diff options
Diffstat (limited to 'compiler/rustc_codegen_llvm/src')
35 files changed, 2049 insertions, 1488 deletions
diff --git a/compiler/rustc_codegen_llvm/src/abi.rs b/compiler/rustc_codegen_llvm/src/abi.rs index cc8b3a1a4e4..a6fd2a7de6b 100644 --- a/compiler/rustc_codegen_llvm/src/abi.rs +++ b/compiler/rustc_codegen_llvm/src/abi.rs @@ -19,6 +19,7 @@ use rustc_target::abi::call::ArgAbi; pub use rustc_target::abi::call::*; use rustc_target::abi::{self, HasDataLayout, Int}; pub use rustc_target::spec::abi::Abi; +use rustc_target::spec::SanitizerSet; use libc::c_uint; use smallvec::SmallVec; @@ -37,7 +38,7 @@ fn should_use_mutable_noalias(cx: &CodegenCx<'_, '_>) -> bool { // LLVM prior to version 12 had known miscompiles in the presence of // noalias attributes (see #54878), but we don't support earlier // versions at all anymore. We now enable mutable noalias by default. - cx.tcx.sess.opts.debugging_opts.mutable_noalias.unwrap_or(true) + cx.tcx.sess.opts.unstable_opts.mutable_noalias.unwrap_or(true) } const ABI_AFFECTING_ATTRIBUTES: [(ArgAttribute, llvm::AttributeKind); 1] = @@ -90,6 +91,13 @@ fn get_attrs<'ll>(this: &ArgAttributes, cx: &CodegenCx<'ll, '_>) -> SmallVec<[&' if regular.contains(ArgAttribute::NoAliasMutRef) && should_use_mutable_noalias(cx) { attrs.push(llvm::AttributeKind::NoAlias.create_attr(cx.llcx)); } + } else if cx.tcx.sess.opts.unstable_opts.sanitizer.contains(SanitizerSet::MEMORY) { + // If we're not optimising, *but* memory sanitizer is on, emit noundef, since it affects + // memory sanitizer's behavior. + + if regular.contains(ArgAttribute::NoUndef) { + attrs.push(llvm::AttributeKind::NoUndef.create_attr(cx.llcx)); + } } attrs @@ -213,7 +221,7 @@ impl<'ll, 'tcx> ArgAbiExt<'ll, 'tcx> for ArgAbi<'tcx, Ty<'tcx>> { OperandValue::Ref(val, None, self.layout.align.abi).store(bx, dst) } else if self.is_unsized_indirect() { bug!("unsized `ArgAbi` must be handled through `store_fn_arg`"); - } else if let PassMode::Cast(cast) = self.mode { + } else if let PassMode::Cast(cast, _) = &self.mode { // FIXME(eddyb): Figure out when the simpler Store is safe, clang // uses it for i16 -> {i8, i8}, but not for i24 -> {i8, i8, i8}. let can_store_through_cast_ptr = false; @@ -283,7 +291,7 @@ impl<'ll, 'tcx> ArgAbiExt<'ll, 'tcx> for ArgAbi<'tcx, Ty<'tcx>> { } PassMode::Direct(_) | PassMode::Indirect { attrs: _, extra_attrs: None, on_stack: _ } - | PassMode::Cast(_) => { + | PassMode::Cast(..) => { let next_arg = next(); self.store(bx, next_arg, dst); } @@ -325,20 +333,18 @@ impl<'ll, 'tcx> FnAbiLlvmExt<'ll, 'tcx> for FnAbi<'tcx, Ty<'tcx>> { fn llvm_type(&self, cx: &CodegenCx<'ll, 'tcx>) -> &'ll Type { // Ignore "extra" args from the call site for C variadic functions. // Only the "fixed" args are part of the LLVM function signature. - let args = if self.c_variadic { &self.args[..self.fixed_count] } else { &self.args }; + let args = + if self.c_variadic { &self.args[..self.fixed_count as usize] } else { &self.args }; - let args_capacity: usize = args.iter().map(|arg| - if arg.pad.is_some() { 1 } else { 0 } + - if let PassMode::Pair(_, _) = arg.mode { 2 } else { 1 } - ).sum(); + // This capacity calculation is approximate. let mut llargument_tys = Vec::with_capacity( - if let PassMode::Indirect { .. } = self.ret.mode { 1 } else { 0 } + args_capacity, + self.args.len() + if let PassMode::Indirect { .. } = self.ret.mode { 1 } else { 0 }, ); - let llreturn_ty = match self.ret.mode { + let llreturn_ty = match &self.ret.mode { PassMode::Ignore => cx.type_void(), PassMode::Direct(_) | PassMode::Pair(..) => self.ret.layout.immediate_llvm_type(cx), - PassMode::Cast(cast) => cast.llvm_type(cx), + PassMode::Cast(cast, _) => cast.llvm_type(cx), PassMode::Indirect { .. } => { llargument_tys.push(cx.type_ptr_to(self.ret.memory_ty(cx))); cx.type_void() @@ -346,12 +352,7 @@ impl<'ll, 'tcx> FnAbiLlvmExt<'ll, 'tcx> for FnAbi<'tcx, Ty<'tcx>> { }; for arg in args { - // add padding - if let Some(ty) = arg.pad { - llargument_tys.push(ty.llvm_type(cx)); - } - - let llarg_ty = match arg.mode { + let llarg_ty = match &arg.mode { PassMode::Ignore => continue, PassMode::Direct(_) => arg.layout.immediate_llvm_type(cx), PassMode::Pair(..) => { @@ -366,7 +367,13 @@ impl<'ll, 'tcx> FnAbiLlvmExt<'ll, 'tcx> for FnAbi<'tcx, Ty<'tcx>> { llargument_tys.push(ptr_layout.scalar_pair_element_llvm_type(cx, 1, true)); continue; } - PassMode::Cast(cast) => cast.llvm_type(cx), + PassMode::Cast(cast, pad_i32) => { + // add padding + if *pad_i32 { + llargument_tys.push(Reg::i32().llvm_type(cx)); + } + cast.llvm_type(cx) + } PassMode::Indirect { attrs: _, extra_attrs: None, on_stack: _ } => { cx.type_ptr_to(arg.memory_ty(cx)) } @@ -391,23 +398,7 @@ impl<'ll, 'tcx> FnAbiLlvmExt<'ll, 'tcx> for FnAbi<'tcx, Ty<'tcx>> { } fn llvm_cconv(&self) -> llvm::CallConv { - match self.conv { - Conv::C | Conv::Rust | Conv::CCmseNonSecureCall => llvm::CCallConv, - Conv::RustCold => llvm::ColdCallConv, - Conv::AmdGpuKernel => llvm::AmdGpuKernel, - Conv::AvrInterrupt => llvm::AvrInterrupt, - Conv::AvrNonBlockingInterrupt => llvm::AvrNonBlockingInterrupt, - Conv::ArmAapcs => llvm::ArmAapcsCallConv, - Conv::Msp430Intr => llvm::Msp430Intr, - Conv::PtxKernel => llvm::PtxKernel, - Conv::X86Fastcall => llvm::X86FastcallCallConv, - Conv::X86Intr => llvm::X86_Intr, - Conv::X86Stdcall => llvm::X86StdcallCallConv, - Conv::X86ThisCall => llvm::X86_ThisCall, - Conv::X86VectorCall => llvm::X86_VectorCall, - Conv::X86_64SysV => llvm::X86_64_SysV, - Conv::X86_64Win64 => llvm::X86_64_Win64, - } + self.conv.into() } fn apply_attrs_llfn(&self, cx: &CodegenCx<'ll, 'tcx>, llfn: &'ll Value) { @@ -426,46 +417,46 @@ impl<'ll, 'tcx> FnAbiLlvmExt<'ll, 'tcx> for FnAbi<'tcx, Ty<'tcx>> { i += 1; i - 1 }; - match self.ret.mode { - PassMode::Direct(ref attrs) => { + match &self.ret.mode { + PassMode::Direct(attrs) => { attrs.apply_attrs_to_llfn(llvm::AttributePlace::ReturnValue, cx, llfn); } - PassMode::Indirect { ref attrs, extra_attrs: _, on_stack } => { + PassMode::Indirect { attrs, extra_attrs: _, on_stack } => { assert!(!on_stack); let i = apply(attrs); let sret = llvm::CreateStructRetAttr(cx.llcx, self.ret.layout.llvm_type(cx)); attributes::apply_to_llfn(llfn, llvm::AttributePlace::Argument(i), &[sret]); } - PassMode::Cast(cast) => { + PassMode::Cast(cast, _) => { cast.attrs.apply_attrs_to_llfn(llvm::AttributePlace::ReturnValue, cx, llfn); } _ => {} } - for arg in &self.args { - if arg.pad.is_some() { - apply(&ArgAttributes::new()); - } - match arg.mode { + for arg in self.args.iter() { + match &arg.mode { PassMode::Ignore => {} - PassMode::Indirect { ref attrs, extra_attrs: None, on_stack: true } => { + PassMode::Indirect { attrs, extra_attrs: None, on_stack: true } => { let i = apply(attrs); let byval = llvm::CreateByValAttr(cx.llcx, arg.layout.llvm_type(cx)); attributes::apply_to_llfn(llfn, llvm::AttributePlace::Argument(i), &[byval]); } - PassMode::Direct(ref attrs) - | PassMode::Indirect { ref attrs, extra_attrs: None, on_stack: false } => { + PassMode::Direct(attrs) + | PassMode::Indirect { attrs, extra_attrs: None, on_stack: false } => { apply(attrs); } - PassMode::Indirect { ref attrs, extra_attrs: Some(ref extra_attrs), on_stack } => { + PassMode::Indirect { attrs, extra_attrs: Some(extra_attrs), on_stack } => { assert!(!on_stack); apply(attrs); apply(extra_attrs); } - PassMode::Pair(ref a, ref b) => { + PassMode::Pair(a, b) => { apply(a); apply(b); } - PassMode::Cast(cast) => { + PassMode::Cast(cast, pad_i32) => { + if *pad_i32 { + apply(&ArgAttributes::new()); + } apply(&cast.attrs); } } @@ -488,17 +479,17 @@ impl<'ll, 'tcx> FnAbiLlvmExt<'ll, 'tcx> for FnAbi<'tcx, Ty<'tcx>> { i += 1; i - 1 }; - match self.ret.mode { - PassMode::Direct(ref attrs) => { + match &self.ret.mode { + PassMode::Direct(attrs) => { attrs.apply_attrs_to_callsite(llvm::AttributePlace::ReturnValue, bx.cx, callsite); } - PassMode::Indirect { ref attrs, extra_attrs: _, on_stack } => { + PassMode::Indirect { attrs, extra_attrs: _, on_stack } => { assert!(!on_stack); let i = apply(bx.cx, attrs); let sret = llvm::CreateStructRetAttr(bx.cx.llcx, self.ret.layout.llvm_type(bx)); attributes::apply_to_callsite(callsite, llvm::AttributePlace::Argument(i), &[sret]); } - PassMode::Cast(cast) => { + PassMode::Cast(cast, _) => { cast.attrs.apply_attrs_to_callsite( llvm::AttributePlace::ReturnValue, &bx.cx, @@ -517,13 +508,10 @@ impl<'ll, 'tcx> FnAbiLlvmExt<'ll, 'tcx> for FnAbi<'tcx, Ty<'tcx>> { } } } - for arg in &self.args { - if arg.pad.is_some() { - apply(bx.cx, &ArgAttributes::new()); - } - match arg.mode { + for arg in self.args.iter() { + match &arg.mode { PassMode::Ignore => {} - PassMode::Indirect { ref attrs, extra_attrs: None, on_stack: true } => { + PassMode::Indirect { attrs, extra_attrs: None, on_stack: true } => { let i = apply(bx.cx, attrs); let byval = llvm::CreateByValAttr(bx.cx.llcx, arg.layout.llvm_type(bx)); attributes::apply_to_callsite( @@ -532,23 +520,22 @@ impl<'ll, 'tcx> FnAbiLlvmExt<'ll, 'tcx> for FnAbi<'tcx, Ty<'tcx>> { &[byval], ); } - PassMode::Direct(ref attrs) - | PassMode::Indirect { ref attrs, extra_attrs: None, on_stack: false } => { + PassMode::Direct(attrs) + | PassMode::Indirect { attrs, extra_attrs: None, on_stack: false } => { apply(bx.cx, attrs); } - PassMode::Indirect { - ref attrs, - extra_attrs: Some(ref extra_attrs), - on_stack: _, - } => { + PassMode::Indirect { attrs, extra_attrs: Some(extra_attrs), on_stack: _ } => { apply(bx.cx, attrs); apply(bx.cx, extra_attrs); } - PassMode::Pair(ref a, ref b) => { + PassMode::Pair(a, b) => { apply(bx.cx, a); apply(bx.cx, b); } - PassMode::Cast(cast) => { + PassMode::Cast(cast, pad_i32) => { + if *pad_i32 { + apply(bx.cx, &ArgAttributes::new()); + } apply(bx.cx, &cast.attrs); } } @@ -569,15 +556,49 @@ impl<'ll, 'tcx> FnAbiLlvmExt<'ll, 'tcx> for FnAbi<'tcx, Ty<'tcx>> { &[cmse_nonsecure_call], ); } + + // Some intrinsics require that an elementtype attribute (with the pointee type of a + // pointer argument) is added to the callsite. + let element_type_index = unsafe { llvm::LLVMRustGetElementTypeArgIndex(callsite) }; + if element_type_index >= 0 { + let arg_ty = self.args[element_type_index as usize].layout.ty; + let pointee_ty = arg_ty.builtin_deref(true).expect("Must be pointer argument").ty; + let element_type_attr = unsafe { + llvm::LLVMRustCreateElementTypeAttr(bx.llcx, bx.layout_of(pointee_ty).llvm_type(bx)) + }; + attributes::apply_to_callsite( + callsite, + llvm::AttributePlace::Argument(element_type_index as u32), + &[element_type_attr], + ); + } } } impl<'tcx> AbiBuilderMethods<'tcx> for Builder<'_, '_, 'tcx> { - fn apply_attrs_callsite(&mut self, fn_abi: &FnAbi<'tcx, Ty<'tcx>>, callsite: Self::Value) { - fn_abi.apply_attrs_callsite(self, callsite) - } - fn get_param(&mut self, index: usize) -> Self::Value { llvm::get_param(self.llfn(), index as c_uint) } } + +impl From<Conv> for llvm::CallConv { + fn from(conv: Conv) -> Self { + match conv { + Conv::C | Conv::Rust | Conv::CCmseNonSecureCall => llvm::CCallConv, + Conv::RustCold => llvm::ColdCallConv, + Conv::AmdGpuKernel => llvm::AmdGpuKernel, + Conv::AvrInterrupt => llvm::AvrInterrupt, + Conv::AvrNonBlockingInterrupt => llvm::AvrNonBlockingInterrupt, + Conv::ArmAapcs => llvm::ArmAapcsCallConv, + Conv::Msp430Intr => llvm::Msp430Intr, + Conv::PtxKernel => llvm::PtxKernel, + Conv::X86Fastcall => llvm::X86FastcallCallConv, + Conv::X86Intr => llvm::X86_Intr, + Conv::X86Stdcall => llvm::X86StdcallCallConv, + Conv::X86ThisCall => llvm::X86_ThisCall, + Conv::X86VectorCall => llvm::X86_VectorCall, + Conv::X86_64SysV => llvm::X86_64_SysV, + Conv::X86_64Win64 => llvm::X86_64_Win64, + } + } +} diff --git a/compiler/rustc_codegen_llvm/src/allocator.rs b/compiler/rustc_codegen_llvm/src/allocator.rs index f935acb1a7e..fed56cdd438 100644 --- a/compiler/rustc_codegen_llvm/src/allocator.rs +++ b/compiler/rustc_codegen_llvm/src/allocator.rs @@ -15,7 +15,7 @@ pub(crate) unsafe fn codegen( module_llvm: &mut ModuleLlvm, module_name: &str, kind: AllocatorKind, - has_alloc_error_handler: bool, + alloc_error_handler_kind: AllocatorKind, ) { let llcx = &*module_llvm.llcx; let llmod = module_llvm.llmod(); @@ -117,8 +117,7 @@ pub(crate) unsafe fn codegen( attributes::apply_to_llfn(llfn, llvm::AttributePlace::Function, &[uwtable]); } - let kind = if has_alloc_error_handler { AllocatorKind::Global } else { AllocatorKind::Default }; - let callee = kind.fn_name(sym::oom); + let callee = alloc_error_handler_kind.fn_name(sym::oom); let callee = llvm::LLVMRustGetOrInsertFunction(llmod, callee.as_ptr().cast(), callee.len(), ty); // -> ! DIFlagNoReturn attributes::apply_to_llfn(callee, llvm::AttributePlace::Function, &[no_return]); @@ -145,7 +144,7 @@ pub(crate) unsafe fn codegen( if tcx.sess.target.default_hidden_visibility { llvm::LLVMRustSetVisibility(ll_g, llvm::Visibility::Hidden); } - let val = tcx.sess.opts.debugging_opts.oom.should_panic(); + let val = tcx.sess.opts.unstable_opts.oom.should_panic(); let llval = llvm::LLVMConstInt(i8, val as u64, False); llvm::LLVMSetInitializer(ll_g, llval); diff --git a/compiler/rustc_codegen_llvm/src/asm.rs b/compiler/rustc_codegen_llvm/src/asm.rs index a53946995ee..219a4f8fa89 100644 --- a/compiler/rustc_codegen_llvm/src/asm.rs +++ b/compiler/rustc_codegen_llvm/src/asm.rs @@ -3,7 +3,6 @@ use crate::builder::Builder; use crate::common::Funclet; use crate::context::CodegenCx; use crate::llvm; -use crate::llvm_util; use crate::type_::Type; use crate::type_of::LayoutLlvmExt; use crate::value::Value; @@ -20,7 +19,6 @@ use rustc_target::asm::*; use libc::{c_char, c_uint}; use smallvec::SmallVec; -use tracing::debug; impl<'ll, 'tcx> AsmBuilderMethods<'tcx> for Builder<'_, 'll, 'tcx> { fn codegen_inline_asm( @@ -132,7 +130,7 @@ impl<'ll, 'tcx> AsmBuilderMethods<'tcx> for Builder<'_, 'll, 'tcx> { op_idx.insert(idx, constraints.len()); constraints.push(reg_to_llvm(reg, Some(&value.layout))); } - InlineAsmOperandRef::InOut { reg, late: _, in_value, out_place: _ } => { + InlineAsmOperandRef::InOut { reg, late, in_value, out_place: _ } => { let value = llvm_fixup_input( self, in_value.immediate(), @@ -140,7 +138,16 @@ impl<'ll, 'tcx> AsmBuilderMethods<'tcx> for Builder<'_, 'll, 'tcx> { &in_value.layout, ); inputs.push(value); - constraints.push(format!("{}", op_idx[&idx])); + + // In the case of fixed registers, we have the choice of + // either using a tied operand or duplicating the constraint. + // We prefer the latter because it matches the behavior of + // Clang. + if late && matches!(reg, InlineAsmRegOrRegClass::Reg(_)) { + constraints.push(format!("{}", reg_to_llvm(reg, Some(&in_value.layout)))); + } else { + constraints.push(format!("{}", op_idx[&idx])); + } } InlineAsmOperandRef::SymFn { instance } => { inputs.push(self.cx.get_fn(instance)); @@ -278,13 +285,13 @@ impl<'ll, 'tcx> AsmBuilderMethods<'tcx> for Builder<'_, 'll, 'tcx> { let mut attrs = SmallVec::<[_; 2]>::new(); if options.contains(InlineAsmOptions::PURE) { if options.contains(InlineAsmOptions::NOMEM) { - attrs.push(llvm::AttributeKind::ReadNone.create_attr(self.cx.llcx)); + attrs.push(llvm::MemoryEffects::None.create_attr(self.cx.llcx)); } else if options.contains(InlineAsmOptions::READONLY) { - attrs.push(llvm::AttributeKind::ReadOnly.create_attr(self.cx.llcx)); + attrs.push(llvm::MemoryEffects::ReadOnly.create_attr(self.cx.llcx)); } attrs.push(llvm::AttributeKind::WillReturn.create_attr(self.cx.llcx)); } else if options.contains(InlineAsmOptions::NOMEM) { - attrs.push(llvm::AttributeKind::InaccessibleMemOnly.create_attr(self.cx.llcx)); + attrs.push(llvm::MemoryEffects::InaccessibleMemOnly.create_attr(self.cx.llcx)); } else { // LLVM doesn't have an attribute to represent ReadOnly + SideEffect } @@ -419,13 +426,6 @@ pub(crate) fn inline_asm_call<'ll>( let constraints_ok = llvm::LLVMRustInlineAsmVerify(fty, cons.as_ptr().cast(), cons.len()); debug!("constraint verification result: {:?}", constraints_ok); if constraints_ok { - if unwind && llvm_util::get_version() < (13, 0, 0) { - bx.cx.sess().span_fatal( - line_spans[0], - "unwinding from inline assembly is only supported on llvm >= 13.", - ); - } - let v = llvm::LLVMRustInlineAsm( fty, asm.as_ptr().cast(), @@ -439,9 +439,9 @@ pub(crate) fn inline_asm_call<'ll>( ); let call = if let Some((dest, catch, funclet)) = dest_catch_funclet { - bx.invoke(fty, v, inputs, dest, catch, funclet) + bx.invoke(fty, None, v, inputs, dest, catch, funclet) } else { - bx.call(fty, v, inputs, None) + bx.call(fty, None, v, inputs, None) }; // Store mark in a metadata node so we can map LLVM errors @@ -505,6 +505,44 @@ fn xmm_reg_index(reg: InlineAsmReg) -> Option<u32> { } } +/// If the register is an AArch64 integer register then return its index. +fn a64_reg_index(reg: InlineAsmReg) -> Option<u32> { + match reg { + InlineAsmReg::AArch64(AArch64InlineAsmReg::x0) => Some(0), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x1) => Some(1), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x2) => Some(2), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x3) => Some(3), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x4) => Some(4), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x5) => Some(5), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x6) => Some(6), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x7) => Some(7), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x8) => Some(8), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x9) => Some(9), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x10) => Some(10), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x11) => Some(11), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x12) => Some(12), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x13) => Some(13), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x14) => Some(14), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x15) => Some(15), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x16) => Some(16), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x17) => Some(17), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x18) => Some(18), + // x19 is reserved + InlineAsmReg::AArch64(AArch64InlineAsmReg::x20) => Some(20), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x21) => Some(21), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x22) => Some(22), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x23) => Some(23), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x24) => Some(24), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x25) => Some(25), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x26) => Some(26), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x27) => Some(27), + InlineAsmReg::AArch64(AArch64InlineAsmReg::x28) => Some(28), + // x29 is reserved + InlineAsmReg::AArch64(AArch64InlineAsmReg::x30) => Some(30), + _ => None, + } +} + /// If the register is an AArch64 vector register then return its index. fn a64_vreg_index(reg: InlineAsmReg) -> Option<u32> { match reg { @@ -535,6 +573,22 @@ fn reg_to_llvm(reg: InlineAsmRegOrRegClass, layout: Option<&TyAndLayout<'_>>) -> 'x' }; format!("{{{}mm{}}}", class, idx) + } else if let Some(idx) = a64_reg_index(reg) { + let class = if let Some(layout) = layout { + match layout.size.bytes() { + 8 => 'x', + _ => 'w', + } + } else { + // We use i32 as the type for discarded outputs + 'w' + }; + if class == 'x' && reg == InlineAsmReg::AArch64(AArch64InlineAsmReg::x30) { + // LLVM doesn't recognize x30. use lr instead. + "{lr}".to_string() + } else { + format!("{{{}{}}}", class, idx) + } } else if let Some(idx) = a64_vreg_index(reg) { let class = if let Some(layout) = layout { match layout.size.bytes() { @@ -550,9 +604,6 @@ fn reg_to_llvm(reg: InlineAsmRegOrRegClass, layout: Option<&TyAndLayout<'_>>) -> 'q' }; format!("{{{}{}}}", class, idx) - } else if reg == InlineAsmReg::AArch64(AArch64InlineAsmReg::x30) { - // LLVM doesn't recognize x30 - "{lr}".to_string() } else if reg == InlineAsmReg::Arm(ArmInlineAsmReg::r14) { // LLVM doesn't recognize r14 "{lr}".to_string() @@ -560,6 +611,8 @@ fn reg_to_llvm(reg: InlineAsmRegOrRegClass, layout: Option<&TyAndLayout<'_>>) -> format!("{{{}}}", reg.name()) } } + // The constraints can be retrieved from + // https://llvm.org/docs/LangRef.html#supported-constraint-code-list InlineAsmRegOrRegClass::RegClass(reg) => match reg { InlineAsmRegClass::AArch64(AArch64InlineAsmRegClass::reg) => "r", InlineAsmRegClass::AArch64(AArch64InlineAsmRegClass::vreg) => "w", @@ -633,6 +686,8 @@ fn modifier_to_llvm( reg: InlineAsmRegClass, modifier: Option<char>, ) -> Option<char> { + // The modifiers can be retrieved from + // https://llvm.org/docs/LangRef.html#asm-template-argument-modifiers match reg { InlineAsmRegClass::AArch64(AArch64InlineAsmRegClass::reg) => modifier, InlineAsmRegClass::AArch64(AArch64InlineAsmRegClass::vreg) diff --git a/compiler/rustc_codegen_llvm/src/attributes.rs b/compiler/rustc_codegen_llvm/src/attributes.rs index 9394d60134f..f3bdacf6085 100644 --- a/compiler/rustc_codegen_llvm/src/attributes.rs +++ b/compiler/rustc_codegen_llvm/src/attributes.rs @@ -12,8 +12,9 @@ use rustc_target::spec::{FramePointer, SanitizerSet, StackProbeType, StackProtec use smallvec::SmallVec; use crate::attributes; +use crate::errors::{MissingFeatures, SanitizerMemtagRequiresMte, TargetFeatureDisableOrEnable}; use crate::llvm::AttributePlace::Function; -use crate::llvm::{self, Attribute, AttributeKind, AttributePlace}; +use crate::llvm::{self, AllocKindFlags, Attribute, AttributeKind, AttributePlace, MemoryEffects}; use crate::llvm_util; pub use rustc_attr::{InlineAttr, InstructionSetAttr, OptimizeAttr}; @@ -35,6 +36,10 @@ pub fn apply_to_callsite(callsite: &Value, idx: AttributePlace, attrs: &[&Attrib /// Get LLVM attribute for the provided inline heuristic. #[inline] fn inline_attr<'ll>(cx: &CodegenCx<'ll, '_>, inline: InlineAttr) -> Option<&'ll Attribute> { + if !cx.tcx.sess.opts.unstable_opts.inline_llvm { + // disable LLVM inlining + return Some(AttributeKind::NoInline.create_attr(cx.llcx)); + } match inline { InlineAttr::Hint => Some(AttributeKind::InlineHint.create_attr(cx.llcx)), InlineAttr::Always => Some(AttributeKind::AlwaysInline.create_attr(cx.llcx)), @@ -56,7 +61,7 @@ pub fn sanitize_attrs<'ll>( no_sanitize: SanitizerSet, ) -> SmallVec<[&'ll Attribute; 4]> { let mut attrs = SmallVec::new(); - let enabled = cx.tcx.sess.opts.debugging_opts.sanitizer - no_sanitize; + let enabled = cx.tcx.sess.opts.unstable_opts.sanitizer - no_sanitize; if enabled.contains(SanitizerSet::ADDRESS) { attrs.push(llvm::AttributeKind::SanitizeAddress.create_attr(cx.llcx)); } @@ -69,13 +74,16 @@ pub fn sanitize_attrs<'ll>( if enabled.contains(SanitizerSet::HWADDRESS) { attrs.push(llvm::AttributeKind::SanitizeHWAddress.create_attr(cx.llcx)); } + if enabled.contains(SanitizerSet::SHADOWCALLSTACK) { + attrs.push(llvm::AttributeKind::ShadowCallStack.create_attr(cx.llcx)); + } if enabled.contains(SanitizerSet::MEMTAG) { // Check to make sure the mte target feature is actually enabled. let features = cx.tcx.global_backend_features(()); let mte_feature = features.iter().map(|s| &s[..]).rfind(|n| ["+mte", "-mte"].contains(&&n[..])); if let None | Some("-mte") = mte_feature { - cx.tcx.sess.err("`-Zsanitizer=memtag` requires `-Ctarget-feature=+mte`"); + cx.tcx.sess.emit_err(SanitizerMemtagRequiresMte); } attrs.push(llvm::AttributeKind::SanitizeMemTag.create_attr(cx.llcx)); @@ -136,7 +144,7 @@ fn probestack_attr<'ll>(cx: &CodegenCx<'ll, '_>) -> Option<&'ll Attribute> { if cx .sess() .opts - .debugging_opts + .unstable_opts .sanitizer .intersects(SanitizerSet::ADDRESS | SanitizerSet::THREAD) { @@ -149,7 +157,7 @@ fn probestack_attr<'ll>(cx: &CodegenCx<'ll, '_>) -> Option<&'ll Attribute> { } // probestack doesn't play nice either with gcov profiling. - if cx.sess().opts.debugging_opts.profile { + if cx.sess().opts.unstable_opts.profile { return None; } @@ -224,6 +232,10 @@ pub(crate) fn default_optimisation_attrs<'ll>( attrs } +fn create_alloc_family_attr(llcx: &llvm::Context) -> &llvm::Attribute { + llvm::CreateAttrStringValue(llcx, "alloc-family", "__rust_alloc") +} + /// Composite function which sets LLVM attributes for function depending on its AST (`#[attribute]`) /// attributes. pub fn from_fn_attrs<'ll, 'tcx>( @@ -246,13 +258,12 @@ pub fn from_fn_attrs<'ll, 'tcx>( OptimizeAttr::Speed => {} } - let inline = 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 - }; + let inline = + if codegen_fn_attrs.inline == InlineAttr::None && instance.def.requires_inline(cx.tcx) { + InlineAttr::Hint + } else { + codegen_fn_attrs.inline + }; to_add.extend(inline_attr(cx, inline)); // The `uwtable` attribute according to LLVM is: @@ -275,7 +286,7 @@ pub fn from_fn_attrs<'ll, 'tcx>( to_add.push(uwtable_attr(cx.llcx)); } - if cx.sess().opts.debugging_opts.profile_sample_use.is_some() { + if cx.sess().opts.unstable_opts.profile_sample_use.is_some() { to_add.push(llvm::CreateAttrString(cx.llcx, "use-sample-profile")); } @@ -292,19 +303,68 @@ pub fn from_fn_attrs<'ll, 'tcx>( to_add.push(AttributeKind::ReturnsTwice.create_attr(cx.llcx)); } if codegen_fn_attrs.flags.contains(CodegenFnAttrFlags::FFI_PURE) { - to_add.push(AttributeKind::ReadOnly.create_attr(cx.llcx)); + to_add.push(MemoryEffects::ReadOnly.create_attr(cx.llcx)); } if codegen_fn_attrs.flags.contains(CodegenFnAttrFlags::FFI_CONST) { - to_add.push(AttributeKind::ReadNone.create_attr(cx.llcx)); + to_add.push(MemoryEffects::None.create_attr(cx.llcx)); } if codegen_fn_attrs.flags.contains(CodegenFnAttrFlags::NAKED) { to_add.push(AttributeKind::Naked.create_attr(cx.llcx)); + // HACK(jubilee): "indirect branch tracking" works by attaching prologues to functions. + // And it is a module-level attribute, so the alternative is pulling naked functions into new LLVM modules. + // Otherwise LLVM's "naked" functions come with endbr prefixes per https://github.com/rust-lang/rust/issues/98768 + to_add.push(AttributeKind::NoCfCheck.create_attr(cx.llcx)); + // Need this for AArch64. + to_add.push(llvm::CreateAttrStringValue(cx.llcx, "branch-target-enforcement", "false")); } - if codegen_fn_attrs.flags.contains(CodegenFnAttrFlags::ALLOCATOR) { + if codegen_fn_attrs.flags.contains(CodegenFnAttrFlags::ALLOCATOR) + || codegen_fn_attrs.flags.contains(CodegenFnAttrFlags::ALLOCATOR_ZEROED) + { + if llvm_util::get_version() >= (15, 0, 0) { + to_add.push(create_alloc_family_attr(cx.llcx)); + // apply to argument place instead of function + let alloc_align = AttributeKind::AllocAlign.create_attr(cx.llcx); + attributes::apply_to_llfn(llfn, AttributePlace::Argument(1), &[alloc_align]); + to_add.push(llvm::CreateAllocSizeAttr(cx.llcx, 0)); + let mut flags = AllocKindFlags::Alloc | AllocKindFlags::Aligned; + if codegen_fn_attrs.flags.contains(CodegenFnAttrFlags::ALLOCATOR) { + flags |= AllocKindFlags::Uninitialized; + } else { + flags |= AllocKindFlags::Zeroed; + } + to_add.push(llvm::CreateAllocKindAttr(cx.llcx, flags)); + } // apply to return place instead of function (unlike all other attributes applied in this function) let no_alias = AttributeKind::NoAlias.create_attr(cx.llcx); attributes::apply_to_llfn(llfn, AttributePlace::ReturnValue, &[no_alias]); } + if codegen_fn_attrs.flags.contains(CodegenFnAttrFlags::REALLOCATOR) { + if llvm_util::get_version() >= (15, 0, 0) { + to_add.push(create_alloc_family_attr(cx.llcx)); + to_add.push(llvm::CreateAllocKindAttr( + cx.llcx, + AllocKindFlags::Realloc | AllocKindFlags::Aligned, + )); + // applies to argument place instead of function place + let allocated_pointer = AttributeKind::AllocatedPointer.create_attr(cx.llcx); + attributes::apply_to_llfn(llfn, AttributePlace::Argument(0), &[allocated_pointer]); + // apply to argument place instead of function + let alloc_align = AttributeKind::AllocAlign.create_attr(cx.llcx); + attributes::apply_to_llfn(llfn, AttributePlace::Argument(2), &[alloc_align]); + to_add.push(llvm::CreateAllocSizeAttr(cx.llcx, 3)); + } + let no_alias = AttributeKind::NoAlias.create_attr(cx.llcx); + attributes::apply_to_llfn(llfn, AttributePlace::ReturnValue, &[no_alias]); + } + if codegen_fn_attrs.flags.contains(CodegenFnAttrFlags::DEALLOCATOR) { + if llvm_util::get_version() >= (15, 0, 0) { + to_add.push(create_alloc_family_attr(cx.llcx)); + to_add.push(llvm::CreateAllocKindAttr(cx.llcx, AllocKindFlags::Free)); + // applies to argument place instead of function place + let allocated_pointer = AttributeKind::AllocatedPointer.create_attr(cx.llcx); + attributes::apply_to_llfn(llfn, AttributePlace::Argument(0), &[allocated_pointer]); + } + } if codegen_fn_attrs.flags.contains(CodegenFnAttrFlags::CMSE_NONSECURE_ENTRY) { to_add.push(llvm::CreateAttrString(cx.llcx, "cmse_nonsecure_entry")); } @@ -330,15 +390,17 @@ pub fn from_fn_attrs<'ll, 'tcx>( ) { let span = cx .tcx - .get_attr(instance.def_id(), sym::target_feature) + .get_attrs(instance.def_id(), sym::target_feature) + .next() .map_or_else(|| cx.tcx.def_span(instance.def_id()), |a| a.span); - let msg = format!