Sync rustc_codegen_gcc

compiler/rustc_codegen_gcc/.github/workflows/ci.yml

@@ -11,7 +11,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        libgccjit_version: ["libgccjit.so", "libgccjit_without_int128.so"]
+        libgccjit_version: ["libgccjit.so", "libgccjit_without_int128.so", "libgccjit12.so"]
 
     steps:
     - uses: actions/checkout@v2
@@ -78,12 +78,21 @@ jobs:
         key: ${{ runner.os }}-cargo-build-target-${{ hashFiles('rust-toolchain') }}
 
     - name: Build
+      if: matrix.libgccjit_version != 'libgccjit12.so'
       run: |
         ./prepare_build.sh
         ./build.sh
         cargo test
         ./clean_all.sh
 
+    - name: Build
+      if: matrix.libgccjit_version == 'libgccjit12.so'
+      run: |
+        ./prepare_build.sh
+        ./build.sh --no-default-features
+        cargo test --no-default-features
+        ./clean_all.sh
+
     - name: Prepare dependencies
       run: |
         git config --global user.email "[email protected]"
@@ -98,6 +107,7 @@ jobs:
         args: --release
 
     - name: Test
+      if: matrix.libgccjit_version != 'libgccjit12.so'
       run: |
         # Enable backtraces for easier debugging
         export RUST_BACKTRACE=1
@@ -107,3 +117,15 @@ jobs:
         export RUN_RUNS=2
 
         ./test.sh --release
+
+    - name: Test
+      if: matrix.libgccjit_version == 'libgccjit12.so'
+      run: |
+        # Enable backtraces for easier debugging
+        export RUST_BACKTRACE=1
+
+        # Reduce amount of benchmark runs as they are slow
+        export COMPILE_RUNS=2
+        export RUN_RUNS=2
+
+        ./test.sh --release --no-default-features

compiler/rustc_codegen_gcc/.gitignore

@@ -13,9 +13,13 @@ perf.data.old
 /rust
 /simple-raytracer
 /regex
+/rand
 gimple*
 *asm
 res
 test-backend
 gcc_path
 benchmarks
+tools/llvm-project
+tools/llvmint
+tools/llvmint-2

compiler/rustc_codegen_gcc/.rustfmt.toml

@@ -0,0 +1 @@
+disable_all_formatting = true

compiler/rustc_codegen_gcc/Cargo.lock

@@ -41,15 +41,15 @@ dependencies = [
 [[package]]
 name = "gccjit"
 version = "1.0.0"
-source = "git+https://github.com/antoyo/gccjit.rs#bdecdecfb8a02ec861a39a350f990faa33bd31c3"
+source = "git+https://github.com/antoyo/gccjit.rs#bdb86fb5092895ff5589726b33250010c64d93f6"
 dependencies = [
  "gccjit_sys",
 ]
 
 [[package]]
 name = "gccjit_sys"
 version = "0.0.1"
-source = "git+https://github.com/antoyo/gccjit.rs#bdecdecfb8a02ec861a39a350f990faa33bd31c3"
+source = "git+https://github.com/antoyo/gccjit.rs#bdb86fb5092895ff5589726b33250010c64d93f6"
 dependencies = [
  "libc 0.1.12",
 ]

compiler/rustc_codegen_gcc/Cargo.toml

@@ -9,9 +9,17 @@ license = "MIT OR Apache-2.0"
 crate-type = ["dylib"]
 
 [[test]]
-name = "lang_tests"
-path = "tests/lib.rs"
+name = "lang_tests_debug"
+path = "tests/lang_tests_debug.rs"
 harness = false
+[[test]]
+name = "lang_tests_release"
+path = "tests/lang_tests_release.rs"
+harness = false
+
+[features]
+default = ["master"]
+master = ["gccjit/master"]
 
 [dependencies]
 gccjit = { git = "https://github.com/antoyo/gccjit.rs" }

compiler/rustc_codegen_gcc/build.sh

@@ -1,11 +1,13 @@
-#!/bin/bash
+#!/usr/bin/env bash
 
 #set -x
 set -e
 
 codegen_channel=debug
 sysroot_channel=debug
 
+flags=
+
 while [[ $# -gt 0 ]]; do
     case $1 in
         --release)
@@ -16,6 +18,15 @@ while [[ $# -gt 0 ]]; do
             sysroot_channel=release
             shift
             ;;
+        --no-default-features)
+            flags="$flags --no-default-features"
+            shift
+            ;;
+        --features)
+            shift
+            flags="$flags --features $1"
+            shift
+            ;;
         *)
             echo "Unknown option $1"
             exit 1
@@ -33,21 +44,13 @@ fi
 export LD_LIBRARY_PATH="$GCC_PATH"
 export LIBRARY_PATH="$GCC_PATH"
 
-features=
-
-if [[ "$1" == "--features" ]]; then
-    shift
-    features="--features $1"
-    shift
-fi
-
 if [[ "$codegen_channel" == "release" ]]; then
     export CHANNEL='release'
-    CARGO_INCREMENTAL=1 cargo rustc --release $features
+    CARGO_INCREMENTAL=1 cargo rustc --release $flags
 else
     echo $LD_LIBRARY_PATH
     export CHANNEL='debug'
-    cargo rustc $features
+    cargo rustc $flags
 fi
 
 source config.sh

compiler/rustc_codegen_gcc/build_sysroot/build_sysroot.sh

@@ -1,4 +1,4 @@
-#!/bin/bash
+#!/usr/bin/env bash
 
 # Requires the CHANNEL env var to be set to `debug` or `release.`
 

compiler/rustc_codegen_gcc/build_sysroot/prepare_sysroot_src.sh

@@ -1,4 +1,4 @@
-#!/bin/bash
+#!/usr/bin/env bash
 set -e
 cd $(dirname "$0")
 

compiler/rustc_codegen_gcc/cargo.sh

@@ -1,4 +1,4 @@
-#!/bin/bash
+#!/usr/bin/env bash
 
 if [ -z $CHANNEL ]; then
 export CHANNEL='debug'
@@ -20,4 +20,4 @@ fi
 cmd=$1
 shift
 
-RUSTDOCFLAGS="$RUSTFLAGS" cargo +${TOOLCHAIN} $cmd --target $TARGET_TRIPLE $@
+RUSTDOCFLAGS="$RUSTFLAGS" cargo +${TOOLCHAIN} $cmd $@

compiler/rustc_codegen_gcc/clean_all.sh

@@ -1,5 +1,6 @@
-#!/bin/bash --verbose
+#!/usr/bin/env bash
 set -e
+set -v
 
 rm -rf target/ build_sysroot/{sysroot/,sysroot_src/,target/,Cargo.lock} perf.data{,.old}
 rm -rf regex/ simple-raytracer/

compiler/rustc_codegen_gcc/config.sh

@@ -2,7 +2,7 @@ set -e
 
 export CARGO_INCREMENTAL=0
 
-if [ -f ./gcc_path ]; then 
+if [ -f ./gcc_path ]; then
     export GCC_PATH=$(cat gcc_path)
 else
     echo 'Please put the path to your custom build of libgccjit in the file `gcc_path`, see Readme.md for details'
@@ -38,7 +38,7 @@ if [[ "$HOST_TRIPLE" != "$TARGET_TRIPLE" ]]; then
    fi
 fi
 
-export RUSTFLAGS="$linker -Cpanic=abort -Csymbol-mangling-version=v0 -Cdebuginfo=2 -Clto=off -Zpanic-abort-tests -Zcodegen-backend=$(pwd)/target/${CHANNEL:-debug}/librustc_codegen_gcc.$dylib_ext --sysroot $(pwd)/build_sysroot/sysroot"
+export RUSTFLAGS="$CG_RUSTFLAGS $linker -Cpanic=abort -Csymbol-mangling-version=v0 -Cdebuginfo=2 -Clto=off -Zpanic-abort-tests -Zcodegen-backend=$(pwd)/target/${CHANNEL:-debug}/librustc_codegen_gcc.$dylib_ext --sysroot $(pwd)/build_sysroot/sysroot"
 
 # FIXME(antoyo): remove once the atomic shim is gone
 if [[ `uname` == 'Darwin' ]]; then

compiler/rustc_codegen_gcc/crate_patches/0002-rand-Disable-failing-test.patch

@@ -0,0 +1,32 @@
+From a8fb97120d71252538b6b026695df40d02696bdb Mon Sep 17 00:00:00 2001
+From: bjorn3 <[email protected]>
+Date: Sat, 15 Aug 2020 20:04:38 +0200
+Subject: [PATCH] [rand] Disable failing test
+
+---
+ src/distributions/uniform.rs | 3 ++-
+ 1 file changed, 2 insertions(+), 1 deletion(-)
+
+diff --git a/src/distributions/uniform.rs b/src/distributions/uniform.rs
+index 480b859..c80bb6f 100644
+--- a/src/distributions/uniform.rs
++++ b/src/distributions/uniform.rs
+@@ -1085,7 +1085,7 @@ mod tests {
+             _ => panic!("`UniformDurationMode` was not serialized/deserialized correctly")
+         }
+     }
+-    
++
+     #[test]
+     #[cfg(feature = "serde1")]
+     fn test_uniform_serialization() {
+@@ -1314,6 +1314,7 @@ mod tests {
+         not(target_arch = "wasm32"),
+         not(target_arch = "asmjs")
+     ))]
++    #[ignore] // FIXME
+     fn test_float_assertions() {
+         use super::SampleUniform;
+         use std::panic::catch_unwind;
+-- 
+2.20.1

compiler/rustc_codegen_gcc/example/std_example.rs

@@ -93,9 +93,10 @@ fn main() {
 
     println!("{:?}", std::intrinsics::caller_location());
 
-    /*unsafe {
+    #[cfg(feature="master")]
+    unsafe {
         test_simd();
-    }*/
+    }
 
     Box::pin(move |mut _task_context| {
         yield ();
@@ -104,15 +105,16 @@ fn main() {
     println!("End");
 }
 
-/*#[target_feature(enable = "sse2")]
+#[cfg(feature="master")]
+#[target_feature(enable = "sse2")]
 unsafe fn test_simd() {
     let x = _mm_setzero_si128();
     let y = _mm_set1_epi16(7);
     let or = _mm_or_si128(x, y);
     let cmp_eq = _mm_cmpeq_epi8(y, y);
     let cmp_lt = _mm_cmplt_epi8(y, y);
 
-    /*assert_eq!(std::mem::transmute::<_, [u16; 8]>(or), [7, 7, 7, 7, 7, 7, 7, 7]);
+    assert_eq!(std::mem::transmute::<_, [u16; 8]>(or), [7, 7, 7, 7, 7, 7, 7, 7]);
     assert_eq!(std::mem::transmute::<_, [u16; 8]>(cmp_eq), [0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff]);
     assert_eq!(std::mem::transmute::<_, [u16; 8]>(cmp_lt), [0, 0, 0, 0, 0, 0, 0, 0]);
 
@@ -124,14 +126,15 @@ unsafe fn test_simd() {
     test_mm_cvtepi8_epi16();
     test_mm_cvtsi128_si64();
 
-    // FIXME(#666) implement `#[rustc_arg_required_const(..)]` support
-    //test_mm_extract_epi8();
+    test_mm_extract_epi8();
+    test_mm_insert_epi16();
 
     let mask1 = _mm_movemask_epi8(dbg!(_mm_setr_epi8(255u8 as i8, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0)));
-    assert_eq!(mask1, 1);*/
-}*/
+    assert_eq!(mask1, 1);
+}
 
-/*#[target_feature(enable = "sse2")]
+#[cfg(feature="master")]
+#[target_feature(enable = "sse2")]
 unsafe fn test_mm_slli_si128() {
     #[rustfmt::skip]
     let a = _mm_setr_epi8(
@@ -155,22 +158,10 @@ unsafe fn test_mm_slli_si128() {
     );
     let r = _mm_slli_si128(a, 16);
     assert_eq_m128i(r, _mm_set1_epi8(0));
-
-    #[rustfmt::skip]
-    let a = _mm_setr_epi8(
-        1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16,
-    );
-    let r = _mm_slli_si128(a, -1);
-    assert_eq_m128i(_mm_set1_epi8(0), r);
-
-    #[rustfmt::skip]
-    let a = _mm_setr_epi8(
-        1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16,
-    );
-    let r = _mm_slli_si128(a, -0x80000000);
-    assert_eq_m128i(r, _mm_set1_epi8(0));
 }
 
+
+#[cfg(feature="master")]
 #[target_feature(enable = "sse2")]
 unsafe fn test_mm_movemask_epi8() {
     #[rustfmt::skip]
@@ -184,6 +175,7 @@ unsafe fn test_mm_movemask_epi8() {
     assert_eq!(r, 0b10100100_00100101);
 }
 
+#[cfg(feature="master")]
 #[target_feature(enable = "avx2")]
 unsafe fn test_mm256_movemask_epi8() {
     let a = _mm256_set1_epi8(-1);
@@ -192,6 +184,7 @@ unsafe fn test_mm256_movemask_epi8() {
     assert_eq!(r, e);
 }
 
