From 30815c536baacc07e925f0aef23a5395883173dc Mon Sep 17 00:00:00 2001 From: Dimitry Andric Date: Thu, 20 Oct 2011 21:10:27 +0000 Subject: Vendor import of llvm release_30 branch r142614: http://llvm.org/svn/llvm-project/llvm/branches/release_30@142614 --- lib/Target/ARM/ARM.h | 21 +- lib/Target/ARM/ARM.td | 23 +- lib/Target/ARM/ARMAddressingModes.h | 595 --- lib/Target/ARM/ARMAsmBackend.cpp | 516 --- lib/Target/ARM/ARMAsmPrinter.cpp | 194 +- lib/Target/ARM/ARMBaseInfo.h | 294 -- lib/Target/ARM/ARMBaseInstrInfo.cpp | 522 ++- lib/Target/ARM/ARMBaseInstrInfo.h | 161 +- lib/Target/ARM/ARMBaseRegisterInfo.cpp | 287 +- lib/Target/ARM/ARMBaseRegisterInfo.h | 19 +- lib/Target/ARM/ARMCodeEmitter.cpp | 35 +- lib/Target/ARM/ARMConstantIslandPass.cpp | 23 +- lib/Target/ARM/ARMConstantPoolValue.cpp | 317 +- lib/Target/ARM/ARMConstantPoolValue.h | 189 +- lib/Target/ARM/ARMExpandPseudoInsts.cpp | 61 +- lib/Target/ARM/ARMFastISel.cpp | 93 +- lib/Target/ARM/ARMFixupKinds.h | 97 - lib/Target/ARM/ARMFrameLowering.cpp | 31 +- lib/Target/ARM/ARMGlobalMerge.cpp | 10 +- lib/Target/ARM/ARMISelDAGToDAG.cpp | 336 +- lib/Target/ARM/ARMISelLowering.cpp | 1295 +++++- lib/Target/ARM/ARMISelLowering.h | 47 +- lib/Target/ARM/ARMInstrFormats.td | 341 +- lib/Target/ARM/ARMInstrInfo.cpp | 26 +- lib/Target/ARM/ARMInstrInfo.td | 2752 ++++++++---- lib/Target/ARM/ARMInstrNEON.td | 153 +- lib/Target/ARM/ARMInstrThumb.td | 489 ++- lib/Target/ARM/ARMInstrThumb2.td | 1957 ++++++--- lib/Target/ARM/ARMInstrVFP.td | 113 +- lib/Target/ARM/ARMLoadStoreOptimizer.cpp | 49 +- lib/Target/ARM/ARMMCCodeEmitter.cpp | 1314 ------ lib/Target/ARM/ARMMCExpr.cpp | 73 - lib/Target/ARM/ARMMCExpr.h | 76 - lib/Target/ARM/ARMMCInstLower.cpp | 2 +- lib/Target/ARM/ARMMachObjectWriter.cpp | 389 -- lib/Target/ARM/ARMRegisterInfo.td | 24 +- lib/Target/ARM/ARMSelectionDAGInfo.cpp | 15 +- lib/Target/ARM/ARMSelectionDAGInfo.h | 17 + lib/Target/ARM/ARMSubtarget.cpp | 5 + lib/Target/ARM/ARMSubtarget.h | 18 + lib/Target/ARM/ARMTargetMachine.cpp | 91 +- lib/Target/ARM/ARMTargetMachine.h | 16 +- lib/Target/ARM/AsmParser/ARMAsmLexer.cpp | 37 +- lib/Target/ARM/AsmParser/ARMAsmParser.cpp | 4071 ++++++++++++++---- lib/Target/ARM/AsmParser/CMakeLists.txt | 9 + lib/Target/ARM/CMakeLists.txt | 48 +- lib/Target/ARM/Disassembler/ARMDisassembler.cpp | 4489 +++++++++++++++++--- lib/Target/ARM/Disassembler/ARMDisassembler.h | 99 - .../ARM/Disassembler/ARMDisassemblerCore.cpp | 3818 ----------------- lib/Target/ARM/Disassembler/ARMDisassemblerCore.h | 336 -- lib/Target/ARM/Disassembler/CMakeLists.txt | 11 +- .../ARM/Disassembler/ThumbDisassemblerCore.h | 2459 ----------- lib/Target/ARM/InstPrinter/ARMInstPrinter.cpp | 399 +- lib/Target/ARM/InstPrinter/ARMInstPrinter.h | 34 +- lib/Target/ARM/InstPrinter/CMakeLists.txt | 8 +- lib/Target/ARM/MCTargetDesc/ARMAddressingModes.h | 667 +++ lib/Target/ARM/MCTargetDesc/ARMAsmBackend.cpp | 531 +++ lib/Target/ARM/MCTargetDesc/ARMBaseInfo.h | 449 ++ lib/Target/ARM/MCTargetDesc/ARMFixupKinds.h | 97 + lib/Target/ARM/MCTargetDesc/ARMMCAsmInfo.cpp | 11 +- lib/Target/ARM/MCTargetDesc/ARMMCCodeEmitter.cpp | 1468 +++++++ lib/Target/ARM/MCTargetDesc/ARMMCExpr.cpp | 73 + lib/Target/ARM/MCTargetDesc/ARMMCExpr.h | 76 + lib/Target/ARM/MCTargetDesc/ARMMCTargetDesc.cpp | 179 +- lib/Target/ARM/MCTargetDesc/ARMMCTargetDesc.h | 21 +- .../ARM/MCTargetDesc/ARMMachObjectWriter.cpp | 388 ++ lib/Target/ARM/MCTargetDesc/CMakeLists.txt | 14 +- lib/Target/ARM/Makefile | 5 +- lib/Target/ARM/NEONMoveFix.cpp | 149 - lib/Target/ARM/TargetInfo/ARMTargetInfo.cpp | 2 +- lib/Target/ARM/TargetInfo/CMakeLists.txt | 8 +- lib/Target/ARM/Thumb1FrameLowering.cpp | 11 +- lib/Target/ARM/Thumb1RegisterInfo.cpp | 34 +- lib/Target/ARM/Thumb2ITBlockPass.cpp | 21 + lib/Target/ARM/Thumb2InstrInfo.cpp | 13 +- lib/Target/ARM/Thumb2SizeReduction.cpp | 21 +- lib/Target/Alpha/AlphaAsmPrinter.cpp | 2 +- lib/Target/Alpha/AlphaISelDAGToDAG.cpp | 7 +- lib/Target/Alpha/AlphaISelLowering.cpp | 8 +- lib/Target/Alpha/AlphaISelLowering.h | 2 +- lib/Target/Alpha/AlphaInstrInfo.cpp | 1 - lib/Target/Alpha/AlphaInstrInfo.td | 2 + lib/Target/Alpha/AlphaRegisterInfo.cpp | 18 +- lib/Target/Alpha/AlphaRegisterInfo.h | 4 - lib/Target/Alpha/AlphaSubtarget.cpp | 1 - lib/Target/Alpha/AlphaTargetMachine.cpp | 12 +- lib/Target/Alpha/AlphaTargetMachine.h | 5 +- lib/Target/Alpha/CMakeLists.txt | 25 +- .../Alpha/MCTargetDesc/AlphaMCTargetDesc.cpp | 35 +- lib/Target/Alpha/MCTargetDesc/CMakeLists.txt | 7 + lib/Target/Alpha/TargetInfo/AlphaTargetInfo.cpp | 2 +- lib/Target/Alpha/TargetInfo/CMakeLists.txt | 8 +- lib/Target/Blackfin/BlackfinAsmPrinter.cpp | 2 +- lib/Target/Blackfin/BlackfinFrameLowering.h | 4 +- lib/Target/Blackfin/BlackfinISelLowering.cpp | 4 +- lib/Target/Blackfin/BlackfinISelLowering.h | 2 +- lib/Target/Blackfin/BlackfinInstrInfo.cpp | 2 +- lib/Target/Blackfin/BlackfinIntrinsicInfo.cpp | 8 +- lib/Target/Blackfin/BlackfinIntrinsicInfo.h | 4 +- lib/Target/Blackfin/BlackfinRegisterInfo.cpp | 18 +- lib/Target/Blackfin/BlackfinRegisterInfo.h | 4 - lib/Target/Blackfin/BlackfinSubtarget.cpp | 2 +- lib/Target/Blackfin/BlackfinTargetMachine.cpp | 12 +- lib/Target/Blackfin/BlackfinTargetMachine.h | 5 +- lib/Target/Blackfin/CMakeLists.txt | 27 +- .../Blackfin/MCTargetDesc/BlackfinMCTargetDesc.cpp | 39 +- lib/Target/Blackfin/MCTargetDesc/CMakeLists.txt | 7 + .../Blackfin/TargetInfo/BlackfinTargetInfo.cpp | 2 +- lib/Target/Blackfin/TargetInfo/CMakeLists.txt | 8 +- lib/Target/CBackend/CBackend.cpp | 180 +- lib/Target/CBackend/CMakeLists.txt | 12 + lib/Target/CBackend/CTargetMachine.h | 5 +- .../CBackend/TargetInfo/CBackendTargetInfo.cpp | 4 +- lib/Target/CBackend/TargetInfo/CMakeLists.txt | 5 + lib/Target/CMakeLists.txt | 8 +- lib/Target/CellSPU/CMakeLists.txt | 27 +- lib/Target/CellSPU/MCTargetDesc/CMakeLists.txt | 7 + .../CellSPU/MCTargetDesc/SPUMCTargetDesc.cpp | 52 +- lib/Target/CellSPU/MCTargetDesc/SPUMCTargetDesc.h | 4 +- lib/Target/CellSPU/SPUAsmPrinter.cpp | 2 +- lib/Target/CellSPU/SPUFrameLowering.cpp | 20 - lib/Target/CellSPU/SPUFrameLowering.h | 14 - lib/Target/CellSPU/SPUISelLowering.cpp | 16 +- lib/Target/CellSPU/SPUISelLowering.h | 6 +- lib/Target/CellSPU/SPUInstrInfo.cpp | 4 +- lib/Target/CellSPU/SPUInstrInfo.td | 10 +- lib/Target/CellSPU/SPURegisterInfo.cpp | 19 +- lib/Target/CellSPU/SPURegisterInfo.h | 6 - lib/Target/CellSPU/SPUSubtarget.cpp | 2 +- lib/Target/CellSPU/SPUTargetMachine.cpp | 25 +- lib/Target/CellSPU/SPUTargetMachine.h | 5 +- lib/Target/CellSPU/TargetInfo/CMakeLists.txt | 8 +- .../CellSPU/TargetInfo/CellSPUTargetInfo.cpp | 2 +- lib/Target/CppBackend/CMakeLists.txt | 7 + lib/Target/CppBackend/CPPBackend.cpp | 141 +- lib/Target/CppBackend/CPPTargetMachine.h | 5 +- lib/Target/CppBackend/TargetInfo/CMakeLists.txt | 4 + .../CppBackend/TargetInfo/CppBackendTargetInfo.cpp | 4 +- lib/Target/MBlaze/AsmParser/CMakeLists.txt | 8 + lib/Target/MBlaze/AsmParser/MBlazeAsmLexer.cpp | 27 +- lib/Target/MBlaze/AsmParser/MBlazeAsmParser.cpp | 24 +- lib/Target/MBlaze/AsmParser/Makefile | 2 +- lib/Target/MBlaze/CMakeLists.txt | 36 +- lib/Target/MBlaze/Disassembler/CMakeLists.txt | 10 +- .../MBlaze/Disassembler/MBlazeDisassembler.cpp | 76 +- .../MBlaze/Disassembler/MBlazeDisassembler.h | 9 +- lib/Target/MBlaze/InstPrinter/CMakeLists.txt | 9 +- .../MBlaze/InstPrinter/MBlazeInstPrinter.cpp | 4 +- lib/Target/MBlaze/InstPrinter/MBlazeInstPrinter.h | 2 +- lib/Target/MBlaze/MBlaze.h | 12 +- lib/Target/MBlaze/MBlazeAsmBackend.cpp | 162 - lib/Target/MBlaze/MBlazeAsmPrinter.cpp | 21 +- lib/Target/MBlaze/MBlazeFrameLowering.cpp | 2 +- lib/Target/MBlaze/MBlazeISelLowering.cpp | 13 +- lib/Target/MBlaze/MBlazeISelLowering.h | 2 +- lib/Target/MBlaze/MBlazeInstrInfo.cpp | 5 +- lib/Target/MBlaze/MBlazeInstrInfo.h | 56 - lib/Target/MBlaze/MBlazeInstrInfo.td | 21 +- lib/Target/MBlaze/MBlazeIntrinsicInfo.cpp | 8 +- lib/Target/MBlaze/MBlazeIntrinsicInfo.h | 4 +- lib/Target/MBlaze/MBlazeMCCodeEmitter.cpp | 222 - lib/Target/MBlaze/MBlazeRegisterInfo.cpp | 172 +- lib/Target/MBlaze/MBlazeRegisterInfo.h | 12 - lib/Target/MBlaze/MBlazeSubtarget.cpp | 2 +- lib/Target/MBlaze/MBlazeTargetMachine.cpp | 50 +- lib/Target/MBlaze/MBlazeTargetMachine.h | 5 +- lib/Target/MBlaze/MBlazeTargetObjectFile.cpp | 2 +- lib/Target/MBlaze/MCTargetDesc/CMakeLists.txt | 13 +- .../MBlaze/MCTargetDesc/MBlazeAsmBackend.cpp | 159 + lib/Target/MBlaze/MCTargetDesc/MBlazeBaseInfo.h | 240 ++ .../MBlaze/MCTargetDesc/MBlazeMCCodeEmitter.cpp | 224 + .../MBlaze/MCTargetDesc/MBlazeMCTargetDesc.cpp | 93 +- .../MBlaze/MCTargetDesc/MBlazeMCTargetDesc.h | 11 + lib/Target/MBlaze/TargetInfo/CMakeLists.txt | 8 +- lib/Target/MBlaze/TargetInfo/MBlazeTargetInfo.cpp | 2 +- lib/Target/MSP430/CMakeLists.txt | 26 +- lib/Target/MSP430/InstPrinter/CMakeLists.txt | 8 +- .../MSP430/InstPrinter/MSP430InstPrinter.cpp | 4 +- lib/Target/MSP430/InstPrinter/MSP430InstPrinter.h | 4 +- lib/Target/MSP430/MCTargetDesc/CMakeLists.txt | 8 + .../MSP430/MCTargetDesc/MSP430MCTargetDesc.cpp | 51 +- .../MSP430/MCTargetDesc/MSP430MCTargetDesc.h | 4 +- lib/Target/MSP430/MSP430AsmPrinter.cpp | 14 +- lib/Target/MSP430/MSP430ISelLowering.cpp | 7 +- lib/Target/MSP430/MSP430ISelLowering.h | 4 +- lib/Target/MSP430/MSP430InstrInfo.cpp | 2 +- lib/Target/MSP430/MSP430RegisterInfo.cpp | 16 +- lib/Target/MSP430/MSP430RegisterInfo.h | 5 - lib/Target/MSP430/MSP430Subtarget.cpp | 2 +- lib/Target/MSP430/MSP430TargetMachine.cpp | 11 +- lib/Target/MSP430/MSP430TargetMachine.h | 5 +- lib/Target/MSP430/TargetInfo/CMakeLists.txt | 8 +- lib/Target/MSP430/TargetInfo/MSP430TargetInfo.cpp | 2 +- lib/Target/Mangler.cpp | 4 +- lib/Target/Mips/CMakeLists.txt | 28 +- lib/Target/Mips/InstPrinter/CMakeLists.txt | 8 +- lib/Target/Mips/InstPrinter/Makefile | 2 +- lib/Target/Mips/InstPrinter/MipsInstPrinter.cpp | 6 +- lib/Target/Mips/InstPrinter/MipsInstPrinter.h | 4 +- lib/Target/Mips/MCTargetDesc/CMakeLists.txt | 13 +- lib/Target/Mips/MCTargetDesc/MipsAsmBackend.cpp | 117 + lib/Target/Mips/MCTargetDesc/MipsBaseInfo.h | 113 + lib/Target/Mips/MCTargetDesc/MipsFixupKinds.h | 90 + lib/Target/Mips/MCTargetDesc/MipsMCAsmInfo.cpp | 3 +- lib/Target/Mips/MCTargetDesc/MipsMCCodeEmitter.cpp | 52 + lib/Target/Mips/MCTargetDesc/MipsMCTargetDesc.cpp | 122 +- lib/Target/Mips/MCTargetDesc/MipsMCTargetDesc.h | 21 +- lib/Target/Mips/Mips.h | 3 + lib/Target/Mips/Mips.td | 32 +- lib/Target/Mips/Mips64InstrInfo.td | 214 + lib/Target/Mips/MipsAsmPrinter.cpp | 42 +- lib/Target/Mips/MipsCallingConv.td | 55 +- lib/Target/Mips/MipsCodeEmitter.cpp | 245 ++ lib/Target/Mips/MipsDelaySlotFiller.cpp | 204 +- lib/Target/Mips/MipsFrameLowering.cpp | 19 +- lib/Target/Mips/MipsFrameLowering.h | 5 +- lib/Target/Mips/MipsISelDAGToDAG.cpp | 185 +- lib/Target/Mips/MipsISelLowering.cpp | 848 ++-- lib/Target/Mips/MipsISelLowering.h | 16 +- lib/Target/Mips/MipsInstrFPU.td | 236 +- lib/Target/Mips/MipsInstrFormats.td | 44 +- lib/Target/Mips/MipsInstrInfo.cpp | 210 +- lib/Target/Mips/MipsInstrInfo.h | 37 +- lib/Target/Mips/MipsInstrInfo.td | 832 ++-- lib/Target/Mips/MipsJITInfo.cpp | 230 + lib/Target/Mips/MipsJITInfo.h | 70 + lib/Target/Mips/MipsMCInstLower.cpp | 64 +- lib/Target/Mips/MipsMCInstLower.h | 5 +- lib/Target/Mips/MipsMCSymbolRefExpr.cpp | 9 +- lib/Target/Mips/MipsMCSymbolRefExpr.h | 7 +- lib/Target/Mips/MipsMachineFunction.h | 9 +- lib/Target/Mips/MipsRegisterInfo.cpp | 205 +- lib/Target/Mips/MipsRegisterInfo.h | 4 - lib/Target/Mips/MipsRegisterInfo.td | 119 +- lib/Target/Mips/MipsRelocations.h | 41 + lib/Target/Mips/MipsSubtarget.cpp | 38 +- lib/Target/Mips/MipsSubtarget.h | 15 +- lib/Target/Mips/MipsTargetMachine.cpp | 73 +- lib/Target/Mips/MipsTargetMachine.h | 48 +- lib/Target/Mips/MipsTargetObjectFile.cpp | 2 +- lib/Target/Mips/TargetInfo/CMakeLists.txt | 8 +- lib/Target/Mips/TargetInfo/MipsTargetInfo.cpp | 16 +- lib/Target/PTX/CMakeLists.txt | 32 +- lib/Target/PTX/InstPrinter/CMakeLists.txt | 13 + lib/Target/PTX/InstPrinter/Makefile | 16 + lib/Target/PTX/InstPrinter/PTXInstPrinter.cpp | 192 + lib/Target/PTX/InstPrinter/PTXInstPrinter.h | 47 + lib/Target/PTX/MCTargetDesc/CMakeLists.txt | 9 + lib/Target/PTX/MCTargetDesc/PTXBaseInfo.h | 63 + lib/Target/PTX/MCTargetDesc/PTXMCTargetDesc.cpp | 53 +- lib/Target/PTX/Makefile | 3 +- lib/Target/PTX/PTX.h | 28 +- lib/Target/PTX/PTX.td | 28 +- lib/Target/PTX/PTXAsmPrinter.cpp | 500 ++- lib/Target/PTX/PTXAsmPrinter.h | 57 + lib/Target/PTX/PTXCallingConv.td | 29 - lib/Target/PTX/PTXFPRoundingModePass.cpp | 179 + lib/Target/PTX/PTXISelDAGToDAG.cpp | 182 +- lib/Target/PTX/PTXISelLowering.cpp | 273 +- lib/Target/PTX/PTXISelLowering.h | 19 +- lib/Target/PTX/PTXInstrFormats.td | 31 +- lib/Target/PTX/PTXInstrInfo.cpp | 82 +- lib/Target/PTX/PTXInstrInfo.td | 1241 +++--- lib/Target/PTX/PTXInstrLoadStore.td | 278 ++ lib/Target/PTX/PTXIntrinsicInstrInfo.td | 78 +- lib/Target/PTX/PTXMCAsmStreamer.cpp | 15 +- lib/Target/PTX/PTXMCInstLower.cpp | 32 + lib/Target/PTX/PTXMFInfoExtract.cpp | 36 +- lib/Target/PTX/PTXMachineFunctionInfo.h | 163 +- lib/Target/PTX/PTXParamManager.cpp | 73 + lib/Target/PTX/PTXParamManager.h | 86 + lib/Target/PTX/PTXRegAlloc.cpp | 58 + lib/Target/PTX/PTXRegisterInfo.cpp | 29 +- lib/Target/PTX/PTXRegisterInfo.h | 18 +- lib/Target/PTX/PTXRegisterInfo.td | 540 +-- lib/Target/PTX/PTXSelectionDAGInfo.cpp | 149 + lib/Target/PTX/PTXSelectionDAGInfo.h | 53 + lib/Target/PTX/PTXSubtarget.cpp | 2 +- lib/Target/PTX/PTXSubtarget.h | 11 +- lib/Target/PTX/PTXTargetMachine.cpp | 318 +- lib/Target/PTX/PTXTargetMachine.h | 60 +- lib/Target/PTX/TargetInfo/CMakeLists.txt | 8 +- lib/Target/PTX/TargetInfo/PTXTargetInfo.cpp | 2 +- lib/Target/PTX/generate-register-td.py | 163 - lib/Target/PowerPC/CMakeLists.txt | 34 +- lib/Target/PowerPC/InstPrinter/CMakeLists.txt | 8 +- lib/Target/PowerPC/InstPrinter/PPCInstPrinter.cpp | 11 +- lib/Target/PowerPC/InstPrinter/PPCInstPrinter.h | 2 +- lib/Target/PowerPC/MCTargetDesc/CMakeLists.txt | 12 + lib/Target/PowerPC/MCTargetDesc/PPCAsmBackend.cpp | 191 + lib/Target/PowerPC/MCTargetDesc/PPCBaseInfo.h | 70 + lib/Target/PowerPC/MCTargetDesc/PPCFixupKinds.h | 45 + lib/Target/PowerPC/MCTargetDesc/PPCMCAsmInfo.cpp | 6 +- .../PowerPC/MCTargetDesc/PPCMCCodeEmitter.cpp | 193 + .../PowerPC/MCTargetDesc/PPCMCTargetDesc.cpp | 115 +- lib/Target/PowerPC/MCTargetDesc/PPCMCTargetDesc.h | 10 + lib/Target/PowerPC/MCTargetDesc/PPCPredicates.cpp | 31 + lib/Target/PowerPC/MCTargetDesc/PPCPredicates.h | 37 + lib/Target/PowerPC/PPC.h | 11 +- lib/Target/PowerPC/PPC.td | 4 +- lib/Target/PowerPC/PPCAsmBackend.cpp | 123 - lib/Target/PowerPC/PPCAsmPrinter.cpp | 14 +- lib/Target/PowerPC/PPCBranchSelector.cpp | 2 +- lib/Target/PowerPC/PPCCodeEmitter.cpp | 4 +- lib/Target/PowerPC/PPCFixupKinds.h | 45 - lib/Target/PowerPC/PPCFrameLowering.cpp | 17 +- lib/Target/PowerPC/PPCFrameLowering.h | 1 - lib/Target/PowerPC/PPCISelDAGToDAG.cpp | 2 +- lib/Target/PowerPC/PPCISelLowering.cpp | 65 +- lib/Target/PowerPC/PPCISelLowering.h | 17 +- lib/Target/PowerPC/PPCInstrInfo.cpp | 36 +- lib/Target/PowerPC/PPCInstrInfo.td | 6 + lib/Target/PowerPC/PPCMCCodeEmitter.cpp | 194 - lib/Target/PowerPC/PPCPredicates.cpp | 31 - lib/Target/PowerPC/PPCPredicates.h | 39 - lib/Target/PowerPC/PPCRegisterInfo.cpp | 79 +- lib/Target/PowerPC/PPCRegisterInfo.h | 8 - lib/Target/PowerPC/PPCSubtarget.cpp | 2 +- lib/Target/PowerPC/PPCTargetMachine.cpp | 71 +- lib/Target/PowerPC/PPCTargetMachine.h | 16 +- lib/Target/PowerPC/TargetInfo/CMakeLists.txt | 8 +- .../PowerPC/TargetInfo/PowerPCTargetInfo.cpp | 2 +- lib/Target/README.txt | 10 + lib/Target/Sparc/CMakeLists.txt | 25 +- lib/Target/Sparc/MCTargetDesc/CMakeLists.txt | 8 + .../Sparc/MCTargetDesc/SparcMCTargetDesc.cpp | 36 +- lib/Target/Sparc/SparcAsmPrinter.cpp | 2 +- lib/Target/Sparc/SparcISelLowering.cpp | 8 +- lib/Target/Sparc/SparcInstrInfo.cpp | 2 +- lib/Target/Sparc/SparcRegisterInfo.cpp | 15 +- lib/Target/Sparc/SparcRegisterInfo.h | 4 - lib/Target/Sparc/SparcSubtarget.cpp | 2 +- lib/Target/Sparc/SparcTargetMachine.cpp | 27 +- lib/Target/Sparc/SparcTargetMachine.h | 16 +- lib/Target/Sparc/TargetInfo/CMakeLists.txt | 8 +- lib/Target/Sparc/TargetInfo/SparcTargetInfo.cpp | 2 +- lib/Target/SystemZ/CMakeLists.txt | 25 +- lib/Target/SystemZ/MCTargetDesc/CMakeLists.txt | 7 + .../SystemZ/MCTargetDesc/SystemZMCTargetDesc.cpp | 39 +- lib/Target/SystemZ/SystemZAsmPrinter.cpp | 4 +- lib/Target/SystemZ/SystemZISelLowering.cpp | 1 + lib/Target/SystemZ/SystemZInstrInfo.cpp | 2 +- lib/Target/SystemZ/SystemZInstrInfo.td | 2 +- lib/Target/SystemZ/SystemZRegisterInfo.cpp | 17 +- lib/Target/SystemZ/SystemZRegisterInfo.h | 4 - lib/Target/SystemZ/SystemZSubtarget.cpp | 2 +- lib/Target/SystemZ/SystemZTargetMachine.cpp | 13 +- lib/Target/SystemZ/SystemZTargetMachine.h | 5 +- lib/Target/SystemZ/TargetInfo/CMakeLists.txt | 8 +- .../SystemZ/TargetInfo/SystemZTargetInfo.cpp | 2 +- lib/Target/Target.cpp | 10 +- lib/Target/TargetAsmInfo.cpp | 23 - lib/Target/TargetAsmLexer.cpp | 14 - lib/Target/TargetData.cpp | 53 +- lib/Target/TargetFrameLowering.cpp | 8 - lib/Target/TargetLoweringObjectFile.cpp | 59 +- lib/Target/TargetMachine.cpp | 67 +- lib/Target/TargetRegisterInfo.cpp | 49 +- lib/Target/X86/AsmParser/CMakeLists.txt | 11 +- lib/Target/X86/AsmParser/X86AsmLexer.cpp | 20 +- lib/Target/X86/AsmParser/X86AsmParser.cpp | 84 +- lib/Target/X86/CMakeLists.txt | 41 +- lib/Target/X86/Disassembler/CMakeLists.txt | 10 +- lib/Target/X86/Disassembler/X86Disassembler.cpp | 77 +- lib/Target/X86/Disassembler/X86Disassembler.h | 26 +- .../X86/Disassembler/X86DisassemblerDecoder.c | 165 +- .../Disassembler/X86DisassemblerDecoderCommon.h | 8 +- lib/Target/X86/InstPrinter/CMakeLists.txt | 9 +- lib/Target/X86/InstPrinter/X86ATTInstPrinter.cpp | 10 +- lib/Target/X86/InstPrinter/X86ATTInstPrinter.h | 2 +- lib/Target/X86/InstPrinter/X86InstComments.cpp | 31 +- lib/Target/X86/InstPrinter/X86IntelInstPrinter.cpp | 7 +- lib/Target/X86/InstPrinter/X86IntelInstPrinter.h | 2 +- lib/Target/X86/MCTargetDesc/CMakeLists.txt | 13 + lib/Target/X86/MCTargetDesc/X86AsmBackend.cpp | 458 ++ lib/Target/X86/MCTargetDesc/X86BaseInfo.h | 548 +++ lib/Target/X86/MCTargetDesc/X86FixupKinds.h | 33 + lib/Target/X86/MCTargetDesc/X86MCCodeEmitter.cpp | 1074 +++++ lib/Target/X86/MCTargetDesc/X86MCTargetDesc.cpp | 338 +- lib/Target/X86/MCTargetDesc/X86MCTargetDesc.h | 45 +- .../X86/MCTargetDesc/X86MachObjectWriter.cpp | 554 +++ lib/Target/X86/README-SSE.txt | 2 +- lib/Target/X86/README.txt | 38 +- lib/Target/X86/SSEDomainFix.cpp | 506 --- lib/Target/X86/TargetInfo/CMakeLists.txt | 8 +- lib/Target/X86/TargetInfo/X86TargetInfo.cpp | 2 +- lib/Target/X86/Utils/CMakeLists.txt | 8 +- lib/Target/X86/Utils/X86ShuffleDecode.cpp | 75 +- lib/Target/X86/Utils/X86ShuffleDecode.h | 20 + lib/Target/X86/X86.h | 30 +- lib/Target/X86/X86.td | 78 +- lib/Target/X86/X86AsmBackend.cpp | 453 -- lib/Target/X86/X86AsmPrinter.cpp | 23 +- lib/Target/X86/X86CodeEmitter.cpp | 