Merge pull request #56 from Kobzol/compiler-builtins

Compiler builtins push

Author: Kobzol

library/compiler-builtins/.github/workflows/main.yaml

@@ -5,15 +5,15 @@ on:
 
 concurrency:
   # Make sure that new pushes cancel running jobs
-  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}
   cancel-in-progress: true
 
 env:
   CARGO_TERM_COLOR: always
   RUSTDOCFLAGS: -Dwarnings
   RUSTFLAGS: -Dwarnings
   RUST_BACKTRACE: full
-  BENCHMARK_RUSTC: nightly-2025-01-16 # Pin the toolchain for reproducable results
+  BENCHMARK_RUSTC: nightly-2025-05-28 # Pin the toolchain for reproducable results
 
 jobs:
   # Determine which tests should be run based on changed files.
@@ -119,7 +119,6 @@ jobs:
         rustup update "$channel" --no-self-update
         rustup default "$channel"
         rustup target add "${{ matrix.target }}"
-        rustup component add llvm-tools-preview
     - uses: taiki-e/install-action@nextest
     - uses: Swatinem/rust-cache@v2
       with:
@@ -196,8 +195,14 @@ jobs:
 
   benchmarks:
     name: Benchmarks
-    runs-on: ubuntu-24.04
     timeout-minutes: 20
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+        - target: x86_64-unknown-linux-gnu
+          os: ubuntu-24.04
+    runs-on: ${{ matrix.os }}
     steps:
     - uses: actions/checkout@master
       with:
@@ -216,12 +221,14 @@ jobs:
         cargo binstall -y iai-callgrind-runner --version "$iai_version"
         sudo apt-get install valgrind
     - uses: Swatinem/rust-cache@v2
+      with:
+        key: ${{ matrix.target }}
 
     - name: Run icount benchmarks
       env:
         GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
         PR_NUMBER: ${{ github.event.pull_request.number }}
-      run: ./ci/bench-icount.sh
+      run: ./ci/bench-icount.sh ${{ matrix.target }}
 
     - name: Upload the benchmark baseline
       uses: actions/upload-artifact@v4

library/compiler-builtins/Cargo.toml

@@ -6,6 +6,7 @@ members = [
     "crates/libm-macros",
     "crates/musl-math-sys",
     "crates/panic-handler",
+    "crates/symbol-check",
     "crates/util",
     "libm",
     "libm-test",

library/compiler-builtins/PUBLISHING.md

@@ -14,3 +14,5 @@ greatly appreciated!
 7. Push the tag
 8. Push the commit
 9. Undo changes to `Cargo.toml` and the `libm` submodule
+
+FOO

library/compiler-builtins/README.md

@@ -5,7 +5,7 @@ This repository contains two main crates:
 * `compiler-builtins`: symbols that the compiler expects to be available at
   link time
 * `libm`: a Rust implementation of C math libraries, used to provide
-  implementations in `ocre`.
+  implementations in `core`.
 
 More details are at [compiler-builtins/README.md](compiler-builtins/README.md)
 and [libm/README.md](libm/README.md).

library/compiler-builtins/builtins-test-intrinsics/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
 name = "builtins-test-intrinsics"
 version = "0.1.0"
-edition = "2021"
+edition = "2024"
 publish = false
 license = "MIT OR Apache-2.0"
 

library/compiler-builtins/builtins-test-intrinsics/src/main.rs

@@ -13,11 +13,14 @@
 #![no_std]
 #![no_main]
 
+// Ensure this `compiler_builtins` gets used, rather than the version injected from the sysroot.
+extern crate compiler_builtins;
 extern crate panic_handler;
 
+// SAFETY: no definitions, only used for linking
 #[cfg(all(not(thumb), not(windows), not(target_arch = "wasm32")))]
 #[link(name = "c")]
-extern "C" {}
+unsafe extern "C" {}
 
 // Every function in this module maps will be lowered to an intrinsic by LLVM, if the platform
 // doesn't have native support for the operation used in the function. ARM has a naming convention
@@ -651,22 +654,23 @@ fn something_with_a_dtor(f: &dyn Fn()) {
 