+#[cfg(feature="master")]
 #[target_feature(enable = "sse2")]
 unsafe fn test_mm_add_epi8() {
     let a = _mm_setr_epi8(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15);
@@ -207,6 +200,7 @@ unsafe fn test_mm_add_epi8() {
     assert_eq_m128i(r, e);
 }
 
+#[cfg(feature="master")]
 #[target_feature(enable = "sse2")]
 unsafe fn test_mm_add_pd() {
     let a = _mm_setr_pd(1.0, 2.0);
@@ -215,25 +209,29 @@ unsafe fn test_mm_add_pd() {
     assert_eq_m128d(r, _mm_setr_pd(6.0, 12.0));
 }
 
+#[cfg(feature="master")]
 fn assert_eq_m128i(x: std::arch::x86_64::__m128i, y: std::arch::x86_64::__m128i) {
     unsafe {
         assert_eq!(std::mem::transmute::<_, [u8; 16]>(x), std::mem::transmute::<_, [u8; 16]>(y));
     }
 }
 
+#[cfg(feature="master")]
 #[target_feature(enable = "sse2")]
 pub unsafe fn assert_eq_m128d(a: __m128d, b: __m128d) {
     if _mm_movemask_pd(_mm_cmpeq_pd(a, b)) != 0b11 {
         panic!("{:?} != {:?}", a, b);
     }
 }
 
+#[cfg(feature="master")]
 #[target_feature(enable = "sse2")]
 unsafe fn test_mm_cvtsi128_si64() {
     let r = _mm_cvtsi128_si64(std::mem::transmute::<[i64; 2], _>([5, 0]));
     assert_eq!(r, 5);
 }
 
+#[cfg(feature="master")]
 #[target_feature(enable = "sse4.1")]
 unsafe fn test_mm_cvtepi8_epi16() {
     let a = _mm_set1_epi8(10);
@@ -246,6 +244,7 @@ unsafe fn test_mm_cvtepi8_epi16() {
     assert_eq_m128i(r, e);
 }
 
+#[cfg(feature="master")]
 #[target_feature(enable = "sse4.1")]
 unsafe fn test_mm_extract_epi8() {
     #[rustfmt::skip]
@@ -254,10 +253,19 @@ unsafe fn test_mm_extract_epi8() {
         8, 9, 10, 11, 12, 13, 14, 15
     );
     let r1 = _mm_extract_epi8(a, 0);
-    let r2 = _mm_extract_epi8(a, 19);
+    let r2 = _mm_extract_epi8(a, 3);
     assert_eq!(r1, 0xFF);
     assert_eq!(r2, 3);
-}*/
+}
+
+#[cfg(all(feature="master", target_arch = "x86_64"))]
+#[target_feature(enable = "sse2")]
+unsafe fn test_mm_insert_epi16() {
+    let a = _mm_setr_epi16(0, 1, 2, 3, 4, 5, 6, 7);
+    let r = _mm_insert_epi16(a, 9, 0);
+    let e = _mm_setr_epi16(9, 1, 2, 3, 4, 5, 6, 7);
+    assert_eq_m128i(r, e);
+}
 
 #[derive(PartialEq)]
 enum LoopState {

compiler/rustc_codegen_gcc/patches/0024-core-Disable-portable-simd-test.patch

@@ -7,167 +7,6 @@ Subject: [PATCH] [core] Disable portable-simd test
  library/core/tests/lib.rs | 1 -
  1 file changed, 1 deletion(-)
 
-diff --git a/library/core/src/lib.rs b/library/core/src/lib.rs
-index aa1ad93..95fbf55 100644
---- a/library/core/src/lib.rs
-+++ b/library/core/src/lib.rs
-@@ -398,23 +398,4 @@ pub mod arch {
-     }
- }
-
--// Pull in the `core_simd` crate directly into libcore. The contents of
--// `core_simd` are in a different repository: rust-lang/portable-simd.
--//
--// `core_simd` depends on libcore, but the contents of this module are
--// set up in such a way that directly pulling it here works such that the
--// crate uses this crate as its libcore.
--#[path = "../../portable-simd/crates/core_simd/src/mod.rs"]
--#[allow(missing_debug_implementations, dead_code, unsafe_op_in_unsafe_fn, unused_unsafe)]
--#[allow(rustdoc::bare_urls)]
--#[unstable(feature = "portable_simd", issue = "86656")]
--mod core_simd;
--
--#[doc = include_str!("../../portable-simd/crates/core_simd/src/core_simd_docs.md")]
--#[unstable(feature = "portable_simd", issue = "86656")]
--pub mod simd {
--    #[unstable(feature = "portable_simd", issue = "86656")]
--    pub use crate::core_simd::simd::*;
--}
--
- include!("primitive_docs.rs");
-diff --git a/library/core/src/slice/mod.rs b/library/core/src/slice/mod.rs
-index cd38c3a..ad632dc 100644
---- a/library/core/src/slice/mod.rs
-+++ b/library/core/src/slice/mod.rs
-@@ -17,6 +17,5 @@ use crate::ptr;
- use crate::result::Result;
- use crate::result::Result::{Err, Ok};
--use crate::simd::{self, Simd};
- use crate::slice;
-
- #[unstable(
-@@ -3475,121 +3474,6 @@ impl<T> [T] {
-         }
-     }
-
--    /// Split a slice into a prefix, a middle of aligned SIMD types, and a suffix.
--    ///
--    /// This is a safe wrapper around [`slice::align_to`], so has the same weak
--    /// postconditions as that method.  You're only assured that
--    /// `self.len() == prefix.len() + middle.len() * LANES + suffix.len()`.
--    ///
--    /// Notably, all of the following are possible:
--    /// - `prefix.len() >= LANES`.
--    /// - `middle.is_empty()` despite `self.len() >= 3 * LANES`.
--    /// - `suffix.len() >= LANES`.
--    ///
--    /// That said, this is a safe method, so if you're only writing safe code,
--    /// then this can at most cause incorrect logic, not unsoundness.
--    ///
--    /// # Panics
--    ///
--    /// This will panic if the size of the SIMD type is different from
--    /// `LANES` times that of the scalar.
--    ///
--    /// At the time of writing, the trait restrictions on `Simd<T, LANES>` keeps
--    /// that from ever happening, as only power-of-two numbers of lanes are
--    /// supported.  It's possible that, in the future, those restrictions might
--    /// be lifted in a way that would make it possible to see panics from this
--    /// method for something like `LANES == 3`.
--    ///
--    /// # Examples
--    ///
--    /// ```
--    /// #![feature(portable_simd)]
--    ///
--    /// let short = &[1, 2, 3];
--    /// let (prefix, middle, suffix) = short.as_simd::<4>();
--    /// assert_eq!(middle, []); // Not enough elements for anything in the middle
--    ///
--    /// // They might be split in any possible way between prefix and suffix
--    /// let it = prefix.iter().chain(suffix).copied();
--    /// assert_eq!(it.collect::<Vec<_>>(), vec![1, 2, 3]);
--    ///
--    /// fn basic_simd_sum(x: &[f32]) -> f32 {
--    ///     use std::ops::Add;
--    ///     use std::simd::f32x4;
--    ///     let (prefix, middle, suffix) = x.as_simd();
--    ///     let sums = f32x4::from_array([
--    ///         prefix.iter().copied().sum(),
--    ///         0.0,
--    ///         0.0,
--    ///         suffix.iter().copied().sum(),
--    ///     ]);
--    ///     let sums = middle.iter().copied().fold(sums, f32x4::add);
--    ///     sums.reduce_sum()
--    /// }
--    ///
--    /// let numbers: Vec<f32> = (1..101).map(|x| x as _).collect();
--    /// assert_eq!(basic_simd_sum(&numbers[1..99]), 4949.0);
--    /// ```
--    #[unstable(feature = "portable_simd", issue = "86656")]
--    pub fn as_simd<const LANES: usize>(&self) -> (&[T], &[Simd<T, LANES>], &[T])
--    where
--        Simd<T, LANES>: AsRef<[T; LANES]>,
--        T: simd::SimdElement,
--        simd::LaneCount<LANES>: simd::SupportedLaneCount,
--    {
--        // These are expected to always match, as vector types are laid out like
--        // arrays per <https://llvm.org/docs/LangRef.html#vector-type>, but we
--        // might as well double-check since it'll optimize away anyhow.
--        assert_eq!(mem::size_of::<Simd<T, LANES>>(), mem::size_of::<[T; LANES]>());
--
--        // SAFETY: The simd types have the same layout as arrays, just with
--        // potentially-higher alignment, so the de-facto transmutes are sound.
--        unsafe { self.align_to() }
--    }
--
--    /// Split a slice into a prefix, a middle of aligned SIMD types, and a suffix.
--    ///
--    /// This is a safe wrapper around [`slice::align_to_mut`], so has the same weak
--    /// postconditions as that method.  You're only assured that
--    /// `self.len() == prefix.len() + middle.len() * LANES + suffix.len()`.
--    ///
--    /// Notably, all of the following are possible:
--    /// - `prefix.len() >= LANES`.
--    /// - `middle.is_empty()` despite `self.len() >= 3 * LANES`.
--    /// - `suffix.len() >= LANES`.
--    ///
--    /// That said, this is a safe method, so if you're only writing safe code,
--    /// then this can at most cause incorrect logic, not unsoundness.
--    ///
--    /// This is the mutable version of [`slice::as_simd`]; see that for examples.
--    ///
--    /// # Panics
--    ///
--    /// This will panic if the size of the SIMD type is different from
--    /// `LANES` times that of the scalar.
--    ///
--    /// At the time of writing, the trait restrictions on `Simd<T, LANES>` keeps
--    /// that from ever happening, as only power-of-two numbers of lanes are
--    /// supported.  It's possible that, in the future, those restrictions might
--    /// be lifted in a way that would make it possible to see panics from this
--    /// method for something like `LANES == 3`.
--    #[unstable(feature = "portable_simd", issue = "86656")]
--    pub fn as_simd_mut<const LANES: usize>(&mut self) -> (&mut [T], &mut [Simd<T, LANES>], &mut [T])
--    where
--        Simd<T, LANES>: AsMut<[T; LANES]>,
--        T: simd::SimdElement,
--        simd::LaneCount<LANES>: simd::SupportedLaneCount,
--    {
--        // These are expected to always match, as vector types are laid out like
--        // arrays per <https://llvm.org/docs/LangRef.html#vector-type>, but we
--        // might as well double-check since it'll optimize away anyhow.
--        assert_eq!(mem::size_of::<Simd<T, LANES>>(), mem::size_of::<[T; LANES]>());
--
--        // SAFETY: The simd types have the same layout as arrays, just with
--        // potentially-higher alignment, so the de-facto transmutes are sound.
--        unsafe { self.align_to_mut() }
--    }
--
-     /// Checks if the elements of this slice are sorted.
-     ///
-     /// That is, for each element `a` and its following element `b`, `a <= b` must hold. If the
 diff --git a/library/core/tests/lib.rs b/library/core/tests/lib.rs
 index 06c7be0..359e2e7 100644
 --- a/library/core/tests/lib.rs
@@ -188,41 +27,3 @@ index 06c7be0..359e2e7 100644
  mod slice;
  mod str;
  mod str_lossy;
-diff --git a/library/std/src/lib.rs b/library/std/src/lib.rs
-index 5dc586d..b6fc48f 100644
---- a/library/std/src/lib.rs
-+++ b/library/std/src/lib.rs
-@@ -312,6 +312,5 @@
- #![feature(panic_can_unwind)]
- #![feature(panic_unwind)]
- #![feature(platform_intrinsics)]
--#![feature(portable_simd)]
- #![feature(prelude_import)]
- #![feature(ptr_as_uninit)]
-@@ -508,23 +508,6 @@ pub mod time;
- #[unstable(feature = "once_cell", issue = "74465")]
- pub mod lazy;
-
--// Pull in `std_float` crate  into libstd. The contents of
--// `std_float` are in a different repository: rust-lang/portable-simd.
--#[path = "../../portable-simd/crates/std_float/src/lib.rs"]
--#[allow(missing_debug_implementations, dead_code, unsafe_op_in_unsafe_fn, unused_unsafe)]
--#[allow(rustdoc::bare_urls)]
--#[unstable(feature = "portable_simd", issue = "86656")]
--mod std_float;
--
--#[doc = include_str!("../../portable-simd/crates/core_simd/src/core_simd_docs.md")]
--#[unstable(feature = "portable_simd", issue = "86656")]
--pub mod simd {
--    #[doc(inline)]
--    pub use crate::std_float::StdFloat;
--    #[doc(inline)]
--    pub use core::simd::*;
--}
--
- #[stable(feature = "futures_api", since = "1.36.0")]
- pub mod task {
-     //! Types and Traits for working with asynchronous tasks.
---
-2.26.2.7.g19db9cfb68
-

compiler/rustc_codegen_gcc/prepare.sh

@@ -1,10 +1,18 @@
-#!/bin/bash --verbose
+#!/usr/bin/env bash
 set -e
+set -v
 
 source prepare_build.sh
 
 cargo install hyperfine || echo "Skipping hyperfine install"
 
