Fix/enable features

Readme.md

@@ -205,6 +205,19 @@ error: failed to build archive: failed to open object file: No such file or dire
 That can be caused by the fact that you try to compile with `lto = "fat"`, but you didn't compile the sysroot with LTO.
 (Not sure if that's the reason since I cannot reproduce anymore. Maybe it happened when forgetting setting `FAT_LTO`.)
 
+### ld: cannot find crtbegin.o
+
+When compiling an executable with libgccijt, if setting the `*LIBRARY_PATH` variables to the install directory, you will get the following errors:
+
+```
+ld: cannot find crtbegin.o: No such file or directory
+ld: cannot find -lgcc: No such file or directory
+ld: cannot find -lgcc: No such file or directory
+libgccjit.so: error: error invoking gcc driver
+```
+
+To fix this, set the variables to `gcc-build/build/gcc`.
+
 ### 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.

messages.ftl

@@ -1,3 +1,7 @@
+codegen_gcc_unknown_ctarget_feature_prefix =
+    unknown feature specified for `-Ctarget-feature`: `{$feature}`
+    .note = features must begin with a `+` to enable or `-` to disable it
+
 codegen_gcc_invalid_minimum_alignment =
     invalid minimum global alignment: {$err}
 
@@ -23,3 +27,15 @@ codegen_gcc_lto_disallowed = lto can only be run for executables, cdylibs and st
 codegen_gcc_lto_dylib = lto cannot be used for `dylib` crate type without `-Zdylib-lto`
 
 codegen_gcc_lto_bitcode_from_rlib = failed to get bitcode from object file for LTO ({$gcc_err})
+
+codegen_gcc_unknown_ctarget_feature =
+    unknown feature specified for `-Ctarget-feature`: `{$feature}`
+    .note = it is still passed through to the codegen backend
+    .possible_feature = you might have meant: `{$rust_feature}`
+    .consider_filing_feature_request = consider filing a feature request
+
+codegen_gcc_missing_features =
+    add the missing features in a `target_feature` attribute
+
+codegen_gcc_target_feature_disable_or_enable =
+    the target features {$features} must all be either enabled or disabled together

src/abi.rs

@@ -3,6 +3,7 @@ use rustc_codegen_ssa::traits::{AbiBuilderMethods, BaseTypeMethods};
 use rustc_data_structures::fx::FxHashSet;
 use rustc_middle::bug;
 use rustc_middle::ty::Ty;
+#[cfg(feature = "master")]
 use rustc_session::config;
 use rustc_target::abi::call::{ArgAttributes, CastTarget, FnAbi, PassMode, Reg, RegKind};
 

src/attributes.rs

@@ -4,72 +4,13 @@ use gccjit::Function;
 use rustc_attr::InstructionSetAttr;
 #[cfg(feature="master")]
 use rustc_attr::InlineAttr;
-use rustc_codegen_ssa::target_features::tied_target_features;
-use rustc_data_structures::fx::FxHashMap;
 use rustc_middle::ty;
 #[cfg(feature="master")]
 use rustc_middle::middle::codegen_fn_attrs::CodegenFnAttrFlags;
-use rustc_session::Session;
 use rustc_span::symbol::sym;
-use smallvec::{smallvec, SmallVec};
 
