2016-08-04 02:17:35 +08:00
|
|
|
; RUN: llc -verify-machineinstrs -mcpu=pwr7 -O3 < %s | FileCheck %s
|
This patch addresses bug 15031.
The common code in the post-RA scheduler to break anti-dependencies on the
critical path contained a flaw. In the reported case, an anti-dependency
between the overlapping registers %X4 and %R4 exists:
%X29<def> = OR8 %X4, %X4
%R4<def>, %X3<def,dead,tied3> = LBZU 1, %X3<kill,tied1>
The unpatched code breaks the dependency by replacing %R4 and its uses
with %R3, the first register on the available list. However, %R3 and
%X3 overlap, so this creates two overlapping definitions on the same
instruction.
The fix is straightforward, preventing selection of a register that
overlaps any other defined register on the same instruction.
The test case is reduced from the bug report, and verifies that we no
longer produce "lbzu 3, 1(3)" when breaking this anti-dependency.
llvm-svn: 173706
2013-01-29 02:36:58 +08:00
|
|
|
|
|
|
|
|
; Test case derived from bug report 15031. The code in the post-RA
|
|
|
|
|
; scheduler to break critical anti-dependencies was failing to check
|
|
|
|
|
; whether an instruction had more than one definition, and ensuring
|
|
|
|
|
; that any additional definitions interfered with the choice of a new
|
|
|
|
|
; register. As a result, this test originally caused this to be
|
|
|
|
|
; generated:
|
|
|
|
|
;
|
|
|
|
|
; lbzu 3, 1(3)
|
|
|
|
|
;
|
|
|
|
|
; which is illegal, since it requires register 3 to both receive the
|
|
|
|
|
; loaded value and receive the updated address. With the fix to bug
|
|
|
|
|
; 15031, a different register is chosen to receive the loaded value.
|
|
|
|
|
|
|
|
|
|
target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-f128:128:128-v128:128:128-n32:64"
|
|
|
|
|
target triple = "powerpc64-unknown-linux-gnu"
|
|
|
|
|
|
|
|
|
|
%"class.llvm::MachineMemOperand" = type { %"struct.llvm::MachinePointerInfo", i64, i32, %"class.llvm::MDNode"*, %"class.llvm::MDNode"* }
|
|
|
|
|
%"struct.llvm::MachinePointerInfo" = type { %"class.llvm::Value"*, i64 }
|
|
|
|
|
%"class.llvm::Value" = type { i32 (...)**, i8, i8, i16, %"class.llvm::Type"*, %"class.llvm::Use"*, %"class.llvm::StringMapEntry"* }
|
|
|
|
|
%"class.llvm::Type" = type { %"class.llvm::LLVMContext"*, i32, i32, %"class.llvm::Type"** }
|
|
|
|
|
%"class.llvm::LLVMContext" = type { %"class.llvm::LLVMContextImpl"* }
|
|
|
|
|
%"class.llvm::LLVMContextImpl" = type opaque
|
|
|
|
|
%"class.llvm::Use" = type { %"class.llvm::Value"*, %"class.llvm::Use"*, %"class.llvm::PointerIntPair" }
|
|
|
|
|
%"class.llvm::PointerIntPair" = type { i64 }
|
|
|
|
|
%"class.llvm::StringMapEntry" = type opaque
|
|
|
|
|
%"class.llvm::MDNode" = type { %"class.llvm::Value", %"class.llvm::FoldingSetImpl::Node", i32, i32 }
|
|
|
|
|
%"class.llvm::FoldingSetImpl::Node" = type { i8* }
|
|
|
|
|
%"class.llvm::MachineInstr" = type { %"class.llvm::ilist_node", %"class.llvm::MCInstrDesc"*, %"class.llvm::MachineBasicBlock"*, %"class.llvm::MachineOperand"*, i32, %"class.llvm::ArrayRecycler<llvm::MachineOperand, 8>::Capacity", i8, i8, i8, %"class.llvm::MachineMemOperand"**, %"class.llvm::DebugLoc" }
|
|
|
|
|
%"class.llvm::ilist_node" = type { %"class.llvm::ilist_half_node", %"class.llvm::MachineInstr"* }
|
|
|
|
|
%"class.llvm::ilist_half_node" = type { %"class.llvm::MachineInstr"* }
|
|
|
|
|
%"class.llvm::MCInstrDesc" = type { i16, i16, i16, i16, i16, i32, i64, i16*, i16*, %"class.llvm::MCOperandInfo"* }
|
|
|
|
|
%"class.llvm::MCOperandInfo" = type { i16, i8, i8, i32 }
|
|
|
|
|
%"class.llvm::MachineBasicBlock" = type { %"class.llvm::ilist_node.0", %"struct.llvm::ilist", %"class.llvm::BasicBlock"*, i32, %"class.llvm::MachineFunction"*, %"class.std::vector.163", %"class.std::vector.163", %"class.std::vector.123", %"class.std::vector.123", i32, i8, i8 }
|
|
|
|
|
%"class.llvm::ilist_node.0" = type { %"class.llvm::ilist_half_node.1", %"class.llvm::MachineBasicBlock"* }
|
|
|
|
|
%"class.llvm::ilist_half_node.1" = type { %"class.llvm::MachineBasicBlock"* }
|
|
|
|
|
%"struct.llvm::ilist" = type { %"class.llvm::iplist" }
|
|
|
|
|
%"class.llvm::iplist" = type { %"struct.llvm::ilist_traits", %"class.llvm::MachineInstr"* }
|
|
|
|
|
%"struct.llvm::ilist_traits" = type { %"class.llvm::ilist_half_node", %"class.llvm::MachineBasicBlock"* }
|
|
|
|
|
%"class.llvm::BasicBlock" = type { %"class.llvm::Value", %"class.llvm::ilist_node.2", %"class.llvm::iplist.4", %"class.llvm::Function"* }
|
|
|
|
|
%"class.llvm::ilist_node.2" = type { %"class.llvm::ilist_half_node.3", %"class.llvm::BasicBlock"* }
|
|
|
|
|
%"class.llvm::ilist_half_node.3" = type { %"class.llvm::BasicBlock"* }
|
|
|
|
|
%"class.llvm::iplist.4" = type { %"struct.llvm::ilist_traits.5", %"class.llvm::Instruction"* }
|
|
|
|
|
%"struct.llvm::ilist_traits.5" = type { %"class.llvm::ilist_half_node.10" }
|
|
|
|
|
%"class.llvm::ilist_half_node.10" = type { %"class.llvm::Instruction"* }
|
|
|
|
|
%"class.llvm::Instruction" = type { %"class.llvm::User", %"class.llvm::ilist_node.193", %"class.llvm::BasicBlock"*, %"class.llvm::DebugLoc" }
|
|
|
|
|
%"class.llvm::User" = type { %"class.llvm::Value", %"class.llvm::Use"*, i32 }
|
|
|
|
|
%"class.llvm::ilist_node.193" = type { %"class.llvm::ilist_half_node.10", %"class.llvm::Instruction"* }
|
|
|
|
|
%"class.llvm::DebugLoc" = type { i32, i32 }
|
|
|
|
|
%"class.llvm::Function" = type { %"class.llvm::GlobalValue", %"class.llvm::ilist_node.27", %"class.llvm::iplist.47", %"class.llvm::iplist.54", %"class.llvm::ValueSymbolTable"*, %"class.llvm::AttributeSet" }
|
|
|
|
|
%"class.llvm::GlobalValue" = type { [52 x i8], [4 x i8], %"class.llvm::Module"*, %"class.std::basic_string" }
