diff --git a/.github/workflows/m68k.yml b/.github/workflows/m68k.yml index 400eb757405..1c864e04413 100644 --- a/.github/workflows/m68k.yml +++ b/.github/workflows/m68k.yml @@ -85,14 +85,14 @@ jobs: - name: Build sample project with target defined as JSON spec run: | ./y.sh prepare --only-libcore --cross - ./y.sh build --sysroot --target-triple m68k-unknown-linux-gnu --target ${{ github.workspace }}/target_specs/m68k-unknown-linux-gnu.json + ./y.sh build --sysroot --features compiler_builtins/no-f16-f128 --target-triple m68k-unknown-linux-gnu --target ${{ github.workspace }}/target_specs/m68k-unknown-linux-gnu.json ./y.sh cargo build --manifest-path=./tests/hello-world/Cargo.toml --target ${{ github.workspace }}/target_specs/m68k-unknown-linux-gnu.json ./y.sh clean all - name: Build run: | ./y.sh prepare --only-libcore --cross - ./y.sh build --sysroot --target-triple m68k-unknown-linux-gnu + ./y.sh build --sysroot --features compiler_builtins/no-f16-f128 --target-triple m68k-unknown-linux-gnu CG_GCC_TEST_TARGET=m68k-unknown-linux-gnu cargo test ./y.sh clean all @@ -107,4 +107,4 @@ jobs: - name: Run tests run: | - ./y.sh test --release --clean --build-sysroot ${{ matrix.commands }} + ./y.sh test --release --clean --build-sysroot --sysroot-features compiler_builtins/no-f16-f128 ${{ matrix.commands }} diff --git a/.github/workflows/stdarch.yml b/.github/workflows/stdarch.yml index 33095bdb5f7..d8818eefa96 100644 --- a/.github/workflows/stdarch.yml +++ b/.github/workflows/stdarch.yml @@ -13,7 +13,7 @@ env: jobs: build: - runs-on: ubuntu-latest + runs-on: ubuntu-24.04 strategy: fail-fast: false @@ -36,6 +36,13 @@ jobs: - name: Install packages run: sudo apt-get install ninja-build ripgrep + # TODO: remove when we have binutils version 2.43 in the repo. + - name: Install more recent binutils + run: | + echo "deb http://archive.ubuntu.com/ubuntu oracular main universe" | sudo tee /etc/apt/sources.list.d/oracular-copies.list + sudo apt-get update + sudo apt-get install binutils + - name: Install Intel Software Development Emulator if: ${{ matrix.cargo_runner }} run: | @@ -96,4 +103,5 @@ jobs: run: | # FIXME: these tests fail when the sysroot is compiled with LTO because of a missing symbol in proc-macro. # TODO: remove --skip test_mm512_stream_ps when stdarch is updated in rustc. - STDARCH_TEST_EVERYTHING=1 CHANNEL=release CARGO_TARGET_X86_64_UNKNOWN_LINUX_GNU_RUNNER="${{ matrix.cargo_runner }}" TARGET=x86_64-unknown-linux-gnu CG_RUSTFLAGS="-Ainternal_features" ./y.sh cargo test --manifest-path build/build_sysroot/sysroot_src/library/stdarch/Cargo.toml -- --skip rtm --skip tbm --skip sse4a --skip test_mm512_stream_ps + # TODO: remove --skip test_tile_ when it's implemented. + STDARCH_TEST_EVERYTHING=1 CHANNEL=release CARGO_TARGET_X86_64_UNKNOWN_LINUX_GNU_RUNNER="${{ matrix.cargo_runner }}" TARGET=x86_64-unknown-linux-gnu CG_RUSTFLAGS="-Ainternal_features --cfg stdarch_intel_sde" ./y.sh cargo test --manifest-path build/build_sysroot/sysroot_src/library/stdarch/Cargo.toml -- --skip rtm --skip tbm --skip sse4a --skip test_mm512_stream_ps --skip test_tile_ diff --git a/.rustfmt.toml b/.rustfmt.toml index 2a35f0230c6..725aec25a07 100644 --- a/.rustfmt.toml +++ b/.rustfmt.toml @@ -1 +1,3 @@ +version = "Two" use_small_heuristics = "Max" +merge_derives = false diff --git a/Cargo.lock b/Cargo.lock index 915229f7e7e..f2a368395b2 100644 --- a/Cargo.lock +++ b/Cargo.lock @@ -79,18 +79,18 @@ dependencies = [ [[package]] name = "gccjit" -version = "2.1.0" +version = "2.2.0" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "62e0ba949ebee07c5cc21f02cb48f28f2c8db7fcbc15fdc5120476a6c43b4636" +checksum = "4bb376e98c82d9284c3a17fc1d6bf9bc921055418950238d7a553c27a7e1f6ab" dependencies = [ "gccjit_sys", ] [[package]] name = "gccjit_sys" -version = "0.2.0" +version = "0.3.0" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "a5bbf85e12c2593772329a9d4e8310271f6706e6045ce4f41b041dd34fba6603" +checksum = "93b4b1be553b5df790bf25ca2a1d6add81727dc29f8d5c8742468ed306d621d1" dependencies = [ "libc", ] diff --git a/Cargo.toml b/Cargo.toml index 5caca63f634..22b953cd2d1 100644 --- a/Cargo.toml +++ b/Cargo.toml @@ -22,7 +22,8 @@ master = ["gccjit/master"] default = ["master"] [dependencies] -gccjit = "2.1" +gccjit = "2.2" +#gccjit = { git = "https://github.com/rust-lang/gccjit.rs" } # Local copy. #gccjit = { path = "../gccjit.rs" } diff --git a/build_system/build_sysroot/Cargo.lock b/build_system/build_sysroot/Cargo.lock index d6ec1f87d01..51bec5aa9e3 100644 --- a/build_system/build_sysroot/Cargo.lock +++ b/build_system/build_sysroot/Cargo.lock @@ -4,12 +4,12 @@ version = 3 [[package]] name = "addr2line" -version = "0.21.0" +version = "0.22.0" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "8a30b2e23b9e17a9f90641c7ab1549cd9b44f296d3ccbf309d2863cfe398a0cb" +checksum = "6e4503c46a5c0c7844e948c9a4d6acd9f50cccb4de1c48eb9e291ea17470c678" dependencies = [ "compiler_builtins", - "gimli", + "gimli 0.29.0", "rustc-std-workspace-alloc", "rustc-std-workspace-core", ] @@ -50,9 +50,9 @@ dependencies = [ [[package]] name = "compiler_builtins" -version = "0.1.109" +version = "0.1.118" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "f11973008a8cf741fe6d22f339eba21fd0ca81e2760a769ba8243ed6c21edd7e" +checksum = "92afe7344b64cccf3662ca26d5d1c0828ab826f04206b97d856e3625e390e4b5" dependencies = [ "rustc-std-workspace-core", ] @@ -97,9 +97,20 @@ dependencies = [ [[package]] name = "gimli" -version = "0.28.1" +version = "0.29.0" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "4271d37baee1b8c7e4b708028c57d816cf9d2434acb33a549475f78c181f6253" +checksum = "40ecd4077b5ae9fd2e9e169b102c6c330d0605168eb0e8bf79952b256dbefffd" +dependencies = [ + "compiler_builtins", + "rustc-std-workspace-alloc", + "rustc-std-workspace-core", +] + +[[package]] +name = "gimli" +version = "0.30.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "e2e1d97fbe9722ba9bbd0c97051c2956e726562b61f86a25a4360398a40edfc9" dependencies = [ "compiler_builtins", "rustc-std-workspace-alloc", @@ -120,9 +131,9 @@ dependencies = [ [[package]] name = "hermit-abi" -version = "0.3.9" +version = "0.4.0" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "d231dfb89cfffdbc30e7fc41579ed6066ad03abda9e567ccafae602b97ec5024" +checksum = "fbf6a919d6cf397374f7dfeeea91d974c7c0a7221d0d0f4f20d859d329e53fcc" dependencies = [ "compiler_builtins", "rustc-std-workspace-alloc", @@ -131,18 +142,18 @@ dependencies = [ [[package]] name = "libc" -version = "0.2.153" +version = "0.2.155" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "9c198f91728a82281a64e1f4f9eeb25d82cb32a5de251c6bd1b5154d63a8e7bd" +checksum = "97b3888a4aecf77e811145cadf6eef5901f4782c53886191b2f693f24761847c" dependencies = [ "rustc-std-workspace-core", ] [[package]] name = "memchr" -version = "2.7.2" +version = "2.7.4" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "6c8640c5d730cb13ebd907d8d04b52f55ac9a2eec55b440c8892f40d56c76c1d" +checksum = "78ca9ab1a0babb1e7d5695e3530886289c18cf2f87ec19a575a0abdce112e3a3" dependencies = [ "compiler_builtins", "rustc-std-workspace-core", @@ -150,9 +161,9 @@ dependencies = [ [[package]] name = "miniz_oxide" -version = "0.7.2" +version = "0.7.4" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "9d811f3e15f28568be3407c8e7fdb6514c1cda3cb30683f15b6a1a1dc4ea14a7" +checksum = "b8a240ddb74feaf34a79a7add65a741f3167852fba007066dcac1ca548d89c08" dependencies = [ "adler", "compiler_builtins", @@ -162,9 +173,9 @@ dependencies = [ [[package]] name = "object" -version = "0.32.2" +version = "0.36.3" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "a6a622008b6e321afc04970976f62ee297fdbaa6f95318ca343e3eebb9648441" +checksum = "27b64972346851a39438c60b341ebc01bba47464ae329e55cf343eb93964efd9" dependencies = [ "compiler_builtins", "memchr", @@ -205,9 +216,9 @@ dependencies = [ [[package]] name = "r-efi" -version = "4.4.0" +version = "4.5.0" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "c47196f636c4cc0634b73b0405323d177753c2e15e866952c64ea22902567a34" +checksum = "e9e935efc5854715dfc0a4c9ef18dc69dee0ec3bf9cc3ab740db831c0fdd86a3" dependencies = [ "compiler_builtins", "rustc-std-workspace-core", @@ -226,9 +237,9 @@ dependencies = [ [[package]] name = "rustc-demangle" -version = "0.1.23" +version = "0.1.24" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "d626bb9dae77e28219937af045c257c28bfd3f69333c512553507f5f9798cb76" +checksum = "719b953e2095829ee67db738b3bfa9fa368c94900df327b3f07fe6e794d2fe1f" dependencies = [ "compiler_builtins", "rustc-std-workspace-core", @@ -310,16 +321,14 @@ dependencies = [ "core", "getopts", "libc", - "panic_abort", - "panic_unwind", "std", ] [[package]] name = "unicode-width" -version = "0.1.12" +version = "0.1.13" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "68f5e5f3158ecfd4b8ff6fe086db7c8467a2dfdac97fe420f2b7c4aa97af66d6" +checksum = "0336d538f7abc86d282a4189614dfaa90810dfc2c6f6427eaf88e16311dd225d" dependencies = [ "compiler_builtins", "rustc-std-workspace-core", @@ -339,12 +348,12 @@ dependencies = [ [[package]] name = "unwinding" -version = "0.2.1" +version = "0.2.2" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "37a19a21a537f635c16c7576f22d0f2f7d63353c1337ad4ce0d8001c7952a25b" +checksum = "dc55842d0db6329a669d55a623c674b02d677b16bfb2d24857d4089d41eba882" dependencies = [ "compiler_builtins", - "gimli", + "gimli 0.30.0", "rustc-std-workspace-core", ] @@ -370,9 +379,9 @@ dependencies = [ [[package]] name = "windows-targets" -version = "0.52.5" +version = "0.52.6" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "6f0713a46559409d202e70e28227288446bf7841d3211583a4b53e3f6d96e7eb" +checksum = "9b724f72796e036ab90c1021d4780d4d3d648aca59e491e6b98e725b84e99973" dependencies = [ "windows_aarch64_gnullvm", "windows_aarch64_msvc", @@ -386,48 +395,48 @@ dependencies = [ [[package]] name = "windows_aarch64_gnullvm" -version = "0.52.5" +version = "0.52.6" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "7088eed71e8b8dda258ecc8bac5fb1153c5cffaf2578fc8ff5d61e23578d3263" +checksum = "32a4622180e7a0ec044bb555404c800bc9fd9ec262ec147edd5989ccd0c02cd3" [[package]] name = "windows_aarch64_msvc" -version = "0.52.5" +version = "0.52.6" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "9985fd1504e250c615ca5f281c3f7a6da76213ebd5ccc9561496568a2752afb6" +checksum = "09ec2a7bb152e2252b53fa7803150007879548bc709c039df7627cabbd05d469" [[package]] name = "windows_i686_gnu" -version = "0.52.5" +version = "0.52.6" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "88ba073cf16d5372720ec942a8ccbf61626074c6d4dd2e745299726ce8b89670" +checksum = "8e9b5ad5ab802e97eb8e295ac6720e509ee4c243f69d781394014ebfe8bbfa0b" [[package]] name = "windows_i686_gnullvm" -version = "0.52.5" +version = "0.52.6" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "87f4261229030a858f36b459e748ae97545d6f1ec60e5e0d6a3d32e0dc232ee9" +checksum = "0eee52d38c090b3caa76c563b86c3a4bd71ef1a819287c19d586d7334ae8ed66" [[package]] name = "windows_i686_msvc" -version = "0.52.5" +version = "0.52.6" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "db3c2bf3d13d5b658be73463284eaf12830ac9a26a90c717b7f771dfe97487bf" +checksum = "240948bc05c5e7c6dabba28bf89d89ffce3e303022809e73deaefe4f6ec56c66" [[package]] name = "windows_x86_64_gnu" -version = "0.52.5" +version = "0.52.6" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "4e4246f76bdeff09eb48875a0fd3e2af6aada79d409d33011886d3e1581517d9" +checksum = "147a5c80aabfbf0c7d901cb5895d1de30ef2907eb21fbbab29ca94c5b08b1a78" [[package]] name = "windows_x86_64_gnullvm" -version = "0.52.5" +version = "0.52.6" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "852298e482cd67c356ddd9570386e2862b5673c85bd5f88df9ab6802b334c596" +checksum = "24d5b23dc417412679681396f2b49f3de8c1473deb516bd34410872eff51ed0d" [[package]] name = "windows_x86_64_msvc" -version = "0.52.5" +version = "0.52.6" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "bec47e5bfd1bff0eeaf6d8b485cc1074891a197ab4225d504cb7a1ab88b02bf0" +checksum = "589f6da84c646204747d1270a2a5661ea66ed1cced2631d546fdfb155959f9ec" diff --git a/build_system/build_sysroot/Cargo.toml b/build_system/build_sysroot/Cargo.toml index e4669923623..24152070e64 100644 --- a/build_system/build_sysroot/Cargo.toml +++ b/build_system/build_sysroot/Cargo.toml @@ -6,9 +6,7 @@ resolver = "2" [dependencies] core = { path = "./sysroot_src/library/core" } -# TODO: after the sync, revert to using version 0.1. -# compiler_builtins = "0.1" -compiler_builtins = "=0.1.109" +compiler_builtins = "0.1" alloc = { path = "./sysroot_src/library/alloc" } std = { path = "./sysroot_src/library/std", features = ["panic_unwind", "backtrace"] } test = { path = "./sysroot_src/library/test" } @@ -19,6 +17,22 @@ rustc-std-workspace-core = { path = "./sysroot_src/library/rustc-std-workspace-c rustc-std-workspace-alloc = { path = "./sysroot_src/library/rustc-std-workspace-alloc" } rustc-std-workspace-std = { path = "./sysroot_src/library/rustc-std-workspace-std" } +# For compiler-builtins we always use a high number of codegen units. +# The goal here is to place every single intrinsic into its own object +# file to avoid symbol clashes with the system libgcc if possible. Note +# that this number doesn't actually produce this many object files, we +# just don't create more than this number of object files. +# +# It's a bit of a bummer that we have to pass this here, unfortunately. +# Ideally this would be specified through an env var to Cargo so Cargo +# knows how many CGUs are for this specific crate, but for now +# per-crate configuration isn't specifiable in the environment. +[profile.dev.package.compiler_builtins] +codegen-units = 10000 + +[profile.release.package.compiler_builtins] +codegen-units = 10000 + [profile.release] debug = "limited" #lto = "fat" # TODO(antoyo): re-enable when the failing LTO tests regarding proc-macros are fixed. diff --git a/build_system/src/build.rs b/build_system/src/build.rs index d465ab7e506..d0ced211a61 100644 --- a/build_system/src/build.rs +++ b/build_system/src/build.rs @@ -1,12 +1,13 @@ -use crate::config::{Channel, ConfigInfo}; -use crate::utils::{ - copy_file, create_dir, get_sysroot_dir, run_command, run_command_with_output_and_env, walk_dir, -}; use std::collections::HashMap; use std::ffi::OsStr; use std::fs; use std::path::Path; +use crate::config::{Channel, ConfigInfo}; +use crate::utils::{ + copy_file, create_dir, get_sysroot_dir, run_command, run_command_with_output_and_env, walk_dir, +}; + #[derive(Default)] struct BuildArg { flags: Vec, @@ -23,16 +24,6 @@ fn new() -> Result, String> { while let Some(arg) = args.next() { match arg.as_str() { - "--features" => { - if let Some(arg) = args.next() { - build_arg.flags.push("--features".to_string()); - build_arg.flags.push(arg.as_str().into()); - } else { - return Err( - "Expected a value after `--features`, found nothing".to_string() - ); - } - } "--sysroot" => { build_arg.build_sysroot = true; } @@ -55,7 +46,6 @@ fn usage() { r#" `build` command help: - --features [arg] : Add a new feature [arg] --sysroot : Build with sysroot"# ); ConfigInfo::show_usage(); @@ -142,6 +132,10 @@ pub fn build_sysroot(env: &HashMap, config: &ConfigInfo) -> Resu } let mut args: Vec<&dyn AsRef> = vec![&"cargo", &"build", &"--target", &config.target]; + for feature in &config.features { + args.push(&"--features"); + args.push(feature); + } if config.no_default_features { rustflags.push_str(" -Csymbol-mangling-version=v0"); diff --git a/build_system/src/clean.rs b/build_system/src/clean.rs index 55f55acf73e..768a78e789e 100644 --- a/build_system/src/clean.rs +++ b/build_system/src/clean.rs @@ -1,8 +1,8 @@ -use crate::utils::{get_sysroot_dir, remove_file, run_command}; - use std::fs::remove_dir_all; use std::path::Path; +use crate::utils::{get_sysroot_dir, remove_file, run_command}; + #[derive(Default)] enum CleanArg { /// `clean all` diff --git a/build_system/src/clone_gcc.rs b/build_system/src/clone_gcc.rs index aee46afaeb0..e28ee873eb6 100644 --- a/build_system/src/clone_gcc.rs +++ b/build_system/src/clone_gcc.rs @@ -1,8 +1,8 @@ +use std::path::{Path, PathBuf}; + use crate::config::ConfigInfo; use crate::utils::{git_clone, run_command_with_output}; -use std::path::{Path, PathBuf}; - fn show_usage() { println!