( - "the target features {} must all be either enabled or disabled together", - f.join(", ") - ); - let mut err = cx.tcx.sess.struct_span_err(span, &msg); - err.help("add the missing features in a `target_feature` attribute"); - err.emit(); + cx.tcx + .sess + .create_err(TargetFeatureDisableOrEnable { + features: f, + span: Some(span), + missing_features: Some(MissingFeatures), + }) + .emit(); return; } diff --git a/compiler/rustc_codegen_llvm/src/back/archive.rs b/compiler/rustc_codegen_llvm/src/back/archive.rs index 8f6438e85ad..0aee1a1439b 100644 --- a/compiler/rustc_codegen_llvm/src/back/archive.rs +++ b/compiler/rustc_codegen_llvm/src/back/archive.rs @@ -1,33 +1,32 @@ //! A helper class for dealing with static archives use std::env; -use std::ffi::{CStr, CString, OsString}; +use std::ffi::{c_char, c_void, CStr, CString, OsString}; use std::io; use std::mem; use std::path::{Path, PathBuf}; use std::ptr; use std::str; +use crate::common; +use crate::errors::{ + DlltoolFailImportLibrary, ErrorCallingDllTool, ErrorCreatingImportLibrary, ErrorWritingDEFFile, +}; use crate::llvm::archive_ro::{ArchiveRO, Child}; use crate::llvm::{self, ArchiveKind, LLVMMachineType, LLVMRustCOFFShortExport}; -use rustc_codegen_ssa::back::archive::ArchiveBuilder; -use rustc_data_structures::temp_dir::MaybeTempDir; -use rustc_session::cstore::{DllCallingConvention, DllImport}; -use rustc_session::Session; +use rustc_codegen_ssa::back::archive::{ + get_native_object_symbols, try_extract_macho_fat_archive, ArArchiveBuilder, + ArchiveBuildFailure, ArchiveBuilder, ArchiveBuilderBuilder, UnknownArchiveKind, +}; -struct ArchiveConfig<'a> { - pub sess: &'a Session, - pub dst: PathBuf, - pub src: Option<PathBuf>, -} +use rustc_session::cstore::DllImport; +use rustc_session::Session; /// Helper for adding many files to an archive. #[must_use = "must call build() to finish building the archive"] -pub struct LlvmArchiveBuilder<'a> { - config: ArchiveConfig<'a>, - removals: Vec<String>, +pub(crate) struct LlvmArchiveBuilder<'a> { + sess: &'a Session, additions: Vec<Addition>, - src_archive: Option<Option<ArchiveRO>>, } enum Addition { @@ -50,10 +49,6 @@ fn is_relevant_child(c: &Child<'_>) -> bool { } } -fn archive_config<'a>(sess: &'a Session, output: &Path, input: Option<&Path>) -> ArchiveConfig<'a> { - ArchiveConfig { sess, dst: output.to_path_buf(), src: input.map(|p| p.to_path_buf()) } -} - /// Map machine type strings to values of LLVM's MachineTypes enum. fn llvm_machine_type(cpu: &str) -> LLVMMachineType { match cpu { @@ -66,46 +61,18 @@ fn llvm_machine_type(cpu: &str) -> LLVMMachineType { } impl<'a> ArchiveBuilder<'a> for LlvmArchiveBuilder<'a> { - /// Creates a new static archive, ready for modifying the archive specified - /// by `config`. - fn new(sess: &'a Session, output: &Path, input: Option<&Path>) -> LlvmArchiveBuilder<'a> { - let config = archive_config(sess, output, input); - LlvmArchiveBuilder { - config, - removals: Vec::new(), - additions: Vec::new(), - src_archive: None, - } - } - - /// Removes a file from this archive - fn remove_file(&mut self, file: &str) { - self.removals.push(file.to_string()); - } - - /// Lists all files in an archive - fn src_files(&mut self) -> Vec<String> { - if self.src_archive().is_none() { - return Vec::new(); + fn add_archive( + &mut self, + archive: &Path, + skip: Box<dyn FnMut(&str) -> bool + 'static>, + ) -> io::Result<()> { + let mut archive = archive.to_path_buf(); + if self.sess.target.llvm_target.contains("-apple-macosx") { + if let Some(new_archive) = try_extract_macho_fat_archive(&self.sess, &archive)? { + archive = new_archive + } } - - let archive = self.src_archive.as_ref().unwrap().as_ref().unwrap(); - - archive - .iter() - .filter_map(|child| child.ok()) - .filter(is_relevant_child) - .filter_map(|child| child.name()) - .filter(|name| !self.removals.iter().any(|x| x == name)) - .map(|name| name.to_owned()) - .collect() - } - - fn add_archive<F>(&mut self, archive: &Path, skip: F) -> io::Result<()> - where - F: FnMut(&str) -> bool + 'static, - { - let archive_ro = match ArchiveRO::open(archive) { + let archive_ro = match ArchiveRO::open(&archive) { Ok(ar) => ar, Err(e) => return Err(io::Error::new(io::ErrorKind::Other, e)), }; @@ -113,7 +80,7 @@ impl<'a> ArchiveBuilder<'a> for LlvmArchiveBuilder<'a> { return Ok(()); } self.additions.push(Addition::Archive { - path: archive.to_path_buf(), + path: archive, archive: archive_ro, skip: Box::new(skip), }); @@ -129,44 +96,53 @@ impl<'a> ArchiveBuilder<'a> for LlvmArchiveBuilder<'a> { /// Combine the provided files, rlibs, and native libraries into a single /// `Archive`. - fn build(mut self) { - let kind = self.llvm_archive_kind().unwrap_or_else(|kind| { - self.config.sess.fatal(&format!("Don't know how to build archive of type: {}", kind)) - }); + fn build(mut self: Box<Self>, output: &Path) -> bool { + match self.build_with_llvm(output) { + Ok(any_members) => any_members, + Err(e) => self.sess.emit_fatal(ArchiveBuildFailure { error: e }), + } + } +} + +pub struct LlvmArchiveBuilderBuilder; - if let Err(e) = self.build_with_llvm(kind) { - self.config.sess.fatal(&format!("failed to build archive: {}", e)); +impl ArchiveBuilderBuilder for LlvmArchiveBuilderBuilder { + fn new_archive_builder<'a>(&self, sess: &'a Session) -> Box<dyn ArchiveBuilder<'a> + 'a> { + if sess.target.arch == "wasm32" || sess.target.arch == "wasm64" { + Box::new(LlvmArchiveBuilder { sess, additions: Vec::new() }) + } else { + Box::new(ArArchiveBuilder::new(sess, get_llvm_object_symbols)) } } - fn inject_dll_import_lib( - &mut self, + fn create_dll_import_lib( + &self, + sess: &Session, lib_name: &str, dll_imports: &[DllImport], - tmpdir: &MaybeTempDir, - ) { + tmpdir: &Path, + is_direct_dependency: bool, + ) -> PathBuf { + let name_suffix = if is_direct_dependency { "_imports" } else { "_imports_indirect" }; let output_path = { - let mut output_path: PathBuf = tmpdir.as_ref().to_path_buf(); - output_path.push(format!("{}_imports", lib_name)); + let mut output_path: PathBuf = tmpdir.to_path_buf(); + output_path.push(format!("{}{}", lib_name, name_suffix)); output_path.with_extension("lib") }; - let target = &self.config.sess.target; - let mingw_gnu_toolchain = target.vendor == "pc" - && target.os == "windows" - && target.env == "gnu" - && target.abi.is_empty(); + let target = &sess.target; + let mingw_gnu_toolchain = common::is_mingw_gnu_toolchain(target); let import_name_and_ordinal_vector: Vec<(String, Option<u16>)> = dll_imports .iter() .map(|import: &DllImport| { - if self.config.sess.target.arch == "x86" { + if sess.target.arch == "x86" { ( - LlvmArchiveBuilder::i686_decorated_name(import, mingw_gnu_toolchain), - import.ordinal, + common::i686_decorated_name(import, mingw_gnu_toolchain, false), + import.ordinal(), ) } else { - (import.name.to_string(), import.ordinal) + (import.name.to_string(), import.ordinal()) } }) .collect(); @@ -178,7 +154,7 @@ impl<'a> ArchiveBuilder<'a> for LlvmArchiveBuilder<'a> { // functions. Therefore, use binutils to create the import library instead, // by writing a .DEF file to the temp dir and calling binutils's dlltool. let def_file_path = - tmpdir.as_ref().join(format!("{}_imports", lib_name)).with_extension("def"); + tmpdir.join(format!("{}{}", lib_name, name_suffix)).with_extension("def"); let def_file_content = format!( "EXPORTS\n{}", @@ -197,11 +173,14 @@ impl<'a> ArchiveBuilder<'a> for LlvmArchiveBuilder<'a> { match std::fs::write(&def_file_path, def_file_content) { Ok(_) => {} Err(e) => { - self.config.sess.fatal(&format!("Error writing .DEF file: {}", e)); + sess.emit_fatal(ErrorWritingDEFFile { error: e }); } }; - let dlltool = find_binutils_dlltool(self.config.sess); + // --no-leading-underscore: For the `import_name_type` feature to work, we need to be + // able to control the *exact* spelling of each of the symbols that are being imported: + // hence we don't want `dlltool` adding leading underscores automatically. + let dlltool = find_binutils_dlltool(sess); let result = std::process::Command::new(dlltool) .args([ "-d", @@ -210,18 +189,20 @@ impl<'a> ArchiveBuilder<'a> for LlvmArchiveBuilder<'a> { lib_name, "-l", output_path.to_str().unwrap(), + "--no-leading-underscore", ]) .output(); match result { Err(e) => { - self.config.sess.fatal(&format!("Error calling dlltool: {}", e)); + sess.emit_fatal(ErrorCallingDllTool { error: e }); + } + Ok(output) if !output.status.success() => { + sess.emit_fatal(DlltoolFailImportLibrary { + stdout: String::from_utf8_lossy(&output.stdout), + stderr: String::from_utf8_lossy(&output.stderr), + }) } - Ok(output) if !output.status.success() => self.config.sess.fatal(&format!( - "Dlltool could not create import library: {}\n{}", - String::from_utf8_lossy(&output.stdout), - String::from_utf8_lossy(&output.stderr) - )), _ => {} } } else { @@ -230,10 +211,10 @@ impl<'a> ArchiveBuilder<'a> for LlvmArchiveBuilder<'a> { let output_path_z = rustc_fs_util::path_to_c_string(&output_path); - tracing::trace!("invoking LLVMRustWriteImportLibrary"); - tracing::trace!(" dll_name {:#?}", dll_name_z); - tracing::trace!(" output_path {}", output_path.display()); - tracing::trace!( + trace!("invoking LLVMRustWriteImportLibrary"); + trace!(" dll_name {:#?}", dll_name_z); + trace!(" output_path {}", output_path.display()); + trace!( " import names: {}", dll_imports .iter() @@ -263,71 +244,93 @@ impl<'a> ArchiveBuilder<'a> for LlvmArchiveBuilder<'a> { output_path_z.as_ptr(), ffi_exports.as_ptr(), ffi_exports.len(), - llvm_machine_type(&self.config.sess.target.arch) as u16, - !self.config.sess.target.is_like_msvc, + llvm_machine_type(&sess.target.arch) as u16, + !sess.target.is_like_msvc, ) }; if result == crate::llvm::LLVMRustResult::Failure { - self.config.sess.fatal(&format!( - "Error creating import library for {}: {}", + sess.emit_fatal(ErrorCreatingImportLibrary { lib_name, - llvm::last_error().unwrap_or("unknown LLVM error".to_string()) - )); + error: llvm::last_error().unwrap_or("unknown LLVM error".to_string()), + }); } }; - self.add_archive(&output_path, |_| false).unwrap_or_else(|e| { - self.config.sess.fatal(&format!( - "failed to add native library {}: {}", - output_path.display(), - e - )); - }); + output_path } } -impl<'a> LlvmArchiveBuilder<'a> { - fn src_archive(&mut self) -> Option<&ArchiveRO> { - if let Some(ref a) = self.src_archive { - return a.as_ref(); +// The object crate doesn't know how to get symbols for LLVM bitcode and COFF bigobj files. +// As such we need to use LLVM for them. +#[deny(unsafe_op_in_unsafe_fn)] +fn get_llvm_object_symbols( + buf: &[u8], + f: &mut dyn FnMut(&[u8]) -> io::Result<()>, +) -> io::Result<bool> { + let is_bitcode = unsafe { llvm::LLVMRustIsBitcode(buf.as_ptr(), buf.len()) }; + + // COFF bigobj file, msvc LTO file or import library. See + // https://github.com/llvm/llvm-project/blob/453f27bc9/llvm/lib/BinaryFormat/Magic.cpp#L38-L51 + let is_unsupported_windows_obj_file = buf.get(0..4) == Some(b"\0\0\xFF\xFF"); + + if is_bitcode || is_unsupported_windows_obj_file { + let mut state = Box::new(f); + + let err = unsafe { + llvm::LLVMRustGetSymbols( + buf.as_ptr(), + buf.len(), + &mut *state as *mut &mut _ as *mut c_void, + callback, + error_callback, + ) + }; + + if err.is_null() { + return Ok(true); + } else { + return Err(unsafe { *Box::from_raw(err as *mut io::Error) }); } - let src = self.config.src.as_ref()?; - self.src_archive = Some(ArchiveRO::open(src).ok()); - self.src_archive.as_ref().unwrap().as_ref() - } - fn llvm_archive_kind(&self) -> Result<ArchiveKind, &str> { - let kind = &*self.config.sess.target.archive_format; - kind.parse().map_err(|_| kind) + unsafe extern "C" fn callback( + state: *mut c_void, + symbol_name: *const c_char, + ) -> *mut c_void { + let f = unsafe { &mut *(state as *mut &mut dyn FnMut(&[u8]) -> io::Result<()>) }; + match f(unsafe { CStr::from_ptr(symbol_name) }.to_bytes()) { + Ok(()) => std::ptr::null_mut(), + Err(err) => Box::into_raw(Box::new(err)) as *mut c_void, + } + } + + unsafe extern "C" fn error_callback(error: *const c_char) -> *mut c_void { + let error = unsafe { CStr::from_ptr(error) }; + Box::into_raw(Box::new(io::Error::new( + io::ErrorKind::Other, + format!("LLVM error: {}", error.to_string_lossy()), + ))) as *mut c_void + } + } else { + get_native_object_symbols(buf, f) } +} + +impl<'a> LlvmArchiveBuilder<'a> { + fn build_with_llvm(&mut self, output: &Path) -> io::Result<bool> { + let kind = &*self.sess.target.archive_format; + let kind = kind + .parse::<ArchiveKind>() + .map_err(|_| kind) + .unwrap_or_else(|kind| self.sess.emit_fatal(UnknownArchiveKind { kind })); - fn build_with_llvm(&mut self, kind: ArchiveKind) -> io::Result<()> { - let removals = mem::take(&mut self.removals); let mut additions = mem::take(&mut self.additions); let mut strings = Vec::new(); let mut members = Vec::new(); - let dst = CString::new(self.config.dst.to_str().unwrap())?; + let dst = CString::new(output.to_str().unwrap())?; unsafe { - if let Some(archive) = self.src_archive() { - for child in archive.iter() { - let child = child.map_err(string_to_io_error)?; - let Some(child_name) = child.name() else { continue }; - if removals.iter().any(|r| r == child_name) { - continue; - } - - let name = CString::new(child_name)?; - members.push(llvm::LLVMRustArchiveMemberNew( - ptr::null(), - name.as_ptr(), - Some(child.raw), - )); - strings.push(name); - } - } for addition in &mut additions { match addition { Addition::File { path, name_in_archive } => { @@ -389,7 +392,7 @@ impl<'a> LlvmArchiveBuilder<'a> { }; Err(io::Error::new(io::ErrorKind::Other, msg)) } else { - Ok(()) + Ok(!members.is_empty()) }; for member in members { llvm::LLVMRustArchiveMemberFree(member); @@ -397,22 +400,6 @@ impl<'a> LlvmArchiveBuilder<'a> { ret } } - - fn i686_decorated_name(import: &DllImport, mingw: bool) -> String { - let name = import.name; - let prefix = if mingw { "" } else { "_" }; - - match import.calling_convention { - DllCallingConvention::C => format!("{}{}", prefix, name), - DllCallingConvention::Stdcall(arg_list_size) => { - format!("{}{}@{}", prefix, name, arg_list_size) - } - DllCallingConvention::Fastcall(arg_list_size) => format!("@{}@{}", name, arg_list_size), - DllCallingConvention::Vectorcall(arg_list_size) => { - format!("{}@@{}", name, arg_list_size) - } - } - } } fn string_to_io_error(s: String) -> io::Error { @@ -421,7 +408,7 @@ fn string_to_io_error(s: String) -> io::Error { fn find_binutils_dlltool(sess: &Session) -> OsString { assert!(sess.target.options.is_like_windows && !sess.target.options.is_like_msvc); - if let Some(dlltool_path) = &sess.opts.debugging_opts.dlltool { + if let Some(dlltool_path) = &sess.opts.unstable_opts.dlltool { return dlltool_path.clone().into_os_string(); } diff --git a/compiler/rustc_codegen_llvm/src/back/lto.rs b/compiler/rustc_codegen_llvm/src/back/lto.rs index 38402e04313..3fa21355b7f 100644 --- a/compiler/rustc_codegen_llvm/src/back/lto.rs +++ b/compiler/rustc_codegen_llvm/src/back/lto.rs @@ -1,15 +1,15 @@ -use crate::back::write::{ - self, save_temp_bitcode, to_llvm_opt_settings, with_llvm_pmb, DiagnosticHandlers, -}; -use crate::llvm::archive_ro::ArchiveRO; -use crate::llvm::{self, build_string, False, True}; -use crate::{llvm_util, LlvmCodegenBackend, ModuleLlvm}; +use crate::back::write::{self, save_temp_bitcode, DiagnosticHandlers}; +use crate::errors::DynamicLinkingWithLTO; +use crate::llvm::{self, build_string}; +use crate::{LlvmCodegenBackend, ModuleLlvm}; +use object::read::archive::ArchiveFile; use rustc_codegen_ssa::back::lto::{LtoModuleCodegen, SerializedModule, ThinModule, ThinShared}; use rustc_codegen_ssa::back::symbol_export; use rustc_codegen_ssa::back::write::{CodegenContext, FatLTOInput, TargetMachineFactoryConfig}; use rustc_codegen_ssa::traits::*; use rustc_codegen_ssa::{looks_like_rust_object_file, ModuleCodegen, ModuleKind}; use rustc_data_structures::fx::FxHashMap; +use rustc_data_structures::memmap::Mmap; use rustc_errors::{FatalError, Handler}; use rustc_hir::def_id::LOCAL_CRATE; use rustc_middle::bug; @@ -17,7 +17,6 @@ use rustc_middle::dep_graph::WorkProduct; use rustc_middle::middle::exported_symbols::{SymbolExportInfo, SymbolExportLevel}; use rustc_session::cgu_reuse_tracker::CguReuse; use rustc_session::config::{self, CrateType, Lto}; -use tracing::{debug, info}; use std::ffi::{CStr, CString}; use std::fs::File; @@ -34,8 +33,8 @@ pub const THIN_LTO_KEYS_INCR_COMP_FILE_NAME: &str = "thin-lto-past-keys.bin"; pub fn crate_type_allows_lto(crate_type: CrateType) -> bool { match crate_type { - CrateType::Executable | CrateType::Staticlib | CrateType::Cdylib => true, - CrateType::Dylib | CrateType::Rlib | CrateType::ProcMacro => false, + CrateType::Executable | CrateType::Dylib | CrateType::Staticlib | CrateType::Cdylib => true, + CrateType::Rlib | CrateType::ProcMacro => false, } } @@ -75,17 +74,6 @@ fn prepare_lto( // with either fat or thin LTO let mut upstream_modules = Vec::new(); if cgcx.lto != Lto::ThinLocal { - if cgcx.opts.cg.prefer_dynamic { - diag_handler - .struct_err("cannot prefer dynamic linking when performing LTO") - .note( - "only 'staticlib', 'bin', and 'cdylib' outputs are \ - supported with LTO", - ) - .emit(); - return Err(FatalError); - } - // Make sure we actually can run LTO for crate_type in cgcx.crate_types.iter() { if !crate_type_allows_lto(*crate_type) { @@ -94,9 +82,19 @@ fn prepare_lto( static library outputs", ); return Err(e); + } else if *crate_type == CrateType::Dylib { + if !cgcx.opts.unstable_opts.dylib_lto { + return Err(diag_handler + .fatal("lto cannot be used for `dylib` crate type without `-Zdylib-lto`")); + } } } + if cgcx.opts.cg.prefer_dynamic && !cgcx.opts.unstable_opts.dylib_lto { + diag_handler.emit_err(DynamicLinkingWithLTO); + return Err(FatalError); + } + for &(cnum, ref path) in cgcx.each_linked_rlib_for_lto.iter() { let exported_symbols = cgcx.exported_symbols.as_ref().expect("needs exported symbols for LTO"); @@ -107,14 +105,24 @@ fn prepare_lto( .extend(exported_symbols[&cnum].iter().filter_map(symbol_filter)); } - let archive = ArchiveRO::open(path).expect("wanted an rlib"); + let archive_data = unsafe { + Mmap::map(std::fs::File::open(&path).expect("couldn't open rlib")) + .expect("couldn't map rlib") + }; + let archive = ArchiveFile::parse(&*archive_data).expect("wanted an rlib"); let obj_files = archive - .iter() - .filter_map(|child| child.ok().and_then(|c| c.name().map(|name| (name, c)))) + .members() + .filter_map(|child| { + child.ok().and_then(|c| { + std::str::from_utf8(c.name()).ok().map(|name| (name.trim(), c)) + }) + }) .filter(|&(name, _)| looks_like_rust_object_file(name)); for (name, child) in obj_files { info!("adding bitcode from {}", name); - match get_bitcode_slice_from_object_data(child.data()) { + match get_bitcode_slice_from_object_data( + child.data(&*archive_data).expect("corrupt rlib"), + ) { Ok(data) => { let module = SerializedModule::FromRlib(data.to_vec()); upstream_modules.push((module, CString::new(name).unwrap())); @@ -199,7 +207,7 @@ pub(crate) fn run_thin( pub(crate) fn prepare_thin(module: ModuleCodegen<ModuleLlvm>) -> (String, ThinBuffer) { let name = module.name.clone(); - let buffer = ThinBuffer::new(module.module_llvm.llmod()); + let buffer = ThinBuffer::new(module.module_llvm.llmod(), true); (name, buffer) } @@ -325,20 +333,6 @@ fn fat_lto( drop(linker); save_temp_bitcode(cgcx, &module, "lto.input"); - // Fat LTO also suffers from the invalid DWARF issue similar to Thin LTO. - // Here we rewrite all `DICompileUnit` pointers if there is only one `DICompileUnit`. - // This only works around the problem when codegen-units = 1. - // Refer to the comments in the `optimize_thin_module` function for more details. - let mut cu1 = ptr::null_mut(); - let mut cu2 = ptr::null_mut(); - unsafe { llvm::LLVMRustLTOGetDICompileUnit(llmod, &mut cu1, &mut cu2) }; - if !cu2.is_null() { - let _timer = - cgcx.prof.generic_activity_with_arg("LLVM_fat_lto_patch_debuginfo", &*module.name); - unsafe { llvm::LLVMRustLTOPatchDICompileUnit(llmod, cu1) }; - save_temp_bitcode(cgcx, &module, "fat-lto-after-patch"); - } - // Internalize everything below threshold to help strip out more modules and such. unsafe { let ptr = symbols_below_threshold.as_ptr(); @@ -579,7 +573,7 @@ pub(crate) fn run_pass_manager( module: &mut ModuleCodegen<ModuleLlvm>, thin: bool, ) -> Result<(), FatalError> { - let _timer = cgcx.prof.extra_verbose_generic_activity("LLVM_lto_optimize", &*module.name); + let _timer = cgcx.prof.verbose_generic_activity_with_arg("LLVM_lto_optimize", &*module.name); let config = cgcx.config(module.kind); // Now we have one massive module inside of llmod. Time to run the @@ -601,61 +595,9 @@ pub(crate) fn run_pass_manager( 1, ); } - if llvm_util::should_use_new_llvm_pass_manager( - &config.new_llvm_pass_manager, - &cgcx.target_arch, - ) { - let opt_stage = if thin { llvm::OptStage::ThinLTO } else { llvm::OptStage::FatLTO }; - let opt_level = config.opt_level.unwrap_or(config::OptLevel::No); - write::optimize_with_new_llvm_pass_manager( - cgcx, - diag_handler, - module, - config, - opt_level, - opt_stage, - )?; - debug!("lto done"); - return Ok(()); - } - - let pm = llvm::LLVMCreatePassManager(); - llvm::LLVMAddAnalysisPasses(module.module_llvm.tm, pm); - - if config.verify_llvm_ir { - let pass = llvm::LLVMRustFindAndCreatePass("verify\0".as_ptr().cast()); - llvm::LLVMRustAddPass(pm, pass.unwrap()); - } - - let opt_level = config - .opt_level - .map(|x| to_llvm_opt_settings(x).0) - .unwrap_or(llvm::CodeGenOptLevel::None); - with_llvm_pmb(module.module_llvm.llmod(), config, opt_level, false, &mut |b| { - if thin { - llvm::LLVMRustPassManagerBuilderPopulateThinLTOPassManager(b, pm); - } else { - llvm::LLVMRustPassManagerBuilderPopulateLTOPassManager( - b, pm, /* Internalize = */ False, /* RunInliner = */ True, - ); - } - }); - - // We always generate bitcode through ThinLTOBuffers, - // which do not support anonymous globals - if config.bitcode_needed() { - let pass = llvm::LLVMRustFindAndCreatePass("name-anon-globals\0".as_ptr().cast()); - llvm::LLVMRustAddPass(pm, pass.unwrap()); - } - - if config.verify_llvm_ir { - let pass = llvm::LLVMRustFindAndCreatePass("verify\0".as_ptr().cast()); - llvm::LLVMRustAddPass(pm, pass.unwrap()); - } - - llvm::LLVMRunPassManager(pm, module.module_llvm.llmod()); - - llvm::LLVMDisposePassManager(pm); + let opt_stage = if thin { llvm::OptStage::ThinLTO } else { llvm::OptStage::FatLTO }; + let opt_level = config.opt_level.unwrap_or(config::OptLevel::No); + write::llvm_optimize(cgcx, diag_handler, module, config, opt_level, opt_stage)?; } debug!