|
|
|
|
|
%"class.llvm::Module" = type { %"class.llvm::LLVMContext"*, %"class.llvm::iplist.11", %"class.llvm::iplist.20", %"class.llvm::iplist.29", %"struct.llvm::ilist.38", %"class.std::basic_string", %"class.llvm::ValueSymbolTable"*, %"class.llvm::OwningPtr", %"class.std::basic_string", %"class.std::basic_string", %"class.std::basic_string", i8* }
|
|
|
|
|
%"class.llvm::iplist.11" = type { %"struct.llvm::ilist_traits.12", %"class.llvm::GlobalVariable"* }
|
|
|
|
|
%"struct.llvm::ilist_traits.12" = type { %"class.llvm::ilist_node.18" }
|
|
|
|
|
%"class.llvm::ilist_node.18" = type { %"class.llvm::ilist_half_node.19", %"class.llvm::GlobalVariable"* }
|
|
|
|
|
%"class.llvm::ilist_half_node.19" = type { %"class.llvm::GlobalVariable"* }
|
|
|
|
|
%"class.llvm::GlobalVariable" = type { %"class.llvm::GlobalValue", %"class.llvm::ilist_node.18", i8 }
|
|
|
|
|
%"class.llvm::iplist.20" = type { %"struct.llvm::ilist_traits.21", %"class.llvm::Function"* }
|
|
|
|
|
%"struct.llvm::ilist_traits.21" = type { %"class.llvm::ilist_node.27" }
|
|
|
|
|
%"class.llvm::ilist_node.27" = type { %"class.llvm::ilist_half_node.28", %"class.llvm::Function"* }
|
|
|
|
|
%"class.llvm::ilist_half_node.28" = type { %"class.llvm::Function"* }
|
|
|
|
|
%"class.llvm::iplist.29" = type { %"struct.llvm::ilist_traits.30", %"class.llvm::GlobalAlias"* }
|
|
|
|
|
%"struct.llvm::ilist_traits.30" = type { %"class.llvm::ilist_node.36" }
|
|
|
|
|
%"class.llvm::ilist_node.36" = type { %"class.llvm::ilist_half_node.37", %"class.llvm::GlobalAlias"* }
|
|
|
|
|
%"class.llvm::ilist_half_node.37" = type { %"class.llvm::GlobalAlias"* }
|
|
|
|
|
%"class.llvm::GlobalAlias" = type { %"class.llvm::GlobalValue", %"class.llvm::ilist_node.36" }
|
|
|
|
|
%"struct.llvm::ilist.38" = type { %"class.llvm::iplist.39" }
|
|
|
|
|
%"class.llvm::iplist.39" = type { %"struct.llvm::ilist_traits.40", %"class.llvm::NamedMDNode"* }
|
|
|
|
|
%"struct.llvm::ilist_traits.40" = type { %"class.llvm::ilist_node.45" }
|
|
|
|
|
%"class.llvm::ilist_node.45" = type { %"class.llvm::ilist_half_node.46", %"class.llvm::NamedMDNode"* }
|
|
|
|
|
%"class.llvm::ilist_half_node.46" = type { %"class.llvm::NamedMDNode"* }
|
|
|
|
|
%"class.llvm::NamedMDNode" = type { %"class.llvm::ilist_node.45", %"class.std::basic_string", %"class.llvm::Module"*, i8* }
|
|
|
|
|
%"class.std::basic_string" = type { %"struct.std::basic_string<char, std::char_traits<char>, std::allocator<char> >::_Alloc_hider" }
|
|
|
|
|
%"struct.std::basic_string<char, std::char_traits<char>, std::allocator<char> >::_Alloc_hider" = type { i8* }
|
|
|
|
|
%"class.llvm::ValueSymbolTable" = type opaque
|
|
|
|
|
%"class.llvm::OwningPtr" = type { %"class.llvm::GVMaterializer"* }
|
|
|
|
|
%"class.llvm::GVMaterializer" = type opaque
|
|
|
|
|
%"class.llvm::iplist.47" = type { %"struct.llvm::ilist_traits.48", %"class.llvm::BasicBlock"* }
|
|
|
|
|
%"struct.llvm::ilist_traits.48" = type { %"class.llvm::ilist_half_node.3" }
|
|
|
|
|
%"class.llvm::iplist.54" = type { %"struct.llvm::ilist_traits.55", %"class.llvm::Argument"* }
|
|
|
|
|
%"struct.llvm::ilist_traits.55" = type { %"class.llvm::ilist_half_node.61" }
|
|
|
|
|
%"class.llvm::ilist_half_node.61" = type { %"class.llvm::Argument"* }
|
|
|
|
|
%"class.llvm::Argument" = type { %"class.llvm::Value", %"class.llvm::ilist_node.192", %"class.llvm::Function"* }
|
|
|
|
|
%"class.llvm::ilist_node.192" = type { %"class.llvm::ilist_half_node.61", %"class.llvm::Argument"* }
|
|
|
|
|
%"class.llvm::AttributeSet" = type { %"class.llvm::AttributeSetImpl"* }
|
|
|
|
|
%"class.llvm::AttributeSetImpl" = type opaque
|
|
|
|
|
%"class.llvm::MachineFunction" = type { %"class.llvm::Function"*, %"class.llvm::TargetMachine"*, %"class.llvm::MCContext"*, %"class.llvm::MachineModuleInfo"*, %"class.llvm::GCModuleInfo"*, %"class.llvm::MachineRegisterInfo"*, %"struct.llvm::MachineFunctionInfo"*, %"class.llvm::MachineFrameInfo"*, %"class.llvm::MachineConstantPool"*, %"class.llvm::MachineJumpTableInfo"*, %"class.std::vector.163", %"class.llvm::BumpPtrAllocator", %"class.llvm::Recycler", %"class.llvm::ArrayRecycler", %"class.llvm::Recycler.180", %"struct.llvm::ilist.181", i32, i32, i8 }
|
|
|
|
|
%"class.llvm::TargetMachine" = type { i32 (...)**, %"class.llvm::Target"*, %"class.std::basic_string", %"class.std::basic_string", %"class.std::basic_string", %"class.llvm::MCCodeGenInfo"*, %"class.llvm::MCAsmInfo"*, i8, %"class.llvm::TargetOptions" }
|
|
|
|
|
%"class.llvm::Target" = type opaque
|
|
|
|
|
%"class.llvm::MCCodeGenInfo" = type opaque
|
|
|
|
|
%"class.llvm::MCAsmInfo" = type opaque
|
|
|
|
|
%"class.llvm::TargetOptions" = type { [2 x i8], i32, i8, i32, i8, %"class.std::basic_string", i32, i32 }
|
|
|
|
|
%"class.llvm::MCContext" = type { %"class.llvm::SourceMgr"*, %"class.llvm::MCAsmInfo"*, %"class.llvm::MCRegisterInfo"*, %"class.llvm::MCObjectFileInfo"*, %"class.llvm::BumpPtrAllocator", %"class.llvm::StringMap", %"class.llvm::StringMap.62", i32, %"class.llvm::DenseMap.63", i8*, %"class.llvm::raw_ostream"*, i8, %"class.std::basic_string", %"class.std::basic_string", %"class.std::vector", %"class.std::vector.70", %"class.llvm::MCDwarfLoc", i8, i8, i32, %"class.llvm::MCSection"*, %"class.llvm::MCSymbol"*, %"class.llvm::MCSymbol"*, %"class.std::vector.75", %"class.llvm::StringRef", %"class.llvm::StringRef", i8, %"class.llvm::DenseMap.80", %"class.std::vector.84", i8*, i8*, i8*, i8 }
|
|
|
|
|
%"class.llvm::SourceMgr" = type opaque
|
|
|
|
|