 use crate::{context::CodegenCx, errors::TiedTargetFeatures};
-
-// Given a map from target_features to whether they are enabled or disabled,
-// ensure only valid combinations are allowed.
-pub fn check_tied_features(sess: &Session, features: &FxHashMap<&str, bool>) -> Option<&'static [&'static str]> {
-    for tied in tied_target_features(sess) {
-        // Tied features must be set to the same value, or not set at all
-        let mut tied_iter = tied.iter();
-        let enabled = features.get(tied_iter.next().unwrap());
-        if tied_iter.any(|feature| enabled != features.get(feature)) {
-            return Some(tied);
-        }
-    }
-    None
-}
-
-// TODO(antoyo): maybe move to a new module gcc_util.
-// To find a list of GCC's names, check https://gcc.gnu.org/onlinedocs/gcc/Function-Attributes.html
-fn to_gcc_features<'a>(sess: &Session, s: &'a str) -> SmallVec<[&'a str; 2]> {
-    let arch = if sess.target.arch == "x86_64" { "x86" } else { &*sess.target.arch };
-    match (arch, s) {
-        ("x86", "sse4.2") => smallvec!["sse4.2", "crc32"],
-        ("x86", "pclmulqdq") => smallvec!["pclmul"],
-        ("x86", "rdrand") => smallvec!["rdrnd"],
-        ("x86", "bmi1") => smallvec!["bmi"],
-        ("x86", "cmpxchg16b") => smallvec!["cx16"],
-        ("x86", "avx512vaes") => smallvec!["vaes"],
-        ("x86", "avx512gfni") => smallvec!["gfni"],
-        ("x86", "avx512vpclmulqdq") => smallvec!["vpclmulqdq"],
-        // NOTE: seems like GCC requires 'avx512bw' for 'avx512vbmi2'.
-        ("x86", "avx512vbmi2") => smallvec!["avx512vbmi2", "avx512bw"],
-        // NOTE: seems like GCC requires 'avx512bw' for 'avx512bitalg'.
-        ("x86", "avx512bitalg") => smallvec!["avx512bitalg", "avx512bw"],
-        ("aarch64", "rcpc2") => smallvec!["rcpc-immo"],
-        ("aarch64", "dpb") => smallvec!["ccpp"],
-        ("aarch64", "dpb2") => smallvec!["ccdp"],
-        ("aarch64", "frintts") => smallvec!["fptoint"],
-        ("aarch64", "fcma") => smallvec!["complxnum"],
-        ("aarch64", "pmuv3") => smallvec!["perfmon"],
-        ("aarch64", "paca") => smallvec!["pauth"],
-        ("aarch64", "pacg") => smallvec!["pauth"],
-        // Rust ties fp and neon together. In LLVM neon implicitly enables fp,
-        // but we manually enable neon when a feature only implicitly enables fp
-        ("aarch64", "f32mm") => smallvec!["f32mm", "neon"],
-        ("aarch64", "f64mm") => smallvec!["f64mm", "neon"],
-        ("aarch64", "fhm") => smallvec!["fp16fml", "neon"],
-        ("aarch64", "fp16") => smallvec!["fullfp16", "neon"],
-        ("aarch64", "jsconv") => smallvec!["jsconv", "neon"],
-        ("aarch64", "sve") => smallvec!["sve", "neon"],
-        ("aarch64", "sve2") => smallvec!["sve2", "neon"],
-        ("aarch64", "sve2-aes") => smallvec!["sve2-aes", "neon"],
-        ("aarch64", "sve2-sm4") => smallvec!["sve2-sm4", "neon"],
-        ("aarch64", "sve2-sha3") => smallvec!["sve2-sha3", "neon"],
-        ("aarch64", "sve2-bitperm") => smallvec!["sve2-bitperm", "neon"],
-        (_, s) => smallvec![s],
-    }
-}
+use crate::gcc_util::{check_tied_features, to_gcc_features};
 
 /// Get GCC attribute for the provided inline heuristic.
 #[cfg(feature="master")]
@@ -153,11 +94,33 @@ pub fn from_fn_attrs<'gcc, 'tcx>(
         }))
         .collect::<Vec<_>>();
 
-    // TODO(antoyo): check if we really need global backend features. (Maybe they could be applied
-    // globally?)
+    // TODO(antoyo): cg_llvm adds global features to each function so that LTO keep them.
+    // Check if GCC requires the same.
     let mut global_features = cx.tcx.global_backend_features(()).iter().map(|s| s.as_str());
     function_features.extend(&mut global_features);
-    let target_features = function_features.join(",");
+    let target_features = function_features
+        .iter()
+        .filter_map(|feature| {
+            // FIXME(antoyo): for some reasons, disabling SSE results in the following error when
+            // compiling Rust for Linux:
+            // SSE register return with SSE disabled
+            // TODO(antoyo): support soft-float and retpoline-external-thunk.
+            if feature.contains("soft-float") || feature.contains("retpoline-external-thunk") || *feature == "-sse" {
+                return None;
+            }
+
+            if feature.starts_with('-') {
+                Some(format!("no{}", feature))
+            }
+            else if feature.starts_with('+') {
+                Some(feature[1..].to_string())
+            }
+            else {
+                Some(feature.to_string())
+            }
+        })
+        .collect::<Vec<_>>()
+        .join(",");
     if !target_features.is_empty() {
         #[cfg(feature="master")]
         func.add_attribute(FnAttribute::Target(&target_features));