( r#" @@ -34,7 +34,7 @@ fn new() -> Result, String> { "--out-path" => match args.next() { Some(path) if !path.is_empty() => out_path = Some(path), _ => { - return Err("Expected an argument after `--out-path`, found nothing".into()) + return Err("Expected an argument after `--out-path`, found nothing".into()); } }, "--help" => { diff --git a/build_system/src/config.rs b/build_system/src/config.rs index 965aedd8be8..e381617be06 100644 --- a/build_system/src/config.rs +++ b/build_system/src/config.rs @@ -1,14 +1,15 @@ +use std::collections::HashMap; +use std::ffi::OsStr; +use std::path::{Path, PathBuf}; +use std::{env as std_env, fs}; + +use boml::types::TomlValue; +use boml::Toml; + use crate::utils::{ create_dir, create_symlink, get_os_name, get_sysroot_dir, run_command_with_output, rustc_version_info, split_args, }; -use std::collections::HashMap; -use std::env as std_env; -use std::ffi::OsStr; -use std::fs; -use std::path::{Path, PathBuf}; - -use boml::{types::TomlValue, Toml}; #[derive(Default, PartialEq, Eq, Clone, Copy, Debug)] pub enum Channel { @@ -54,7 +55,7 @@ pub fn new(config_file: &Path) -> Result { config.gcc_path = Some(value.as_str().to_string()) } ("gcc-path", _) => { - return failed_config_parsing(config_file, "Expected a string for `gcc-path`") + return failed_config_parsing(config_file, "Expected a string for `gcc-path`"); } ("download-gccjit", TomlValue::Boolean(value)) => { config.download_gccjit = Some(*value) @@ -63,7 +64,7 @@ pub fn new(config_file: &Path) -> Result { return failed_config_parsing( config_file, "Expected a boolean for `download-gccjit`", - ) + ); } _ => return failed_config_parsing(config_file, &format!("Unknown key `{}`", key)), } @@ -73,7 +74,7 @@ pub fn new(config_file: &Path) -> Result { return failed_config_parsing( config_file, "At least one of `gcc-path` or `download-gccjit` value must be set", - ) + ); } (Some(_), Some(true)) => { println!( @@ -97,7 +98,7 @@ pub fn new(config_file: &Path) -> Result { } } -#[derive(Default, Debug)] +#[derive(Default, Debug, Clone)] pub struct ConfigInfo { pub target: String, pub target_triple: String, @@ -122,6 +123,7 @@ pub struct ConfigInfo { pub no_download: bool, pub no_default_features: bool, pub backend: Option, + pub features: Vec, } impl ConfigInfo { @@ -132,6 +134,13 @@ pub fn parse_argument( args: &mut impl Iterator, ) -> Result { match arg { + "--features" => { + if let Some(arg) = args.next() { + self.features.push(arg); + } else { + return Err("Expected a value after `--features`, found nothing".to_string()); + } + } "--target" => { if let Some(arg) = args.next() { self.target = arg; @@ -144,7 +153,7 @@ pub fn parse_argument( _ => { return Err( "Expected a value after `--target-triple`, found nothing".to_string() - ) + ); } }, "--out-dir" => match args.next() { @@ -158,7 +167,7 @@ pub fn parse_argument( self.config_file = Some(arg.to_string()); } _ => { - return Err("Expected a value after `--config-file`, found nothing".to_string()) + return Err("Expected a value after `--config-file`, found nothing".to_string()); } }, "--release-sysroot" => self.sysroot_release_channel = true, @@ -169,7 +178,7 @@ pub fn parse_argument( self.cg_gcc_path = Some(arg.into()); } _ => { - return Err("Expected a value after `--cg_gcc-path`, found nothing".to_string()) + return Err("Expected a value after `--cg_gcc-path`, found nothing".to_string()); } }, "--use-backend" => match args.next() { @@ -277,7 +286,7 @@ pub fn setup_gcc_path(&mut self) -> Result<(), String> { self.gcc_path = match gcc_path { Some(path) => path, None => { - return Err(format!("missing `gcc-path` value from `{}`", config_file.display(),)) + return Err(format!("missing `gcc-path` value from `{}`", config_file.display(),)); } }; Ok(()) @@ -442,6 +451,7 @@ pub fn setup( pub fn show_usage() { println!( "\ + --features [arg] : Add a new feature [arg] --target-triple [arg] : Set the target triple to [arg] --target [arg] : Set the target to [arg] --out-dir : Location where the files will be generated diff --git a/build_system/src/fmt.rs b/build_system/src/fmt.rs index 43644ba19b3..de310a6a30f 100644 --- a/build_system/src/fmt.rs +++ b/build_system/src/fmt.rs @@ -1,7 +1,8 @@ -use crate::utils::run_command_with_output; use std::ffi::OsStr; use std::path::Path; +use crate::utils::run_command_with_output; + fn show_usage() { println!( r#" diff --git a/build_system/src/main.rs b/build_system/src/main.rs index d678fd75344..3a860e2b136 100644 --- a/build_system/src/main.rs +++ b/build_system/src/main.rs @@ -1,5 +1,4 @@ -use std::env; -use std::process; +use std::{env, process}; mod build; mod clean; diff --git a/build_system/src/prepare.rs b/build_system/src/prepare.rs index 00aa632165e..d14639afee5 100644 --- a/build_system/src/prepare.rs +++ b/build_system/src/prepare.rs @@ -1,12 +1,12 @@ +use std::fs; +use std::path::{Path, PathBuf}; + use crate::rustc_info::get_rustc_path; use crate::utils::{ cargo_install, create_dir, get_sysroot_dir, git_clone_root_dir, remove_file, run_command, run_command_with_output, walk_dir, }; -use std::fs; -use std::path::{Path, PathBuf}; - fn prepare_libcore( sysroot_path: &Path, libgccjit12_patches: bool, diff --git a/build_system/src/rust_tools.rs b/build_system/src/rust_tools.rs index 242fa7ef949..105f5eebe24 100644 --- a/build_system/src/rust_tools.rs +++ b/build_system/src/rust_tools.rs @@ -1,13 +1,13 @@ +use std::collections::HashMap; +use std::ffi::OsStr; +use std::path::PathBuf; + use crate::config::ConfigInfo; use crate::utils::{ get_toolchain, run_command_with_output_and_env_no_err, rustc_toolchain_version_info, rustc_version_info, }; -use std::collections::HashMap; -use std::ffi::OsStr; -use std::path::PathBuf; - fn args(command: &str) -> Result>, String> { // We skip the binary and the "cargo"/"rustc" option. if let Some("--help") = std::env::args().skip(2).next().as_deref() { diff --git a/build_system/src/test.rs b/build_system/src/test.rs index 8d088a3aac3..dd09de24aa3 100644 --- a/build_system/src/test.rs +++ b/build_system/src/test.rs @@ -1,3 +1,10 @@ +use std::collections::HashMap; +use std::ffi::OsStr; +use std::fs::{remove_dir_all, File}; +use std::io::{BufRead, BufReader}; +use std::path::{Path, PathBuf}; +use std::str::FromStr; + use crate::build; use crate::config::{Channel, ConfigInfo}; use crate::utils::{ @@ -6,13 +13,6 @@ split_args, walk_dir, }; -use std::collections::HashMap; -use std::ffi::OsStr; -use std::fs::{remove_dir_all, File}; -use std::io::{BufRead, BufReader}; -use std::path::{Path, PathBuf}; -use std::str::FromStr; - type Env = HashMap; type Runner = fn(&Env, &TestArg) -> Result<(), String>; type Runners = HashMap<&'static str, (&'static str, Runner)>; @@ -92,6 +92,7 @@ struct TestArg { current_part: Option, sysroot_panic_abort: bool, config_info: ConfigInfo, + sysroot_features: Vec, } impl TestArg { @@ -109,7 +110,7 @@ fn new() -> Result, String> { test_arg.flags.extend_from_slice(&["--features".into(), feature]); } _ => { - return Err("Expected an argument after `--features`, found nothing".into()) + return Err("Expected an argument after `--features`, found nothing".into()); } }, "--use-system-gcc" => { @@ -127,6 +128,14 @@ fn new() -> Result, String> { "--sysroot-panic-abort" => { test_arg.sysroot_panic_abort = true; } + "--sysroot-features" => match args.next() { + Some(feature) if !feature.is_empty() => { + test_arg.sysroot_features.push(feature); + } + _ => { + return Err(format!("Expected an argument after `{}`, found nothing", arg)); + } + }, "--help" => { show_usage(); return Ok(None); @@ -250,7 +259,9 @@ fn mini_tests(env: &Env, args: &TestArg) -> Result<(), String> { fn build_sysroot(env: &Env, args: &TestArg) -> Result<(), String> { // FIXME: create a function "display_if_not_quiet" or something along the line. println!("[BUILD] sysroot"); - build::build_sysroot(env, &args.config_info)?; + let mut config = args.config_info.clone(); + config.features.extend(args.sysroot_features.iter().cloned()); + build::build_sysroot(env, &config)?; Ok(()) } @@ -458,11 +469,7 @@ fn setup_rustc(env: &mut Env, args: &TestArg) -> Result { .map_err(|error| format!("Failed to retrieve cargo path: {:?}", error)) .and_then(|cargo| { let cargo = cargo.trim().to_owned(); - if cargo.is_empty() { - Err(format!("`cargo` path is empty")) - } else { - Ok(cargo) - } + if cargo.is_empty() { Err(format!("`cargo` path is empty")) } else { Ok(cargo) } })?; let rustc = String::from_utf8( run_command_with_env(&[&"rustup", &toolchain, &"which", &"rustc"], rust_dir, Some(env))? @@ -471,11 +478,7 @@ fn setup_rustc(env: &mut Env, args: &TestArg) -> Result { .map_err(|error| format!("Failed to retrieve rustc path: {:?}", error)) .and_then(|rustc| { let rustc = rustc.trim().to_owned(); - if rustc.is_empty() { - Err(format!("`rustc` path is empty")) - } else { - Ok(rustc) - } + if rustc.is_empty() { Err(format!("`rustc` path is empty")) } else { Ok(rustc) } })?; let llvm_filecheck = match run_command_with_env( &[ @@ -634,7 +637,8 @@ fn test_projects(env: &Env, args: &TestArg) -> Result<(), String> { "https://github.com/BurntSushi/memchr", "https://github.com/dtolnay/itoa", "https://github.com/rust-lang/cfg-if", - "https://github.com/rust-lang-nursery/lazy-static.rs", + //"https://github.com/rust-lang-nursery/lazy-static.rs", // TODO: re-enable when the + //failing test is fixed upstream. //"https://github.com/marshallpierce/rust-base64", // FIXME: one test is OOM-killed. // TODO: ignore the base64 test that is OOM-killed. "https://github.com/time-rs/time", diff --git a/build_system/src/utils.rs b/build_system/src/utils.rs index 3bba8df6c65..e338d1b4992 100644 --- a/build_system/src/utils.rs +++ b/build_system/src/utils.rs @@ -175,11 +175,7 @@ pub fn cargo_install(to_install: &str) -> Result<(), String> { pub fn get_os_name() -> Result { let output = run_command(&[&"uname"], None)?; let name = std::str::from_utf8(&output.stdout).unwrap_or("").trim().to_string(); - if !name.is_empty() { - Ok(name) - } else { - Err("Failed to retrieve the OS name".to_string()) - } + if !name.is_empty() { Ok(name) } else { Err("Failed to retrieve the OS name".to_string()) } } #[derive(Default, PartialEq)] diff --git a/doc/debugging-gcc-lto.md b/doc/debugging-gcc-lto.md deleted file mode 100644 index 93b150d7686..00000000000 --- a/doc/debugging-gcc-lto.md +++ /dev/null @@ -1,3 +0,0 @@ -# How to debug GCC LTO - -Run do the command with `-v -save-temps` and then extract the `lto1` line from the output and run that under the debugger. diff --git a/doc/debugging.md b/doc/debugging.md new file mode 100644 index 00000000000..6ff4edf8877 --- /dev/null +++ b/doc/debugging.md @@ -0,0 +1,38 @@ +# Debugging + +## How to debug GCC LTO + +Run do the command with `-v -save-temps` and then extract the `lto1` line from the output and run that under the debugger. + +## How to debug stdarch tests that cannot be ran locally + +First, run the tests normally: + +---- +cd build/build_sysroot/sysroot_src/library/stdarch/ +STDARCH_TEST_EVERYTHING=1 CARGO_TARGET_X86_64_UNKNOWN_LINUX_GNU_RUNNER="sde -future -rtm_mode full --" TARGET=x86_64-unknown-linux-gnu ../../../../../y.sh cargo test +---- + +It will show the command it ran, something like this: + +---- + process didn't exit successfully: `sde -future -rtm_mode full -- /home/user/projects/rustc_codegen_gcc/build/build_sysroot/sysroot_src/library/stdarch/target/debug/deps/core_arch-fd2d75f89baae5c6` (signal: 11, SIGSEGV: invalid memory reference) +---- + +Then add the `-debug` flag to it: + +---- +sde -debug -future -rtm_mode full -- /home/user/projects/rustc_codegen_gcc/build/build_sysroot/sysroot_src/library/stdarch/target/debug/deps/core_arch-fd2d75f89baae5c6 +---- + +To see the symbols in `gdb`, specify the executable in your command: + +---- +gdb /home/user/projects/rustc_codegen_gcc/build/build_sysroot/sysroot_src/library/stdarch/target/debug/deps/core_arch-fd2d75f89baae5c6 +---- + +and then write the `gdb` command that `sde` tells you to use, something like: + +---- +target remote :51299 +---- diff --git a/example/mini_core.rs b/example/mini_core.rs index a48c0a4450c..f47bfdad131 100644 --- a/example/mini_core.rs +++ b/example/mini_core.rs @@ -1,5 +1,5 @@ #![feature( - no_core, lang_items, intrinsics, unboxed_closures, type_ascription, extern_types, + no_core, lang_items, intrinsics, unboxed_closures, extern_types, decl_macro, rustc_attrs, transparent_unions, auto_traits, freeze_impls, thread_local )] diff --git a/example/mini_core_hello_world.rs b/example/mini_core_hello_world.rs index 5a7ddc4cd7f..9f096e90220 100644 --- a/example/mini_core_hello_world.rs +++ b/example/mini_core_hello_world.rs @@ -430,6 +430,7 @@ pub enum E2 { V4, } +#[allow(unreachable_patterns)] fn check_niche_behavior () { if let E1::V2 { .. } = (E1::V1 { f: true }) { intrinsics::abort(); diff --git a/example/std_example.rs b/example/std_example.rs index 8ab8fcc525e..9e43b4635f0 100644 --- a/example/std_example.rs +++ b/example/std_example.rs @@ -1,5 +1,5 @@ #![allow(internal_features)] -#![feature(core_intrinsics, coroutines, coroutine_trait, is_sorted, stmt_expr_attributes)] +#![feature(core_intrinsics, coroutines, coroutine_trait, stmt_expr_attributes)] #[cfg(feature="master")] #[cfg(target_arch="x86_64")] diff --git a/libgccjit.version b/libgccjit.version index 23ca7f02215..b9bbbd324c3 100644 --- a/libgccjit.version +++ b/libgccjit.version @@ -1 +1 @@ -341be3b7d7ac6976cfed8ed59da3573c040d0776 +e744a9459d33864067214741daf5c5bc2a7b88c6 diff --git a/patches/0022-core-Disable-not-compiling-tests.patch b/patches/0022-core-Disable-not-compiling-tests.patch index ea1a5a8d355..b2ab05691ec 100644 --- a/patches/0022-core-Disable-not-compiling-tests.patch +++ b/patches/0022-core-Disable-not-compiling-tests.patch @@ -1,26 +1,24 @@ -From f6befc4bb51d84f5f1cf35938a168c953d421350 Mon Sep 17 00:00:00 2001 -From: bjorn3 -Date: Sun, 24 Nov 2019 15:10:23 +0100 +From 18793c6109890493ceb3ff36549849a36e3d8022 Mon Sep 17 00:00:00 2001 +From: None +Date: Sun, 1 Sep 2024 11:42:17 -0400 Subject: [PATCH] [core] Disable not compiling tests --- - library/core/tests/Cargo.toml | 8 ++++++++ - library/core/tests/num/flt2dec/mod.rs | 1 - - library/core/tests/num/int_macros.rs | 2 ++ - library/core/tests/num/uint_macros.rs | 2 ++ - library/core/tests/ptr.rs | 2 ++ - library/core/tests/slice.rs | 2 ++ - 6 files changed, 16 insertions(+), 1 deletion(-) + library/core/tests/Cargo.toml | 14 ++++++++++++++ + library/core/tests/lib.rs | 1 + + 2 files changed, 15 insertions(+) create mode 100644 library/core/tests/Cargo.toml diff --git a/library/core/tests/Cargo.toml b/library/core/tests/Cargo.toml new file mode 100644 -index 0000000..46fd999 +index 0000000..ca326ac --- /dev/null +++ b/library/core/tests/Cargo.toml -@@ -0,0 +1,12 @@ +@@ -0,0 +1,14 @@ ++[workspace] ++ +[package] -+name = "core" ++name = "coretests" +version = "0.0.0" +edition = "2021" + @@ -32,11 +30,14 @@ index 0000000..46fd999 +rand = { version = "0.8.5", default-features = false } +rand_xorshift = { version = "0.3.0", default-features = false } diff --git a/library/core/tests/lib.rs b/library/core/tests/lib.rs -index 42a26ae..5ac1042 100644 +index 1e336bf..5800ebb 100644 --- a/library/core/tests/lib.rs +++ b/library/core/tests/lib.rs -@@ -1,3 +1,4 @@ +@@ -1,4 +1,5 @@ + // tidy-alphabetical-start +#![cfg(test)] - #![feature(alloc_layout_extra)] - #![feature(array_chunks)] - #![feature(array_ptr_get)] + #![cfg_attr(bootstrap, feature(offset_of_nested))] + #![cfg_attr(target_has_atomic = "128", feature(integer_atomics))] + #![cfg_attr(test, feature(cfg_match))] +-- +2.46.0 diff --git a/patches/libgccjit12/0001-core-Disable-portable-simd-test.patch b/patches/libgccjit12/0001-core-Disable-portable-simd-test.patch index c060300f44f..da664d51b63 100644 --- a/patches/libgccjit12/0001-core-Disable-portable-simd-test.patch +++ b/patches/libgccjit12/0001-core-Disable-portable-simd-test.patch @@ -11,15 +11,15 @@ diff --git a/library/core/tests/lib.rs b/library/core/tests/lib.rs index b71786c..cf484d5 100644 --- a/library/core/tests/lib.rs +++ b/library/core/tests/lib.rs -@@ -95,7 +95,6 @@ - #![feature(never_type)] - #![feature(unwrap_infallible)] +@@ -87,7 +87,6 @@ + #![feature(numfmt)] + #![feature(pattern)] #![feature(pointer_is_aligned_to)] -#![feature(portable_simd)] #![feature(ptr_metadata)] - #![feature(unsized_tuple_coercion)] - #![feature(const_option)] -@@ -157,7 +156,6 @@ mod pin; + #![feature(slice_from_ptr_range)] + #![feature(slice_internals)] +@@ -155,7 +154,6 @@ mod pin; mod pin_macro; mod ptr; mod result; diff --git a/rust-toolchain b/rust-toolchain index 3c83f4b4608..dca3b0c22e4 100644 --- a/rust-toolchain +++ b/rust-toolchain @@ -1,3 +1,3 @@ [toolchain] -channel = "nightly-2024-07-02" +channel = "nightly-2024-08-11" components = ["rust-src", "rustc-dev", "llvm-tools-preview"] diff --git a/src/abi.rs b/src/abi.rs index 166dd080cf2..0a99e7213be 100644 --- a/src/abi.rs +++ b/src/abi.rs @@ -26,11 +26,7 @@ fn get_param(&mut self, index: usize) -> Self::Value { } else { false }; - if on_stack { - param.to_lvalue().get_address(None) - } else { - param.to_rvalue() - } + if on_stack { param.to_lvalue().get_address(None) } else { param.to_rvalue() } } } diff --git a/src/archive.rs b/src/archive.rs index 73ff0c37b66..0cee05f1cea 100644 --- a/src/archive.rs +++ b/src/archive.rs @@ -1,27 +1,24 @@ -use std::path::{Path, PathBuf}; +use std::path::Path; use rustc_codegen_ssa::back::archive::{ - get_native_object_symbols, ArArchiveBuilder, ArchiveBuilder, ArchiveBuilderBuilder, + ArArchiveBuilder, ArchiveBuilder, ArchiveBuilderBuilder, DEFAULT_OBJECT_READER, }; use rustc_session::Session; -use rustc_session::cstore::DllImport; - pub(crate) struct ArArchiveBuilderBuilder; impl ArchiveBuilderBuilder for ArArchiveBuilderBuilder { fn new_archive_builder<'a>(&self, sess: &'a Session) -> Box { - Box::new(ArArchiveBuilder::new(sess, get_native_object_symbols)) + Box::new(ArArchiveBuilder::new(sess, &DEFAULT_OBJECT_READER)) } fn create_dll_import_lib( &self, _sess: &Session, _lib_name: &str, - _dll_imports: &[DllImport], - _tmpdir: &Path, - _is_direct_dependency: bool, - ) -> PathBuf { + _import_name_and_ordinal_vector: Vec<(String, Option)>, + _output_path: &Path, + ) { unimplemented!("creating dll imports is not yet supported"); } } diff --git a/src/asm.rs b/src/asm.rs index aa485846cd4..7c135289958 100644 --- a/src/asm.rs +++ b/src/asm.rs @@ -1,3 +1,5 @@ +use std::borrow::Cow; + use gccjit::{LValue, RValue, ToRValue, Type}; use rustc_ast::ast::{InlineAsmOptions, InlineAsmTemplatePiece}; use rustc_codegen_ssa::mir::operand::OperandValue; @@ -6,13 +8,11 @@ AsmBuilderMethods, AsmMethods, BaseTypeMethods, BuilderMethods, GlobalAsmOperandRef, InlineAsmOperandRef, }; - -use rustc_middle::{bug, ty::Instance}; +use rustc_middle::bug; +use rustc_middle::ty::Instance; use rustc_span::Span; use rustc_target::asm::*; -use std::borrow::Cow; - use crate::builder::Builder; use crate::callee::get_fn; use crate::context::CodegenCx; @@ -858,11 +858,7 @@ fn modifier_to_gcc( InlineAsmRegClass::AArch64(AArch64InlineAsmRegClass::reg) => modifier, InlineAsmRegClass::AArch64(AArch64InlineAsmRegClass::vreg) | InlineAsmRegClass::AArch64(AArch64InlineAsmRegClass::vreg_low16) => { - if modifier == Some('v') { - None - } else { - modifier - } + if modifier == Some('v') { None } else { modifier } } InlineAsmRegClass::AArch64(AArch64InlineAsmRegClass::preg) => { unreachable!("clobber-only") diff --git a/src/attributes.rs b/src/attributes.rs index 27f21107eda..5fdf2680aac 100644 --- a/src/attributes.rs +++ b/src/attributes.rs @@ -9,8 +9,9 @@ use rustc_middle::ty; use rustc_span::symbol::sym; +use crate::context::CodegenCx; +use crate::errors::TiedTargetFeatures; use crate::gcc_util::{check_tied_features, to_gcc_features}; -use crate::{context::CodegenCx, errors::TiedTargetFeatures}; /// Get GCC attribute for the provided inline heuristic. #[cfg(feature = "master")] @@ -74,7 +75,7 @@ pub fn from_fn_attrs<'gcc, 'tcx>( let function_features = codegen_fn_attrs .target_features .iter() - .map(|features| features.as_str()) + .map(|features| features.name.as_str()) .collect::>(); if let Some(features) = check_tied_features( diff --git a/src/base.rs b/src/base.rs index e3558b6b875..2eaab3ed00c 100644 --- a/src/base.rs +++ b/src/base.rs @@ -19,8 +19,7 @@ use crate::builder::Builder; use crate::context::CodegenCx; -use crate::{gcc_util, new_context, LockedTargetInfo}; -use crate::{GccContext, SyncContext}; +use crate::{gcc_util, new_context, GccContext, LockedTargetInfo, SyncContext}; #[cfg(feature = "master")] pub fn visibility_to_gcc(linkage: Visibility) -> gccjit::Visibility { diff --git a/src/builder.rs b/src/builder.rs index 38e147599c9..9936bc1f5f2 100644 --- a/src/builder.rs +++ b/src/builder.rs @@ -28,9 +28,8 @@ use rustc_middle::ty::{Instance, ParamEnv, Ty, TyCtxt}; use rustc_span::def_id::DefId; use rustc_span::Span; -use rustc_target::abi::{ - self, call::FnAbi, Align, HasDataLayout, Size, TargetDataLayout, WrappingRange, -}; +use rustc_target::abi::call::FnAbi; +use rustc_target::abi::{self, Align, HasDataLayout, Size, TargetDataLayout, WrappingRange}; use rustc_target::spec::{HasTargetSpec, HasWasmCAbiOpt, Target, WasmCAbi}; use crate::common::{type_is_pointer, SignType, TypeReflection}; @@ -142,7 +141,7 @@ fn compare_exchange( ) -> RValue<'gcc> { let size = get_maybe_pointer_size(src); let compare_exchange = - self.context.get_builtin_function(&format!("__atomic_compare_exchange_{}", size)); + self.context.get_builtin_function(format!("__atomic_compare_exchange_{}", size)); let order = self.context.new_rvalue_from_int(self.i32_type, order.to_gcc()); let failure_order = self.context.new_rvalue_from_int(self.i32_type, failure_order.to_gcc()); let weak = self.context.new_rvalue_from_int(self.bool_type, weak as i32); @@ -271,10 +270,12 @@ fn check_ptr_call<'b>( actual_val.dereference(self.location).to_rvalue() } } else { + // FIXME: this condition seems wrong: it will pass when both types are not + // a vector. assert!( (!expected_ty.is_vector() || actual_ty.is_vector()) && (expected_ty.is_vector() || !actual_ty.is_vector()), - "{:?} ({}) -> {:?} ({}), index: {:?}[{}]", + "{:?} (is vector: {}) -> {:?} (is vector: {}), Function: {:?}[{}]", actual_ty, actual_ty.is_vector(), expected_ty, @@ -284,6 +285,8 @@ fn check_ptr_call<'b>( ); // TODO(antoyo): perhaps use __builtin_convertvector for vector casting. // TODO: remove bitcast now that vector types can be compared? + // ==> We use bitcast to avoid having to do many manual casts from e.g. __m256i to __v32qi (in + // the case of _mm256_aesenc_epi128). self.bitcast(actual_val, expected_ty) } } else { @@ -329,7 +332,7 @@ fn function_call( let result = current_func.new_local( self.location, return_type, - &format!("returnValue{}", self.next_value_counter()), + format!("returnValue{}", self.next_value_counter()), ); self.block.add_assignment( self.location, @@ -367,6 +370,7 @@ fn function_ptr_call( let args = { let function_address_names = self.function_address_names.borrow(); let original_function_name = function_address_names.get(&func_ptr); + func_ptr = llvm::adjust_function(self.context, &func_name, func_ptr, args); llvm::adjust_intrinsic_arguments( self, gcc_func, @@ -397,7 +401,7 @@ fn function_ptr_call( let result = current_func.new_local( self.location, return_value.get_type(), - &format!("ptrReturnValue{}", self.next_value_counter()), + format!("ptrReturnValue{}", self.next_value_counter()), ); self.block.add_assignment(self.location, result, return_value); result.to_rvalue() @@ -439,7 +443,7 @@ pub fn overflow_call( let result = current_func.new_local( self.location, return_type, - &format!("overflowReturnValue{}", self.next_value_counter()), + format!("overflowReturnValue{}", self.next_value_counter()), ); self.block.add_assignment( self.location, @@ -924,7 +928,7 @@ fn alloca(&mut self, size: Size, align: Align) -> RValue<'gcc> { let ty = self.cx.type_array(self.cx.type_i8(), size.bytes()).get_aligned(align.bytes()); // TODO(antoyo): It might be better to return a LValue, but fixing the rustc API is non-trivial. self.current_func() - .new_local(self.location, ty, &format!("stack_var_{}", self.next_value_counter())) + .new_local(self.location, ty, format!("stack_var_{}", self.next_value_counter())) .get_address(self.location) } @@ -950,7 +954,7 @@ fn load(&mut self, pointee_ty: Type<'gcc>, ptr: RValue<'gcc>, align: Align) -> R let loaded_value = function.new_local( self.location, aligned_type, - &format!("loadedValue{}", self.next_value_counter()), + format!("loadedValue{}", self.next_value_counter()), ); block.add_assignment(self.location, loaded_value, deref); loaded_value.to_rvalue() @@ -971,7 +975,7 @@ fn atomic_load( // TODO(antoyo): use ty. // TODO(antoyo): handle alignment. let atomic_load = - self.context.get_builtin_function(&format!("__atomic_load_{}", size.bytes())); + self.context.get_builtin_function(format!("__atomic_load_{}", size.bytes())); let ordering = self.context.new_rvalue_from_int(self.i32_type, order.to_gcc()); let volatile_const_void_ptr_type = @@ -1030,11 +1034,7 @@ fn scalar_load_metadata<'a, 'gcc, 'tcx>( let llty = place.layout.scalar_pair_element_gcc_type(self, i); let load = self.load(llty, llptr, align); scalar_load_metadata(self, load, scalar); - if scalar.is_bool() { - self.trunc(load, self.type_i1()) - } else { - load - } + if scalar.is_bool() { self.trunc(load, self.type_i1()) } else { load } }; OperandValue::Pair( @@ -1131,7 +1131,7 @@ fn atomic_store( ) { // TODO(antoyo): handle alignment. let atomic_store = - self.context.get_builtin_function(&format!("__atomic_store_{}", size.bytes())); + self.context.get_builtin_function(format!("__atomic_store_{}", size.bytes())); let ordering = self.context.new_rvalue_from_int(self.i32_type, order.to_gcc()); let volatile_const_void_ptr_type = self.context.new_type::<()>().make_volatile().make_pointer(); @@ -1782,18 +1782,10 @@ fn fptoint_sat( // This already happens today with u128::MAX = 2^128 - 1 > f32::MAX. let int_max = |signed: bool, int_width: u64| -> u128 { let shift_amount = 128 - int_width; - if signed { - i128::MAX as u128 >> shift_amount - } else { - u128::MAX >> shift_amount - } + if signed { i128::MAX as u128 >> shift_amount } else { u128::MAX >> shift_amount } }; let int_min = |signed: bool, int_width: u64| -> i128 { - if signed { - i128::MIN >> (128 - int_width) - } else { - 0 - } + if signed { i128::MIN >> (128 - int_width) } else { 0 } }; let compute_clamp_bounds_single = |signed: bool, int_width: u64| -> (u128, u128) { diff --git a/src/common.rs b/src/common.rs index 35699346b29..7a456e1c5d6 100644 --- a/src/common.rs +++ b/src/common.rs @@ -1,5 +1,4 @@ -use gccjit::LValue; -use gccjit::{RValue, ToRValue, Type}; +use gccjit::{LValue, RValue, ToRValue, Type}; use rustc_codegen_ssa::traits::{BaseTypeMethods, ConstMethods, MiscMethods, StaticMethods}; use rustc_middle::mir::interpret::{ConstAllocation, GlobalAlloc, Scalar}; use rustc_middle::mir::Mutability; @@ -58,11 +57,7 @@ pub fn type_is_pointer(typ: Type<'_>) -> bool { impl<'gcc, 'tcx> ConstMethods<'tcx> for CodegenCx<'gcc, 'tcx> { fn const_null(&self, typ: Type<'gcc>) -> RValue<'gcc> { - if type_is_pointer(typ) { - self.context.new_null(typ) - } else { - self.const_int(typ, 0) - } + if type_is_pointer(typ) { self.context.new_null(typ) } else { self.const_int(typ, 0) } } fn const_undef(&self, typ: Type<'gcc>) -> RValue<'gcc> { @@ -221,7 +216,7 @@ fn scalar_to_backend(&self, cv: Scalar, layout: abi::Scalar, ty: Type<'gcc>) -> } value } - GlobalAlloc::Function(fn_instance) => self.get_fn_addr(fn_instance), + GlobalAlloc::Function { instance, .. } => self.get_fn_addr(instance), GlobalAlloc::VTable(ty, trait_ref) => { let alloc = self .tcx diff --git a/src/consts.rs b/src/consts.rs index ba7e08e33ef..483b2355c52 100644 --- a/src/consts.rs +++ b/src/consts.rs @@ -3,14 +3,14 @@ use gccjit::{Function, GlobalKind, LValue, RValue, ToRValue, Type}; use rustc_codegen_ssa::traits::{BaseTypeMethods, ConstMethods, StaticMethods}; use rustc_hir::def::DefKind; -use rustc_middle::bug; use rustc_middle::middle::codegen_fn_attrs::{CodegenFnAttrFlags, CodegenFnAttrs}; use rustc_middle::mir::interpret::{ self, read_target_uint, ConstAllocation, ErrorHandled, Scalar as InterpScalar, }; -use rustc_middle::span_bug; +use rustc_middle::mir::mono::Linkage; use rustc_middle::ty::layout::LayoutOf; use rustc_middle::ty::{self, Instance}; +use rustc_middle::{bug, span_bug}; use rustc_span::def_id::DefId; use rustc_target::abi::{self, Align, HasDataLayout, Primitive, Size, WrappingRange}; @@ -257,7 +257,7 @@ pub(crate) fn get_static_inner(&self, def_id: DefId, gcc_type: Type<'gcc>) -> LV if !self.tcx.is_reachable_non_generic(def_id) { #[cfg(feature = "master")] - global.add_string_attribute(VarAttribute::Visibility(Visibility::Hidden)); + global.add_attribute(VarAttribute::Visibility(Visibility::Hidden)); } global @@ -385,6 +385,11 @@ fn check_and_apply_linkage<'gcc, 'tcx>( let global1 = cx.declare_global_with_linkage(sym, cx.type_i8(), base::global_linkage_to_gcc(linkage)); + if linkage == Linkage::ExternalWeak { + #[cfg(feature = "master")] + global1.add_attribute(VarAttribute::Weak); + } + // Declare an internal global `extern_with_linkage_foo` which // is initialized with the address of `foo`. If `foo` is // discarded during linking (for example, if `foo` has weak diff --git a/src/context.rs b/src/context.rs index 4458ca84bbb..9756f0faae2 100644 --- a/src/context.rs +++ b/src/context.rs @@ -6,8 +6,7 @@ use rustc_codegen_ssa::base::wants_msvc_seh; use rustc_codegen_ssa::errors as ssa_errors; use rustc_codegen_ssa::traits::{BackendTypes, BaseTypeMethods, MiscMethods}; -use rustc_data_structures::base_n::ToBaseN; -use rustc_data_structures::base_n::ALPHANUMERIC_ONLY; +use rustc_data_structures::base_n::{ToBaseN, ALPHANUMERIC_ONLY}; use rustc_data_structures::fx::{FxHashMap, FxHashSet}; use rustc_middle::mir::mono::CodegenUnit; use rustc_middle::span_bug; @@ -17,15 +16,16 @@ }; use rustc_middle::ty::{self, Instance, ParamEnv, PolyExistentialTraitRef, Ty, TyCtxt}; use rustc_session::Session; -use rustc_span::{source_map::respan, Span}; -use rustc_target::abi::{ - call::FnAbi, HasDataLayout, PointeeInfo, Size, TargetDataLayout, VariantIdx, -}; +use rustc_span::source_map::respan; +use rustc_span::{Span, DUMMY_SP}; +use rustc_target::abi::call::FnAbi; +use rustc_target::abi::{HasDataLayout, PointeeInfo, Size, TargetDataLayout, VariantIdx}; use