+git clone https://github.com/rust-random/rand.git || echo "rust-random/rand has already been cloned"
+pushd rand
+git checkout -- .
+git checkout 0f933f9c7176e53b2a3c7952ded484e1783f0bf1
+git am ../crate_patches/*-rand-*.patch
+popd
+
 git clone https://github.com/rust-lang/regex.git || echo "rust-lang/regex has already been cloned"
 pushd regex
 git checkout -- .

compiler/rustc_codegen_gcc/prepare_build.sh

@@ -1,4 +1,5 @@
-#!/bin/bash --verbose
+#!/usr/bin/env bash
 set -e
+set -v
 
 ./build_sysroot/prepare_sysroot_src.sh

compiler/rustc_codegen_gcc/rust-toolchain

@@ -1,3 +1,3 @@
 [toolchain]
-channel = "nightly-2022-03-26"
+channel = "nightly-2022-06-06"
 components = ["rust-src", "rustc-dev", "llvm-tools-preview"]

compiler/rustc_codegen_gcc/rustc_patches/compile_test.patch

@@ -0,0 +1,14 @@
+diff --git a/src/tools/compiletest/src/header.rs b/src/tools/compiletest/src/header.rs
+index 887d27fd6dca4..2c2239f2b83d1 100644
+--- a/src/tools/compiletest/src/header.rs
++++ b/src/tools/compiletest/src/header.rs
+@@ -806,8 +806,8 @@ pub fn make_test_description<R: Read>(
+     cfg: Option<&str>,
+ ) -> test::TestDesc {
+     let mut ignore = false;
+     #[cfg(not(bootstrap))]
+-    let ignore_message: Option<String> = None;
++    let ignore_message: Option<&str> = None;
+     let mut should_fail = false;
+
+     let rustc_has_profiler_support = env::var_os("RUSTC_PROFILER_SUPPORT").is_some();

compiler/rustc_codegen_gcc/rustup.sh

@@ -1,4 +1,4 @@
-#!/bin/bash
+#!/usr/bin/env bash
 
 set -e
 

compiler/rustc_codegen_gcc/src/asm.rs

@@ -13,6 +13,7 @@ use std::borrow::Cow;
 use crate::builder::Builder;
 use crate::context::CodegenCx;
 use crate::type_of::LayoutGccExt;
+use crate::callee::get_fn;
 
 
 // Rust asm! and GCC Extended Asm semantics differ substantially.
@@ -116,7 +117,6 @@ impl<'a, 'gcc, 'tcx> AsmBuilderMethods<'tcx> for Builder<'a, 'gcc, 'tcx> {
         let asm_arch = self.tcx.sess.asm_arch.unwrap();
         let is_x86 = matches!(asm_arch, InlineAsmArch::X86 | InlineAsmArch::X86_64);
         let att_dialect = is_x86 && options.contains(InlineAsmOptions::ATT_SYNTAX);
-        let intel_dialect = is_x86 && !options.contains(InlineAsmOptions::ATT_SYNTAX);
 
         // GCC index of an output operand equals its position in the array
         let mut outputs = vec![];
@@ -348,9 +348,24 @@ impl<'a, 'gcc, 'tcx> AsmBuilderMethods<'tcx> for Builder<'a, 'gcc, 'tcx> {
                     // processed in the previous pass
                 }
 
-                InlineAsmOperandRef::Const { .. }
-                | InlineAsmOperandRef::SymFn { .. }
-                | InlineAsmOperandRef::SymStatic { .. } => {
+                InlineAsmOperandRef::SymFn { instance } => {
+                    inputs.push(AsmInOperand {
+                        constraint: "X".into(),
+                        rust_idx,
+                        val: self.cx.rvalue_as_function(get_fn(self.cx, instance))
+                            .get_address(None),
+                    });
+                }
+
+                InlineAsmOperandRef::SymStatic { def_id } => {
+                    inputs.push(AsmInOperand {
+                        constraint: "X".into(),
+                        rust_idx,
+                        val: self.cx.get_static(def_id).get_address(None),
+                    });
+                }
+
+                InlineAsmOperandRef::Const { .. } => {
                     // processed in the previous pass
                 }
             }
@@ -359,7 +374,7 @@ impl<'a, 'gcc, 'tcx> AsmBuilderMethods<'tcx> for Builder<'a, 'gcc, 'tcx> {
         // 3. Build the template string
 
         let mut template_str = String::with_capacity(estimate_template_length(template, constants_len, att_dialect));
-        if !intel_dialect {
+        if att_dialect {
             template_str.push_str(ATT_SYNTAX_INS);
         }
 
@@ -444,7 +459,7 @@ impl<'a, 'gcc, 'tcx> AsmBuilderMethods<'tcx> for Builder<'a, 'gcc, 'tcx> {
             }
         }
 
-        if !intel_dialect {
+        if att_dialect {
             template_str.push_str(INTEL_SYNTAX_INS);
         }
 
@@ -588,7 +603,7 @@ fn reg_to_gcc(reg: InlineAsmRegOrRegClass) -> ConstraintOrRegister {
             InlineAsmRegClass::X86(X86InlineAsmRegClass::xmm_reg)
             | InlineAsmRegClass::X86(X86InlineAsmRegClass::ymm_reg) => "x",
             InlineAsmRegClass::X86(X86InlineAsmRegClass::zmm_reg) => "v",
-            InlineAsmRegClass::X86(X86InlineAsmRegClass::kreg) => unimplemented!(),
+            InlineAsmRegClass::X86(X86InlineAsmRegClass::kreg) => "Yk",
             InlineAsmRegClass::X86(X86InlineAsmRegClass::kreg0) => unimplemented!(),
             InlineAsmRegClass::Wasm(WasmInlineAsmRegClass::local) => unimplemented!(),
             InlineAsmRegClass::X86(
@@ -672,8 +687,8 @@ impl<'gcc, 'tcx> AsmMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
         let asm_arch = self.tcx.sess.asm_arch.unwrap();
 
         // Default to Intel syntax on x86
-        let intel_syntax = matches!(asm_arch, InlineAsmArch::X86 | InlineAsmArch::X86_64)
-            && !options.contains(InlineAsmOptions::ATT_SYNTAX);
+        let att_dialect = matches!(asm_arch, InlineAsmArch::X86 | InlineAsmArch::X86_64)
+            && options.contains(InlineAsmOptions::ATT_SYNTAX);
 
         // Build the template string
         let mut template_str = String::new();
@@ -723,11 +738,11 @@ impl<'gcc, 'tcx> AsmMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
         }
 
         let template_str =
-            if intel_syntax {
-                format!("{}\n\t.intel_syntax noprefix", template_str)
+            if att_dialect {
+                format!(".att_syntax\n\t{}\n\t.intel_syntax noprefix", template_str)
             }
             else {
-                format!(".att_syntax\n\t{}\n\t.intel_syntax noprefix", template_str)
+                template_str
             };
         // NOTE: seems like gcc will put the asm in the wrong section, so set it to .text manually.
         let template_str = format!(".pushsection .text\n{}\n.popsection", template_str);

compiler/rustc_codegen_gcc/src/base.rs

@@ -78,9 +78,19 @@ pub fn compile_codegen_unit<'tcx>(tcx: TyCtxt<'tcx>, cgu_name: Symbol, supports_
         let context = Context::default();
         // TODO(antoyo): only set on x86 platforms.
         context.add_command_line_option("-masm=intel");
+        // TODO(antoyo): only add the following cli argument if the feature is supported.
+        context.add_command_line_option("-msse2");
+        context.add_command_line_option("-mavx2");
+        context.add_command_line_option("-msha");
+        context.add_command_line_option("-mpclmul");
+        // FIXME(antoyo): the following causes an illegal instruction on vmovdqu64 in std_example on my CPU.
+        // Only add if the CPU supports it.
+        //context.add_command_line_option("-mavx512f");
         for arg in &tcx.sess.opts.cg.llvm_args {
             context.add_command_line_option(arg);
         }
+        // NOTE: This is needed to compile the file src/intrinsic/archs.rs during a bootstrap of rustc.
+        context.add_command_line_option("-fno-var-tracking-assignments");
         // NOTE: an optimization (https://github.com/rust-lang/rustc_codegen_gcc/issues/53).
         context.add_command_line_option("-fno-semantic-interposition");
         // NOTE: Rust relies on LLVM not doing TBAA (https://github.com/rust-lang/unsafe-code-guidelines/issues/292).

compiler/rustc_codegen_gcc/src/common.rs

@@ -121,8 +121,8 @@ impl<'gcc, 'tcx> ConstMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
         unimplemented!();
     }
 
-    fn const_real(&self, _t: Type<'gcc>, _val: f64) -> RValue<'gcc> {
-        unimplemented!();
+    fn const_real(&self, typ: Type<'gcc>, val: f64) -> RValue<'gcc> {
+        self.context.new_rvalue_from_double(typ, val)
     }
 
     fn const_str(&self, s: Symbol) -> (RValue<'gcc>, RValue<'gcc>) {
@@ -279,6 +279,21 @@ impl<'gcc, 'tcx> SignType<'gcc, 'tcx> for Type<'gcc> {
         else if self.is_u128(cx) {
             cx.i128_type
         }
+        else if self.is_uchar(cx) {
+            cx.char_type
+        }
+        else if self.is_ushort(cx) {
+            cx.short_type
+        }
+        else if self.is_uint(cx) {
+            cx.int_type
+        }
+        else if self.is_ulong(cx) {
+            cx.long_type
+        }
+        else if self.is_ulonglong(cx) {
+            cx.longlong_type
+        }
         else {
             self.clone()
         }
@@ -300,6 +315,21 @@ impl<'gcc, 'tcx> SignType<'gcc, 'tcx> for Type<'gcc> {
         else if self.is_i128(cx) {
             cx.u128_type
         }
+        else if self.is_char(cx) {
+            cx.uchar_type
+        }
+        else if self.is_short(cx) {
+            cx.ushort_type
+        }
+        else if self.is_int(cx) {
+            cx.uint_type
+        }
+        else if self.is_long(cx) {
+            cx.ulong_type
+        }
+        else if self.is_longlong(cx) {
+            cx.ulonglong_type
+        }
         else {
             self.clone()
         }
@@ -312,6 +342,11 @@ pub trait TypeReflection<'gcc, 'tcx>  {
     fn is_uint(&self, cx: &CodegenCx<'gcc, 'tcx>) -> bool;
     fn is_ulong(&self, cx: &CodegenCx<'gcc, 'tcx>) -> bool;
     fn is_ulonglong(&self, cx: &CodegenCx<'gcc, 'tcx>) -> bool;
+    fn is_char(&self, cx: &CodegenCx<'gcc, 'tcx>) -> bool;
+    fn is_short(&self, cx: &CodegenCx<'gcc, 'tcx>) -> bool;
+    fn is_int(&self, cx: &CodegenCx<'gcc, 'tcx>) -> bool;
+    fn is_long(&self, cx: &CodegenCx<'gcc, 'tcx>) -> bool;
+    fn is_longlong(&self, cx: &CodegenCx<'gcc, 'tcx>) -> bool;
 
     fn is_i8(&self, cx: &CodegenCx<'gcc, 'tcx>) -> bool;
     fn is_u8(&self, cx: &CodegenCx<'gcc, 'tcx>) -> bool;
@@ -326,15 +361,17 @@ pub trait TypeReflection<'gcc, 'tcx>  {
 
     fn is_f32(&self, cx: &CodegenCx<'gcc, 'tcx>) -> bool;
     fn is_f64(&self, cx: &CodegenCx<'gcc, 'tcx>) -> bool;
+
+    fn is_vector(&self) -> bool;
 }
 
 impl<'gcc, 'tcx> TypeReflection<'gcc, 'tcx> for Type<'gcc> {
     fn is_uchar(&self, cx: &CodegenCx<'gcc, 'tcx>) -> bool {
-        self.unqualified() == cx.u8_type
+        self.unqualified() == cx.uchar_type
     }
 
     fn is_ushort(&self, cx: &CodegenCx<'gcc, 'tcx>) -> bool {
-        self.unqualified() == cx.u16_type
+        self.unqualified() == cx.ushort_type
     }
 
     fn is_uint(&self, cx: &CodegenCx<'gcc, 'tcx>) -> bool {
@@ -349,6 +386,26 @@ impl<'gcc, 'tcx> TypeReflection<'gcc, 'tcx> for Type<'gcc> {
         self.unqualified() == cx.ulonglong_type
     }
 
