Tale as old as the world - there's an ABI mismatch:
https://github.com/rust-lang/compiler-builtins/pull/527
Fortunately, newest GCCs (from v11, it seems) actually provide most of
those intrinsics (even for f64!), so that's pretty cool.
(the only intrinsics not provided by GCC are `__powisf2` & `__powidf2`,
but our codegen for AVR doesn't emit those anyway.)
Fixes https://github.com/rust-lang/rust/issues/118079.
Like SGX, Xous does not have any libc to link against. As a result,
memory intrinsics need to be available as part of `compiler_builtins`
Signed-off-by: Sean Cross <sean@xobs.io>
they dereference raw pointers, so the caller needs to make sure the
pointer is valid.
note that this requires changing `maybe_use_optimized_c_shim` to support
unsafe functions.
This has a very long history, summarized in
https://github.com/rust-lang/rust/issues/109064. This port is a very
minimal subset of `aarch64/lse.S` from LLVM's compiler-rt. In
particular, it is missing the following:
1. Any form of runtime dispatch between LL/SC and LSE.
Determining which version of the intrinsics to use
requires one of the following:
i) `getauxval` from glibc. It's unclear whether `compiler_builtins` is
allowed to depend on libc at all, and musl doesn't even support
getauxval. Don't enshrine the requirement "de-facto" by making it
required for outline-atomics.
ii) kernel support. Linux and FreeBSD have limited support, but it
requires an extremely recent kernel version and doesn't work at all under QEMU (https://github.com/rust-lang/rust/issues/109064#issuecomment-1494939904).
Instead, we hard-code LL/SC intrinsics. Users who want LSE support
should use the LLVM compiler-rt (if you're building from source in
rust-lang/rust, make sure you have `src/llvm-project` checked out
locally. the goal is to soon add a new `optimized-compiler-builtins`
option so this is easier to discover).
2. The global `___aarch64_have_lse_atomics` CTOR, required to do runtime
dispatch. Thom Chiviolani has this to say about global CTORs:
> static ctors are problems because we are pretty eager about dead code elim
> in general if you have a module that isnt directly reference we will probably not have its static ctors
> also, while llvm has a super robust way to have a static ctor (theres s special "appending global" to use for c++), we dont use that and just have people make a #[used] static in a special section
> 1. the robust way kinda requires rust knowing that the argument is a static ctor (maybe a #[rustc_static_ctor] attribute). it also would be... finnicky, since on windows we actually care beyond being a static ctor, that we run as part in a specific group of ctors, which means a very specific section (one for TLS and the other for, uh, i dont remember)
> 2. we still actually have to codegen the cgu that isn't referenced. but maybe we could remember that it has that attribute and use that
So while this is possible in theory, it's decidedly non-trivial, and
needs invasive changes to rust itself. In any case, it doesn't matter
until we decide the story around libc.
3. The 16-byte (i128) version of compare_and_swap. This wouldn't be
*too* hard to add, but it would be hard to test. The way I tested the
existing code was not just with unit tests but also by loading it as a
path dependency and running `x test core` - the latter caught several
bugs the unit tests didn't catch (because I originally wrote the tests
wrong). So I am slightly nervous about adding a 16-byte version that is
much more poorly tested than the other intrinsics.
Same story as always, i.e. ABI mismatch:
- https://github.com/rust-lang/compiler-builtins/pull/462
- https://github.com/rust-lang/compiler-builtins/pull/466
- https://github.com/rust-lang/compiler-builtins/pull/513
I've made sure the changes work by rendering a Mandelbrot fractal:
```rust
#[arduino_hal::entry]
fn main() -> ! {
let dp = arduino_hal::Peripherals::take().unwrap();
let pins = arduino_hal::pins!(dp);
let mut serial = arduino_hal::default_serial!(dp, pins, 57600);
mandelbrot(&mut serial, 60, 40, -2.05, -1.12, 0.47, 1.12, 100);
loop {
//
}
}
fn mandelbrot<T>(
output: &mut T,
viewport_width: i64,
viewport_height: i64,
x1: f32,
y1: f32,
x2: f32,
y2: f32,
max_iterations: i64,
) where
T: uWrite,
{
for viewport_y in 0..viewport_height {
let y0 = y1 + (y2 - y1) * ((viewport_y as f32) / (viewport_height as f32));
for viewport_x in 0..viewport_width {
let x0 = x1 + (x2 - x1) * ((viewport_x as f32) / (viewport_width as f32));
let mut x = 0.0;
let mut y = 0.0;
let mut iterations = max_iterations;
while x * x + y * y <= 4.0 && iterations > 0 {
let xtemp = x * x - y * y + x0;
y = 2.0 * x * y + y0;
x = xtemp;
iterations -= 1;
}
let ch = "#%=-:,. "
.chars()
.nth((8.0 * ((iterations as f32) / (max_iterations as f32))) as _)
.unwrap();
_ = ufmt::uwrite!(output, "{}", ch);
}
_ = ufmt::uwriteln!(output, "");
}
}
```
... where without avr_skips, the code printed an image full of only `#`.