%"class.llvm::MCRegisterInfo" = type { %"struct.llvm::MCRegisterDesc"*, i32, i32, i32, %"class.llvm::MCRegisterClass"*, i32, i32, [2 x i16]*, i16*, i8*, i16*, i32, i16*, i32, i32, i32, i32, %"struct.llvm::MCRegisterInfo::DwarfLLVMRegPair"*, %"struct.llvm::MCRegisterInfo::DwarfLLVMRegPair"*, %"struct.llvm::MCRegisterInfo::DwarfLLVMRegPair"*, %"struct.llvm::MCRegisterInfo::DwarfLLVMRegPair"*, %"class.llvm::DenseMap" }
|
|
|
|
|
%"struct.llvm::MCRegisterDesc" = type { i32, i32, i32, i32, i32, i32 }
|
|
|
|
|
%"class.llvm::MCRegisterClass" = type { i8*, i16*, i8*, i16, i16, i16, i16, i16, i8, i8 }
|
|
|
|
|
%"struct.llvm::MCRegisterInfo::DwarfLLVMRegPair" = type { i32, i32 }
|
|
|
|
|
%"class.llvm::DenseMap" = type { %"struct.std::pair"*, i32, i32, i32 }
|
|
|
|
|
%"struct.std::pair" = type { i32, i32 }
|
|
|
|
|
%"class.llvm::MCObjectFileInfo" = type opaque
|
|
|
|
|
%"class.llvm::BumpPtrAllocator" = type { i64, i64, %"class.llvm::SlabAllocator"*, %"class.llvm::MemSlab"*, i8*, i8*, i64 }
|
|
|
|
|
%"class.llvm::SlabAllocator" = type { i32 (...)** }
|
|
|
|
|
%"class.llvm::MemSlab" = type { i64, %"class.llvm::MemSlab"* }
|
|
|
|
|
%"class.llvm::StringMap" = type { %"class.llvm::StringMapImpl", %"class.llvm::BumpPtrAllocator"* }
|
|
|
|
|
%"class.llvm::StringMapImpl" = type { %"class.llvm::StringMapEntryBase"**, i32, i32, i32, i32 }
|
|
|
|
|
%"class.llvm::StringMapEntryBase" = type { i32 }
|
|
|
|
|
%"class.llvm::StringMap.62" = type { %"class.llvm::StringMapImpl", %"class.llvm::BumpPtrAllocator"* }
|
|
|
|
|
%"class.llvm::DenseMap.63" = type { %"struct.std::pair.66"*, i32, i32, i32 }
|
|
|
|
|
%"struct.std::pair.66" = type opaque
|
|
|
|
|
%"class.llvm::raw_ostream" = type { i32 (...)**, i8*, i8*, i8*, i32 }
|
|
|
|
|
%"class.std::vector" = type { %"struct.std::_Vector_base" }
|
|
|
|
|
%"struct.std::_Vector_base" = type { %"struct.std::_Vector_base<llvm::MCDwarfFile *, std::allocator<llvm::MCDwarfFile *> >::_Vector_impl" }
|
|
|
|
|
%"struct.std::_Vector_base<llvm::MCDwarfFile *, std::allocator<llvm::MCDwarfFile *> >::_Vector_impl" = type { %"class.llvm::MCDwarfFile"**, %"class.llvm::MCDwarfFile"**, %"class.llvm::MCDwarfFile"** }
|
|
|
|
|
%"class.llvm::MCDwarfFile" = type { %"class.llvm::StringRef", i32 }
|
|
|
|
|
%"class.llvm::StringRef" = type { i8*, i64 }
|
|
|
|
|
%"class.std::vector.70" = type { %"struct.std::_Vector_base.71" }
|
|
|
|
|
%"struct.std::_Vector_base.71" = type { %"struct.std::_Vector_base<llvm::StringRef, std::allocator<llvm::StringRef> >::_Vector_impl" }
|
|
|
|
|
%"struct.std::_Vector_base<llvm::StringRef, std::allocator<llvm::StringRef> >::_Vector_impl" = type { %"class.llvm::StringRef"*, %"class.llvm::StringRef"*, %"class.llvm::StringRef"* }
|
|
|
|
|
%"class.llvm::MCDwarfLoc" = type { i32, i32, i32, i32, i32, i32 }
|
|
|
|
|
%"class.llvm::MCSection" = type opaque
|
|
|
|
|
%"class.llvm::MCSymbol" = type { %"class.llvm::StringRef", %"class.llvm::MCSection"*, %"class.llvm::MCExpr"*, i8 }
|
|
|
|
|
%"class.llvm::MCExpr" = type opaque
|
|
|
|
|
%"class.std::vector.75" = type { %"struct.std::_Vector_base.76" }
|
|
|
|
|
%"struct.std::_Vector_base.76" = type { %"struct.std::_Vector_base<const llvm::MCGenDwarfLabelEntry *, std::allocator<const llvm::MCGenDwarfLabelEntry *> >::_Vector_impl" }
|
|
|
|
|
%"struct.std::_Vector_base<const llvm::MCGenDwarfLabelEntry *, std::allocator<const llvm::MCGenDwarfLabelEntry *> >::_Vector_impl" = type { %"class.llvm::MCGenDwarfLabelEntry"**, %"class.llvm::MCGenDwarfLabelEntry"**, %"class.llvm::MCGenDwarfLabelEntry"** }
|
|
|
|
|
%"class.llvm::MCGenDwarfLabelEntry" = type { %"class.llvm::StringRef", i32, i32, %"class.llvm::MCSymbol"* }
|
|
|
|
|
%"class.llvm::DenseMap.80" = type { %"struct.std::pair.83"*, i32, i32, i32 }
|
|
|
|
|
%"struct.std::pair.83" = type { %"class.llvm::MCSection"*, %"class.llvm::MCLineSection"* }
|
|
|
|
|
%"class.llvm::MCLineSection" = type { %"class.std::vector.215" }
|
|
|
|
|
%"class.std::vector.215" = type { %"struct.std::_Vector_base.216" }
|
|
|
|
|
%"struct.std::_Vector_base.216" = type { %"struct.std::_Vector_base<llvm::MCLineEntry, std::allocator<llvm::MCLineEntry> >::_Vector_impl" }
|
|
|
|
|
%"struct.std::_Vector_base<llvm::MCLineEntry, std::allocator<llvm::MCLineEntry> >::_Vector_impl" = type { %"class.llvm::MCLineEntry"*, %"class.llvm::MCLineEntry"*, %"class.llvm::MCLineEntry"* }
|
|
|
|
|
%"class.llvm::MCLineEntry" = type { %"class.llvm::MCDwarfLoc", %"class.llvm::MCSymbol"* }
|
|
|
|
|
%"class.std::vector.84" = type { %"struct.std::_Vector_base.85" }
|
|
|
|
|
%"struct.std::_Vector_base.85" = type { %"struct.std::_Vector_base<const llvm::MCSection *, std::allocator<const llvm::MCSection *> >::_Vector_impl" }
|
|
|
|
|
%"struct.std::_Vector_base<const llvm::MCSection *, std::allocator<const llvm::MCSection *> >::_Vector_impl" = type { %"class.llvm::MCSection"**, %"class.llvm::MCSection"**, %"class.llvm::MCSection"** }
|
|
|
|
|
%"class.llvm::MachineModuleInfo" = type { %"class.llvm::ImmutablePass", %"class.llvm::MCContext", %"class.llvm::Module"*, %"class.llvm::MachineModuleInfoImpl"*, %"class.std::vector.95", i32, %"class.std::vector.100", %"class.llvm::DenseMap.110", %"class.llvm::DenseMap.114", i32, %"class.std::vector.118", %"class.std::vector.123", %"class.std::vector.123", %"class.std::vector.128", %"class.llvm::SmallPtrSet", %"class.llvm::MMIAddrLabelMap"*, i8, i8, i8, i8, %"class.llvm::SmallVector.133" }
|
|
|
|
|
%"class.llvm::ImmutablePass" = type { %"class.llvm::ModulePass" }
|
|
|
|
|
%"class.llvm::ModulePass" = type { %"class.llvm::Pass" }
|
|
|
|
|
%"class.llvm::Pass" = type { i32 (...)**, %"class.llvm::AnalysisResolver"*, i8*, i32 }
|
|
|
|
|
%"class.llvm::AnalysisResolver" = type { %"class.std::vector.89", %"class.llvm::PMDataManager"* }
|
|
|
|
|
%"class.std::vector.89" = type { %"struct.std::_Vector_base.90" }