+    fn is_char(&self, cx: &CodegenCx<'gcc, 'tcx>) -> bool {
+        self.unqualified() == cx.char_type
+    }
+
+    fn is_short(&self, cx: &CodegenCx<'gcc, 'tcx>) -> bool {
+        self.unqualified() == cx.short_type
+    }
+
+    fn is_int(&self, cx: &CodegenCx<'gcc, 'tcx>) -> bool {
+        self.unqualified() == cx.int_type
+    }
+
+    fn is_long(&self, cx: &CodegenCx<'gcc, 'tcx>) -> bool {
+        self.unqualified() == cx.long_type
+    }
+
+    fn is_longlong(&self, cx: &CodegenCx<'gcc, 'tcx>) -> bool {
+        self.unqualified() == cx.longlong_type
+    }
+
     fn is_i8(&self, cx: &CodegenCx<'gcc, 'tcx>) -> bool {
         self.unqualified() == cx.i8_type
     }
@@ -396,4 +453,21 @@ impl<'gcc, 'tcx> TypeReflection<'gcc, 'tcx> for Type<'gcc> {
     fn is_f64(&self, cx: &CodegenCx<'gcc, 'tcx>) -> bool {
         self.unqualified() == cx.context.new_type::<f64>()
     }
+
+    fn is_vector(&self) -> bool {
+        let mut typ = self.clone();
+        loop {
+            if typ.dyncast_vector().is_some() {
+                return true;
+            }
+
+            let old_type = typ;
+            typ = typ.unqualified();
+            if old_type == typ {
+                break;
+            }
+        }
+
+        false
+    }
 }

compiler/rustc_codegen_gcc/src/consts.rs

@@ -25,7 +25,9 @@ impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
                 }
             }
         }
-        self.context.new_bitcast(None, value, typ)
+        // NOTE: since bitcast makes a value non-constant, don't bitcast if not necessary as some
+        // SIMD builtins require a constant value.
+        self.bitcast_if_needed(value, typ)
     }
 }
 
@@ -45,7 +47,10 @@ impl<'gcc, 'tcx> StaticMethods for CodegenCx<'gcc, 'tcx> {
             }
         }
         let global_value = self.static_addr_of_mut(cv, align, kind);
-        // TODO(antoyo): set global constant.
+        #[cfg(feature = "master")]
+        self.global_lvalues.borrow().get(&global_value)
+            .expect("`static_addr_of_mut` did not add the global to `self.global_lvalues`")
+            .global_set_readonly();
         self.const_globals.borrow_mut().insert(cv, global_value);
         global_value
     }
@@ -79,20 +84,15 @@ impl<'gcc, 'tcx> StaticMethods for CodegenCx<'gcc, 'tcx> {
 
         // TODO(antoyo): set alignment.
 
-        let value =
-            if value.get_type() != gcc_type {
-                self.context.new_bitcast(None, value, gcc_type)
-            }
-            else {
-                value
-            };
+        let value = self.bitcast_if_needed(value, gcc_type);
         global.global_set_initializer_rvalue(value);
 
         // As an optimization, all shared statics which do not have interior
         // mutability are placed into read-only memory.
         if !is_mutable {
             if self.type_is_freeze(ty) {
-                // TODO(antoyo): set global constant.
+                #[cfg(feature = "master")]
+                global.global_set_readonly();
             }
         }
 
@@ -171,8 +171,9 @@ impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
                 Some(kind) if !self.tcx.sess.fewer_names() => {
                     let name = self.generate_local_symbol_name(kind);
                     // TODO(antoyo): check if it's okay that no link_section is set.
-                    // TODO(antoyo): set alignment here as well.
-                    let global = self.declare_private_global(&name[..], self.val_ty(cv));
+
+                    let typ = self.val_ty(cv).get_aligned(align.bytes());
+                    let global = self.declare_private_global(&name[..], typ);
                     global
                 }
                 _ => {

compiler/rustc_codegen_gcc/src/context.rs

@@ -35,6 +35,7 @@ pub struct CodegenCx<'gcc, 'tcx> {
     pub normal_function_addresses: RefCell<FxHashSet<RValue<'gcc>>>,
 
     pub functions: RefCell<FxHashMap<String, Function<'gcc>>>,
+    pub intrinsics: RefCell<FxHashMap<String, Function<'gcc>>>,
 
     pub tls_model: gccjit::TlsModel,
 
@@ -53,10 +54,15 @@ pub struct CodegenCx<'gcc, 'tcx> {
     pub u128_type: Type<'gcc>,
     pub usize_type: Type<'gcc>,
 
+    pub char_type: Type<'gcc>,
+    pub uchar_type: Type<'gcc>,
+    pub short_type: Type<'gcc>,
+    pub ushort_type: Type<'gcc>,
     pub int_type: Type<'gcc>,
     pub uint_type: Type<'gcc>,
     pub long_type: Type<'gcc>,
     pub ulong_type: Type<'gcc>,
+    pub longlong_type: Type<'gcc>,
     pub ulonglong_type: Type<'gcc>,
     pub sizet_type: Type<'gcc>,
 
@@ -145,10 +151,15 @@ impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
         let float_type = context.new_type::<f32>();
         let double_type = context.new_type::<f64>();
 
+        let char_type = context.new_c_type(CType::Char);
+        let uchar_type = context.new_c_type(CType::UChar);
+        let short_type = context.new_c_type(CType::Short);
+        let ushort_type = context.new_c_type(CType::UShort);
         let int_type = context.new_c_type(CType::Int);
         let uint_type = context.new_c_type(CType::UInt);
         let long_type = context.new_c_type(CType::Long);
         let ulong_type = context.new_c_type(CType::ULong);
+        let longlong_type = context.new_c_type(CType::LongLong);
         let ulonglong_type = context.new_c_type(CType::ULongLong);
         let sizet_type = context.new_c_type(CType::SizeT);
 
@@ -184,6 +195,7 @@ impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
             current_func: RefCell::new(None),
             normal_function_addresses: Default::default(),
             functions: RefCell::new(functions),
+            intrinsics: RefCell::new(FxHashMap::default()),
 
             tls_model,
 
@@ -200,10 +212,15 @@ impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
             u32_type,
             u64_type,
             u128_type,
+            char_type,
+            uchar_type,
+            short_type,
+            ushort_type,
             int_type,
             uint_type,
             long_type,
             ulong_type,
+            longlong_type,
             ulonglong_type,
             sizet_type,
 
@@ -269,16 +286,25 @@ impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
     }
 
     pub fn is_native_int_type_or_bool(&self, typ: Type<'gcc>) -> bool {
-        self.is_native_int_type(typ) || typ == self.bool_type
+        self.is_native_int_type(typ) || typ.is_compatible_with(self.bool_type)
     }
 
     pub fn is_int_type_or_bool(&self, typ: Type<'gcc>) -> bool {
-        self.is_native_int_type(typ) || self.is_non_native_int_type(typ) || typ == self.bool_type
+        self.is_native_int_type(typ) || self.is_non_native_int_type(typ) || typ.is_compatible_with(self.bool_type)
     }
 
     pub fn sess(&self) -> &Session {
         &self.tcx.sess
     }
+
+    pub fn bitcast_if_needed(&self, value: RValue<'gcc>, expected_type: Type<'gcc>) -> RValue<'gcc> {
+        if value.get_type() != expected_type {
+            self.context.new_bitcast(None, value, expected_type)
+        }
+        else {
+            value
+        }
+    }
 }
 
 impl<'gcc, 'tcx> BackendTypes for CodegenCx<'gcc, 'tcx> {
@@ -306,8 +332,16 @@ impl<'gcc, 'tcx> MiscMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
     }
 
     fn get_fn_addr(&self, instance: Instance<'tcx>) -> RValue<'gcc> {
-        let func = get_fn(self, instance);
-        let func = self.rvalue_as_function(func);
+        let func_name = self.tcx.symbol_name(instance).name;
+
+        let func =
+            if self.intrinsics.borrow().contains_key(func_name) {
+                self.intrinsics.borrow()[func_name].clone()
+            }
+            else {
+                let func = get_fn(self, instance);
+                self.rvalue_as_function(func)
+            };
         let ptr = func.get_address(None);
 
         // TODO(antoyo): don't do this twice: i.e. in declare_fn and here.

compiler/rustc_codegen_gcc/src/declare.rs

@@ -11,7 +11,7 @@ use crate::intrinsic::llvm;
 impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
     pub fn get_or_insert_global(&self, name: &str, ty: Type<'gcc>, is_tls: bool, link_section: Option<Symbol>) -> LValue<'gcc> {
         if self.globals.borrow().contains_key(name) {
-            let typ = self.globals.borrow().get(name).expect("global").get_type();
+            let typ = self.globals.borrow()[name].get_type();
             let global = self.context.new_global(None, GlobalKind::Imported, typ, name);
             if is_tls {
                 global.set_tls_model(self.tls_model);
@@ -103,11 +103,13 @@ impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
 /// update the declaration and return existing Value instead.
 fn declare_raw_fn<'gcc>(cx: &CodegenCx<'gcc, '_>, name: &str, _callconv: () /*llvm::CallConv*/, return_type: Type<'gcc>, param_types: &[Type<'gcc>], variadic: bool) -> Function<'gcc> {
     if name.starts_with("llvm.") {
-        return llvm::intrinsic(name, cx);
+        let intrinsic = llvm::intrinsic(name, cx);
+        cx.intrinsics.borrow_mut().insert(name.to_string(), intrinsic);
+        return intrinsic;
     }
     let func =
         if cx.functions.borrow().contains_key(name) {
-            *cx.functions.borrow().get(name).expect("function")
+            cx.functions.borrow()[name]
         }
         else {
             let params: Vec<_> = param_types.into_iter().enumerate()

compiler/rustc_codegen_gcc/src/int.rs

@@ -153,8 +153,15 @@ impl<'a, 'gcc, 'tcx> Builder<'a, 'gcc, 'tcx> {
         let a_type = a.get_type();
         let b_type = b.get_type();
         if self.is_native_int_type_or_bool(a_type) && self.is_native_int_type_or_bool(b_type) {
-            if a.get_type() != b.get_type() {
-                b = self.context.new_cast(None, b, a.get_type());
+            if a_type != b_type {
+                if a_type.is_vector() {
+                    // Vector types need to be bitcast.
+                    // TODO(antoyo): perhaps use __builtin_convertvector for vector casting.
+                    b = self.context.new_bitcast(None, b, a.get_type());
+                }
+                else {
+                    b = self.context.new_cast(None, b, a.get_type());
+                }
             }
             self.context.new_binary_op(None, operation, a_type, a, b)
         }
@@ -593,7 +600,10 @@ impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
         let b_type = b.get_type();
         let a_native = self.is_native_int_type_or_bool(a_type);
         let b_native = self.is_native_int_type_or_bool(b_type);
-        if a_native && b_native {
+        if a_type.is_vector() && b_type.is_vector() {
+            self.context.new_binary_op(None, operation, a_type, a, b)
+        }
+        else if a_native && b_native {
             if a_type != b_type {
                 b = self.context.new_cast(None, b, a_type);
             }
@@ -639,6 +649,8 @@ impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
         else {
             // Since u128 and i128 are the only types that can be unsupported, we know the type of
             // value and the destination type have the same size, so a bitcast is fine.
+
+            // TODO(antoyo): perhaps use __builtin_convertvector for vector casting.
             self.context.new_bitcast(None, value, dest_typ)
         }
     }

compiler/rustc_codegen_gcc/src/intrinsic/llvm.rs

@@ -1,22 +1,250 @@
-use gccjit::Function;
+use std::borrow::Cow;
 
-use crate::context::CodegenCx;
+use gccjit::{Function, FunctionPtrType, RValue, ToRValue};
 