rustc_target::spec::{HasTargetSpec, HasWasmCAbiOpt, Target, TlsModel, WasmCAbi}; use crate::callee::get_fn; use crate::common::SignType; +#[cfg_attr(not(feature = "master"), allow(dead_code))] pub struct CodegenCx<'gcc, 'tcx> { pub codegen_unit: &'tcx CodegenUnit<'tcx>, pub context: &'gcc Context<'gcc>, @@ -228,48 +228,14 @@ pub fn new( "__builtin_umul_overflow", "__builtin_usubll_overflow", "__builtin_usub_overflow", - "sqrtf", - "sqrt", "__builtin_powif", "__builtin_powi", - "sinf", - "sin", - "cosf", - "cos", - "powf", - "pow", - "expf", - "exp", - "exp2f", - "exp2", - "logf", - "log", - "log10f", - "log10", - "log2f", - "log2", - "fmaf", - "fma", "fabsf", "fabs", - "fminf", - "fmin", - "fmaxf", - "fmax", "copysignf", "copysign", - "floorf", - "floor", - "ceilf", - "ceil", - "truncf", - "trunc", - "rintf", - "rint", "nearbyintf", "nearbyint", - "roundf", - "round", ]; for builtin in builtins.iter() { @@ -494,6 +460,7 @@ fn eh_personality(&self) -> RValue<'gcc> { ty::ParamEnv::reveal_all(), def_id, ty::List::empty(), + DUMMY_SP, ); let symbol_name = tcx.symbol_name(instance).name; diff --git a/src/debuginfo.rs b/src/debuginfo.rs index 87638642883..c2adca34748 100644 --- a/src/debuginfo.rs +++ b/src/debuginfo.rs @@ -1,3 +1,5 @@ +use std::ops::Range; + use gccjit::{Location, RValue}; use rustc_codegen_ssa::mir::debuginfo::{DebugScope, FunctionDebugContext, VariableKind}; use rustc_codegen_ssa::traits::{DebugInfoBuilderMethods, DebugInfoMethods}; @@ -10,7 +12,6 @@ use rustc_span::{BytePos, Pos, SourceFile, SourceFileAndLine, Span, Symbol}; use rustc_target::abi::call::FnAbi; use rustc_target::abi::Size; -use std::ops::Range; use crate::builder::Builder; use crate::context::CodegenCx; diff --git a/src/declare.rs b/src/declare.rs index a2b158ee0a7..aa7709488e6 100644 --- a/src/declare.rs +++ b/src/declare.rs @@ -168,7 +168,15 @@ fn declare_raw_fn<'gcc>( variadic: bool, ) -> Function<'gcc> { if name.starts_with("llvm.") { - let intrinsic = llvm::intrinsic(name, cx); + let intrinsic = match name { + "llvm.fma.f16" => { + // fma is not a target builtin, but a normal builtin, so we handle it differently + // here. + cx.context.get_builtin_function("fma") + } + _ => llvm::intrinsic(name, cx), + }; + cx.intrinsics.borrow_mut().insert(name.to_string(), intrinsic); return intrinsic; } diff --git a/src/gcc_util.rs b/src/gcc_util.rs index 53877e8ff7f..5308ccdb614 100644 --- a/src/gcc_util.rs +++ b/src/gcc_util.rs @@ -1,11 +1,10 @@ #[cfg(feature = "master")] use gccjit::Context; -use smallvec::{smallvec, SmallVec}; - use rustc_data_structures::fx::FxHashMap; use rustc_middle::bug; use rustc_session::Session; use rustc_target::target_features::RUSTC_SPECIFIC_FEATURES; +use smallvec::{smallvec, SmallVec}; use crate::errors::{ PossibleFeature, TargetFeatureDisableOrEnable, UnknownCTargetFeature, @@ -66,8 +65,8 @@ pub(crate) fn global_gcc_features(sess: &Session, diagnostics: bool) -> Vec Builder<'a, 'gcc, 'tcx> { pub fn gcc_urem(&self, a: RValue<'gcc>, b: RValue<'gcc>) -> RValue<'gcc> { @@ -266,7 +261,9 @@ pub fn gcc_checked_binop( lhs: ::Value, rhs: ::Value, ) -> (::Value, ::Value) { - use rustc_middle::ty::{Int, IntTy::*, Uint, UintTy::*}; + use rustc_middle::ty::IntTy::*; + use rustc_middle::ty::UintTy::*; + use rustc_middle::ty::{Int, Uint}; let new_kind = match *typ.kind() { Int(t @ Isize) => Int(t.normalize(self.tcx.sess.target.pointer_width)), @@ -738,7 +735,7 @@ pub fn gcc_bswap(&mut self, mut arg: RValue<'gcc>, width: u64) -> RValue<'gcc> { // TODO(antoyo): check if it's faster to use string literals and a // match instead of format!. - let bswap = self.cx.context.get_builtin_function(&format!("__builtin_bswap{}", width)); + let bswap = self.cx.context.get_builtin_function(format!("__builtin_bswap{}", width)); // FIXME(antoyo): this cast should not be necessary. Remove // when having proper sized integer types. let param_type = bswap.get_param(0).to_rvalue().get_type(); diff --git a/src/intrinsic/archs.rs b/src/intrinsic/archs.rs index f7500933789..b8d1cde1d5d 100644 --- a/src/intrinsic/archs.rs +++ b/src/intrinsic/archs.rs @@ -31,8 +31,11 @@ "llvm.AMDGPU.trig.preop.v2f64" => "__builtin_amdgpu_trig_preop", "llvm.AMDGPU.trig.preop.v4f32" => "__builtin_amdgpu_trig_preop", // aarch64 + "llvm.aarch64.chkfeat" => "__builtin_arm_chkfeat", "llvm.aarch64.dmb" => "__builtin_arm_dmb", "llvm.aarch64.dsb" => "__builtin_arm_dsb", + "llvm.aarch64.gcspopm" => "__builtin_arm_gcspopm", + "llvm.aarch64.gcsss" => "__builtin_arm_gcsss", "llvm.aarch64.isb" => "__builtin_arm_isb", "llvm.aarch64.prefetch" => "__builtin_arm_prefetch", "llvm.aarch64.sve.aesd" => "__builtin_sve_svaesd_u8", @@ -80,7 +83,6 @@ "llvm.amdgcn.dot4.f32.fp8.fp8" => "__builtin_amdgcn_dot4_f32_fp8_fp8", "llvm.amdgcn.ds.add.gs.reg.rtn" => "__builtin_amdgcn_ds_add_gs_reg_rtn", "llvm.amdgcn.ds.bpermute" => "__builtin_amdgcn_ds_bpermute", - "llvm.amdgcn.ds.fadd.v2bf16" => "__builtin_amdgcn_ds_atomic_fadd_v2bf16", "llvm.amdgcn.ds.gws.barrier" => "__builtin_amdgcn_ds_gws_barrier", "llvm.amdgcn.ds.gws.init" => "__builtin_amdgcn_ds_gws_init", "llvm.amdgcn.ds.gws.sema.br" => "__builtin_amdgcn_ds_gws_sema_br", @@ -96,6 +98,7 @@ "llvm.amdgcn.fdot2.f16.f16" => "__builtin_amdgcn_fdot2_f16_f16", "llvm.amdgcn.fdot2.f32.bf16" => "__builtin_amdgcn_fdot2_f32_bf16", "llvm.amdgcn.fmul.legacy" => "__builtin_amdgcn_fmul_legacy", + "llvm.amdgcn.global.load.lds" => "__builtin_amdgcn_global_load_lds", "llvm.amdgcn.groupstaticsize" => "__builtin_amdgcn_groupstaticsize", "llvm.amdgcn.iglp.opt" => "__builtin_amdgcn_iglp_opt", "llvm.amdgcn.implicit.buffer.ptr" => "__builtin_amdgcn_implicit_buffer_ptr", @@ -154,16 +157,11 @@ "llvm.amdgcn.mqsad.u32.u8" => "__builtin_amdgcn_mqsad_u32_u8", "llvm.amdgcn.msad.u8" => "__builtin_amdgcn_msad_u8", "llvm.amdgcn.perm" => "__builtin_amdgcn_perm", - "llvm.amdgcn.permlane16" => "__builtin_amdgcn_permlane16", "llvm.amdgcn.permlane16.var" => "__builtin_amdgcn_permlane16_var", - "llvm.amdgcn.permlane64" => "__builtin_amdgcn_permlane64", - "llvm.amdgcn.permlanex16" => "__builtin_amdgcn_permlanex16", "llvm.amdgcn.permlanex16.var" => "__builtin_amdgcn_permlanex16_var", "llvm.amdgcn.qsad.pk.u16.u8" => "__builtin_amdgcn_qsad_pk_u16_u8", "llvm.amdgcn.queue.ptr" => "__builtin_amdgcn_queue_ptr", "llvm.amdgcn.rcp.legacy" => "__builtin_amdgcn_rcp_legacy", - "llvm.amdgcn.readfirstlane" => "__builtin_amdgcn_readfirstlane", - "llvm.amdgcn.readlane" => "__builtin_amdgcn_readlane", "llvm.amdgcn.rsq.legacy" => "__builtin_amdgcn_rsq_legacy", "llvm.amdgcn.s.barrier" => "__builtin_amdgcn_s_barrier", "llvm.amdgcn.s.barrier.init" => "__builtin_amdgcn_s_barrier_init", @@ -192,6 +190,8 @@ "llvm.amdgcn.s.setreg" => "__builtin_amdgcn_s_setreg", "llvm.amdgcn.s.sleep" => "__builtin_amdgcn_s_sleep", "llvm.amdgcn.s.sleep.var" => "__builtin_amdgcn_s_sleep_var", + "llvm.amdgcn.s.ttracedata" => "__builtin_amdgcn_s_ttracedata", + "llvm.amdgcn.s.ttracedata.imm" => "__builtin_amdgcn_s_ttracedata_imm", "llvm.amdgcn.s.wait.event.export.ready" => "__builtin_amdgcn_s_wait_event_export_ready", "llvm.amdgcn.s.waitcnt" => "__builtin_amdgcn_s_waitcnt", "llvm.amdgcn.s.wakeup.barrier" => "__builtin_amdgcn_s_wakeup_barrier", @@ -227,7 +227,6 @@ "llvm.amdgcn.workgroup.id.x" => "__builtin_amdgcn_workgroup_id_x", "llvm.amdgcn.workgroup.id.y" => "__builtin_amdgcn_workgroup_id_y", "llvm.amdgcn.workgroup.id.z" => "__builtin_amdgcn_workgroup_id_z", - "llvm.amdgcn.writelane" => "__builtin_amdgcn_writelane", // arm "llvm.arm.cdp" => "__builtin_arm_cdp", "llvm.arm.cdp2" => "__builtin_arm_cdp2", @@ -4536,10 +4535,18 @@ "llvm.nvvm.div.rz.d" => "__nvvm_div_rz_d", "llvm.nvvm.div.rz.f" => "__nvvm_div_rz_f", "llvm.nvvm.div.rz.ftz.f" => "__nvvm_div_rz_ftz_f", + "llvm.nvvm.e4m3x2.to.f16x2.rn" => "__nvvm_e4m3x2_to_f16x2_rn", + "llvm.nvvm.e4m3x2.to.f16x2.rn.relu" => "__nvvm_e4m3x2_to_f16x2_rn_relu", + "llvm.nvvm.e5m2x2.to.f16x2.rn" => "__nvvm_e5m2x2_to_f16x2_rn", + "llvm.nvvm.e5m2x2.to.f16x2.rn.relu" => "__nvvm_e5m2x2_to_f16x2_rn_relu", "llvm.nvvm.ex2.approx.d" => "__nvvm_ex2_approx_d", "llvm.nvvm.ex2.approx.f" => "__nvvm_ex2_approx_f", "llvm.nvvm.ex2.approx.ftz.f" => "__nvvm_ex2_approx_ftz_f", "llvm.nvvm.exit" => "__nvvm_exit", + "llvm.nvvm.f16x2.to.e4m3x2.rn" => "__nvvm_f16x2_to_e4m3x2_rn", + "llvm.nvvm.f16x2.to.e4m3x2.rn.relu" => "__nvvm_f16x2_to_e4m3x2_rn_relu", + "llvm.nvvm.f16x2.to.e5m2x2.rn" => "__nvvm_f16x2_to_e5m2x2_rn", + "llvm.nvvm.f16x2.to.e5m2x2.rn.relu" => "__nvvm_f16x2_to_e5m2x2_rn_relu", "llvm.nvvm.f2bf16.rn" => "__nvvm_f2bf16_rn", "llvm.nvvm.f2bf16.rn.relu" => "__nvvm_f2bf16_rn_relu", "llvm.nvvm.f2bf16.rz" => "__nvvm_f2bf16_rz", @@ -4582,6 +4589,10 @@ "llvm.nvvm.fabs.d" => "__nvvm_fabs_d", "llvm.nvvm.fabs.f" => "__nvvm_fabs_f", "llvm.nvvm.fabs.ftz.f" => "__nvvm_fabs_ftz_f", + "llvm.nvvm.ff.to.e4m3x2.rn" => "__nvvm_ff_to_e4m3x2_rn", + "llvm.nvvm.ff.to.e4m3x2.rn.relu" => "__nvvm_ff_to_e4m3x2_rn_relu", + "llvm.nvvm.ff.to.e5m2x2.rn" => "__nvvm_ff_to_e5m2x2_rn", + "llvm.nvvm.ff.to.e5m2x2.rn.relu" => "__nvvm_ff_to_e5m2x2_rn_relu", "llvm.nvvm.ff2bf16x2.rn" => "__nvvm_ff2bf16x2_rn", "llvm.nvvm.ff2bf16x2.rn.relu" => "__nvvm_ff2bf16x2_rn_relu", "llvm.nvvm.ff2bf16x2.rz" => "__nvvm_ff2bf16x2_rz", @@ -4866,6 +4877,7 @@ "llvm.nvvm.round.ftz.f" => "__nvvm_round_ftz_f", "llvm.nvvm.rsqrt.approx.d" => "__nvvm_rsqrt_approx_d", "llvm.nvvm.rsqrt.approx.f" => "__nvvm_rsqrt_approx_f", + "llvm.nvvm.rsqrt.approx.ftz.d" => "__nvvm_rsqrt_approx_ftz_d", "llvm.nvvm.rsqrt.approx.ftz.f" => "__nvvm_rsqrt_approx_ftz_f", "llvm.nvvm.sad.i" => "__nvvm_sad_i", "llvm.nvvm.sad.ll" => "__nvvm_sad_ll", @@ -5164,6 +5176,8 @@ // ppc "llvm.ppc.addex" => "__builtin_ppc_addex", "llvm.ppc.addf128.round.to.odd" => "__builtin_addf128_round_to_odd", + "llvm.ppc.addg6s" => "__builtin_addg6s", + "llvm.ppc.addg6sd" => "__builtin_ppc_addg6s", "llvm.ppc.altivec.crypto.vcipher" => "__builtin_altivec_crypto_vcipher", "llvm.ppc.altivec.crypto.vcipherlast" => "__builtin_altivec_crypto_vcipherlast", "llvm.ppc.altivec.crypto.vncipher" => "__builtin_altivec_crypto_vncipher", @@ -5461,6 +5475,10 @@ "llvm.ppc.bcdsub" => "__builtin_ppc_bcdsub", "llvm.ppc.bcdsub.p" => "__builtin_ppc_bcdsub_p", "llvm.ppc.bpermd" => "__builtin_bpermd", + "llvm.ppc.cbcdtd" => "__builtin_cbcdtd", + "llvm.ppc.cbcdtdd" => "__builtin_ppc_cbcdtd", + "llvm.ppc.cdtbcd" => "__builtin_cdtbcd", + "llvm.ppc.cdtbcdd" => "__builtin_ppc_cdtbcd", "llvm.ppc.cfuged" => "__builtin_cfuged", "llvm.ppc.cmpeqb" => "__builtin_ppc_cmpeqb", "llvm.ppc.cmprb" => "__builtin_ppc_cmprb", @@ -5627,7 +5645,6 @@ "llvm.ppc.qpx.qvstfs" => "__builtin_qpx_qvstfs", "llvm.ppc.qpx.qvstfsa" => "__builtin_qpx_qvstfsa", "llvm.ppc.readflm" => "__builtin_readflm", - "llvm.ppc.rldimi" => "__builtin_ppc_rldimi", "llvm.ppc.rlwimi" => "__builtin_ppc_rlwimi", "llvm.ppc.rlwnm" => "__builtin_ppc_rlwnm", "llvm.ppc.scalar.extract.expq" => "__builtin_vsx_scalar_extract_expq", @@ -7210,29 +7227,6 @@ "llvm.ve.vl.xorm.MMM" => "__builtin_ve_vl_xorm_MMM", "llvm.ve.vl.xorm.mmm" => "__builtin_ve_vl_xorm_mmm", // x86 - "llvm.x86.3dnow.pavgusb" => "__builtin_ia32_pavgusb", - "llvm.x86.3dnow.pf2id" => "__builtin_ia32_pf2id", - "llvm.x86.3dnow.pfacc" => "__builtin_ia32_pfacc", - "llvm.x86.3dnow.pfadd" => "__builtin_ia32_pfadd", - "llvm.x86.3dnow.pfcmpeq" => "__builtin_ia32_pfcmpeq", - "llvm.x86.3dnow.pfcmpge" => "__builtin_ia32_pfcmpge", - "llvm.x86.3dnow.pfcmpgt" => "__builtin_ia32_pfcmpgt", - "llvm.x86.3dnow.pfmax" => "__builtin_ia32_pfmax", - "llvm.x86.3dnow.pfmin" => "__builtin_ia32_pfmin", - "llvm.x86.3dnow.pfmul" => "__builtin_ia32_pfmul", - "llvm.x86.3dnow.pfrcp" => "__builtin_ia32_pfrcp", - "llvm.x86.3dnow.pfrcpit1" => "__builtin_ia32_pfrcpit1", - "llvm.x86.3dnow.pfrcpit2" => "__builtin_ia32_pfrcpit2", - "llvm.x86.3dnow.pfrsqit1" => "__builtin_ia32_pfrsqit1", - "llvm.x86.3dnow.pfrsqrt" => "__builtin_ia32_pfrsqrt", - "llvm.x86.3dnow.pfsub" => "__builtin_ia32_pfsub", - "llvm.x86.3dnow.pfsubr" => "__builtin_ia32_pfsubr", - "llvm.x86.3dnow.pi2fd" => "__builtin_ia32_pi2fd", - "llvm.x86.3dnow.pmulhrw" => "__builtin_ia32_pmulhrw", - "llvm.x86.3dnowa.pf2iw" => "__builtin_ia32_pf2iw", - "llvm.x86.3dnowa.pfnacc" => "__builtin_ia32_pfnacc", - "llvm.x86.3dnowa.pfpnacc" => "__builtin_ia32_pfpnacc", - "llvm.x86.3dnowa.pi2fw" => "__builtin_ia32_pi2fw", "llvm.x86.aadd32" => "__builtin_ia32_aadd32", "llvm.x86.aadd64" => "__builtin_ia32_aadd64", "llvm.x86.aand32" => "__builtin_ia32_aand32", @@ -7334,6 +7328,207 @@ "llvm.x86.avx.vtestz.ps.256" => "__builtin_ia32_vtestzps256", "llvm.x86.avx.vzeroall" => "__builtin_ia32_vzeroall", "llvm.x86.avx.vzeroupper" => "__builtin_ia32_vzeroupper", + "llvm.x86.avx10.mask.vcvt2ps2phx.128" => "__builtin_ia32_vcvt2ps2phx128_mask", + "llvm.x86.avx10.mask.vcvt2ps2phx.256" => "__builtin_ia32_vcvt2ps2phx256_mask", + "llvm.x86.avx10.mask.vcvt2ps2phx.512" => "__builtin_ia32_vcvt2ps2phx512_mask", + "llvm.x86.avx10.mask.vcvtbiasph2bf8128" => "__builtin_ia32_vcvtbiasph2bf8_128_mask", + "llvm.x86.avx10.mask.vcvtbiasph2bf8256" => "__builtin_ia32_vcvtbiasph2bf8_256_mask", + "llvm.x86.avx10.mask.vcvtbiasph2bf8512" => "__builtin_ia32_vcvtbiasph2bf8_512_mask", + "llvm.x86.avx10.mask.vcvtbiasph2bf8s128" => "__builtin_ia32_vcvtbiasph2bf8s_128_mask", + "llvm.x86.avx10.mask.vcvtbiasph2bf8s256" => "__builtin_ia32_vcvtbiasph2bf8s_256_mask", + "llvm.x86.avx10.mask.vcvtbiasph2bf8s512" => "__builtin_ia32_vcvtbiasph2bf8s_512_mask", + "llvm.x86.avx10.mask.vcvtbiasph2hf8128" => "__builtin_ia32_vcvtbiasph2hf8_128_mask", + "llvm.x86.avx10.mask.vcvtbiasph2hf8256" => "__builtin_ia32_vcvtbiasph2hf8_256_mask", + "llvm.x86.avx10.mask.vcvtbiasph2hf8512" => "__builtin_ia32_vcvtbiasph2hf8_512_mask", + "llvm.x86.avx10.mask.vcvtbiasph2hf8s128" => "__builtin_ia32_vcvtbiasph2hf8s_128_mask", + "llvm.x86.avx10.mask.vcvtbiasph2hf8s256" => "__builtin_ia32_vcvtbiasph2hf8s_256_mask", + "llvm.x86.avx10.mask.vcvtbiasph2hf8s512" => "__builtin_ia32_vcvtbiasph2hf8s_512_mask", + "llvm.x86.avx10.mask.vcvthf82ph128" => "__builtin_ia32_vcvthf8_2ph128_mask", + "llvm.x86.avx10.mask.vcvthf82ph256" => "__builtin_ia32_vcvthf8_2ph256_mask", + "llvm.x86.avx10.mask.vcvthf82ph512" => "__builtin_ia32_vcvthf8_2ph512_mask", + "llvm.x86.avx10.mask.vcvtneph2bf8128" => "__builtin_ia32_vcvtneph2bf8_128_mask", + "llvm.x86.avx10.mask.vcvtneph2bf8256" => "__builtin_ia32_vcvtneph2bf8_256_mask", + "llvm.x86.avx10.mask.vcvtneph2bf8512" => "__builtin_ia32_vcvtneph2bf8_512_mask", + "llvm.x86.avx10.mask.vcvtneph2bf8s128" => "__builtin_ia32_vcvtneph2bf8s_128_mask", + "llvm.x86.avx10.mask.vcvtneph2bf8s256" => "__builtin_ia32_vcvtneph2bf8s_256_mask", + "llvm.x86.avx10.mask.vcvtneph2bf8s512" => "__builtin_ia32_vcvtneph2bf8s_512_mask", + "llvm.x86.avx10.mask.vcvtneph2hf8128" => "__builtin_ia32_vcvtneph2hf8_128_mask", + "llvm.x86.avx10.mask.vcvtneph2hf8256" => "__builtin_ia32_vcvtneph2hf8_256_mask", + "llvm.x86.avx10.mask.vcvtneph2hf8512" => "__builtin_ia32_vcvtneph2hf8_512_mask", + "llvm.x86.avx10.mask.vcvtneph2hf8s128" => "__builtin_ia32_vcvtneph2hf8s_128_mask", + "llvm.x86.avx10.mask.vcvtneph2hf8s256" => "__builtin_ia32_vcvtneph2hf8s_256_mask", + "llvm.x86.avx10.mask.vcvtneph2hf8s512" => "__builtin_ia32_vcvtneph2hf8s_512_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvtpd2dq256" => "__builtin_ia32_vcvtpd2dq256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvtpd2ph256" => "__builtin_ia32_vcvtpd2ph256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvtpd2ps256" => "__builtin_ia32_vcvtpd2ps256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvtpd2qq256" => "__builtin_ia32_vcvtpd2qq256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvtpd2udq256" => "__builtin_ia32_vcvtpd2udq256