src/base.rs

@@ -1,15 +1,12 @@
 use std::collections::HashSet;
 use std::env;
-use std::sync::Arc;
 use std::time::Instant;
 
 use gccjit::{
     Context,
     FunctionType,
     GlobalKind,
 };
-#[cfg(feature="master")]
-use gccjit::TargetInfo;
 use rustc_middle::dep_graph;
 use rustc_middle::ty::TyCtxt;
 #[cfg(feature="master")]
@@ -22,8 +19,7 @@ use rustc_codegen_ssa::traits::DebugInfoMethods;
 use rustc_session::config::DebugInfo;
 use rustc_span::Symbol;
 
-#[cfg(not(feature="master"))]
-use crate::TargetInfo;
+use crate::{LockedTargetInfo, gcc_util};
 use crate::GccContext;
 use crate::builder::Builder;
 use crate::context::CodegenCx;
@@ -70,7 +66,7 @@ pub fn linkage_to_gcc(linkage: Linkage) -> FunctionType {
     }
 }
 
-pub fn compile_codegen_unit(tcx: TyCtxt<'_>, cgu_name: Symbol, target_info: Arc<TargetInfo>) -> (ModuleCodegen<GccContext>, u64) {
+pub fn compile_codegen_unit(tcx: TyCtxt<'_>, cgu_name: Symbol, target_info: LockedTargetInfo) -> (ModuleCodegen<GccContext>, u64) {
     let prof_timer = tcx.prof.generic_activity("codegen_module");
     let start_time = Instant::now();
 
@@ -89,7 +85,7 @@ pub fn compile_codegen_unit(tcx: TyCtxt<'_>, cgu_name: Symbol, target_info: Arc<
     // the time we needed for codegenning it.
     let cost = time_to_codegen.as_secs() * 1_000_000_000 + time_to_codegen.subsec_nanos() as u64;
 
-    fn module_codegen(tcx: TyCtxt<'_>, (cgu_name, target_info): (Symbol, Arc<TargetInfo>)) -> ModuleCodegen<GccContext> {
+    fn module_codegen(tcx: TyCtxt<'_>, (cgu_name, target_info): (Symbol, LockedTargetInfo)) -> ModuleCodegen<GccContext> {
         let cgu = tcx.codegen_unit(cgu_name);
         // Instantiate monomorphizations without filling out definitions yet...
         let context = Context::default();
@@ -102,32 +98,16 @@ pub fn compile_codegen_unit(tcx: TyCtxt<'_>, cgu_name: Symbol, target_info: Arc<
             .map(|string| &string[1..])
             .collect();
 
-        let add_cpu_feature_flag = |feature: &str| {
-            // FIXME(antoyo): some tests cause a segfault in GCC when not enabling all these
-            // features.
-            if (true || target_info.cpu_supports(feature)) && !disabled_features.contains(feature) {
-                context.add_command_line_option(&format!("-m{}", feature));
-            }
-        };
-
         // TODO(antoyo): only set on x86 platforms.
         context.add_command_line_option("-masm=intel");
 
-        let features = ["sse2", "avx", "avx2", "sha", "fma", "gfni", "f16c", "aes", "bmi2", "rtm",
-            "vaes", "vpclmulqdq", "xsavec",
-        ];
-
-        for feature in &features {
-            add_cpu_feature_flag(feature);
+        if !disabled_features.contains("avx") {
+            // NOTE: we always enable AVX because the equivalent of llvm.x86.sse2.cmp.pd in GCC for
+            // SSE2 is multiple builtins, so we use the AVX __builtin_ia32_cmppd instead.
+            // FIXME(antoyo): use the proper builtins for llvm.x86.sse2.cmp.pd and similar.
+            context.add_command_line_option("-mavx");
         }
 