-pub fn intrinsic<'gcc, 'tcx>(name: &str, cx: &CodegenCx<'gcc, 'tcx>) -> Function<'gcc> {
-    let _gcc_name =
-        match name {
-            "llvm.x86.xgetbv" => {
-                let gcc_name = "__builtin_trap";
-                let func = cx.context.get_builtin_function(gcc_name);
-                cx.functions.borrow_mut().insert(gcc_name.to_string(), func);
-                return func;
+use crate::{context::CodegenCx, builder::Builder};
+
+pub fn adjust_intrinsic_arguments<'a, 'b, 'gcc, 'tcx>(builder: &Builder<'a, 'gcc, 'tcx>, gcc_func: FunctionPtrType<'gcc>, mut args: Cow<'b, [RValue<'gcc>]>, func_name: &str) -> Cow<'b, [RValue<'gcc>]> {
+    // Some LLVM intrinsics do not map 1-to-1 to GCC intrinsics, so we add the missing
+    // arguments here.
+    if gcc_func.get_param_count() != args.len() {
+        match &*func_name {
+            "__builtin_ia32_pmuldq512_mask" | "__builtin_ia32_pmuludq512_mask"
+                // FIXME(antoyo): the following intrinsics has 4 (or 5) arguments according to the doc, but is defined with 2 (or 3) arguments in library/stdarch/crates/core_arch/src/x86/avx512f.rs.
+                | "__builtin_ia32_pmaxsd512_mask" | "__builtin_ia32_pmaxsq512_mask" | "__builtin_ia32_pmaxsq256_mask"
+                | "__builtin_ia32_pmaxsq128_mask" | "__builtin_ia32_maxps512_mask" | "__builtin_ia32_maxpd512_mask"
+                | "__builtin_ia32_pmaxud512_mask" | "__builtin_ia32_pmaxuq512_mask" | "__builtin_ia32_pmaxuq256_mask"
+                | "__builtin_ia32_pmaxuq128_mask"
+                | "__builtin_ia32_pminsd512_mask" | "__builtin_ia32_pminsq512_mask" | "__builtin_ia32_pminsq256_mask"
+                | "__builtin_ia32_pminsq128_mask" | "__builtin_ia32_minps512_mask" | "__builtin_ia32_minpd512_mask"
+                | "__builtin_ia32_pminud512_mask" | "__builtin_ia32_pminuq512_mask" | "__builtin_ia32_pminuq256_mask"
+                | "__builtin_ia32_pminuq128_mask" | "__builtin_ia32_sqrtps512_mask" | "__builtin_ia32_sqrtpd512_mask"
+                => {
+                    // TODO: refactor by separating those intrinsics outside of this branch.
+                    let add_before_last_arg =
+                        match &*func_name {
+                            "__builtin_ia32_maxps512_mask" | "__builtin_ia32_maxpd512_mask"
+                                | "__builtin_ia32_minps512_mask" | "__builtin_ia32_minpd512_mask"
+                                | "__builtin_ia32_sqrtps512_mask" | "__builtin_ia32_sqrtpd512_mask" => true,
+                            _ => false,
+                        };
+                    let new_first_arg_is_zero =
+                        match &*func_name {
+                            "__builtin_ia32_pmaxuq256_mask" | "__builtin_ia32_pmaxuq128_mask"
+                                | "__builtin_ia32_pminuq256_mask" | "__builtin_ia32_pminuq128_mask" => true,
+                            _ => false
+                        };
+                    let arg3_index =
+                        match &*func_name {
+                            "__builtin_ia32_sqrtps512_mask" | "__builtin_ia32_sqrtpd512_mask" => 1,
+                            _ => 2,
+                        };
+                    let mut new_args = args.to_vec();
+                    let arg3_type = gcc_func.get_param_type(arg3_index);
+                    let first_arg =
+                        if new_first_arg_is_zero {
+                            let vector_type = arg3_type.dyncast_vector().expect("vector type");
+                            let zero = builder.context.new_rvalue_zero(vector_type.get_element_type());
+                            let num_units = vector_type.get_num_units();
+                            builder.context.new_rvalue_from_vector(None, arg3_type, &vec![zero; num_units])
+                        }
+                        else {
+                            builder.current_func().new_local(None, arg3_type, "undefined_for_intrinsic").to_rvalue()
+                        };
+                    if add_before_last_arg {
+                        new_args.insert(new_args.len() - 1, first_arg);
+                    }
+                    else {
+                        new_args.push(first_arg);
+                    }
+                    let arg4_index =
+                        match &*func_name {
+                            "__builtin_ia32_sqrtps512_mask" | "__builtin_ia32_sqrtpd512_mask" => 2,
+                            _ => 3,
+                        };
+                    let arg4_type = gcc_func.get_param_type(arg4_index);
+                    let minus_one = builder.context.new_rvalue_from_int(arg4_type, -1);
+                    if add_before_last_arg {
+                        new_args.insert(new_args.len() - 1, minus_one);
+                    }
+                    else {
+                        new_args.push(minus_one);
+                    }
+                    args = new_args.into();
+                },
+                "__builtin_ia32_pternlogd512_mask" | "__builtin_ia32_pternlogd256_mask"
+                    | "__builtin_ia32_pternlogd128_mask" | "__builtin_ia32_pternlogq512_mask"
+                    | "__builtin_ia32_pternlogq256_mask" | "__builtin_ia32_pternlogq128_mask" => {
+                        let mut new_args = args.to_vec();
+                        let arg5_type = gcc_func.get_param_type(4);
+                        let minus_one = builder.context.new_rvalue_from_int(arg5_type, -1);
+                        new_args.push(minus_one);
+                        args = new_args.into();
+                    },
+                    "__builtin_ia32_vfmaddps512_mask" | "__builtin_ia32_vfmaddpd512_mask" => {
+                        let mut new_args = args.to_vec();
+
+                        let mut last_arg = None;
+                        if args.len() == 4 {
+                            last_arg = new_args.pop();
+                        }
+
+                        let arg4_type = gcc_func.get_param_type(3);
+                        let minus_one = builder.context.new_rvalue_from_int(arg4_type, -1);
+                        new_args.push(minus_one);
+
+                        if args.len() == 3 {
+                            // Both llvm.fma.v16f32 and llvm.x86.avx512.vfmadd.ps.512 maps to
+                            // the same GCC intrinsic, but the former has 3 parameters and the
+                            // latter has 4 so it doesn't require this additional argument.
+                            let arg5_type = gcc_func.get_param_type(4);
+                            new_args.push(builder.context.new_rvalue_from_int(arg5_type, 4));
+                        }
+
+                        if let Some(last_arg) = last_arg {
+                            new_args.push(last_arg);
+                        }
+
+                        args = new_args.into();
+                    },
+                    "__builtin_ia32_addps512_mask" | "__builtin_ia32_addpd512_mask"
+                        | "__builtin_ia32_subps512_mask" | "__builtin_ia32_subpd512_mask"
+                        | "__builtin_ia32_mulps512_mask" | "__builtin_ia32_mulpd512_mask"
+                        | "__builtin_ia32_divps512_mask" | "__builtin_ia32_divpd512_mask" => {
+                        let mut new_args = args.to_vec();
+                        let last_arg = new_args.pop().expect("last arg");
+                        let arg3_type = gcc_func.get_param_type(2);
+                        let undefined = builder.current_func().new_local(None, arg3_type, "undefined_for_intrinsic").to_rvalue();
+                        new_args.push(undefined);
+                        let arg4_type = gcc_func.get_param_type(3);
+                        let minus_one = builder.context.new_rvalue_from_int(arg4_type, -1);
+                        new_args.push(minus_one);
+                        new_args.push(last_arg);
+                        args = new_args.into();
+                    },
+                    "__builtin_ia32_vfmaddsubps512_mask" | "__builtin_ia32_vfmaddsubpd512_mask" => {
+                        let mut new_args = args.to_vec();
+                        let last_arg = new_args.pop().expect("last arg");
+                        let arg4_type = gcc_func.get_param_type(3);
+                        let minus_one = builder.context.new_rvalue_from_int(arg4_type, -1);
+                        new_args.push(minus_one);
+                        new_args.push(last_arg);
+                        args = new_args.into();
+                    },
+                    _ => (),
+        }
+    }
+
+    args
+}
+
+pub fn ignore_arg_cast(func_name: &str, index: usize, args_len: usize) -> bool {
+    // NOTE: these intrinsics have missing parameters before the last one, so ignore the
+    // last argument type check.
+    // FIXME(antoyo): find a way to refactor in order to avoid this hack.
+    match func_name {
+        "__builtin_ia32_maxps512_mask" | "__builtin_ia32_maxpd512_mask"
+            | "__builtin_ia32_minps512_mask" | "__builtin_ia32_minpd512_mask" | "__builtin_ia32_sqrtps512_mask"
+            | "__builtin_ia32_sqrtpd512_mask" | "__builtin_ia32_addps512_mask" | "__builtin_ia32_addpd512_mask"
+            | "__builtin_ia32_subps512_mask" | "__builtin_ia32_subpd512_mask"
+            | "__builtin_ia32_mulps512_mask" | "__builtin_ia32_mulpd512_mask"
+            | "__builtin_ia32_divps512_mask" | "__builtin_ia32_divpd512_mask"
+            | "__builtin_ia32_vfmaddsubps512_mask" | "__builtin_ia32_vfmaddsubpd512_mask" => {
+                if index == args_len - 1 {
+                    return true;
+                }
             },
-            // NOTE: this doc specifies the equivalent GCC builtins: http://huonw.github.io/llvmint/llvmint/x86/index.html
-            "llvm.x86.sse2.cmp.pd" => "__builtin_ia32_cmppd",
-            "llvm.x86.sse2.movmsk.pd" => "__builtin_ia32_movmskpd",
-            "llvm.x86.sse2.pmovmskb.128" => "__builtin_ia32_pmovmskb128",
-            _ => unimplemented!("unsupported LLVM intrinsic {}", name)
-        };
-
-    unimplemented!();
+        "__builtin_ia32_vfmaddps512_mask" | "__builtin_ia32_vfmaddpd512_mask" => {
+            // Since there are two LLVM intrinsics that map to each of these GCC builtins and only
+            // one of them has a missing parameter before the last one, we check the number of
+            // arguments to distinguish those cases.
+            if args_len == 4 && index == args_len - 1 {
+                return true;
+            }
+        },
+        _ => (),
+    }
+
+    false
+}
+
+#[cfg(not(feature="master"))]
+pub fn intrinsic<'gcc, 'tcx>(name: &str, cx: &CodegenCx<'gcc, 'tcx>) -> Function<'gcc> {
+    match name {
+        "llvm.x86.xgetbv" => {
+            let gcc_name = "__builtin_trap";
+            let func = cx.context.get_builtin_function(gcc_name);
+            cx.functions.borrow_mut().insert(gcc_name.to_string(), func);
+            return func;
+        },
+        _ => unimplemented!("unsupported LLVM intrinsic {}", name),
+    }
+}
+
+#[cfg(feature="master")]
+pub fn intrinsic<'gcc, 'tcx>(name: &str, cx: &CodegenCx<'gcc, 'tcx>) -> Function<'gcc> {
+    let gcc_name = match name {
+        "llvm.x86.xgetbv" => "__builtin_ia32_xgetbv",
+        // NOTE: this doc specifies the equivalent GCC builtins: http://huonw.github.io/llvmint/llvmint/x86/index.html
+        "llvm.sqrt.v2f64" => "__builtin_ia32_sqrtpd",
+        "llvm.x86.avx512.pmul.dq.512" => "__builtin_ia32_pmuldq512_mask",
+        "llvm.x86.avx512.pmulu.dq.512" => "__builtin_ia32_pmuludq512_mask",
+        "llvm.x86.avx512.mask.pmaxs.q.256" => "__builtin_ia32_pmaxsq256_mask",
+        "llvm.x86.avx512.mask.pmaxs.q.128" => "__builtin_ia32_pmaxsq128_mask",
+        "llvm.x86.avx512.max.ps.512" => "__builtin_ia32_maxps512_mask",
+        "llvm.x86.avx512.max.pd.512" => "__builtin_ia32_maxpd512_mask",
+        "llvm.x86.avx512.mask.pmaxu.q.256" => "__builtin_ia32_pmaxuq256_mask",
+        "llvm.x86.avx512.mask.pmaxu.q.128" => "__builtin_ia32_pmaxuq128_mask",
+        "llvm.x86.avx512.mask.pmins.q.256" => "__builtin_ia32_pminsq256_mask",
+        "llvm.x86.avx512.mask.pmins.q.128" => "__builtin_ia32_pminsq128_mask",
+        "llvm.x86.avx512.min.ps.512" => "__builtin_ia32_minps512_mask",
+        "llvm.x86.avx512.min.pd.512" => "__builtin_ia32_minpd512_mask",
+        "llvm.x86.avx512.mask.pminu.q.256" => "__builtin_ia32_pminuq256_mask",
+        "llvm.x86.avx512.mask.pminu.q.128" => "__builtin_ia32_pminuq128_mask",
+        "llvm.fma.v16f32" => "__builtin_ia32_vfmaddps512_mask",
+        "llvm.fma.v8f64" => "__builtin_ia32_vfmaddpd512_mask",
+        "llvm.x86.avx512.vfmaddsub.ps.512" => "__builtin_ia32_vfmaddsubps512_mask",
+        "llvm.x86.avx512.vfmaddsub.pd.512" => "__builtin_ia32_vfmaddsubpd512_mask",
+        "llvm.x86.avx512.pternlog.d.512" => "__builtin_ia32_pternlogd512_mask",
+        "llvm.x86.avx512.pternlog.d.256" => "__builtin_ia32_pternlogd256_mask",
+        "llvm.x86.avx512.pternlog.d.128" => "__builtin_ia32_pternlogd128_mask",
+        "llvm.x86.avx512.pternlog.q.512" => "__builtin_ia32_pternlogq512_mask",
+        "llvm.x86.avx512.pternlog.q.256" => "__builtin_ia32_pternlogq256_mask",
+        "llvm.x86.avx512.pternlog.q.128" => "__builtin_ia32_pternlogq128_mask",
+        "llvm.x86.avx512.add.ps.512" => "__builtin_ia32_addps512_mask",
+        "llvm.x86.avx512.add.pd.512" => "__builtin_ia32_addpd512_mask",
+        "llvm.x86.avx512.sub.ps.512" => "__builtin_ia32_subps512_mask",
+        "llvm.x86.avx512.sub.pd.512" => "__builtin_ia32_subpd512_mask",
+        "llvm.x86.avx512.mul.ps.512" => "__builtin_ia32_mulps512_mask",
+        "llvm.x86.avx512.mul.pd.512" => "__builtin_ia32_mulpd512_mask",
+        "llvm.x86.avx512.div.ps.512" => "__builtin_ia32_divps512_mask",
+        "llvm.x86.avx512.div.pd.512" => "__builtin_ia32_divpd512_mask",
+        "llvm.x86.avx512.vfmadd.ps.512" => "__builtin_ia32_vfmaddps512_mask",
+        "llvm.x86.avx512.vfmadd.pd.512" => "__builtin_ia32_vfmaddpd512_mask",
+
+        // The above doc points to unknown builtins for the following, so override them:
+        "llvm.x86.avx2.gather.d.d" => "__builtin_ia32_gathersiv4si",
+        "llvm.x86.avx2.gather.d.d.256" => "__builtin_ia32_gathersiv8si",
+        "llvm.x86.avx2.gather.d.ps" => "__builtin_ia32_gathersiv4sf",
+        "llvm.x86.avx2.gather.d.ps.256" => "__builtin_ia32_gathersiv8sf",
+        "llvm.x86.avx2.gather.d.q" => "__builtin_ia32_gathersiv2di",
+        "llvm.x86.avx2.gather.d.q.256" => "__builtin_ia32_gathersiv4di",
+        "llvm.x86.avx2.gather.d.pd" => "__builtin_ia32_gathersiv2df",
+        "llvm.x86.avx2.gather.d.pd.256" => "__builtin_ia32_gathersiv4df",
+        "llvm.x86.avx2.gather.q.d" => "__builtin_ia32_gatherdiv4si",
+        "llvm.x86.avx2.gather.q.d.256" => "__builtin_ia32_gatherdiv4si256",
+        "llvm.x86.avx2.gather.q.ps" => "__builtin_ia32_gatherdiv4sf",
+        "llvm.x86.avx2.gather.q.ps.256" => "__builtin_ia32_gatherdiv4sf256",
+        "llvm.x86.avx2.gather.q.q" => "__builtin_ia32_gatherdiv2di",
+        "llvm.x86.avx2.gather.q.q.256" => "__builtin_ia32_gatherdiv4di",
+        "llvm.x86.avx2.gather.q.pd" => "__builtin_ia32_gatherdiv2df",
+        "llvm.x86.avx2.gather.q.pd.256" => "__builtin_ia32_gatherdiv4df",
+        "" => "",
+        // NOTE: this file is generated by https://github.com/GuillaumeGomez/llvmint/blob/master/generate_list.py
+        _ => include!("archs.rs"),
+    };
+
+    let func = cx.context.get_target_builtin_function(gcc_name);
+    cx.functions.borrow_mut().insert(gcc_name.to_string(), func);
+    func
 }