|
|
|
|
|
%"struct.std::_Vector_base.90" = type { %"struct.std::_Vector_base<std::pair<const void *, llvm::Pass *>, std::allocator<std::pair<const void *, llvm::Pass *> > >::_Vector_impl" }
|
|
|
|
|
%"struct.std::_Vector_base<std::pair<const void *, llvm::Pass *>, std::allocator<std::pair<const void *, llvm::Pass *> > >::_Vector_impl" = type { %"struct.std::pair.94"*, %"struct.std::pair.94"*, %"struct.std::pair.94"* }
|
|
|
|
|
%"struct.std::pair.94" = type { i8*, %"class.llvm::Pass"* }
|
|
|
|
|
%"class.llvm::PMDataManager" = type opaque
|
|
|
|
|
%"class.llvm::MachineModuleInfoImpl" = type { i32 (...)** }
|
|
|
|
|
%"class.std::vector.95" = type { %"struct.std::_Vector_base.96" }
|
|
|
|
|
%"struct.std::_Vector_base.96" = type { %"struct.std::_Vector_base<llvm::MachineMove, std::allocator<llvm::MachineMove> >::_Vector_impl" }
|
|
|
|
|
%"struct.std::_Vector_base<llvm::MachineMove, std::allocator<llvm::MachineMove> >::_Vector_impl" = type { %"class.llvm::MachineMove"*, %"class.llvm::MachineMove"*, %"class.llvm::MachineMove"* }
|
|
|
|
|
%"class.llvm::MachineMove" = type { %"class.llvm::MCSymbol"*, %"class.llvm::MachineLocation", %"class.llvm::MachineLocation" }
|
|
|
|
|
%"class.llvm::MachineLocation" = type { i8, i32, i32 }
|
|
|
|
|
%"class.std::vector.100" = type { %"struct.std::_Vector_base.101" }
|
|
|
|
|
%"struct.std::_Vector_base.101" = type { %"struct.std::_Vector_base<llvm::LandingPadInfo, std::allocator<llvm::LandingPadInfo> >::_Vector_impl" }
|
|
|
|
|
%"struct.std::_Vector_base<llvm::LandingPadInfo, std::allocator<llvm::LandingPadInfo> >::_Vector_impl" = type { %"struct.llvm::LandingPadInfo"*, %"struct.llvm::LandingPadInfo"*, %"struct.llvm::LandingPadInfo"* }
|
|
|
|
|
%"struct.llvm::LandingPadInfo" = type { %"class.llvm::MachineBasicBlock"*, %"class.llvm::SmallVector", %"class.llvm::SmallVector", %"class.llvm::MCSymbol"*, %"class.llvm::Function"*, %"class.std::vector.105" }
|
|
|
|
|
%"class.llvm::SmallVector" = type { %"class.llvm::SmallVectorImpl", %"struct.llvm::SmallVectorStorage" }
|
|
|
|
|
%"class.llvm::SmallVectorImpl" = type { %"class.llvm::SmallVectorTemplateBase" }
|
|
|
|
|
%"class.llvm::SmallVectorTemplateBase" = type { %"class.llvm::SmallVectorTemplateCommon" }
|
|
|
|
|
%"class.llvm::SmallVectorTemplateCommon" = type { %"class.llvm::SmallVectorBase", %"struct.llvm::AlignedCharArrayUnion" }
|
|
|
|
|
%"class.llvm::SmallVectorBase" = type { i8*, i8*, i8* }
|
|
|
|
|
%"struct.llvm::AlignedCharArrayUnion" = type { %"struct.llvm::AlignedCharArray" }
|
|
|
|
|
%"struct.llvm::AlignedCharArray" = type { [8 x i8] }
|
|
|
|
|
%"struct.llvm::SmallVectorStorage" = type { i8 }
|
|
|
|
|
%"class.std::vector.105" = type { %"struct.std::_Vector_base.106" }
|
|
|
|
|
%"struct.std::_Vector_base.106" = type { %"struct.std::_Vector_base<int, std::allocator<int> >::_Vector_impl" }
|
|
|
|
|
%"struct.std::_Vector_base<int, std::allocator<int> >::_Vector_impl" = type { i32*, i32*, i32* }
|
|
|
|
|
%"class.llvm::DenseMap.110" = type { %"struct.std::pair.113"*, i32, i32, i32 }
|
|
|
|
|
%"struct.std::pair.113" = type { %"class.llvm::MCSymbol"*, %"class.llvm::SmallVector.206" }
|
|
|
|
|
%"class.llvm::SmallVector.206" = type { [28 x i8], %"struct.llvm::SmallVectorStorage.207" }
|
|
|
|
|
%"struct.llvm::SmallVectorStorage.207" = type { [3 x %"struct.llvm::AlignedCharArrayUnion.198"] }
|
|
|
|
|
%"struct.llvm::AlignedCharArrayUnion.198" = type { %"struct.llvm::AlignedCharArray.199" }
|
|
|
|
|
%"struct.llvm::AlignedCharArray.199" = type { [4 x i8] }
|
|
|
|
|
%"class.llvm::DenseMap.114" = type { %"struct.std::pair.117"*, i32, i32, i32 }
|
|
|
|
|
%"struct.std::pair.117" = type { %"class.llvm::MCSymbol"*, i32 }
|
|
|
|
|
%"class.std::vector.118" = type { %"struct.std::_Vector_base.119" }
|
|
|
|
|
%"struct.std::_Vector_base.119" = type { %"struct.std::_Vector_base<const llvm::GlobalVariable *, std::allocator<const llvm::GlobalVariable *> >::_Vector_impl" }
|
|
|
|
|
%"struct.std::_Vector_base<const llvm::GlobalVariable *, std::allocator<const llvm::GlobalVariable *> >::_Vector_impl" = type { %"class.llvm::GlobalVariable"**, %"class.llvm::GlobalVariable"**, %"class.llvm::GlobalVariable"** }
|
|
|
|
|
%"class.std::vector.123" = type { %"struct.std::_Vector_base.124" }
|
|
|
|
|
%"struct.std::_Vector_base.124" = type { %"struct.std::_Vector_base<unsigned int, std::allocator<unsigned int> >::_Vector_impl" }
|
|
|
|
|
%"struct.std::_Vector_base<unsigned int, std::allocator<unsigned int> >::_Vector_impl" = type { i32*, i32*, i32* }
|
|
|
|
|
%"class.std::vector.128" = type { %"struct.std::_Vector_base.129" }
|
|
|
|
|
%"struct.std::_Vector_base.129" = type { %"struct.std::_Vector_base<const llvm::Function *, std::allocator<const llvm::Function *> >::_Vector_impl" }
|
|
|
|
|
%"struct.std::_Vector_base<const llvm::Function *, std::allocator<const llvm::Function *> >::_Vector_impl" = type { %"class.llvm::Function"**, %"class.llvm::Function"**, %"class.llvm::Function"** }
|
|
|
|
|
%"class.llvm::SmallPtrSet" = type { %"class.llvm::SmallPtrSetImpl", [33 x i8*] }
|
|
|
|
|
%"class.llvm::SmallPtrSetImpl" = type { i8**, i8**, i32, i32, i32 }
|
|
|
|
|
%"class.llvm::MMIAddrLabelMap" = type opaque
|
|
|
|
|
%"class.llvm::SmallVector.133" = type { %"class.llvm::SmallVectorImpl.134", %"struct.llvm::SmallVectorStorage.139" }
|
|
|
|
|
%"class.llvm::SmallVectorImpl.134" = type { %"class.llvm::SmallVectorTemplateBase.135" }
|
|
|
|
|
%"class.llvm::SmallVectorTemplateBase.135" = type { %"class.llvm::SmallVectorTemplateCommon.136" }
|
|
|
|
|
%"class.llvm::SmallVectorTemplateCommon.136" = type { %"class.llvm::SmallVectorBase", %"struct.llvm::AlignedCharArrayUnion.137" }
