();


note: Yeah, this is a fun dialect difference of AIX assembly that we don't handle well

folkertdev

thanks for taking a look!

folkertdev · 2025-03-05T09:18:34Z

compiler/rustc_codegen_ssa/src/mir/naked_asm.rs

+            //
+            // Consequently, we try our best here but cannot do as good a job as for other targets.
+
+            // FIXME: start a section. `.csect` is not currently implemented in LLVM


doesn't look like it https://godbolt.org/z/4zfvahexs, neither .section or .pushsection is recognized.

folkertdev · 2025-03-05T09:19:52Z

compiler/rustc_codegen_ssa/src/mir/naked_asm.rs

+                        // FIXME: .weak_definition is accepted by the assembly parser, but is not
+                        // documented https://www.ibm.com/docs/en/ssw_aix_71/assembler/assembler_pdf.pdf
+                        // For all I know, it might just be ignored.
+                        //
+                        // On the other hand `.weak` is documented, but rejected as an "unknown directive".
+                        writeln!(w, ".weak_definition {asm_name}")?;


@daltenty do you have thoughts here? Leaving the fixme is probably fine for now, I'm guessing the .weak_definition is somehow supported by the base asm parser implementation? Do you know whether this has any effect?

Unfortunately this doesn't have an effect it seems, and it crashes the backend if you actually try to emit and object: https://godbolt.org/z/rderToPx4

Not sure what we can do in this case, we may just be missing too much of the LLVM side at the moment unfortunately.

well I'll just make this generate a decent error message then saying that it's not currently supported, and link to an llvm issue to track

rustbot · 2025-03-07T10:29:39Z

Some changes occurred in compiler/rustc_codegen_ssa

cc @WaffleLapkin

folkertdev · 2025-03-07T10:31:57Z

Allright, I think I at least have a good picture of the limitations now (hopefully they are documented well) and this PR tests for some of the weird edge cases (e.g. how alignment is handled), in case those change. So, I think this is the best we can to today, and handling of xcoff can be improved when LLVM improves.

compiler/rustc_codegen_ssa/src/mir/naked_asm.rs

Noratrieb · 2025-03-12T20:11:13Z

looks good (or well, as good as it can look given the bad assembly support :D).
it's just a tier 3 target, and a target maintainer reviewed it as well.
@bors r+

bors · 2025-03-12T20:11:16Z

📌 Commit f35bda3 has been approved by Noratrieb

It is now in the queue for this repository.

…trieb attempt to support `BinaryFormat::Xcoff` in `naked_asm!` Fixes rust-lang#137219 So, the inline assembly support for xcoff is extremely limited. The LLVM [XCOFFAsmParser](https://github.com/llvm/llvm-project/blob/1b25c0c4da968fe78921ce77736e5baef4db75e3/llvm/lib/MC/MCParser/XCOFFAsmParser.cpp) does not support many of the attributes that LLVM itself emits, and that should exist based on [the assembler docs](https://www.ibm.com/docs/en/ssw_aix_71/assembler/assembler_pdf.pdf). It also does accept some that should not exist based on those docs. So, I've tried to do the best I can given those limitations. At least it's better than emitting the directives for elf and having that fail somewhere deep in LLVM. Given that inline assembly for this target is incomplete (under `asm_experimental_arch`), I think that's OK (and again I don't see how we can do better given the limitations in LLVM). r? `@Noratrieb` (given that you reviewed rust-lang#136637) It seems reasonable to ping the [`powerpc64-ibm-aix` target maintainers](https://doc.rust-lang.org/rustc/platform-support/aix.html), hopefully they have thoughts too: `@daltenty` `@gilamn5tr`

…trieb attempt to support `BinaryFormat::Xcoff` in `naked_asm!` Fixes rust-lang#137219 So, the inline assembly support for xcoff is extremely limited. The LLVM [XCOFFAsmParser](https://github.com/llvm/llvm-project/blob/1b25c0c4da968fe78921ce77736e5baef4db75e3/llvm/lib/MC/MCParser/XCOFFAsmParser.cpp) does not support many of the attributes that LLVM itself emits, and that should exist based on [the assembler docs](https://www.ibm.com/docs/en/ssw_aix_71/assembler/assembler_pdf.pdf). It also does accept some that should not exist based on those docs. So, I've tried to do the best I can given those limitations. At least it's better than emitting the directives for elf and having that fail somewhere deep in LLVM. Given that inline assembly for this target is incomplete (under `asm_experimental_arch`), I think that's OK (and again I don't see how we can do better given the limitations in LLVM). r? ``@Noratrieb`` (given that you reviewed rust-lang#136637) It seems reasonable to ping the [`powerpc64-ibm-aix` target maintainers](https://doc.rust-lang.org/rustc/platform-support/aix.html), hopefully they have thoughts too: ``@daltenty`` ``@gilamn5tr``

…iaskrgr Rollup of 7 pull requests Successful merges: - rust-lang#137816 (attempt to support `BinaryFormat::Xcoff` in `naked_asm!`) - rust-lang#138109 (make precise capturing args in rustdoc Json typed) - rust-lang#138343 (Enable `f16` tests for `powf`) - rust-lang#138356 (bump libc to 0.2.171 to fix xous) - rust-lang#138371 (Update compiletest's `has_asm_support` to match rustc) - rust-lang#138380 (ci: add runners for vanilla LLVM 20) - rust-lang#138404 (Cleanup sysroot locating a bit) r? `@ghost` `@rustbot` modify labels: rollup