compiler/rustc_codegen_gcc/src/intrinsic/mod.rs

@@ -967,34 +967,55 @@ impl<'a, 'gcc, 'tcx> Builder<'a, 'gcc, 'tcx> {
     }
 
     fn saturating_add(&mut self, lhs: RValue<'gcc>, rhs: RValue<'gcc>, signed: bool, width: u64) -> RValue<'gcc> {
-        let func = self.current_func.borrow().expect("func");
-
+        let result_type = lhs.get_type();
         if signed {
-            // Algorithm from: https://stackoverflow.com/a/56531252/389119
-            let after_block = func.new_block("after");
-            let func_name =
-                match width {
-                    8 => "__builtin_add_overflow",
-                    16 => "__builtin_add_overflow",
-                    32 => "__builtin_sadd_overflow",
-                    64 => "__builtin_saddll_overflow",
-                    128 => "__builtin_add_overflow",
-                    _ => unreachable!(),
-                };
-            let overflow_func = self.context.get_builtin_function(func_name);
-            let result_type = lhs.get_type();
+            // Based on algorithm from: https://stackoverflow.com/a/56531252/389119
+            let func = self.current_func.borrow().expect("func");
             let res = func.new_local(None, result_type, "saturating_sum");
-            let overflow = self.overflow_call(overflow_func, &[lhs, rhs, res.get_address(None)], None);
+            let supports_native_type = self.is_native_int_type(result_type);
+            let overflow =
+                if supports_native_type {
+                    let func_name =
+                        match width {
+                            8 => "__builtin_add_overflow",
+                            16 => "__builtin_add_overflow",
+                            32 => "__builtin_sadd_overflow",
+                            64 => "__builtin_saddll_overflow",
+                            128 => "__builtin_add_overflow",
+                            _ => unreachable!(),
+                        };
+                    let overflow_func = self.context.get_builtin_function(func_name);
+                    self.overflow_call(overflow_func, &[lhs, rhs, res.get_address(None)], None)
+                }
+                else {
+                    let func_name =
+                        match width {
+                            128 => "__rust_i128_addo",
+                            _ => unreachable!(),
+                        };
+                    let param_a = self.context.new_parameter(None, result_type, "a");
+                    let param_b = self.context.new_parameter(None, result_type, "b");
+                    let result_field = self.context.new_field(None, result_type, "result");
+                    let overflow_field = self.context.new_field(None, self.bool_type, "overflow");
+                    let return_type = self.context.new_struct_type(None, "result_overflow", &[result_field, overflow_field]);
+                    let func = self.context.new_function(None, FunctionType::Extern, return_type.as_type(), &[param_a, param_b], func_name, false);
+                    let result = self.context.new_call(None, func, &[lhs, rhs]);
+                    let overflow = result.access_field(None, overflow_field);
+                    let int_result = result.access_field(None, result_field);
+                    self.llbb().add_assignment(None, res, int_result);
+                    overflow
+                };
 
             let then_block = func.new_block("then");
+            let after_block = func.new_block("after");
 
-            let unsigned_type = self.context.new_int_type(width as i32 / 8, false);
-            let shifted = self.context.new_cast(None, lhs, unsigned_type) >> self.context.new_rvalue_from_int(unsigned_type, width as i32 - 1);
-            let uint_max = self.context.new_unary_op(None, UnaryOp::BitwiseNegate, unsigned_type,
-                self.context.new_rvalue_from_int(unsigned_type, 0)
-            );
-            let int_max = uint_max >> self.context.new_rvalue_one(unsigned_type);
-            then_block.add_assignment(None, res, self.context.new_cast(None, shifted + int_max, result_type));
+            // Return `result_type`'s maximum or minimum value on overflow
+            // NOTE: convert the type to unsigned to have an unsigned shift.
+            let unsigned_type = result_type.to_unsigned(&self.cx);
+            let shifted = self.gcc_lshr(self.gcc_int_cast(lhs, unsigned_type), self.gcc_int(unsigned_type, width as i64 - 1));
+            let uint_max = self.gcc_not(self.gcc_int(unsigned_type, 0));
+            let int_max = self.gcc_lshr(uint_max, self.gcc_int(unsigned_type, 1));
+            then_block.add_assignment(None, res, self.gcc_int_cast(self.gcc_add(shifted, int_max), result_type));
             then_block.end_with_jump(None, after_block);
 
             self.llbb().end_with_conditional(None, overflow, then_block, after_block);
@@ -1007,19 +1028,18 @@ impl<'a, 'gcc, 'tcx> Builder<'a, 'gcc, 'tcx> {
         }
         else {
             // Algorithm from: http://locklessinc.com/articles/sat_arithmetic/
-            let res = lhs + rhs;
-            let res_type = res.get_type();
-            let cond = self.context.new_comparison(None, ComparisonOp::LessThan, res, lhs);
-            let value = self.context.new_unary_op(None, UnaryOp::Minus, res_type, self.context.new_cast(None, cond, res_type));
-            res | value
+            let res = self.gcc_add(lhs, rhs);
+            let cond = self.gcc_icmp(IntPredicate::IntULT, res, lhs);
+            let value = self.gcc_neg(self.gcc_int_cast(cond, result_type));
+            self.gcc_or(res, value)
         }
     }
 
     // Algorithm from: https://locklessinc.com/articles/sat_arithmetic/
     fn saturating_sub(&mut self, lhs: RValue<'gcc>, rhs: RValue<'gcc>, signed: bool, width: u64) -> RValue<'gcc> {
+        let result_type = lhs.get_type();
         if signed {
-            // Also based on algorithm from: https://stackoverflow.com/a/56531252/389119
-            let result_type = lhs.get_type();
+            // Based on algorithm from: https://stackoverflow.com/a/56531252/389119
             let func = self.current_func.borrow().expect("func");
             let res = func.new_local(None, result_type, "saturating_diff");
             let supports_native_type = self.is_native_int_type(result_type);
@@ -1059,6 +1079,7 @@ impl<'a, 'gcc, 'tcx> Builder<'a, 'gcc, 'tcx> {
             let then_block = func.new_block("then");
             let after_block = func.new_block("after");
 
+            // Return `result_type`'s maximum or minimum value on overflow
             // NOTE: convert the type to unsigned to have an unsigned shift.
             let unsigned_type = result_type.to_unsigned(&self.cx);
             let shifted = self.gcc_lshr(self.gcc_int_cast(lhs, unsigned_type), self.gcc_int(unsigned_type, width as i64 - 1));
@@ -1076,11 +1097,10 @@ impl<'a, 'gcc, 'tcx> Builder<'a, 'gcc, 'tcx> {
             res.to_rvalue()
         }
         else {
-            let res = lhs - rhs;
-            let comparison = self.context.new_comparison(None, ComparisonOp::LessThanEquals, res, lhs);
-            let comparison = self.context.new_cast(None, comparison, lhs.get_type());
-            let unary_op = self.context.new_unary_op(None, UnaryOp::Minus, comparison.get_type(), comparison);
-            self.and(res, unary_op)
+            let res = self.gcc_sub(lhs, rhs);
+            let comparison = self.gcc_icmp(IntPredicate::IntULE, res, lhs);
+            let value = self.gcc_neg(self.gcc_int_cast(comparison, result_type));
+            self.gcc_and(res, value)
         }
     }
 }

compiler/rustc_codegen_gcc/src/lib.rs

@@ -203,7 +203,7 @@ impl WriteBackendMethods for GccCodegenBackend {
     fn run_fat_lto(_cgcx: &CodegenContext<Self>, mut modules: Vec<FatLTOInput<Self>>, _cached_modules: Vec<(SerializedModule<Self::ModuleBuffer>, WorkProduct)>) -> Result<LtoModuleCodegen<Self>, FatalError> {
         // TODO(antoyo): implement LTO by sending -flto to libgccjit and adding the appropriate gcc linker plugins.
         // NOTE: implemented elsewhere.
-        // TODO: what is implemented elsewhere ^ ?
+        // TODO(antoyo): what is implemented elsewhere ^ ?
         let module =
             match modules.remove(0) {
                 FatLTOInput::InMemory(module) => module,
@@ -301,7 +301,22 @@ pub fn target_features(sess: &Session) -> Vec<Symbol> {
         )
         .filter(|_feature| {
             // TODO(antoyo): implement a way to get enabled feature in libgccjit.
-            false
+            // Probably using the equivalent of __builtin_cpu_supports.
+            #[cfg(feature="master")]
+            {
+                _feature.contains("sse") || _feature.contains("avx")
+            }
+            #[cfg(not(feature="master"))]
+            {
+                false
+            }
+            /*
+               adx, aes, avx, avx2, avx512bf16, avx512bitalg, avx512bw, avx512cd, avx512dq, avx512er, avx512f, avx512gfni,
+               avx512ifma, avx512pf, avx512vaes, avx512vbmi, avx512vbmi2, avx512vl, avx512vnni, avx512vp2intersect, avx512vpclmulqdq,
+               avx512vpopcntdq, bmi1, bmi2, cmpxchg16b, ermsb, f16c, fma, fxsr, lzcnt, movbe, pclmulqdq, popcnt, rdrand, rdseed, rtm,
+               sha, sse, sse2, sse3, sse4.1, sse4.2, sse4a, ssse3, tbm, xsave, xsavec, xsaveopt, xsaves
+             */
+            //false
         })
         .map(|feature| Symbol::intern(feature))
         .collect()

compiler/rustc_codegen_gcc/src/type_.rs

@@ -3,10 +3,11 @@ use std::convert::TryInto;
 use gccjit::{RValue, Struct, Type};
 use rustc_codegen_ssa::traits::{BaseTypeMethods, DerivedTypeMethods};
 use rustc_codegen_ssa::common::TypeKind;
-use rustc_middle::bug;
+use rustc_middle::{bug, ty};
 use rustc_middle::ty::layout::TyAndLayout;
 use rustc_target::abi::{AddressSpace, Align, Integer, Size};
 
+use crate::common::TypeReflection;
 use crate::context::CodegenCx;
 use crate::type_of::LayoutGccExt;
 
@@ -60,6 +61,17 @@ impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
         let ity = Integer::approximate_align(self, align);
         self.type_from_integer(ity)
     }
+
+    pub fn type_vector(&self, ty: Type<'gcc>, len: u64) -> Type<'gcc> {
+        self.context.new_vector_type(ty, len)
+    }
+
+    pub fn type_float_from_ty(&self, t: ty::FloatTy) -> Type<'gcc> {
+        match t {
+            ty::FloatTy::F32 => self.type_f32(),
+            ty::FloatTy::F64 => self.type_f64(),
+        }
+    }
 }
 
 impl<'gcc, 'tcx> BaseTypeMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
@@ -103,16 +115,19 @@ impl<'gcc, 'tcx> BaseTypeMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
         self.context.new_function_pointer_type(None, return_type, params, false)
     }
 
