Autodiff Upstreaming - rustc_codegen_ssa, rustc_middle
This PR should not be merged until the rustc_codegen_llvm part is merged.
I will also alter it a little based on what get's shaved off from the cg_llvm PR,
and address some of the feedback I received in the other PR (including cleanups).
I am putting it already up to
1) Discuss with `@jieyouxu` if there is more work needed to add tests to this and
2) Pray that there is someone reviewing who can tell me why some of my autodiff invocations get lost.
Re 1: My test require fat-lto. I also modify the compilation pipeline. So if there are any other llvm-ir tests in the same compilation unit then I will likely break them. Luckily there are two groups who currently have the same fat-lto requirement for their GPU code which I have for my autodiff code and both groups have some plans to enable support for thin-lto. Once either that work pans out, I'll copy it over for this feature. I will also work on not changing the optimization pipeline for functions not differentiated, but that will require some thoughts and engineering, so I think it would be good to be able to run the autodiff tests isolated from the rest for now. Can you guide me here please?
For context, here are some of my tests in the samples folder: https://github.com/EnzymeAD/rustbook
Re 2: This is a pretty serious issue, since it effectively prevents publishing libraries making use of autodiff: https://github.com/EnzymeAD/rust/issues/173. For some reason my dummy code persists till the end, so the code which calls autodiff, deletes the dummy, and inserts the code to compute the derivative never gets executed. To me it looks like the rustc_autodiff attribute just get's dropped, but I don't know WHY? Any help would be super appreciated, as rustc queries look a bit voodoo to me.
Tracking:
- https://github.com/rust-lang/rust/issues/124509
r? `@jieyouxu`
Cast global variables to default address space
Pointers for variables all need to be in the same address space for correct compilation. Therefore ensure that even if a global variable is created in a different address space, it is casted to the default address space before its value is used.
This is necessary for the amdgpu target and others where the default address space for global variables is not 0.
For example `core` does not compile in debug mode when not casting the address space to the default one because it tries to emit the following (simplified) LLVM IR, containing a type mismatch:
```llvm
`@alloc_0` = addrspace(1) constant <{ [6 x i8] }> <{ [6 x i8] c"bit.rs" }>, align 1
`@alloc_1` = addrspace(1) constant <{ ptr }> <{ ptr addrspace(1) `@alloc_0` }>, align 8
; ^ here a struct containing a `ptr` is needed, but it is created using a `ptr addrspace(1)`
```
For this to compile, we need to insert a constant `addrspacecast` before we use a global variable:
```llvm
`@alloc_0` = addrspace(1) constant <{ [6 x i8] }> <{ [6 x i8] c"bit.rs" }>, align 1
`@alloc_1` = addrspace(1) constant <{ ptr }> <{ ptr addrspacecast (ptr addrspace(1) `@alloc_0` to ptr) }>, align 8
```
As vtables are global variables as well, they are also created with an `addrspacecast`. In the SSA backend, after a vtable global is created, metadata is added to it. To add metadata, we need the non-casted global variable. Therefore we strip away an addrspacecast if there is one, to get the underlying global.
Tracking issue: #135024
Make our `DIFlags` match `LLVMDIFlags` in the LLVM-C API
In order to be able to use a mixture of LLVM-C and C++ bindings for debuginfo, our Rust-side `DIFlags` needs to have the same layout as LLVM-C's `LLVMDIFlags`, and we also need to be able to convert it to the `DIFlags` accepted by LLVM's C++ API.
Internally, LLVM converts between the two types with a simple cast. We can't necessarily rely on that always being true, and LLVM doesn't expose a conversion function, so we have two potential options:
- Convert each bit/subvalue individually
- Statically assert that doing a cast is actually fine
As long as both types do remain the same under the hood (which seems likely), the static-assert-and-cast approach is easier and faster. If the static assertions ever start failing against some future version of LLVM, we'll have to switch over to the convert-each-subvalue approach, which is a bit more error-prone.
---
Extracted from #134009, though this PR ended up choosing the static-assert-and-cast approach over the convert-each-subvalue approach.
Add gpu-kernel calling convention
The amdgpu-kernel calling convention was reverted in commit f6b21e90d1 (#120495 and https://github.com/rust-lang/rust-analyzer/pull/16463) due to inactivity in the amdgpu target.
Introduce a `gpu-kernel` calling convention that translates to `ptx_kernel` or `amdgpu_kernel`, depending on the target that rust compiles for.
Tracking issue: #135467
amdgpu target tracking issue: #135024
The amdgpu-kernel calling convention was reverted in commit
f6b21e90d1 due to inactivity in the amdgpu
target.
Introduce a `gpu-kernel` calling convention that translates to
`ptx_kernel` or `amdgpu_kernel`, depending on the target that rust
compiles for.