("lto done"); Ok(()) @@ -709,9 +651,9 @@ unsafe impl Send for ThinBuffer {} unsafe impl Sync for ThinBuffer {} impl ThinBuffer { - pub fn new(m: &llvm::Module) -> ThinBuffer { + pub fn new(m: &llvm::Module, is_thin: bool) -> ThinBuffer { unsafe { - let buffer = llvm::LLVMRustThinLTOBufferCreate(m); + let buffer = llvm::LLVMRustThinLTOBufferCreate(m, is_thin); ThinBuffer(buffer) } } @@ -769,7 +711,7 @@ pub unsafe fn optimize_thin_module( // an error. let mut cu1 = ptr::null_mut(); let mut cu2 = ptr::null_mut(); - llvm::LLVMRustLTOGetDICompileUnit(llmod, &mut cu1, &mut cu2); + llvm::LLVMRustThinLTOGetDICompileUnit(llmod, &mut cu1, &mut cu2); if !cu2.is_null() { let msg = "multiple source DICompileUnits found"; return Err(write::llvm_err(&diag_handler, msg)); @@ -858,7 +800,7 @@ pub unsafe fn optimize_thin_module( let _timer = cgcx .prof .generic_activity_with_arg("LLVM_thin_lto_patch_debuginfo", thin_module.name()); - llvm::LLVMRustLTOPatchDICompileUnit(llmod, cu1); + llvm::LLVMRustThinLTOPatchDICompileUnit(llmod, cu1); save_temp_bitcode(cgcx, &module, "thin-lto-after-patch"); } diff --git a/compiler/rustc_codegen_llvm/src/back/write.rs b/compiler/rustc_codegen_llvm/src/back/write.rs index ab8874d796d..97d0de47b3a 100644 --- a/compiler/rustc_codegen_llvm/src/back/write.rs +++ b/compiler/rustc_codegen_llvm/src/back/write.rs @@ -5,7 +5,7 @@ use crate::back::profiling::{ use crate::base; use crate::common; use crate::consts; -use crate::llvm::{self, DiagnosticInfo, PassManager, SMDiagnostic}; +use crate::llvm::{self, DiagnosticInfo, PassManager}; use crate::llvm_util; use crate::type_::Type; use crate::LlvmCodegenBackend; @@ -21,14 +21,12 @@ use rustc_data_structures::profiling::SelfProfilerRef; use rustc_data_structures::small_c_str::SmallCStr; use rustc_errors::{FatalError, Handler, Level}; use rustc_fs_util::{link_or_copy, path_to_c_string}; -use rustc_middle::bug; use rustc_middle::ty::TyCtxt; use rustc_session::config::{self, Lto, OutputType, Passes, SplitDwarfKind, SwitchWithOptPath}; use rustc_session::Session; use rustc_span::symbol::sym; use rustc_span::InnerSpan; use rustc_target::spec::{CodeModel, RelocModel, SanitizerSet, SplitDebuginfo}; -use tracing::debug; use libc::{c_char, c_int, c_uint, c_void, size_t}; use std::ffi::CString; @@ -107,7 +105,7 @@ pub fn create_target_machine(tcx: TyCtxt<'_>, mod_name: &str) -> &'static mut ll let split_dwarf_file = if tcx.sess.target_can_use_split_dwarf() { tcx.output_filenames(()).split_dwarf_path( tcx.sess.split_debuginfo(), - tcx.sess.opts.debugging_opts.split_dwarf_kind, + tcx.sess.opts.unstable_opts.split_dwarf_kind, Some(mod_name), ) } else { @@ -182,9 +180,9 @@ pub fn target_machine_factory( let use_softfp = sess.opts.cg.soft_float; let ffunction_sections = - sess.opts.debugging_opts.function_sections.unwrap_or(sess.target.function_sections); + sess.opts.unstable_opts.function_sections.unwrap_or(sess.target.function_sections); let fdata_sections = ffunction_sections; - let funique_section_names = !sess.opts.debugging_opts.no_unique_section_names; + let funique_section_names = !sess.opts.unstable_opts.no_unique_section_names; let code_model = to_llvm_code_model(sess.code_model()); @@ -202,15 +200,15 @@ pub fn target_machine_factory( let features = CString::new(target_features.join(",")).unwrap(); let abi = SmallCStr::new(&sess.target.llvm_abiname); let trap_unreachable = - sess.opts.debugging_opts.trap_unreachable.unwrap_or(sess.target.trap_unreachable); - let emit_stack_size_section = sess.opts.debugging_opts.emit_stack_sizes; + sess.opts.unstable_opts.trap_unreachable.unwrap_or(sess.target.trap_unreachable); + let emit_stack_size_section = sess.opts.unstable_opts.emit_stack_sizes; let asm_comments = sess.asm_comments(); let relax_elf_relocations = - sess.opts.debugging_opts.relax_elf_relocations.unwrap_or(sess.target.relax_elf_relocations); + sess.opts.unstable_opts.relax_elf_relocations.unwrap_or(sess.target.relax_elf_relocations); let use_init_array = - !sess.opts.debugging_opts.use_ctors_section.unwrap_or(sess.target.use_ctors_section); + !sess.opts.unstable_opts.use_ctors_section.unwrap_or(sess.target.use_ctors_section); let path_mapping = sess.source_map().path_mapping().clone(); @@ -304,7 +302,6 @@ impl<'a> DiagnosticHandlers<'a> { remark_passes.as_ptr(), remark_passes.len(), ); - llvm::LLVMRustSetInlineAsmDiagnosticHandler(llcx, inline_asm_handler, data.cast()); DiagnosticHandlers { data, llcx, old_handler } } } @@ -312,9 +309,7 @@ impl<'a> DiagnosticHandlers<'a> { impl<'a> Drop for DiagnosticHandlers<'a> { fn drop(&mut self) { - use std::ptr::null_mut; unsafe { - llvm::LLVMRustSetInlineAsmDiagnosticHandler(self.llcx, inline_asm_handler, null_mut()); llvm::LLVMRustContextSetDiagnosticHandler(self.llcx, self.old_handler); drop(Box::from_raw(self.data)); } @@ -342,16 +337,6 @@ fn report_inline_asm( cgcx.diag_emitter.inline_asm_error(cookie as u32, msg, level, source); } -unsafe extern "C" fn inline_asm_handler(diag: &SMDiagnostic, user: *const c_void, cookie: c_uint) { - if user.is_null() { - return; - } - let (cgcx, _) = *(user as *const (&CodegenContext<LlvmCodegenBackend>, &Handler)); - - let smdiag = llvm::diagnostic::SrcMgrDiagnostic::unpack(diag); - report_inline_asm(cgcx, smdiag.message, smdiag.level, cookie, smdiag.source); -} - unsafe extern "C" fn diagnostic_handler(info: &DiagnosticInfo, user: *mut c_void) { if user.is_null() { return; @@ -423,7 +408,15 @@ fn get_pgo_sample_use_path(config: &ModuleConfig) -> Option<CString> { .map(|path_buf| CString::new(path_buf.to_string_lossy().as_bytes()).unwrap()) } -pub(crate) unsafe fn optimize_with_new_llvm_pass_manager( +fn get_instr_profile_output_path(config: &ModuleConfig) -> Option<CString> { + if config.instrument_coverage { + Some(CString::new("default_%m_%p.profraw").unwrap()) + } else { + None + } +} + +pub(crate) unsafe fn llvm_optimize( cgcx: &CodegenContext<LlvmCodegenBackend>, diag_handler: &Handler, module: &ModuleCodegen<ModuleLlvm>, @@ -438,6 +431,7 @@ pub(crate) unsafe fn optimize_with_new_llvm_pass_manager( let pgo_use_path = get_pgo_use_path(config); let pgo_sample_use_path = get_pgo_sample_use_path(config); let is_lto = opt_stage == llvm::OptStage::ThinLTO || opt_stage == llvm::OptStage::FatLTO; + let instr_profile_output_path = get_instr_profile_output_path(config); // Sanitizer instrumentation is only inserted during the pre-link optimization stage. let sanitizer_options = if !is_lto { Some(llvm::SanitizerOptions { @@ -470,7 +464,7 @@ pub(crate) unsafe fn optimize_with_new_llvm_pass_manager( // FIXME: NewPM doesn't provide a facility to pass custom InlineParams. // We would have to add upstream support for this first, before we can support // config.inline_threshold and our more aggressive default thresholds. - let result = llvm::LLVMRustOptimizeWithNewPassManager( + let result = llvm::LLVMRustOptimize( module.module_llvm.llmod(), &*module.module_llvm.tm, to_pass_builder_opt_level(opt_level), @@ -488,6 +482,7 @@ pub(crate) unsafe fn optimize_with_new_llvm_pass_manager( pgo_gen_path.as_ref().map_or(std::ptr::null(), |s| s.as_ptr()), pgo_use_path.as_ref().map_or(std::ptr::null(), |s| s.as_ptr()), config.instrument_coverage, + instr_profile_output_path.as_ref().map_or(std::ptr::null(), |s| s.as_ptr()), config.instrument_gcov, pgo_sample_use_path.as_ref().map_or(std::ptr::null(), |s| s.as_ptr()), config.debug_info_for_profiling, @@ -513,18 +508,11 @@ pub(crate) unsafe fn optimize( let llmod = module.module_llvm.llmod(); let llcx = &*module.module_llvm.llcx; - let tm = &*module.module_llvm.tm; let _handlers = DiagnosticHandlers::new(cgcx, diag_handler, llcx); let module_name = module.name.clone(); let module_name = Some(&module_name[..]); - if let Some(false) = config.new_llvm_pass_manager && llvm_util::get_version() >= (15, 0, 0) { - diag_handler.warn( - "ignoring `-Z new-llvm-pass-manager=no`, which is no longer supported with LLVM 15", - ); - } - if config.emit_no_opt_bc { let out = cgcx.output_filenames.temp_path_ext("no-opt.bc", module_name); let out = path_to_c_string(&out); @@ -532,191 +520,24 @@ pub(crate) unsafe fn optimize( } if let Some(opt_level) = config.opt_level { - if llvm_util::should_use_new_llvm_pass_manager( - &config.new_llvm_pass_manager, - &cgcx.target_arch, - ) { - let opt_stage = match cgcx.lto { - Lto::Fat => llvm::OptStage::PreLinkFatLTO, - Lto::Thin | Lto::ThinLocal => llvm::OptStage::PreLinkThinLTO, - _ if cgcx.opts.cg.linker_plugin_lto.enabled() => llvm::OptStage::PreLinkThinLTO, - _ => llvm::OptStage::PreLinkNoLTO, - }; - return optimize_with_new_llvm_pass_manager( - cgcx, - diag_handler, - module, - config, - opt_level, - opt_stage, - ); - } - - if cgcx.prof.llvm_recording_enabled() { - diag_handler - .warn("`-Z self-profile-events = llvm` requires `-Z new-llvm-pass-manager`"); - } - - // Create the two optimizing pass managers. These mirror what clang - // does, and are by populated by LLVM's default PassManagerBuilder. - // Each manager has a different set of passes, but they also share - // some common passes. - let fpm = llvm::LLVMCreateFunctionPassManagerForModule(llmod); - let mpm = llvm::LLVMCreatePassManager(); - - { - let find_pass = |pass_name: &str| { - let pass_name = SmallCStr::new(pass_name); - llvm::LLVMRustFindAndCreatePass(pass_name.as_ptr()) - }; - - if config.verify_llvm_ir { - // Verification should run as the very first pass. - llvm::LLVMRustAddPass(fpm, find_pass("verify").unwrap()); - } - - let mut extra_passes = Vec::new(); - let mut have_name_anon_globals_pass = false; - - for pass_name in &config.passes { - if pass_name == "lint" { - // Linting should also be performed early, directly on the generated IR. - llvm::LLVMRustAddPass(fpm, find_pass("lint").unwrap()); - continue; - } - - if let Some(pass) = find_pass(pass_name) { - extra_passes.push(pass); - } else { - diag_handler.warn(&format!("unknown pass `{}`, ignoring", pass_name)); - } - - if pass_name == "name-anon-globals" { - have_name_anon_globals_pass = true; - } - } - - // Instrumentation must be inserted before optimization, - // otherwise LLVM may optimize some functions away which - // breaks llvm-cov. - // - // This mirrors what Clang does in lib/CodeGen/BackendUtil.cpp. - if config.instrument_gcov { - llvm::LLVMRustAddPass(mpm, find_pass("insert-gcov-profiling").unwrap()); - } - if config.instrument_coverage { - llvm::LLVMRustAddPass(mpm, find_pass("instrprof").unwrap()); - } - if config.debug_info_for_profiling { - llvm::LLVMRustAddPass(mpm, find_pass("add-discriminators").unwrap()); - } - - add_sanitizer_passes(config, &mut extra_passes); - - // Some options cause LLVM bitcode to be emitted, which uses ThinLTOBuffers, so we need - // to make sure we run LLVM's NameAnonGlobals pass when emitting bitcode; otherwise - // we'll get errors in LLVM. - let using_thin_buffers = config.bitcode_needed(); - if !config.no_prepopulate_passes { - llvm::LLVMAddAnalysisPasses(tm, fpm); - llvm::LLVMAddAnalysisPasses(tm, mpm); - let opt_level = to_llvm_opt_settings(opt_level).0; - let prepare_for_thin_lto = cgcx.lto == Lto::Thin - || cgcx.lto == Lto::ThinLocal - || (cgcx.lto != Lto::Fat && cgcx.opts.cg.linker_plugin_lto.enabled()); - with_llvm_pmb(llmod, config, opt_level, prepare_for_thin_lto, &mut |b| { - llvm::LLVMRustAddLastExtensionPasses( - b, - extra_passes.as_ptr(), - extra_passes.len() as size_t, - ); - llvm::LLVMRustPassManagerBuilderPopulateFunctionPassManager(b, fpm); - llvm::LLVMRustPassManagerBuilderPopulateModulePassManager(b, mpm); - }); - - have_name_anon_globals_pass = have_name_anon_globals_pass || prepare_for_thin_lto; - if using_thin_buffers && !prepare_for_thin_lto { - llvm::LLVMRustAddPass(mpm, find_pass("name-anon-globals").unwrap()); - have_name_anon_globals_pass = true; - } - } else { - // If we don't use the standard pipeline, directly populate the MPM - // with the extra passes. - for pass in extra_passes { - llvm::LLVMRustAddPass(mpm, pass); - } - } - - if using_thin_buffers && !have_name_anon_globals_pass { - // As described above, this will probably cause an error in LLVM - if config.no_prepopulate_passes { - diag_handler.err( - "The current compilation is going to use thin LTO buffers \ - without running LLVM's NameAnonGlobals pass. \ - This will likely cause errors in LLVM. Consider adding \ - -C passes=name-anon-globals to the compiler command line.", - ); - } else { - bug!( - "We are using thin LTO buffers without running the NameAnonGlobals pass. \ - This will likely cause errors in LLVM and should never happen." - ); - } - } - } - - diag_handler.abort_if_errors(); - - // Finally, run the actual optimization passes - { - let _timer = cgcx.prof.extra_verbose_generic_activity( - "LLVM_module_optimize_function_passes", - &*module.name, - ); - llvm::LLVMRustRunFunctionPassManager(fpm, llmod); - } - { - let _timer = cgcx.prof.extra_verbose_generic_activity( - "LLVM_module_optimize_module_passes", - &*module.name, - ); - llvm::LLVMRunPassManager(mpm, llmod); - } - - // Deallocate managers that we're now done with - llvm::LLVMDisposePassManager(fpm); - llvm::LLVMDisposePassManager(mpm); + let opt_stage = match cgcx.lto { + Lto::Fat => llvm::OptStage::PreLinkFatLTO, + Lto::Thin | Lto::ThinLocal => llvm::OptStage::PreLinkThinLTO, + _ if cgcx.opts.cg.linker_plugin_lto.enabled() => llvm::OptStage::PreLinkThinLTO, + _ => llvm::OptStage::PreLinkNoLTO, + }; + return llvm_optimize(cgcx, diag_handler, module, config, opt_level, opt_stage); } Ok(()) } -unsafe fn add_sanitizer_passes(config: &ModuleConfig, passes: &mut Vec<&'static mut llvm::Pass>) { - if config.sanitizer.contains(SanitizerSet::ADDRESS) { - let recover = config.sanitizer_recover.contains(SanitizerSet::ADDRESS); - passes.push(llvm::LLVMRustCreateAddressSanitizerFunctionPass(recover)); - passes.push(llvm::LLVMRustCreateModuleAddressSanitizerPass(recover)); - } - if config.sanitizer.contains(SanitizerSet::MEMORY) { - let track_origins = config.sanitizer_memory_track_origins as c_int; - let recover = config.sanitizer_recover.contains(SanitizerSet::MEMORY); - passes.push(llvm::LLVMRustCreateMemorySanitizerPass(track_origins, recover)); - } - if config.sanitizer.contains(SanitizerSet::THREAD) { - passes.push(llvm::LLVMRustCreateThreadSanitizerPass()); - } - if config.sanitizer.contains(SanitizerSet::HWADDRESS) { - let recover = config.sanitizer_recover.contains(SanitizerSet::HWADDRESS); - passes.push(llvm::LLVMRustCreateHWAddressSanitizerPass(recover)); - } -} - pub(crate) fn link( cgcx: &CodegenContext<LlvmCodegenBackend>, diag_handler: &Handler, mut modules: Vec<ModuleCodegen<ModuleLlvm>>, ) -> Result<ModuleCodegen<ModuleLlvm>, FatalError> { use super::lto::{Linker, ModuleBuffer}; - // Sort the modules by name to ensure to ensure deterministic behavior. + // Sort the modules by name to ensure deterministic behavior. modules.sort_by(|a, b| a.name.cmp(&b.name)); let (first, elements) = modules.split_first().expect("Bug! modules must contain at least one module."); @@ -790,7 +611,7 @@ pub(crate) unsafe fn codegen( let _timer = cgcx .prof .generic_activity_with_arg("LLVM_module_codegen_make_bitcode", &*module.name); - let thin = ThinBuffer::new(llmod); + let thin = ThinBuffer::new(llmod, config.emit_thin_lto); let data = thin.data(); if let Some(bitcode_filename) = bc_out.file_name() { @@ -944,11 +765,21 @@ pub(crate) unsafe fn codegen( drop(handlers); } + // `.dwo` files are only emitted if: + // + // - Object files are being emitted (i.e. bitcode only or metadata only compilations will not + // produce dwarf objects, even if otherwise enabled) + // - Target supports Split DWARF + // - Split debuginfo is enabled + // - Split DWARF kind is `split` (i.e. debuginfo is split into `.dwo` files, not different + // sections in the `.o` files). + let dwarf_object_emitted = matches!(config.emit_obj, EmitObj::ObjectCode(_)) + && cgcx.target_can_use_split_dwarf + && cgcx.split_debuginfo != SplitDebuginfo::Off + && cgcx.split_dwarf_kind == SplitDwarfKind::Split; Ok(module.into_compiled_module( config.emit_obj != EmitObj::None, - cgcx.target_can_use_split_dwarf - && cgcx.split_debuginfo != SplitDebuginfo::Off - && cgcx.split_dwarf_kind == SplitDwarfKind::Split, + dwarf_object_emitted, config.emit_bc, &cgcx.output_filenames, )) @@ -1076,72 +907,6 @@ unsafe fn embed_bitcode( } } -pub unsafe fn with_llvm_pmb( - llmod: &llvm::Module, - config: &ModuleConfig, - opt_level: llvm::CodeGenOptLevel, - prepare_for_thin_lto: bool, - f: &mut dyn FnMut(&llvm::PassManagerBuilder), -) { - use std::ptr; - - // Create the PassManagerBuilder for LLVM. We configure it with - // reasonable defaults and prepare it to actually populate the pass - // manager. - let builder = llvm::LLVMRustPassManagerBuilderCreate(); - let opt_size = config.opt_size.map_or(llvm::CodeGenOptSizeNone, |x| to_llvm_opt_settings(x).1); - let inline_threshold = config.inline_threshold; - let pgo_gen_path = get_pgo_gen_path(config); - let pgo_use_path = get_pgo_use_path(config); - let pgo_sample_use_path = get_pgo_sample_use_path(config); - - llvm::LLVMRustConfigurePassManagerBuilder( - builder, - opt_level, - config.merge_functions, - config.vectorize_slp, - config.vectorize_loop, - prepare_for_thin_lto, - pgo_gen_path.as_ref().map_or(ptr::null(), |s| s.as_ptr()), - pgo_use_path.as_ref().map_or(ptr::null(), |s| s.as_ptr()), - pgo_sample_use_path.as_ref().map_or(ptr::null(), |s| s.as_ptr()), - opt_size as c_int, - ); - - llvm::LLVMRustAddBuilderLibraryInfo(builder, llmod, config.no_builtins); - - // Here we match what clang does (kinda). For O0 we only inline - // always-inline functions (but don't add lifetime intrinsics), at O1 we - // inline with lifetime intrinsics, and O2+ we add an inliner with a - // thresholds copied from clang. - match (opt_level, opt_size, inline_threshold) { - (.., Some(t)) => { - llvm::LLVMRustPassManagerBuilderUseInlinerWithThreshold(builder, t); - } - (llvm::CodeGenOptLevel::Aggressive, ..) => { - llvm::LLVMRustPassManagerBuilderUseInlinerWithThreshold(builder, 275); - } - (_, llvm::CodeGenOptSizeDefault, _) => { - llvm::LLVMRustPassManagerBuilderUseInlinerWithThreshold(builder, 75); - } - (_, llvm::CodeGenOptSizeAggressive, _) => { - llvm::LLVMRustPassManagerBuilderUseInlinerWithThreshold(builder, 25); - } - (llvm::CodeGenOptLevel::None, ..) => { - llvm::LLVMRustAddAlwaysInlinePass(builder, config.emit_lifetime_markers); - } - (llvm::CodeGenOptLevel::Less, ..) => { - llvm::LLVMRustAddAlwaysInlinePass(builder, config.emit_lifetime_markers); - } - (llvm::CodeGenOptLevel::Default, ..) => { - llvm::LLVMRustPassManagerBuilderUseInlinerWithThreshold(builder, 225); - } - } - - f(builder); - llvm::LLVMRustPassManagerBuilderDispose(builder); -} - // Create a `__imp_<symbol> = &symbol` global for every public static `symbol`. // This is required to satisfy `dllimport` references to static data in .rlibs // when using MSVC linker. We do this only for data, as linker can fix up diff --git a/compiler/rustc_codegen_llvm/src/base.rs b/compiler/rustc_codegen_llvm/src/base.rs index 86f92dc0239..5b2bbdb4bde 100644 --- a/compiler/rustc_codegen_llvm/src/base.rs +++ b/compiler/rustc_codegen_llvm/src/base.rs @@ -19,6 +19,8 @@ use crate::context::CodegenCx; use crate::llvm; use crate::value::Value; +use cstr::cstr; + use rustc_codegen_ssa::base::maybe_create_entry_wrapper; use rustc_codegen_ssa::mono_item::MonoItemExt; use rustc_codegen_ssa::traits::*; @@ -107,11 +109,14 @@ pub fn compile_codegen_unit(tcx: TyCtxt<'_>, cgu_name: Symbol) -> (ModuleCodegen } // Create the llvm.used and llvm.compiler.used variables. - if !cx.used_statics().borrow().is_empty() { - cx.create_used_variable() + if !cx.used_statics.borrow().is_empty() { + cx.create_used_variable_impl(cstr!("llvm.used"), &*cx.used_statics.borrow()); } - if !cx.compiler_used_statics().borrow().is_empty() { - cx.create_compiler_used_variable() + if !cx.compiler_used_statics.borrow().is_empty() { + cx.create_used_variable_impl( + cstr!("llvm.compiler.used"), + &*cx.compiler_used_statics.borrow(), + ); } // Run replace-all-uses-with for statics that need it. This must diff --git a/compiler/rustc_codegen_llvm/src/builder.rs b/compiler/rustc_codegen_llvm/src/builder.rs index c41a41980eb..77dd15ef4d8 100644 --- a/compiler/rustc_codegen_llvm/src/builder.rs +++ b/compiler/rustc_codegen_llvm/src/builder.rs @@ -1,15 +1,14 @@ +use crate::abi::FnAbiLlvmExt; use crate::attributes; use crate::common::Funclet; use crate::context::CodegenCx; -use crate::llvm::{self, BasicBlock, False}; -use crate::llvm::{AtomicOrdering, AtomicRmwBinOp, SynchronizationScope}; -use crate::llvm_util; +use crate::llvm::{self, AtomicOrdering, AtomicRmwBinOp, BasicBlock}; use crate::type_::Type; use crate::type_of::LayoutLlvmExt; use crate::value::Value; use cstr::cstr; use libc::{c_char, c_uint}; -use rustc_codegen_ssa::common::{IntPredicate, RealPredicate, TypeKind}; +use rustc_codegen_ssa::common::{IntPredicate, RealPredicate, SynchronizationScope, TypeKind}; use rustc_codegen_ssa::mir::operand::{OperandRef, OperandValue}; use rustc_codegen_ssa::mir::place::PlaceRef; use rustc_codegen_ssa::traits::*; @@ -28,7 +27,6 @@ use std::ffi::CStr; use std::iter; use std::ops::Deref; use std::ptr; -use tracing::debug; // All Builders must have an llfn associated with them #[must_use] @@ -217,6 +215,7 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> { fn invoke( &mut self, llty: &'ll Type, + fn_abi: Option<&FnAbi<'tcx, Ty<'tcx>>>, llfn: &'ll Value, args: &[&'ll Value], then: &'ll BasicBlock, @@ -229,7 +228,7 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> { let bundle = funclet.map(|funclet| funclet.bundle()); let bundle = bundle.as_ref().map(|b| &*b.raw); - unsafe { + let invoke = unsafe { llvm::LLVMRustBuildInvoke( self.llbuilder, llty, @@ -241,7 +240,11 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> { bundle, UNNAMED, ) + }; + if let Some(fn_abi) = fn_abi { + fn_abi.apply_attrs_callsite(self, invoke); } + invoke } fn unreachable(&mut self) { @@ -362,11 +365,14 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> { Int(I64) => "llvm.ssub.with.overflow.i64", Int(I128) => "llvm.ssub.with.overflow.i128", - Uint(U8) => "llvm.usub.with.overflow.i8", - Uint(U16) => "llvm.usub.with.overflow.i16", - Uint(U32) => "llvm.usub.with.overflow.i32", - Uint(U64) => "llvm.usub.with.overflow.i64", - Uint(U128) => "llvm.usub.with.overflow.i128", + Uint(_) => { + // Emit sub and icmp instead of llvm.usub.with.overflow. LLVM considers these + // to be the canonical form. It will attempt to reform llvm.usub.with.overflow + // in the backend if profitable. + let sub = self.sub(lhs, rhs); + let cmp = self.icmp(IntPredicate::IntULT, lhs, rhs); + return (sub, cmp); + } _ => unreachable!(), }, @@ -408,20 +414,17 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> { fn alloca(&mut self, ty: &'ll Type, align: Align) -> &'ll Value { let mut bx = Builder::with_cx(self.cx); bx.position_at_start(unsafe { llvm::LLVMGetFirstBasicBlock(self.llfn()) }); - bx.dynamic_alloca(ty, align) - } - - fn dynamic_alloca(&mut self, ty: &'ll Type, align: Align) -> &'ll Value { unsafe { - let alloca = llvm::LLVMBuildAlloca(self.llbuilder, ty, UNNAMED); + let alloca = llvm::LLVMBuildAlloca(bx.llbuilder, ty, UNNAMED); llvm::LLVMSetAlignment(alloca, align.bytes() as c_uint); alloca } } - fn array_alloca(&mut self, ty: &'ll Type, len: &'ll Value, align: Align) -> &'ll Value { + fn byte_array_alloca(&mut self, len: &'ll Value, align: Align) -> &'ll Value { unsafe { - let alloca = llvm::LLVMBuildArrayAlloca(self.llbuilder, ty, len, UNNAMED); + let alloca = + llvm::LLVMBuildArrayAlloca(self.llbuilder, self.cx().type_i8(), len, UNNAMED); llvm::LLVMSetAlignment(alloca, align.bytes() as c_uint); alloca } @@ -464,15 +467,15 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> { } } + #[instrument(level = "trace", skip(self))] fn load_operand(&mut self, place: PlaceRef<'tcx, &'ll Value>) -> OperandRef<'tcx, &'ll Value> { - debug!("PlaceRef::load: {:?}", place); - assert_eq!(place.llextra.is_some(), place.layout.is_unsized()); if place.layout.is_zst() { return OperandRef::new_zst(self, place.layout); } + #[instrument(level = "trace", skip(bx))] fn scalar_load_metadata<'a, 'll, 'tcx>( bx: &mut Builder<'a, 'll, 'tcx>, load: &'ll Value, @@ -553,15 +556,15 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> { } fn write_operand_repeatedly( - mut self, + &mut self, cg_elem: OperandRef<'tcx, &'ll Value>, count: u64, dest: PlaceRef<'tcx, &'ll Value>, - ) -> Self { + ) { let zero = self.const_usize(0); let count = self.const_usize(count); - let start = dest.project_index(&mut self, zero).llval; - let end = dest.project_index(&mut self, count).llval; + let start = dest.project_index(self, zero).llval; + let end = dest.project_index(self, count).llval; let header_bb = self.append_sibling_block("repeat_loop_header"); let body_bb = self.append_sibling_block("repeat_loop_body"); @@ -589,7 +592,7 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> { body_bx.br(header_bb); header_bx.add_incoming_to_phi(current, next, body_bb); - Self::build(self.cx, next_bb) + *self = Self::build(self.cx, next_bb); } fn range_metadata(&mut self, load: &'ll Value, range: WrappingRange) { @@ -626,32 +629,6 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> { } } - fn type_metadata(&mut self, function: &'ll Value, typeid: String) { - let typeid_metadata = self.typeid_metadata(typeid); - let v = [self.const_usize(0), typeid_metadata]; - unsafe { - llvm::LLVMGlobalSetMetadata( - function, - llvm::MD_type as c_uint, - llvm::LLVMValueAsMetadata(llvm::LLVMMDNodeInContext( - self.cx.llcx, - v.as_ptr(), - v.len() as c_uint, - )), - ) - } - } - - fn typeid_metadata(&mut self, typeid: String) -> Self::Value { - unsafe { - llvm::LLVMMDStringInContext( - self.cx.llcx, - typeid.as_ptr() as *const c_char, - typeid.as_bytes().len() as c_uint, - ) - } - } - fn store(&mut self, val: &'ll Value, ptr: &'ll Value, align: Align) -> &'ll Value { self.store_with_flags(val, ptr, align, MemFlags::empty()) } @@ -752,11 +729,11 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> { unsafe { llvm::LLVMBuildSExt(self.llbuilder, val, dest_ty, UNNAMED) } } - fn fptoui_sat(&mut self, val: &'ll Value, dest_ty: &'ll Type) -> Option<&'ll Value> { + fn fptoui_sat(&mut self, val: &'ll Value, dest_ty: &'ll Type) -> &'ll Value { self.fptoint_sat(false, val, dest_ty) } - fn fptosi_sat(&mut self, val: &'ll Value, dest_ty: &'ll Type) -> Option<&'ll Value> { + fn fptosi_sat(&mut self, val: &'ll Value, dest_ty: &'ll Type) -> &'ll Value { self.fptoint_sat(true, val, dest_ty) } @@ -1070,15 +1047,17 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> { ) -> &'ll Value { let weak = if weak { llvm::True } else { llvm::False }; unsafe { - llvm::LLVMRustBuildAtomicCmpXchg( + let value = llvm::LLVMBuildAtomicCmpXchg( self.llbuilder, dst, cmp, src, AtomicOrdering::from_generic(order), AtomicOrdering::from_generic(failure_order), - weak, - ) + llvm::False, // SingleThreaded + ); + llvm::LLVMSetWeak(value, weak); + value } } fn atomic_rmw( @@ -1095,7 +1074,7 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> { dst, src, AtomicOrdering::from_generic(order), - False, + llvm::False, // SingleThreaded ) } } @@ -1103,13 +1082,18 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> { fn atomic_fence( &mut self, order: rustc_codegen_ssa::common::AtomicOrdering, - scope: rustc_codegen_ssa::common::SynchronizationScope, + scope: SynchronizationScope, ) { + let single_threaded = match scope { + SynchronizationScope::SingleThread => llvm::True, + SynchronizationScope::CrossThread => llvm::False, + }; unsafe { - llvm::LLVMRustBuildAtomicFence( + llvm::LLVMBuildFence( self.llbuilder, AtomicOrdering::from_generic(order), - SynchronizationScope::from_generic(scope), + single_threaded, + UNNAMED, ); } } @@ -1167,6 +1151,7 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> { fn call( &mut self, llty: &'ll Type, + fn_abi: Option<&FnAbi<'tcx, Ty<'tcx>>>, llfn: &'ll Value, args: &[&'ll Value], funclet: Option<&Funclet<'ll>>, @@ -1177,7 +1162,7 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> { let bundle = funclet.map(|funclet| funclet.bundle()); let bundle = bundle.as_ref().map(|b| &*b.raw); - unsafe { + let call = unsafe { llvm::LLVMRustBuildCall( self.llbuilder, llty, @@ -1186,7 +1171,11 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> { args.len() as c_uint, bundle, ) + }; + if let Some(fn_abi) = fn_abi { + fn_abi.apply_attrs_callsite(self, call); } + call } fn zext(&mut self, val: &'ll Value, dest_ty: &'ll Type) -> &'ll Value { @@ -1419,7 +1408,7 @@ impl<'a, 'll, 'tcx> Builder<'a, 'll, 'tcx> { pub(crate) fn call_intrinsic(&mut self, intrinsic: &str, args: &[&'ll Value]) -> &'ll Value { let (ty, f) = self.cx.get_intrinsic(intrinsic); - self.call(ty, f, args, None) + self.call(ty, None, f, args, None) } fn call_lifetime_intrinsic(&mut self, intrinsic: &str, ptr: &'ll Value, size: Size) { @@ -1456,51 +1445,32 @@ impl<'a, 'll, 'tcx> Builder<'a, 'll, 'tcx> { } } - fn fptoint_sat_broken_in_llvm(&self) -> bool { - match self.tcx.sess.target.arch.as_ref() { - // FIXME - https://bugs.llvm.org/show_bug.cgi?id=50083 - "riscv64" => llvm_util::get_version() < (13, 0, 0), - _ => false, - } - } - - fn fptoint_sat( - &mut self, - signed: bool, - val: &'ll Value, - dest_ty: &'ll Type, - ) -> Option<&'ll Value> { - if !self.fptoint_sat_broken_in_llvm() { - let src_ty = self.cx.val_ty(val); - let (float_ty, int_ty, vector_length) = if self.cx.type_kind(src_ty) == TypeKind::Vector - { - assert_eq!(self.cx.vector_length(src_ty), self.cx.vector_length(dest_ty)); - ( - self.cx.element_type(src_ty), - self.cx.element_type(dest_ty), - Some(self.cx.vector_length(src_ty)), - ) - } else { - (src_ty, dest_ty, None) - }; - let float_width = self.cx.float_width(float_ty); - let int_width = self.cx.int_width(int_ty); - - let instr = if signed { "fptosi" } else { "fptoui" }; - let name = if let Some(vector_length) = vector_length { - format!( - "llvm.{}.sat.v{}i{}.v{}f{}", - instr, vector_length, int_width, vector_length, float_width - ) - } else { - format!("llvm.{}.sat.i{}.f{}", instr, int_width, float_width) - }; - let f = - self.declare_cfn(&name, llvm::UnnamedAddr::No, self.type_func(&[src_ty], dest_ty)); - Some(self.call(self.type_func(&[src_ty], dest_ty), f, &[val], None)) + fn fptoint_sat(&mut self, signed: bool, val: &'ll Value, dest_ty: &'ll Type) -> &'ll Value { + let src_ty = self.cx.val_ty(val); + let (float_ty, int_ty, vector_length) = if self.cx.type_kind(src_ty) == TypeKind::Vector { + assert_eq!(self.cx.vector_length(src_ty), self.cx.vector_length(dest_ty)); + ( + self.cx.element_type(src_ty), + self.cx.element_type(dest_ty), + Some(self.cx.vector_length(src_ty)), + ) } else { - None - } + (src_ty, dest_ty, None) + }; + let float_width = self.cx.float_width(float_ty); + let int_width = self.cx.int_width(int_ty); + + let instr = if signed { "fptosi" } else { "fptoui" }; + let name = if let Some(vector_length) = vector_length { + format!