|
|
|
|
|
%"struct.llvm::AlignedCharArrayUnion.137" = type { %"struct.llvm::AlignedCharArray.138" }
|
|
|
|
|
%"struct.llvm::AlignedCharArray.138" = type { [40 x i8] }
|
|
|
|
|
%"struct.llvm::SmallVectorStorage.139" = type { [3 x %"struct.llvm::AlignedCharArrayUnion.137"] }
|
|
|
|
|
%"class.llvm::GCModuleInfo" = type opaque
|
|
|
|
|
%"class.llvm::MachineRegisterInfo" = type { %"class.llvm::TargetRegisterInfo"*, i8, i8, %"class.llvm::IndexedMap", %"class.llvm::IndexedMap.146", %"class.llvm::MachineOperand"**, %"class.llvm::BitVector", %"class.llvm::BitVector", %"class.llvm::BitVector", %"class.std::vector.147", %"class.std::vector.123" }
|
|
|
|
|
%"class.llvm::TargetRegisterInfo" = type { i32 (...)**, %"class.llvm::MCRegisterInfo", %"struct.llvm::TargetRegisterInfoDesc"*, i8**, i32*, %"class.llvm::TargetRegisterClass"**, %"class.llvm::TargetRegisterClass"** }
|
|
|
|
|
%"struct.llvm::TargetRegisterInfoDesc" = type { i32, i8 }
|
|
|
|
|
%"class.llvm::TargetRegisterClass" = type { %"class.llvm::MCRegisterClass"*, i32*, i32*, i16*, %"class.llvm::TargetRegisterClass"**, void (%"class.llvm::ArrayRef"*, %"class.llvm::MachineFunction"*)* }
|
|
|
|
|
%"class.llvm::ArrayRef" = type { i16*, i64 }
|
|
|
|
|
%"class.llvm::IndexedMap" = type { %"class.std::vector.140", %"struct.std::pair.145", %"struct.llvm::VirtReg2IndexFunctor" }
|
|
|
|
|
%"class.std::vector.140" = type { %"struct.std::_Vector_base.141" }
|
|
|
|
|
%"struct.std::_Vector_base.141" = type { %"struct.std::_Vector_base<std::pair<const llvm::TargetRegisterClass *, llvm::MachineOperand *>, std::allocator<std::pair<const llvm::TargetRegisterClass *, llvm::MachineOperand *> > >::_Vector_impl" }
|
|
|
|
|
%"struct.std::_Vector_base<std::pair<const llvm::TargetRegisterClass *, llvm::MachineOperand *>, std::allocator<std::pair<const llvm::TargetRegisterClass *, llvm::MachineOperand *> > >::_Vector_impl" = type { %"struct.std::pair.145"*, %"struct.std::pair.145"*, %"struct.std::pair.145"* }
|
|
|
|
|
%"struct.std::pair.145" = type { %"class.llvm::TargetRegisterClass"*, %"class.llvm::MachineOperand"* }
|
|
|
|
|
%"class.llvm::MachineOperand" = type { i8, [3 x i8], %union.anon, %"class.llvm::MachineInstr"*, %union.anon.188 }
|
|
|
|
|
%union.anon = type { i32 }
|
|
|
|
|
%union.anon.188 = type { %struct.anon }
|
|
|
|
|
%struct.anon = type { %"class.llvm::MachineOperand"*, %"class.llvm::MachineOperand"* }
|
|
|
|
|
%"struct.llvm::VirtReg2IndexFunctor" = type { i8 }
|
|
|
|
|
%"class.llvm::IndexedMap.146" = type { %"class.std::vector.147", %"struct.std::pair.152", %"struct.llvm::VirtReg2IndexFunctor" }
|
|
|
|
|
%"class.std::vector.147" = type { %"struct.std::_Vector_base.148" }
|
|
|
|
|
%"struct.std::_Vector_base.148" = type { %"struct.std::_Vector_base<std::pair<unsigned int, unsigned int>, std::allocator<std::pair<unsigned int, unsigned int> > >::_Vector_impl" }
|
|
|
|
|
%"struct.std::_Vector_base<std::pair<unsigned int, unsigned int>, std::allocator<std::pair<unsigned int, unsigned int> > >::_Vector_impl" = type { %"struct.std::pair.152"*, %"struct.std::pair.152"*, %"struct.std::pair.152"* }
|
|
|
|
|
%"struct.std::pair.152" = type { i32, i32 }
|
|
|
|
|
%"class.llvm::BitVector" = type { i64*, i32, i32 }
|
|
|
|
|
%"struct.llvm::MachineFunctionInfo" = type { i32 (...)** }
|
|
|
|
|
%"class.llvm::MachineFrameInfo" = type opaque
|
|
|
|
|
%"class.llvm::MachineConstantPool" = type { %"class.llvm::DataLayout"*, i32, %"class.std::vector.153", %"class.llvm::DenseSet" }
|
|
|
|
|
%"class.llvm::DataLayout" = type opaque
|
|
|
|
|
%"class.std::vector.153" = type { %"struct.std::_Vector_base.154" }
|
|
|
|
|
%"struct.std::_Vector_base.154" = type { %"struct.std::_Vector_base<llvm::MachineConstantPoolEntry, std::allocator<llvm::MachineConstantPoolEntry> >::_Vector_impl" }
|
|
|
|
|
%"struct.std::_Vector_base<llvm::MachineConstantPoolEntry, std::allocator<llvm::MachineConstantPoolEntry> >::_Vector_impl" = type { %"class.llvm::MachineConstantPoolEntry"*, %"class.llvm::MachineConstantPoolEntry"*, %"class.llvm::MachineConstantPoolEntry"* }
|
|
|
|
|
%"class.llvm::MachineConstantPoolEntry" = type { %union.anon.158, i32 }
|
|
|
|
|
%union.anon.158 = type { %"class.llvm::Constant"* }
|
|
|
|
|
%"class.llvm::Constant" = type { %"class.llvm::User" }
|
|
|
|
|
%"class.llvm::DenseSet" = type { %"class.llvm::DenseMap.159" }
|
|
|
|
|
%"class.llvm::DenseMap.159" = type { %"struct.std::pair.162"*, i32, i32, i32 }
|
|
|
|
|
%"struct.std::pair.162" = type { %"class.llvm::MachineConstantPoolValue"*, i8 }
|
|
|
|
|
%"class.llvm::MachineConstantPoolValue" = type { i32 (...)**, %"class.llvm::Type"* }
|
|
|
|
|
%"class.llvm::MachineJumpTableInfo" = type opaque
|
|
|
|
|
%"class.std::vector.163" = type { %"struct.std::_Vector_base.164" }
|
|
|
|
|
%"struct.std::_Vector_base.164" = type { %"struct.std::_Vector_base<llvm::MachineBasicBlock *, std::allocator<llvm::MachineBasicBlock *> >::_Vector_impl" }
|
|
|
|
|
%"struct.std::_Vector_base<llvm::MachineBasicBlock *, std::allocator<llvm::MachineBasicBlock *> >::_Vector_impl" = type { %"class.llvm::MachineBasicBlock"**, %"class.llvm::MachineBasicBlock"**, %"class.llvm::MachineBasicBlock"** }
|
|
|
|
|
%"class.llvm::Recycler" = type { %"class.llvm::iplist.168" }
|
|
|
|
|
%"class.llvm::iplist.168" = type { %"struct.llvm::ilist_traits.169", %"struct.llvm::RecyclerStruct"* }
|
|
|
|
|
%"struct.llvm::ilist_traits.169" = type { %"struct.llvm::RecyclerStruct" }
|
|
|
|
|
%"struct.llvm::RecyclerStruct" = type { %"struct.llvm::RecyclerStruct"*, %"struct.llvm::RecyclerStruct"* }
|
|
|
|
|
%"class.llvm::ArrayRecycler" = type { %"class.llvm::SmallVector.174" }
|
|
|
|
|