See llvm/llvm-project#121851
For LLVM 20+, this function (`renameModuleForThinLTO`) has no return
value. For prior versions of LLVM, this never failed, but had a
signature which allowed an error value people were handling.
[Debuginfo] Force enum `DISCR_*` to `static const u64` to allow for inspection via LLDB
see [here](https://rust-lang.zulipchat.com/#narrow/channel/317568-t-compiler.2Fwg-debugging/topic/Revamping.20Debuginfo/near/486614878) for more info.
This change mainly helps `*-msvc` debugged with LLDB. Currently, LLDB cannot inspect `static` struct fields, so the intended visualization for enums is only borderline functional, and niche enums with ranges of discriminant cannot be determined at all .
LLDB *can* inspect `static const` values (though for whatever reason, non-enum/non-u64 consts don't work).
This change adds the `LLVMRustDIBuilderCreateQualifiedType` to the rust FFI layer to wrap the discr type with a `const` modifier, as well as forcing all generated integer enum `DISCR_*` values to be u64's. Those values will only ever be used by debugger visualizers anyway, so it shouldn't be a huge deal, but I left a fixme comment for it just in case.. The `tag` also still properly reflects the discriminant type, so no information is lost.
Pointers for variables all need to be in the same address space for
correct compilation. Therefore ensure that even if a global variable is
created in a different address space, it is casted to the default
address space before its value is used.
This is necessary for the amdgpu target and others where the default
address space for global variables is not 0.
For example `core` does not compile in debug mode when not casting the
address space to the default one because it tries to emit the following
(simplified) LLVM IR, containing a type mismatch:
```llvm
@alloc_0 = addrspace(1) constant <{ [6 x i8] }> <{ [6 x i8] c"bit.rs" }>, align 1
@alloc_1 = addrspace(1) constant <{ ptr }> <{ ptr addrspace(1) @alloc_0 }>, align 8
; ^ here a struct containing a `ptr` is needed, but it is created using a `ptr addrspace(1)`
```
For this to compile, we need to insert a constant `addrspacecast` before
we use a global variable:
```llvm
@alloc_0 = addrspace(1) constant <{ [6 x i8] }> <{ [6 x i8] c"bit.rs" }>, align 1
@alloc_1 = addrspace(1) constant <{ ptr }> <{ ptr addrspacecast (ptr addrspace(1) @alloc_0 to ptr) }>, align 8
```
As vtables are global variables as well, they are also created with an
`addrspacecast`. In the SSA backend, after a vtable global is created,
metadata is added to it. To add metadata, we need the non-casted global
variable. Therefore we strip away an addrspacecast if there is one, to
get the underlying global.
In the LLVM-C API, boolean values are passed as `typedef int LLVMBool`, but our
Rust-side typedef was using `c_uint` instead.
Signed and unsigned integers have the same ABI on most platforms, but that
isn't universally true, so we should prefer to be consistent with LLVM.
Pass end position of span through inline ASM cookie
Before this PR, only the start position of the span was passed though the inline ASM cookie to diagnostics. LLVM 19 has full support for 64-bit inline ASM cookies; this PR uses that to pass the end position of the span in the upper 32 bits, meaning inline ASM diagnostics now point at the entire line the error occurred on, not just the first character of it.
Allow disabling ASan instrumentation for globals
AddressSanitizer adds instrumentation to global variables unless the [`no_sanitize_address`](https://llvm.org/docs/LangRef.html#global-attributes) attribute is set on them.
This commit extends the existing `#[no_sanitize(address)]` attribute to set this; previously it only had the desired effect on functions.
(cc https://github.com/rust-lang/rust/issues/39699)
CFI: Append debug location to CFI blocks
Currently we're not appending debug locations to the inserted CFI blocks. This shows up in #132615 and #100783. This change fixes that by passing down the debug location to the CFI type-test generation and appending it to the blocks.
Credits also belong to `@jakos-sec` who worked with me on this.
LLVM does not expect to ever see multiple dbg_declares for the same variable at the same
location with different values. proc-macros make it possible for arbitrary code,
including multiple calls that get inlined, to happen at any given location in the source
code. Add discriminators when that happens so these locations are different to LLVM.
This may interfere with the AddDiscriminators pass in LLVM, which is added by the
unstable flag -Zdebug-info-for-profiling.
Fixes#131944
Trim and tidy includes in `rustc_llvm`
These includes tend to accumulate over time, and are usually only removed when something breaks in a new LLVM version, so it's nice to clean them up manually once in a while.
General strategy used for this PR:
- Remove all includes from `LLVMWrapper.h` that aren't needed by the header itself, transplanting them to individual source files as necessary.
- For each source file, temporarily remove each include if doing so doesn't cause a compile error.