36 +- lib/Target/X86/X86ELFWriterInfo.cpp | 2 +- lib/Target/X86/X86FastISel.cpp | 93 +- lib/Target/X86/X86FixupKinds.h | 33 - lib/Target/X86/X86FloatingPoint.cpp | 34 + lib/Target/X86/X86FrameLowering.cpp | 625 ++- lib/Target/X86/X86FrameLowering.h | 7 +- lib/Target/X86/X86ISelDAGToDAG.cpp | 16 +- lib/Target/X86/X86ISelLowering.cpp | 2940 ++++++++++--- lib/Target/X86/X86ISelLowering.h | 62 +- lib/Target/X86/X86InstrArithmetic.td | 96 +- lib/Target/X86/X86InstrCompiler.td | 102 +- lib/Target/X86/X86InstrExtension.td | 4 +- lib/Target/X86/X86InstrFormats.td | 8 +- lib/Target/X86/X86InstrFragmentsSIMD.td | 84 +- lib/Target/X86/X86InstrInfo.cpp | 1838 ++++---- lib/Target/X86/X86InstrInfo.h | 553 +-- lib/Target/X86/X86InstrInfo.td | 209 +- lib/Target/X86/X86InstrSSE.td | 4174 +++++++++++------- lib/Target/X86/X86InstrSystem.td | 114 +- lib/Target/X86/X86InstrVMX.td | 10 +- lib/Target/X86/X86MCCodeEmitter.cpp | 1044 ----- lib/Target/X86/X86MCInstLower.cpp | 23 +- lib/Target/X86/X86MachObjectWriter.cpp | 554 --- lib/Target/X86/X86MachineFunctionInfo.h | 20 +- lib/Target/X86/X86RegisterInfo.cpp | 165 +- lib/Target/X86/X86RegisterInfo.h | 25 +- lib/Target/X86/X86RegisterInfo.td | 11 +- lib/Target/X86/X86SelectionDAGInfo.cpp | 2 +- lib/Target/X86/X86Subtarget.cpp | 78 +- lib/Target/X86/X86Subtarget.h | 36 +- lib/Target/X86/X86TargetMachine.cpp | 173 +- lib/Target/X86/X86TargetMachine.h | 22 +- lib/Target/X86/X86TargetObjectFile.cpp | 76 - lib/Target/X86/X86TargetObjectFile.h | 22 - lib/Target/X86/X86VZeroUpper.cpp | 105 + lib/Target/XCore/CMakeLists.txt | 25 +- lib/Target/XCore/MCTargetDesc/CMakeLists.txt | 7 + .../XCore/MCTargetDesc/XCoreMCTargetDesc.cpp | 48 +- lib/Target/XCore/TargetInfo/CMakeLists.txt | 8 +- lib/Target/XCore/TargetInfo/XCoreTargetInfo.cpp | 2 +- lib/Target/XCore/XCoreAsmPrinter.cpp | 60 +- lib/Target/XCore/XCoreFrameLowering.cpp | 11 +- lib/Target/XCore/XCoreFrameLowering.h | 2 - lib/Target/XCore/XCoreISelDAGToDAG.cpp | 11 +- lib/Target/XCore/XCoreISelLowering.cpp | 37 +- lib/Target/XCore/XCoreISelLowering.h | 5 +- lib/Target/XCore/XCoreInstrInfo.cpp | 12 +- lib/Target/XCore/XCoreInstrInfo.h | 5 + lib/Target/XCore/XCoreInstrInfo.td | 81 +- lib/Target/XCore/XCoreRegisterInfo.cpp | 15 +- lib/Target/XCore/XCoreRegisterInfo.h | 5 - lib/Target/XCore/XCoreSubtarget.cpp | 2 +- lib/Target/XCore/XCoreTargetMachine.cpp | 10 +- lib/Target/XCore/XCoreTargetMachine.h | 5 +- 448 files changed, 38648 insertions(+), 28015 deletions(-) delete mode 100644 lib/Target/ARM/ARMAddressingModes.h delete mode 100644 lib/Target/ARM/ARMAsmBackend.cpp delete mode 100644 lib/Target/ARM/ARMBaseInfo.h delete mode 100644 lib/Target/ARM/ARMFixupKinds.h delete mode 100644 lib/Target/ARM/ARMMCCodeEmitter.cpp delete mode 100644 lib/Target/ARM/ARMMCExpr.cpp delete mode 100644 lib/Target/ARM/ARMMCExpr.h delete mode 100644 lib/Target/ARM/ARMMachObjectWriter.cpp delete mode 100644 lib/Target/ARM/Disassembler/ARMDisassembler.h delete mode 100644 lib/Target/ARM/Disassembler/ARMDisassemblerCore.cpp delete mode 100644 lib/Target/ARM/Disassembler/ARMDisassemblerCore.h delete mode 100644 lib/Target/ARM/Disassembler/ThumbDisassemblerCore.h create mode 100644 lib/Target/ARM/MCTargetDesc/ARMAddressingModes.h create mode 100644 lib/Target/ARM/MCTargetDesc/ARMAsmBackend.cpp create mode 100644 lib/Target/ARM/MCTargetDesc/ARMBaseInfo.h create mode 100644 lib/Target/ARM/MCTargetDesc/ARMFixupKinds.h create mode 100644 lib/Target/ARM/MCTargetDesc/ARMMCCodeEmitter.cpp create mode 100644 lib/Target/ARM/MCTargetDesc/ARMMCExpr.cpp create mode 100644 lib/Target/ARM/MCTargetDesc/ARMMCExpr.h create mode 100644 lib/Target/ARM/MCTargetDesc/ARMMachObjectWriter.cpp delete mode 100644 lib/Target/ARM/NEONMoveFix.cpp delete mode 100644 lib/Target/MBlaze/MBlazeAsmBackend.cpp delete mode 100644 lib/Target/MBlaze/MBlazeMCCodeEmitter.cpp create mode 100644 lib/Target/MBlaze/MCTargetDesc/MBlazeAsmBackend.cpp create mode 100644 lib/Target/MBlaze/MCTargetDesc/MBlazeBaseInfo.h create mode 100644 lib/Target/MBlaze/MCTargetDesc/MBlazeMCCodeEmitter.cpp create mode 100644 lib/Target/Mips/MCTargetDesc/MipsAsmBackend.cpp create mode 100644 lib/Target/Mips/MCTargetDesc/MipsBaseInfo.h create mode 100644 lib/Target/Mips/MCTargetDesc/MipsFixupKinds.h create mode 100644 lib/Target/Mips/MCTargetDesc/MipsMCCodeEmitter.cpp create mode 100644 lib/Target/Mips/Mips64InstrInfo.td create mode 100644 lib/Target/Mips/MipsCodeEmitter.cpp create mode 100644 lib/Target/Mips/MipsJITInfo.cpp create mode 100644 lib/Target/Mips/MipsJITInfo.h create mode 100644 lib/Target/Mips/MipsRelocations.h create mode 100644 lib/Target/PTX/InstPrinter/CMakeLists.txt create mode 100644 lib/Target/PTX/InstPrinter/Makefile