Simonas Kazlauskas 72fb4379d5 Adjust -Ctarget-cpu=native handling in cg_llvm
When cg_llvm encounters the `-Ctarget-cpu=native` it computes an
explciit set of features that applies to the target in order to
correctly compile code for the host CPU (because e.g. `skylake` alone is
not sufficient to tell if some of the instructions are available or
not).

However there were a couple of issues with how we did this. Firstly, the
order in which features were overriden wasn't quite right – conceptually
you'd expect `-Ctarget-cpu=native` option to override the features that
are implicitly set by the target definition. However due to how other
`-Ctarget-cpu` values are handled we must adopt the following order
of priority:

* Features from -Ctarget-cpu=*; are overriden by
* Features implied by --target; are overriden by
* Features from -Ctarget-feature; are overriden by
* function specific features.

Another problem was in that the function level `target-features`
attribute would overwrite the entire set of the globally enabled
features, rather than just the features the
`#[target_feature(enable/disable)]` specified. With something like
`-Ctarget-cpu=native` we'd end up in a situation wherein a function
without `#[target_feature(enable)]` annotation would have a broader
set of features compared to a function with one such attribute. This
turned out to be a cause of heavy run-time regressions in some code
using these function-level attributes in conjunction with
`-Ctarget-cpu=native`, for example.

With this PR rustc is more careful about specifying the entire set of
features for functions that use `#[target_feature(enable/disable)]` or
`#[instruction_set]` attributes.

Sadly testing the original reproducer for this behaviour is quite
impossible – we cannot rely on `-Ctarget-cpu=native` to be anything in
particular on developer or CI machines.
2021-03-16 21:32:55 +02:00