( + "llvm.{}.sat.v{}i{}.v{}f{}", + instr, vector_length, int_width, vector_length, float_width + ) + } else { + format!("llvm.{}.sat.i{}.f{}", instr, int_width, float_width) + }; + let f = self.declare_cfn(&name, llvm::UnnamedAddr::No, self.type_func(&[src_ty], dest_ty)); + self.call(self.type_func(&[src_ty], dest_ty), None, f, &[val], None) } pub(crate) fn landing_pad( diff --git a/compiler/rustc_codegen_llvm/src/callee.rs b/compiler/rustc_codegen_llvm/src/callee.rs index ac423a22703..70ff5c9617b 100644 --- a/compiler/rustc_codegen_llvm/src/callee.rs +++ b/compiler/rustc_codegen_llvm/src/callee.rs @@ -6,14 +6,14 @@ use crate::abi::FnAbiLlvmExt; use crate::attributes; +use crate::common; use crate::context::CodegenCx; use crate::llvm; use crate::value::Value; use rustc_codegen_ssa::traits::*; -use tracing::debug; use rustc_middle::ty::layout::{FnAbiOf, HasTyCtxt}; -use rustc_middle::ty::{self, Instance, TypeFoldable}; +use rustc_middle::ty::{self, Instance, TypeVisitable}; /// Codegens a reference to a fn/method item, monomorphizing and /// inlining as it goes. @@ -79,13 +79,31 @@ pub fn get_fn<'ll, 'tcx>(cx: &CodegenCx<'ll, 'tcx>, instance: Instance<'tcx>) -> llfn } } else { - let llfn = cx.declare_fn(sym, fn_abi); + let instance_def_id = instance.def_id(); + let llfn = if tcx.sess.target.arch == "x86" && + let Some(dllimport) = common::get_dllimport(tcx, instance_def_id, sym) + { + // Fix for https://github.com/rust-lang/rust/issues/104453 + // On x86 Windows, LLVM uses 'L' as the prefix for any private + // global symbols, so when we create an undecorated function symbol + // that begins with an 'L' LLVM misinterprets that as a private + // global symbol that it created and so fails the compilation at a + // later stage since such a symbol must have a definition. + // + // To avoid this, we set the Storage Class to "DllImport" so that + // LLVM will prefix the name with `__imp_`. Ideally, we'd like the + // existing logic below to set the Storage Class, but it has an + // exemption for MinGW for backwards compatability. + let llfn = cx.declare_fn(&common::i686_decorated_name(&dllimport, common::is_mingw_gnu_toolchain(&tcx.sess.target), true), fn_abi); + unsafe { llvm::LLVMSetDLLStorageClass(llfn, llvm::DLLStorageClass::DllImport); } + llfn + } else { + cx.declare_fn(sym, fn_abi) + }; debug!("get_fn: not casting pointer!"); attributes::from_fn_attrs(cx, llfn, instance); - let instance_def_id = instance.def_id(); - // Apply an appropriate linkage/visibility value to our item that we // just declared. // @@ -174,7 +192,8 @@ pub fn get_fn<'ll, 'tcx>(cx: &CodegenCx<'ll, 'tcx>, instance: Instance<'tcx>) -> // MinGW: For backward compatibility we rely on the linker to decide whether it // should use dllimport for functions. if cx.use_dll_storage_attrs - && tcx.is_dllimport_foreign_item(instance_def_id) + && let Some(library) = tcx.native_library(instance_def_id) + && library.kind.is_dllimport() && !matches!(tcx.sess.target.env.as_ref(), "gnu" | "uclibc") { llvm::LLVMSetDLLStorageClass(llfn, llvm::DLLStorageClass::DllImport); diff --git a/compiler/rustc_codegen_llvm/src/common.rs b/compiler/rustc_codegen_llvm/src/common.rs index b69d7a000ee..acee9134fb9 100644 --- a/compiler/rustc_codegen_llvm/src/common.rs +++ b/compiler/rustc_codegen_llvm/src/common.rs @@ -10,15 +10,17 @@ use crate::value::Value; use rustc_ast::Mutability; use rustc_codegen_ssa::mir::place::PlaceRef; use rustc_codegen_ssa::traits::*; +use rustc_hir::def_id::DefId; use rustc_middle::bug; use rustc_middle::mir::interpret::{ConstAllocation, GlobalAlloc, Scalar}; use rustc_middle::ty::layout::{LayoutOf, TyAndLayout}; -use rustc_middle::ty::ScalarInt; -use rustc_span::symbol::Symbol; +use rustc_middle::ty::TyCtxt; +use rustc_session::cstore::{DllCallingConvention, DllImport, PeImportNameType}; use rustc_target::abi::{self, AddressSpace, HasDataLayout, Pointer, Size}; +use rustc_target::spec::Target; use libc::{c_char, c_uint}; -use tracing::debug; +use std::fmt::Write; /* * A note on nomenclature of linking: "extern", "foreign", and "upcall". @@ -109,8 +111,7 @@ impl<'ll> CodegenCx<'ll, '_> { pub fn const_get_elt(&self, v: &'ll Value, idx: u64) -> &'ll Value { unsafe { assert_eq!(idx as c_uint as u64, idx); - let us = &[idx as c_uint]; - let r = llvm::LLVMConstExtractValue(v, us.as_ptr(), us.len() as c_uint); + let r = llvm::LLVMGetAggregateElement(v, idx as c_uint).unwrap(); debug!("const_get_elt(v={:?}, idx={}, r={:?})", v, idx, r); @@ -181,22 +182,27 @@ impl<'ll, 'tcx> ConstMethods<'tcx> for CodegenCx<'ll, 'tcx> { unsafe { llvm::LLVMConstReal(t, val) } } - fn const_str(&self, s: Symbol) -> (&'ll Value, &'ll Value) { - let s_str = s.as_str(); - let str_global = *self.const_str_cache.borrow_mut().entry(s).or_insert_with(|| { - let sc = self.const_bytes(s_str.as_bytes()); - let sym = self.generate_local_symbol_name("str"); - let g = self.define_global(&sym, self.val_ty(sc)).unwrap_or_else(|| { - bug!("symbol `{}` is already defined", sym); - }); - unsafe { - llvm::LLVMSetInitializer(g, sc); - llvm::LLVMSetGlobalConstant(g, True); - llvm::LLVMRustSetLinkage(g, llvm::Linkage::InternalLinkage); - } - g - }); - let len = s_str.len(); + fn const_str(&self, s: &str) -> (&'ll Value, &'ll Value) { + let str_global = *self + .const_str_cache + .borrow_mut() + .raw_entry_mut() + .from_key(s) + .or_insert_with(|| { + let sc = self.const_bytes(s.as_bytes()); + let sym = self.generate_local_symbol_name("str"); + let g = self.define_global(&sym, self.val_ty(sc)).unwrap_or_else(|| { + bug!("symbol `{}` is already defined", sym); + }); + unsafe { + llvm::LLVMSetInitializer(g, sc); + llvm::LLVMSetGlobalConstant(g, True); + llvm::LLVMRustSetLinkage(g, llvm::Linkage::InternalLinkage); + } + (s.to_owned(), g) + }) + .1; + let len = s.len(); let cs = consts::ptrcast( str_global, self.type_ptr_to(self.layout_of(self.tcx.types.str_).llvm_type(self)), @@ -209,7 +215,11 @@ impl<'ll, 'tcx> ConstMethods<'tcx> for CodegenCx<'ll, 'tcx> { } fn const_to_opt_uint(&self, v: &'ll Value) -> Option<u64> { - try_as_const_integral(v).map(|v| unsafe { llvm::LLVMConstIntGetZExtValue(v) }) + try_as_const_integral(v).and_then(|v| unsafe { + let mut i = 0u64; + let success = llvm::LLVMRustConstIntGetZExtValue(v, &mut i); + success.then_some(i) + }) } fn const_to_opt_u128(&self, v: &'ll Value, sign_ext: bool) -> Option<u128> { @@ -223,10 +233,6 @@ impl<'ll, 'tcx> ConstMethods<'tcx> for CodegenCx<'ll, 'tcx> { fn scalar_to_backend(&self, cv: Scalar, layout: abi::Scalar, llty: &'ll Type) -> &'ll Value { let bitsize = if layout.is_bool() { 1 } else { layout.size(self).bits() }; match cv { - Scalar::Int(ScalarInt::ZST) => { - assert_eq!(0, layout.size(self).bytes()); - self.const_undef(self.type_ix(0)) - } Scalar::Int(int) => { let data = int.assert_bits(layout.size(self)); let llval = self.const_uint_big(self.type_ix(bitsize), data); @@ -255,6 +261,15 @@ impl<'ll, 'tcx> ConstMethods<'tcx> for CodegenCx<'ll, 'tcx> { self.get_fn_addr(fn_instance.polymorphize(self.tcx)), self.data_layout().instruction_address_space, ), + GlobalAlloc::VTable(ty, trait_ref) => { + let alloc = self + .tcx + .global_alloc(self.tcx.vtable_allocation((ty, trait_ref))) + .unwrap_memory(); + let init = const_alloc_to_llvm(self, alloc); + let value = self.static_addr_of(init, alloc.inner().align, None); + (value, AddressSpace::DATA) + } GlobalAlloc::Static(def_id) => { assert!(self.tcx.is_static(def_id)); assert!(!self.tcx.is_thread_local_static(def_id)); @@ -346,3 +361,74 @@ fn hi_lo_to_u128(lo: u64, hi: u64) -> u128 { fn try_as_const_integral(v: &Value) -> Option<&ConstantInt> { unsafe { llvm::LLVMIsAConstantInt(v) } } + +pub(crate) fn get_dllimport<'tcx>( + tcx: TyCtxt<'tcx>, + id: DefId, + name: &str, +) -> Option<&'tcx DllImport> { + tcx.native_library(id) + .map(|lib| lib.dll_imports.iter().find(|di| di.name.as_str() == name)) + .flatten() +} + +pub(crate) fn is_mingw_gnu_toolchain(target: &Target) -> bool { + target.vendor == "pc" && target.os == "windows" && target.env == "gnu" && target.abi.is_empty() +} + +pub(crate) fn i686_decorated_name( + dll_import: &DllImport, + mingw: bool, + disable_name_mangling: bool, +) -> String { + let name = dll_import.name.as_str(); + + let (add_prefix, add_suffix) = match dll_import.import_name_type { + Some(PeImportNameType::NoPrefix) => (false, true), + Some(PeImportNameType::Undecorated) => (false, false), + _ => (true, true), + }; + + // Worst case: +1 for disable name mangling, +1 for prefix, +4 for suffix (@@__). + let mut decorated_name = String::with_capacity(name.len() + 6); + + if disable_name_mangling { + // LLVM uses a binary 1 ('\x01') prefix to a name to indicate that mangling needs to be disabled. + decorated_name.push('\x01'); + } + + let prefix = if add_prefix && dll_import.is_fn { + match dll_import.calling_convention { + DllCallingConvention::C | DllCallingConvention::Vectorcall(_) => None, + DllCallingConvention::Stdcall(_) => (!mingw + || dll_import.import_name_type == Some(PeImportNameType::Decorated)) + .then_some('_'), + DllCallingConvention::Fastcall(_) => Some('@'), + } + } else if !dll_import.is_fn && !mingw { + // For static variables, prefix with '_' on MSVC. + Some('_') + } else { + None + }; + if let Some(prefix) = prefix { + decorated_name.push(prefix); + } + + decorated_name.push_str(name); + + if add_suffix && dll_import.is_fn { + match dll_import.calling_convention { + DllCallingConvention::C => {} + DllCallingConvention::Stdcall(arg_list_size) + | DllCallingConvention::Fastcall(arg_list_size) => { + write!(&mut decorated_name, "@{}", arg_list_size).unwrap(); + } + DllCallingConvention::Vectorcall(arg_list_size) => { + write!(&mut decorated_name, "@@{}", arg_list_size).unwrap(); + } + } + } + + decorated_name +} diff --git a/compiler/rustc_codegen_llvm/src/consts.rs b/compiler/rustc_codegen_llvm/src/consts.rs index 5bbbfe9a4ab..3626aa901c0 100644 --- a/compiler/rustc_codegen_llvm/src/consts.rs +++ b/compiler/rustc_codegen_llvm/src/consts.rs @@ -1,6 +1,7 @@ use crate::base; -use crate::common::CodegenCx; +use crate::common::{self, CodegenCx}; use crate::debuginfo; +use crate::errors::{InvalidMinimumAlignment, SymbolAlreadyDefined}; use crate::llvm::{self, True}; use crate::llvm_util; use crate::type_::Type; @@ -19,29 +20,27 @@ use rustc_middle::mir::mono::MonoItem; use rustc_middle::ty::layout::LayoutOf; use rustc_middle::ty::{self, Instance, Ty}; use rustc_middle::{bug, span_bug}; +use rustc_session::config::Lto; use rustc_target::abi::{ AddressSpace, Align, HasDataLayout, Primitive, Scalar, Size, WrappingRange, }; use std::ops::Range; -use tracing::debug; pub fn const_alloc_to_llvm<'ll>(cx: &CodegenCx<'ll, '_>, alloc: ConstAllocation<'_>) -> &'ll Value { let alloc = alloc.inner(); - let mut llvals = Vec::with_capacity(alloc.relocations().len() + 1); + let mut llvals = Vec::with_capacity(alloc.provenance().ptrs().len() + 1); let dl = cx.data_layout(); let pointer_size = dl.pointer_size.bytes() as usize; - // Note: this function may call `inspect_with_uninit_and_ptr_outside_interpreter`, - // so `range` must be within the bounds of `alloc` and not contain or overlap a relocation. + // Note: this function may call `inspect_with_uninit_and_ptr_outside_interpreter`, so `range` + // must be within the bounds of `alloc` and not contain or overlap a pointer provenance. fn append_chunks_of_init_and_uninit_bytes<'ll, 'a, 'b>( llvals: &mut Vec<&'ll Value>, cx: &'a CodegenCx<'ll, 'b>, alloc: &'a Allocation, range: Range<usize>, ) { - let chunks = alloc - .init_mask() - .range_as_init_chunks(Size::from_bytes(range.start), Size::from_bytes(range.end)); + let chunks = alloc.init_mask().range_as_init_chunks(range.clone().into()); let chunk_to_llval = move |chunk| match chunk { InitChunk::Init(range) => { @@ -64,7 +63,7 @@ pub fn const_alloc_to_llvm<'ll>(cx: &CodegenCx<'ll, '_>, alloc: ConstAllocation< // See https://github.com/rust-lang/rust/issues/84565. 1 } else { - cx.sess().opts.debugging_opts.uninit_const_chunk_threshold + cx.sess().opts.unstable_opts.uninit_const_chunk_threshold }; let allow_uninit_chunks = chunks.clone().take(max.saturating_add(1)).count() <= max; @@ -79,12 +78,12 @@ pub fn const_alloc_to_llvm<'ll>(cx: &CodegenCx<'ll, '_>, alloc: ConstAllocation< } let mut next_offset = 0; - for &(offset, alloc_id) in alloc.relocations().iter() { + for &(offset, alloc_id) in alloc.provenance().ptrs().iter() { let offset = offset.bytes(); assert_eq!(offset as usize as u64, offset); let offset = offset as usize; if offset > next_offset { - // This `inspect` is okay since we have checked that it is not within a relocation, it + // This `inspect` is okay since we have checked that there is no provenance, it // is within the bounds of the allocation, and it doesn't affect interpreter execution // (we inspect the result after interpreter execution). append_chunks_of_init_and_uninit_bytes(&mut llvals, cx, alloc, next_offset..offset); @@ -93,7 +92,7 @@ pub fn const_alloc_to_llvm<'ll>(cx: &CodegenCx<'ll, '_>, alloc: ConstAllocation< dl.endian, // This `inspect` is okay since it is within the bounds of the allocation, it doesn't // affect interpreter execution (we inspect the result after interpreter execution), - // and we properly interpret the relocation as a relocation pointer offset. + // and we properly interpret the provenance as a relocation pointer offset. alloc.inspect_with_uninit_and_ptr_outside_interpreter(offset..(offset + pointer_size)), ) .expect("const_alloc_to_llvm: could not read relocation pointer") @@ -101,7 +100,9 @@ pub fn const_alloc_to_llvm<'ll>(cx: &CodegenCx<'ll, '_>, alloc: ConstAllocation< let address_space = match cx.tcx.global_alloc(alloc_id) { GlobalAlloc::Function(..) => cx.data_layout().instruction_address_space, - GlobalAlloc::Static(..) | GlobalAlloc::Memory(..) => AddressSpace::DATA, + GlobalAlloc::Static(..) | GlobalAlloc::Memory(..) | GlobalAlloc::VTable(..) => { + AddressSpace::DATA + } }; llvals.push(cx.scalar_to_backend( @@ -119,7 +120,7 @@ pub fn const_alloc_to_llvm<'ll>(cx: &CodegenCx<'ll, '_>, alloc: ConstAllocation< } if alloc.len() >= next_offset { let range = next_offset..alloc.len(); - // This `inspect` is okay since we have check that it is after all relocations, it is + // This `inspect` is okay since we have check that it is after all provenance, it is // within the bounds of the allocation, and it doesn't affect interpreter execution (we // inspect the result after interpreter execution). append_chunks_of_init_and_uninit_bytes(&mut llvals, cx, alloc, range); @@ -144,7 +145,7 @@ fn set_global_alignment<'ll>(cx: &CodegenCx<'ll, '_>, gv: &'ll Value, mut align: match Align::from_bits(min) { Ok(min) => align = align.max(min), Err(err) => { - cx.sess().err(&format!("invalid minimum global alignment: {}", err)); + cx.sess().emit_err(InvalidMinimumAlignment { err }); } } } @@ -158,28 +159,15 @@ fn check_and_apply_linkage<'ll, 'tcx>( attrs: &CodegenFnAttrs, ty: Ty<'tcx>, sym: &str, - span_def_id: DefId, + def_id: DefId, ) -> &'ll Value { let llty = cx.layout_of(ty).llvm_type(cx); - if let Some(linkage) = attrs.linkage { + if let Some(linkage) = attrs.import_linkage { debug!("get_static: sym={} linkage={:?}", sym, linkage); - // If this is a static with a linkage specified, then we need to handle - // it a little specially. The typesystem prevents things like &T and - // extern "C" fn() from being non-null, so we can't just declare a - // static and call it a day. Some linkages (like weak) will make it such - // that the static actually has a null value. - let llty2 = if let ty::RawPtr(ref mt) = ty.kind() { - cx.layout_of(mt.ty).llvm_type(cx) - } else { - cx.sess().span_fatal( - cx.tcx.def_span(span_def_id), - "must have type `*const T` or `*mut T` due to `#[linkage]` attribute", - ) - }; unsafe { // Declare a symbol `foo` with the desired linkage. - let g1 = cx.declare_global(sym, llty2); + let g1 = cx.declare_global(sym, cx.type_i8()); llvm::LLVMRustSetLinkage(g1, base::linkage_to_llvm(linkage)); // Declare an internal global `extern_with_linkage_foo` which @@ -191,15 +179,19 @@ fn check_and_apply_linkage<'ll, 'tcx>( let mut real_name = "_rust_extern_with_linkage_".to_string(); real_name.push_str(sym); let g2 = cx.define_global(&real_name, llty).unwrap_or_else(|| { - cx.sess().span_fatal( - cx.tcx.def_span(span_def_id), - &format!("symbol `{}` is already defined", &sym), - ) + cx.sess().emit_fatal(SymbolAlreadyDefined { + span: cx.tcx.def_span(def_id), + symbol_name: sym, + }) }); llvm::LLVMRustSetLinkage(g2, llvm::Linkage::InternalLinkage); - llvm::LLVMSetInitializer(g2, g1); + llvm::LLVMSetInitializer(g2, cx.const_ptrcast(g1, llty)); g2 } + } else if cx.tcx.sess.target.arch == "x86" && + let Some(dllimport) = common::get_dllimport(cx.tcx, def_id, sym) + { + cx.declare_global(&common::i686_decorated_name(&dllimport, common::is_mingw_gnu_toolchain(&cx.tcx.sess.target), true), llty) } else { // Generate an external declaration. // FIXME(nagisa): investigate whether it can be changed into define_global @@ -293,12 +285,23 @@ impl<'ll> CodegenCx<'ll, '_> { llvm::set_thread_local_mode(g, self.tls_model); } + let dso_local = unsafe { self.should_assume_dso_local(g, true) }; + if dso_local { + unsafe { + llvm::LLVMRustSetDSOLocal(g, true); + } + } + if !def_id.is_local() { let needs_dll_storage_attr = self.use_dll_storage_attrs && !self.tcx.is_foreign_item(def_id) && + // Local definitions can never be imported, so we must not apply + // the DLLImport annotation. + !dso_local && // ThinLTO can't handle this workaround in all cases, so we don't // emit the attrs. Instead we make them unnecessary by disallowing // dynamic linking when linker plugin based LTO is enabled. - !self.tcx.sess.opts.cg.linker_plugin_lto.enabled(); + !self.tcx.sess.opts.cg.linker_plugin_lto.enabled() && + self.tcx.sess.lto() != Lto::Thin; // If this assertion triggers, there's something wrong with commandline // argument validation. @@ -327,19 +330,16 @@ impl<'ll> CodegenCx<'ll, '_> { } } - if self.use_dll_storage_attrs && self.tcx.is_dllimport_foreign_item(def_id) { + if self.use_dll_storage_attrs + && let Some(library) = self.tcx.native_library(def_id) + && library.kind.is_dllimport() + { // For foreign (native) libs we know the exact storage type to use. unsafe { llvm::LLVMSetDLLStorageClass(g, llvm::DLLStorageClass::DllImport); } } - unsafe { - if self.should_assume_dso_local(g, true) { - llvm::LLVMRustSetDSOLocal(g, true); - } - } - self.instances.borrow_mut().insert(instance, g); g } @@ -473,7 +473,7 @@ impl<'ll> StaticMethods for CodegenCx<'ll, '_> { // // We could remove this hack whenever we decide to drop macOS 10.10 support. if self.tcx.sess.target.is_like_osx { - // The `inspect` method is okay here because we checked relocations, and + // The `inspect` method is okay here because we checked for provenance, and // because we are doing this access to inspect the final interpreter state // (not as part of the interpreter execution). // @@ -481,7 +481,7 @@ impl<'ll> StaticMethods for CodegenCx<'ll, '_> { // happens to be zero. Instead, we should only check the value of defined bytes // and set all undefined bytes to zero if this allocation is headed for the // BSS. - let all_bytes_are_zero = alloc.relocations().is_empty() + let all_bytes_are_zero = alloc.provenance().ptrs().is_empty() && alloc .inspect_with_uninit_and_ptr_outside_interpreter(0..alloc.len()) .iter() @@ -505,9 +505,9 @@ impl<'ll> StaticMethods for CodegenCx<'ll, '_> { section.as_str().as_ptr().cast(), section.as_str().len() as c_uint, ); - assert!(alloc.relocations().is_empty()); + assert!(alloc.provenance().ptrs().is_empty()); - // The `inspect` method is okay here because we checked relocations, and + // The `inspect` method is okay here because we checked for provenance, and // because we are doing this access to inspect the final interpreter state (not // as part of the interpreter execution). let bytes = @@ -535,10 +535,20 @@ impl<'ll> StaticMethods for CodegenCx<'ll, '_> { // The semantics of #[used] in Rust only require the symbol to make it into the // object file. It is explicitly allowed for the linker to strip the symbol if it - // is dead. As such, use llvm.compiler.used instead of llvm.used. + // is dead, which means we are allowed use `llvm.compiler.used` instead of + // `llvm.used` here. + // // Additionally, https://reviews.llvm.org/D97448 in LLVM 13 started emitting unique // sections with SHF_GNU_RETAIN flag for llvm.used symbols, which may trigger bugs - // in some versions of the gold linker. + // in the handling of `.init_array` (the static constructor list) in versions of + // the gold linker (prior to the one released with binutils 2.36). + // + // That said, we only ever emit these when compiling for ELF targets, unless + // `#[used(compiler)]` is explicitly requested. This is to avoid similar breakage + // on other targets, in particular MachO targets have *their* static constructor + // lists broken if `llvm.compiler.used` is emitted rather than llvm.used. However, + // that check happens when assigning the `CodegenFnAttrFlags` in `rustc_hir_analysis`, + // so we don't need to take care of it here. self.add_compiler_used_global(g); } if attrs.flags.contains(CodegenFnAttrFlags::USED_LINKER) { diff --git a/compiler/rustc_codegen_llvm/src/context.rs b/compiler/rustc_codegen_llvm/src/context.rs index b5c31fcebe0..4dcc7cd5447 100644 --- a/compiler/rustc_codegen_llvm/src/context.rs +++ b/compiler/rustc_codegen_llvm/src/context.rs @@ -3,6 +3,7 @@ use crate::back::write::to_llvm_code_model; use crate::callee::get_fn; use crate::coverageinfo; use crate::debuginfo; +use crate::errors::BranchProtectionRequiresAArch64; use crate::llvm; use crate::llvm_util; use crate::type_::Type; @@ -26,7 +27,7 @@ use rustc_session::config::{BranchProtection, CFGuard, CFProtection}; use rustc_session::config::{CrateType, DebugInfo, PAuthKey, PacRet}; use rustc_session::Session; use rustc_span::source_map::Span; -use rustc_span::symbol::Symbol; +use rustc_span::source_map::Spanned; use rustc_target::abi::{ call::FnAbi, HasDataLayout, PointeeInfo, Size, TargetDataLayout, VariantIdx, }; @@ -56,7 +57,7 @@ pub struct CodegenCx<'ll, 'tcx> { pub vtables: RefCell<FxHashMap<(Ty<'tcx>, Option<ty::PolyExistentialTraitRef<'tcx>>), &'ll Value>>, /// Cache of constant strings, - pub const_str_cache: RefCell<FxHashMap<Symbol, &'ll Value>>, + pub const_str_cache: RefCell<FxHashMap<String, &'ll Value>>, /// Reverse-direction for const ptrs cast from globals. /// @@ -143,17 +144,6 @@ pub unsafe fn create_module<'ll>( let mut target_data_layout = sess.target.data_layout.to_string(); let llvm_version = llvm_util::get_version(); - if llvm_version < (13, 0, 0) { - if sess.target.arch == "powerpc64" { - target_data_layout = target_data_layout.replace("-S128", ""); - } - if sess.target.arch == "wasm32" { - target_data_layout = "e-m:e-p:32:32-i64:64-n32:64-S128".to_string(); - } - if sess.target.arch == "wasm64" { - target_data_layout = "e-m:e-p:64:64-i64:64-n32:64-S128".to_string(); - } - } if llvm_version < (14, 0, 0) { if sess.target.llvm_target == "i686-pc-windows-msvc" || sess.target.llvm_target == "i586-pc-windows-msvc" @@ -166,6 +156,15 @@ pub unsafe fn create_module<'ll>( target_data_layout = target_data_layout.replace("-p10:8:8-p20:8:8", ""); } } + if llvm_version < (16, 0, 0) { + if sess.target.arch == "s390x" { + target_data_layout = target_data_layout.replace("-v128:64", ""); + } + + if sess.target.arch == "riscv64" { + target_data_layout = target_data_layout.replace("-n32:64-", "-n64-"); + } + } // Ensure the data-layout values hardcoded remain the defaults. if sess.target.is_builtin { @@ -276,9 +275,9 @@ pub unsafe fn create_module<'ll>( } } - if let Some(BranchProtection { bti, pac_ret }) = sess.opts.debugging_opts.branch_protection { + if let Some(BranchProtection { bti, pac_ret }) = sess.opts.unstable_opts.branch_protection { if sess.target.arch != "aarch64" { - sess.err("-Zbranch-protection is only supported on aarch64"); + sess.emit_err(BranchProtectionRequiresAArch64); } else { llvm::LLVMRustAddModuleFlag( llmod, @@ -309,7 +308,7 @@ pub unsafe fn create_module<'ll>( } // Pass on the control-flow protection flags to LLVM (equivalent to `-fcf-protection` in Clang). - if let CFProtection::Branch | CFProtection::Full = sess.opts.debugging_opts.cf_protection { + if let CFProtection::Branch | CFProtection::Full = sess.opts.unstable_opts.cf_protection { llvm::LLVMRustAddModuleFlag( llmod, llvm::LLVMModFlagBehavior::Override, @@ -317,7 +316,7 @@ pub unsafe fn create_module<'ll>( 1, ) } - if let CFProtection::Return | CFProtection::Full = sess.opts.debugging_opts.cf_protection { + if let CFProtection::Return | CFProtection::Full = sess.opts.unstable_opts.cf_protection { llvm::LLVMRustAddModuleFlag( llmod, llvm::LLVMModFlagBehavior::Override, @@ -326,7 +325,7 @@ pub unsafe fn create_module<'ll>( ) } - if sess.opts.debugging_opts.virtual_function_elimination { + if sess.opts.unstable_opts.virtual_function_elimination { llvm::LLVMRustAddModuleFlag( llmod, llvm::LLVMModFlagBehavior::Error, @@ -465,7 +464,7 @@ impl<'ll, 'tcx> CodegenCx<'ll, 'tcx> { self.coverage_cx.as_ref() } - fn create_used_variable_impl(&self, name: &'static CStr, values: &[&'ll Value]) { + pub(crate) fn create_used_variable_impl(&self, name: &'static CStr, values: &[&'ll Value]) { let section = cstr!("llvm.metadata"); let array = self.const_array(self.type_ptr_to(self.type_i8()), values); @@ -563,14 +562,6 @@ impl<'ll, 'tcx> MiscMethods<'tcx> for CodegenCx<'ll, 'tcx> { self.codegen_unit } - fn used_statics(&self) -> &RefCell<Vec<&'ll Value>> { - &self.used_statics - } - - fn compiler_used_statics(&self) -> &RefCell<Vec<&'ll Value>> { - &self.compiler_used_statics - } - fn set_frame_pointer_type(&self, llfn: &'ll Value) { if let Some(attr) = attributes::frame_pointer_type_attr(self) { attributes::apply_to_llfn(llfn, llvm::AttributePlace::Function, &[attr]); @@ -584,20 +575,15 @@ impl<'ll, 'tcx> MiscMethods<'tcx> for CodegenCx<'ll, 'tcx> { attributes::apply_to_llfn(llfn, llvm::AttributePlace::Function, &attrs); } - fn create_used_variable(&self) { - self.create_used_variable_impl(cstr!("llvm.used"), &*self.used_statics.borrow()); - } - - fn create_compiler_used_variable(&self) { - self.create_used_variable_impl( - cstr!("llvm.compiler.used"), - &*self.compiler_used_statics.borrow(), - ); - } - fn declare_c_main(&self, fn_type: Self::Type) -> Option<Self::Function> { - if self.get_declared_value("main").is_none() { - Some(self.declare_cfn("main", llvm::UnnamedAddr::Global, fn_type)) + let entry_name = self.sess().target.entry_name.as_ref(); + if self.get_declared_value(entry_name).is_none() { + Some(self.declare_entry_fn( + entry_name, + self.sess().target.entry_abi.into(), + llvm::UnnamedAddr::Global, + fn_type, + )) } else { // If the symbol already exists, it is an error: for example, the user wrote // #[no_mangle] extern "C" fn main(..) {..} @@ -898,6 +884,9 @@ impl<'ll> CodegenCx<'ll, '_> { ifn!("llvm.dbg.declare", fn(t_metadata, t_metadata) -> void); ifn!("llvm.dbg.value", fn(t_metadata, t_i64, t_metadata) -> void); } + + ifn!("llvm.ptrmask", fn(i8p, t_isize) -> i8p); + None } @@ -906,7 +895,7 @@ impl<'ll> CodegenCx<'ll, '_> { return eh_catch_typeinfo; } let tcx = self.tcx; - assert!(self.sess().target.is_like_emscripten); + assert!