- If a “required” include looks like it shouldn't be needed, try replacing it with its sub-includes, then trim that list.
- After doing all of the above, go back and re-add any removed include if the file does actually use things defined in that header, even if the header happens to also be included by something else.
Simplify FFI calls for `-Ztime-llvm-passes` and `-Zprint-codegen-stats`
The existing code for these unstable LLVM-infodump flags was jumping through hoops to pass an allocated C string across the FFI boundary, when it's much simpler to just write to a `&RustString` instead.
AddressSanitizer adds instrumentation to global variables unless the
[`no_sanitize_address`](https://llvm.org/docs/LangRef.html#global-attributes)
attribute is set on them.
This commit extends the existing `#[no_sanitize(address)]` attribute to
set this; previously it only had the desired effect on functions.
cg_llvm: Clean up FFI calls for operand bundles
All of these FFI functions have equivalents in the stable LLVM-C API, though `LLVMBuildCallBr` requires a temporary polyfill on LLVM 18.
This PR also creates a clear split between `OperandBundleOwned` and `OperandBundle`, and updates the internals of the owner to be a little less terrifying.
cg_llvm: Clean up FFI calls for setting module flags
This is a combination of several inter-related changes to how module flags are set:
- Remove some unnecessary code for setting an `"LTOPostLink"` flag, which has been obsolete since LLVM 17.
- Define our own enum instead of relying on enum values defined by LLVM's unstable C++ API.
- Use safe wrapper functions to set module flags, instead of direct `unsafe` calls.
- Consistently pass pointer/length strings instead of C strings.
- Remove or shrink some `unsafe` blocks.
- Don't rely on enum values defined by LLVM's C++ API
- Use safe wrapper functions instead of direct `unsafe` calls
- Consistently pass pointer/length strings instead of C strings
correct LLVMRustCreateThinLTOData arg types
`LLVMRustCreateThinLTOData` defined in rust as
```rust
pub fn LLVMRustCreateThinLTOData(
Modules: *const ThinLTOModule,
NumModules: c_uint,
PreservedSymbols: *const *const c_char,
PreservedSymbolsLen: c_uint,
) -> Option<&'static mut ThinLTOData>;
```
but in cpp as
```cpp
extern "C" LLVMRustThinLTOData *
LLVMRustCreateThinLTOData(LLVMRustThinLTOModule *modules, int num_modules,
const char **preserved_symbols, int num_symbols) {
```
(note `c_unit` vs `int` types). Let it be actually `size_t`.
Also fixes return type of `LLVMRustDIBuilderCreateOpLLVMFragment` to uint64_t as other similar functions around, which should be correct, i assume.
coverage: Pass coverage mappings to LLVM as separate structs
Instead of trying to cram *N* different kinds of coverage mapping data into a single list for FFI, pass *N* different lists of simpler structs.
This avoids the need to fill unused fields with dummy values, and avoids the need to tag structs with their underlying kind. It also lets us call the dedicated LLVM constructors for each different mapping type, instead of having to go through the complex general-purpose constructor.
Even though this adds multiple new structs to the FFI surface area, the resulting C++ code is simpler and shorter.
---
I've structured this mostly as a single atomic patch, rather than a series of incremental changes, because that avoids the need to make fiddly fixes to code that is about to be deleted anyway.
Migrate `llvm::set_comdat` and `llvm::SetUniqueComdat` to LLVM-C FFI.
Note, now we can call `llvm::set_comdat` only when the target actually
supports adding comdat. As this has no convenient LLVM-C API, we
implement this as `TargetOptions::supports_comdat`.
Co-authored-by: Stuart Cook <Zalathar@users.noreply.github.com>
llvm: replace some deprecated functions
`LLVMMDStringInContext` and `LLVMMDNodeInContext` are deprecated, replace them with `LLVMMDStringInContext2` and `LLVMMDNodeInContext2`.
Also replace `Value` with `Metadata` in some function signatures for better consistency.
Because constants are currently emitted *before* the prologue, leaving the
debug location on the IRBuilder spills onto other instructions in the prologue
and messes up both line numbers as well as the point LLVM chooses to be the
prologue end.