…iaskrgr Rollup of 6 pull requests Successful merges: - rust-lang#137816 (attempt to support `BinaryFormat::Xcoff` in `naked_asm!`) - rust-lang#138109 (make precise capturing args in rustdoc Json typed) - rust-lang#138343 (Enable `f16` tests for `powf`) - rust-lang#138356 (bump libc to 0.2.171 to fix xous) - rust-lang#138371 (Update compiletest's `has_asm_support` to match rustc) - rust-lang#138404 (Cleanup sysroot locating a bit) r? `@ghost` `@rustbot` modify labels: rollup

Rollup merge of rust-lang#137816 - folkertdev:naked-asm-xcoff, r=Noratrieb attempt to support `BinaryFormat::Xcoff` in `naked_asm!` Fixes rust-lang#137219 So, the inline assembly support for xcoff is extremely limited. The LLVM [XCOFFAsmParser](https://github.com/llvm/llvm-project/blob/1b25c0c4da968fe78921ce77736e5baef4db75e3/llvm/lib/MC/MCParser/XCOFFAsmParser.cpp) does not support many of the attributes that LLVM itself emits, and that should exist based on [the assembler docs](https://www.ibm.com/docs/en/ssw_aix_71/assembler/assembler_pdf.pdf). It also does accept some that should not exist based on those docs. So, I've tried to do the best I can given those limitations. At least it's better than emitting the directives for elf and having that fail somewhere deep in LLVM. Given that inline assembly for this target is incomplete (under `asm_experimental_arch`), I think that's OK (and again I don't see how we can do better given the limitations in LLVM). r? ```@Noratrieb``` (given that you reviewed rust-lang#136637) It seems reasonable to ping the [`powerpc64-ibm-aix` target maintainers](https://doc.rust-lang.org/rustc/platform-support/aix.html), hopefully they have thoughts too: ```@daltenty``` ```@gilamn5tr```

…iaskrgr Rollup of 6 pull requests Successful merges: - rust-lang#137816 (attempt to support `BinaryFormat::Xcoff` in `naked_asm!`) - rust-lang#138109 (make precise capturing args in rustdoc Json typed) - rust-lang#138343 (Enable `f16` tests for `powf`) - rust-lang#138356 (bump libc to 0.2.171 to fix xous) - rust-lang#138371 (Update compiletest's `has_asm_support` to match rustc) - rust-lang#138404 (Cleanup sysroot locating a bit) r? `@ghost` `@rustbot` modify labels: rollup

move naked function assembly tests to their own directory

c620d76

rustbot assigned Noratrieb Feb 28, 2025

rustbot added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. labels Feb 28, 2025

daltenty reviewed Mar 5, 2025

View reviewed changes

folkertdev commented Mar 5, 2025

View reviewed changes

folkertdev mentioned this pull request Mar 7, 2025

powerpc64-ibm-aix: the .weak assembler directive is not supported llvm/llvm-project#130269

Open

folkertdev force-pushed the naked-asm-xcoff branch from 2577662 to 5e1f4d5 Compare March 7, 2025 10:29

folkertdev force-pushed the naked-asm-xcoff branch from 5e1f4d5 to c2d240f Compare March 7, 2025 10:34

WaffleLapkin reviewed Mar 8, 2025

View reviewed changes

compiler/rustc_codegen_ssa/src/mir/naked_asm.rs Outdated Show resolved Hide resolved

support XCOFF in naked_asm!

f35bda3

folkertdev force-pushed the naked-asm-xcoff branch from c2d240f to f35bda3 Compare March 8, 2025 13:11

bors added S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Mar 12, 2025

matthiaskrgr mentioned this pull request Mar 13, 2025

Rollup of 8 pull requests #138444

Closed

matthiaskrgr mentioned this pull request Mar 13, 2025

Rollup of 7 pull requests #138445

Closed

matthiaskrgr mentioned this pull request Mar 13, 2025

Rollup of 6 pull requests #138450

Merged

bors merged commit 762acf5 into rust-lang:master Mar 13, 2025
6 checks passed

rustbot added this to the 1.87.0 milestone Mar 13, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

attempt to support `BinaryFormat::Xcoff` in `naked_asm!` #137816

attempt to support `BinaryFormat::Xcoff` in `naked_asm!` #137816

folkertdev commented Feb 28, 2025

daltenty left a comment

daltenty Mar 5, 2025

folkertdev Mar 5, 2025

daltenty Mar 5, 2025

folkertdev left a comment

folkertdev Mar 5, 2025

folkertdev Mar 5, 2025

daltenty Mar 7, 2025

folkertdev Mar 7, 2025

rustbot commented Mar 7, 2025

folkertdev commented Mar 7, 2025

Noratrieb commented Mar 12, 2025

bors commented Mar 12, 2025

attempt to support BinaryFormat::Xcoff in naked_asm! #137816

attempt to support BinaryFormat::Xcoff in naked_asm! #137816

Conversation

folkertdev commented Feb 28, 2025

daltenty left a comment

Choose a reason for hiding this comment

daltenty Mar 5, 2025

Choose a reason for hiding this comment

folkertdev Mar 5, 2025

Choose a reason for hiding this comment

daltenty Mar 5, 2025

Choose a reason for hiding this comment

folkertdev left a comment

Choose a reason for hiding this comment

folkertdev Mar 5, 2025

Choose a reason for hiding this comment

folkertdev Mar 5, 2025

Choose a reason for hiding this comment

daltenty Mar 7, 2025

Choose a reason for hiding this comment

folkertdev Mar 7, 2025

Choose a reason for hiding this comment

rustbot commented Mar 7, 2025

folkertdev commented Mar 7, 2025

Noratrieb commented Mar 12, 2025

bors commented Mar 12, 2025

attempt to support `BinaryFormat::Xcoff` in `naked_asm!` #137816

attempt to support `BinaryFormat::Xcoff` in `naked_asm!` #137816