(self.sess().target.os == "emscripten"); let eh_catch_typeinfo = match tcx.lang_items().eh_catch_typeinfo() { Some(def_id) => self.get_static(def_id), _ => { @@ -970,7 +959,7 @@ impl<'tcx> LayoutOfHelpers<'tcx> for CodegenCx<'_, 'tcx> { #[inline] fn handle_layout_err(&self, err: LayoutError<'tcx>, span: Span, ty: Ty<'tcx>) -> ! { if let LayoutError::SizeOverflow(_) = err { - self.sess().span_fatal(span, &err.to_string()) + self.sess().emit_fatal(Spanned { span, node: err }) } else { span_bug!(span, "failed to get layout for `{}`: {}", ty, err) } @@ -988,7 +977,7 @@ impl<'tcx> FnAbiOfHelpers<'tcx> for CodegenCx<'_, 'tcx> { fn_abi_request: FnAbiRequest<'tcx>, ) -> ! { if let FnAbiError::Layout(LayoutError::SizeOverflow(_)) = err { - self.sess().span_fatal(span, &err.to_string()) + self.sess().emit_fatal(Spanned { span, node: err }) } else { match fn_abi_request { FnAbiRequest::OfFnPtr { sig, extra_args } => { diff --git a/compiler/rustc_codegen_llvm/src/coverageinfo/mapgen.rs b/compiler/rustc_codegen_llvm/src/coverageinfo/mapgen.rs index 58f391692c4..86580d05d41 100644 --- a/compiler/rustc_codegen_llvm/src/coverageinfo/mapgen.rs +++ b/compiler/rustc_codegen_llvm/src/coverageinfo/mapgen.rs @@ -1,5 +1,6 @@ use crate::common::CodegenCx; use crate::coverageinfo; +use crate::errors::InstrumentCoverageRequiresLLVM12; use crate::llvm; use llvm::coverageinfo::CounterMappingRegion; @@ -16,8 +17,6 @@ use rustc_middle::ty::TyCtxt; use std::ffi::CString; -use tracing::debug; - /// Generates and exports the Coverage Map. /// /// Rust Coverage Map generation supports LLVM Coverage Mapping Format versions @@ -39,7 +38,7 @@ pub fn finalize<'ll, 'tcx>(cx: &CodegenCx<'ll, 'tcx>) { // LLVM 12. let version = coverageinfo::mapping_version(); if version < 4 { - tcx.sess.fatal("rustc option `-C instrument-coverage` requires LLVM 12 or higher."); + tcx.sess.emit_fatal(InstrumentCoverageRequiresLLVM12); } debug!("Generating coverage map for CodegenUnit: `{}`", cx.codegen_unit.name()); @@ -131,7 +130,7 @@ impl CoverageMapGenerator { // LLVM Coverage Mapping Format version 6 (zero-based encoded as 5) // requires setting the first filename to the compilation directory. // Since rustc generates coverage maps with relative paths, the - // compilation directory can be combined with the the relative paths + // compilation directory can be combined with the relative paths // to get absolute paths, if needed. let working_dir = tcx .sess @@ -175,7 +174,7 @@ impl CoverageMapGenerator { counter_regions.sort_unstable_by_key(|(_counter, region)| *region); for (counter, region) in counter_regions { let CodeRegion { file_name, start_line, start_col, end_line, end_col } = *region; - let same_file = current_file_name.as_ref().map_or(false, |p| *p == file_name); + let same_file = current_file_name.map_or(false, |p| p == file_name); if !same_file { if current_file_name.is_some() { current_file_id += 1; diff --git a/compiler/rustc_codegen_llvm/src/coverageinfo/mod.rs b/compiler/rustc_codegen_llvm/src/coverageinfo/mod.rs index 98ba38356a4..ace15cfb024 100644 --- a/compiler/rustc_codegen_llvm/src/coverageinfo/mod.rs +++ b/compiler/rustc_codegen_llvm/src/coverageinfo/mod.rs @@ -28,7 +28,6 @@ use std::cell::RefCell; use std::ffi::CString; use std::iter; -use tracing::debug; pub mod mapgen; @@ -38,7 +37,7 @@ const VAR_ALIGN_BYTES: usize = 8; /// A context object for maintaining all state needed by the coverageinfo module. pub struct CrateCoverageContext<'ll, 'tcx> { - // Coverage data for each instrumented function identified by DefId. + /// Coverage data for each instrumented function identified by DefId. pub(crate) function_coverage_map: RefCell<FxHashMap<Instance<'tcx>, FunctionCoverage<'tcx>>>, pub(crate) pgo_func_name_var_map: RefCell<FxHashMap<Instance<'tcx>, &'ll llvm::Value>>, } diff --git a/compiler/rustc_codegen_llvm/src/debuginfo/gdb.rs b/compiler/rustc_codegen_llvm/src/debuginfo/gdb.rs index 5186aee57fb..80fd9726fc7 100644 --- a/compiler/rustc_codegen_llvm/src/debuginfo/gdb.rs +++ b/compiler/rustc_codegen_llvm/src/debuginfo/gdb.rs @@ -55,7 +55,7 @@ pub fn get_or_insert_gdb_debug_scripts_section_global<'ll>(cx: &CodegenCx<'ll, ' // The initial byte `4` instructs GDB that the following pretty printer // is defined inline as opposed to in a standalone file. section_contents.extend_from_slice(b"\x04"); - let vis_name = format!("pretty-printer-{}-{}\n", crate_name.as_str(), index); + let vis_name = format!("pretty-printer-{}-{}\n", crate_name, index); section_contents.extend_from_slice(vis_name.as_bytes()); section_contents.extend_from_slice(&visualizer.src); diff --git a/compiler/rustc_codegen_llvm/src/debuginfo/metadata.rs b/compiler/rustc_codegen_llvm/src/debuginfo/metadata.rs index d5f39a45670..d87117dffdc 100644 --- a/compiler/rustc_codegen_llvm/src/debuginfo/metadata.rs +++ b/compiler/rustc_codegen_llvm/src/debuginfo/metadata.rs @@ -42,7 +42,6 @@ use rustc_span::{self, FileNameDisplayPreference, SourceFile}; use rustc_symbol_mangling::typeid_for_trait_ref; use rustc_target::abi::{Align, Size}; use smallvec::smallvec; -use tracing::debug; use libc::{c_char, c_longlong, c_uint}; use std::borrow::Cow; @@ -51,7 +50,6 @@ use std::hash::{Hash, Hasher}; use std::iter; use std::path::{Path, PathBuf}; use std::ptr; -use tracing::instrument; impl PartialEq for llvm::Metadata { fn eq(&self, other: &Self) -> bool { @@ -114,6 +112,7 @@ macro_rules! return_if_di_node_created_in_meantime { } /// Extract size and alignment from a TyAndLayout. +#[inline] fn size_and_align_of<'tcx>(ty_and_layout: TyAndLayout<'tcx>) -> (Size, Align) { (ty_and_layout.size, ty_and_layout.align.abi) } @@ -824,7 +823,7 @@ pub fn build_compile_unit_di_node<'ll, 'tcx>( output_filenames .split_dwarf_path( tcx.sess.split_debuginfo(), - tcx.sess.opts.debugging_opts.split_dwarf_kind, + tcx.sess.opts.unstable_opts.split_dwarf_kind, Some(codegen_unit_name), ) // We get a path relative to the working directory from split_dwarf_path @@ -881,15 +880,15 @@ pub fn build_compile_unit_di_node<'ll, 'tcx>( split_name.len(), kind, 0, - tcx.sess.opts.debugging_opts.split_dwarf_inlining, + tcx.sess.opts.unstable_opts.split_dwarf_inlining, ); - if tcx.sess.opts.debugging_opts.profile { + if tcx.sess.opts.unstable_opts.profile { let cu_desc_metadata = llvm::LLVMRustMetadataAsValue(debug_context.llcontext, unit_metadata); let default_gcda_path = &output_filenames.with_extension("gcda"); let gcda_path = - tcx.sess.opts.debugging_opts.profile_emit.as_ref().unwrap_or(default_gcda_path); + tcx.sess.opts.unstable_opts.profile_emit.as_ref().unwrap_or(default_gcda_path); let gcov_cu_info = [ path_to_mdstring(debug_context.llcontext, &output_filenames.with_extension("gcno")), @@ -999,7 +998,7 @@ fn build_struct_type_di_node<'ll, 'tcx>( .iter() .enumerate() .map(|(i, f)| { - let field_name = if variant_def.ctor_kind == CtorKind::Fn { + let field_name = if variant_def.ctor_kind() == Some(CtorKind::Fn) { // This is a tuple struct tuple_field_name(i) } else { @@ -1420,7 +1419,7 @@ fn build_vtable_type_di_node<'ll, 'tcx>( cx, type_map::stub( cx, - Stub::VtableTy { vtable_holder }, + Stub::VTableTy { vtable_holder }, unique_type_id, &vtable_type_name, (size, pointer_align), @@ -1499,24 +1498,18 @@ fn vcall_visibility_metadata<'ll, 'tcx>( // If there is not LTO and the visibility in public, we have to assume that the vtable can // be seen from anywhere. With multiple CGUs, the vtable is quasi-public. (Lto::No | Lto::ThinLocal, Visibility::Public, _) - | (Lto::No, Visibility::Restricted(_) | Visibility::Invisible, false) => { - VCallVisibility::Public - } + | (Lto::No, Visibility::Restricted(_), false) => VCallVisibility::Public, // With LTO and a quasi-public visibility, the usages of the functions of the vtable are // all known by the `LinkageUnit`. // FIXME: LLVM only supports this optimization for `Lto::Fat` currently. Once it also // supports `Lto::Thin` the `VCallVisibility` may have to be adjusted for those. (Lto::Fat | Lto::Thin, Visibility::Public, _) - | ( - Lto::ThinLocal | Lto::Thin | Lto::Fat, - Visibility::Restricted(_) | Visibility::Invisible, - false, - ) => VCallVisibility::LinkageUnit, + | (Lto::ThinLocal | Lto::Thin | Lto::Fat, Visibility::Restricted(_), false) => { + VCallVisibility::LinkageUnit + } // If there is only one CGU, private vtables can only be seen by that CGU/translation unit // and therefore we know of all usages of functions in the vtable. - (_, Visibility::Restricted(_) | Visibility::Invisible, true) => { - VCallVisibility::TranslationUnit - } + (_, Visibility::Restricted(_), true) => VCallVisibility::TranslationUnit, }; let trait_ref_typeid = typeid_for_trait_ref(cx.tcx, trait_ref); @@ -1559,7 +1552,7 @@ pub fn create_vtable_di_node<'ll, 'tcx>( ) { // FIXME(flip1995): The virtual function elimination optimization only works with full LTO in // LLVM at the moment. - if cx.sess().opts.debugging_opts.virtual_function_elimination && cx.sess().lto() == Lto::Fat { + if cx.sess().opts.unstable_opts.virtual_function_elimination && cx.sess().lto() == Lto::Fat { vcall_visibility_metadata(cx, ty, poly_trait_ref, vtable); } diff --git a/compiler/rustc_codegen_llvm/src/debuginfo/metadata/enums/cpp_like.rs b/compiler/rustc_codegen_llvm/src/debuginfo/metadata/enums/cpp_like.rs index d6e2c8ccdf4..53e8a291d1e 100644 --- a/compiler/rustc_codegen_llvm/src/debuginfo/metadata/enums/cpp_like.rs +++ b/compiler/rustc_codegen_llvm/src/debuginfo/metadata/enums/cpp_like.rs @@ -1,19 +1,21 @@ use std::borrow::Cow; use libc::c_uint; -use rustc_codegen_ssa::debuginfo::{ - type_names::compute_debuginfo_type_name, wants_c_like_enum_debuginfo, +use rustc_codegen_ssa::{ + debuginfo::{type_names::compute_debuginfo_type_name, wants_c_like_enum_debuginfo}, + traits::ConstMethods, }; + +use rustc_index::vec::IndexVec; use rustc_middle::{ bug, ty::{ self, layout::{LayoutOf, TyAndLayout}, - util::Discr, - AdtDef, GeneratorSubsts, + AdtDef, GeneratorSubsts, Ty, }, }; -use rustc_target::abi::{Size, TagEncoding, VariantIdx, Variants}; +use rustc_target::abi::{Align, Endian, Size, TagEncoding, VariantIdx, Variants}; use smallvec::smallvec; use crate::{ @@ -21,9 +23,9 @@ use crate::{ debuginfo::{ metadata::{ build_field_di_node, closure_saved_names_of_captured_variables, - enums::tag_base_type, - file_metadata, generator_layout_and_saved_local_names, size_and_align_of, - type_map::{self, UniqueTypeId}, + enums::{tag_base_type, DiscrResult}, + file_metadata, generator_layout_and_saved_local_names, size_and_align_of, type_di_node, + type_map::{self, Stub, UniqueTypeId}, unknown_file_metadata, DINodeCreationResult, SmallVec, NO_GENERICS, NO_SCOPE_METADATA, UNKNOWN_LINE_NUMBER, }, @@ -35,59 +37,161 @@ use crate::{ }, }; -/// In CPP-like mode, we generate a union of structs for each variant and an -/// explicit discriminant field roughly equivalent to the following C/C++ code: +// The names of the associated constants in each variant wrapper struct. +// These have to match up with the names being used in `intrinsic.natvis`. +const ASSOC_CONST_DISCR_NAME: &str = "NAME"; +const ASSOC_CONST_DISCR_EXACT: &str = "DISCR_EXACT"; +const ASSOC_CONST_DISCR_BEGIN: &str = "DISCR_BEGIN"; +const ASSOC_CONST_DISCR_END: &str = "DISCR_END"; + +const ASSOC_CONST_DISCR128_EXACT_LO: &str = "DISCR128_EXACT_LO"; +const ASSOC_CONST_DISCR128_EXACT_HI: &str = "DISCR128_EXACT_HI"; +const ASSOC_CONST_DISCR128_BEGIN_LO: &str = "DISCR128_BEGIN_LO"; +const ASSOC_CONST_DISCR128_BEGIN_HI: &str = "DISCR128_BEGIN_HI"; +const ASSOC_CONST_DISCR128_END_LO: &str = "DISCR128_END_LO"; +const ASSOC_CONST_DISCR128_END_HI: &str = "DISCR128_END_HI"; + +// The name of the tag field in the top-level union +const TAG_FIELD_NAME: &str = "tag"; +const TAG_FIELD_NAME_128_LO: &str = "tag128_lo"; +const TAG_FIELD_NAME_128_HI: &str = "tag128_hi"; + +// We assign a "virtual" discriminant value to the sole variant of +// a single-variant enum. +const SINGLE_VARIANT_VIRTUAL_DISR: u64 = 0; + +/// In CPP-like mode, we generate a union with a field for each variant and an +/// explicit tag field. The field of each variant has a struct type +/// that encodes the discrimiant of the variant and it's data layout. +/// The union also has a nested enumeration type that is only used for encoding +/// variant names in an efficient way. Its enumerator values do _not_ correspond +/// to the enum's discriminant values. +/// It's roughly equivalent to the following C/C++ code: /// /// ```c -/// union enum$<{fully-qualified-name}> { -/// struct {variant 0 name} { -/// <variant 0 fields> +/// union enum2$<{fully-qualified-name}> { +/// struct Variant0 { +/// struct {name-of-variant-0} { +/// <variant 0 fields> +/// } value; +/// +/// static VariantNames NAME = {name-of-variant-0}; +/// static int_type DISCR_EXACT = {discriminant-of-variant-0}; /// } variant0; +/// /// <other variant structs> -/// {name} discriminant; +/// +/// int_type tag; +/// +/// enum VariantNames { +/// <name-of-variant-0> = 0, // The numeric values are variant index, +/// <name-of-variant-1> = 1, // not discriminant values. +/// <name-of-variant-2> = 2, +/// ... +/// } /// } /// ``` /// -/// As you can see, the type name is wrapped `enum$`. This way we can have a -/// single NatVis rule for handling all enums. +/// As you can see, the type name is wrapped in `enum2$<_>`. This way we can +/// have a single NatVis rule for handling all enums. The `2` in `enum2$<_>` +/// is an encoding version tag, so that debuggers can decide to decode this +/// differently than the previous `enum$<_>` encoding emitted by earlier +/// compiler versions. /// -/// At the LLVM IR level this looks like +/// Niche-tag enums have one special variant, usually called the +/// "untagged variant". This variant has a field that +/// doubles as the tag of the enum. The variant is active when the value of +/// that field is within a pre-defined range. Therefore the variant struct +/// has a `DISCR_BEGIN` and `DISCR_END` field instead of `DISCR_EXACT` in +/// that case. Both `DISCR_BEGIN` and `DISCR_END` are inclusive bounds. +/// Note that these ranges can wrap around, so that `DISCR_END < DISCR_BEGIN`. /// -/// ```txt -/// DW_TAG_union_type (top-level type for enum) -/// DW_TAG_member (member for variant 1) -/// DW_TAG_member (member for variant 2) -/// DW_TAG_member (member for variant 3) -/// DW_TAG_structure_type (type of variant 1) -/// DW_TAG_structure_type (type of variant 2) -/// DW_TAG_structure_type (type of variant 3) -/// DW_TAG_enumeration_type (type of tag) -/// ``` +/// Single-variant enums don't actually have a tag field. In this case we +/// emit a static tag field (that always has the value 0) so we can use the +/// same representation (and NatVis). /// -/// The above encoding applies for enums with a direct tag. For niche-tag we have to do things -/// differently in order to allow a NatVis visualizer to extract all the information needed: -/// We generate a union of two fields, one for the dataful variant -/// and one that just points to the discriminant (which is some field within the dataful variant). -/// We also create a DW_TAG_enumeration_type DIE that contains tag values for the non-dataful -/// variants and make the discriminant field that type. We then use NatVis to render the enum type -/// correctly in Windbg/VS. This will generate debuginfo roughly equivalent to the following C: +/// For niche-layout enums it's possible to have a 128-bit tag. NatVis, VS, and +/// WinDbg (the main targets for CPP-like debuginfo at the moment) don't support +/// 128-bit integers, so all values involved get split into two 64-bit fields. +/// Instead of the `tag` field, we generate two fields `tag128_lo` and `tag128_hi`, +/// Instead of `DISCR_EXACT`, we generate `DISCR128_EXACT_LO` and `DISCR128_EXACT_HI`, +/// and so on. /// -/// ```c -/// union enum$<{name}, {min niche}, {max niche}, {dataful variant name}> { -/// struct <dataful variant name> { -/// <fields in dataful variant> -/// } dataful_variant; -/// enum Discriminant$ { -/// <non-dataful variants> -/// } discriminant; +/// +/// The following pseudocode shows how to decode an enum value in a debugger: +/// +/// ```text +/// +/// fn find_active_variant(enum_value) -> (VariantName, VariantValue) { +/// let is_128_bit = enum_value.has_field("tag128_lo"); +/// +/// if !is_128_bit { +/// // Note: `tag` can be a static field for enums with only one +/// // inhabited variant. +/// let tag = enum_value.field("tag").value; +/// +/// // For each variant, check if it is a match. Only one of them will match, +/// // so if we find it we can return it immediately. +/// for variant_field in enum_value.fields().filter(|f| f.name.starts_with("variant")) { +/// if variant_field.has_field("DISCR_EXACT") { +/// // This variant corresponds to a single tag value +/// if variant_field.field("DISCR_EXACT").value == tag { +/// return (variant_field.field("NAME"), variant_field.value); +/// } +/// } else { +/// // This is a range variant +/// let begin = variant_field.field("DISCR_BEGIN"); +/// let end = variant_field.field("DISCR_END"); +/// +/// if is_in_range(tag, begin, end) { +/// return (variant_field.field("NAME"), variant_field.value); +/// } +/// } +/// } +/// } else { +/// // Basically the same as with smaller tags, we just have to +/// // stitch the values together. +/// let tag: u128 = (enum_value.field("tag128_lo").value as u128) | +/// (enum_value.field("tag128_hi").value as u128 << 64); +/// +/// for variant_field in enum_value.fields().filter(|f| f.name.starts_with("variant")) { +/// if variant_field.has_field("DISCR128_EXACT_LO") { +/// let discr_exact = (variant_field.field("DISCR128_EXACT_LO" as u128) | +/// (variant_field.field("DISCR128_EXACT_HI") as u128 << 64); +/// +/// // This variant corresponds to a single tag value +/// if discr_exact.value == tag { +/// return (variant_field.field("NAME"), variant_field.value); +/// } +/// } else { +/// // This is a range variant +/// let begin = (variant_field.field("DISCR128_BEGIN_LO").value as u128) | +/// (variant_field.field("DISCR128_BEGIN_HI").value as u128 << 64); +/// let end = (variant_field.field("DISCR128_END_LO").value as u128) | +/// (variant_field.field("DISCR128_END_HI").value as u128 << 64); +/// +/// if is_in_range(tag, begin, end) { +/// return (variant_field.field("NAME"), variant_field.value); +/// } +/// } +/// } +/// } +/// +/// // We should have found an active variant at this point. +/// unreachable!(); /// } -/// ``` /// -/// The NatVis in `intrinsic.natvis` matches on the type name `enum$<*, *, *, *>` -/// and evaluates `this.discriminant`. If the value is between the min niche and max -/// niche, then the enum is in the dataful variant and `this.dataful_variant` is -/// rendered. Otherwise, the enum is in one of the non-dataful variants. In that -/// case, we just need to render the name of the `this.discriminant` enum. +/// // Check if a value is within the given range +/// // (where the range might wrap around the value space) +/// fn is_in_range(value, start, end) -> bool { +/// if start < end { +/// value >= start && value <= end +/// } else { +/// value >= start || value <= end +/// } +/// } +/// +/// ``` pub(super) fn build_enum_type_di_node<'ll, 'tcx>( cx: &CodegenCx<'ll, 'tcx>, unique_type_id: UniqueTypeId<'tcx>, @@ -135,27 +239,28 @@ pub(super) fn build_enum_type_di_node<'ll, 'tcx>( ref variants, tag_field, .. - } => build_union_fields_for_direct_tag_enum( + } => build_union_fields_for_enum( cx, enum_adt_def, enum_type_and_layout, enum_type_di_node, - &mut variants.indices(), + variants.indices(), tag_field, + None, ), Variants::Multiple { - tag_encoding: TagEncoding::Niche { dataful_variant, .. }, + tag_encoding: TagEncoding::Niche { untagged_variant, .. }, ref variants, tag_field, .. - } => build_union_fields_for_niche_tag_enum( + } => build_union_fields_for_enum( cx, enum_adt_def, enum_type_and_layout, enum_type_di_node, - dataful_variant, - &mut variants.indices(), + variants.indices(), tag_field, + Some(untagged_variant), ), } }, @@ -217,137 +322,344 @@ fn build_single_variant_union_fields<'ll, 'tcx>( let variant_layout = enum_type_and_layout.for_variant(cx, variant_index); let variant_struct_type_di_node = super::build_enum_variant_struct_type_di_node( cx, - enum_type_and_layout.ty, + enum_type_and_layout, enum_type_di_node, variant_index, enum_adt_def.variant(variant_index), variant_layout, ); - // NOTE: The field name of the union is the same as the variant name, not "variant0". - let variant_name = enum_adt_def.variant(variant_index).name.as_str(); + let tag_base_type = cx.tcx.types.u32; + let tag_base_type_di_node = type_di_node(cx, tag_base_type); + let tag_base_type_align = cx.align_of(tag_base_type); + + let variant_names_type_di_node = build_variant_names_type_di_node( + cx, + enum_type_di_node, + std::iter::once(( + variant_index, + Cow::from(enum_adt_def.variant(variant_index).name.as_str()), + )), + ); - smallvec![build_field_di_node( + let variant_struct_type_wrapper_di_node = build_variant_struct_wrapper_type_di_node( cx, + enum_type_and_layout, enum_type_di_node, - variant_name, - // NOTE: We use the size and align of the entire type, not from variant_layout - // since the later is sometimes smaller (if it has fewer fields). - size_and_align_of(enum_type_and_layout), - Size::ZERO, - DIFlags::FlagZero, + variant_index, + None, variant_struct_type_di_node, - )] + variant_names_type_di_node, + tag_base_type_di_node, + tag_base_type, + DiscrResult::NoDiscriminant, + ); + + smallvec![ + build_field_di_node( + cx, + enum_type_di_node, + &variant_union_field_name(variant_index), + // NOTE: We use the size and align of the entire type, not from variant_layout + // since the later is sometimes smaller (if it has fewer fields). + size_and_align_of(enum_type_and_layout), + Size::ZERO, + DIFlags::FlagZero, + variant_struct_type_wrapper_di_node, + ), + unsafe { + llvm::LLVMRustDIBuilderCreateStaticMemberType( + DIB(cx), + enum_type_di_node, + TAG_FIELD_NAME.as_ptr().cast(), + TAG_FIELD_NAME.len(), + unknown_file_metadata(cx), + UNKNOWN_LINE_NUMBER, + variant_names_type_di_node, + DIFlags::FlagZero, + Some(cx.const_u64(SINGLE_VARIANT_VIRTUAL_DISR)), + tag_base_type_align.bits() as u32, + ) + } + ] } -fn build_union_fields_for_direct_tag_enum<'ll, 'tcx>( +fn build_union_fields_for_enum<'ll, 'tcx>( cx: &CodegenCx<'ll, 'tcx>, enum_adt_def: AdtDef<'tcx>, enum_type_and_layout: TyAndLayout<'tcx>, enum_type_di_node: &'ll DIType, - variant_indices: &mut dyn Iterator<Item = VariantIdx>, + variant_indices: impl Iterator<Item = VariantIdx> + Clone, tag_field: usize, + untagged_variant_index: Option<VariantIdx>, ) -> SmallVec<&'ll DIType> { + let tag_base_type = super::tag_base_type(cx, enum_type_and_layout); + + let variant_names_type_di_node = build_variant_names_type_di_node( + cx, + enum_type_di_node, + variant_indices.clone().map(|variant_index| { + let variant_name = Cow::from(enum_adt_def.variant(variant_index).name.as_str()); + (variant_index, variant_name) + }), + ); + let variant_field_infos: SmallVec<VariantFieldInfo<'ll>> = variant_indices .map(|variant_index| { let variant_layout = enum_type_and_layout.for_variant(cx, variant_index); + let variant_def = enum_adt_def.variant(variant_index); + + let variant_struct_type_di_node = super::build_enum_variant_struct_type_di_node( + cx, + enum_type_and_layout, + enum_type_di_node, + variant_index, + variant_def, + variant_layout, + ); + VariantFieldInfo { variant_index, - variant_struct_type_di_node: super::build_enum_variant_struct_type_di_node( - cx, - enum_type_and_layout.ty, - enum_type_di_node, - variant_index, - enum_adt_def.variant(variant_index), - variant_layout, - ), + variant_struct_type_di_node, source_info: None, + discr: super::compute_discriminant_value(cx, enum_type_and_layout, variant_index), } }) .collect(); - let discr_type_name = cx.tcx.item_name(enum_adt_def.did()); - let tag_base_type = super::tag_base_type(cx, enum_type_and_layout); - let discr_type_di_node = super::build_enumeration_type_di_node( - cx, - discr_type_name.as_str(), - tag_base_type, - &mut enum_adt_def.discriminants(cx.tcx).map(|(variant_index, discr)| { - (discr, Cow::from(enum_adt_def.variant(variant_index).name.as_str())) - }), - enum_type_di_node, - ); - build_union_fields_for_direct_tag_enum_or_generator( cx, enum_type_and_layout, enum_type_di_node, &variant_field_infos, - discr_type_di_node, + variant_names_type_di_node, + tag_base_type, tag_field, + untagged_variant_index, ) } -fn build_union_fields_for_niche_tag_enum<'ll, 'tcx>( +// The base type of the VariantNames DW_AT_enumeration_type is always the same. +// It has nothing to do with the tag of the enum and just has to be big enough +// to hold all variant names. +fn variant_names_enum_base_type<'ll, 'tcx>(cx: &CodegenCx<'ll, 'tcx>) -> Ty<'tcx> { + cx.tcx.types.u32 +} + +/// This function builds a DW_AT_enumeration_type that contains an entry for +/// each variant. Note that this has nothing to do with the discriminant. The +/// numeric value of each enumerator corresponds to the variant index. The +/// type is only used for efficiently encoding the name of each variant in +/// debuginfo. +fn build_variant_names_type_di_node<'ll, 'tcx>( cx: &CodegenCx<'ll, 'tcx>, - enum_adt_def: AdtDef<'tcx>, - enum_type_and_layout: TyAndLayout<'tcx>, - enum_type_di_node: &'ll DIType, - dataful_variant_index: VariantIdx, - variant_indices: &mut dyn Iterator<Item = VariantIdx>, - tag_field: usize, -) -> SmallVec<&'ll DIType> { - let dataful_variant_struct_type_di_node = super::build_enum_variant_struct_type_di_node( + containing_scope: &'ll DIType, + variants: impl Iterator<Item = (VariantIdx, Cow<'tcx, str>)>, +) -> &'ll DIType { + // Create an enumerator for each variant. + super::build_enumeration_type_di_node( cx, - enum_type_and_layout.ty, - enum_type_di_node, - dataful_variant_index, - &enum_adt_def.variant(dataful_variant_index), - enum_type_and_layout.for_variant(cx, dataful_variant_index), - ); + "VariantNames", + variant_names_enum_base_type(cx), + variants.map(|(variant_index, variant_name)| (variant_name, variant_index.as_u32().into())), + containing_scope, + ) +} - let tag_base_type = super::tag_base_type(cx, enum_type_and_layout); - // Create an DW_TAG_enumerator for each variant except the dataful one. - let discr_type_di_node = super::build_enumeration_type_di_node( +fn build_variant_struct_wrapper_type_di_node<'ll, 'tcx>( + cx: &CodegenCx<'ll, 'tcx>, + enum_or_generator_type_and_layout: TyAndLayout<'tcx>, + enum_or_generator_type_di_node: &'ll DIType, + variant_index: VariantIdx, + untagged_variant_index: Option<VariantIdx>, + variant_struct_type_di_node: &'ll DIType, + variant_names_type_di_node: &'ll DIType, + tag_base_type_di_node: &'ll DIType, + tag_base_type: Ty<'tcx>, + discr: DiscrResult, +) -> &'ll DIType { + type_map::build_type_with_children( cx, - "Discriminant$", - tag_base_type, - &mut variant_indices.filter_map(|variant_index| { - if let Some(discr_val) = - super::compute_discriminant_value(cx, enum_type_and_layout, variant_index) - { - let discr = Discr { val: discr_val as u128, ty: tag_base_type }; - let variant_name = Cow::from(enum_adt_def.variant(variant_index).name.as_str()); - Some((discr, variant_name)) - } else { - debug_assert_eq!