-        // TODO(antoyo): only add the following cli arguments if the feature is supported.
-        context.add_command_line_option("-mpclmul");
-        context.add_command_line_option("-mfma4");
-        context.add_command_line_option("-m64");
-        context.add_command_line_option("-mbmi");
-        //context.add_command_line_option("-mavxvnni"); // The CI doesn't support this option.
-
         for arg in &tcx.sess.opts.cg.llvm_args {
             context.add_command_line_option(arg);
         }
@@ -145,6 +125,11 @@ pub fn compile_codegen_unit(tcx: TyCtxt<'_>, cgu_name: Symbol, target_info: Arc<
             context.add_command_line_option("-fno-pie");
         }
 
+        let target_cpu = gcc_util::target_cpu(tcx.sess);
+        if target_cpu != "generic" {
+            context.add_command_line_option(&format!("-march={}", target_cpu));
+        }
+
         if tcx.sess.opts.unstable_opts.function_sections.unwrap_or(tcx.sess.target.function_sections) {
             context.add_command_line_option("-ffunction-sections");
             context.add_command_line_option("-fdata-sections");

src/errors.rs

@@ -1,8 +1,36 @@
-use rustc_errors::{DiagnosticArgValue, IntoDiagnosticArg};
-use rustc_macros::Diagnostic;
+use rustc_errors::{
+    DiagnosticArgValue, DiagnosticBuilder, ErrorGuaranteed, Handler, IntoDiagnostic, IntoDiagnosticArg,
+};
+use rustc_macros::{Diagnostic, Subdiagnostic};
 use rustc_span::Span;
 use std::borrow::Cow;
 
+use crate::fluent_generated as fluent;
+
+#[derive(Diagnostic)]
+#[diag(codegen_gcc_unknown_ctarget_feature_prefix)]
+#[note]
+pub(crate) struct UnknownCTargetFeaturePrefix<'a> {
+    pub feature: &'a str,
+}
+
+#[derive(Diagnostic)]
+#[diag(codegen_gcc_unknown_ctarget_feature)]
+#[note]
+pub(crate) struct UnknownCTargetFeature<'a> {
+    pub feature: &'a str,
+    #[subdiagnostic]
+    pub rust_feature: PossibleFeature<'a>,
+}
+
+#[derive(Subdiagnostic)]
+pub(crate) enum PossibleFeature<'a> {
+    #[help(codegen_gcc_possible_feature)]
+    Some { rust_feature: &'a str },
+    #[help(codegen_gcc_consider_filing_feature_request)]
+    None,
+}
+
 struct ExitCode(Option<i32>);
 
 impl IntoDiagnosticArg for ExitCode {
@@ -71,3 +99,27 @@ pub(crate) struct LtoDylib;
 pub(crate) struct LtoBitcodeFromRlib {
     pub gcc_err: String,
 }
+
+pub(crate) struct TargetFeatureDisableOrEnable<'a> {
+    pub features: &'a [&'a str],
+    pub span: Option<Span>,
+    pub missing_features: Option<MissingFeatures>,
+}
+
+#[derive(Subdiagnostic)]
+#[help(codegen_gcc_missing_features)]
+pub(crate) struct MissingFeatures;
+
+impl IntoDiagnostic<'_, ErrorGuaranteed> for TargetFeatureDisableOrEnable<'_> {
+    fn into_diagnostic(self, sess: &'_ Handler) -> DiagnosticBuilder<'_, ErrorGuaranteed> {
+        let mut diag = sess.struct_err(fluent::codegen_gcc_target_feature_disable_or_enable);
+        if let Some(span) = self.span {
+            diag.set_span(span);
+        };
+        if let Some(missing_features) = self.missing_features {
+            diag.subdiagnostic(missing_features);
+        }
+        diag.set_arg("features", self.features.join(", "));
+        diag
+    }
+}