%"class.llvm::SmallVector.174" = type { %"class.llvm::SmallVectorImpl.175", %"struct.llvm::SmallVectorStorage.179" }
|
|
|
|
|
%"class.llvm::SmallVectorImpl.175" = type { %"class.llvm::SmallVectorTemplateBase.176" }
|
|
|
|
|
%"class.llvm::SmallVectorTemplateBase.176" = type { %"class.llvm::SmallVectorTemplateCommon.177" }
|
|
|
|
|
%"class.llvm::SmallVectorTemplateCommon.177" = type { %"class.llvm::SmallVectorBase", %"struct.llvm::AlignedCharArrayUnion.178" }
|
|
|
|
|
%"struct.llvm::AlignedCharArrayUnion.178" = type { %"struct.llvm::AlignedCharArray" }
|
|
|
|
|
%"struct.llvm::SmallVectorStorage.179" = type { [7 x %"struct.llvm::AlignedCharArrayUnion.178"] }
|
|
|
|
|
%"class.llvm::Recycler.180" = type { %"class.llvm::iplist.168" }
|
|
|
|
|
%"struct.llvm::ilist.181" = type { %"class.llvm::iplist.182" }
|
|
|
|
|
%"class.llvm::iplist.182" = type { %"struct.llvm::ilist_traits.183", %"class.llvm::MachineBasicBlock"* }
|
|
|
|
|
%"struct.llvm::ilist_traits.183" = type { %"class.llvm::ilist_half_node.1" }
|
|
|
|
|
%"class.llvm::ArrayRecycler<llvm::MachineOperand, 8>::Capacity" = type { i8 }
|
|
|
|
|
%"class.llvm::ConstantInt" = type { %"class.llvm::Constant", %"class.llvm::APInt" }
|
|
|
|
|
%"class.llvm::APInt" = type { i32, %union.anon.189 }
|
|
|
|
|
%union.anon.189 = type { i64 }
|
|
|
|
|
%"class.llvm::ConstantFP" = type { %"class.llvm::Constant", %"class.llvm::APFloat" }
|
|
|
|
|
%"class.llvm::APFloat" = type { %"struct.llvm::fltSemantics"*, %"union.llvm::APFloat::Significand", i16, i8 }
|
|
|
|
|
%"struct.llvm::fltSemantics" = type opaque
|
|
|
|
|
%"union.llvm::APFloat::Significand" = type { i64 }
|
|
|
|
|
%"class.llvm::BlockAddress" = type { %"class.llvm::Constant" }
|
|
|
|
|
%"class.llvm::hash_code" = type { i64 }
|
|
|
|
|
%"struct.llvm::hashing::detail::hash_combine_recursive_helper" = type { [64 x i8], %"struct.llvm::hashing::detail::hash_state", i64 }
|
|
|
|
|
%"struct.llvm::hashing::detail::hash_state" = type { i64, i64, i64, i64, i64, i64, i64, i64 }
|
|
|
|
|
%"class.llvm::PrintReg" = type { %"class.llvm::TargetRegisterInfo"*, i32, i32 }
|
|
|
|
|
%"class.llvm::PseudoSourceValue" = type { %"class.llvm::Value" }
|
|
|
|
|
%"class.llvm::FoldingSetNodeID" = type { %"class.llvm::SmallVector.194" }
|
|
|
|
|
%"class.llvm::SmallVector.194" = type { [28 x i8], %"struct.llvm::SmallVectorStorage.200" }
|
|
|
|
|
%"struct.llvm::SmallVectorStorage.200" = type { [31 x %"struct.llvm::AlignedCharArrayUnion.198"] }
|
|
|
|
|
%"struct.llvm::ArrayRecycler<llvm::MachineOperand, 8>::FreeList" = type { %"struct.llvm::ArrayRecycler<llvm::MachineOperand, 8>::FreeList"* }
|
|
|
|
|
%"class.llvm::ilist_iterator.202" = type { %"class.llvm::MachineInstr"* }
|
|
|
|
|
%"class.llvm::TargetInstrInfo" = type { i32 (...)**, [28 x i8], i32, i32 }
|
|
|
|
|
%"struct.std::pair.203" = type { i8, i8 }
|
|
|
|
|
%"class.llvm::SmallVectorImpl.195" = type { %"class.llvm::SmallVectorTemplateBase.196" }
|
|
|
|
|
%"class.llvm::SmallVectorTemplateBase.196" = type { %"class.llvm::SmallVectorTemplateCommon.197" }
|
|
|
|
|
%"class.llvm::SmallVectorTemplateCommon.197" = type { %"class.llvm::SmallVectorBase", %"struct.llvm::AlignedCharArrayUnion.198" }
|
|
|
|
|
%"class.llvm::AliasAnalysis" = type { i32 (...)**, %"class.llvm::DataLayout"*, %"class.llvm::TargetLibraryInfo"*, %"class.llvm::AliasAnalysis"* }
|
|
|
|
|
%"class.llvm::TargetLibraryInfo" = type opaque
|
|
|
|
|
%"struct.llvm::AliasAnalysis::Location" = type { %"class.llvm::Value"*, i64, %"class.llvm::MDNode"* }
|
|
|
|
|
%"class.llvm::DIVariable" = type { %"class.llvm::DIDescriptor" }
|
|
|
|
|
%"class.llvm::DIDescriptor" = type { %"class.llvm::MDNode"* }
|
|
|
|
|
%"class.llvm::DIScope" = type { %"class.llvm::DIDescriptor" }
|
|
|
|
|
%"class.llvm::ArrayRef.208" = type { i32*, i64 }
|
|
|
|
|
%"class.llvm::SmallVector.209" = type { %"class.llvm::SmallVectorImpl.210", %"struct.llvm::SmallVectorStorage.214" }
|
|
|
|
|
%"class.llvm::SmallVectorImpl.210" = type { %"class.llvm::SmallVectorTemplateBase.211" }
|
|
|
|
|
%"class.llvm::SmallVectorTemplateBase.211" = type { %"class.llvm::SmallVectorTemplateCommon.212" }
|
|
|
|
|
%"class.llvm::SmallVectorTemplateCommon.212" = type { %"class.llvm::SmallVectorBase", %"struct.llvm::AlignedCharArrayUnion.213" }
|
|
|
|
|
%"struct.llvm::AlignedCharArrayUnion.213" = type { %"struct.llvm::AlignedCharArray" }
|
|
|
|
|
%"struct.llvm::SmallVectorStorage.214" = type { [7 x %"struct.llvm::AlignedCharArrayUnion.213"] }
|
|
|
|
|
%"class.llvm::Twine" = type { %"union.llvm::Twine::Child", %"union.llvm::Twine::Child", i8, i8 }
|
|
|
|
|
%"union.llvm::Twine::Child" = type { %"class.llvm::Twine"* }
|
|
|
|
|
%"struct.std::random_access_iterator_tag" = type { i8 }
|
|
|
|
|
|
|
|
|
|
declare void @_ZN4llvm19MachineRegisterInfo27removeRegOperandFromUseListEPNS_14MachineOperandE(%"class.llvm::MachineRegisterInfo"*, %"class.llvm::MachineOperand"*)
|
|
|
|
|
|
|
|
|
|
declare void @_ZN4llvm19MachineRegisterInfo22addRegOperandToUseListEPNS_14MachineOperandE(%"class.llvm::MachineRegisterInfo"*, %"class.llvm::MachineOperand"*)
|
|
|
|
|
|
|
|
|
|
declare zeroext i32 @_ZNK4llvm14MCRegisterInfo9getSubRegEjj(%"class.llvm::MCRegisterInfo"*, i32 zeroext, i32 zeroext)
|
|
|
|
|
|
|
|
|
|
define void @_ZN4llvm14MachineOperand12substPhysRegEjRKNS_18TargetRegisterInfoE(%"class.llvm::MachineOperand"* %this, i32 zeroext %Reg, %"class.llvm::TargetRegisterInfo"* %TRI) align 2 {
|
|
|
|
|
entry:
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%SubReg_TargetFlags.i = getelementptr inbounds %"class.llvm::MachineOperand", %"class.llvm::MachineOperand"* %this, i64 0, i32 1
|
This patch addresses bug 15031.