(variant_index, dataful_variant_index); - None - } - }), - enum_type_di_node, - ); - - smallvec![ - build_field_di_node( - cx, - enum_type_di_node, - "dataful_variant", - size_and_align_of(enum_type_and_layout), - Size::ZERO, - DIFlags::FlagZero, - dataful_variant_struct_type_di_node, - ), - build_field_di_node( + type_map::stub( cx, - enum_type_di_node, - "discriminant", - cx.size_and_align_of(tag_base_type), - enum_type_and_layout.fields.offset(tag_field), + Stub::Struct, + UniqueTypeId::for_enum_variant_struct_type_wrapper( + cx.tcx, + enum_or_generator_type_and_layout.ty, + variant_index, + ), + &variant_struct_wrapper_type_name(variant_index), + // NOTE: We use size and align of enum_type, not from variant_layout: + size_and_align_of(enum_or_generator_type_and_layout), + Some(enum_or_generator_type_di_node), DIFlags::FlagZero, - discr_type_di_node, ), - ] + |cx, wrapper_struct_type_di_node| { + enum DiscrKind { + Exact(u64), + Exact128(u128), + Range(u64, u64), + Range128(u128, u128), + } + + let (tag_base_type_size, tag_base_type_align) = cx.size_and_align_of(tag_base_type); + let is_128_bits = tag_base_type_size.bits() > 64; + + let discr = match discr { + DiscrResult::NoDiscriminant => DiscrKind::Exact(SINGLE_VARIANT_VIRTUAL_DISR), + DiscrResult::Value(discr_val) => { + if is_128_bits { + DiscrKind::Exact128(discr_val) + } else { + debug_assert_eq!(discr_val, discr_val as u64 as u128); + DiscrKind::Exact(discr_val as u64) + } + } + DiscrResult::Range(min, max) => { + assert_eq!(Some(variant_index), untagged_variant_index); + if is_128_bits { + DiscrKind::Range128(min, max) + } else { + debug_assert_eq!(min, min as u64 as u128); + debug_assert_eq!(max, max as u64 as u128); + DiscrKind::Range(min as u64, max as u64) + } + } + }; + + let mut fields = SmallVec::new(); + + // We always have a field for the value + fields.push(build_field_di_node( + cx, + wrapper_struct_type_di_node, + "value", + size_and_align_of(enum_or_generator_type_and_layout), + Size::ZERO, + DIFlags::FlagZero, + variant_struct_type_di_node, + )); + + let build_assoc_const = + |name: &str, type_di_node: &'ll DIType, value: u64, align: Align| unsafe { + llvm::LLVMRustDIBuilderCreateStaticMemberType( + DIB(cx), + wrapper_struct_type_di_node, + name.as_ptr().cast(), + name.len(), + unknown_file_metadata(cx), + UNKNOWN_LINE_NUMBER, + type_di_node, + DIFlags::FlagZero, + Some(cx.const_u64(value)), + align.bits() as u32, + ) + }; + + // We also always have an associated constant for the discriminant value + // of the variant. + fields.push(build_assoc_const( + ASSOC_CONST_DISCR_NAME, + variant_names_type_di_node, + variant_index.as_u32() as u64, + cx.align_of(variant_names_enum_base_type(cx)), + )); + + // Emit the discriminant value (or range) corresponding to the variant. + match discr { + DiscrKind::Exact(discr_val) => { + fields.push(build_assoc_const( + ASSOC_CONST_DISCR_EXACT, + tag_base_type_di_node, + discr_val, + tag_base_type_align, + )); + } + DiscrKind::Exact128(discr_val) => { + let align = cx.align_of(cx.tcx.types.u64); + let type_di_node = type_di_node(cx, cx.tcx.types.u64); + let Split128 { hi, lo } = split_128(discr_val); + + fields.push(build_assoc_const( + ASSOC_CONST_DISCR128_EXACT_LO, + type_di_node, + lo, + align, + )); + + fields.push(build_assoc_const( + ASSOC_CONST_DISCR128_EXACT_HI, + type_di_node, + hi, + align, + )); + } + DiscrKind::Range(begin, end) => { + fields.push(build_assoc_const( + ASSOC_CONST_DISCR_BEGIN, + tag_base_type_di_node, + begin, + tag_base_type_align, + )); + + fields.push(build_assoc_const( + ASSOC_CONST_DISCR_END, + tag_base_type_di_node, + end, + tag_base_type_align, + )); + } + DiscrKind::Range128(begin, end) => { + let align = cx.align_of(cx.tcx.types.u64); + let type_di_node = type_di_node(cx, cx.tcx.types.u64); + let Split128 { hi: begin_hi, lo: begin_lo } = split_128(begin); + let Split128 { hi: end_hi, lo: end_lo } = split_128(end); + + fields.push(build_assoc_const( + ASSOC_CONST_DISCR128_BEGIN_HI, + type_di_node, + begin_hi, + align, + )); + + fields.push(build_assoc_const( + ASSOC_CONST_DISCR128_BEGIN_LO, + type_di_node, + begin_lo, + align, + )); + + fields.push(build_assoc_const( + ASSOC_CONST_DISCR128_END_HI, + type_di_node, + end_hi, + align, + )); + + fields.push(build_assoc_const( + ASSOC_CONST_DISCR128_END_LO, + type_di_node, + end_lo, + align, + )); + } + } + + fields + }, + NO_GENERICS, + ) + .di_node +} + +struct Split128 { + hi: u64, + lo: u64, +} + +fn split_128(value: u128) -> Split128 { + Split128 { hi: (value >> 64) as u64, lo: value as u64 } } fn build_union_fields_for_direct_tag_generator<'ll, 'tcx>( @@ -369,6 +681,29 @@ fn build_union_fields_for_direct_tag_generator<'ll, 'tcx>( let common_upvar_names = closure_saved_names_of_captured_variables(cx.tcx, generator_def_id); let variant_range = generator_substs.variant_range(generator_def_id, cx.tcx); + let variant_count = (variant_range.start.as_u32()..variant_range.end.as_u32()).len(); + + let tag_base_type = tag_base_type(cx, generator_type_and_layout); + + let variant_names_type_di_node = build_variant_names_type_di_node( + cx, + generator_type_di_node, + variant_range + .clone() + .map(|variant_index| (variant_index, GeneratorSubsts::variant_name(variant_index))), + ); + + let discriminants: IndexVec<VariantIdx, DiscrResult> = { + let discriminants_iter = generator_substs.discriminants(generator_def_id, cx.tcx); + let mut discriminants: IndexVec<VariantIdx, DiscrResult> = + IndexVec::with_capacity(variant_count); + for (variant_index, discr) in discriminants_iter { + // Assert that the index in the IndexMap matches up with the given VariantIdx. + assert_eq!(variant_index, discriminants.next_index()); + discriminants.push(DiscrResult::Value(discr.val)); + } + discriminants + }; // Build the type node for each field. let variant_field_infos: SmallVec<VariantFieldInfo<'ll>> = variant_range @@ -391,29 +726,24 @@ fn build_union_fields_for_direct_tag_generator<'ll, 'tcx>( None }; - VariantFieldInfo { variant_index, variant_struct_type_di_node, source_info } + VariantFieldInfo { + variant_index, + variant_struct_type_di_node, + source_info, + discr: discriminants[variant_index], + } }) .collect(); - let tag_base_type = tag_base_type(cx, generator_type_and_layout); - let discr_type_name = "Discriminant$"; - let discr_type_di_node = super::build_enumeration_type_di_node( - cx, - discr_type_name, - tag_base_type, - &mut generator_substs - .discriminants(generator_def_id, cx.tcx) - .map(|(variant_index, discr)| (discr, GeneratorSubsts::variant_name(variant_index))), - generator_type_di_node, - ); - build_union_fields_for_direct_tag_enum_or_generator( cx, generator_type_and_layout, generator_type_di_node, &variant_field_infos[..], - discr_type_di_node, + variant_names_type_di_node, + tag_base_type, tag_field, + None, ) } @@ -425,8 +755,11 @@ fn build_union_fields_for_direct_tag_enum_or_generator<'ll, 'tcx>( enum_type_di_node: &'ll DIType, variant_field_infos: &[VariantFieldInfo<'ll>], discr_type_di_node: &'ll DIType, + tag_base_type: Ty<'tcx>, tag_field: usize, + untagged_variant_index: Option<VariantIdx>, ) -> SmallVec<&'ll DIType> { + let tag_base_type_di_node = type_di_node(cx, tag_base_type); let mut unions_fields = SmallVec::with_capacity(variant_field_infos.len() + 1); // We create a field in the union for each variant ... @@ -438,6 +771,19 @@ fn build_union_fields_for_direct_tag_enum_or_generator<'ll, 'tcx>( let field_name = variant_union_field_name(variant_member_info.variant_index); let (size, align) = size_and_align_of(enum_type_and_layout); + let variant_struct_type_wrapper = build_variant_struct_wrapper_type_di_node( + cx, + enum_type_and_layout, + enum_type_di_node, + variant_member_info.variant_index, + untagged_variant_index, + variant_member_info.variant_struct_type_di_node, + discr_type_di_node, + tag_base_type_di_node, + tag_base_type, + variant_member_info.discr, + ); + // We use LLVMRustDIBuilderCreateMemberType() member type directly because // the build_field_di_node() function does not support specifying a source location, // which is something that we don't do anywhere else. @@ -456,7 +802,7 @@ fn build_union_fields_for_direct_tag_enum_or_generator<'ll, 'tcx>( // Union fields are always at offset zero Size::ZERO.bits(), DIFlags::FlagZero, - variant_member_info.variant_struct_type_di_node, + variant_struct_type_wrapper, ) } })); @@ -466,16 +812,53 @@ fn build_union_fields_for_direct_tag_enum_or_generator<'ll, 'tcx>( cx.size_and_align_of(super::tag_base_type(cx, enum_type_and_layout)) ); - // ... and a field for the discriminant. - unions_fields.push(build_field_di_node( - cx, - enum_type_di_node, - "discriminant", - cx.size_and_align_of(enum_type_and_layout.field(cx, tag_field).ty), - enum_type_and_layout.fields.offset(tag_field), - DIFlags::FlagZero, - discr_type_di_node, - )); + // ... and a field for the tag. If the tag is 128 bits wide, this will actually + // be two 64-bit fields. + let is_128_bits = cx.size_of(tag_base_type).bits() > 64; + + if is_128_bits { + let type_di_node = type_di_node(cx, cx.tcx.types.u64); + let size_and_align = cx.size_and_align_of(cx.tcx.types.u64); + + let (lo_offset, hi_offset) = match cx.tcx.data_layout.endian { + Endian::Little => (0, 8), + Endian::Big => (8, 0), + }; + + let tag_field_offset = enum_type_and_layout.fields.offset(tag_field).bytes(); + let lo_offset = Size::from_bytes(tag_field_offset + lo_offset); + let hi_offset = Size::from_bytes(tag_field_offset + hi_offset); + + unions_fields.push(build_field_di_node( + cx, + enum_type_di_node, + TAG_FIELD_NAME_128_LO, + size_and_align, + lo_offset, + DIFlags::FlagZero, + type_di_node, + )); + + unions_fields.push(build_field_di_node( + cx, + enum_type_di_node, + TAG_FIELD_NAME_128_HI, + size_and_align, + hi_offset, + DIFlags::FlagZero, + type_di_node, + )); + } else { + unions_fields.push(build_field_di_node( + cx, + enum_type_di_node, + TAG_FIELD_NAME, + cx.size_and_align_of(enum_type_and_layout.field(cx, tag_field).ty), + enum_type_and_layout.fields.offset(tag_field), + DIFlags::FlagZero, + tag_base_type_di_node, + )); + } unions_fields } @@ -485,6 +868,7 @@ struct VariantFieldInfo<'ll> { variant_index: VariantIdx, variant_struct_type_di_node: &'ll DIType, source_info: Option<(&'ll DIFile, c_uint)>, + discr: DiscrResult, } fn variant_union_field_name(variant_index: VariantIdx) -> Cow<'static, str> { @@ -512,3 +896,29 @@ fn variant_union_field_name(variant_index: VariantIdx) -> Cow<'static, str> { .map(|&s| Cow::from(s)) .unwrap_or_else(|| format!("variant{}", variant_index.as_usize()).into()) } + +fn variant_struct_wrapper_type_name(variant_index: VariantIdx) -> Cow<'static, str> { + const PRE_ALLOCATED: [&str; 16] = [ + "Variant0", + "Variant1", + "Variant2", + "Variant3", + "Variant4", + "Variant5", + "Variant6", + "Variant7", + "Variant8", + "Variant9", + "Variant10", + "Variant11", + "Variant12", + "Variant13", + "Variant14", + "Variant15", + ]; + + PRE_ALLOCATED + .get(variant_index.as_usize()) + .map(|&s| Cow::from(s)) + .unwrap_or_else(|| format!("Variant{}", variant_index.as_usize()).into()) +} diff --git a/compiler/rustc_codegen_llvm/src/debuginfo/metadata/enums/mod.rs b/compiler/rustc_codegen_llvm/src/debuginfo/metadata/enums/mod.rs index 73e01d0453b..564ab351bd4 100644 --- a/compiler/rustc_codegen_llvm/src/debuginfo/metadata/enums/mod.rs +++ b/compiler/rustc_codegen_llvm/src/debuginfo/metadata/enums/mod.rs @@ -10,7 +10,6 @@ use rustc_middle::{ ty::{ self, layout::{IntegerExt, LayoutOf, PrimitiveExt, TyAndLayout}, - util::Discr, AdtDef, GeneratorSubsts, Ty, VariantDef, }, }; @@ -90,8 +89,9 @@ fn build_c_style_enum_di_node<'ll, 'tcx>( cx, &compute_debuginfo_type_name(cx.tcx, enum_type_and_layout.ty, false), tag_base_type(cx, enum_type_and_layout), - &mut enum_adt_def.discriminants(cx.tcx).map(|(variant_index, discr)| { - (discr, Cow::from(enum_adt_def.variant(variant_index).name.as_str())) + enum_adt_def.discriminants(cx.tcx).map(|(variant_index, discr)| { + let name = Cow::from(enum_adt_def.variant(variant_index).name.as_str()); + (name, discr.val) }), containing_scope, ), @@ -145,14 +145,11 @@ fn tag_base_type<'ll, 'tcx>( /// This is a helper function and does not register anything in the type map by itself. /// /// `variants` is an iterator of (discr-value, variant-name). -/// -// NOTE: Handling of discriminant values is somewhat inconsistent. They can appear as u128, -// u64, and i64. Here everything gets mapped to i64 because that's what LLVM's API expects. fn build_enumeration_type_di_node<'ll, 'tcx>( cx: &CodegenCx<'ll, 'tcx>, type_name: &str, base_type: Ty<'tcx>, - variants: &mut dyn Iterator<Item = (Discr<'tcx>, Cow<'tcx, str>)>, + enumerators: impl Iterator<Item = (Cow<'tcx, str>, u128)>, containing_scope: &'ll DIType, ) -> &'ll DIType { let is_unsigned = match base_type.kind() { @@ -160,24 +157,22 @@ fn build_enumeration_type_di_node<'ll, 'tcx>( ty::Uint(_) => true, _ => bug!("build_enumeration_type_di_node() called with non-integer tag type."), }; + let (size, align) = cx.size_and_align_of(base_type); - let enumerator_di_nodes: SmallVec<Option<&'ll DIType>> = variants - .map(|(discr, variant_name)| { - unsafe { - Some(llvm::LLVMRustDIBuilderCreateEnumerator( - DIB(cx), - variant_name.as_ptr().cast(), - variant_name.len(), - // FIXME: what if enumeration has i128 discriminant? - discr.val as i64, - is_unsigned, - )) - } + let enumerator_di_nodes: SmallVec<Option<&'ll DIType>> = enumerators + .map(|(name, value)| unsafe { + let value = [value as u64, (value >> 64) as u64]; + Some(llvm::LLVMRustDIBuilderCreateEnumerator( + DIB(cx), + name.as_ptr().cast(), + name.len(), + value.as_ptr(), + size.bits() as libc::c_uint, + is_unsigned, + )) }) .collect(); - let (size, align) = cx.size_and_align_of(base_type); - unsafe { llvm::LLVMRustDIBuilderCreateEnumerationType( DIB(cx), @@ -247,30 +242,34 @@ fn build_enumeration_type_di_node<'ll, 'tcx>( /// and a DW_TAG_member for each field (but not the discriminant). fn build_enum_variant_struct_type_di_node<'ll, 'tcx>( cx: &CodegenCx<'ll, 'tcx>, - enum_type: Ty<'tcx>, + enum_type_and_layout: TyAndLayout<'tcx>, enum_type_di_node: &'ll DIType, variant_index: VariantIdx, variant_def: &VariantDef, variant_layout: TyAndLayout<'tcx>, ) -> &'ll DIType { - debug_assert_eq!(variant_layout.ty, enum_type); + debug_assert_eq!(variant_layout.ty, enum_type_and_layout.ty); type_map::build_type_with_children( cx, type_map::stub( cx, Stub::Struct, - UniqueTypeId::for_enum_variant_struct_type(cx.tcx, enum_type, variant_index), + UniqueTypeId::for_enum_variant_struct_type( + cx.tcx, + enum_type_and_layout.ty, + variant_index, + ), variant_def.name.as_str(), // NOTE: We use size and align of enum_type, not from variant_layout: - cx.size_and_align_of(enum_type), + size_and_align_of(enum_type_and_layout), Some(enum_type_di_node), DIFlags::FlagZero, ), |cx, struct_type_di_node| { (0..variant_layout.fields.count()) .map(|field_index| { - let field_name = if variant_def.ctor_kind != CtorKind::Fn { + let field_name = if variant_def.ctor_kind() != Some(CtorKind::Fn) { // Fields have names Cow::from(variant_def.fields[field_index].name.as_str()) } else { @@ -290,9 +289,9 @@ fn build_enum_variant_struct_type_di_node<'ll, 'tcx>( type_di_node(cx, field_layout.ty), ) }) - .collect() + .collect::<SmallVec<_>>() }, - |cx| build_generic_type_param_di_nodes(cx, enum_type), + |cx| build_generic_type_param_di_nodes(cx, enum_type_and_layout.ty), ) .di_node } @@ -398,39 +397,60 @@ pub fn build_generator_variant_struct_type_di_node<'ll, 'tcx>( .di_node } +#[derive(Copy, Clone)] +enum DiscrResult { + NoDiscriminant, + Value(u128), + Range(u128, u128), +} + +impl DiscrResult { + fn opt_single_val(&self) -> Option<u128> { + if let Self::Value(d) = *self { Some(d) } else { None } + } +} + /// Returns the discriminant value corresponding to the variant index. /// /// Will return `None` if there is less than two variants (because then the enum won't have) -/// a tag, and if this is the dataful variant of a niche-layout enum (because then there is no +/// a tag, and if this is the untagged variant of a niche-layout enum (because then there is no /// single discriminant value). fn compute_discriminant_value<'ll, 'tcx>( cx: &CodegenCx<'ll, 'tcx>, enum_type_and_layout: TyAndLayout<'tcx>, variant_index: VariantIdx, -) -> Option<u64> { +) -> DiscrResult { match enum_type_and_layout.layout.variants() { - &Variants::Single { .. } => None, - &Variants::Multiple { tag_encoding: TagEncoding::Direct, .. } => Some( - enum_type_and_layout.ty.discriminant_for_variant(cx.tcx, variant_index).unwrap().val - as u64, + &Variants::Single { .. } => DiscrResult::NoDiscriminant, + &Variants::Multiple { tag_encoding: TagEncoding::Direct, .. } => DiscrResult::Value( + enum_type_and_layout.ty.discriminant_for_variant(cx.tcx, variant_index).unwrap().val, ), &Variants::Multiple { - tag_encoding: TagEncoding::Niche { ref niche_variants, niche_start, dataful_variant }, + tag_encoding: TagEncoding::Niche { ref niche_variants, niche_start, untagged_variant }, tag, .. } => { - if variant_index == dataful_variant { - None + if variant_index == untagged_variant { + let valid_range = enum_type_and_layout + .for_variant(cx, variant_index) + .largest_niche + .as_ref() + .unwrap() + .valid_range; + + let min = valid_range.start.min(valid_range.end); + let min = tag.size(cx).truncate(min); + + let max = valid_range.start.max(valid_range.end); + let max = tag.size(cx).truncate(max); + + DiscrResult::Range(min, max) } else { let value = (variant_index.as_u32() as u128) .wrapping_sub(niche_variants.start().as_u32() as u128) .wrapping_add(niche_start); let value = tag.size(cx).truncate(value); - // NOTE(eddyb) do *NOT* remove this assert, until - // we pass the full 128-bit value to LLVM, otherwise - // truncation will be silent and remain undetected. - assert_eq!(value as u64 as u128, value); - Some(value as u64) + DiscrResult::Value(value) } } } diff --git a/compiler/rustc_codegen_llvm/src/debuginfo/metadata/enums/native.rs b/compiler/rustc_codegen_llvm/src/debuginfo/metadata/enums/native.rs index f1935e0ec31..becbccc434d 100644 --- a/compiler/rustc_codegen_llvm/src/debuginfo/metadata/enums/native.rs +++ b/compiler/rustc_codegen_llvm/src/debuginfo/metadata/enums/native.rs @@ -88,7 +88,7 @@ pub(super) fn build_enum_type_di_node<'ll, 'tcx>( variant_name: Cow::from(enum_adt_def.variant(variant_index).name.as_str()), variant_struct_type_di_node: super::build_enum_variant_struct_type_di_node( cx, - enum_type, + enum_type_and_layout, enum_type_di_node, variant_index, enum_adt_def.variant(variant_index), @@ -378,7 +378,7 @@ fn build_discr_member_di_node<'ll, 'tcx>( /// /// The DW_AT_discr_value is optional, and is omitted if /// - This is the only variant of a univariant enum (i.e. their is no discriminant) -/// - This is the "dataful" variant of a niche-layout enum +/// - This is the "untagged" variant of a niche-layout enum /// (where only the other variants are identified by a single value) /// /// There is only ever a single member, the type of which is a struct that describes the @@ -413,7 +413,13 @@ fn build_enum_variant_member_di_node<'ll, 'tcx>( enum_type_and_layout.size.bits(), enum_type_and_layout.align.abi.bits() as u32, Size::ZERO.bits(), - discr_value.map(|v| cx.const_u64(v)), + discr_value.opt_single_val().map(|value| { + // NOTE(eddyb) do *NOT* remove this assert, until + // we pass the full 128-bit value to LLVM, otherwise + // truncation will be silent and remain undetected. + assert_eq!(value as u64 as u128, value); + cx.const_u64(value as u64) + }), DIFlags::FlagZero, variant_member_info.variant_struct_type_di_node, ) diff --git a/compiler/rustc_codegen_llvm/src/debuginfo/metadata/type_map.rs b/compiler/rustc_codegen_llvm/src/debuginfo/metadata/type_map.rs index 87fbb737ea8..e30622cbdce 100644 --- a/compiler/rustc_codegen_llvm/src/debuginfo/metadata/type_map.rs +++ b/compiler/rustc_codegen_llvm/src/debuginfo/metadata/type_map.rs @@ -47,6 +47,8 @@ pub(super) enum UniqueTypeId<'tcx> { VariantPart(Ty<'tcx>, private::HiddenZst), /// The ID for the artificial struct type describing a single enum variant. VariantStructType(Ty<'tcx>, VariantIdx, private::HiddenZst), + /// The ID for the additional wrapper struct type describing an enum variant in CPP-like mode. + VariantStructTypeCppLikeWrapper(Ty<'tcx>, VariantIdx, private::HiddenZst), /// The ID of the artificial type we create for VTables. VTableTy(Ty<'tcx>, Option<PolyExistentialTraitRef<'tcx>>, private::HiddenZst), } @@ -71,6 +73,15 @@ impl<'tcx> UniqueTypeId<'tcx> { UniqueTypeId::VariantStructType(enum_ty, variant_idx, private::HiddenZst) } + pub fn for_enum_variant_struct_type_wrapper( + tcx: TyCtxt<'tcx>, + enum_ty: Ty<'tcx>, + variant_idx: VariantIdx, + ) -> Self { + debug_assert_eq!(enum_ty, tcx.normalize_erasing_regions(ParamEnv::reveal_all(), enum_ty)); + UniqueTypeId::VariantStructTypeCppLikeWrapper(enum_ty, variant_idx, private::HiddenZst) + } + pub fn for_vtable_ty( tcx: TyCtxt<'tcx>, self_type: Ty<'tcx>, @@ -93,8 +104,9 @@ impl<'tcx> UniqueTypeId<'tcx> { /// Right now this takes the form of a hex-encoded opaque hash value. pub fn generate_unique_id_string(self, tcx: TyCtxt<'tcx>) -> String { let mut hasher = StableHasher::new(); - let mut hcx = tcx.create_stable_hashing_context(); - hcx.while_hashing_spans(false, |hcx| self.hash_stable(hcx, &mut hasher)); + tcx.with_stable_hashing_context(|mut hcx| { + hcx.while_hashing_spans(false, |hcx| self.hash_stable(hcx, &mut hasher)) + }); hasher.finish::<Fingerprint>().to_hex() } @@ -145,7 +157,7 @@ impl<'ll> DINodeCreationResult<'ll> { pub enum Stub<'ll> { Struct, Union, - VtableTy { vtable_holder: &'ll DIType }, + VTableTy { vtable_holder: &'ll DIType }, } pub struct StubInfo<'ll, 'tcx> { @@ -179,9 +191,9 @@ pub(super) fn stub<'ll, 'tcx>( let unique_type_id_str = unique_type_id.generate_unique_id_string(cx.tcx); let metadata = match kind { - Stub::Struct | Stub::VtableTy { .. } => { + Stub::Struct | Stub::VTableTy { .. } => { let vtable_holder = match kind { - Stub::VtableTy { vtable_holder } => Some(vtable_holder), + Stub::VTableTy { vtable_holder } => Some(vtable_holder), _ => None, }; unsafe { diff --git a/compiler/rustc_codegen_llvm/src/debuginfo/mod.rs b/compiler/rustc_codegen_llvm/src/debuginfo/mod.rs index 71699b5cf38..ca7a07d8391 100644 --- a/compiler/rustc_codegen_llvm/src/debuginfo/mod.rs +++ b/compiler/rustc_codegen_llvm/src/debuginfo/mod.rs @@ -27,7 +27,7 @@ use rustc_index::vec::IndexVec; use rustc_middle::mir; use rustc_middle::ty::layout::LayoutOf; use rustc_middle::ty::subst::{GenericArgKind, SubstsRef}; -use rustc_middle::ty::{self, Instance, ParamEnv, Ty, TypeFoldable}; +use rustc_middle::ty::{self, Instance, ParamEnv, Ty, TypeVisitable}; use rustc_session::config::{self, DebugInfo}; use rustc_session::Session; use rustc_span::symbol::Symbol; @@ -39,7 +39,7 @@ use smallvec::SmallVec; use std::cell::OnceCell; use std::cell::RefCell; use std::iter; -use tracing::debug; +use std::ops::Range; mod create_scope_map; pub mod gdb; @@ -97,23 +97,26 @@ impl<'ll, 'tcx> CodegenUnitDebugContext<'ll, 'tcx> { unsafe { llvm::LLVMRustDIBuilderFinalize(self.builder); - // Debuginfo generation in LLVM by default uses a higher - // version of dwarf than macOS currently understands. We can - // instruct LLVM to emit an older version of dwarf, however, - // for macOS to understand. For more info see #11352 - // This can be overridden using --llvm-opts -dwarf-version,N. - // Android has the same issue (#22398) - if let Some(version) = sess.target.dwarf_version { + if !sess.target.is_like_msvc { + // Debuginfo generation in LLVM by default uses a higher + // version of dwarf than macOS currently understands. We can + // instruct LLVM to emit an older version of dwarf, however, + // for macOS to understand. For more info see #11352 + // This can be overridden using --llvm-opts -dwarf-version,N. + // Android has the same issue (#22398) + let dwarf_version = sess + .opts + .unstable_opts + .dwarf_version + .unwrap_or(sess.target.default_dwarf_version); llvm::LLVMRustAddModuleFlag( self.llmod, llvm::LLVMModFlagBehavior::Warning, "Dwarf Version\0".as_ptr().cast(), - version, - ) - } - - // Indicate that we want CodeView debug information on MSVC - if sess.target.is_like_msvc { + dwarf_version, + ); + } else { + // Indicate that we want CodeView debug information on MSVC llvm::LLVMRustAddModuleFlag( self.llmod, llvm::LLVMModFlagBehavior::Warning, @@ -161,12 +164,14 @@ impl<'ll> DebugInfoBuilderMethods for Builder<'_, 'll, '_> { variable_alloca: Self::Value, direct_offset: Size, indirect_offsets: &[Size], + fragment: Option<Range<Size>>, ) { - // Convert the direct and indirect offsets to address ops. + // Convert the direct and indirect offsets and fragment byte range to address ops. // FIXME(eddyb) use `const`s instead of getting the values via FFI, // the values should match the ones in the DWARF standard anyway. let op_deref = || unsafe { llvm::LLVMRustDIBuilderCreateOpDeref() }; let op_plus_uconst = || unsafe { llvm::LLVMRustDIBuilderCreateOpPlusUconst() }; + let op_llvm_fragment = || unsafe { llvm::LLVMRustDIBuilderCreateOpLLVMFragment() }; let mut addr_ops = SmallVec::<[u64; 8]>::new(); if direct_offset.bytes() > 0 { @@ -180,6 +185,13 @@ impl<'ll> DebugInfoBuilderMethods for Builder<'_, 'll, '_> { addr_ops.push(offset.bytes() as u64); } } + if let Some(fragment) = fragment { + // `DW_OP_LLVM_fragment` takes as arguments the fragment's + // offset and size, both of them in bits. + addr_ops.push(op_llvm_fragment()); + addr_ops.push(fragment.start.bits() as u64); + addr_ops.push((fragment.end - fragment.start).bits() as u64); + } unsafe { // FIXME(eddyb) replace `llvm.dbg.declare` with `llvm.dbg.addr`. diff --git a/compiler/rustc_codegen_llvm/src/debuginfo/utils.rs b/compiler/rustc_codegen_llvm/src/debuginfo/utils.rs index 8f243673907..5cd0e1cb63a 100644 --- a/compiler/rustc_codegen_llvm/src/debuginfo/utils.rs +++ b/compiler/rustc_codegen_llvm/src/debuginfo/utils.rs @@ -6,7 +6,7 @@ use super::CodegenUnitDebugContext; use rustc_hir::def_id::DefId; use rustc_middle::ty::layout::{HasParamEnv, LayoutOf}; use rustc_middle::ty::{self, DefIdTree, Ty}; -use tracing::trace; +use trace; use crate::common::CodegenCx; use crate::llvm; @@ -72,7 +72,7 @@ pub(crate) fn fat_pointer_kind<'ll, 'tcx>( layout.is_unsized() ); - if !layout.is_unsized() { + if layout.is_sized() { return None; } diff --git a/compiler/rustc_codegen_llvm/src/declare.rs b/compiler/rustc_codegen_llvm/src/declare.rs index 5a5c4f7f860..dc21a02cec4 100644 --- a/compiler/rustc_codegen_llvm/src/declare.rs +++ b/compiler/rustc_codegen_llvm/src/declare.rs @@ -18,9 +18,10 @@ use crate::llvm; use crate::llvm::AttributePlace::Function; use crate::type_::Type; use crate::value::Value; +use rustc_codegen_ssa::traits::TypeMembershipMethods; use rustc_middle::ty::Ty; +use rustc_symbol_mangling::typeid::typeid_for_fnabi; use smallvec::SmallVec; -use tracing::debug; /// Declare a function. /// @@ -31,6 +32,7 @@ fn declare_raw_fn<'ll>( name: &str, callconv: llvm::CallConv, unnamed: llvm::UnnamedAddr, + visibility: llvm::Visibility, ty: &'ll Type, ) -> &'ll Value { debug!("declare_raw_fn(name={:?}, ty={:?