 #[unsafe(no_mangle)]
 #[cfg(not(thumb))]
-fn main(_argc: core::ffi::c_int, _argv: *const *const u8) -> core::ffi::c_int {
+extern "C" fn main(_argc: core::ffi::c_int, _argv: *const *const u8) -> core::ffi::c_int {
     run();
     0
 }
 
 #[unsafe(no_mangle)]
 #[cfg(thumb)]
-pub fn _start() -> ! {
+extern "C" fn _start() -> ! {
     run();
     loop {}
 }
 
+// SAFETY: no definitions, only used for linking
 #[cfg(windows)]
 #[link(name = "kernel32")]
 #[link(name = "msvcrt")]
-extern "C" {}
+unsafe extern "C" {}
 
 // ARM targets need these symbols
 #[unsafe(no_mangle)]

library/compiler-builtins/builtins-test/Cargo.toml

@@ -10,19 +10,19 @@ license = "MIT AND Apache-2.0 WITH LLVM-exception AND (MIT OR Apache-2.0)"
 # For fuzzing tests we want a deterministic seedable RNG. We also eliminate potential
 # problems with system RNGs on the variety of platforms this crate is tested on.
 # `xoshiro128**` is used for its quality, size, and speed at generating `u32` shift amounts.
-rand_xoshiro = "0.6"
+rand_xoshiro = "0.7"
 # To compare float builtins against
-rustc_apfloat = "0.2.1"
+rustc_apfloat = "0.2.2"
 # Really a dev dependency, but dev dependencies can't be optional
-iai-callgrind = { version = "0.14.0", optional = true }
+iai-callgrind = { version = "0.14.1", optional = true }
 
 [dependencies.compiler_builtins]
 path = "../compiler-builtins"
 default-features = false
 features = ["unstable-public-internals"]
 
 [dev-dependencies]
-criterion = { version = "0.5.1", default-features = false, features = ["cargo_bench_support"] }
+criterion = { version = "0.6.0", default-features = false, features = ["cargo_bench_support"] }
 paste = "1.0.15"
 
 [target.'cfg(all(target_arch = "arm", not(any(target_env = "gnu", target_env = "musl")), target_os = "linux"))'.dev-dependencies]

library/compiler-builtins/builtins-test/benches/float_cmp.rs

@@ -1,27 +1,38 @@
 #![cfg_attr(f128_enabled, feature(f128))]
 
 use builtins_test::float_bench;
-use compiler_builtins::float::cmp;
+use compiler_builtins::float::cmp::{self, CmpResult};
 use criterion::{Criterion, criterion_main};
 
 /// `gt` symbols are allowed to return differing results, they just get compared
 /// to 0.
-fn gt_res_eq(a: i32, b: i32) -> bool {
+fn gt_res_eq(mut a: CmpResult, mut b: CmpResult) -> bool {
+    // FIXME: Our CmpResult used to be `i32`, but GCC/LLVM expect `isize`. on 64-bit platforms,
+    // this means the top half of the word may be garbage if built with an old version of
+    // `compiler-builtins`, so add a hack around this.
+    //
+    // This can be removed once a version of `compiler-builtins` with the return type fix makes
+    // it upstream.
+    if size_of::<CmpResult>() == 8 {
+        a = a as i32 as CmpResult;
+        b = b as i32 as CmpResult;
+    }
+
     let a_lt_0 = a <= 0;
     let b_lt_0 = b <= 0;
     (a_lt_0 && b_lt_0) || (!a_lt_0 && !b_lt_0)
 }
 
 float_bench! {
     name: cmp_f32_gt,
-    sig: (a: f32, b: f32) -> i32,
+    sig: (a: f32, b: f32) -> CmpResult,
     crate_fn: cmp::__gtsf2,
     sys_fn: __gtsf2,
     sys_available: all(),
     output_eq: gt_res_eq,
     asm: [
         #[cfg(target_arch = "x86_64")] {
-            let ret: i32;
+            let ret: CmpResult;
             asm!(
                 "xor     {ret:e}, {ret:e}",
                 "ucomiss {a}, {b}",
@@ -36,7 +47,7 @@ float_bench! {
         };
 