-    fn type_struct(&self, fields: &[Type<'gcc>], _packed: bool) -> Type<'gcc> {
+    fn type_struct(&self, fields: &[Type<'gcc>], packed: bool) -> Type<'gcc> {
         let types = fields.to_vec();
         if let Some(typ) = self.struct_types.borrow().get(fields) {
             return typ.clone();
         }
         let fields: Vec<_> = fields.iter().enumerate()
             .map(|(index, field)| self.context.new_field(None, *field, &format!("field{}_TODO", index)))
             .collect();
-        // TODO(antoyo): use packed.
         let typ = self.context.new_struct_type(None, "struct", &fields).as_type();
+        if packed {
+            #[cfg(feature="master")]
+            typ.set_packed();
+        }
         self.struct_types.borrow_mut().insert(types, typ);
         typ
     }
@@ -127,7 +142,7 @@ impl<'gcc, 'tcx> BaseTypeMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
         else if typ.is_compatible_with(self.double_type) {
             TypeKind::Double
         }
-        else if typ.dyncast_vector().is_some() {
+        else if typ.is_vector() {
             TypeKind::Vector
         }
         else {
@@ -141,7 +156,7 @@ impl<'gcc, 'tcx> BaseTypeMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
     }
 
     fn type_ptr_to_ext(&self, ty: Type<'gcc>, _address_space: AddressSpace) -> Type<'gcc> {
-        // TODO(antoyo): use address_space
+        // TODO(antoyo): use address_space, perhaps with TYPE_ADDR_SPACE?
         ty.make_pointer()
     }
 
@@ -167,10 +182,10 @@ impl<'gcc, 'tcx> BaseTypeMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
     fn float_width(&self, typ: Type<'gcc>) -> usize {
         let f32 = self.context.new_type::<f32>();
         let f64 = self.context.new_type::<f64>();
-        if typ == f32 {
+        if typ.is_compatible_with(f32) {
             32
         }
-        else if typ == f64 {
+        else if typ.is_compatible_with(f64) {
             64
         }
         else {
@@ -197,12 +212,15 @@ impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
         self.type_array(self.type_from_integer(unit), size / unit_size)
     }
 
-    pub fn set_struct_body(&self, typ: Struct<'gcc>, fields: &[Type<'gcc>], _packed: bool) {
-        // TODO(antoyo): use packed.
+    pub fn set_struct_body(&self, typ: Struct<'gcc>, fields: &[Type<'gcc>], packed: bool) {
         let fields: Vec<_> = fields.iter().enumerate()
             .map(|(index, field)| self.context.new_field(None, *field, &format!("field_{}", index)))
             .collect();
         typ.set_fields(None, &fields);
+        if packed {
+            #[cfg(feature="master")]
+            typ.as_type().set_packed();
+        }
     }
 
     pub fn type_named_struct(&self, name: &str) -> Struct<'gcc> {
@@ -229,6 +247,10 @@ impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
 
         self.context.new_array_type(None, ty, len)
     }
+
+    pub fn type_bool(&self) -> Type<'gcc> {
+        self.context.new_type::<bool>()
+    }
 }
 
 pub fn struct_fields<'gcc, 'tcx>(cx: &CodegenCx<'gcc, 'tcx>, layout: TyAndLayout<'tcx>) -> (Vec<Type<'gcc>>, bool) {

compiler/rustc_codegen_gcc/src/type_of.rs

@@ -24,6 +24,30 @@ impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
             I128 => self.type_u128(),
         }
     }
+
+    #[cfg(feature="master")]
+    pub fn type_int_from_ty(&self, t: ty::IntTy) -> Type<'gcc> {
+        match t {
+            ty::IntTy::Isize => self.type_isize(),
+            ty::IntTy::I8 => self.type_i8(),
+            ty::IntTy::I16 => self.type_i16(),
+            ty::IntTy::I32 => self.type_i32(),
+            ty::IntTy::I64 => self.type_i64(),
+            ty::IntTy::I128 => self.type_i128(),
+        }
+    }
+
+    #[cfg(feature="master")]
+    pub fn type_uint_from_ty(&self, t: ty::UintTy) -> Type<'gcc> {
+        match t {
+            ty::UintTy::Usize => self.type_isize(),
+            ty::UintTy::U8 => self.type_i8(),
+            ty::UintTy::U16 => self.type_i16(),
+            ty::UintTy::U32 => self.type_i32(),
+            ty::UintTy::U64 => self.type_i64(),
+            ty::UintTy::U128 => self.type_i128(),
+        }
+    }
 }
 
 pub fn uncached_gcc_type<'gcc, 'tcx>(cx: &CodegenCx<'gcc, 'tcx>, layout: TyAndLayout<'tcx>, defer: &mut Option<(Struct<'gcc>, TyAndLayout<'tcx>)>) -> Type<'gcc> {

compiler/rustc_codegen_gcc/test.sh

@@ -1,4 +1,4 @@
-#!/bin/bash
+#!/usr/bin/env bash
 
 # TODO(antoyo): rewrite to cargo-make (or just) or something like that to only rebuild the sysroot when needed?
 
@@ -14,25 +14,87 @@ fi
 export LD_LIBRARY_PATH="$GCC_PATH"
 export LIBRARY_PATH="$GCC_PATH"
 
-features=
-
-if [[ "$1" == "--features" ]]; then
-    shift
-    features="--features $1"
-    shift
-fi
-
-if [[ "$1" == "--release" ]]; then
+flags=
+gcc_master_branch=1
+channel="debug"
+func=all
+build_only=0
+
+while [[ $# -gt 0 ]]; do
+    case $1 in
+        --release)
+            codegen_channel=release
+            shift
+            ;;
+        --release-sysroot)
+            sysroot_channel=release
+            shift
+            ;;
+        --no-default-features)
+            gcc_master_branch=0
+            flags="$flags --no-default-features"
+            shift
+            ;;
+        --features)
+            shift
+            flags="$flags --features $1"
+            shift
+            ;;
+        --release)
+            channel="release"
+            shift
+            ;;
+        "--test-rustc")
+            func=test_rustc
+            shift
+            ;;
+
+        "--test-libcore")
+            func=test_libcore
+            shift
+            ;;
+
+        "--clean-ui-tests")
+            func=clean_ui_tests
+            shift
+            ;;
+
+        "--std-tests")
+            func=std_tests
+            shift
+            ;;
+
+        "--extended-tests")
+            func=extended_sysroot_tests
+            shift
+            ;;
+
+        "--build-sysroot")
+            func=build_sysroot
+            shift
+            ;;
+        "--build")
+            build_only=1
+            shift
+            ;;
+        *)
+            echo "Unknown option $1"
+            exit 1
+            ;;
+    esac
+done
+
+if [[ $channel == "release" ]]; then
     export CHANNEL='release'
-    CARGO_INCREMENTAL=1 cargo rustc --release $features
+    CARGO_INCREMENTAL=1 cargo rustc --release $flags
     shift
 else
     echo $LD_LIBRARY_PATH
     export CHANNEL='debug'
-    cargo rustc $features
+    cargo rustc $flags
 fi
 
-if [[ "$1" == "--build" ]]; then
+if (( $build_only == 1 )); then
     exit
 fi
 
@@ -78,7 +140,11 @@ function std_tests() {
     $RUN_WRAPPER ./target/out/dst_field_align || (echo $?; false)
 
     echo "[AOT] std_example"
-    $RUSTC example/std_example.rs --crate-type bin --target $TARGET_TRIPLE
+    std_flags="--cfg feature=\"master\""
+    if (( $gcc_master_branch == 0 )); then
+        std_flags=""
+    fi
+    $RUSTC example/std_example.rs --crate-type bin --target $TARGET_TRIPLE $std_flags
     $RUN_WRAPPER ./target/out/std_example --target $TARGET_TRIPLE
 
     echo "[AOT] subslice-patterns-const-eval"
@@ -97,25 +163,6 @@ function std_tests() {
 #echo "[BUILD] sysroot in release mode"
 #./build_sysroot/build_sysroot.sh --release
 
-# TODO(antoyo): uncomment when it works.
-#pushd simple-raytracer
-#if [[ "$HOST_TRIPLE" = "$TARGET_TRIPLE" ]]; then
-    #echo "[BENCH COMPILE] ebobby/simple-raytracer"
-    #hyperfine --runs ${RUN_RUNS:-10} --warmup 1 --prepare "rm -r target/*/debug || true" \
-    #"RUSTFLAGS='' cargo build --target $TARGET_TRIPLE" \
-    #"../cargo.sh build"
-
-    #echo "[BENCH RUN] ebobby/simple-raytracer"
-    #cp ./target/*/debug/main ./raytracer_cg_gccjit
-    #hyperfine --runs ${RUN_RUNS:-10} ./raytracer_cg_llvm ./raytracer_cg_gccjit
-#else
-    #echo "[BENCH COMPILE] ebobby/simple-raytracer (skipped)"
-    #echo "[COMPILE] ebobby/simple-raytracer"
-    #../cargo.sh build
-    #echo "[BENCH RUN] ebobby/simple-raytracer (skipped)"
-#fi
-#popd
-
 function test_libcore() {
     pushd build_sysroot/sysroot_src/library/core/tests
     echo "[TEST] libcore"
@@ -124,19 +171,6 @@ function test_libcore() {
     popd
 }
 
-# TODO(antoyo): uncomment when it works.
-#pushd regex
-#echo "[TEST] rust-lang/regex example shootout-regex-dna"
-#../cargo.sh clean
-## Make sure `[codegen mono items] start` doesn't poison the diff
-#../cargo.sh build --example shootout-regex-dna
-#cat examples/regexdna-input.txt | ../cargo.sh run --example shootout-regex-dna | grep -v "Spawned thread" > res.txt
-#diff -u res.txt examples/regexdna-output.txt
-
-#echo "[TEST] rust-lang/regex tests"
-#../cargo.sh test --tests -- --exclude-should-panic --test-threads 1 -Zunstable-options
-#popd
-
 #echo
 #echo "[BENCH COMPILE] mod_bench"
 
@@ -153,6 +187,44 @@ function test_libcore() {
 #echo "[BENCH RUN] mod_bench"
 #hyperfine --runs ${RUN_RUNS:-10} ./target/out/mod_bench{,_inline} ./target/out/mod_bench_llvm_*
 
+function extended_sysroot_tests() {
+    if (( $gcc_master_branch == 0 )); then
+        return
+    fi
+
+    pushd rand
+    cargo clean
+    echo "[TEST] rust-random/rand"
+    ../cargo.sh test --workspace
+    popd
+
+    #pushd simple-raytracer
+    #echo "[BENCH COMPILE] ebobby/simple-raytracer"
+    #hyperfine --runs "${RUN_RUNS:-10}" --warmup 1 --prepare "cargo clean" \
+    #"RUSTC=rustc RUSTFLAGS='' cargo build" \
+    #"../cargo.sh build"
+
+    #echo "[BENCH RUN] ebobby/simple-raytracer"
+    #cp ./target/debug/main ./raytracer_cg_gcc
+    #hyperfine --runs "${RUN_RUNS:-10}" ./raytracer_cg_llvm ./raytracer_cg_gcc
+    #popd
+
+    pushd regex
+    echo "[TEST] rust-lang/regex example shootout-regex-dna"
+    cargo clean
+    export CG_RUSTFLAGS="--cap-lints warn" # newer aho_corasick versions throw a deprecation warning
+    # Make sure `[codegen mono items] start` doesn't poison the diff
+    ../cargo.sh build --example shootout-regex-dna
+    cat examples/regexdna-input.txt \
+        | ../cargo.sh run --example shootout-regex-dna \
+        | grep -v "Spawned thread" > res.txt
+    diff -u res.txt examples/regexdna-output.txt
+
+    echo "[TEST] rust-lang/regex tests"
+    ../cargo.sh test --tests -- --exclude-should-panic --test-threads 1 -Zunstable-options -q
+    popd
+}
+
 function test_rustc() {
     echo
     echo "[TEST] rust-lang/rust"
@@ -165,23 +237,7 @@ function test_rustc() {
     git checkout $(rustc -V | cut -d' ' -f3 | tr -d '(')
     export RUSTFLAGS=
 
-    git apply - <<EOF
-diff --git a/src/tools/compiletest/src/header.rs b/src/tools/compiletest/src/header.rs
-index 887d27fd6dca4..2c2239f2b83d1 100644
---- a/src/tools/compiletest/src/header.rs
-+++ b/src/tools/compiletest/src/header.rs
-@@ -806,8 +806,8 @@ pub fn make_test_description<R: Read>(
-     cfg: Option<&str>,
- ) -> test::TestDesc {
-     let mut ignore = false;
-     #[cfg(not(bootstrap))]
--    let ignore_message: Option<String> = None;
-+    let ignore_message: Option<&str> = None;
-     let mut should_fail = false;
-
-     let rustc_has_profiler_support = env::var_os("RUSTC_PROFILER_SUPPORT").is_some();
-
-EOF
+    git apply ../rustc_patches/compile_test.patch || true
 
     rm config.toml || true
 
@@ -205,7 +261,7 @@ EOF
 
     git checkout -- src/test/ui/issues/auxiliary/issue-3136-a.rs # contains //~ERROR, but shouldn't be removed
 