create mode 100644 lib/Target/PTX/InstPrinter/PTXInstPrinter.cpp create mode 100644 lib/Target/PTX/InstPrinter/PTXInstPrinter.h create mode 100644 lib/Target/PTX/MCTargetDesc/PTXBaseInfo.h create mode 100644 lib/Target/PTX/PTXAsmPrinter.h delete mode 100644 lib/Target/PTX/PTXCallingConv.td create mode 100644 lib/Target/PTX/PTXFPRoundingModePass.cpp create mode 100644 lib/Target/PTX/PTXInstrLoadStore.td create mode 100644 lib/Target/PTX/PTXMCInstLower.cpp create mode 100644 lib/Target/PTX/PTXParamManager.cpp create mode 100644 lib/Target/PTX/PTXParamManager.h create mode 100644 lib/Target/PTX/PTXRegAlloc.cpp create mode 100644 lib/Target/PTX/PTXSelectionDAGInfo.cpp create mode 100644 lib/Target/PTX/PTXSelectionDAGInfo.h delete mode 100755 lib/Target/PTX/generate-register-td.py create mode 100644 lib/Target/PowerPC/MCTargetDesc/PPCAsmBackend.cpp create mode 100644 lib/Target/PowerPC/MCTargetDesc/PPCBaseInfo.h create mode 100644 lib/Target/PowerPC/MCTargetDesc/PPCFixupKinds.h create mode 100644 lib/Target/PowerPC/MCTargetDesc/PPCMCCodeEmitter.cpp create mode 100644 lib/Target/PowerPC/MCTargetDesc/PPCPredicates.cpp create mode 100644 lib/Target/PowerPC/MCTargetDesc/PPCPredicates.h delete mode 100644 lib/Target/PowerPC/PPCAsmBackend.cpp delete mode 100644 lib/Target/PowerPC/PPCFixupKinds.h delete mode 100644 lib/Target/PowerPC/PPCMCCodeEmitter.cpp delete mode 100644 lib/Target/PowerPC/PPCPredicates.cpp delete mode 100644 lib/Target/PowerPC/PPCPredicates.h delete mode 100644 lib/Target/TargetAsmInfo.cpp delete mode 100644 lib/Target/TargetAsmLexer.cpp create mode 100644 lib/Target/X86/MCTargetDesc/X86AsmBackend.cpp create mode 100644 lib/Target/X86/MCTargetDesc/X86BaseInfo.h create mode 100644 lib/Target/X86/MCTargetDesc/X86FixupKinds.h create mode 100644 lib/Target/X86/MCTargetDesc/X86MCCodeEmitter.cpp create mode 100644 lib/Target/X86/MCTargetDesc/X86MachObjectWriter.cpp delete mode 100644 lib/Target/X86/SSEDomainFix.cpp delete mode 100644 lib/Target/X86/X86AsmBackend.cpp delete mode 100644 lib/Target/X86/X86FixupKinds.h delete mode 100644 lib/Target/X86/X86MCCodeEmitter.cpp delete mode 100644 lib/Target/X86/X86MachObjectWriter.cpp create mode 100644 lib/Target/X86/X86VZeroUpper.cpp (limited to 'lib/Target') diff --git a/lib/Target/ARM/ARM.h b/lib/Target/ARM/ARM.h index 08dc340f8541d..16d0da3b8ac72 100644 --- a/lib/Target/ARM/ARM.h +++ b/lib/Target/ARM/ARM.h @@ -15,7 +15,7 @@ #ifndef TARGET_ARM_H #define TARGET_ARM_H -#include "ARMBaseInfo.h" +#include "MCTargetDesc/ARMBaseInfo.h" #include "MCTargetDesc/ARMMCTargetDesc.h" #include "llvm/Support/DataTypes.h" #include "llvm/Support/ErrorHandling.h" @@ -29,19 +29,7 @@ class ARMBaseTargetMachine; class FunctionPass; class JITCodeEmitter; class MachineInstr; -class MCCodeEmitter; class MCInst; -class MCInstrInfo; -class MCObjectWriter; -class MCSubtargetInfo; -class TargetAsmBackend; -class formatted_raw_ostream; - -MCCodeEmitter *createARMMCCodeEmitter(const MCInstrInfo &MCII, - const MCSubtargetInfo &STI, - MCContext &Ctx); - -TargetAsmBackend *createARMAsmBackend(const Target &, const std::string &); FunctionPass *createARMISelDag(ARMBaseTargetMachine &TM, CodeGenOpt::Level OptLevel); @@ -53,7 +41,6 @@ FunctionPass *createARMLoadStoreOptimizationPass(bool PreAlloc = false); FunctionPass *createARMExpandPseudoPass(); FunctionPass *createARMGlobalMergePass(const TargetLowering* tli); FunctionPass *createARMConstantIslandPass(); -FunctionPass *createNEONMoveFixPass(); FunctionPass *createMLxExpansionPass(); FunctionPass *createThumb2ITBlockPass(); FunctionPass *createThumb2SizeReductionPass(); @@ -61,12 +48,6 @@ FunctionPass *createThumb2SizeReductionPass(); void LowerARMMachineInstrToMCInst(const MachineInstr *MI, MCInst &OutMI, ARMAsmPrinter &AP); -/// createARMMachObjectWriter - Construct an ARM Mach-O object writer. -MCObjectWriter *createARMMachObjectWriter(raw_ostream &OS, - bool Is64Bit, - uint32_t CPUType, - uint32_t CPUSubtype); - } // end namespace llvm; #endif diff --git a/lib/Target/ARM/ARM.td b/lib/Target/ARM/ARM.td index cf333ccd49ba6..5c727ad6e3439 100644 --- a/lib/Target/ARM/ARM.td +++ b/lib/Target/ARM/ARM.td @@ -23,6 +23,9 @@ include "llvm/Target/Target.td" def ModeThumb : SubtargetFeature<"thumb-mode", "InThumbMode", "true", "Thumb mode">; +def ModeNaCl : SubtargetFeature<"nacl-mode", "InNaClMode", "true", + "Native client mode">; + //===----------------------------------------------------------------------===// // ARM Subtarget features. // @@ -85,12 +88,16 @@ def FeatureAvoidPartialCPSR : SubtargetFeature<"avoid-partial-cpsr", /// Some M architectures don't have the DSP extension (v7E-M vs. v7M) def FeatureDSPThumb2 : SubtargetFeature<"t2dsp", "Thumb2DSP", "true", - "Supports