374 lines
14 KiB
Rust

//! Set and unset common attributes on LLVM values.
use std::ffi::CString;
use cstr::cstr;
use rustc_codegen_ssa::traits::*;
use rustc_data_structures::fx::FxHashMap;
use rustc_data_structures::small_c_str::SmallCStr;
use rustc_hir::def_id::DefId;
use rustc_middle::middle::codegen_fn_attrs::CodegenFnAttrFlags;
use rustc_middle::ty::layout::HasTyCtxt;
use rustc_middle::ty::query::Providers;
use rustc_middle::ty::{self, TyCtxt};
use rustc_session::config::{OptLevel, SanitizerSet};
use rustc_session::Session;
use rustc_target::spec::StackProbeType;
use crate::attributes;
use crate::llvm::AttributePlace::Function;
use crate::llvm::{self, Attribute};
use crate::llvm_util;
pub use rustc_attr::{InlineAttr, InstructionSetAttr, OptimizeAttr};
use crate::context::CodegenCx;
use crate::value::Value;
/// Mark LLVM function to use provided inline heuristic.
#[inline]
fn inline(cx: &CodegenCx<'ll, '_>, val: &'ll Value, inline: InlineAttr) {
use self::InlineAttr::*;
match inline {
Hint => Attribute::InlineHint.apply_llfn(Function, val),
Always => Attribute::AlwaysInline.apply_llfn(Function, val),
Never => {
if cx.tcx().sess.target.arch != "amdgpu" {
Attribute::NoInline.apply_llfn(Function, val);
}
}
None => {}
};
}
/// Apply LLVM sanitize attributes.
#[inline]
pub fn sanitize(cx: &CodegenCx<'ll, '_>, no_sanitize: SanitizerSet, llfn: &'ll Value) {
let enabled = cx.tcx.sess.opts.debugging_opts.sanitizer - no_sanitize;
if enabled.contains(SanitizerSet::ADDRESS) {
llvm::Attribute::SanitizeAddress.apply_llfn(Function, llfn);
}
if enabled.contains(SanitizerSet::MEMORY) {
llvm::Attribute::SanitizeMemory.apply_llfn(Function, llfn);
}
if enabled.contains(SanitizerSet::THREAD) {
llvm::Attribute::SanitizeThread.apply_llfn(Function, llfn);
}
if enabled.contains(SanitizerSet::HWADDRESS) {
llvm::Attribute::SanitizeHWAddress.apply_llfn(Function, llfn);
}
}
/// Tell LLVM to emit or not emit the information necessary to unwind the stack for the function.
#[inline]
pub fn emit_uwtable(val: &'ll Value, emit: bool) {
Attribute::UWTable.toggle_llfn(Function, val, emit);
}
/// Tell LLVM if this function should be 'naked', i.e., skip the epilogue and prologue.
#[inline]
fn naked(val: &'ll Value, is_naked: bool) {
Attribute::Naked.toggle_llfn(Function, val, is_naked);
}
pub fn set_frame_pointer_elimination(cx: &CodegenCx<'ll, '_>, llfn: &'ll Value) {
if cx.sess().must_not_eliminate_frame_pointers() {
llvm::AddFunctionAttrStringValue(
llfn,
llvm::AttributePlace::Function,
cstr!("frame-pointer"),
cstr!("all"),
);
}
}
/// Tell LLVM what instrument function to insert.
#[inline]
fn set_instrument_function(cx: &CodegenCx<'ll, '_>, llfn: &'ll Value) {
if cx.sess().instrument_mcount() {
// Similar to `clang -pg` behavior. Handled by the
// `post-inline-ee-instrument` LLVM pass.
// The function name varies on platforms.
// See test/CodeGen/mcount.c in clang.
let mcount_name = CString::new(cx.sess().target.mcount.as_str().as_bytes()).unwrap();
llvm::AddFunctionAttrStringValue(
llfn,
llvm::AttributePlace::Function,
cstr!("instrument-function-entry-inlined"),
&mcount_name,
);
}
}
fn set_probestack(cx: &CodegenCx<'ll, '_>, llfn: &'ll Value) {
// Currently stack probes seem somewhat incompatible with the address
// sanitizer and thread sanitizer. With asan we're already protected from
// stack overflow anyway so we don't really need stack probes regardless.
if cx
.sess()
.opts
.debugging_opts
.sanitizer
.intersects(SanitizerSet::ADDRESS | SanitizerSet::THREAD)
{
return;
}
// probestack doesn't play nice either with `-C profile-generate`.
if cx.sess().opts.cg.profile_generate.enabled() {
return;
}
// probestack doesn't play nice either with gcov profiling.
if cx.sess().opts.debugging_opts.profile {
return;
}
let attr_value = match cx.sess().target.stack_probes {
StackProbeType::None => None,
// Request LLVM to generate the probes inline. If the given LLVM version does not support
// this, no probe is generated at all (even if the attribute is specified).