-    rm -r src/test/ui/{abi*,extern/,panic-runtime/,panics/,unsized-locals/,proc-macro/,threads-sendsync/,thinlto/,simd*,borrowck/,test*,*lto*.rs} || true
+    rm -r src/test/ui/{abi*,extern/,panic-runtime/,panics/,unsized-locals/,proc-macro/,threads-sendsync/,thinlto/,borrowck/,test*,*lto*.rs} || true
     for test in $(rg --files-with-matches "catch_unwind|should_panic|thread|lto" src/test/ui); do
       rm $test
     done
@@ -222,33 +278,14 @@ function clean_ui_tests() {
     find rust/build/x86_64-unknown-linux-gnu/test/ui/ -name stamp -exec rm -rf {} \;
 }
 
-case $1 in
-    "--test-rustc")
-        test_rustc
-        ;;
-
-    "--test-libcore")
-        test_libcore
-        ;;
-
-    "--clean-ui-tests")
-        clean_ui_tests
-        ;;
-
-    "--std-tests")
-        std_tests
-        ;;
-
-    "--build-sysroot")
-        build_sysroot
-        ;;
-
-    *)
-        clean
-        mini_tests
-        build_sysroot
-        std_tests
-        test_libcore
-        test_rustc
-        ;;
-esac
+function all() {
+    clean
+    mini_tests
+    build_sysroot
+    std_tests
+    test_libcore
+    extended_sysroot_tests
+    test_rustc
+}
+
+$func

compiler/rustc_codegen_gcc/tests/lib.rs → compiler/rustc_codegen_gcc/tests/lang_tests_common.rs

@@ -1,3 +1,4 @@
+//! The common code for `tests/lang_tests_*.rs`
 use std::{
     env::{self, current_dir},
     path::PathBuf,
@@ -7,7 +8,15 @@ use std::{
 use lang_tester::LangTester;
 use tempfile::TempDir;
 
-fn main() {
+/// Controls the compile options (e.g., optimization level) used to compile
+/// test code.
+#[allow(dead_code)] // Each test crate picks one variant
+pub enum Profile {
+    Debug,
+    Release,
+}
+
+pub fn main_inner(profile: Profile) {
     let tempdir = TempDir::new().expect("temp dir");
     let current_dir = current_dir().expect("current dir");
     let current_dir = current_dir.to_str().expect("current dir").to_string();
@@ -42,6 +51,15 @@ fn main() {
                 "-o", exe.to_str().expect("to_str"),
                 path.to_str().expect("to_str"),
             ]);
+            match profile {
+                Profile::Debug => {}
+                Profile::Release => {
+                    compiler.args(&[
+                        "-C", "opt-level=3",
+                        "-C", "lto=no",
+                    ]);
+                }
+            }
             // Test command 2: run `tempdir/x`.
             let runtime = Command::new(exe);
             vec![("Compiler", compiler), ("Run-time", runtime)]

compiler/rustc_codegen_gcc/tests/lang_tests_debug.rs

@@ -0,0 +1,5 @@
+mod lang_tests_common;
+
+fn main() {
+    lang_tests_common::main_inner(lang_tests_common::Profile::Debug);
+}

compiler/rustc_codegen_gcc/tests/lang_tests_release.rs

@@ -0,0 +1,5 @@
+mod lang_tests_common;
+
+fn main() {
+    lang_tests_common::main_inner(lang_tests_common::Profile::Release);
+}

compiler/rustc_codegen_gcc/tests/run/int_overflow.rs

@@ -1,7 +1,7 @@
 // Compiler:
 //
 // Run-time:
-//   stdout: Panicking
+//   stdout: Success
 //   status: signal
 
 #![allow(unused_attributes)]
@@ -64,7 +64,9 @@ mod intrinsics {
 #[no_mangle]
 pub fn panic(_msg: &str) -> ! {
     unsafe {
-        libc::puts("Panicking\0" as *const str as *const u8);
+        // Panicking is expected iff overflow checking is enabled.
+        #[cfg(debug_assertions)]
+        libc::puts("Success\0" as *const str as *const u8);
         libc::fflush(libc::stdout);
         intrinsics::abort();
     }
@@ -124,6 +126,15 @@ impl Add for isize {
 #[start]
 fn main(mut argc: isize, _argv: *const *const u8) -> isize {
     let int = 9223372036854775807isize;
-    let int = int + argc;
+    let int = int + argc;  // overflow
+
+    // If overflow checking is disabled, we should reach here.
+    #[cfg(not(debug_assertions))]
+    unsafe {
+        libc::puts("Success\0" as *const str as *const u8);
+        libc::fflush(libc::stdout);
+        intrinsics::abort();
+    }
+
     int
 }

compiler/rustc_codegen_gcc/tools/generate_intrinsics.py

@@ -0,0 +1,238 @@
+import json
+import os
+import re
+import sys
+import subprocess
+from os import walk
+
+
+def run_command(command, cwd=None):
+    p = subprocess.Popen(command, cwd=cwd)
+    if p.wait() != 0:
+        print("command `{}` failed...".format(" ".join(command)))
+        sys.exit(1)
+
+
+def clone_repository(repo_name, path, repo_url, sub_path=None):
+    if os.path.exists(path):
+        while True:
+            choice = input("There is already a `{}` folder, do you want to update it? [y/N]".format(path))
+            if choice == "" or choice.lower() == "n":
+                print("Skipping repository update.")
+                return
+            elif choice.lower() == "y":
+                print("Updating repository...")
+                run_command(["git", "pull", "origin"], cwd=path)
+                return
+            else:
+                print("Didn't understand answer...")
+    print("Cloning {} repository...".format(repo_name))
+    if sub_path is None:
+        run_command(["git", "clone", repo_url, "--depth", "1", path])
+    else:
+        run_command(["git", "clone", repo_url, "--filter=tree:0", "--no-checkout", path])
+        run_command(["git", "sparse-checkout", "init"], cwd=path)
+        run_command(["git", "sparse-checkout", "set", "add", sub_path], cwd=path)
+        run_command(["git", "checkout"], cwd=path)
+
+
+def append_intrinsic(array, intrinsic_name, translation):
+    array.append((intrinsic_name, translation))
+
+
+def extract_instrinsics(intrinsics, file):
+    print("Extracting intrinsics from `{}`...".format(file))
+    with open(file, "r", encoding="utf8") as f:
+        content = f.read()
+
+    lines = content.splitlines()
+    pos = 0
+    current_arch = None
+    while pos < len(lines):
+        line = lines[pos].strip()
+        if line.startswith("let TargetPrefix ="):
+            current_arch = line.split('"')[1].strip()
+            if len(current_arch) == 0:
+                current_arch = None
+        elif current_arch is None:
+            pass
+        elif line == "}":
+            current_arch = None
+        elif line.startswith("def "):
+            content = ""
+            while not content.endswith(";") and not content.endswith("}") and pos < len(lines):
+                line = lines[pos].split(" // ")[0].strip()
+                content += line
+                pos += 1
+            entries = re.findall('GCCBuiltin<"(\\w+)">', content)
+            if len(entries) > 0:
+                intrinsic = content.split("def ")[1].strip().split(":")[0].strip()
+                intrinsic = intrinsic.split("_")
+                if len(intrinsic) < 2 or intrinsic[0] != "int":
+                    continue
+                intrinsic[0] = "llvm"
+                intrinsic = ".".join(intrinsic)
+                if current_arch not in intrinsics:
+                    intrinsics[current_arch] = []
+                for entry in entries:
+                    append_intrinsic(intrinsics[current_arch], intrinsic, entry)
+            continue
+        pos += 1
+        continue
+    print("Done!")
+
+
+def extract_instrinsics_from_llvm(llvm_path, intrinsics):
+    files = []
+    intrinsics_path = os.path.join(llvm_path, "llvm/include/llvm/IR")
+    for (dirpath, dirnames, filenames) in walk(intrinsics_path):
+        files.extend([os.path.join(intrinsics_path, f) for f in filenames if f.endswith(".td")])
+
+    for file in files:
+        extract_instrinsics(intrinsics, file)
+
+
+def append_translation(json_data, p, array):
+    it = json_data["index"][p]
+    content = it["docs"].split('`')
+    if len(content) != 5:
+        return
+    append_intrinsic(array, content[1], content[3])
+
+
+def extract_instrinsics_from_llvmint(llvmint, intrinsics):
+    archs = [
+        "AMDGPU",
+        "aarch64",
+        "arm",
+        "cuda",
+        "hexagon",
+        "mips",
+        "nvvm",
+        "ppc",
+        "ptx",
+        "x86",
+        "xcore",
+    ]
+
+    json_file = os.path.join(llvmint, "target/doc/llvmint.json")
+    # We need to regenerate the documentation!
+    run_command(
+        ["cargo", "rustdoc", "--", "-Zunstable-options", "--output-format", "json"],
+        cwd=llvmint,
+    )
+    with open(json_file, "r", encoding="utf8") as f:
+        json_data = json.loads(f.read())
+    for p in json_data["paths"]:
+        it = json_data["paths"][p]
+        if it["crate_id"] != 0:
+            # This is from an external crate.
+            continue
+        if it["kind"] != "function":
+            # We're only looking for functions.
+            continue
+        # if len(it["path"]) == 2:
+        #   # This is a "general" intrinsic, not bound to a specific arch.
+        #   append_translation(json_data, p, general)
+        #   continue
+        if len(it["path"]) != 3 or it["path"][1] not in archs:
+            continue
+        arch = it["path"][1]
+        if arch not in intrinsics:
+            intrinsics[arch] = []
+        append_translation(json_data, p, intrinsics[arch])
+
+
+def fill_intrinsics(intrinsics, from_intrinsics, all_intrinsics):
+    for arch in from_intrinsics:
+        if arch not in intrinsics:
+            intrinsics[arch] = []
+        for entry in from_intrinsics[arch]:
+            if entry[0] in all_intrinsics:
+                if all_intrinsics[entry[0]] == entry[1]:
+                    # This is a "full" duplicate, both the LLVM instruction and the GCC
+                    # translation are the same.
+                    continue
+                intrinsics[arch].append((entry[0], entry[1], True))
+            else:
+                intrinsics[arch].append((entry[0], entry[1], False))
+                all_intrinsics[entry[0]] = entry[1]
+
+
+def update_intrinsics(llvm_path, llvmint, llvmint2):
+    intrinsics_llvm = {}
+    intrinsics_llvmint = {}
+    all_intrinsics = {}
+
+    extract_instrinsics_from_llvm(llvm_path, intrinsics_llvm)
+    extract_instrinsics_from_llvmint(llvmint, intrinsics_llvmint)
+    extract_instrinsics_from_llvmint(llvmint2, intrinsics_llvmint)
+
+    intrinsics = {}
+    # We give priority to translations from LLVM over the ones from llvmint.
+    fill_intrinsics(intrinsics, intrinsics_llvm, all_intrinsics)
+    fill_intrinsics(intrinsics, intrinsics_llvmint, all_intrinsics)
+
+    archs = [arch for arch in intrinsics]
+    archs.sort()
+
+    output_file = os.path.join(
+        os.path.dirname(os.path.abspath(__file__)),
+        "../src/intrinsic/archs.rs",
+    )
+    print("Updating content of `{}`...".format(output_file))
+    with open(output_file, "w", encoding="utf8") as out:
+        out.write("// File generated by `rustc_codegen_gcc/tools/generate_intrinsics.py`\n")
+        out.write("// DO NOT EDIT IT!\n")
+        out.write("match name {\n")
+        for arch in archs:
+            if len(intrinsics[arch]) == 0:
+                continue
+            intrinsics[arch].sort(key=lambda x: (x[0], x[2]))
+            out.write('    // {}\n'.format(arch))
+            for entry in intrinsics[arch]:
+                if entry[2] == True: # if it is a duplicate
+                    out.write('    // [DUPLICATE]: "{}" => "{}",\n'.format(entry[0], entry[1]))
+                else:
+                    out.write('    "{}" => "{}",\n'.format(entry[0], entry[1]))
+        out.write('    _ => unimplemented!("***** unsupported LLVM intrinsic {}", name),\n')
+        out.write("}\n")
+    print("Done!")
+
+
+def main():
+    llvm_path = os.path.join(
+        os.path.dirname(os.path.abspath(__file__)),
+        "llvm-project",
+    )
+    llvmint_path = os.path.join(
+        os.path.dirname(os.path.abspath(__file__)),
+        "llvmint",
+    )
+    llvmint2_path = os.path.join(
+        os.path.dirname(os.path.abspath(__file__)),
+        "llvmint-2",
+    )
+
+    # First, we clone the LLVM repository if it's not already here.
+    clone_repository(
+        "llvm-project",
+        llvm_path,
+        "https://github.com/llvm/llvm-project",
+        sub_path="llvm/include/llvm/IR",
+    )
+    clone_repository(
+        "llvmint",
+        llvmint_path,
+        "https://github.com/GuillaumeGomez/llvmint",
+    )
+    clone_repository(
+        "llvmint2",
+        llvmint2_path,
+        "https://github.com/antoyo/llvmint",
+    )
+    update_intrinsics(llvm_path, llvmint_path, llvmint2_path)
+
+
+if __name__ == "__main__":
+    sys.exit(main())