v7 DSP instructions in Thumb2.">; + "Supports v7 DSP instructions in Thumb2">; // Multiprocessing extension. def FeatureMP : SubtargetFeature<"mp", "HasMPExtension", "true", "Supports Multiprocessing extension">; +// M-series ISA? +def FeatureMClass : SubtargetFeature<"mclass", "IsMClass", "true", + "Is microcontroller profile ('M' series)">; + // ARM ISAs. def HasV4TOps : SubtargetFeature<"v4t", "HasV4TOps", "true", "Support ARM v4T instructions">; @@ -105,7 +112,7 @@ def HasV6Ops : SubtargetFeature<"v6", "HasV6Ops", "true", [HasV5TEOps]>; def HasV6T2Ops : SubtargetFeature<"v6t2", "HasV6T2Ops", "true", "Support ARM v6t2 instructions", - [HasV6Ops, FeatureThumb2, FeatureDSPThumb2]>; + [HasV6Ops, FeatureThumb2]>; def HasV7Ops : SubtargetFeature<"v7", "HasV7Ops", "true", "Support ARM v7 instructions", [HasV6T2Ops]>; @@ -182,12 +189,14 @@ def : Processor<"mpcore", ARMV6Itineraries, [HasV6Ops, FeatureVFP2, // V6M Processors. def : Processor<"cortex-m0", ARMV6Itineraries, [HasV6Ops, FeatureNoARM, - FeatureDB]>; + FeatureDB, FeatureMClass]>; // V6T2 Processors. -def : Processor<"arm1156t2-s", ARMV6Itineraries, [HasV6T2Ops]>; +def : Processor<"arm1156t2-s", ARMV6Itineraries, [HasV6T2Ops, + FeatureDSPThumb2]>; def : Processor<"arm1156t2f-s", ARMV6Itineraries, [HasV6T2Ops, FeatureVFP2, - FeatureHasSlowFPVMLx]>; + FeatureHasSlowFPVMLx, + FeatureDSPThumb2]>; // V7a Processors. def : Processor<"cortex-a8", CortexA8Itineraries, @@ -203,14 +212,14 @@ def : Processor<"cortex-a9-mp", CortexA9Itineraries, // V7M Processors. def : ProcNoItin<"cortex-m3", [HasV7Ops, FeatureThumb2, FeatureNoARM, FeatureDB, - FeatureHWDiv]>; + FeatureHWDiv, FeatureMClass]>; // V7EM Processors. def : ProcNoItin<"cortex-m4", [HasV7Ops, FeatureThumb2, FeatureNoARM, FeatureDB, FeatureHWDiv, FeatureDSPThumb2, FeatureT2XtPk, FeatureVFP2, - FeatureVFPOnlySP]>; + FeatureVFPOnlySP, FeatureMClass]>; //===----------------------------------------------------------------------===// // Register File Description diff --git a/lib/Target/ARM/ARMAddressingModes.h b/lib/Target/ARM/ARMAddressingModes.h deleted file mode 100644 index 595708fa78811..0000000000000 --- a/lib/Target/ARM/ARMAddressingModes.h +++ /dev/null @@ -1,595 +0,0 @@ -//===- ARMAddressingModes.h - ARM Addressing Modes --------------*- C++ -*-===// -// -// The LLVM Compiler Infrastructure -// -// This file is distributed under the University of Illinois Open Source -// License. See LICENSE.TXT for details. -// -//===----------------------------------------------------------------------===// -// -// This file contains the ARM addressing mode implementation stuff. -// -//===----------------------------------------------------------------------===// - -#ifndef LLVM_TARGET_ARM_ARMADDRESSINGMODES_H -#define LLVM_TARGET_ARM_ARMADDRESSINGMODES_H - -#include "llvm/CodeGen/SelectionDAGNodes.h" -#include "llvm/Support/MathExtras.h" -#include - -namespace llvm { - -/// ARM_AM - ARM Addressing Mode Stuff -namespace ARM_AM { - enum ShiftOpc { - no_shift = 0, - asr, - lsl, - lsr, - ror, - rrx - }; - - enum AddrOpc { - add = '+', sub = '-' - }; - - static inline const char *getAddrOpcStr(AddrOpc Op) { - return Op == sub ? "-" : ""; - } - - static inline const char *getShiftOpcStr(ShiftOpc Op) { - switch (Op) { - default: assert(0 && "Unknown shift opc!"); - case ARM_AM::asr: return "asr"; - case ARM_AM::lsl: return "lsl"; - case ARM_AM::lsr: return "lsr"; - case ARM_AM::ror: return "ror"; - case ARM_AM::rrx: return "rrx"; - } - } - - static inline unsigned getShiftOpcEncoding(ShiftOpc Op) { - switch (Op) { - default: assert(0 && "Unknown shift opc!"); - case ARM_AM::asr: return 2; - case ARM_AM::lsl: return 0; - case ARM_AM::lsr: return 1; - case ARM_AM::ror: return 3; - } - } - - static inline ShiftOpc getShiftOpcForNode(SDValue N) { - switch (N.getOpcode()) { - default: return ARM_AM::no_shift; - case ISD::SHL: return ARM_AM::lsl; - case ISD::SRL: return ARM_AM::lsr; - case ISD::SRA: return ARM_AM::asr; - case ISD::ROTR: return ARM_AM::ror; - //case ISD::ROTL: // Only if imm -> turn into ROTR. - // Can't handle RRX here, because it would require folding a flag into - // the addressing mode. :( This causes us to miss certain things. - //case ARMISD::RRX: return ARM_AM::rrx; - } - } - - enum AMSubMode { - bad_am_submode = 0, - ia, - ib, - da, - db - }; - - static inline const char *getAMSubModeStr(AMSubMode Mode) { - switch (Mode) { - default: assert(0 && "Unknown addressing sub-mode!"); - case ARM_AM::ia: return "ia"; - case ARM_AM::ib: return "ib"; - case ARM_AM::da: return "da"; - case ARM_AM::db: return "db"; - } - } - - /// rotr32 - Rotate a 32-bit unsigned value right by a specified # bits. - /// - static