The common code in the post-RA scheduler to break anti-dependencies on the
critical path contained a flaw. In the reported case, an anti-dependency
between the overlapping registers %X4 and %R4 exists:
%X29<def> = OR8 %X4, %X4
%R4<def>, %X3<def,dead,tied3> = LBZU 1, %X3<kill,tied1>
The unpatched code breaks the dependency by replacing %R4 and its uses
with %R3, the first register on the available list. However, %R3 and
%X3 overlap, so this creates two overlapping definitions on the same
instruction.
The fix is straightforward, preventing selection of a register that
overlaps any other defined register on the same instruction.
The test case is reduced from the bug report, and verifies that we no
longer produce "lbzu 3, 1(3)" when breaking this anti-dependency.
llvm-svn: 173706
2013-01-29 02:36:58 +08:00
|
|
|
%0 = bitcast [3 x i8]* %SubReg_TargetFlags.i to i24*
|
2015-02-28 05:17:42 +08:00
|
|
|
%bf.load.i = load i24, i24* %0, align 1
|
This patch addresses bug 15031.
The common code in the post-RA scheduler to break anti-dependencies on the
critical path contained a flaw. In the reported case, an anti-dependency
between the overlapping registers %X4 and %R4 exists:
%X29<def> = OR8 %X4, %X4
%R4<def>, %X3<def,dead,tied3> = LBZU 1, %X3<kill,tied1>
The unpatched code breaks the dependency by replacing %R4 and its uses
with %R3, the first register on the available list. However, %R3 and
%X3 overlap, so this creates two overlapping definitions on the same
instruction.
The fix is straightforward, preventing selection of a register that
overlaps any other defined register on the same instruction.
The test case is reduced from the bug report, and verifies that we no
longer produce "lbzu 3, 1(3)" when breaking this anti-dependency.
llvm-svn: 173706
2013-01-29 02:36:58 +08:00
|
|
|
%bf.lshr.i = lshr i24 %bf.load.i, 12
|
|
|
|
|
%tobool = icmp eq i24 %bf.lshr.i, 0
|
|
|
|
|
br i1 %tobool, label %if.end, label %if.then
|
|
|
|
|
|
|
|
|
|
if.then: ; preds = %entry
|
|
|
|
|
%bf.cast.i = zext i24 %bf.lshr.i to i32
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%add.ptr = getelementptr inbounds %"class.llvm::TargetRegisterInfo", %"class.llvm::TargetRegisterInfo"* %TRI, i64 0, i32 1
|
This patch addresses bug 15031.
The common code in the post-RA scheduler to break anti-dependencies on the
critical path contained a flaw. In the reported case, an anti-dependency
between the overlapping registers %X4 and %R4 exists:
%X29<def> = OR8 %X4, %X4
%R4<def>, %X3<def,dead,tied3> = LBZU 1, %X3<kill,tied1>
The unpatched code breaks the dependency by replacing %R4 and its uses
with %R3, the first register on the available list. However, %R3 and
%X3 overlap, so this creates two overlapping definitions on the same
instruction.
The fix is straightforward, preventing selection of a register that
overlaps any other defined register on the same instruction.
The test case is reduced from the bug report, and verifies that we no
longer produce "lbzu 3, 1(3)" when breaking this anti-dependency.
llvm-svn: 173706
2013-01-29 02:36:58 +08:00
|
|
|
%call3 = tail call zeroext i32 @_ZNK4llvm14MCRegisterInfo9getSubRegEjj(%"class.llvm::MCRegisterInfo"* %add.ptr, i32 zeroext %Reg, i32 zeroext %bf.cast.i)
|
2015-02-28 05:17:42 +08:00
|
|
|
%bf.load.i10 = load i24, i24* %0, align 1
|
This patch addresses bug 15031.
The common code in the post-RA scheduler to break anti-dependencies on the
critical path contained a flaw. In the reported case, an anti-dependency
between the overlapping registers %X4 and %R4 exists:
%X29<def> = OR8 %X4, %X4
%R4<def>, %X3<def,dead,tied3> = LBZU 1, %X3<kill,tied1>
The unpatched code breaks the dependency by replacing %R4 and its uses
with %R3, the first register on the available list. However, %R3 and
%X3 overlap, so this creates two overlapping definitions on the same
instruction.
The fix is straightforward, preventing selection of a register that
overlaps any other defined register on the same instruction.
The test case is reduced from the bug report, and verifies that we no
longer produce "lbzu 3, 1(3)" when breaking this anti-dependency.
llvm-svn: 173706
2013-01-29 02:36:58 +08:00
|
|
|
%bf.clear.i = and i24 %bf.load.i10, 4095
|
|
|
|
|
store i24 %bf.clear.i, i24* %0, align 1
|
|
|
|
|
br label %if.end
|
|
|
|
|
|
|
|
|
|
if.end: ; preds = %entry, %if.then
|
|
|
|
|
%Reg.addr.0 = phi i32 [ %call3, %if.then ], [ %Reg, %entry ]
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%RegNo.i.i = getelementptr inbounds %"class.llvm::MachineOperand", %"class.llvm::MachineOperand"* %this, i64 0, i32 2, i32 0
|
2015-02-28 05:17:42 +08:00
|
|
|
%1 = load i32, i32* %RegNo.i.i, align 4
|
This patch addresses bug 15031.
The common code in the post-RA scheduler to break anti-dependencies on the
critical path contained a flaw. In the reported case, an anti-dependency
between the overlapping registers %X4 and %R4 exists:
%X29<def> = OR8 %X4, %X4
%R4<def>, %X3<def,dead,tied3> = LBZU 1, %X3<kill,tied1>
The unpatched code breaks the dependency by replacing %R4 and its uses
with %R3, the first register on the available list. However, %R3 and
%X3 overlap, so this creates two overlapping definitions on the same
instruction.
The fix is straightforward, preventing selection of a register that
overlaps any other defined register on the same instruction.
The test case is reduced from the bug report, and verifies that we no
longer produce "lbzu 3, 1(3)" when breaking this anti-dependency.
llvm-svn: 173706
2013-01-29 02:36:58 +08:00
|
|
|
%cmp.i = icmp eq i32 %1, %Reg.addr.0
|
|
|
|
|
br i1 %cmp.i, label %_ZN4llvm14MachineOperand6setRegEj.exit, label %if.end.i
|
|
|
|
|
|
|
|
|
|
if.end.i: ; preds = %if.end
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%ParentMI.i.i = getelementptr inbounds %"class.llvm::MachineOperand", %"class.llvm::MachineOperand"* %this, i64 0, i32 3
|
2015-02-28 05:17:42 +08:00
|
|
|
%2 = load %"class.llvm::MachineInstr"*, %"class.llvm::MachineInstr"** %ParentMI.i.i, align 8
|
This patch addresses bug 15031.
The common code in the post-RA scheduler to break anti-dependencies on the
critical path contained a flaw. In the reported case, an anti-dependency
between the overlapping registers %X4 and %R4 exists:
%X29<def> = OR8 %X4, %X4
%R4<def>, %X3<def,dead,tied3> = LBZU 1, %X3<kill,tied1>
The unpatched code breaks the dependency by replacing %R4 and its uses
with %R3, the first register on the available list. However, %R3 and
%X3 overlap, so this creates two overlapping definitions on the same
instruction.