Note that because libgcc doesn't provide implementations for f64, using
those (e.g. swapping f32 to f64 in the code above) will cause linking to
fail:
```
undefined reference to `__divdf3'
undefined reference to `__muldf3'
undefined reference to `__gedf2'
undefined reference to `__fixunsdfsi'
undefined reference to `__gtdf2'
```
Ideally compiler-builtins could jump right in and provide those, but f64
also require a special calling convention which hasn't been yet exposed
through LLVM.
Note that because using 64-bit floats on an 8-bit target is a pretty
niche thing to do, and because f64 floats don't work correctly anyway at
the moment (due to this ABI mismatch), we're not actually breaking
anything by skipping those functions, since any code that currently uses
f64 on AVR works by accident.
Closes https://github.com/rust-lang/rust/issues/108489.
When enabled, the weak-intrinsics feature will cause all intrinsics
functions to be marked with weak linkage (i.e. `#[linkage = "weak"])
so that they can be replaced at link time by a stronger symbol.
This can be set to use C++ intrinsics from the compiler-rt library,
as it will avoid Rust's implementation replacing the compiler-rt
implementation as long as the compiler-rt symbols are linked as
strong symbols. Typically this requires the compiler-rt library to
be explicitly specified in the link command.
Addresses https://github.com/rust-lang/compiler-builtins/issues/525.
Without weak-intrinsics, from nm:
```
00000000 W __aeabi_memclr8 // Is explicitly weak
00000000 T __udivsi3 // Is not.
```
With weak-intrinsics, from nm:
```
00000000 W __aeabi_memclr8 // Is explicitly weak
00000000 W __udivsi3 // Is weak due to weak-intrinsics
```
This commit follows the same logic as:
- https://github.com/rust-lang/compiler-builtins/pull/462
- https://github.com/rust-lang/compiler-builtins/pull/466
I've tested the changes by preparing a simple program:
```rust
fn calc() -> ... {
let x = hint::black_box(4u...); // 4u8, 4u16, 4u32, 4u64, 4u128 + signed
let y = hint::black_box(1u32);
// x >> y
// x << y
}
fn main() -> ! {
let dp = arduino_hal::Peripherals::take().unwrap();
let pins = arduino_hal::pins!(dp);
let mut serial = arduino_hal::default_serial!(dp, pins, 57600);
for b in calc().to_le_bytes() {
_ = ufmt::uwrite!(&mut serial, "{} ", b);
}
_ = ufmt::uwriteln!(&mut serial, "");
loop {
//
}
}
```
... switching types & operators in `calc()`, and observing the results;
what I ended up with was:
```
u32 << u32 - ok
u64 << u32 - ok
u128 << u32 - error (undefined reference to `__ashlti3')
i32 >> u32 - ok
i64 >> u32 - ok
i128 >> u32 - error (undefined reference to `__ashrti3')
u32 >> u32 - ok
u64 >> u32 - ok
u128 >> u32 - error (undefined reference to `__lshrti3')
(where "ok" = compiles and returns correct results)
```
As with multiplication and division, so do in here 128-bit operations
not work, because avr-gcc's standard library doesn't provide them (at
the same time, requiring that specific calling convention, making it
pretty difficult for compiler-builtins to jump in).
I think 128-bit operations non-working on an 8-bit controller is an
acceptable trade-off - 😇 - and so the entire fix in here is
just about skipping those functions.
int_util.c includes stdlib.h if `_WIN32` is defined. When compiling
the UEFI targets with clang they are treated as Windows targets (e.g. if
the Rust target is x86_64-unknown-uefi, the clang target is
x86_64-unknown-windows-gnu). So stdlib.h gets included, even though we
are compilling with `-ffreestanding` and don't want stdlib.h to be
used. That file may not be present, or an incompatible version might be
installed leading to typedef redefinition errors.
The contents of stdlib.h aren't actually needed for these targets anyway
(due to `__STDC_HOSTED__` being 0), so create a minimal stdlib.h in
`build.rs` when `target_os == uefi` and add it to the include path.
The UEFI targets link with `/SAFESEH`. That requires that objects have a
symbol called [`@feat.00`]. Clang adds that symbol for COFF targets if
the input is a C file, but not if the input is an ASM file. That doesn't
prevent compiler_builtins or rustc from building, but using the
resulting rustc to compile something that references one of the objects
lacking `@feat.00` will result in a linker error.
Fix by removing all the `.S` implementations when `target_os == uefi`.
[`@feat.00`]: https://learn.microsoft.com/en-us/windows/win32/debug/pe-format#the-sxdata-section
The conversion functions from i128/u128 to f32/f64 have the
`unadjusted_on_win64` attribute, but it is disabled starting with
LLVM14. This seems to be the correct thing to do for Win64, but for some
reason x86_64-unknown-uefi is different, despite generally using the
same ABI as Win64.
The "xous" operating system is enturely Rust-based, meaning it has no
libc. Therefore, it relies on `compiler-builtins` for all intrinsics.
Unfortunately, there are not yet Rust equivalents for all C functions.
For example, triganometric functions are still missing. In the meantime,
enable C replacements for these functions so that Rust programs compiled
for Xous can call these functions.
Signed-off-by: Sean Cross <sean@xobs.io>
x86_64 can load unaligned words in a single cache line as fast as
aligned words. Even when crossing cache or page boundaries it is just as
fast to do an unaligned word read instead of multiple byte reads.
Also add a couple more tests & benchmarks.