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvtpd2uqq256" => "__builtin_ia32_vcvtpd2uqq256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvtph2dq256" => "__builtin_ia32_vcvtph2dq256_round_mask", + "llvm.x86.avx10.mask.vcvtph2ibs128" => "__builtin_ia32_vcvtph2ibs128_mask", + "llvm.x86.avx10.mask.vcvtph2ibs256" => "__builtin_ia32_vcvtph2ibs256_mask", + "llvm.x86.avx10.mask.vcvtph2ibs512" => "__builtin_ia32_vcvtph2ibs512_mask", + "llvm.x86.avx10.mask.vcvtph2iubs128" => "__builtin_ia32_vcvtph2iubs128_mask", + "llvm.x86.avx10.mask.vcvtph2iubs256" => "__builtin_ia32_vcvtph2iubs256_mask", + "llvm.x86.avx10.mask.vcvtph2iubs512" => "__builtin_ia32_vcvtph2iubs512_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvtph2pd256" => "__builtin_ia32_vcvtph2pd256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvtph2psx256" => "__builtin_ia32_vcvtph2psx256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvtph2qq256" => "__builtin_ia32_vcvtph2qq256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvtph2udq256" => "__builtin_ia32_vcvtph2udq256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvtph2uqq256" => "__builtin_ia32_vcvtph2uqq256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvtph2uw256" => "__builtin_ia32_vcvtph2uw256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvtph2w256" => "__builtin_ia32_vcvtph2w256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvtps2dq256" => "__builtin_ia32_vcvtps2dq256_round_mask", + "llvm.x86.avx10.mask.vcvtps2ibs128" => "__builtin_ia32_vcvtps2ibs128_mask", + "llvm.x86.avx10.mask.vcvtps2ibs256" => "__builtin_ia32_vcvtps2ibs256_mask", + "llvm.x86.avx10.mask.vcvtps2ibs512" => "__builtin_ia32_vcvtps2ibs512_mask", + "llvm.x86.avx10.mask.vcvtps2iubs128" => "__builtin_ia32_vcvtps2iubs128_mask", + "llvm.x86.avx10.mask.vcvtps2iubs256" => "__builtin_ia32_vcvtps2iubs256_mask", + "llvm.x86.avx10.mask.vcvtps2iubs512" => "__builtin_ia32_vcvtps2iubs512_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvtps2pd256" => "__builtin_ia32_vcvtps2pd256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvtps2ph256" => "__builtin_ia32_vcvtps2ph256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvtps2phx256" => "__builtin_ia32_vcvtps2phx256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvtps2qq256" => "__builtin_ia32_vcvtps2qq256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvtps2udq256" => "__builtin_ia32_vcvtps2udq256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvtps2uqq256" => "__builtin_ia32_vcvtps2uqq256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvttpd2dq256" => "__builtin_ia32_vcvttpd2dq256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvttpd2qq256" => "__builtin_ia32_vcvttpd2qq256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvttpd2udq256" => "__builtin_ia32_vcvttpd2udq256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvttpd2uqq256" => "__builtin_ia32_vcvttpd2uqq256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvttph2dq256" => "__builtin_ia32_vcvttph2dq256_round_mask", + "llvm.x86.avx10.mask.vcvttph2ibs128" => "__builtin_ia32_vcvttph2ibs128_mask", + "llvm.x86.avx10.mask.vcvttph2ibs256" => "__builtin_ia32_vcvttph2ibs256_mask", + "llvm.x86.avx10.mask.vcvttph2ibs512" => "__builtin_ia32_vcvttph2ibs512_mask", + "llvm.x86.avx10.mask.vcvttph2iubs128" => "__builtin_ia32_vcvttph2iubs128_mask", + "llvm.x86.avx10.mask.vcvttph2iubs256" => "__builtin_ia32_vcvttph2iubs256_mask", + "llvm.x86.avx10.mask.vcvttph2iubs512" => "__builtin_ia32_vcvttph2iubs512_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvttph2qq256" => "__builtin_ia32_vcvttph2qq256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvttph2udq256" => "__builtin_ia32_vcvttph2udq256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvttph2uqq256" => "__builtin_ia32_vcvttph2uqq256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvttph2uw256" => "__builtin_ia32_vcvttph2uw256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvttph2w256" => "__builtin_ia32_vcvttph2w256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvttps2dq256" => "__builtin_ia32_vcvttps2dq256_round_mask", + "llvm.x86.avx10.mask.vcvttps2ibs128" => "__builtin_ia32_vcvttps2ibs128_mask", + "llvm.x86.avx10.mask.vcvttps2ibs256" => "__builtin_ia32_vcvttps2ibs256_mask", + "llvm.x86.avx10.mask.vcvttps2ibs512" => "__builtin_ia32_vcvttps2ibs512_mask", + "llvm.x86.avx10.mask.vcvttps2iubs128" => "__builtin_ia32_vcvttps2iubs128_mask", + "llvm.x86.avx10.mask.vcvttps2iubs256" => "__builtin_ia32_vcvttps2iubs256_mask", + "llvm.x86.avx10.mask.vcvttps2iubs512" => "__builtin_ia32_vcvttps2iubs512_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvttps2qq256" => "__builtin_ia32_vcvttps2qq256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvttps2udq256" => "__builtin_ia32_vcvttps2udq256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vcvttps2uqq256" => "__builtin_ia32_vcvttps2uqq256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vfcmaddcph256" => "__builtin_ia32_vfcmaddcph256_round_mask3", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vfcmulcph256" => "__builtin_ia32_vfcmulcph256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vfixupimmpd256" => "__builtin_ia32_vfixupimmpd256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vfixupimmps256" => "__builtin_ia32_vfixupimmps256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vfmaddcph256" => "__builtin_ia32_vfmaddcph256_round_mask3", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vfmulcph256" => "__builtin_ia32_vfmulcph256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vgetexppd256" => "__builtin_ia32_vgetexppd256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vgetexpph256" => "__builtin_ia32_vgetexpph256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vgetexpps256" => "__builtin_ia32_vgetexpps256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vgetmantpd256" => "__builtin_ia32_vgetmantpd256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vgetmantph256" => "__builtin_ia32_vgetmantph256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vgetmantps256" => "__builtin_ia32_vgetmantps256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vminmaxpd.round" => "__builtin_ia32_vminmaxpd512_round_mask", + "llvm.x86.avx10.mask.vminmaxpd128" => "__builtin_ia32_vminmaxpd128_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vminmaxpd256.round" => "__builtin_ia32_vminmaxpd256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vminmaxph.round" => "__builtin_ia32_vminmaxph512_round_mask", + "llvm.x86.avx10.mask.vminmaxph128" => "__builtin_ia32_vminmaxph128_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vminmaxph256.round" => "__builtin_ia32_vminmaxph256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vminmaxps.round" => "__builtin_ia32_vminmaxps512_round_mask", + "llvm.x86.avx10.mask.vminmaxps128" => "__builtin_ia32_vminmaxps128_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vminmaxps256.round" => "__builtin_ia32_vminmaxps256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vminmaxsd.round" => "__builtin_ia32_vminmaxsd_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vminmaxsh.round" => "__builtin_ia32_vminmaxsh_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vminmaxss.round" => "__builtin_ia32_vminmaxss_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vrangepd256" => "__builtin_ia32_vrangepd256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vrangeps256" => "__builtin_ia32_vrangeps256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vreducepd256" => "__builtin_ia32_vreducepd256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vreduceph256" => "__builtin_ia32_vreduceph256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vreduceps256" => "__builtin_ia32_vreduceps256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vrndscalepd256" => "__builtin_ia32_vrndscalepd256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vrndscaleph256" => "__builtin_ia32_vrndscaleph256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vrndscaleps256" => "__builtin_ia32_vrndscaleps256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vscalefpd256" => "__builtin_ia32_vscalefpd256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vscalefph256" => "__builtin_ia32_vscalefph256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.mask.vscalefps256" => "__builtin_ia32_vscalefps256_round_mask", + // [INVALID CONVERSION]: "llvm.x86.avx10.maskz.vfcmaddcph256" => "__builtin_ia32_vfcmaddcph256_round_maskz", + // [INVALID CONVERSION]: "llvm.x86.avx10.maskz.vfixupimmpd256" => "__builtin_ia32_vfixupimmpd256_round_maskz", + // [INVALID CONVERSION]: "llvm.x86.avx10.maskz.vfixupimmps256" => "__builtin_ia32_vfixupimmps256_round_maskz", + // [INVALID CONVERSION]: "llvm.x86.avx10.maskz.vfmaddcph256" => "__builtin_ia32_vfmaddcph256_round_maskz", + "llvm.x86.avx10.vaddpd256" => "__builtin_ia32_vaddpd256_round", + "llvm.x86.avx10.vaddph256" => "__builtin_ia32_vaddph256_round", + "llvm.x86.avx10.vaddps256" => "__builtin_ia32_vaddps256_round", + "llvm.x86.avx10.vcvtne2ph2bf8128" => "__builtin_ia32_vcvtne2ph2bf8_128", + "llvm.x86.avx10.vcvtne2ph2bf8256" => "__builtin_ia32_vcvtne2ph2bf8_256", + "llvm.x86.avx10.vcvtne2ph2bf8512" => "__builtin_ia32_vcvtne2ph2bf8_512", + "llvm.x86.avx10.vcvtne2ph2bf8s128" => "__builtin_ia32_vcvtne2ph2bf8s_128", + "llvm.x86.avx10.vcvtne2ph2bf8s256" => "__builtin_ia32_vcvtne2ph2bf8s_256", + "llvm.x86.avx10.vcvtne2ph2bf8s512" => "__builtin_ia32_vcvtne2ph2bf8s_512", + "llvm.x86.avx10.vcvtne2ph2hf8128" => "__builtin_ia32_vcvtne2ph2hf8_128", + "llvm.x86.avx10.vcvtne2ph2hf8256" => "__builtin_ia32_vcvtne2ph2hf8_256", + "llvm.x86.avx10.vcvtne2ph2hf8512" => "__builtin_ia32_vcvtne2ph2hf8_512", + "llvm.x86.avx10.vcvtne2ph2hf8s128" => "__builtin_ia32_vcvtne2ph2hf8s_128", + "llvm.x86.avx10.vcvtne2ph2hf8s256" => "__builtin_ia32_vcvtne2ph2hf8s_256", + "llvm.x86.avx10.vcvtne2ph2hf8s512" => "__builtin_ia32_vcvtne2ph2hf8s_512", + "llvm.x86.avx10.vcvtnebf162ibs128" => "__builtin_ia32_vcvtnebf162ibs128", + "llvm.x86.avx10.vcvtnebf162ibs256" => "__builtin_ia32_vcvtnebf162ibs256", + "llvm.x86.avx10.vcvtnebf162ibs512" => "__builtin_ia32_vcvtnebf162ibs512", + "llvm.x86.avx10.vcvtnebf162iubs128" => "__builtin_ia32_vcvtnebf162iubs128", + "llvm.x86.avx10.vcvtnebf162iubs256" => "__builtin_ia32_vcvtnebf162iubs256", + "llvm.x86.avx10.vcvtnebf162iubs512" => "__builtin_ia32_vcvtnebf162iubs512", + "llvm.x86.avx10.vcvttnebf162ibs128" => "__builtin_ia32_vcvttnebf162ibs128", + "llvm.x86.avx10.vcvttnebf162ibs256" => "__builtin_ia32_vcvttnebf162ibs256", + "llvm.x86.avx10.vcvttnebf162ibs512" => "__builtin_ia32_vcvttnebf162ibs512", + "llvm.x86.avx10.vcvttnebf162iubs128" => "__builtin_ia32_vcvttnebf162iubs128", + "llvm.x86.avx10.vcvttnebf162iubs256" => "__builtin_ia32_vcvttnebf162iubs256", + "llvm.x86.avx10.vcvttnebf162iubs512" => "__builtin_ia32_vcvttnebf162iubs512", + "llvm.x86.avx10.vdivpd256" => "__builtin_ia32_vdivpd256_round", + "llvm.x86.avx10.vdivph256" => "__builtin_ia32_vdivph256_round", + "llvm.x86.avx10.vdivps256" => "__builtin_ia32_vdivps256_round", + "llvm.x86.avx10.vdpphps.128" => "__builtin_ia32_vdpphps128", + "llvm.x86.avx10.vdpphps.256" => "__builtin_ia32_vdpphps256", + "llvm.x86.avx10.vdpphps.512" => "__builtin_ia32_vdpphps512", + "llvm.x86.avx10.vfmaddsubpd256" => "__builtin_ia32_vfmaddsubpd256_round", + "llvm.x86.avx10.vfmaddsubph256" => "__builtin_ia32_vfmaddsubph256_round", + "llvm.x86.avx10.vfmaddsubps256" => "__builtin_ia32_vfmaddsubps256_round", + "llvm.x86.avx10.vmaxpd256" => "__builtin_ia32_vmaxpd256_round", + "llvm.x86.avx10.vmaxph256" => "__builtin_ia32_vmaxph256_round", + "llvm.x86.avx10.vmaxps256" => "__builtin_ia32_vmaxps256_round", + "llvm.x86.avx10.vminmaxnepbf16128" => "__builtin_ia32_vminmaxnepbf16128", + "llvm.x86.avx10.vminmaxnepbf16256" => "__builtin_ia32_vminmaxnepbf16256", + "llvm.x86.avx10.vminmaxnepbf16512" => "__builtin_ia32_vminmaxnepbf16512", + "llvm.x86.avx10.vminmaxpd128" => "__builtin_ia32_vminmaxpd128", + "llvm.x86.avx10.vminmaxpd256" => "__builtin_ia32_vminmaxpd256", + "llvm.x86.avx10.vminmaxph128" => "__builtin_ia32_vminmaxph128", + "llvm.x86.avx10.vminmaxph256" => "__builtin_ia32_vminmaxph256", + "llvm.x86.avx10.vminmaxps128" => "__builtin_ia32_vminmaxps128", + "llvm.x86.avx10.vminmaxps256" => "__builtin_ia32_vminmaxps256", + "llvm.x86.avx10.vminpd256" => "__builtin_ia32_vminpd256_round", + "llvm.x86.avx10.vminph256" => "__builtin_ia32_vminph256_round", + "llvm.x86.avx10.vminps256" => "__builtin_ia32_vminps256_round", + "llvm.x86.avx10.vmpsadbw.512" => "__builtin_ia32_mpsadbw512", + "llvm.x86.avx10.vmulpd256" => "__builtin_ia32_vmulpd256_round", + "llvm.x86.avx10.vmulph256" => "__builtin_ia32_vmulph256_round", + "llvm.x86.avx10.vmulps256" => "__builtin_ia32_vmulps256_round", + "llvm.x86.avx10.vpdpbssd.512" => "__builtin_ia32_vpdpbssd512", + "llvm.x86.avx10.vpdpbssds.512" => "__builtin_ia32_vpdpbssds512", + "llvm.x86.avx10.vpdpbsud.512" => "__builtin_ia32_vpdpbsud512", + "llvm.x86.avx10.vpdpbsuds.512" => "__builtin_ia32_vpdpbsuds512", + "llvm.x86.avx10.vpdpbuud.512" => "__builtin_ia32_vpdpbuud512", + "llvm.x86.avx10.vpdpbuuds.512" => "__builtin_ia32_vpdpbuuds512", + "llvm.x86.avx10.vpdpwsud.512" => "__builtin_ia32_vpdpwsud512", + "llvm.x86.avx10.vpdpwsuds.512" => "__builtin_ia32_vpdpwsuds512", + "llvm.x86.avx10.vpdpwusd.512" => "__builtin_ia32_vpdpwusd512", + "llvm.x86.avx10.vpdpwusds.512" => "__builtin_ia32_vpdpwusds512", + "llvm.x86.avx10.vpdpwuud.512" => "__builtin_ia32_vpdpwuud512", + "llvm.x86.avx10.vpdpwuuds.512" => "__builtin_ia32_vpdpwuuds512", + "llvm.x86.avx10.vsqrtpd256" => "__builtin_ia32_vsqrtpd256_round", + "llvm.x86.avx10.vsqrtph256" => "__builtin_ia32_vsqrtph256_round", + "llvm.x86.avx10.vsqrtps256" => "__builtin_ia32_vsqrtps256_round", + "llvm.x86.avx10.vsubpd256" => "__builtin_ia32_vsubpd256_round", + "llvm.x86.avx10.vsubph256" => "__builtin_ia32_vsubph256_round", + "llvm.x86.avx10.vsubps256" => "__builtin_ia32_vsubps256_round", "llvm.x86.avx2.gather.d.d" => "__builtin_ia32_gatherd_d", "llvm.x86.avx2.gather.d.d.256" => "__builtin_ia32_gatherd_d256", "llvm.x86.avx2.gather.d.pd" => "__builtin_ia32_gatherd_pd", @@ -8738,10 +8933,10 @@ "llvm.x86.avx512.rcp14.ss" => "__builtin_ia32_rcp14ss_mask", "llvm.x86.avx512.rcp28.pd" => "__builtin_ia32_rcp28pd_mask", "llvm.x86.avx512.rcp28.ps" => "__builtin_ia32_rcp28ps_mask", - // [INVALID CONVERSION]: "llvm.x86.avx512.rcp28.sd" => "__builtin_ia32_rcp28sd_round_mask", - // [DUPLICATE]: "llvm.x86.avx512.rcp28.sd" => "__builtin_ia32_rcp28sd_mask", - // [INVALID CONVERSION]: "llvm.x86.avx512.rcp28.ss" => "__builtin_ia32_rcp28ss_round_mask", - // [DUPLICATE]: "llvm.x86.avx512.rcp28.ss" => "__builtin_ia32_rcp28ss_mask", + "llvm.x86.avx512.rcp28.sd" => "__builtin_ia32_rcp28sd_mask", + // [DUPLICATE]: "llvm.x86.avx512.rcp28.sd" => "__builtin_ia32_rcp28sd_round_mask", + "llvm.x86.avx512.rcp28.ss" => "__builtin_ia32_rcp28ss_mask", + // [DUPLICATE]: "llvm.x86.avx512.rcp28.ss" => "__builtin_ia32_rcp28ss_round_mask", "llvm.x86.avx512.rndscale.sd" => "__builtin_ia32_rndscalesd", "llvm.x86.avx512.rndscale.ss" => "__builtin_ia32_rndscaless", "llvm.x86.avx512.rsqrt14.pd.128" => "__builtin_ia32_rsqrt14pd128_mask", @@ -8754,10 +8949,10 @@ "llvm.x86.avx512.rsqrt14.ss" => "__builtin_ia32_rsqrt14ss_mask", "llvm.x86.avx512.rsqrt28.pd" => "__builtin_ia32_rsqrt28pd_mask", "llvm.x86.avx512.rsqrt28.ps" => "__builtin_ia32_rsqrt28ps_mask", - // [INVALID CONVERSION]: "llvm.x86.avx512.rsqrt28.sd" => "__builtin_ia32_rsqrt28sd_round_mask", - // [DUPLICATE]: "llvm.x86.avx512.rsqrt28.sd" => "__builtin_ia32_rsqrt28sd_mask", - // [INVALID