StackProbeType::Inline => Some(cstr!("inline-asm")),
// Flag our internal `__rust_probestack` function as the stack probe symbol.
// This is defined in the `compiler-builtins` crate for each architecture.
StackProbeType::Call => Some(cstr!("__rust_probestack")),
// Pick from the two above based on the LLVM version.
StackProbeType::InlineOrCall { min_llvm_version_for_inline } => {
if llvm_util::get_version() < min_llvm_version_for_inline {
Some(cstr!("__rust_probestack"))
} else {
Some(cstr!("inline-asm"))
}
}
};
if let Some(attr_value) = attr_value {
llvm::AddFunctionAttrStringValue(
llfn,
llvm::AttributePlace::Function,
cstr!("probe-stack"),
attr_value,
);
}
}
pub fn apply_target_cpu_attr(cx: &CodegenCx<'ll, '_>, llfn: &'ll Value) {
let target_cpu = SmallCStr::new(llvm_util::target_cpu(cx.tcx.sess));
llvm::AddFunctionAttrStringValue(
llfn,
llvm::AttributePlace::Function,
cstr!("target-cpu"),
target_cpu.as_c_str(),
);
}
pub fn apply_tune_cpu_attr(cx: &CodegenCx<'ll, '_>, llfn: &'ll Value) {
if let Some(tune) = llvm_util::tune_cpu(cx.tcx.sess) {
let tune_cpu = SmallCStr::new(tune);
llvm::AddFunctionAttrStringValue(
llfn,
llvm::AttributePlace::Function,
cstr!("tune-cpu"),
tune_cpu.as_c_str(),
);
}
}
/// Sets the `NonLazyBind` LLVM attribute on a given function,
/// assuming the codegen options allow skipping the PLT.
pub fn non_lazy_bind(sess: &Session, llfn: &'ll Value) {
// Don't generate calls through PLT if it's not necessary
if !sess.needs_plt() {
Attribute::NonLazyBind.apply_llfn(Function, llfn);
}
}
pub(crate) fn default_optimisation_attrs(sess: &Session, llfn: &'ll Value) {
match sess.opts.optimize {
OptLevel::Size => {
llvm::Attribute::MinSize.unapply_llfn(Function, llfn);
llvm::Attribute::OptimizeForSize.apply_llfn(Function, llfn);
llvm::Attribute::OptimizeNone.unapply_llfn(Function, llfn);
}
OptLevel::SizeMin => {
llvm::Attribute::MinSize.apply_llfn(Function, llfn);
llvm::Attribute::OptimizeForSize.apply_llfn(Function, llfn);
llvm::Attribute::OptimizeNone.unapply_llfn(Function, llfn);
}
OptLevel::No => {
llvm::Attribute::MinSize.unapply_llfn(Function, llfn);
llvm::Attribute::OptimizeForSize.unapply_llfn(Function, llfn);
llvm::Attribute::OptimizeNone.unapply_llfn(Function, llfn);
}
_ => {}
}
}
/// Composite function which sets LLVM attributes for function depending on its AST (`#[attribute]`)
/// attributes.
pub fn from_fn_attrs(cx: &CodegenCx<'ll, 'tcx>, llfn: &'ll Value, instance: ty::Instance<'tcx>) {
let codegen_fn_attrs = cx.tcx.codegen_fn_attrs(instance.def_id());
match codegen_fn_attrs.optimize {
OptimizeAttr::None => {
default_optimisation_attrs(cx.tcx.sess, llfn);
}
OptimizeAttr::Speed => {
llvm::Attribute::MinSize.unapply_llfn(Function, llfn);
llvm::Attribute::OptimizeForSize.unapply_llfn(Function, llfn);
llvm::Attribute::OptimizeNone.unapply_llfn(Function, llfn);
}
OptimizeAttr::Size => {
llvm::Attribute::MinSize.apply_llfn(Function, llfn);
llvm::Attribute::OptimizeForSize.apply_llfn(Function, llfn);
llvm::Attribute::OptimizeNone.unapply_llfn(Function, llfn);
}
}
let inline_attr = if codegen_fn_attrs.flags.contains(CodegenFnAttrFlags::NAKED) {
InlineAttr::Never
} else if codegen_fn_attrs.inline == InlineAttr::None && instance.def.requires_inline(cx.tcx) {
InlineAttr::Hint
} else {
codegen_fn_attrs.inline
};
inline(cx, llfn, inline_attr);
// The `uwtable` attribute according to LLVM is:
//
// This attribute indicates that the ABI being targeted requires that an
// unwind table entry be produced for this function even if we can show
// that no exceptions passes by it. This is normally the case for the
// ELF x86-64 abi, but it can be disabled for some compilation units.
//
// Typically when we're compiling with `-C panic=abort` (which implies this
// `no_landing_pads` check) we don't need `uwtable` because we can't
// generate any exceptions! On Windows, however, exceptions include other
// events such as illegal instructions, segfaults, etc. This means that on
// Windows we end up still needing the `uwtable` attribute even if the `-C
