* Use a no-asm feature instead of an asm feature
This works better as core/alloc/std have trouble supporting default
featues in this crate.
Signed-off-by: Joe Richey <joerichey@google.com>
* Have no-asm disable arm assembly intrinsics
Signed-off-by: Joe Richey <joerichey@google.com>
* Reorganize mem functions
This reduces the amount of platform-specific code
Signed-off-by: Joe Richey <joerichey@google.com>
* Use ERMSB implementations if the feature is set
Signed-off-by: Joe Richey <joerichey@google.com>
* Add non-aligned benchmarks
Signed-off-by: Joe Richey <joerichey@google.com>
* mem: Move mem* functions to separate directory
Signed-off-by: Joe Richey <joerichey@google.com>
* memcpy: Create separate memcpy.rs file
Signed-off-by: Joe Richey <joerichey@google.com>
* benches: Add benchmarks for mem* functions
This allows comparing the "normal" implementations to the
implementations provided by this crate.
Signed-off-by: Joe Richey <joerichey@google.com>
* mem: Add REP MOVSB/STOSB implementations
The assembly generated seems correct:
https://rust.godbolt.org/z/GGnec8
Signed-off-by: Joe Richey <joerichey@google.com>
* mem: Add documentations for REP string insturctions
Signed-off-by: Joe Richey <joerichey@google.com>
* Use quad-word rep string instructions
Signed-off-by: Joe Richey <joerichey@google.com>
* Prevent panic when compiled in debug mode
Signed-off-by: Joe Richey <joerichey@google.com>
* Add tests for mem* functions
Signed-off-by: Joe Richey <joerichey@google.com>
* Add build/test with the "asm" feature
Signed-off-by: Joe Richey <joerichey@google.com>
* Add byte length to Bencher
Signed-off-by: Joe Richey <joerichey@google.com>
1. In UEFI x86 arch, probestack need triple underscore.
2. In UEFI, probestack function do things like _chkstk(in MSVC).
MSVC x32's _chkstk and cygwin/mingw's _alloca adjust %esp themselves
MSVC x64's __chkstk and cygwin/mingw's ___chkstk_ms do not adjust
%rsp themselves.
But current probestack doesn't adjust esp. And LLVM doesn't generate
sub %eax, %esp after probestack. So we adjust esp in probestack like
MSVC x32's _chkstk.
* add 32 bit shift instructions to src/int/shift.rs
__ashlsi3
__ashrsi3
__lshrsi3
* add int_impl! for 16 bit numbers and large_int! for i32 and u32
* add tests in testcrate/build.rs
https://github.com/rust-lang/rust/pull/72700 caused the existing
`allow(improper_ctypes)` guard to stop working, we now need
`allow(improper_ctypes_definitions)` instead.
We keep the old one to avoid any issues with older nightlies.
Signed-off-by: Joe Richey <joerichey@google.com>
* Expand wasm32 testing on CI
Run the full `run.sh` test script to get full assertions, including that
nothing in the wasm compiler-builtins is panicking. Unfortunately it's
currently panicking, so this is good to weed out!
* Update libm
* Allow FFI-unsafe warnings for u128/i128
Handle new warnings on nightly, and we shouldn't need to worry about
these with compiler-builtins since this is tied to a particular compiler.
* Clean up crate attributes
* No need for stability marker
* Rustdoc docs not used for this crate
* Remove old build-system related cruft from rustc itself.
* Run `cargo fmt`
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
