As of LLVM 9.0, certain calls to memcmp may be converted to bcmp, which I guess
could save a single subtraction on some architectures. [1]
bcmp is just like memcmp except instead of returning the difference between the
two differing bytes, it returns non-zero instead. As such, memcmp is a valid
implementation of bcmp.
If we care about size, bcmp should just call memcmp.
If we care about speed, we can change bcmp to look like this instead:
```rust
pub unsafe extern "C" fn bcmp(s1: *const u8, s2: *const u8, n: usize) -> i32 {
let mut i = 0;
while i < n {
let a = *s1.offset(i as isize);
let b = *s2.offset(i as isize);
if a != b {
return 1;
}
i += 1;
}
0
}
```
In this PR I do not address any changes which may or may not be needed for arm
aebi as I lack proper test hardware.
[1]: https://releases.llvm.org/9.0.0/docs/ReleaseNotes.html#noteworthy-optimizations
This turns the following backtrace,
```
>> bt
#0 0x0000555555576f73 in __rust_probestack () at /cargo/registry/src/github.com-1ecc6299db9ec823/compiler_builtins-0.1.14/src/probestack.rs:55
Backtrace stopped: Cannot access memory at address 0x7fffff7fedf0
```
To this:
```
>>> bt
#0 0x0000555555574e47 in __rust_probestack ()
#1 0x00005555555595ba in test::main ()
#2 0x00005555555594f3 in std::rt::lang_start::{{closure}} ()
#3 0x0000555555561ae3 in std::panicking::try::do_call ()
#4 0x000055555556595a in __rust_maybe_catch_panic ()
#5 0x000055555555af9b in std::rt::lang_start_internal ()
#6 0x00005555555594d5 in std::rt::lang_start ()
#7 0x000055555555977b in main ()
```
The -Zlower-128bit-ops feature is completely broken, as libcore needs
those lang items to compile with this feature, but they are only
provided by compiler_builtins, which itself depends on libcore.
According to rust-lang/rust#58969 the feature never got finished.
This commit removes the associated lang items and replaces them with
normal unmangled functions, when there is no existing intrinsic. This
makes it easier for alternative codegen backends to implement 128bit
integer support.
This commit tweaks the implementation of the synthetic
`#[use_c_shim_if]` attribute, renaming it to
`#[maybe_use_optimized_c_shim]` in the process. This no longer requires
specifying a `#[cfg]` clause indicating when the optimized intrinsic
should be used, but rather this is inferred and printed from the build
script.
The build script will now print out appropriate `#[cfg]` directives for
rustc to indicate what intrinsics it's compiling. This should remove the
need for us to keep the build script and the source in sync, but rather
the build script can simply take care of everything.
This commit fixes a bug accidentally introduced in #285 where some
lingering references remained to `#[cfg(thumbv6m)]` but this, since the
historical revert, was renamed to `#[cfg(thumb_1)]`. This caused on the
thumbv6m platform for the intrinsics to be accidentally omitted because
the build script didn't actually compile them but the Rust code thought
the C code was in use.
After correcting the `#[cfg]` statements the CI configuration for the
`thumb*` family of targets was all updated. The support for xargo
testing was removed from `run.sh` since it had long since bitrotted, and
the script was updated to simply build the intrinsics example to attempt
to link for each of these targets. This in turn exposed the bug locally
and allowed to confirm a fix once the `#[cfg]` statements were
corrected.
cc rust-lang/rust#60782
Looks like our tests weren't quite testing compiler-builtins when it was
compiled with unmangled symbols, so update the tests to catch this and
then fix the compilation of the `__clzsi2` intrinsic to use the C
version if it's compiled.
Looks like LLVM optimizes programs like:
fn foo(a: u8) -> f32 {
2.0f32.powf(a as f32)
}
to actually invoking `ldexpf`, so let's be sure to include bindings so
there's not undefined symbols.
Armv8-M Baseline, ie thumbv8m.base-none-eabi, is a superset of the
Armv6-M architecture profile. As it shares almost the same instruction
set, this commit copies the configuration for thumbv6m-none-eabi to
enable it.
Compilation on rustc 1.33.0-nightly (c76f3c374 2019-01-18) failed with
```
error: the feature `cfg_target_vendor` has been stable since 1.33.0 and no longer requires an attribute to enable
--> src/lib.rs:19:12
|
19 | #![feature(cfg_target_vendor)]
| ^^^^^^^^^^^^^^^^^
|
```
Removed the attribute to make it compile.
Adds generic conversion from a wider to a narrower IEEE-754
floating-point type.
Implement `__truncdfsf2` and `__truncdfsf2vfp` and associated test-cases.
This was an accidental regression introduced in #252 by removing compilation of
C files without adjusting the `#[use_c_shim_if]` directives. This restores the
compilation of the assembly files and updates the `#[use_c_shim_if]` directives.