The fix is straightforward, preventing selection of a register that
overlaps any other defined register on the same instruction.
The test case is reduced from the bug report, and verifies that we no
longer produce "lbzu 3, 1(3)" when breaking this anti-dependency.
llvm-svn: 173706
2013-01-29 02:36:58 +08:00
|
|
|
%tobool.i = icmp eq %"class.llvm::MachineInstr"* %2, null
|
|
|
|
|
br i1 %tobool.i, label %if.end13.i, label %if.then3.i
|
|
|
|
|
|
|
|
|
|
if.then3.i: ; preds = %if.end.i
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%Parent.i.i = getelementptr inbounds %"class.llvm::MachineInstr", %"class.llvm::MachineInstr"* %2, i64 0, i32 2
|
2015-02-28 05:17:42 +08:00
|
|
|
%3 = load %"class.llvm::MachineBasicBlock"*, %"class.llvm::MachineBasicBlock"** %Parent.i.i, align 8
|
This patch addresses bug 15031.
The common code in the post-RA scheduler to break anti-dependencies on the
critical path contained a flaw. In the reported case, an anti-dependency
between the overlapping registers %X4 and %R4 exists:
%X29<def> = OR8 %X4, %X4
%R4<def>, %X3<def,dead,tied3> = LBZU 1, %X3<kill,tied1>
The unpatched code breaks the dependency by replacing %R4 and its uses
with %R3, the first register on the available list. However, %R3 and
%X3 overlap, so this creates two overlapping definitions on the same
instruction.
The fix is straightforward, preventing selection of a register that
overlaps any other defined register on the same instruction.
The test case is reduced from the bug report, and verifies that we no
longer produce "lbzu 3, 1(3)" when breaking this anti-dependency.
llvm-svn: 173706
2013-01-29 02:36:58 +08:00
|
|
|
%tobool5.i = icmp eq %"class.llvm::MachineBasicBlock"* %3, null
|
|
|
|
|
br i1 %tobool5.i, label %if.end13.i, label %if.then6.i
|
|
|
|
|
|
|
|
|
|
if.then6.i: ; preds = %if.then3.i
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%xParent.i.i = getelementptr inbounds %"class.llvm::MachineBasicBlock", %"class.llvm::MachineBasicBlock"* %3, i64 0, i32 4
|
2015-02-28 05:17:42 +08:00
|
|
|
%4 = load %"class.llvm::MachineFunction"*, %"class.llvm::MachineFunction"** %xParent.i.i, align 8
|
This patch addresses bug 15031.
The common code in the post-RA scheduler to break anti-dependencies on the
critical path contained a flaw. In the reported case, an anti-dependency
between the overlapping registers %X4 and %R4 exists:
%X29<def> = OR8 %X4, %X4
%R4<def>, %X3<def,dead,tied3> = LBZU 1, %X3<kill,tied1>
The unpatched code breaks the dependency by replacing %R4 and its uses
with %R3, the first register on the available list. However, %R3 and
%X3 overlap, so this creates two overlapping definitions on the same
instruction.
The fix is straightforward, preventing selection of a register that
overlaps any other defined register on the same instruction.
The test case is reduced from the bug report, and verifies that we no
longer produce "lbzu 3, 1(3)" when breaking this anti-dependency.
llvm-svn: 173706
2013-01-29 02:36:58 +08:00
|
|
|
%tobool8.i = icmp eq %"class.llvm::MachineFunction"* %4, null
|
|
|
|
|
br i1 %tobool8.i, label %if.end13.i, label %if.then9.i
|
|
|
|
|
|
|
|
|
|
if.then9.i: ; preds = %if.then6.i
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%RegInfo.i.i = getelementptr inbounds %"class.llvm::MachineFunction", %"class.llvm::MachineFunction"* %4, i64 0, i32 5
|
2015-02-28 05:17:42 +08:00
|
|
|
%5 = load %"class.llvm::MachineRegisterInfo"*, %"class.llvm::MachineRegisterInfo"** %RegInfo.i.i, align 8
|
This patch addresses bug 15031.
The common code in the post-RA scheduler to break anti-dependencies on the
critical path contained a flaw. In the reported case, an anti-dependency
between the overlapping registers %X4 and %R4 exists:
%X29<def> = OR8 %X4, %X4
%R4<def>, %X3<def,dead,tied3> = LBZU 1, %X3<kill,tied1>
The unpatched code breaks the dependency by replacing %R4 and its uses
with %R3, the first register on the available list. However, %R3 and
%X3 overlap, so this creates two overlapping definitions on the same
instruction.
The fix is straightforward, preventing selection of a register that
overlaps any other defined register on the same instruction.
The test case is reduced from the bug report, and verifies that we no
longer produce "lbzu 3, 1(3)" when breaking this anti-dependency.
llvm-svn: 173706
2013-01-29 02:36:58 +08:00
|
|
|
tail call void @_ZN4llvm19MachineRegisterInfo27removeRegOperandFromUseListEPNS_14MachineOperandE(%"class.llvm::MachineRegisterInfo"* %5, %"class.llvm::MachineOperand"* %this)
|
2013-08-22 06:20:53 +08:00
|
|
|
store i32 %Reg.addr.0, i32* %RegNo.i.i, align 4
|
This patch addresses bug 15031.
The common code in the post-RA scheduler to break anti-dependencies on the
critical path contained a flaw. In the reported case, an anti-dependency
between the overlapping registers %X4 and %R4 exists:
%X29<def> = OR8 %X4, %X4
%R4<def>, %X3<def,dead,tied3> = LBZU 1, %X3<kill,tied1>
The unpatched code breaks the dependency by replacing %R4 and its uses
with %R3, the first register on the available list. However, %R3 and
%X3 overlap, so this creates two overlapping definitions on the same
instruction.
The fix is straightforward, preventing selection of a register that
overlaps any other defined register on the same instruction.
The test case is reduced from the bug report, and verifies that we no
longer produce "lbzu 3, 1(3)" when breaking this anti-dependency.
llvm-svn: 173706
2013-01-29 02:36:58 +08:00
|
|
|
tail call void @_ZN4llvm19MachineRegisterInfo22addRegOperandToUseListEPNS_14MachineOperandE(%"class.llvm::MachineRegisterInfo"* %5, %"class.llvm::MachineOperand"* %this)
|
|
|
|
|
br label %_ZN4llvm14MachineOperand6setRegEj.exit
|
|
|
|
|
|
|
|
|
|
if.end13.i: ; preds = %if.then6.i, %if.then3.i, %if.end.i
|
2013-08-22 06:20:53 +08:00
|
|
|
store i32 %Reg.addr.0, i32* %RegNo.i.i, align 4
|
This patch addresses bug 15031.
The common code in the post-RA scheduler to break anti-dependencies on the
critical path contained a flaw. In the reported case, an anti-dependency
between the overlapping registers %X4 and %R4 exists:
%X29<def> = OR8 %X4, %X4
%R4<def>, %X3<def,dead,tied3> = LBZU 1, %X3<kill,tied1>
The unpatched code breaks the dependency by replacing %R4 and its uses
with %R3, the first register on the available list. However, %R3 and
%X3 overlap, so this creates two overlapping definitions on the same
instruction.
The fix is straightforward, preventing selection of a register that
overlaps any other defined register on the same instruction.
The test case is reduced from the bug report, and verifies that we no
longer produce "lbzu 3, 1(3)" when breaking this anti-dependency.
llvm-svn: 173706
2013-01-29 02:36:58 +08:00
|
|
|
br label %_ZN4llvm14MachineOperand6setRegEj.exit
|
|
|
|
|
|
|
|
|
|
_ZN4llvm14MachineOperand6setRegEj.exit: ; preds = %if.end, %if.then9.i, %if.end13.i
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; CHECK-NOT: lbzu 3, 1(3)
|