Example LLVM IR (irrelevant IR elided):
Before:
define internal { i64, i64 } @_ZN3tmp3Foo18var_return_opt_try17he02116165b0fc08cE(ptr align 8 %self) !dbg !347 {
start:
%self.dbg.spill = alloca [8 x i8], align 8
%_0 = alloca [16 x i8], align 8
%residual.dbg.spill = alloca [0 x i8], align 1
#dbg_declare(ptr %residual.dbg.spill, !353, !DIExpression(), !357)
store ptr %self, ptr %self.dbg.spill, align 8, !dbg !357
#dbg_declare(ptr %self.dbg.spill, !350, !DIExpression(), !358)
After:
define internal { i64, i64 } @_ZN3tmp3Foo18var_return_opt_try17h00b17d08874ddd90E(ptr align 8 %self) !dbg !347 {
start:
%self.dbg.spill = alloca [8 x i8], align 8
%_0 = alloca [16 x i8], align 8
%residual.dbg.spill = alloca [0 x i8], align 1
#dbg_declare(ptr %residual.dbg.spill, !353, !DIExpression(), !357)
store ptr %self, ptr %self.dbg.spill, align 8
#dbg_declare(ptr %self.dbg.spill, !350, !DIExpression(), !358)
Note in particular how !357 from %residual.dbg.spill's dbg_declare no longer
falls through onto the store to %self.dbg.spill. This fixes argument values
at entry when the constant is a ZST (e.g. <Option as Try>::Residual). This
fixes#130003 (but note that it does *not* fix issues with argument values and
non-ZST constants, which emit their own stores that have debug info on them,
like #128945).
Add `select_unpredictable` to force LLVM to use CMOV
Since https://reviews.llvm.org/D118118, LLVM will no longer turn CMOVs into branches if it comes from a `select` marked with an `unpredictable` metadata attribute.
This PR introduces `core::intrinsics::select_unpredictable` which emits such a `select` and uses it in the implementation of `binary_search_by`.
Since https://reviews.llvm.org/D118118, LLVM will no longer turn CMOVs
into branches if it comes from a `select` marked with an `unpredictable`
metadata attribute.
This PR introduces `core::intrinsics::select_unpredictable` which emits
such a `select` and uses it in the implementation of `binary_search_by`.
Sync ar_archive_writer to LLVM 18.1.3
From LLVM 15.0.0-rc3. This adds support for COFF archives containing Arm64EC object files and has various fixes for AIX big archive files.
The return value changed from an Instruction to a DbgRecord in
LLVM 19. As we don't actually use the result, drop the return
value entirely to support both.
Since this codegen flag now only controls LLVM-generated comments rather than
all assembly comments, make the name more accurate (and also match Clang).
If we don't do this, some versions of LLVM (at least 17, experimentally)
will double-emit some error messages, which is how I noticed this. Given
that it seems to be costing some extra work, let's only request the
summary bitcode production if we'll actually bother writing it down,
otherwise skip it.
Typical uses of ThinLTO don't have any use for this as a standalone
file, but distributed ThinLTO uses this to make the linker phase more
efficient. With clang you'd do something like `clang -flto=thin
-fthin-link-bitcode=foo.indexing.o -c foo.c` and then get both foo.o
(full of bitcode) and foo.indexing.o (just the summary or index part of
the bitcode). That's then usable by a two-stage linking process that's
more friendly to distributed build systems like bazel, which is why I'm
working on this area.
I talked some to @teresajohnson about naming in this area, as things
seem to be a little confused between various blog posts and build
systems. "bitcode index" and "bitcode summary" tend to be a little too
ambiguous, and she tends to use "thin link bitcode" and "minimized
bitcode" (which matches the descriptions in LLVM). Since the clang
option is thin-link-bitcode, I went with that to try and not add a new
spelling in the world.
Per @dtolnay, you can work around the lack of this by using `lld
--thinlto-index-only` to do the indexing on regular .o files of
bitcode, but that is a bit wasteful on actions when we already have all
the information in rustc and could just write out the matching minimized
bitcode. I didn't test that at all in our infrastructure, because by the
time I learned that I already had this patch largely written.
Set writable and dead_on_unwind attributes for sret arguments
Set the `writable` and `dead_on_unwind` attributes for `sret` arguments. This allows call slot optimization to remove more memcpy's.
See https://llvm.org/docs/LangRef.html#parameter-attributes for the specification of these attributes. In short, the statement we're making here is that:
* The return slot is writable.
* The return slot will not be read if the function unwinds.
Fixes https://github.com/rust-lang/rust/issues/90595.
LLVM's default bad-alloc handler may throw if exceptions are enabled,
and `operator new` isn't hooked at all by default. Now we register our
own handler that prints a message similar to fatal errors, then aborts.
We also call the function that registers the C++ `std::new_handler`.
Add asm goto support to `asm!`
Tracking issue: #119364
This PR implements asm-goto support, using the syntax described in "future possibilities" section of [RFC2873](https://rust-lang.github.io/rfcs/2873-inline-asm.html#asm-goto).
Currently I have only implemented the `label` part, not the `fallthrough` part (i.e. fallthrough is implicit). This doesn't reduce the expressive though, since you can use label-break to get arbitrary control flow or simply set a value and rely on jump threading optimisation to get the desired control flow. I can add that later if deemed necessary.
r? ``@Amanieu``
cc ``@ojeda``
