Optimize Ord trait implementation for bool
Casting the booleans to `i8`s and converting their difference into `Ordering` generates better assembly than casting them to `u8`s and comparing them.
Fixes#66780
#### Comparison([Godbolt link](https://rust.godbolt.org/z/PjBpvF))
##### Old assembly:
```asm
example::boolean_cmp:
mov ecx, edi
xor ecx, esi
test esi, esi
mov eax, 255
cmove eax, ecx
test edi, edi
cmovne eax, ecx
ret
```
##### New assembly:
```asm
example::boolean_cmp:
mov eax, edi
sub al, sil
ret
```
##### Old LLVM-MCA statistics:
```
Iterations: 100
Instructions: 800
Total Cycles: 234
Total uOps: 1000
Dispatch Width: 6
uOps Per Cycle: 4.27
IPC: 3.42
Block RThroughput: 1.7
```
##### New LLVM-MCA statistics:
```
Iterations: 100
Instructions: 300
Total Cycles: 110
Total uOps: 500
Dispatch Width: 6
uOps Per Cycle: 4.55
IPC: 2.73
Block RThroughput: 1.0
```
LLVM 7 is over a year old, which should be plenty for compatibility. The
last LLVM 6 holdout was llvm-emscripten, which went away in #65501.
I've also included a fix for LLVM 8 lacking `MemorySanitizerOptions`,
which was broken by #66522.
This optimization depends on inlining for the identity
conversions introduced by the lowering of the `?`.
To take advantage of `SimplifyArmIdentity`, `-Z mir-opt-level=2`
is required because that triggers the inlining MIR optimization.
perf.rlo shows that running the `ConstProp` pass results in
across-the-board wins regardless of debug or opt complilation mode. As a
result, we're turning it on to get the compile time benefits.
`ConstProp` doesn't currently intern the memory used by its `Machine` so
we can't yet propagate allocations which is why
`ConstProp::should_const_prop()` checks if the value being propagated is
a scalar or not.
Re-enable Emscripten's exception handling support
Passes LLVM codegen and Emscripten link-time flags for exception
handling if and only if the panic strategy is `unwind`. Sets the
default panic strategy for Emscripten targets to `unwind`. Re-enables
tests that depend on unwinding support for Emscripten, including
`should_panic` tests.
r? @alexcrichton
Passes LLVM codegen and Emscripten link-time flags for exception
handling if and only if the panic strategy is `unwind`. Sets the
default panic strategy for Emscripten targets to `unwind`. Re-enables
tests that depend on unwinding support for Emscripten, including
`should_panic` tests.
Use revisions to run the EFIABI in multiple configurations, compiling
for each supported UEFI platform, and checking the ABI generated in the
LLVM IR is correct.
Use no_core to make it easier to test.
Adds a new ABI for the EFIAPI calls. This ABI should reflect the latest
version of the UEFI specification at the time of commit (UEFI spec 2.8,
URL below). The specification says that for x86_64, we should follow the
win64 ABI, while on all other supported platforms (ia32, itanium, arm,
arm64 and risc-v), we should follow the C ABI.
To simplify the implementation, we will simply follow the C ABI on all
platforms except x86_64, even those technically unsupported by the UEFI
specification.
https://uefi.org/sites/default/files/resources/UEFI_Spec_2_8_final.pdf
- Compatible with Emscripten 1.38.46-upstream or later upstream.
- Refactors the Emscripten target spec to share code with other wasm
targets.
- Replaces the old incorrect wasm32 C call ABI with the correct one,
preserving the old one as wasm32_bindgen_compat for wasm-bindgen
compatibility.
- Updates the varargs ABI used by Emscripten and deletes the old one.
- Removes the obsolete wasm32-experimental-emscripten target.
- Uses EMCC_CFLAGS on CI to avoid the timeout problems with #63649.
LLVM assumes that a thread will eventually cause side effect. This is
not true in Rust if a loop or recursion does nothing in its body,
causing undefined behavior even in common cases like `loop {}`.
Inserting llvm.sideeffect fixes the undefined behavior.
As a micro-optimization, only insert llvm.sideeffect when jumping back
in blocks or calling a function.
A patch for LLVM is expected to allow empty non-terminate code by
default and fix this issue from LLVM side.
https://github.com/rust-lang/rust/issues/28728
On architectures such as powerpc64 that use extend_integer_width_to in
their C ABI processing, integer parameters shorter than the native
register width will be annotated with the ArgAttribute::SExt or
ArgAttribute::ZExt attribute, and that attribute will be included in the
generated LLVM IR.
In this test, all relevant parameters are `i32`, which will get the
`signext` annotation on the relevant 64-bit architectures. Match both
the annotated and non-annotated case, but enforce that the annotation is
applied consistently.
Update to LLVM 9 trunk
Following the preparatory changes in #62474, this updates the LLVM submodule to https://github.com/rust-lang/llvm-project/tree/rustc/9.0-2019-07-12 and:
* Changes the LLVM Rust bindings to account for the new SubtargetSubTypeKV.
* Adjusts a codegen test for the new form of the byval attribute that takes a type.
* Makes a PGO codegen test more liberal with regard to order and linkage.
* Builds InstrProfilingPlatformWindows.c as part of libprofiler_builtins.
* Moves registration of additional passes (in particular sanitizers) to the end of the module pass manager.
* Disables LLDB on builders.
r? @alexcrichton
Don't require a specific order for the per-function globals, and
don't require the locals to have private linkage (apparently
internal linkage is also possible).
Refactor C FFI variadics to more closely match their C counterparts, and add Clone implementation
We had to make some changes to expose `va_copy` and `va_end` directly to users (mainly for C2Rust, but not exclusively):
- redefine the Rust variadic structures to more closely correspond to C: `VaList` now matches `va_list`, and `VaListImpl` matches `__va_list_tag`
- add `Clone` for `VaListImpl`
- add explicit `as_va_list()` conversion function from `VaListImpl` to `VaList`
- add deref coercion from `VaList` to `VaListImpl`
- add support for the `asmjs` target
All these changes were needed for use cases like:
```Rust
let mut ap2 = va_copy(ap);
vprintf(fmt, ap2);
va_end(&mut ap2);
```
Help LLVM better optimize slice::Iter(Mut)::len
r? @RalfJung
I've included a codegen test that fails without this change as a demonstration of usefulness.
