Generalized operand.rs#nontemporal_store and fixed tidy issues
Generalized operand.rs#nontemporal_store's implem even more
With a BuilderMethod trait implemented by Builder for LLVM
Cleaned builder.rs : no more code duplication, no more ValueTrait
Full traitification of builder.rs
Generalized FunctionCx
Added ValueTrait and first change
Generalize CondegenCx
Generalized the Builder struct defined in librustc_codegen_llvm/builder.rs
Add mem::forget_unsized() for forgetting unsized values
~~Allows passing values of `T: ?Sized` types to `mem::drop` and `mem::forget`.~~
Adds `mem::forget_unsized()` that accepts `T: ?Sized`.
I had to revert the PR that removed the `forget` intrinsic and replaced it with `ManuallyDrop`: https://github.com/rust-lang/rust/pull/40559
We can't use `ManuallyDrop::new()` here because it needs `T: Sized` and we don't have support for unsized return values yet (will we ever?).
r? @eddyb
Fix emission of niche-filling discriminant values
Bug #55606 points out a regression introduced by #54004; namely that
an assertion can erroneously fire when a niche-filling discriminant
value is emitted.
This fixes the bug by removing the assertion, and furthermore by
arranging for the discriminant value to be masked according to the
size of the niche. This makes handling the discriminant a bit simpler
for debuggers.
The test case is from Jonathan Turner.
Closes#55606
Support for the program data address space option of LLVM's Target Datalayout
This was introduced recently (specifically, for AVR, cc @dylanmckay).
(I came up with this when attempting to run [avr-rust](https://github.com/avr-rust/rust) rebased on the latest [rust-lang](https://github.com/rust-lang/rust) commits. If this requires a different design, some additional discussions, or is not something to pursue right now, I'd be happy to close this PR).
Note that this somewhat overlaps with @DiamondLovesYou's #51576, I think, although the implementation here is significantly simpler: Since the address space applies to _all_ program data, we can just check the pointee's type whenever we create an LLVM pointer type. If it is a function we use the program data address space; if not we use the default address space.
cc @eddyb, who has been reviewing #51576
Ref: https://llvm.org/docs/LangRef.html#data-layout
Cleanup codegen_llvm/back
- improve allocations
- use `Cow<'static, str>` where applicable
- use `to_owned` instead of `to_string` with string literals
- remove a redundant `continue`
- possible minor speedup in logic
- use `mem::replace` instead of `swap` where more concise
- remove `'static` from consts
- improve common patterns
- remove explicit `return`s
- whitespace & formatting fixes
Implement rotate using funnel shift on LLVM >= 7
Implement the rotate_left and rotate_right operations using
llvm.fshl and llvm.fshr if they are available (LLVM >= 7).
Originally I wanted to expose the funnel_shift_left and
funnel_shift_right intrinsics and implement rotate_left and
rotate_right on top of them. However, emulation of funnel
shifts requires emitting a conditional to check for zero shift
amount, which is not necessary for rotates. I was uncomfortable
doing that here, as I don't want to rely on LLVM to optimize
away that conditional (and for variable rotates, I'm not sure it
can). We should revisit that question when we raise our minimum
version requirement to LLVM 7 and don't need emulation code
anymore.
Fixes#52457.
Update emscripten
This updates emscripten to 1.38.15, which is based on LLVM 6.0.1 and would allow us to drop code for handling LLVM 4.
The main issue I ran into is that exporting statics through `EXPORTED_FUNCTIONS` no longer works. As far as I understand exporting non-functions doesn't really make sense under emscripten anyway, so I've modified the symbol export code to not even try.
Closes#52323.
Support memcpy/memmove with differing src/dst alignment
If LLVM 7 is used, generate memcpy/memmove with differing src/dst alignment. I've added new FFI functions to construct these through the builder API, which is more convenient than dealing with differing intrinsic signatures depending on the LLVM version.
Fixes#49740.
Run name-anon-globals after LTO passes as well
If we're going to emit bitcode (through ThinLTOBuffer), then we need to ensure that anon globals are named. This was already done after optimization passes, but also has to happen after LTO passes, as we always emit the final result in a ThinLTO-compatible manner.
I added the test as `run-make`. The important bit is that we emit bitcode in some way (e.g. `--crate-type rlib` or `--emit=llvm-bc`). Please tell me if there is a better way to test for that.
Fixes#51947
If we're going to emit bitcode (through ThinLTOBuffer), then we
need to ensure that anon globals are named. This was already done
after optimization passes, but also has to happen after LTO passes,
as we always emit the final result in a ThinLTO-compatible manner.
Fixes#51947.
Bug #55606 points out a regression introduced by #54004; namely that
an assertion can erroneously fire when a niche-filling discriminant
value is emitted.
This fixes the bug by removing the assertion, and furthermore by
arranging for the discriminant value to be masked according to the
size of the niche. This makes handling the discriminant a bit simpler
for debuggers.
The test case is from Jonathan Turner.
Closes#55606