// panic=abort` flag is passed.
//
// You can also find more info on why Windows always requires uwtables here:
// https://bugzilla.mozilla.org/show_bug.cgi?id=1302078
if cx.sess().must_emit_unwind_tables() {
attributes::emit_uwtable(llfn, true);
}
set_frame_pointer_elimination(cx, llfn);
set_instrument_function(cx, llfn);
set_probestack(cx, llfn);
if codegen_fn_attrs.flags.contains(CodegenFnAttrFlags::COLD) {
Attribute::Cold.apply_llfn(Function, llfn);
}
if codegen_fn_attrs.flags.contains(CodegenFnAttrFlags::FFI_RETURNS_TWICE) {
Attribute::ReturnsTwice.apply_llfn(Function, llfn);
}
if codegen_fn_attrs.flags.contains(CodegenFnAttrFlags::FFI_PURE) {
Attribute::ReadOnly.apply_llfn(Function, llfn);
}
if codegen_fn_attrs.flags.contains(CodegenFnAttrFlags::FFI_CONST) {
Attribute::ReadNone.apply_llfn(Function, llfn);
}
if codegen_fn_attrs.flags.contains(CodegenFnAttrFlags::NAKED) {
naked(llfn, true);
}
if codegen_fn_attrs.flags.contains(CodegenFnAttrFlags::ALLOCATOR) {
Attribute::NoAlias.apply_llfn(llvm::AttributePlace::ReturnValue, llfn);
}
if codegen_fn_attrs.flags.contains(CodegenFnAttrFlags::CMSE_NONSECURE_ENTRY) {
llvm::AddFunctionAttrString(llfn, Function, cstr!("cmse_nonsecure_entry"));
}
sanitize(cx, codegen_fn_attrs.no_sanitize, llfn);
// Always annotate functions with the target-cpu they are compiled for.
// Without this, ThinLTO won't inline Rust functions into Clang generated
// functions (because Clang annotates functions this way too).
apply_target_cpu_attr(cx, llfn);
// tune-cpu is only conveyed through the attribute for our purpose.
// The target doesn't care; the subtarget reads our attribute.
apply_tune_cpu_attr(cx, llfn);
let function_features = codegen_fn_attrs
.target_features
.iter()
.map(|f| {
let feature = &f.as_str();
format!("+{}", llvm_util::to_llvm_feature(cx.tcx.sess, feature))
})
.chain(codegen_fn_attrs.instruction_set.iter().map(|x| match x {
InstructionSetAttr::ArmA32 => "-thumb-mode".to_string(),
InstructionSetAttr::ArmT32 => "+thumb-mode".to_string(),
}))
.collect::<Vec<String>>();
if !function_features.is_empty() {
let mut global_features = llvm_util::llvm_global_features(cx.tcx.sess);
global_features.extend(function_features.into_iter());
let features = global_features.join(",");
let val = CString::new(features).unwrap();
llvm::AddFunctionAttrStringValue(
llfn,
llvm::AttributePlace::Function,
cstr!("target-features"),
&val,
);
}
// Note that currently the `wasm-import-module` doesn't do anything, but
// eventually LLVM 7 should read this and ferry the appropriate import
// module to the output file.
if cx.tcx.sess.target.arch == "wasm32" {
if let Some(module) = wasm_import_module(cx.tcx, instance.def_id()) {
llvm::AddFunctionAttrStringValue(
llfn,
llvm::AttributePlace::Function,
cstr!("wasm-import-module"),
&module,
);
let name =
codegen_fn_attrs.link_name.unwrap_or_else(|| cx.tcx.item_name(instance.def_id()));
let name = CString::new(&name.as_str()[..]).unwrap();
llvm::AddFunctionAttrStringValue(
llfn,
llvm::AttributePlace::Function,
cstr!("wasm-import-name"),
&name,
);
}
}
}
pub fn provide_both(providers: &mut Providers) {
providers.wasm_import_module_map = |tcx, cnum| {
// Build up a map from DefId to a `NativeLib` structure, where
// `NativeLib` internally contains information about
// `#[link(wasm_import_module = "...")]` for example.
let native_libs = tcx.native_libraries(cnum);
let def_id_to_native_lib = native_libs
.iter()
.filter_map(|lib| lib.foreign_module.map(|id| (id, lib)))
.collect::<FxHashMap<_, _>>();
let mut ret = FxHashMap::default();
for (def_id, lib) in tcx.foreign_modules(cnum).iter() {
let module = def_id_to_native_lib.get(&def_id).and_then(|s| s.wasm_import_module);
let module = match module {
Some(s) => s,
None => continue,
};
ret.extend(lib.foreign_items.iter().map(|id| {
assert_eq!(id.krate, cnum);
(*id, module.to_string())
}));
}
ret
};
}
fn wasm_import_module(tcx: TyCtxt<'_>, id: DefId) -> Option<CString> {
tcx.wasm_import_module_map(id.krate).get(&id).map(|s| CString::new(&s[..]).unwrap())
}