CONVERSION]: "llvm.x86.avx512.rsqrt28.ss" => "__builtin_ia32_rsqrt28ss_round_mask", - // [DUPLICATE]: "llvm.x86.avx512.rsqrt28.ss" => "__builtin_ia32_rsqrt28ss_mask", + "llvm.x86.avx512.rsqrt28.sd" => "__builtin_ia32_rsqrt28sd_mask", + // [DUPLICATE]: "llvm.x86.avx512.rsqrt28.sd" => "__builtin_ia32_rsqrt28sd_round_mask", + "llvm.x86.avx512.rsqrt28.ss" => "__builtin_ia32_rsqrt28ss_mask", + // [DUPLICATE]: "llvm.x86.avx512.rsqrt28.ss" => "__builtin_ia32_rsqrt28ss_round_mask", "llvm.x86.avx512.scatter.dpd.512" => "__builtin_ia32_scattersiv8df", "llvm.x86.avx512.scatter.dpi.512" => "__builtin_ia32_scattersiv16si", "llvm.x86.avx512.scatter.dpq.512" => "__builtin_ia32_scattersiv8di", @@ -9082,75 +9277,6 @@ "llvm.x86.lwpval64" => "__builtin_ia32_lwpval64", "llvm.x86.mmx.emms" => "__builtin_ia32_emms", "llvm.x86.mmx.femms" => "__builtin_ia32_femms", - "llvm.x86.mmx.maskmovq" => "__builtin_ia32_maskmovq", - "llvm.x86.mmx.movnt.dq" => "__builtin_ia32_movntq", - "llvm.x86.mmx.packssdw" => "__builtin_ia32_packssdw", - "llvm.x86.mmx.packsswb" => "__builtin_ia32_packsswb", - "llvm.x86.mmx.packuswb" => "__builtin_ia32_packuswb", - "llvm.x86.mmx.padd.b" => "__builtin_ia32_paddb", - "llvm.x86.mmx.padd.d" => "__builtin_ia32_paddd", - "llvm.x86.mmx.padd.q" => "__builtin_ia32_paddq", - "llvm.x86.mmx.padd.w" => "__builtin_ia32_paddw", - "llvm.x86.mmx.padds.b" => "__builtin_ia32_paddsb", - "llvm.x86.mmx.padds.w" => "__builtin_ia32_paddsw", - "llvm.x86.mmx.paddus.b" => "__builtin_ia32_paddusb", - "llvm.x86.mmx.paddus.w" => "__builtin_ia32_paddusw", - "llvm.x86.mmx.palignr.b" => "__builtin_ia32_palignr", - "llvm.x86.mmx.pand" => "__builtin_ia32_pand", - "llvm.x86.mmx.pandn" => "__builtin_ia32_pandn", - "llvm.x86.mmx.pavg.b" => "__builtin_ia32_pavgb", - "llvm.x86.mmx.pavg.w" => "__builtin_ia32_pavgw", - "llvm.x86.mmx.pcmpeq.b" => "__builtin_ia32_pcmpeqb", - "llvm.x86.mmx.pcmpeq.d" => "__builtin_ia32_pcmpeqd", - "llvm.x86.mmx.pcmpeq.w" => "__builtin_ia32_pcmpeqw", - "llvm.x86.mmx.pcmpgt.b" => "__builtin_ia32_pcmpgtb", - "llvm.x86.mmx.pcmpgt.d" => "__builtin_ia32_pcmpgtd", - "llvm.x86.mmx.pcmpgt.w" => "__builtin_ia32_pcmpgtw", - "llvm.x86.mmx.pextr.w" => "__builtin_ia32_vec_ext_v4hi", - "llvm.x86.mmx.pinsr.w" => "__builtin_ia32_vec_set_v4hi", - "llvm.x86.mmx.pmadd.wd" => "__builtin_ia32_pmaddwd", - "llvm.x86.mmx.pmaxs.w" => "__builtin_ia32_pmaxsw", - "llvm.x86.mmx.pmaxu.b" => "__builtin_ia32_pmaxub", - "llvm.x86.mmx.pmins.w" => "__builtin_ia32_pminsw", - "llvm.x86.mmx.pminu.b" => "__builtin_ia32_pminub", - "llvm.x86.mmx.pmovmskb" => "__builtin_ia32_pmovmskb", - "llvm.x86.mmx.pmulh.w" => "__builtin_ia32_pmulhw", - "llvm.x86.mmx.pmulhu.w" => "__builtin_ia32_pmulhuw", - "llvm.x86.mmx.pmull.w" => "__builtin_ia32_pmullw", - "llvm.x86.mmx.pmulu.dq" => "__builtin_ia32_pmuludq", - "llvm.x86.mmx.por" => "__builtin_ia32_por", - "llvm.x86.mmx.psad.bw" => "__builtin_ia32_psadbw", - "llvm.x86.mmx.psll.d" => "__builtin_ia32_pslld", - "llvm.x86.mmx.psll.q" => "__builtin_ia32_psllq", - "llvm.x86.mmx.psll.w" => "__builtin_ia32_psllw", - "llvm.x86.mmx.pslli.d" => "__builtin_ia32_pslldi", - "llvm.x86.mmx.pslli.q" => "__builtin_ia32_psllqi", - "llvm.x86.mmx.pslli.w" => "__builtin_ia32_psllwi", - "llvm.x86.mmx.psra.d" => "__builtin_ia32_psrad", - "llvm.x86.mmx.psra.w" => "__builtin_ia32_psraw", - "llvm.x86.mmx.psrai.d" => "__builtin_ia32_psradi", - "llvm.x86.mmx.psrai.w" => "__builtin_ia32_psrawi", - "llvm.x86.mmx.psrl.d" => "__builtin_ia32_psrld", - "llvm.x86.mmx.psrl.q" => "__builtin_ia32_psrlq", - "llvm.x86.mmx.psrl.w" => "__builtin_ia32_psrlw", - "llvm.x86.mmx.psrli.d" => "__builtin_ia32_psrldi", - "llvm.x86.mmx.psrli.q" => "__builtin_ia32_psrlqi", - "llvm.x86.mmx.psrli.w" => "__builtin_ia32_psrlwi", - "llvm.x86.mmx.psub.b" => "__builtin_ia32_psubb", - "llvm.x86.mmx.psub.d" => "__builtin_ia32_psubd", - "llvm.x86.mmx.psub.q" => "__builtin_ia32_psubq", - "llvm.x86.mmx.psub.w" => "__builtin_ia32_psubw", - "llvm.x86.mmx.psubs.b" => "__builtin_ia32_psubsb", - "llvm.x86.mmx.psubs.w" => "__builtin_ia32_psubsw", - "llvm.x86.mmx.psubus.b" => "__builtin_ia32_psubusb", - "llvm.x86.mmx.psubus.w" => "__builtin_ia32_psubusw", - "llvm.x86.mmx.punpckhbw" => "__builtin_ia32_punpckhbw", - "llvm.x86.mmx.punpckhdq" => "__builtin_ia32_punpckhdq", - "llvm.x86.mmx.punpckhwd" => "__builtin_ia32_punpckhwd", - "llvm.x86.mmx.punpcklbw" => "__builtin_ia32_punpcklbw", - "llvm.x86.mmx.punpckldq" => "__builtin_ia32_punpckldq", - "llvm.x86.mmx.punpcklwd" => "__builtin_ia32_punpcklwd", - "llvm.x86.mmx.pxor" => "__builtin_ia32_pxor", "llvm.x86.monitorx" => "__builtin_ia32_monitorx", "llvm.x86.movdir64b" => "__builtin_ia32_movdir64b", "llvm.x86.mwaitx" => "__builtin_ia32_mwaitx", @@ -9193,16 +9319,10 @@ "llvm.x86.sse.comile.ss" => "__builtin_ia32_comile", "llvm.x86.sse.comilt.ss" => "__builtin_ia32_comilt", "llvm.x86.sse.comineq.ss" => "__builtin_ia32_comineq", - "llvm.x86.sse.cvtpd2pi" => "__builtin_ia32_cvtpd2pi", - "llvm.x86.sse.cvtpi2pd" => "__builtin_ia32_cvtpi2pd", - "llvm.x86.sse.cvtpi2ps" => "__builtin_ia32_cvtpi2ps", - "llvm.x86.sse.cvtps2pi" => "__builtin_ia32_cvtps2pi", "llvm.x86.sse.cvtsi2ss" => "__builtin_ia32_cvtsi2ss", "llvm.x86.sse.cvtsi642ss" => "__builtin_ia32_cvtsi642ss", "llvm.x86.sse.cvtss2si" => "__builtin_ia32_cvtss2si", "llvm.x86.sse.cvtss2si64" => "__builtin_ia32_cvtss2si64", - "llvm.x86.sse.cvttpd2pi" => "__builtin_ia32_cvttpd2pi", - "llvm.x86.sse.cvttps2pi" => "__builtin_ia32_cvttps2pi", "llvm.x86.sse.cvttss2si" => "__builtin_ia32_cvttss2si", "llvm.x86.sse.cvttss2si64" => "__builtin_ia32_cvttss2si64", "llvm.x86.sse.div.ss" => "__builtin_ia32_divss", @@ -9212,7 +9332,6 @@ "llvm.x86.sse.min.ss" => "__builtin_ia32_minss", "llvm.x86.sse.movmsk.ps" => "__builtin_ia32_movmskps", "llvm.x86.sse.mul.ss" => "__builtin_ia32_mulss", - "llvm.x86.sse.pshuf.w" => "__builtin_ia32_pshufw", "llvm.x86.sse.rcp.ps" => "__builtin_ia32_rcpps", "llvm.x86.sse.rcp.ss" => "__builtin_ia32_rcpss", "llvm.x86.sse.rsqrt.ps" => "__builtin_ia32_rsqrtps", @@ -9398,35 +9517,20 @@ "llvm.x86.sse4a.insertqi" => "__builtin_ia32_insertqi", "llvm.x86.sse4a.movnt.sd" => "__builtin_ia32_movntsd", "llvm.x86.sse4a.movnt.ss" => "__builtin_ia32_movntss", - "llvm.x86.ssse3.pabs.b" => "__builtin_ia32_pabsb", "llvm.x86.ssse3.pabs.b.128" => "__builtin_ia32_pabsb128", - "llvm.x86.ssse3.pabs.d" => "__builtin_ia32_pabsd", "llvm.x86.ssse3.pabs.d.128" => "__builtin_ia32_pabsd128", - "llvm.x86.ssse3.pabs.w" => "__builtin_ia32_pabsw", "llvm.x86.ssse3.pabs.w.128" => "__builtin_ia32_pabsw128", - "llvm.x86.ssse3.phadd.d" => "__builtin_ia32_phaddd", "llvm.x86.ssse3.phadd.d.128" => "__builtin_ia32_phaddd128", - "llvm.x86.ssse3.phadd.sw" => "__builtin_ia32_phaddsw", "llvm.x86.ssse3.phadd.sw.128" => "__builtin_ia32_phaddsw128", - "llvm.x86.ssse3.phadd.w" => "__builtin_ia32_phaddw", "llvm.x86.ssse3.phadd.w.128" => "__builtin_ia32_phaddw128", - "llvm.x86.ssse3.phsub.d" => "__builtin_ia32_phsubd", "llvm.x86.ssse3.phsub.d.128" => "__builtin_ia32_phsubd128", - "llvm.x86.ssse3.phsub.sw" => "__builtin_ia32_phsubsw", "llvm.x86.ssse3.phsub.sw.128" => "__builtin_ia32_phsubsw128", - "llvm.x86.ssse3.phsub.w" => "__builtin_ia32_phsubw", "llvm.x86.ssse3.phsub.w.128" => "__builtin_ia32_phsubw128", - "llvm.x86.ssse3.pmadd.ub.sw" => "__builtin_ia32_pmaddubsw", "llvm.x86.ssse3.pmadd.ub.sw.128" => "__builtin_ia32_pmaddubsw128", - "llvm.x86.ssse3.pmul.hr.sw" => "__builtin_ia32_pmulhrsw", "llvm.x86.ssse3.pmul.hr.sw.128" => "__builtin_ia32_pmulhrsw128", - "llvm.x86.ssse3.pshuf.b" => "__builtin_ia32_pshufb", "llvm.x86.ssse3.pshuf.b.128" => "__builtin_ia32_pshufb128", - "llvm.x86.ssse3.psign.b" => "__builtin_ia32_psignb", "llvm.x86.ssse3.psign.b.128" => "__builtin_ia32_psignb128", - "llvm.x86.ssse3.psign.d" => "__builtin_ia32_psignd", "llvm.x86.ssse3.psign.d.128" => "__builtin_ia32_psignd128", - "llvm.x86.ssse3.psign.w" => "__builtin_ia32_psignw", "llvm.x86.ssse3.psign.w.128" => "__builtin_ia32_psignw128", "llvm.x86.sttilecfg" => "__builtin_ia32_tile_storeconfig", "llvm.x86.stui" => "__builtin_ia32_stui", diff --git a/src/intrinsic/llvm.rs b/src/intrinsic/llvm.rs index a1270482219..7a8fe134cfc 100644 --- a/src/intrinsic/llvm.rs +++ b/src/intrinsic/llvm.rs @@ -1,9 +1,43 @@ use std::borrow::Cow; +use gccjit::{CType, Context}; use gccjit::{Function, FunctionPtrType, RValue, ToRValue, UnaryOp}; use rustc_codegen_ssa::traits::BuilderMethods; -use crate::{builder::Builder, context::CodegenCx}; +use crate::builder::Builder; +use crate::context::CodegenCx; + +#[cfg_attr(not(feature = "master"), allow(unused_variables))] +pub fn adjust_function<'gcc>( + context: &'gcc Context<'gcc>, + func_name: &str, + func_ptr: RValue<'gcc>, + args: &[RValue<'gcc>], +) -> RValue<'gcc> { + // FIXME: we should not need this hack: this is required because both _mm_fcmadd_sch + // and _mm_mask3_fcmadd_round_sch calls llvm.x86.avx512fp16.mask.vfcmadd.csh and we + // seem to need to map this one LLVM intrinsic to 2 different GCC builtins. + #[cfg(feature = "master")] + match func_name { + "__builtin_ia32_vfcmaddcsh_mask3_round" => { + if format!("{:?}", args[3]).ends_with("255") { + return context + .get_target_builtin_function("__builtin_ia32_vfcmaddcsh_mask_round") + .get_address(None); + } + } + "__builtin_ia32_vfmaddcsh_mask3_round" => { + if format!("{:?}", args[3]).ends_with("255") { + return context + .get_target_builtin_function("__builtin_ia32_vfmaddcsh_mask_round") + .get_address(None); + } + } + _ => (), + } + + func_ptr +} pub fn adjust_intrinsic_arguments<'a, 'b, 'gcc, 'tcx>( builder: &Builder<'a, 'gcc, 'tcx>, @@ -12,6 +46,11 @@ pub fn adjust_intrinsic_arguments<'a, 'b, 'gcc, 'tcx>( func_name: &str, original_function_name: Option<&String>, ) -> Cow<'b, [RValue<'gcc>]> { + // TODO: this might not be a good way to workaround the missing tile builtins. + if func_name == "__builtin_trap" { + return vec![].into(); + } + // 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() { @@ -146,7 +185,11 @@ pub fn adjust_intrinsic_arguments<'a, 'b, 'gcc, 'tcx>( | "__builtin_ia32_psrav16hi_mask" | "__builtin_ia32_psrav8hi_mask" | "__builtin_ia32_permvarhi256_mask" - | "__builtin_ia32_permvarhi128_mask" => { + | "__builtin_ia32_permvarhi128_mask" + | "__builtin_ia32_maxph128_mask" + | "__builtin_ia32_maxph256_mask" + | "__builtin_ia32_minph128_mask" + | "__builtin_ia32_minph256_mask" => { let mut new_args = args.to_vec(); let arg3_type = gcc_func.get_param_type(2); let vector_type = arg3_type.dyncast_vector().expect("vector type"); @@ -181,7 +224,19 @@ pub fn adjust_intrinsic_arguments<'a, 'b, 'gcc, 'tcx>( | "__builtin_ia32_vplzcntd_128_mask" | "__builtin_ia32_vplzcntq_512_mask" | "__builtin_ia32_vplzcntq_256_mask" - | "__builtin_ia32_vplzcntq_128_mask" => { + | "__builtin_ia32_vplzcntq_128_mask" + | "__builtin_ia32_cvtqq2pd128_mask" + | "__builtin_ia32_cvtqq2pd256_mask" + | "__builtin_ia32_cvtqq2ps256_mask" + | "__builtin_ia32_cvtuqq2pd128_mask" + | "__builtin_ia32_cvtuqq2pd256_mask" + | "__builtin_ia32_cvtuqq2ps256_mask" + | "__builtin_ia32_vcvtw2ph128_mask" + | "__builtin_ia32_vcvtw2ph256_mask" + | "__builtin_ia32_vcvtuw2ph128_mask" + | "__builtin_ia32_vcvtuw2ph256_mask" + | "__builtin_ia32_vcvtdq2ph256_mask" + | "__builtin_ia32_vcvtudq2ph256_mask" => { let mut new_args = args.to_vec(); // Remove last arg as it doesn't seem to be used in GCC and is always false. new_args.pop(); @@ -280,7 +335,11 @@ pub fn adjust_intrinsic_arguments<'a, 'b, 'gcc, 'tcx>( new_args.push(last_arg); args = new_args.into(); } - "__builtin_ia32_vfmaddsubps512_mask" | "__builtin_ia32_vfmaddsubpd512_mask" => { + "__builtin_ia32_vfmaddsubps512_mask" + | "__builtin_ia32_vfmaddsubpd512_mask" + | "__builtin_ia32_cmpsh_mask_round" + | "__builtin_ia32_vfmaddph512_mask" + | "__builtin_ia32_vfmaddsubph512_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); @@ -303,9 +362,8 @@ pub fn adjust_intrinsic_arguments<'a, 'b, 'gcc, 'tcx>( | "__builtin_ia32_vpermi2varpd128_mask" | "__builtin_ia32_vpmadd52huq512_mask" | "__builtin_ia32_vpmadd52luq512_mask" - | "__builtin_ia32_vpmadd52huq256_mask" - | "__builtin_ia32_vpmadd52luq256_mask" - | "__builtin_ia32_vpmadd52huq128_mask" => { + | "__builtin_ia32_vfmaddsubph128_mask" + | "__builtin_ia32_vfmaddsubph256_mask" => { let mut new_args = args.to_vec(); let arg4_type = gcc_func.get_param_type(3); let minus_one = builder.context.new_rvalue_from_int(arg4_type, -1); @@ -354,7 +412,14 @@ pub fn adjust_intrinsic_arguments<'a, 'b, 'gcc, 'tcx>( let minus_one = builder.context.new_rvalue_from_int(arg4_type, -1); args = vec![new_args[1], new_args[0], new_args[2], minus_one].into(); } - "__builtin_ia32_xrstor" | "__builtin_ia32_xsavec" => { + "__builtin_ia32_xrstor" + | "__builtin_ia32_xrstor64" + | "__builtin_ia32_xsavec" + | "__builtin_ia32_xsavec64" + | "__builtin_ia32_xsave" + | "__builtin_ia32_xsave64" + | "__builtin_ia32_xsaveopt" + | "__builtin_ia32_xsaveopt64" => { let new_args = args.to_vec(); let thirty_two = builder.context.new_rvalue_from_int(new_args[1].get_type(), 32); let arg2 = new_args[1] << thirty_two | new_args[2]; @@ -377,11 +442,76 @@ pub fn adjust_intrinsic_arguments<'a, 'b, 'gcc, 'tcx>( ); args = vec![arg.get_address(None)].into(); } + "__builtin_ia32_cvtqq2pd512_mask" + | "__builtin_ia32_cvtqq2ps512_mask" + | "__builtin_ia32_cvtuqq2pd512_mask" + | "__builtin_ia32_cvtuqq2ps512_mask" + | "__builtin_ia32_sqrtph512_mask_round" + | "__builtin_ia32_vcvtw2ph512_mask_round" + | "__builtin_ia32_vcvtuw2ph512_mask_round" + | "__builtin_ia32_vcvtdq2ph512_mask_round" + | "__builtin_ia32_vcvtudq2ph512_mask_round" + | "__builtin_ia32_vcvtqq2ph512_mask_round" + | "__builtin_ia32_vcvtuqq2ph512_mask_round" => { + let mut old_args = args.to_vec(); + let mut new_args = vec![]; + new_args.push(old_args.swap_remove(0)); + let arg2_type = gcc_func.get_param_type(1); + let vector_type = arg2_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(); + let first_arg = + builder.context.new_rvalue_from_vector(None, arg2_type, &vec![zero; num_units]); + new_args.push(first_arg); + let arg3_type = gcc_func.get_param_type(2); + let minus_one = builder.context.new_rvalue_from_int(arg3_type, -1); + new_args.push(minus_one); + new_args.push(old_args.swap_remove(0)); + args = new_args.into(); + } + "__builtin_ia32_addph512_mask_round" + | "__builtin_ia32_subph512_mask_round" + | "__builtin_ia32_mulph512_mask_round" + | "__builtin_ia32_divph512_mask_round" + | "__builtin_ia32_maxph512_mask_round" + | "__builtin_ia32_minph512_mask_round" => { + 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 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(); + let first_arg = + builder.context.new_rvalue_from_vector(None, arg3_type, &vec![zero; num_units]); + new_args.push(first_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(); + } + // NOTE: the LLVM intrinsics receive 3 floats, but the GCC builtin requires 3 vectors. + "__builtin_ia32_vfmaddsh3_mask" => { + let new_args = args.to_vec(); + let arg1_type = gcc_func.get_param_type(0); + let arg2_type = gcc_func.get_param_type(1); + let arg3_type = gcc_func.get_param_type(2); + let arg4_type = gcc_func.get_param_type(3); + let a = builder.context.new_rvalue_from_vector(None, arg1_type, &[new_args[0]; 8]); + let b = builder.context.new_rvalue_from_vector(None, arg2_type, &[new_args[1]; 8]); + let c = builder.context.new_rvalue_from_vector(None, arg3_type, &[new_args[2]; 8]); + let arg4 = builder.context.new_rvalue_from_int(arg4_type, -1); + args = vec![a, b, c, arg4, new_args[3]].into(); + } _ => (), } } else { match func_name { - "__builtin_ia32_rndscaless_mask_round" | "__builtin_ia32_rndscalesd_mask_round" => { + "__builtin_ia32_rndscaless_mask_round" + | "__builtin_ia32_rndscalesd_mask_round" + | "__builtin_ia32_reducesh_mask_round" => { let new_args = args.to_vec(); let arg3_type = gcc_func.get_param_type(2); let arg3 = builder.context.new_cast(None, new_args[4], arg3_type); @@ -389,7 +519,7 @@ pub fn adjust_intrinsic_arguments<'a, 'b, 'gcc, 'tcx>( let arg4 = builder.context.new_bitcast(None, new_args[2], arg4_type); args = vec![new_args[0], new_args[1], arg3, arg4, new_args[3], new_args[5]].into(); } - // NOTE: the LLVM intrinsic receives 3 floats, but the GCC builtin requires 3 vectors. + // NOTE: the LLVM intrinsics receive 3 floats, but the GCC builtin requires 3 vectors. // FIXME: the intrinsics like _mm_mask_fmadd_sd should probably directly call the GCC // intrinsic to avoid this. "__builtin_ia32_vfmaddss3_round" => { @@ -472,6 +602,52 @@ pub fn adjust_intrinsic_arguments<'a, 'b, 'gcc, 'tcx>( let new_args = args.to_vec(); args = vec![new_args[1], new_args[0], new_args[2]].into(); } + "__builtin_ia32_rangesd128_mask_round" + | "__builtin_ia32_rangess128_mask_round" + | "__builtin_ia32_reducesd_mask_round" + | "__builtin_ia32_reducess_mask_round" => { + let new_args = args.to_vec(); + args = vec![ + new_args[0], + new_args[1], + new_args[4], + new_args[2], + new_args[3], + new_args[5], + ] + .into(); + } + "__builtin_ia32_rndscalesh_mask_round" => { + let new_args = args.to_vec(); + args = vec![ + new_args[0], + new_args[1], + new_args[4], + new_args[2], + new_args[3], + new_args[5], + ] + .into(); + } + "fma" => { + let mut new_args = args.to_vec(); + new_args[0] = builder.context.new_cast(None, new_args[0], builder.double_type); + new_args[1] = builder.context.new_cast(None, new_args[1], builder.double_type); + new_args[2] = builder.context.new_cast(None, new_args[2], builder.double_type); + args = new_args.into(); + } + "__builtin_ia32_sqrtsh_mask_round" + | "__builtin_ia32_vcvtss2sh_mask_round" + | "__builtin_ia32_vcvtsd2sh_mask_round" + | "__builtin_ia32_vcvtsh2ss_mask_round" + | "__builtin_ia32_vcvtsh2sd_mask_round" + | "__builtin_ia32_rcpsh_mask" + | "__builtin_ia32_rsqrtsh_mask" => { + // The first two arguments are inverted, so swap them. + let mut new_args = args.to_vec(); + new_args.swap(0, 1); + args = new_args.into(); + } _ => (), } } @@ -488,7 +664,9 @@ pub fn adjust_intrinsic_return_value<'a, 'gcc, 'tcx>( orig_args: &[RValue<'gcc>], ) -> RValue<'gcc> { match func_name { - "__builtin_ia32_vfmaddss3_round" | "__builtin_ia32_vfmaddsd3_round" => { + "__builtin_ia32_vfmaddss3_round" + | "__builtin_ia32_vfmaddsd3_round" + | "__builtin_ia32_vfmaddsh3_mask" => { #[cfg(feature = "master")] { let zero = builder.context.new_rvalue_zero(builder.int_type); @@ -547,6 +725,10 @@ pub fn adjust_intrinsic_return_value<'a, 'gcc, 'tcx>( &[random_number, success_variable.to_rvalue()], ); } + "fma" => { + let f16_type = builder.context.new_c_type(CType::Float16); + return_value = builder.context.new_cast(None, return_value, f16_type); + } _ => (), } @@ -780,7 +962,9 @@ pub fn intrinsic<'gcc, 'tcx>(name: &str, cx: &CodegenCx<'gcc, 'tcx>) -> Function "llvm.x86.avx512.mask.cmp.b.256" => "__builtin_ia32_cmpb256_mask", "llvm.x86.avx512.mask.cmp.b.128" => "__builtin_ia32_cmpb128_mask", "llvm.x86.xrstor" => "__builtin_ia32_xrstor", + "llvm.x86.xrstor64" => "__builtin_ia32_xrstor64", "llvm.x86.xsavec" => "__builtin_ia32_xsavec", + "llvm.x86.xsavec64" => "__builtin_ia32_xsavec64", "llvm.x86.addcarry.32" => "__builtin_ia32_addcarryx_u32", "llvm.x86.subborrow.32" => "__builtin_ia32_sbb_u32", "llvm.x86.avx512.mask.compress.store.w.512" => "__builtin_ia32_compressstoreuhi512_mask", @@ -969,9 +1153,9 @@ pub fn intrinsic<'gcc, 'tcx>(name: &str, cx: &CodegenCx<'gcc, 'tcx>) -> Function "llvm.x86.avx512.dbpsadbw.128" => "__builtin_ia32_dbpsadbw128_mask", "llvm.x86.avx512.vpmadd52h.uq.512" => "__builtin_ia32_vpmadd52huq512_mask", "llvm.x86.avx512.vpmadd52l.uq.512" => "__builtin_ia32_vpmadd52luq512_mask", - "llvm.x86.avx512.vpmadd52h.uq.256" => "__builtin_ia32_vpmadd52huq256_mask", - "llvm.x86.avx512.vpmadd52l.uq.256" => "__builtin_ia32_vpmadd52luq256_mask", - "llvm.x86.avx512.vpmadd52h.uq.128" => "__builtin_ia32_vpmadd52huq128_mask", + "llvm.x86.avx512.vpmadd52h.uq.256" => "__builtin_ia32_vpmadd52huq256", + "llvm.x86.avx512.vpmadd52l.uq.256" => "__builtin_ia32_vpmadd52luq256", + "llvm.x86.avx512.vpmadd52h.uq.128" => "__builtin_ia32_vpmadd52huq128", "llvm.x86.avx512.vpdpwssd.512" => "__builtin_ia32_vpdpwssd_v16si", "llvm.x86.avx512.vpdpwssd.256" => "__builtin_ia32_vpdpwssd_v8si", "llvm.x86.avx512.vpdpwssd.128" => "__builtin_ia32_vpdpwssd_v4si", @@ -984,6 +1168,180 @@ pub fn intrinsic<'gcc, 'tcx>(name: &str, cx: &CodegenCx<'gcc, 'tcx>) -> Function "llvm.x86.avx512.vpdpbusds.512" => "__builtin_ia32_vpdpbusds_v16si", "llvm.x86.avx512.vpdpbusds.256" => "__builtin_ia32_vpdpbusds_v8si", "llvm.x86.avx512.vpdpbusds.128" => "__builtin_ia32_vpdpbusds_v4si", + "llvm.x86.xsave" => "__builtin_ia32_xsave", + "llvm.x86.xsave64" => "__builtin_ia32_xsave64", + "llvm.x86.xsaveopt" => "__builtin_ia32_xsaveopt", + "llvm.x86.xsaveopt64" => "__builtin_ia32_xsaveopt64", + "llvm.x86.avx512.mask.loadu.w.512" => "__builtin_ia32_loaddquhi512_mask", + "llvm.x86.avx512.mask.loadu.b.512" => "__builtin_ia32_loaddquqi512_mask", + "llvm.x86.avx512.mask.loadu.w.256" => "__builtin_ia32_loaddquhi256_mask", + "llvm.x86.avx512.mask.loadu.b.256" => "__builtin_ia32_loaddquqi256_mask", + "llvm.x86.avx512.mask.loadu.w.128" => "__builtin_ia32_loaddquhi128_mask", + "llvm.x86.avx512.mask.loadu.b.128" => "__builtin_ia32_loaddquqi128_mask", + "llvm.x86.avx512.mask.storeu.w.512" => "__builtin_ia32_storedquhi512_mask", + "llvm.x86.avx512.mask.storeu.b.512" => "__builtin_ia32_storedquqi512_mask", + "llvm.x86.avx512.mask.storeu.w.256" => "__builtin_ia32_storedquhi256_mask", + "llvm.x86.avx512.mask.storeu.b.256" => "__builtin_ia32_storedquqi256_mask", + "llvm.x86.avx512.mask.storeu.w.128" => "__builtin_ia32_storedquhi128_mask", + "llvm.x86.avx512.mask.storeu.b.128" => "__builtin_ia32_storedquqi128_mask", + "llvm.x86.avx512.mask.expand.load.w.512" => "__builtin_ia32_expandloadhi512_mask", + "llvm.x86.avx512.mask.expand.load.w.256" => "__builtin_ia32_expandloadhi256_mask", + "llvm.x86.avx512.mask.expand.load.w.128" => "__builtin_ia32_expandloadhi128_mask", + "llvm.x86.avx512.mask.expand.load.b.512" => "__builtin_ia32_expandloadqi512_mask", + "llvm.x86.avx512.mask.expand.load.b.256" => "__builtin_ia32_expandloadqi256_mask", + "llvm.x86.avx512.mask.expand.load.b.128" => "__builtin_ia32_expandloadqi128_mask", + "llvm.x86.avx512.sitofp.round.v8f64.v8i64" => "__builtin_ia32_cvtqq2pd512_mask", + "llvm.x86.avx512.sitofp.round.v2f64.v2i64" => "__builtin_ia32_cvtqq2pd128_mask", + "llvm.x86.avx512.sitofp.round.v4f64.v4i64" => "__builtin_ia32_cvtqq2pd256_mask", + "llvm.x86.avx512.sitofp.round.v8f32.v8i64" => "__builtin_ia32_cvtqq2ps512_mask", + "llvm.x86.avx512.sitofp.round.v4f32.v4i64" => "__builtin_ia32_cvtqq2ps256_mask", + "llvm.x86.avx512.uitofp.round.v8f64.v8u64" => "__builtin_ia32_cvtuqq2pd512_mask", + "llvm.x86.avx512.uitofp.round.v2f64.v2u64" => "__builtin_ia32_cvtuqq2pd128_mask", + "llvm.x86.avx512.uitofp.round.v4f64.v4u64" => "__builtin_ia32_cvtuqq2pd256_mask", + "llvm.x86.avx512.uitofp.round.v8f32.v8u64" => "__builtin_ia32_cvtuqq2ps512_mask", + "llvm.x86.avx512.uitofp.round.v4f32.v4u64" => "__builtin_ia32_cvtuqq2ps256_mask", + "llvm.x86.avx512.mask.reduce.pd.512" => "__builtin_ia32_reducepd512_mask_round", + "llvm.x86.avx512.mask.reduce.ps.512" => "__builtin_ia32_reduceps512_mask_round", + "llvm.x86.avx512.mask.reduce.sd" => "__builtin_ia32_reducesd_mask_round", + "llvm.x86.avx512.mask.reduce.ss" => "__builtin_ia32_reducess_mask_round", + "llvm.x86.avx512.mask.loadu.d.256" => "__builtin_ia32_loaddqusi256_mask", + "llvm.x86.avx512.mask.loadu.q.256" => "__builtin_ia32_loaddqudi256_mask", + "llvm.x86.avx512.mask.loadu.ps.256" => "__builtin_ia32_loadups256_mask", + "llvm.x86.avx512.mask.loadu.pd.256" => "__builtin_ia32_loadupd256_mask", + "llvm.x86.avx512.mask.loadu.d.128" => "__builtin_ia32_loaddqusi128_mask", + "llvm.x86.avx512.mask.loadu.q.128" => "__builtin_ia32_loaddqudi128_mask", + "llvm.x86.avx512.mask.loadu.ps.128" => "__builtin_ia32_loadups128_mask", + "llvm.x86.avx512.mask.loadu.pd.128" => "__builtin_ia32_loadupd128_mask", + "llvm.x86.avx512.mask.load.d.512" => "__builtin_ia32_movdqa32load512_mask", + "llvm.x86.avx512.mask.load.q.512" => "__builtin_ia32_movdqa64load512_mask", + "llvm.x86.avx512.mask.load.ps.512" => "__builtin_ia32_loadaps512_mask", + "llvm.x86.avx512.mask.load.pd.512" => "__builtin_ia32_loadapd512_mask", + "llvm.x86.avx512.mask.load.d.256" => "__builtin_ia32_movdqa32load256_mask", + "llvm.x86.avx512.mask.load.q.256" => "__builtin_ia32_movdqa64load256_mask", + "llvm.x86.avx512fp16.mask.cmp.sh" => "__builtin_ia32_cmpsh_mask_round", + "llvm.x86.avx512fp16.vcomi.sh" => "__builtin_ia32_cmpsh_mask_round", + "llvm.x86.avx512fp16.add.ph.512" => "__builtin_ia32_addph512_mask_round", + "llvm.x86.avx512fp16.sub.ph.512" => "__builtin_ia32_subph512_mask_round", + "llvm.x86.avx512fp16.mul.ph.512" => "__builtin_ia32_mulph512_mask_round", + "llvm.x86.avx512fp16.div.ph.512" => "__builtin_ia32_divph512_mask_round", + "llvm.x86.avx512fp16.mask.vfmul.cph.512" => "__builtin_ia32_vfmulcph512_mask_round", + "llvm.x86.avx512fp16.mask.vfmul.csh" => "__builtin_ia32_vfmulcsh_mask_round", + "llvm.x86.avx512fp16.mask.vfcmul.cph.512" => "__builtin_ia32_vfcmulcph512_mask_round", + "llvm.x86.avx512fp16.mask.vfcmul.csh" => "__builtin_ia32_vfcmulcsh_mask_round", + "llvm.x86.avx512fp16.mask.vfmadd.cph.512" => "__builtin_ia32_vfmaddcph512_mask3_round", + "llvm.x86.avx512fp16.maskz.vfmadd.cph.512" => "__builtin_ia32_vfmaddcph512_maskz_round", + "llvm.x86.avx512fp16.mask.vfmadd.csh" => "__builtin_ia32_vfmaddcsh_mask3_round", + "llvm.x86.avx512fp16.maskz.vfmadd.csh" => "__builtin_ia32_vfmaddcsh_maskz_round", + "llvm.x86.avx512fp16.mask.vfcmadd.cph.512" => "__builtin_ia32_vfcmaddcph512_mask3_round", + "llvm.x86.avx512fp16.maskz.vfcmadd.cph.512" => "__builtin_ia32_vfcmaddcph512_maskz_round", + "llvm.x86.avx512fp16.mask.vfcmadd.csh" => "__builtin_ia32_vfcmaddcsh_mask3_round", + "llvm.x86.avx512fp16.maskz.vfcmadd.csh" => "__builtin_ia32_vfcmaddcsh_maskz_round", + "llvm.x86.avx512fp16.vfmadd.ph.512" => "__builtin_ia32_vfmaddph512_mask", + "llvm.x86.avx512fp16.vcvtsi642sh" => "__builtin_ia32_vcvtsi2sh64_round", + "llvm.x86.avx512fp16.vcvtusi642sh" => "__builtin_ia32_vcvtusi2sh64_round", + "llvm.x86.avx512fp16.vcvtsh2si64" => "__builtin_ia32_vcvtsh2si64_round", + "llvm.x86.avx512fp16.vcvtsh2usi64" => "__builtin_ia32_vcvtsh2usi64_round", + "llvm.x86.avx512fp16.vcvttsh2si64" => "__builtin_ia32_vcvttsh2si64_round", + "llvm.x86.avx512fp16.vcvttsh2usi64" => "__builtin_ia32_vcvttsh2usi64_round", + "llvm.x86.avx512.mask.load.ps.256" => "__builtin_ia32_loadaps256_mask", + "llvm.x86.avx512.mask.load.pd.256" => "__builtin_ia32_loadapd256_mask", + "llvm.x86.avx512.mask.load.d.128" => "__builtin_ia32_movdqa32load128_mask", + "llvm.x86.avx512.mask.load.q.128" => "__builtin_ia32_movdqa64load128_mask", + "llvm.x86.avx512.mask.load.ps.128" => "__builtin_ia32_loadaps128_mask", + "llvm.x86.avx512.mask.load.pd.128" => "__builtin_ia32_loadapd128_mask", + "llvm.x86.avx512.mask.storeu.d.256" => "__builtin_ia32_storedqusi256_mask", + "llvm.x86.avx512.mask.storeu.q.256" => "__builtin_ia32_storedqudi256_mask", + "llvm.x86.avx512.mask.storeu.ps.256" => "__builtin_ia32_storeups256_mask", + "llvm.x86.avx512.mask.storeu.pd.256" => "__builtin_ia32_storeupd256_mask", + "llvm.x86.avx512.mask.storeu.d.128" => "__builtin_ia32_storedqusi128_mask", + "llvm.x86.avx512.mask.storeu.q.128" => "__builtin_ia32_storedqudi128_mask", + "llvm.x86.avx512.mask.storeu.ps.128" => "__builtin_ia32_storeups128_mask", + "llvm.x86.avx512.mask.storeu.pd.128" => "__builtin_ia32_storeupd128_mask", + "llvm.x86.avx512.mask.store.d.512" => "__builtin_ia32_movdqa32store512_mask", + "llvm.x86.avx512.mask.store.q.512" => "__builtin_ia32_movdqa64store512_mask", + "llvm.x86.avx512.mask.store.ps.512" => "__builtin_ia32_storeaps512_mask", + "llvm.x86.avx512.mask.store.pd.512" => "__builtin_ia32_storeapd512_mask", + "llvm.x86.avx512.mask.store.d.256" => "__builtin_ia32_movdqa32store256_mask", + "llvm.x86.avx512.mask.store.q.256" => "__builtin_ia32_movdqa64store256_mask", + "llvm.x86.avx512.mask.store.ps.256" => "__builtin_ia32_storeaps256_mask", + "llvm.x86.avx512.mask.store.pd.256" => "__builtin_ia32_storeapd256_mask", + "llvm.x86.avx512.mask.store.d.128" => "__builtin_ia32_movdqa32store128_mask", + "llvm.x86.avx512.mask.store.q.128" => "__builtin_ia32_movdqa64store128_mask", + "llvm.x86.avx512.mask.store.ps.128" => "__builtin_ia32_storeaps128_mask", + "llvm.x86.avx512.mask.store.pd.128" => "__builtin_ia32_storeapd128_mask", + "llvm.x86.avx512fp16.vfmadd.f16" => "__builtin_ia32_vfmaddsh3_mask", + "llvm.x86.avx512fp16.vfmaddsub.ph.128" => "__builtin_ia32_vfmaddsubph128_mask", + "llvm.x86.avx512fp16.vfmaddsub.ph.256" => "__builtin_ia32_vfmaddsubph256_mask", + "llvm.x86.avx512fp16.vfmaddsub.ph.512" => "__builtin_ia32_vfmaddsubph512_mask", + "llvm.x86.avx512fp16.sqrt.ph.512" => "__builtin_ia32_sqrtph512_mask_round", + "llvm.x86.avx512fp16.mask.sqrt.sh" => "__builtin_ia32_sqrtsh_mask_round", + "llvm.x86.avx512fp16.max.ph.128" => "__builtin_ia32_maxph128_mask", + "llvm.x86.avx512fp16.max.ph.256" => "__builtin_ia32_maxph256_mask", + "llvm.x86.avx512fp16.max.ph.512" => "__builtin_ia32_maxph512_mask_round", + "llvm.x86.avx512fp16.min.ph.128" => "__builtin_ia32_minph128_mask", + "llvm.x86.avx512fp16.min.ph.256" => "__builtin_ia32_minph256_mask", + "llvm.x86.avx512fp16.min.ph.512" => "__builtin_ia32_minph512_mask_round", + "llvm.x86.avx512fp16.mask.getexp.sh" => "__builtin_ia32_getexpsh_mask_round", + "llvm.x86.avx512fp16.mask.rndscale.ph.128" => "__builtin_ia32_rndscaleph128_mask", + "llvm.x86.avx512fp16.mask.rndscale.ph.256" => "__builtin_ia32_rndscaleph256_mask", + "llvm.x86.avx512fp16.mask.rndscale.ph.512" => "__builtin_ia32_rndscaleph512_mask_round", + "llvm.x86.avx512fp16.mask.scalef.ph.512" => "__builtin_ia32_scalefph512_mask_round", + "llvm.x86.avx512fp16.mask.reduce.ph.512" => "__builtin_ia32_reduceph512_mask_round", + "llvm.x86.avx512fp16.mask.reduce.sh" => "__builtin_ia32_reducesh_mask_round", + "llvm.x86.avx512.sitofp.round.v8f16.v8i16" => "__builtin_ia32_vcvtw2ph128_mask", + "llvm.x86.avx512.sitofp.round.v16f16.v16i16" => "__builtin_ia32_vcvtw2ph256_mask", + "llvm.x86.avx512.sitofp.round.v32f16.v32i16" => "__builtin_ia32_vcvtw2ph512_mask_round", + "llvm.x86.avx512.uitofp.round.v8f16.v8u16" => "__builtin_ia32_vcvtuw2ph128_mask", + "llvm.x86.avx512.uitofp.round.v16f16.v16u16" => "__builtin_ia32_vcvtuw2ph256_mask", + "llvm.x86.avx512.uitofp.round.v32f16.v32u16" => "__builtin_ia32_vcvtuw2ph512_mask_round", + "llvm.x86.avx512.sitofp.round.v8f16.v8i32" => "__builtin_ia32_vcvtdq2ph256_mask", + "llvm.x86.avx512.sitofp.round.v16f16.v16i32" => "__builtin_ia32_vcvtdq2ph512_mask_round", + "llvm.x86.avx512fp16.vcvtsi2sh" => "__builtin_ia32_vcvtsi2sh32_round", + "llvm.x86.avx512.uitofp.round.v8f16.v8u32" => "__builtin_ia32_vcvtudq2ph256_mask", + "llvm.x86.avx512.uitofp.round.v16f16.v16u32" => "__builtin_ia32_vcvtudq2ph512_mask_round", + "llvm.x86.avx512fp16.vcvtusi2sh" => "__builtin_ia32_vcvtusi2sh32_round", + "llvm.x86.avx512.sitofp.round.v8f16.v8i64" => "__builtin_ia32_vcvtqq2ph512_mask_round", + "llvm.x86.avx512.uitofp.round.v8f16.v8u64" => "__builtin_ia32_vcvtuqq2ph512_mask_round", + "llvm.x86.avx512fp16.mask.vcvtps2phx.512" => "__builtin_ia32_vcvtps2phx512_mask_round", + "llvm.x86.avx512fp16.mask.vcvtpd2ph.512" => "__builtin_ia32_vcvtpd2ph512_mask_round", + "llvm.x86.avx512fp16.mask.vcvtph2uw.512" => "__builtin_ia32_vcvtph2uw512_mask_round", + "llvm.x86.avx512fp16.mask.vcvttph2w.512" => "__builtin_ia32_vcvttph2w512_mask_round", + "llvm.x86.avx512fp16.mask.vcvttph2uw.512" => "__builtin_ia32_vcvttph2uw512_mask_round", + "llvm.x86.avx512fp16.mask.vcvtph2dq.512" => "__builtin_ia32_vcvtph2dq512_mask_round", + "llvm.x86.avx512fp16.vcvtsh2si32" => "__builtin_ia32_vcvtsh2si32_round", + "llvm.x86.avx512fp16.mask.vcvtph2udq.512" => "__builtin_ia32_vcvtph2udq512_mask_round", + "llvm.x86.avx512fp16.vcvtsh2usi32" => "__builtin_ia32_vcvtsh2usi32_round", + "llvm.x86.avx512fp16.mask.vcvttph2dq.512" => "__builtin_ia32_vcvttph2dq512_mask_round", + "llvm.x86.avx512fp16.vcvttsh2si32" => "__builtin_ia32_vcvttsh2si32_round", + "llvm.x86.avx512fp16.mask.vcvttph2udq.512" => "__builtin_ia32_vcvttph2udq512_mask_round", + "llvm.x86.avx512fp16.vcvttsh2usi32" => "__builtin_ia32_vcvttsh2usi32_round", + "llvm.x86.avx512fp16.mask.vcvtph2qq.512" => "__builtin_ia32_vcvtph2qq512_mask_round", + "llvm.x86.avx512fp16.mask.vcvtph2uqq.512" => "__builtin_ia32_vcvtph2uqq512_mask_round", + "llvm.x86.avx512fp16.mask.vcvttph2qq.512" => "__builtin_ia32_vcvttph2qq512_mask_round", + "llvm.x86.avx512fp16.mask.vcvttph2uqq.512" => "__builtin_ia32_vcvttph2uqq512_mask_round", + "llvm.x86.avx512fp16.mask.vcvtph2psx.512" => "__builtin_ia32_vcvtph2psx512_mask_round", + "llvm.x86.avx512fp16.mask.vcvtph2pd.512" => "__builtin_ia32_vcvtph2pd512_mask_round", + "llvm.x86.avx512fp16.mask.vfcmadd.cph.256" => "__builtin_ia32_vfcmaddcph256_mask3", + "llvm.x86.avx512fp16.mask.vfmadd.cph.256" => "__builtin_ia32_vfmaddcph256_mask3", + "llvm.x86.avx512fp16.mask.vfcmadd.cph.128" => "__builtin_ia32_vfcmaddcph128_mask3", + "llvm.x86.avx512fp16.mask.vfmadd.cph.128" => "__builtin_ia32_vfmaddcph128_mask3", + + // TODO: support the tile builtins: + "llvm.x86.ldtilecfg" => "__builtin_trap", + "llvm.x86.sttilecfg" => "__builtin_trap", + "llvm.x86.tileloadd64" => "__builtin_trap", + "llvm.x86.tilerelease" => "__builtin_trap", + "llvm.x86.tilestored64" => "__builtin_trap", + "llvm.x86.tileloaddt164" => "__builtin_trap", + "llvm.x86.tilezero" => "__builtin_trap", + "llvm.x86.tdpbf16ps" => "__builtin_trap", + "llvm.x86.tdpbssd" => "__builtin_trap", + "llvm.x86.tdpbsud" => "__builtin_trap", + "llvm.x86.tdpbusd" => "__builtin_trap", + "llvm.x86.tdpbuud" => "__builtin_trap", + "llvm.x86.tdpfp16ps" => "__builtin_trap", + "llvm.x86.tcmmimfp16ps" => "__builtin_trap", + "llvm.x86.tcmmrlfp16ps" => "__builtin_trap", // NOTE: this file is generated by https://github.com/GuillaumeGomez/llvmint/blob/master/generate_list.py _ => include!("archs.rs"), diff --git a/src/intrinsic/mod.rs b/src/intrinsic/mod.rs index 839ebf3f298..d95a7782fa0 100644 --- a/src/intrinsic/mod.rs +++ b/src/intrinsic/mod.rs @@ -127,20 +127,13 @@ fn codegen_intrinsic_call( // https://github.com/rust-lang/rust-clippy/issues/12497 // and leave `else if use_integer_compare` to be placed "as is". #[allow(clippy::suspicious_else_formatting)] - let llval = match name { + let value = match name { _ if simple.is_some() => { - // FIXME(antoyo): remove this cast when the API supports function. - let func = unsafe { - std::mem::transmute::, RValue<'gcc>>(simple.expect("simple")) - }; - self.call( - self.type_void(), - None, - None, + let func = simple.expect("simple function"); + self.cx.context.new_call( + self.location, func, &args.iter().map(|arg| arg.immediate()).collect::>(), - None, - None, ) } sym::likely => self.expect(args[0].immediate(), true), @@ -383,7 +376,7 @@ fn codegen_intrinsic_call( _ if name_str.starts_with("simd_") => { match generic_simd_intrinsic(self, name, callee_ty, args, ret_ty, llret_ty, span) { - Ok(llval) => llval, + Ok(value) => value, Err(()) => return Ok(()), } } @@ -396,9 +389,9 @@ fn codegen_intrinsic_call( if let PassMode::Cast { cast: ref ty, .. } = fn_abi.ret.mode { let ptr_llty = self.type_ptr_to(ty.gcc_type(self)); let ptr = self.pointercast(result.val.llval, ptr_llty); - self.store(llval, ptr, result.val.align); + self.store(value, ptr, result.val.align); } else { - OperandRef::from_immediate_or_packed_pair(self, llval, result.layout) + OperandRef::from_immediate_or_packed_pair(self, value, result.layout) .val .store(self, result); } @@ -677,11 +670,7 @@ fn bit_reverse(&mut self, width: u64, value: RValue<'gcc>) -> RValue<'gcc> { let step3 = self.or(left, right); // Fourth step. - if width == 8 { - step3 - } else { - self.gcc_bswap(step3, width) - } + if width == 8 { step3 } else { self.gcc_bswap(step3, width) } } 128 => { // TODO(antoyo): find a more efficient implementation? diff --git a/src/intrinsic/simd.rs b/src/intrinsic/simd.rs index ba214a9c24c..79b345982c6 100644 --- a/src/intrinsic/simd.rs +++ b/src/intrinsic/simd.rs @@ -1,10 +1,8 @@ use std::iter::FromIterator; -use gccjit::ToRValue; -use gccjit::{BinaryOp, RValue, Type}; +use gccjit::{BinaryOp, RValue, ToRValue, Type}; #[cfg(feature = "master")] use gccjit::{ComparisonOp, UnaryOp}; - use rustc_codegen_ssa::base::compare_simd_types; use rustc_codegen_ssa::common::{IntPredicate, TypeKind}; #[cfg(feature = "master")] @@ -203,7 +201,7 @@ macro_rules! require_simd { bx.context.new_bitcast(None, shuffled, v_type) }; - if name == sym::simd_bswap || name == sym::simd_bitreverse { + if matches!(name, sym::simd_bswap | sym::simd_bitreverse | sym::simd_ctpop) { require!( bx.type_kind(bx.element_type(llret_ty)) == TypeKind::Integer, InvalidMonomorphization::UnsupportedOperation { span, name, in_ty, in_elem } @@ -214,6 +212,22 @@ macro_rules! require_simd { return Ok(simd_bswap(bx, args[0].immediate())); } + let simd_ctpop = |bx: &mut Builder<'a, 'gcc, 'tcx>, vector: RValue<'gcc>| -> RValue<'gcc> { + let mut vector_elements = vec![]; + let elem_ty = bx.element_type(llret_ty); + for i in 0..in_len { + let index = bx.context.new_rvalue_from_long(bx.ulong_type, i as i64); + let element = bx.extract_element(vector, index).to_rvalue(); + let result = bx.context.new_cast(None, bx.pop_count(element), elem_ty); + vector_elements.push(result); + } + bx.context.new_rvalue_from_vector(None, llret_ty, &vector_elements) + }; + + if name == sym::simd_ctpop { + return Ok(simd_ctpop(bx, args[0].immediate())); + } + // We use a different algorithm from non-vector bitreverse to take advantage of most // processors' vector shuffle units. It works like this: // 1. Generate pre-reversed low and high nibbles as a vector. @@ -725,11 +739,12 @@ macro_rules! return_error { return Err(()); }}; } - let (elem_ty_str, elem_ty) = if let ty::Float(ref f) = *in_elem.kind() { + let (elem_ty_str, elem_ty, cast_type) = if let ty::Float(ref f) = *in_elem.kind() { let elem_ty = bx.cx.type_float_from_ty(*f); match f.bit_width() { - 32 => ("f", elem_ty), - 64 => ("", elem_ty), + 16 => ("", elem_ty, Some(bx.cx.double_type)), + 32 => ("f", elem_ty, None), + 64 => ("", elem_ty, None), _ => { return_error!(InvalidMonomorphization::FloatingPointVector { span, @@ -765,10 +780,7 @@ macro_rules! return_error { _ => return_error!(InvalidMonomorphization::UnrecognizedIntrinsic { span, name }), }; let builtin_name = format!("{}{}", intr_name, elem_ty_str); - let funcs = bx.cx.functions.borrow(); - let function = funcs - .get(&builtin_name) - .unwrap_or_else(|| panic!("unable to find builtin function {}", builtin_name)); + let function = bx.context.get_builtin_function(builtin_name); // TODO(antoyo): add platform-specific behavior here for architectures that have these // intrinsics as instructions (for instance, gpus) @@ -776,17 +788,28 @@ macro_rules! return_error { for i in 0..in_len { let index = bx.context.new_rvalue_from_long(bx.ulong_type, i as i64); // we have to treat fpowi specially, since fpowi's second argument is always an i32 - let arguments = if name == sym::simd_fpowi { - vec![ + let mut arguments = vec![]; + if name == sym::simd_fpowi { + arguments = vec![ bx.extract_element(args[0].immediate(), index).to_rvalue(), args[1].immediate(), - ] + ]; } else { - args.iter() - .map(|arg| bx.extract_element(arg.immediate(), index).to_rvalue()) - .collect() + for arg in args { + let mut element = bx.extract_element(arg.immediate(), index).to_rvalue(); + // FIXME: it would probably be better to not have casts here and use the proper + // instructions. + if let Some(typ) = cast_type { + element = bx.context.new_cast(None, element, typ); + } + arguments.push(element); + } }; - vector_elements.push(bx.context.new_call(None, *function, &arguments)); + let mut result = bx.context.new_call(None, function, &arguments); + if cast_type.is_some() { + result = bx.context.new_cast(None, result, elem_ty); + } + vector_elements.push(result); } let c = bx.context.new_rvalue_from_vector(None, vec_ty, &vector_elements); Ok(c) diff --git a/src/lib.rs b/src/lib.rs index c11a183acde..7b0d594ddc1 100644 --- a/src/lib.rs +++ b/src/lib.rs @@ -79,14 +79,11 @@ use std::sync::atomic::AtomicBool; #[cfg(not(feature = "master"))] use std::sync::atomic::Ordering; -use std::sync::Arc; -use std::sync::Mutex; +use std::sync::{Arc, Mutex}; -use back::lto::ThinBuffer; -use back::lto::ThinData; +use back::lto::{ThinBuffer, ThinData}; use errors::LTONotSupported; -use gccjit::CType; -use gccjit::{Context, OptimizationLevel}; +use gccjit::{CType, Context, OptimizationLevel}; #[cfg(feature = "master")] use gccjit::{TargetInfo, Version}; use rustc_ast::expand::allocator::AllocatorKind; @@ -489,7 +486,7 @@ pub fn target_features( sess.target .supported_target_features() .iter() - .filter_map(|&(feature, gate)| { + .filter_map(|&(feature, gate, _)| { if sess.is_nightly_build() || allow_unstable || gate.is_stable() { Some(feature) } else { diff --git a/src/mono_item.rs b/src/mono_item.rs index 44657ad4f6e..ba81dea49d5 100644 --- a/src/mono_item.rs +++ b/src/mono_item.rs @@ -9,10 +9,9 @@ use rustc_middle::ty::layout::{FnAbiOf, LayoutOf}; use rustc_middle::ty::{self, Instance, TypeVisitableExt}; -use crate::attributes; -use crate::base; use crate::context::CodegenCx; use crate::type_of::LayoutGccExt; +use crate::{attributes, base}; impl<'gcc, 'tcx> PreDefineMethods<'tcx> for CodegenCx<'gcc, 'tcx> { #[cfg_attr(not(feature = "master"), allow(unused_variables))] @@ -38,7 +37,7 @@ fn predefine_static( let is_tls = attrs.flags.contains(CodegenFnAttrFlags::THREAD_LOCAL); let global = self.define_global(symbol_name, gcc_type, is_tls, attrs.link_section); #[cfg(feature = "master")] - global.add_string_attribute(VarAttribute::Visibility(base::visibility_to_gcc(visibility))); + global.add_attribute(VarAttribute::Visibility(base::visibility_to_gcc(visibility))); // TODO(antoyo): set linkage. self.instances.borrow_mut().insert(instance, global); diff --git a/tests/failing-ice-tests.txt b/tests/failing-ice-tests.txt index 2084f86b62e..ff1b6f14894 100644 --- a/tests/failing-ice-tests.txt +++ b/tests/failing-ice-tests.txt @@ -34,3 +34,7 @@ tests/ui/sepcomp/sepcomp-unwind.rs tests/ui/extern/issue-64655-extern-rust-must-allow-unwind.rs tests/ui/extern/issue-64655-allow-unwind-when-calling-panic-directly.rs tests/ui/unwind-no-uwtable.rs +tests/ui/delegation/fn-header.rs +tests/ui/simd/intrinsic/generic-arithmetic-pass.rs +tests/ui/simd/masked-load-store.rs +tests/ui/rfcs/rfc-2632-const-trait-impl/effects/minicore.rs diff --git a/tests/failing-ui-tests.txt b/tests/failing-ui-tests.txt index 5a55bdb156e..56b51275a53 100644 --- a/tests/failing-ui-tests.txt +++ b/tests/failing-ui-tests.txt @@ -95,3 +95,29 @@ tests/ui/simd/intrinsic/generic-arithmetic-pass.rs tests/ui/backtrace/backtrace.rs tests/ui/lifetimes/tail-expr-lock-poisoning.rs tests/ui/runtime/rt-explody-panic-payloads.rs +tests/ui/codegen/equal-pointers-unequal/as-cast/function.rs +tests/ui/codegen/equal-pointers-unequal/as-cast/basic.rs +tests/ui/codegen/equal-pointers-unequal/as-cast/inline1.rs +tests/ui/codegen/equal-pointers-unequal/as-cast/print.rs +tests/ui/codegen/equal-pointers-unequal/as-cast/inline2.rs +tests/ui/codegen/equal-pointers-unequal/as-cast/print3.rs +tests/ui/codegen/equal-pointers-unequal/as-cast/segfault.rs +tests/ui/codegen/equal-pointers-unequal/exposed-provenance/function.rs +tests/ui/codegen/equal-pointers-unequal/exposed-provenance/basic.rs +tests/ui/codegen/equal-pointers-unequal/as-cast/zero.rs +tests/ui/codegen/equal-pointers-unequal/exposed-provenance/inline1.rs +tests/ui/codegen/equal-pointers-unequal/exposed-provenance/print.rs +tests/ui/codegen/equal-pointers-unequal/exposed-provenance/print3.rs +tests/ui/codegen/equal-pointers-unequal/exposed-provenance/inline2.rs +tests/ui/codegen/equal-pointers-unequal/exposed-provenance/segfault.rs +tests/ui/codegen/equal-pointers-unequal/exposed-provenance/zero.rs +tests/ui/codegen/equal-pointers-unequal/strict-provenance/basic.rs +tests/ui/codegen/equal-pointers-unequal/strict-provenance/function.rs +tests/ui/codegen/equal-pointers-unequal/strict-provenance/print.rs +tests/ui/codegen/equal-pointers-unequal/strict-provenance/inline1.rs +tests/ui/codegen/equal-pointers-unequal/strict-provenance/print3.rs +tests/ui/codegen/equal-pointers-unequal/strict-provenance/inline2.rs +tests/ui/codegen/equal-pointers-unequal/strict-provenance/segfault.rs +tests/ui/codegen/equal-pointers-unequal/strict-provenance/zero.rs +tests/ui/sanitizer/cfi-sized-associated-ty.rs +tests/ui/sanitizer/cfi-can-reveal-opaques.rs diff --git a/tests/run/array.rs b/tests/run/array.rs index 3fe8917c9a3..432f11ad8d4 100644 --- a/tests/run/array.rs +++ b/tests/run/array.rs @@ -31,6 +31,7 @@ impl Copy for i32 {} impl Copy for u8 {} impl Copy for i8 {} impl Copy for i16 {} +impl Copy for *mut T {} #[lang = "receiver"] trait Receiver { diff --git a/tests/run/closure.rs b/tests/run/closure.rs index 355f0acee74..00e61cc001f 100644 --- a/tests/run/closure.rs +++ b/tests/run/closure.rs @@ -33,6 +33,7 @@ impl Copy for i32 {} impl Copy for u32 {} impl Copy for u8 {} impl Copy for i8 {} +impl Copy for *mut T {} #[lang = "receiver"] trait Receiver { diff --git a/tests/run/condition.rs b/tests/run/condition.rs index 1b3ae6dc004..7b05b7decd3 100644 --- a/tests/run/condition.rs +++ b/tests/run/condition.rs @@ -34,6 +34,7 @@ impl Copy for i16 {} impl Copy for char {} impl Copy for i8 {} impl Copy for u8 {} +impl Copy for *mut T {} #[lang = "receiver"] trait Receiver { diff --git a/tests/run/fun_ptr.rs b/tests/run/fun_ptr.rs index 96030359772..4e96f376555 100644 --- a/tests/run/fun_ptr.rs +++ b/tests/run/fun_ptr.rs @@ -28,6 +28,7 @@ impl Copy for i32 {} impl Copy for u8 {} impl Copy for i8 {} impl Copy for i16 {} +impl Copy for *mut T {} #[lang = "receiver"] trait Receiver { diff --git a/tests/run/ptr_cast.rs b/tests/run/ptr_cast.rs index 09d77abe27c..a94279182d6 100644 --- a/tests/run/ptr_cast.rs +++ b/tests/run/ptr_cast.rs @@ -28,6 +28,7 @@ impl Copy for i32 {} impl Copy for u8 {} impl Copy for i8 {} impl Copy for i16 {} +impl Copy for *mut T {} #[lang = "receiver"] trait Receiver { diff --git a/tests/run/slice.rs b/tests/run/slice.rs index 1262c86c810..e86fc823a1a 100644 --- a/tests/run/slice.rs +++ b/tests/run/slice.rs @@ -26,6 +26,7 @@ impl Copy for isize {} impl Copy for usize {} impl Copy for i32 {} impl Copy for u32 {} +impl Copy for *mut T {} #[lang = "receiver"] trait Receiver { diff --git a/tests/run/static.rs b/tests/run/static.rs index e7c46ae3fcc..6247e08f5e3 100644 --- a/tests/run/static.rs +++ b/tests/run/static.rs @@ -34,6 +34,7 @@ trait Copy { } impl Copy for isize {} +impl Copy for *mut T {} #[lang = "receiver"] trait Receiver {