         #[cfg(target_arch = "aarch64")] {
-            let ret: i32;
+            let ret: CmpResult;
             asm!(
                 "fcmp    {a:s}, {b:s}",
                 "cset    {ret:w}, gt",
@@ -53,13 +64,13 @@ float_bench! {
 
 float_bench! {
     name: cmp_f32_unord,
-    sig: (a: f32, b: f32) -> i32,
+    sig: (a: f32, b: f32) -> CmpResult,
     crate_fn: cmp::__unordsf2,
     sys_fn: __unordsf2,
     sys_available: all(),
     asm: [
         #[cfg(target_arch = "x86_64")] {
-            let ret: i32;
+            let ret: CmpResult;
             asm!(
                 "xor     {ret:e}, {ret:e}",
                 "ucomiss {a}, {b}",
@@ -74,7 +85,7 @@ float_bench! {
         };
 
         #[cfg(target_arch = "aarch64")] {
-            let ret: i32;
+            let ret: CmpResult;
             asm!(
                 "fcmp    {a:s}, {b:s}",
                 "cset    {ret:w}, vs",
@@ -91,14 +102,14 @@ float_bench! {
 
 float_bench! {
     name: cmp_f64_gt,
-    sig: (a: f64, b: f64) -> i32,
+    sig: (a: f64, b: f64) -> CmpResult,
     crate_fn: cmp::__gtdf2,
     sys_fn: __gtdf2,
     sys_available: all(),
     output_eq: gt_res_eq,
     asm: [
         #[cfg(target_arch = "x86_64")] {
-            let ret: i32;
+            let ret: CmpResult;
             asm!(
                 "xor     {ret:e}, {ret:e}",
                 "ucomisd {a}, {b}",
@@ -113,7 +124,7 @@ float_bench! {
         };
 
         #[cfg(target_arch = "aarch64")] {
-            let ret: i32;
+            let ret: CmpResult;
             asm!(
                 "fcmp    {a:d}, {b:d}",
                 "cset {ret:w}, gt",
@@ -130,13 +141,13 @@ float_bench! {
 
 float_bench! {
     name: cmp_f64_unord,
-    sig: (a: f64, b: f64) -> i32,
+    sig: (a: f64, b: f64) -> CmpResult,
     crate_fn: cmp::__unorddf2,
     sys_fn: __unorddf2,
     sys_available: all(),
     asm: [
         #[cfg(target_arch = "x86_64")] {
-            let ret: i32;
+            let ret: CmpResult;
             asm!(
                 "xor     {ret:e}, {ret:e}",
                 "ucomisd {a}, {b}",
@@ -151,7 +162,7 @@ float_bench! {
         };
 
         #[cfg(target_arch = "aarch64")] {
-            let ret: i32;
+            let ret: CmpResult;
             asm!(
                 "fcmp    {a:d}, {b:d}",
                 "cset    {ret:w}, vs",
@@ -168,7 +179,7 @@ float_bench! {
 
 float_bench! {
     name: cmp_f128_gt,
-    sig: (a: f128, b: f128) -> i32,
+    sig: (a: f128, b: f128) -> CmpResult,
     crate_fn: cmp::__gttf2,
     crate_fn_ppc: cmp::__gtkf2,
     sys_fn: __gttf2,
@@ -180,7 +191,7 @@ float_bench! {
 
 float_bench! {
     name: cmp_f128_unord,
-    sig: (a: f128, b: f128) -> i32,
+    sig: (a: f128, b: f128) -> CmpResult,
     crate_fn: cmp::__unordtf2,
     crate_fn_ppc: cmp::__unordkf2,
     sys_fn: __unordtf2,

library/compiler-builtins/builtins-test/src/bench.rs

@@ -358,8 +358,8 @@ impl_testio!(float f16);
 impl_testio!(float f32, f64);
 #[cfg(f128_enabled)]
 impl_testio!(float f128);
-impl_testio!(int i16, i32, i64, i128);
-impl_testio!(int u16, u32, u64, u128);
+impl_testio!(int i8, i16, i32, i64, i128, isize);
+impl_testio!(int u8, u16, u32, u64, u128, usize);
 impl_testio!((float, int)(f32, i32));
 impl_testio!((float, int)(f64, i32));
 #[cfg(f128_enabled)]