})", name, ty); @@ -40,6 +42,7 @@ fn declare_raw_fn<'ll>( llvm::SetFunctionCallConv(llfn, callconv); llvm::SetUnnamedAddress(llfn, unnamed); + llvm::set_visibility(llfn, visibility); let mut attrs = SmallVec::<[_; 4]>::new(); @@ -77,7 +80,36 @@ impl<'ll, 'tcx> CodegenCx<'ll, 'tcx> { unnamed: llvm::UnnamedAddr, fn_type: &'ll Type, ) -> &'ll Value { - declare_raw_fn(self, name, llvm::CCallConv, unnamed, fn_type) + // Declare C ABI functions with the visibility used by C by default. + let visibility = if self.tcx.sess.target.default_hidden_visibility { + llvm::Visibility::Hidden + } else { + llvm::Visibility::Default + }; + + declare_raw_fn(self, name, llvm::CCallConv, unnamed, visibility, fn_type) + } + + /// Declare an entry Function + /// + /// The ABI of this function can change depending on the target (although for now the same as + /// `declare_cfn`) + /// + /// If there’s a value with the same name already declared, the function will + /// update the declaration and return existing Value instead. + pub fn declare_entry_fn( + &self, + name: &str, + callconv: llvm::CallConv, + unnamed: llvm::UnnamedAddr, + fn_type: &'ll Type, + ) -> &'ll Value { + let visibility = if self.tcx.sess.target.default_hidden_visibility { + llvm::Visibility::Hidden + } else { + llvm::Visibility::Default + }; + declare_raw_fn(self, name, callconv, unnamed, visibility, fn_type) } /// Declare a Rust function. @@ -94,9 +126,16 @@ impl<'ll, 'tcx> CodegenCx<'ll, 'tcx> { name, fn_abi.llvm_cconv(), llvm::UnnamedAddr::Global, + llvm::Visibility::Default, fn_abi.llvm_type(self), ); fn_abi.apply_attrs_llfn(self, llfn); + + if self.tcx.sess.is_sanitizer_cfi_enabled() { + let typeid = typeid_for_fnabi(self.tcx, fn_abi); + self.set_type_metadata(llfn, typeid); + } + llfn } diff --git a/compiler/rustc_codegen_llvm/src/errors.rs b/compiler/rustc_codegen_llvm/src/errors.rs new file mode 100644 index 00000000000..af9f31fc324 --- /dev/null +++ b/compiler/rustc_codegen_llvm/src/errors.rs @@ -0,0 +1,120 @@ +use std::borrow::Cow; + +use rustc_errors::fluent; +use rustc_errors::DiagnosticBuilder; +use rustc_errors::ErrorGuaranteed; +use rustc_errors::Handler; +use rustc_errors::IntoDiagnostic; +use rustc_macros::{Diagnostic, Subdiagnostic}; +use rustc_span::Span; + +#[derive(Diagnostic)] +#[diag(codegen_llvm_unknown_ctarget_feature_prefix)] +#[note] +pub(crate) struct UnknownCTargetFeaturePrefix<'a> { + pub feature: &'a str, +} + +#[derive(Diagnostic)] +#[diag(codegen_llvm_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(possible_feature)] + Some { rust_feature: &'a str }, + #[help(consider_filing_feature_request)] + None, +} + +#[derive(Diagnostic)] +#[diag(codegen_llvm_error_creating_import_library)] +pub(crate) struct ErrorCreatingImportLibrary<'a> { + pub lib_name: &'a str, + pub error: String, +} + +#[derive(Diagnostic)] +#[diag(codegen_llvm_instrument_coverage_requires_llvm_12)] +pub(crate) struct InstrumentCoverageRequiresLLVM12; + +#[derive(Diagnostic)] +#[diag(codegen_llvm_symbol_already_defined)] +pub(crate) struct SymbolAlreadyDefined<'a> { + #[primary_span] + pub span: Span, + pub symbol_name: &'a str, +} + +#[derive(Diagnostic)] +#[diag(codegen_llvm_branch_protection_requires_aarch64)] +pub(crate) struct BranchProtectionRequiresAArch64; + +#[derive(Diagnostic)] +#[diag(codegen_llvm_invalid_minimum_alignment)] +pub(crate) struct InvalidMinimumAlignment { + pub err: String, +} + +#[derive(Diagnostic)] +#[diag(codegen_llvm_sanitizer_memtag_requires_mte)] +pub(crate) struct SanitizerMemtagRequiresMte; + +#[derive(Diagnostic)] +#[diag(codegen_llvm_error_writing_def_file)] +pub(crate) struct ErrorWritingDEFFile { + pub error: std::io::Error, +} + +#[derive(Diagnostic)] +#[diag(codegen_llvm_error_calling_dlltool)] +pub(crate) struct ErrorCallingDllTool { + pub error: std::io::Error, +} + +#[derive(Diagnostic)] +#[diag(codegen_llvm_dlltool_fail_import_library)] +pub(crate) struct DlltoolFailImportLibrary<'a> { + pub stdout: Cow<'a, str>, + pub stderr: Cow<'a, str>, +} + +#[derive(Diagnostic)] +#[diag(codegen_llvm_dynamic_linking_with_lto)] +#[note] +pub(crate) struct DynamicLinkingWithLTO; + +#[derive(Diagnostic)] +#[diag(codegen_llvm_fail_parsing_target_machine_config_to_target_machine)] +pub(crate) struct FailParsingTargetMachineConfigToTargetMachine { + pub error: String, +} + +pub(crate) struct TargetFeatureDisableOrEnable<'a> { + pub features: &'a [&'a str], + pub span: Option<Span>, + pub missing_features: Option<MissingFeatures>, +} + +#[derive(Subdiagnostic)] +#[help(codegen_llvm_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_llvm_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 + } +} diff --git a/compiler/rustc_codegen_llvm/src/intrinsic.rs b/compiler/rustc_codegen_llvm/src/intrinsic.rs index a18f5b9dd7f..2f5dd519b26 100644 --- a/compiler/rustc_codegen_llvm/src/intrinsic.rs +++ b/compiler/rustc_codegen_llvm/src/intrinsic.rs @@ -71,6 +71,7 @@ fn get_simple_intrinsic<'ll>( sym::nearbyintf64 => "llvm.nearbyint.f64", sym::roundf32 => "llvm.round.f32", sym::roundf64 => "llvm.round.f64", + sym::ptr_mask => "llvm.ptrmask", _ => return None, }; Some(cx.get_intrinsic(llvm_name)) @@ -107,6 +108,7 @@ impl<'ll, 'tcx> IntrinsicCallMethods<'tcx> for Builder<'_, 'll, 'tcx> { let (simple_ty, simple_fn) = simple.unwrap(); self.call( simple_ty, + None, simple_fn, &args.iter().map(|arg| arg.immediate()).collect::<Vec<_>>(), None, @@ -161,7 +163,7 @@ impl<'ll, 'tcx> IntrinsicCallMethods<'tcx> for Builder<'_, 'll, 'tcx> { sym::volatile_load | sym::unaligned_volatile_load => { let tp_ty = substs.type_at(0); let ptr = args[0].immediate(); - let load = if let PassMode::Cast(ty) = fn_abi.ret.mode { + let load = if let PassMode::Cast(ty, _) = &fn_abi.ret.mode { let llty = ty.llvm_type(self); let ptr = self.pointercast(ptr, self.type_ptr_to(llty)); self.volatile_load(llty, ptr) @@ -338,17 +340,26 @@ impl<'ll, 'tcx> IntrinsicCallMethods<'tcx> for Builder<'_, 'll, 'tcx> { sym::black_box => { args[0].val.store(self, result); - + let result_val_span = [result.llval]; // We need to "use" the argument in some way LLVM can't introspect, and on // targets that support it we can typically leverage inline assembly to do // this. LLVM's interpretation of inline assembly is that it's, well, a black // box. This isn't the greatest implementation since it probably deoptimizes // more than we want, but it's so far good enough. + // + // For zero-sized types, the location pointed to by the result may be + // uninitialized. Do not "use" the result in this case; instead just clobber + // the memory. + let (constraint, inputs): (&str, &[_]) = if result.layout.is_zst() { + ("~{memory}", &[]) + } else { + ("r,~{memory}", &result_val_span) + }; crate::asm::inline_asm_call( self, "", - "r,~{memory}", - &[result.llval], + constraint, + inputs, self.type_void(), true, false, @@ -374,7 +385,7 @@ impl<'ll, 'tcx> IntrinsicCallMethods<'tcx> for Builder<'_, 'll, 'tcx> { }; if !fn_abi.ret.is_ignore() { - if let PassMode::Cast(ty) = fn_abi.ret.mode { + if let PassMode::Cast(ty, _) = &fn_abi.ret.mode { let ptr_llty = self.type_ptr_to(ty.llvm_type(self)); let ptr = self.pointercast(result.llval, ptr_llty); self.store(llval, ptr, result.align); @@ -434,14 +445,14 @@ fn try_intrinsic<'ll>( ) { if bx.sess().panic_strategy() == PanicStrategy::Abort { let try_func_ty = bx.type_func(&[bx.type_i8p()], bx.type_void()); - bx.call(try_func_ty, try_func, &[data], None); + bx.call(try_func_ty, None, try_func, &[data], None); // Return 0 unconditionally from the intrinsic call; // we can never unwind. let ret_align = bx.tcx().data_layout.i32_align.abi; bx.store(bx.const_i32(0), dest, ret_align); } else if wants_msvc_seh(bx.sess()) { codegen_msvc_try(bx, try_func, data, catch_func, dest); - } else if bx.sess().target.is_like_emscripten { + } else if bx.sess().target.os == "emscripten" { codegen_emcc_try(bx, try_func, data, catch_func, dest); } else { codegen_gnu_try(bx, try_func, data, catch_func, dest); @@ -533,7 +544,7 @@ fn codegen_msvc_try<'ll>( let ptr_align = bx.tcx().data_layout.pointer_align.abi; let slot = bx.alloca(bx.type_i8p(), ptr_align); let try_func_ty = bx.type_func(&[bx.type_i8p()], bx.type_void()); - bx.invoke(try_func_ty, try_func, &[data], normal, catchswitch, None); + bx.invoke(try_func_ty, None, try_func, &[data], normal, catchswitch, None); bx.switch_to_block(normal); bx.ret(bx.const_i32(0)); @@ -577,7 +588,7 @@ fn codegen_msvc_try<'ll>( let funclet = bx.catch_pad(cs, &[tydesc, flags, slot]); let ptr = bx.load(bx.type_i8p(), slot, ptr_align); let catch_ty = bx.type_func(&[bx.type_i8p(), bx.type_i8p()], bx.type_void()); - bx.call(catch_ty, catch_func, &[data, ptr], Some(&funclet)); + bx.call(catch_ty, None, catch_func, &[data, ptr], Some(&funclet)); bx.catch_ret(&funclet, caught); // The flag value of 64 indicates a "catch-all". @@ -585,7 +596,7 @@ fn codegen_msvc_try<'ll>( let flags = bx.const_i32(64); let null = bx.const_null(bx.type_i8p()); let funclet = bx.catch_pad(cs, &[null, flags, null]); - bx.call(catch_ty, catch_func, &[data, null], Some(&funclet)); + bx.call(catch_ty, None, catch_func, &[data, null], Some(&funclet)); bx.catch_ret(&funclet, caught); bx.switch_to_block(caught); @@ -594,7 +605,7 @@ fn codegen_msvc_try<'ll>( // Note that no invoke is used here because by definition this function // can't panic (that's what it's catching). - let ret = bx.call(llty, llfn, &[try_func, data, catch_func], None); + let ret = bx.call(llty, None, llfn, &[try_func, data, catch_func], None); let i32_align = bx.tcx().data_layout.i32_align.abi; bx.store(ret, dest, i32_align); } @@ -637,7 +648,7 @@ fn codegen_gnu_try<'ll>( let data = llvm::get_param(bx.llfn(), 1); let catch_func = llvm::get_param(bx.llfn(), 2); let try_func_ty = bx.type_func(&[bx.type_i8p()], bx.type_void()); - bx.invoke(try_func_ty, try_func, &[data], then, catch, None); + bx.invoke(try_func_ty, None, try_func, &[data], then, catch, None); bx.switch_to_block(then); bx.ret(bx.const_i32(0)); @@ -655,13 +666,13 @@ fn codegen_gnu_try<'ll>( bx.add_clause(vals, tydesc); let ptr = bx.extract_value(vals, 0); let catch_ty = bx.type_func(&[bx.type_i8p(), bx.type_i8p()], bx.type_void()); - bx.call(catch_ty, catch_func, &[data, ptr], None); + bx.call(catch_ty, None, catch_func, &[data, ptr], None); bx.ret(bx.const_i32(1)); }); // Note that no invoke is used here because by definition this function // can't panic (that's what it's catching). - let ret = bx.call(llty, llfn, &[try_func, data, catch_func], None); + let ret = bx.call(llty, None, llfn, &[try_func, data, catch_func], None); let i32_align = bx.tcx().data_layout.i32_align.abi; bx.store(ret, dest, i32_align); } @@ -701,7 +712,7 @@ fn codegen_emcc_try<'ll>( let data = llvm::get_param(bx.llfn(), 1); let catch_func = llvm::get_param(bx.llfn(), 2); let try_func_ty = bx.type_func(&[bx.type_i8p()], bx.type_void()); - bx.invoke(try_func_ty, try_func, &[data], then, catch, None); + bx.invoke(try_func_ty, None, try_func, &[data], then, catch, None); bx.switch_to_block(then); bx.ret(bx.const_i32(0)); @@ -740,13 +751,13 @@ fn codegen_emcc_try<'ll>( let catch_data = bx.bitcast(catch_data, bx.type_i8p()); let catch_ty = bx.type_func(&[bx.type_i8p(), bx.type_i8p()], bx.type_void()); - bx.call(catch_ty, catch_func, &[data, catch_data], None); + bx.call(catch_ty, None, catch_func, &[data, catch_data], None); bx.ret(bx.const_i32(1)); }); // Note that no invoke is used here because by definition this function // can't panic (that's what it's catching). - let ret = bx.call(llty, llfn, &[try_func, data, catch_func], None); + let ret = bx.call(llty, None, llfn, &[try_func, data, catch_func], None); let i32_align = bx.tcx().data_layout.i32_align.abi; bx.store(ret, dest, i32_align); } @@ -1216,8 +1227,13 @@ fn generic_simd_intrinsic<'ll, 'tcx>( }; let llvm_name = &format!("llvm.{0}.v{1}{2}", intr_name, in_len, elem_ty_str); let f = bx.declare_cfn(llvm_name, llvm::UnnamedAddr::No, fn_ty); - let c = - bx.call(fn_ty, f, &args.iter().map(|arg| arg.immediate()).collect::<Vec<_>>(), None); + let c = bx.call( + fn_ty, + None, + f, + &args.iter().map(|arg| arg.immediate()).collect::<Vec<_>>(), + None, + ); Ok(c) } @@ -1416,8 +1432,13 @@ fn generic_simd_intrinsic<'ll, 'tcx>( llvm_elem_vec_ty, ); let f = bx.declare_cfn(&llvm_intrinsic, llvm::UnnamedAddr::No, fn_ty); - let v = - bx.call(fn_ty, f, &[args[1].immediate(), alignment, mask, args[0].immediate()], None); + let v = bx.call( + fn_ty, + None, + f, + &[args[1].immediate(), alignment, mask, args[0].immediate()], + None, + ); return Ok(v); } @@ -1479,7 +1500,7 @@ fn generic_simd_intrinsic<'ll, 'tcx>( let (_, element_ty1) = arg_tys[1].simd_size_and_type(bx.tcx()); let (_, element_ty2) = arg_tys[2].simd_size_and_type(bx.tcx()); let (pointer_count, underlying_ty) = match element_ty1.kind() { - ty::RawPtr(p) if p.ty == in_elem && p.mutbl == hir::Mutability::Mut => { + ty::RawPtr(p) if p.ty == in_elem && p.mutbl.is_mut() => { (ptr_count(element_ty1), non_ptr(element_ty1)) } _ => { @@ -1542,8 +1563,13 @@ fn generic_simd_intrinsic<'ll, 'tcx>( let fn_ty = bx.type_func(&[llvm_elem_vec_ty, llvm_pointer_vec_ty, alignment_ty, mask_ty], ret_t); let f = bx.declare_cfn(&llvm_intrinsic, llvm::UnnamedAddr::No, fn_ty); - let v = - bx.call(fn_ty, f, &[args[0].immediate(), args[1].immediate(), alignment, mask], None); + let v = bx.call( + fn_ty, + None, + f, + &[args[0].immediate(), args[1].immediate(), alignment, mask], + None, + ); return Ok(v); } @@ -1704,6 +1730,97 @@ unsupported {} from `{}` with element `{}` of size `{}` to `{}`"#, bitwise_red!(simd_reduce_all: vector_reduce_and, true); bitwise_red!(simd_reduce_any: vector_reduce_or, true); + if name == sym::simd_cast_ptr { + require_simd!(ret_ty, "return"); + let (out_len, out_elem) = ret_ty.simd_size_and_type(bx.tcx()); + require!( + in_len == out_len, + "expected return type with length {} (same as input type `{}`), \ + found `{}` with length {}", + in_len, + in_ty, + ret_ty, + out_len + ); + + match in_elem.kind() { + ty::RawPtr(p) => { + let (metadata, check_sized) = p.ty.ptr_metadata_ty(bx.tcx, |ty| { + bx.tcx.normalize_erasing_regions(ty::ParamEnv::reveal_all(), ty) + }); + assert!(!check_sized); // we are in codegen, so we shouldn't see these types + require!(metadata.is_unit(), "cannot cast fat pointer `{}`", in_elem) + } + _ => return_error!("expected pointer, got `{}`", in_elem), + } + match out_elem.kind() { + ty::RawPtr(p) => { + let (metadata, check_sized) = p.ty.ptr_metadata_ty(bx.tcx, |ty| { + bx.tcx.normalize_erasing_regions(ty::ParamEnv::reveal_all(), ty) + }); + assert!(!check_sized); // we are in codegen, so we shouldn't see these types + require!(metadata.is_unit(), "cannot cast to fat pointer `{}`", out_elem) + } + _ => return_error!("expected pointer, got `{}`", out_elem), + } + + if in_elem == out_elem { + return Ok(args[0].immediate()); + } else { + return Ok(bx.pointercast(args[0].immediate(), llret_ty)); + } + } + + if name == sym::simd_expose_addr { + require_simd!(ret_ty, "return"); + let (out_len, out_elem) = ret_ty.simd_size_and_type(bx.tcx()); + require!( + in_len == out_len, + "expected return type with length {} (same as input type `{}`), \ + found `{}` with length {}", + in_len, + in_ty, + ret_ty, + out_len + ); + + match in_elem.kind() { + ty::RawPtr(_) => {} + _ => return_error!("expected pointer, got `{}`", in_elem), + } + match out_elem.kind() { + ty::Uint(ty::UintTy::Usize) => {} + _ => return_error!("expected `usize`, got `{}`", out_elem), + } + + return Ok(bx.ptrtoint(args[0].immediate(), llret_ty)); + } + + if name == sym::simd_from_exposed_addr { + require_simd!(ret_ty, "return"); + let (out_len, out_elem) = ret_ty.simd_size_and_type(bx.tcx()); + require!( + in_len == out_len, + "expected return type with length {} (same as input type `{}`), \ + found `{}` with length {}", + in_len, + in_ty, + ret_ty, + out_len + ); + + match in_elem.kind() { + ty::Uint(ty::UintTy::Usize) => {} + _ => return_error!("expected `usize`, got `{}`", in_elem), + } + match out_elem.kind() { + ty::RawPtr(_) => {} + _ => return_error!("expected pointer, got `{}`", out_elem), + } + + return Ok(bx.inttoptr(args[0].immediate(), llret_ty)); + } + if name == sym::simd_cast || name == sym::simd_as { require_simd!(ret_ty, "return"); let (out_len, out_elem) = ret_ty.simd_size_and_type(bx.tcx()); @@ -1900,7 +2017,7 @@ unsupported {} from `{}` with element `{}` of size `{}` to `{}`"#, let fn_ty = bx.type_func(&[vec_ty, vec_ty], vec_ty); let f = bx.declare_cfn(llvm_intrinsic, llvm::UnnamedAddr::No, fn_ty); - let v = bx.call(fn_ty, f, &[lhs, rhs], None); + let v = bx.call(fn_ty, None, f, &[lhs, rhs], None); return Ok(v); } diff --git a/compiler/rustc_codegen_llvm/src/lib.rs b/compiler/rustc_codegen_llvm/src/lib.rs index 6713a756735..246e82545c8 100644 --- a/compiler/rustc_codegen_llvm/src/lib.rs +++ b/compiler/rustc_codegen_llvm/src/lib.rs @@ -5,19 +5,24 @@ //! This API is completely unstable and subject to change. #![doc(html_root_url = "https://doc.rust-lang.org/nightly/nightly-rustc/")] +#![feature(hash_raw_entry)] #![feature(let_chains)] -#![feature(let_else)] #![feature(extern_types)] #![feature(once_cell)] #![feature(iter_intersperse)] #![recursion_limit = "256"] #![allow(rustc::potential_query_instability)] +#![deny(rustc::untranslatable_diagnostic)] +#![deny(rustc::diagnostic_outside_of_impl)] #[macro_use] extern crate rustc_macros; +#[macro_use] +extern crate tracing; use back::write::{create_informational_target_machine, create_target_machine}; +use errors::FailParsingTargetMachineConfigToTargetMachine; pub use llvm_util::target_features; use rustc_ast::expand::allocator::AllocatorKind; use rustc_codegen_ssa::back::lto::{LtoModuleCodegen, SerializedModule, ThinModule}; @@ -60,6 +65,7 @@ mod context; mod coverageinfo; mod debuginfo; mod declare; +mod errors; mod intrinsic; // The following is a work around that replaces `pub mod llvm;` and that fixes issue 53912. @@ -106,11 +112,11 @@ impl ExtraBackendMethods for LlvmCodegenBackend { tcx: TyCtxt<'tcx>, module_name: &str, kind: AllocatorKind, - has_alloc_error_handler: bool, + alloc_error_handler_kind: AllocatorKind, ) -> ModuleLlvm { let mut module_llvm = ModuleLlvm::new_metadata(tcx, module_name); unsafe { - allocator::codegen(tcx, &mut module_llvm, module_name, kind, has_alloc_error_handler); + allocator::codegen(tcx, &mut module_llvm, module_name, kind, alloc_error_handler_kind); } module_llvm } @@ -129,12 +135,6 @@ impl ExtraBackendMethods for LlvmCodegenBackend { ) -> TargetMachineFactoryFn<Self> { back::write::target_machine_factory(sess, optlvl, target_features) } - fn target_cpu<'b>(&self, sess: &'b Session) -> &'b str { - llvm_util::target_cpu(sess) - } - fn tune_cpu<'b>(&self, sess: &'b Session) -> Option<&'b str> { - llvm_util::tune_cpu(sess) - } fn spawn_thread<F, T>(time_trace: bool, f: F) -> std::thread::JoinHandle<T> where @@ -168,7 +168,6 @@ impl ExtraBackendMethods for LlvmCodegenBackend { impl WriteBackendMethods for LlvmCodegenBackend { type Module = ModuleLlvm; type ModuleBuffer = back::lto::ModuleBuffer; - type Context = llvm::Context; type TargetMachine = &'static mut llvm::TargetMachine; type ThinData = back::lto::ThinData; type ThinBuffer = back::lto::ThinBuffer; @@ -323,8 +322,8 @@ impl CodegenBackend for LlvmCodegenBackend { llvm_util::print_version(); } - fn target_features(&self, sess: &Session) -> Vec<Symbol> { - target_features(sess) + fn target_features(&self, sess: &Session, allow_unstable: bool) -> Vec<Symbol> { + target_features(sess, allow_unstable) } fn codegen_crate<'tcx>( @@ -354,7 +353,7 @@ impl CodegenBackend for LlvmCodegenBackend { .join(sess); sess.time("llvm_dump_timing_file", || { - if sess.opts.debugging_opts.llvm_time_trace { + if sess.opts.unstable_opts.llvm_time_trace { let file_name = outputs.with_extension("llvm_timings.json"); llvm_util::time_trace_profiler_finish(&file_name); } @@ -369,12 +368,12 @@ impl CodegenBackend for LlvmCodegenBackend { codegen_results: CodegenResults, outputs: &OutputFilenames, ) -> Result<(), ErrorGuaranteed> { - use crate::back::archive::LlvmArchiveBuilder; + use crate::back::archive::LlvmArchiveBuilderBuilder; use rustc_codegen_ssa::back::link::link_binary; // Run the linker on any artifacts that resulted from the LLVM run. // This should produce either a finished executable or library. - link_binary::<LlvmArchiveBuilder<'_>>(sess, &codegen_results, outputs) + link_binary(sess, &LlvmArchiveBuilderBuilder, &codegen_results, outputs) } } @@ -417,7 +416,7 @@ impl ModuleLlvm { let tm = match (cgcx.tm_factory)(tm_factory_config) { Ok(m) => m, Err(e) => { - handler.struct_err(&e).emit(); + handler.emit_err(FailParsingTargetMachineConfigToTargetMachine { error: e }); return Err(FatalError); } }; diff --git a/compiler/rustc_codegen_llvm/src/llvm/archive_ro.rs b/compiler/rustc_codegen_llvm/src/llvm/archive_ro.rs index 64db4f7462d..7d948970223 100644 --- a/compiler/rustc_codegen_llvm/src/llvm/archive_ro.rs +++ b/compiler/rustc_codegen_llvm/src/llvm/archive_ro.rs @@ -83,17 +83,6 @@ impl<'a> Child<'a> { } } } - - pub fn data(&self) -> &'a [u8] { - unsafe { - let mut data_len = 0; - let data_ptr = super::LLVMRustArchiveChildData(self.raw, &mut data_len); - if data_ptr.is_null() { - panic!("failed to read data from archive child"); - } - slice::from_raw_parts(data_ptr as *const u8, data_len as usize) - } - } } impl<'a> Drop for Child<'a> { diff --git a/compiler/rustc_codegen_llvm/src/llvm/diagnostic.rs b/compiler/rustc_codegen_llvm/src/llvm/diagnostic.rs index e2fa5e488ed..45de284d22a 100644 --- a/compiler/rustc_codegen_llvm/src/llvm/diagnostic.rs +++ b/compiler/rustc_codegen_llvm/src/llvm/diagnostic.rs @@ -20,19 +20,6 @@ pub enum OptimizationDiagnosticKind { OptimizationRemarkOther, } -impl OptimizationDiagnosticKind { - pub fn describe(self) -> &'static str { - match self { - OptimizationRemark | OptimizationRemarkOther => "remark", - OptimizationMissed => "missed", - OptimizationAnalysis => "analysis", - OptimizationAnalysisFPCommute => "floating-point", - OptimizationAnalysisAliasing => "aliasing", - OptimizationFailure => "failure", - } - } -} - pub struct OptimizationDiagnostic<'ll> { pub kind: OptimizationDiagnosticKind, pub pass_name: String, diff --git a/compiler/rustc_codegen_llvm/src/llvm/ffi.rs b/compiler/rustc_codegen_llvm/src/llvm/ffi.rs index b831423994f..8a9392255b8 100644 --- a/compiler/rustc_codegen_llvm/src/llvm/ffi.rs +++ b/compiler/rustc_codegen_llvm/src/llvm/ffi.rs @@ -35,7 +35,7 @@ pub enum LLVMRustResult { pub struct LLVMRustCOFFShortExport { pub name: *const c_char, pub ordinal_present: bool, - // value of `ordinal` only important when `ordinal_present` is true + /// value of `ordinal` only important when `ordinal_present` is true pub ordinal: u16, } @@ -183,7 +183,6 @@ pub enum AttributeKind { OptimizeNone = 24, ReturnsTwice = 25, ReadNone = 26, - InaccessibleMemOnly = 27, SanitizeHWAddress = 28, WillReturn = 29, StackProtectReq = 30, @@ -191,6 +190,11 @@ pub enum AttributeKind { StackProtect = 32, NoUndef = 33, SanitizeMemTag = 34, + NoCfCheck = 35, + ShadowCallStack = 36, + AllocSize = 37, + AllocatedPointer = 38, + AllocAlign = 39, } /// LLVMIntPredicate @@ -395,27 +399,6 @@ impl AtomicOrdering { } } -/// LLVMRustSynchronizationScope -#[derive(Copy, Clone)] -#[repr(C)] -pub enum SynchronizationScope { - SingleThread, - CrossThread, -} - -impl SynchronizationScope { - pub fn from_generic(sc: rustc_codegen_ssa::common::SynchronizationScope) -> Self { - match sc { - rustc_codegen_ssa::common::SynchronizationScope::SingleThread => { - SynchronizationScope::SingleThread - } - rustc_codegen_ssa::common::SynchronizationScope::CrossThread => { - SynchronizationScope::CrossThread - } - } - } -} - /// LLVMRustFileType #[derive(Copy, Clone)] #[repr(C)] @@ -567,16 +550,6 @@ pub enum ArchiveKind { K_COFF, } -/// LLVMRustPassKind -#[derive(Copy, Clone, PartialEq, Debug)] -#[repr(C)] -#[allow(dead_code)] // Variants constructed by C++. -pub enum PassKind { - Other, - Function, - Module, -} - // LLVMRustThinLTOData extern "C" { pub type ThinLTOData; @@ -587,10 +560,6 @@ extern "C" { pub type ThinLTOBuffer; } -// LLVMRustModuleNameCallback -pub type ThinLTOModuleNameCallback = - unsafe extern "C" fn(*mut c_void, *const c_char, *const c_char); - /// LLVMRustThinLTOModule #[repr(C)] pub struct ThinLTOModule { @@ -620,6 +589,15 @@ pub enum ChecksumKind { SHA256, } +/// LLVMRustMemoryEffects +#[derive(Copy, Clone)] +#[repr(C)] +pub enum MemoryEffects { + None, + ReadOnly, + InaccessibleMemOnly, +} + extern "C" { type Opaque; } @@ -656,9 +634,6 @@ extern "C" { } #[repr(C)] pub struct Builder<'a>(InvariantOpaque<'a>); -extern "C" { - pub type MemoryBuffer; -} #[repr(C)] pub struct PassManager<'a>(InvariantOpaque<'a>); extern "C" { @@ -984,6 +959,22 @@ pub mod debuginfo { } } +use bitflags::bitflags; +// These values **must** match with LLVMRustAllocKindFlags +bitflags! { + #[repr(transparent)] + #[derive(Default)] + pub struct AllocKindFlags : u64 { + const Unknown = 0; + const Alloc = 1; + const Realloc = 1 << 1; + const Free = 1 << 2; + const Uninitialized = 1 << 3; + const Zeroed = 1 << 4; + const Aligned = 1 << 5; + } +} + extern "C" { pub type ModuleBuffer; } @@ -992,6 +983,9 @@ pub type SelfProfileBeforePassCallback = unsafe extern "C" fn(*mut c_void, *const c_char, *const c_char); pub type SelfProfileAfterPassCallback = unsafe extern "C" fn(*mut c_void); +pub type GetSymbolsCallback = unsafe extern "C" fn(*mut c_void, *const c_char) -> *mut c_void; +pub type GetSymbolsErrorCallback = unsafe extern "C" fn(*const c_char) -> *mut c_void; + extern "C" { pub fn LLVMRustInstallFatalErrorHandler(); pub fn LLVMRustDisableSystemDialogsOnCrash(); @@ -1011,7 +1005,6 @@ extern "C" { pub fn LLVMSetDataLayout(M: &Module, Triple: *const c_char); /// See Module::setModuleInlineAsm. - pub fn LLVMSetModuleInlineAsm2(M: &Module, Asm: *const c_char, AsmLen: size_t); pub fn LLVMRustAppendModuleInlineAsm(M: &Module, Asm: *const c_char, AsmLen: size_t); /// See llvm::LLVMTypeKind::getTypeID. @@ -1093,7 +1086,7 @@ extern "C" { pub fn LLVMConstInt(IntTy: &Type, N: c_ulonglong, SignExtend: Bool) -> &Value; pub fn LLVMConstIntOfArbitraryPrecision(IntTy: &Type, Wn: c_uint, Ws: *const u64) -> &Value; pub fn LLVMConstReal(RealTy: &Type, N: f64) -> &Value; - pub fn LLVMConstIntGetZExtValue(ConstantVal: &ConstantInt) -> c_ulonglong; + pub fn LLVMRustConstIntGetZExtValue(ConstantVal: &ConstantInt, Value: &mut u64) -> bool; pub fn LLVMRustConstInt128Get( ConstantVal: &ConstantInt, SExt: bool, @@ -1134,11 +1127,7 @@ extern "C" { pub fn LLVMConstIntToPtr<'a>(ConstantVal: &'a Value, ToType: &'a Type) -> &'a Value; pub fn LLVMConstBitCast<'a>(ConstantVal: &'a Value, ToType: &'a Type) -> &'a Value; pub fn LLVMConstPointerCast<'a>(ConstantVal: &'a Value, ToType: &'a Type) -> &'a Value; - pub fn LLVMConstExtractValue( - AggConstant: &Value, - IdxList: *const c_uint, - NumIdx: c_uint, - ) -> &Value; + pub fn LLVMGetAggregateElement(ConstantVal: &Value, Idx: c_uint) -> Option<&Value>; // Operations on global variables, functions, and aliases (globals) pub fn LLVMIsDeclaration(Global: &Value) -> Bool; @@ -1169,7 +1158,6 @@ extern "C" { pub fn LLVMGetInitializer(GlobalVar: &Value) -> Option<&Value>; pub fn LLVMSetInitializer<'a>(GlobalVar: &'a Value, ConstantVal: &'a Value); pub fn LLVMIsThreadLocal(GlobalVar: &Value) -> Bool; - pub fn LLVMSetThreadLocal(GlobalVar: &Value, IsThreadLocal: Bool); pub fn LLVMSetThreadLocalMode(GlobalVar: &Value, Mode: ThreadLocalMode); pub fn LLVMIsGlobalConstant(GlobalVar: &Value) -> Bool; pub fn LLVMSetGlobalConstant(GlobalVar: &Value, IsConstant: Bool); @@ -1194,7 +1182,11 @@ extern "C" { pub fn LLVMRustCreateDereferenceableOrNullAttr(C: &Context, bytes: u64) -> &Attribute; pub fn LLVMRustCreateByValAttr<'a>(C: &'a Context, ty: &'a Type) -> &'a Attribute; pub fn LLVMRustCreateStructRetAttr<'a>(C: &'a Context, ty: &'a Type) -> &'a Attribute; + pub fn LLVMRustCreateElementTypeAttr<'a>(C: &'a Context, ty: &'a Type) -> &'a Attribute; pub fn LLVMRustCreateUWTableAttr(C: &Context, async_: bool) -> &Attribute; + pub fn LLVMRustCreateAllocSizeAttr(C: &Context, size_arg: u32) -> &Attribute; + pub fn LLVMRustCreateAllocKindAttr(C: &Context, size_arg: u64) -> &Attribute; + pub fn LLVMRustCreateMemoryEffectsAttr(C: &Context, effects: MemoryEffects) -> &Attribute; // Operations on functions pub fn LLVMRustGetOrInsertFunction<'a>( @@ -1623,7 +1615,7 @@ extern "C" { B: &Builder<'a>, Val: &'a Value, DestTy: &'a Type, - IsSized: bool, + IsSigned: bool, ) -> &'a Value; // Comparisons @@ -1781,16 +1773,18 @@ extern "C" { Order: AtomicOrdering, ) -> &'a Value; - pub fn LLVMRustBuildAtomicCmpXchg<'a>( + pub fn LLVMBuildAtomicCmpXchg<'a>( B: &Builder<'a>, LHS: &'a Value, CMP: &'a Value, RHS: &'a Value, Order: AtomicOrdering, FailureOrder: AtomicOrdering, - Weak: Bool, + SingleThreaded: Bool, ) -> &'a Value; + pub fn LLVMSetWeak(CmpXchgInst: &Value, IsWeak: Bool); + pub fn LLVMBuildAtomicRMW<'a>( B: &Builder<'a>, Op: AtomicRmwBinOp, @@ -1800,27 +1794,19 @@ extern "C" { SingleThreaded: Bool, ) -> &'a Value; - pub fn LLVMRustBuildAtomicFence( - B: &Builder<'_>, + pub fn LLVMBuildFence<'a>( + B: &Builder<'a>, Order: AtomicOrdering, - Scope: SynchronizationScope, - ); + SingleThreaded: Bool, + Name: *const c_char, + ) -> &'a Value; /// Writes a module to the specified path. Returns 0 on success. pub fn LLVMWriteBitcodeToFile(M: &Module, Path: *const c_char) -> c_int; - /// Creates a pass manager. + /// Creates a legacy pass manager -- only used for final codegen. pub fn LLVMCreatePassManager<'a>() -> &'a mut PassManager<'a>; - /// Creates a function-by-function pass manager - pub fn LLVMCreateFunctionPassManagerForModule(M: &Module) -> &mut PassManager<'_>; - - /// Disposes a pass manager. - pub fn LLVMDisposePassManager<'a>(PM: &'a mut PassManager<'a>); - - /// Runs a pass manager on a module. - pub fn LLVMRunPassManager<'a>(PM: &PassManager<'a>, M: &'a Module) -> Bool; - pub fn LLVMInitializePasses(); pub fn LLVMTimeTraceProfilerInitialize(); @@ -1831,32 +1817,6 @@ extern "C" { pub fn LLVMAddAnalysisPasses<'a>(T: &'a TargetMachine, PM: &PassManager<'a>); - pub fn LLVMRustPassManagerBuilderCreate() -> &'static mut PassManagerBuilder; - pub fn LLVMRustPassManagerBuilderDispose(PMB: &'static mut PassManagerBuilder); - pub fn LLVMRustPassManagerBuilderUseInlinerWithThreshold( - PMB: &PassManagerBuilder, - threshold: c_uint, - ); - pub fn LLVMRustPassManagerBuilderPopulateModulePassManager( - PMB: &PassManagerBuilder, - PM: &PassManager<'_>, - ); - - pub fn LLVMRustPassManagerBuilderPopulateFunctionPassManager( - PMB: &PassManagerBuilder, - PM: &PassManager<'_>, - ); - pub fn LLVMRustPassManagerBuilderPopulateLTOPassManager( - PMB: &PassManagerBuilder, - PM: &PassManager<'_>, - Internalize: Bool, - RunInliner: Bool, - ); - pub fn LLVMRustPassManagerBuilderPopulateThinLTOPassManager( - PMB: &PassManagerBuilder, - PM: &PassManager<'_>, - ); - pub fn LLVMGetHostCPUFeatures() -> *mut c_char; pub fn LLVMDisposeMessage(message: *mut c_char); @@ -2078,6 +2038,19 @@ extern "C" { Ty: &'a DIType, ) -> &'a DIType; + pub fn LLVMRustDIBuilderCreateStaticMemberType<'a>( + Builder: &DIBuilder<'a>, + Scope: &'a DIDescriptor, + Name: *const c_char, + NameLen: size_t, + File: &'a DIFile, + LineNo: c_uint, + Ty: &'a DIType, + Flags: DIFlags, + val: Option<&'a Value>, + AlignInBits: u32, + ) -> &'a DIDerivedType; + pub fn LLVMRustDIBuilderCreateLexicalBlock<'a>( Builder: &DIBuilder<'a>, Scope: &'a DIScope, @@ -2157,7 +2130,8 @@ extern "C" { Builder: &DIBuilder<'a>, Name: *const c_char, NameLen: size_t, - Value: i64, + Value: *const u64, + SizeInBits: c_uint, IsUnsigned: bool, ) -> &'a DIEnumerator; @@ -2240,6 +2214,7 @@ extern "C" { ) -> &'a DILocation; pub fn LLVMRustDIBuilderCreateOpDeref() -> u64; pub fn LLVMRustDIBuilderCreateOpPlusUconst() -> u64; + pub fn LLVMRustDIBuilderCreateOpLLVMFragment() -> u64; #[allow(improper_ctypes)] pub fn LLVMRustWriteTypeToString(Type: &Type, s: &RustString); @@ -2248,23 +2223,6 @@ extern "C" { pub fn LLVMIsAConstantInt(value_ref: &Value) -> Option<&ConstantInt>; - pub fn LLVMRustPassKind(Pass: &Pass) -> PassKind; - pub fn LLVMRustFindAndCreatePass(Pass: *const c_char) -> Option<&'static mut Pass>; - pub fn LLVMRustCreateAddressSanitizerFunctionPass(Recover: bool) -> &'static mut Pass; - pub fn LLVMRustCreateModuleAddressSanitizerPass(Recover: bool) -> &'static mut Pass; - pub fn LLVMRustCreateMemorySanitizerPass( - TrackOrigins: c_int, - Recover: bool, - ) -> &'static mut Pass; - pub fn LLVMRustCreateThreadSanitizerPass() -> &'static mut Pass; - pub fn LLVMRustCreateHWAddressSanitizerPass(Recover: bool) -> &'static mut Pass; - pub fn LLVMRustAddPass(PM: &PassManager<'_>, Pass: &'static mut Pass); - pub fn LLVMRustAddLastExtensionPasses( - PMB: &PassManagerBuilder, - Passes: *const &'static mut Pass, - NumPasses: size_t, - ); - pub fn LLVMRustHasFeature(T: &TargetMachine, s: *const c_char) -> bool; pub fn LLVMRustPrintTargetCPUs(T: &TargetMachine); @@ -2298,29 +2256,11 @@ extern "C" { SplitDwarfFile: *const c_char, ) -> Option<&'static mut TargetMachine>; pub fn LLVMRustDisposeTargetMachine(T: &'static mut TargetMachine); - pub fn LLVMRustAddBuilderLibraryInfo<'a>( - PMB: &'a PassManagerBuilder, - M: &'a Module, - DisableSimplifyLibCalls: bool, - ); - pub fn LLVMRustConfigurePassManagerBuilder( - PMB: &PassManagerBuilder, - OptLevel: CodeGenOptLevel, - MergeFunctions: bool, - SLPVectorize: bool, - LoopVectorize: bool, - PrepareForThinLTO: bool, - PGOGenPath: *const c_char, - PGOUsePath: *const c_char, - PGOSampleUsePath: *const c_char, - SizeLevel: c_int, - ); pub fn LLVMRustAddLibraryInfo<'a>( PM: &PassManager<'a>, M: &'a Module, DisableSimplifyLibCalls: bool, ); - pub fn LLVMRustRunFunctionPassManager<'a>(PM: &PassManager<'a>, M: &'a Module); pub fn LLVMRustWriteOutputFile<'a>( T: &'a TargetMachine, PM: &PassManager<'a>, @@ -2329,7 +2269,7 @@ extern "C" { DwoOutput: *const c_char, FileType: FileType, ) -> LLVMRustResult; - pub fn LLVMRustOptimizeWithNewPassManager<'a>( + pub fn LLVMRustOptimize<'a>( M: &'a Module, TM: &'a TargetMachine, OptLevel: PassBuilderOptLevel, @@ -2347,6 +2287,7 @@ extern "C" { PGOGenPath: *const c_char, PGOUsePath: *const c_char, InstrumentCoverage: bool, + InstrProfileOutput: *const c_char, InstrumentGCOV: bool, PGOSampleUsePath: *const c_char, DebugInfoForProfiling: bool, @@ -2365,9 +2306,7 @@ extern "C" { ) -> LLVMRustResult; pub fn LLVMRustSetLLVMOptions(Argc: c_int, Argv: *const *const c_char); pub fn LLVMRustPrintPasses(); - pub fn LLVMRustGetInstructionCount(M: &Module) -> u32; pub fn LLVMRustSetNormalizedTarget(M: &Module, triple: *const c_char); - pub fn LLVMRustAddAlwaysInlinePass(P: &PassManagerBuilder, AddLifetimes: bool); pub fn LLVMRustRunRestrictionPass(M: &Module, syms: *const *const c_char, len: size_t); pub fn LLVMRustOpenArchive(path: *const c_char) -> Option<&'static mut Archive>; @@ -2376,7 +2315,6 @@ extern "C" { AIR: &ArchiveIterator<'a>, ) -> Option<&'a mut ArchiveChild<'a>>; pub fn LLVMRustArchiveChildName(ACR: &ArchiveChild<'_>, size: &mut size_t) -> *const c_char; - pub fn LLVMRustArchiveChildData(ACR: &ArchiveChild<'_>, size: &mut size_t) -> *const c_char; pub fn LLVMRustArchiveChildFree<'a>(ACR: &'a mut ArchiveChild<'a>); pub fn LLVMRustArchiveIteratorFree<'a>(AIR: &'a mut ArchiveIterator<'a>); pub fn LLVMRustDestroyArchive(AR: &'static mut Archive); @@ -2411,12 +2349,6 @@ extern "C" { cookie_out: &mut c_uint, ) -> &'a SMDiagnostic; - pub fn LLVMRustSetInlineAsmDiagnosticHandler( - C: &Context, - H: InlineAsmDiagHandlerTy, - CX: *mut c_void, - ); - #[allow(improper_ctypes)] pub fn LLVMRustUnpackSMDiagnostic( d: &SMDiagnostic, @@ -2463,7 +2395,6 @@ extern "C" { pub fn LLVMRustPositionBuilderAtStart<'a>(B: &Builder<'a>, BB: &'a BasicBlock); pub fn LLVMRustSetComdat<'a>(M: &'a Module, V: &'a Value, Name: *const c_char, NameLen: size_t); - pub fn LLVMRustUnsetComdat(V: &Value); pub fn LLVMRustSetModulePICLevel(M: &Module); pub fn LLVMRustSetModulePIELevel(M: &Module); pub fn LLVMRustSetModuleCodeModel(M: &Module, Model: CodeModel); @@ -2473,7 +2404,7 @@ extern "C" { pub fn LLVMRustModuleBufferFree(p: &'static mut ModuleBuffer); pub fn LLVMRustModuleCost(M: &Module) -> u64; - pub fn LLVMRustThinLTOBufferCreate(M: &Module) -> &'static mut ThinLTOBuffer; + pub fn LLVMRustThinLTOBufferCreate(M: &Module, is_thin: bool) -> &'static mut ThinLTOBuffer; pub fn LLVMRustThinLTOBufferFree(M: &'static mut ThinLTOBuffer); pub fn LLVMRustThinLTOBufferPtr(M: &ThinLTOBuffer) -> *const c_char; pub fn LLVMRustThinLTOBufferLen(M: &ThinLTOBuffer) -> size_t; @@ -2495,11 +2426,6 @@ extern "C" { Module: &Module, Target: &TargetMachine, ) -> bool; - pub fn LLVMRustGetThinLTOModuleImports( - Data: *const ThinLTOData, - ModuleNameCallback: ThinLTOModuleNameCallback, - CallbackPayload: *mut c_void, - ); pub fn LLVMRustFreeThinLTOData(Data: &'static mut ThinLTOData); pub fn LLVMRustParseBitcodeForLTO( Context: &Context, @@ -2512,8 +2438,12 @@ extern "C" { len: usize, out_len: &mut usize, ) -> *const u8; - pub fn LLVMRustLTOGetDICompileUnit(M: &Module, CU1: &mut *mut c_void, CU2: &mut *mut c_void); - pub fn LLVMRustLTOPatchDICompileUnit(M: &Module, CU: *mut c_void); + pub fn LLVMRustThinLTOGetDICompileUnit( + M: &Module, + CU1: &mut *mut c_void, + CU2: &mut *mut c_void, + ); + pub fn LLVMRustThinLTOPatchDICompileUnit(M: &Module, CU: *mut c_void); pub fn LLVMRustLinkerNew(M: &Module) -> &mut Linker<'_>; pub fn LLVMRustLinkerAdd( @@ -2545,4 +2475,16 @@ extern "C" { #[allow(improper_ctypes)] pub fn LLVMRustGetMangledName(V: &Value, out: &RustString); + + pub fn LLVMRustGetElementTypeArgIndex(CallSite: &Value) -> i32; + + pub fn LLVMRustIsBitcode(ptr: *const u8, len: usize) -> bool; + + pub fn LLVMRustGetSymbols( + buf_ptr: *const u8, + buf_len: usize, + state: *mut c_void, + callback: GetSymbolsCallback, + error_callback: GetSymbolsErrorCallback, + ) -> *mut c_void; } diff --git a/compiler/rustc_codegen_llvm/src/llvm/mod.rs b/compiler/rustc_codegen_llvm/src/llvm/mod.rs index 48fbc1de8ee..f820e752371 100644 --- a/compiler/rustc_codegen_llvm/src/llvm/mod.rs +++ b/compiler/rustc_codegen_llvm/src/llvm/mod.rs @@ -95,6 +95,14 @@ pub fn CreateUWTableAttr(llcx: &Context, async_: bool) -> &Attribute { unsafe { LLVMRustCreateUWTableAttr(llcx, async_) } } +pub fn CreateAllocSizeAttr(llcx: &Context, size_arg: u32) -> &Attribute { + unsafe { LLVMRustCreateAllocSizeAttr(llcx, size_arg) } +} + +pub fn CreateAllocKindAttr(llcx: &Context, kind_arg: AllocKindFlags) -> &Attribute { + unsafe { LLVMRustCreateAllocKindAttr(llcx, kind_arg.bits()) } +} + #[derive(Copy, Clone)] pub enum AttributePlace { ReturnValue, @@ -158,12 +166,6 @@ pub fn SetUniqueComdat(llmod: &Module, val: &Value) { } } -pub fn UnsetComdat(val: &Value) { - unsafe { - LLVMRustUnsetComdat(val); - } -} - pub fn SetUnnamedAddress(global: &Value, unnamed: UnnamedAddr) { unsafe { LLVMSetUnnamedAddress(global, unnamed); @@ -183,6 +185,13 @@ impl AttributeKind { } } +impl MemoryEffects { + /// Create an LLVM Attribute with these memory effects. + pub fn create_attr(self, llcx: &Context) -> &Attribute { + unsafe { LLVMRustCreateMemoryEffectsAttr(llcx, self) } + } +} + pub fn set_section(llglobal: &Value, section_name: &str) { let section_name_cstr = CString::new(section_name).expect("unexpected CString error"); unsafe { diff --git a/compiler/rustc_codegen_llvm/src/llvm_util.rs b/compiler/rustc_codegen_llvm/src/llvm_util.rs index ce6c6e3215c..2fa602520dc 100644 --- a/compiler/rustc_codegen_llvm/src/llvm_util.rs +++ b/compiler/rustc_codegen_llvm/src/llvm_util.rs @@ -1,7 +1,10 @@ use crate::back::write::create_informational_target_machine; -use crate::{llvm, llvm_util}; +use crate::errors::{ + PossibleFeature, TargetFeatureDisableOrEnable, UnknownCTargetFeature, + UnknownCTargetFeaturePrefix, +}; +use crate::llvm; use libc::c_int; -use libloading::Library; use rustc_codegen_ssa::target_features::{ supported_target_features, tied_target_features, RUSTC_SPECIFIC_FEATURES, }; @@ -15,9 +18,7 @@ use rustc_span::symbol::Symbol; use rustc_target::spec::{MergeFunctions, PanicStrategy}; use smallvec::{smallvec, SmallVec}; use std::ffi::{CStr, CString}; -use tracing::debug; -use std::mem; use std::path::Path; use std::ptr; use std::slice; @@ -87,22 +88,12 @@ unsafe fn configure_llvm(sess: &Session) { add("-debug-pass=Structure", false); } if sess.target.generate_arange_section - && !sess.opts.debugging_opts.no_generate_arange_section + && !sess.opts.unstable_opts.no_generate_arange_section { add("-generate-arange-section", false); } - // Disable the machine outliner by default in LLVM versions 11 and LLVM - // version 12, where it leads to miscompilation. - // - // Ref: - // - https://github.com/rust-lang/rust/issues/85351 - // - https://reviews.llvm.org/D103167 - if llvm_util::get_version() < (13, 0, 0) { - add("-enable-machine-outliner=never", false); - } - - match sess.opts.debugging_opts.merge_functions.unwrap_or(sess.target.merge_functions) { + match sess.opts.unstable_opts.merge_functions.unwrap_or(sess.target.merge_functions) { MergeFunctions::Disabled | MergeFunctions::Trampolines => {} MergeFunctions::Aliases => { add("-mergefunc-use-aliases", false); @@ -125,28 +116,12 @@ unsafe fn configure_llvm(sess: &Session) { } } - if sess.opts.debugging_opts.llvm_time_trace { + if sess.opts.unstable_opts.llvm_time_trace { llvm::LLVMTimeTraceProfilerInitialize(); } llvm::LLVMInitializePasses(); - // Use the legacy plugin registration if we don't use the new pass manager - if !should_use_new_llvm_pass_manager( - &sess.opts.debugging_opts.new_llvm_pass_manager, - &sess.target.arch, - ) { - // Register LLVM plugins by loading them into the compiler process. - for plugin in &sess.opts.debugging_opts.llvm_plugins { - let lib = Library::new(plugin).unwrap_or_else(|e| bug!("couldn't load plugin: {}", e)); - debug!("LLVM plugin loaded successfully {:?} ({})", lib, plugin); - - // Intentionally leak the dynamic library. We can't ever unload it - // since the library can make things that will live arbitrarily long. - mem::forget(lib); - } - } - rustc_llvm::initialize_available_targets(); llvm::LLVMRustSetLLVMOptions(llvm_args.len() as c_int, llvm_args.as_ptr()); @@ -165,6 +140,10 @@ pub fn time_trace_profiler_finish(file_name: &Path) { // // To find a list of LLVM's names, check llvm-project/llvm/include/llvm/Support/*TargetParser.def // where the * matches the architecture's name +// +// For targets not present in the above location, see llvm-project/llvm/lib/Target/{ARCH}/*.td +// where `{ARCH}` is the architecture name. Look for instances of `SubtargetFeature`. +// // Beware to not use the llvm github project for this, but check the git submodule // found in src/llvm-project // Though note that Rust can also be build with an external precompiled version of LLVM @@ -184,6 +163,9 @@ pub fn to_llvm_features<'a>(sess: &Session, s: &'a str) -> SmallVec<[&'a str; 2] ("x86", "rdrand") => smallvec!["rdrnd"], ("x86", "bmi1") => smallvec!["bmi"], ("x86", "cmpxchg16b") => smallvec!["cx16"], + // FIXME: These aliases are misleading, and should be removed before avx512_target_feature is + // stabilized. They must remain until std::arch switches off them. + // rust#100752 ("x86", "avx512vaes") => smallvec!["vaes"], ("x86", "avx512gfni") => smallvec!["gfni"], ("x86", "avx512vpclmulqdq") => smallvec!["vpclmulqdq"], @@ -212,8 +194,8 @@ pub fn to_llvm_features<'a>(sess: &Session, s: &'a str) -> SmallVec<[&'a str; 2] } } -// Given a map from target_features to whether they are enabled or disabled, -// ensure only valid combinations are allowed. +/// 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>, @@ -231,28 +213,31 @@ pub fn check_tied_features( return None; } -// Used to generate cfg variables and apply features -// Must express features in the way Rust understands them -pub fn target_features(sess: &Session) -> Vec<Symbol> { +/// Used to generate cfg variables and apply features +/// Must express features in the way Rust understands them +pub fn target_features(sess: &Session, allow_unstable: bool) -> Vec<Symbol> { let target_machine = create_informational_target_machine(sess); - let mut features: Vec<Symbol> = - supported_target_features(sess) - .iter() - .filter_map(|&(feature, gate)| { - if sess.is_nightly_build() || gate.is_none() { Some(feature) } else { None } - }) - .filter(|feature| { - // check that all features in a given smallvec are enabled - for llvm_feature in to_llvm_features(sess, feature) { - let cstr = SmallCStr::new(llvm_feature); - if !unsafe { llvm::LLVMRustHasFeature(target_machine, cstr.as_ptr()) } { - return false; - } + let mut features: Vec<Symbol> = supported_target_features(sess) + .iter() + .filter_map(|&(feature, gate)| { + if sess.is_nightly_build() || allow_unstable || gate.is_none() { + Some(feature) + } else { + None + } + }) + .filter(|feature| { + // check that all features in a given smallvec are enabled + for llvm_feature in to_llvm_features(sess, feature) { + let cstr = SmallCStr::new(llvm_feature); + if !unsafe { llvm::LLVMRustHasFeature(target_machine, cstr.as_ptr()) } { + return false; } - true - }) - .map(|feature| Symbol::intern(feature)) - .collect(); + } + true + }) + .map(|feature| Symbol::intern(feature)) + .collect(); // LLVM 14 changed the ABI for i128 arguments to __float/__fix builtins on Win64 // (see https://reviews.llvm.org/D110413). This unstable target feature is intended for use @@ -307,30 +292,33 @@ fn llvm_target_features(tm: &llvm::TargetMachine) -> Vec<(&str, &str)> { } fn print_target_features(sess: &Session, tm: &llvm::TargetMachine) { - let mut target_features = llvm_target_features(tm); + let mut llvm_target_features = llvm_target_features(tm); + let mut known_llvm_target_features = FxHashSet::<&'static str>::default(); let mut rustc_target_features = supported_target_features(sess) .iter() - .filter_map(|(feature, _gate)| { - for llvm_feature in to_llvm_features(sess, *feature) { + .map(|(feature, _gate)| { + let desc = if let Some(llvm_feature) = to_llvm_features(sess, *feature).first() { // LLVM asserts that these are sorted. LLVM and Rust both use byte comparison for these strings. - match target_features.binary_search_by_key(&llvm_feature, |(f, _d)| f).ok().map( - |index| { - let (_f, desc) = target_features.remove(index); - (*feature, desc) - }, - ) { - Some(v) => return Some(v), - None => {} + match llvm_target_features.binary_search_by_key(&llvm_feature, |(f, _d)| f).ok() { + Some(index) => { + known_llvm_target_features.insert(llvm_feature); + llvm_target_features[index].1 + } + None => "", } - } - None + } else { + "" + }; + (*feature, desc) }) .collect::<Vec<_>>(); rustc_target_features.extend_from_slice(&[( "crt-static", "Enables C Run-time Libraries to be statically linked", )]); - let max_feature_len = target_features + llvm_target_features.retain(|(f, _d)| !known_llvm_target_features.contains(f)); + + let max_feature_len = llvm_target_features .iter() .chain(rustc_target_features.iter()) .map(|(feature, _desc)| feature.len()) @@ -342,10 +330,10 @@ fn print_target_features(sess: &Session, tm: &llvm::TargetMachine) { println!(" {1:0$} - {2}.", max_feature_len, feature, desc); } println!("\nCode-generation features supported by LLVM for this target:"); - for (feature, desc) in &target_features { + for (feature, desc) in &llvm_target_features { println!(" {1:0$} - {2}.", max_feature_len, feature, desc); } - if target_features.is_empty() { + if llvm_target_features.is_empty() { println!(" Target features listing is not supported by this LLVM version."); } println!("\nUse +feature to enable a feature, or -feature to disable it."); @@ -437,6 +425,8 @@ pub(crate) fn global_llvm_features(sess: &Session, diagnostics: bool) -> Vec<Str .features .split(',') .filter(|v| !v.is_empty() && backend_feature_name(v).is_some()) + // Drop +atomics-32 feature introduced in LLVM 15. + .filter(|v| *v != "+atomics-32" || get_version() >= (15, 0, 0)) .map(String::from), ); @@ -454,12 +444,7 @@ pub(crate) fn global_llvm_features(sess: &Session, diagnostics: bool) -> Vec<Str Some(c @ '+' | c @ '-') => c, Some(_) => { if diagnostics { - let mut diag = sess.struct_warn(&format!( - "unknown feature specified for `-Ctarget-feature`: `{}`", - s - )); - diag.note("features must begin with a `+` to enable or `-` to disable it"); - diag.emit(); + sess.emit_warning(UnknownCTargetFeaturePrefix { feature: s }); } return None; } @@ -476,17 +461,15 @@ pub(crate) fn global_llvm_features(sess: &Session, diagnostics: bool) -> Vec<Str None } }); - let mut diag = sess.struct_warn(&format!( - "unknown feature specified for `-Ctarget-feature`: `{}`", - feature - )); - diag.note("it is still passed through to the codegen backend"); - if let Some(rust_feature) = rust_feature { - diag.help(&format!("you might have meant: `{}`", rust_feature)); + let unknown_feature = if let Some(rust_feature) = rust_feature { + UnknownCTargetFeature { + feature, + rust_feature: PossibleFeature::Some { rust_feature }, + } } else { - diag.note("consider filing a feature request"); - } - diag.emit(); + UnknownCTargetFeature { feature, rust_feature: PossibleFeature::None } + }; + sess.emit_warning(unknown_feature); } if diagnostics { @@ -511,11 +494,17 @@ pub(crate) fn global_llvm_features(sess: &Session, diagnostics: bool) -> Vec<Str .flatten(); features.extend(feats); + // FIXME: Move v8a to target definition list when earliest supported LLVM is 14. + if get_version() >= (14, 0, 0) && sess.target.arch == "aarch64" { + features.push("+v8a".into()); + } + if diagnostics && let Some(f) = check_tied_features(sess, &featsmap) { - sess.err(&format!( - "target features {} must all be enabled or disabled together", - f.join(", ") - )); + sess.emit_err(TargetFeatureDisableOrEnable { + features: f, + span: None, + missing_features: None, + }); } features @@ -538,22 +527,6 @@ fn backend_feature_name(s: &str) -> Option<&str> { } pub fn tune_cpu(sess: &Session) -> Option<&str> { - let name = sess.opts.debugging_opts.tune_cpu.as_ref()?; + let name = sess.opts.unstable_opts.tune_cpu.as_ref()?; Some(handle_native(name)) } - -pub(crate) fn should_use_new_llvm_pass_manager(user_opt: &Option<bool>, target_arch: &str) -> bool { - // The new pass manager is enabled by default for LLVM >= 13. - // This matches Clang, which also enables it since Clang 13. - - // Since LLVM 15, the legacy pass manager is no longer supported. - if llvm_util::get_version() >= (15, 0, 0) { - return true; - } - - // There are some perf issues with the new pass manager when targeting - // s390x with LLVM 13, so enable the new pass manager only with LLVM 14. - // See https://github.com/rust-lang/rust/issues/89609. - let min_version = if target_arch == "s390x" { 14 } else { 13 }; - user_opt.unwrap_or_else(|| llvm_util::get_version() >= (min_version, 0, 0)) -} diff --git a/compiler/rustc_codegen_llvm/src/mono_item.rs b/compiler/rustc_codegen_llvm/src/mono_item.rs index a3053742aad..76f692b2016 100644 --- a/compiler/rustc_codegen_llvm/src/mono_item.rs +++ b/compiler/rustc_codegen_llvm/src/mono_item.rs @@ -1,6 +1,7 @@ use crate::attributes; use crate::base; use crate::context::CodegenCx; +use crate::errors::SymbolAlreadyDefined; use crate::llvm; use crate::type_of::LayoutLlvmExt; use rustc_codegen_ssa::traits::*; @@ -8,10 +9,9 @@ use rustc_hir::def_id::{DefId, LOCAL_CRATE}; pub use rustc_middle::mir::mono::MonoItem; use rustc_middle::mir::mono::{Linkage, Visibility}; use rustc_middle::ty::layout::{FnAbiOf, LayoutOf}; -use rustc_middle::ty::{self, Instance, TypeFoldable}; +use rustc_middle::ty::{self, Instance, TypeVisitable}; use rustc_session::config::CrateType; use rustc_target::spec::RelocModel; -use tracing::debug; impl<'tcx> PreDefineMethods<'tcx> for CodegenCx<'_, 'tcx> { fn predefine_static( @@ -26,10 +26,8 @@ impl<'tcx> PreDefineMethods<'tcx> for CodegenCx<'_, 'tcx> { let llty = self.layout_of(ty).llvm_type(self); let g = self.define_global(symbol_name, llty).unwrap_or_else(|| { - self.sess().span_fatal( - self.tcx.def_span(def_id), - &format!("symbol `{}` is already defined", symbol_name), - ) + self.sess() + .emit_fatal(SymbolAlreadyDefined { span: self.tcx.def_span(def_id), symbol_name }) }); unsafe { diff --git a/compiler/rustc_codegen_llvm/src/type_.rs b/compiler/rustc_codegen_llvm/src/type_.rs index cf2d3c423c3..5772b7e1d81 100644 --- a/compiler/rustc_codegen_llvm/src/type_.rs +++ b/compiler/rustc_codegen_llvm/src/type_.rs @@ -19,7 +19,7 @@ use rustc_target::abi::{AddressSpace, Align, Integer, Size}; use std::fmt; use std::ptr; -use libc::c_uint; +use libc::{c_char, c_uint}; impl PartialEq for Type { fn eq(&self, other: &Self) -> bool { @@ -127,10 +127,6 @@ impl<'ll> CodegenCx<'ll, '_> { pub(crate) fn type_variadic_func(&self, args: &[&'ll Type], ret: &'ll Type) -> &'ll Type { unsafe { llvm::LLVMFunctionType(ret, args.as_ptr(), args.len() as c_uint, True) } } - - pub(crate) fn type_array(&self, ty: &'ll Type, len: u64) -> &'ll Type { - unsafe { llvm::LLVMRustArrayType(ty, len) } - } } impl<'ll, 'tcx> BaseTypeMethods<'tcx> for CodegenCx<'ll, 'tcx> { @@ -231,6 +227,10 @@ impl<'ll, 'tcx> BaseTypeMethods<'tcx> for CodegenCx<'ll, 'tcx> { fn val_ty(&self, v: &'ll Value) -> &'ll Type { common::val_ty(v) } + + fn type_array(&self, ty: &'ll Type, len: u64) -> &'ll Type { + unsafe { llvm::LLVMRustArrayType(ty, len) } + } } impl Type { @@ -238,7 +238,7 @@ impl Type { unsafe { llvm::LLVMInt8TypeInContext(llcx) } } - // Creates an integer type with the given number of bits, e.g., i24 + /// Creates an integer type with the given number of bits, e.g., i24 pub fn ix_llcx(llcx: &llvm::Context, num_bits: u64) -> &Type { unsafe { llvm::LLVMIntTypeInContext(llcx, num_bits as c_uint) } } @@ -289,3 +289,31 @@ impl<'ll, 'tcx> LayoutTypeMethods<'tcx> for CodegenCx<'ll, 'tcx> { ty.llvm_type(self) } } + +impl<'ll, 'tcx> TypeMembershipMethods<'tcx> for CodegenCx<'ll, 'tcx> { + fn set_type_metadata(&self, function: &'ll Value, typeid: String) { + let typeid_metadata = self.typeid_metadata(typeid); + let v = [self.const_usize(0), typeid_metadata]; + unsafe { + llvm::LLVMGlobalSetMetadata( + function, + llvm::MD_type as c_uint, + llvm::LLVMValueAsMetadata(llvm::LLVMMDNodeInContext( + self.llcx, + v.as_ptr(), + v.len() as c_uint, + )), + ) + } + } + + fn typeid_metadata(&self, typeid: String) -> &'ll Value { + unsafe { + llvm::LLVMMDStringInContext( + self.llcx, + typeid.as_ptr() as *const c_char, + typeid.len() as c_uint, + ) + } + } +} diff --git a/compiler/rustc_codegen_llvm/src/type_of.rs b/compiler/rustc_codegen_llvm/src/type_of.rs index 86280523631..182adf81785 100644 --- a/compiler/rustc_codegen_llvm/src/type_of.rs +++ b/compiler/rustc_codegen_llvm/src/type_of.rs @@ -6,12 +6,11 @@ use rustc_codegen_ssa::traits::*; use rustc_middle::bug; use rustc_middle::ty::layout::{FnAbiOf, LayoutOf, TyAndLayout}; use rustc_middle::ty::print::{with_no_trimmed_paths, with_no_visible_paths}; -use rustc_middle::ty::{self, Ty, TypeFoldable}; +use rustc_middle::ty::{self, Ty, TypeVisitable}; use rustc_target::abi::{Abi, AddressSpace, Align, FieldsShape}; use rustc_target::abi::{Int, Pointer, F32, F64}; use rustc_target::abi::{PointeeInfo, Scalar, Size, TyAbiInterface, Variants}; use smallvec::{smallvec, SmallVec}; -use tracing::debug; use std::fmt::Write; @@ -141,7 +140,7 @@ fn struct_llfields<'a, 'tcx>( prev_effective_align = effective_field_align; } let padding_used = result.len() > field_count; - if !layout.is_unsized() && field_count > 0 { + if layout.is_sized() && field_count > 0 { if offset > layout.size { bug!("layout: {:#?} stride: {:?} offset: {:?}", layout, layout.size, offset); } |