inline unsigned rotr32(unsigned Val, unsigned Amt) { - assert(Amt < 32 && "Invalid rotate amount"); - return (Val >> Amt) | (Val << ((32-Amt)&31)); - } - - /// rotl32 - Rotate a 32-bit unsigned value left by a specified # bits. - /// - static inline unsigned rotl32(unsigned Val, unsigned Amt) { - assert(Amt < 32 && "Invalid rotate amount"); - return (Val << Amt) | (Val >> ((32-Amt)&31)); - } - - //===--------------------------------------------------------------------===// - // Addressing Mode #1: shift_operand with registers - //===--------------------------------------------------------------------===// - // - // This 'addressing mode' is used for arithmetic instructions. It can - // represent things like: - // reg - // reg [asr|lsl|lsr|ror|rrx] reg - // reg [asr|lsl|lsr|ror|rrx] imm - // - // This is stored three operands [rega, regb, opc]. The first is the base - // reg, the second is the shift amount (or reg0 if not present or imm). The - // third operand encodes the shift opcode and the imm if a reg isn't present. - // - static inline unsigned getSORegOpc(ShiftOpc ShOp, unsigned Imm) { - return ShOp | (Imm << 3); - } - static inline unsigned getSORegOffset(unsigned Op) { - return Op >> 3; - } - static inline ShiftOpc getSORegShOp(unsigned Op) { - return (ShiftOpc)(Op & 7); - } - - /// getSOImmValImm - Given an encoded imm field for the reg/imm form, return - /// the 8-bit imm value. - static inline unsigned getSOImmValImm(unsigned Imm) { - return Imm & 0xFF; - } - /// getSOImmValRot - Given an encoded imm field for the reg/imm form, return - /// the rotate amount. - static inline unsigned getSOImmValRot(unsigned Imm) { - return (Imm >> 8) * 2; - } - - /// getSOImmValRotate - Try to handle Imm with an immediate shifter operand, - /// computing the rotate amount to use. If this immediate value cannot be - /// handled with a single shifter-op, determine a good rotate amount that will - /// take a maximal chunk of bits out of the immediate. - static inline unsigned getSOImmValRotate(unsigned Imm) { - // 8-bit (or less) immediates are trivially shifter_operands with a rotate - // of zero. - if ((Imm & ~255U) == 0) return 0; - - // Use CTZ to compute the rotate amount. - unsigned TZ = CountTrailingZeros_32(Imm); - - // Rotate amount must be even. Something like 0x200 must be rotated 8 bits, - // not 9. - unsigned RotAmt = TZ & ~1; - - // If we can handle this spread, return it. - if ((rotr32(Imm, RotAmt) & ~255U) == 0) - return (32-RotAmt)&31; // HW rotates right, not left. - - // For values like 0xF000000F, we should ignore the low 6 bits, then - // retry the hunt. - if (Imm & 63U) { - unsigned TZ2 = CountTrailingZeros_32(Imm & ~63U); - unsigned RotAmt2 = TZ2 & ~1; - if ((rotr32(Imm, RotAmt2) & ~255U) == 0) - return (32-RotAmt2)&31; // HW rotates right, not left. - } - - // Otherwise, we have no way to cover this span of bits with a single - // shifter_op immediate. Return a chunk of bits that will be useful to - // handle. - return (32-RotAmt)&31; // HW rotates right, not left. - } - - /// getSOImmVal - Given a 32-bit immediate, if it is something that can fit - /// into an shifter_operand immediate operand, return the 12-bit encoding for - /// it. If not, return -1. - static inline int getSOImmVal(unsigned Arg) { - // 8-bit (or less) immediates are trivially shifter_operands with a rotate - // of zero. - if ((Arg & ~255U) == 0) return Arg; - - unsigned RotAmt = getSOImmValRotate(Arg); - - // If this cannot be handled with a single shifter_op, bail out. - if (rotr32(~255U, RotAmt) & Arg) - return -1; - - // Encode this correctly. - return rotl32(Arg, RotAmt) | ((RotAmt>>1) << 8); - } - - /// isSOImmTwoPartVal - Return true if the specified value can be obtained by - /// or'ing together two SOImmVal's. - static inline bool isSOImmTwoPartVal(unsigned V) { - // If this can be handled with a single shifter_op, bail out. - V = rotr32(~255U, getSOImmValRotate(V)) & V; - if (V == 0) - return false; - - // If this can be handled with two shifter_op's, accept. - V = rotr32(~255U, getSOImmValRotate(V)) & V; - return V == 0; - } - - /// getSOImmTwoPartFirst - If V is a value that satisfies isSOImmTwoPartVal, - /// return the first chunk of it. - static inline unsigned getSOImmTwoPartFirst(unsigned V) { - return rotr32(255U, getSOImmValRotate(V)) & V; - } - - /// getSOImmTwoPartSecond - If V is a value that satisfies isSOImmTwoPartVal, - /// return the second chunk of it. - static inline unsigned getSOImmTwoPartSecond(unsigned V) { - // Mask out the first hunk. - V = rotr32(~255U, getSOImmValRotate(V)) & V; - - // Take what's left. - assert(V == (rotr32(255U, getSOImmValRotate(V)) & V)); - return V; - } - - /// getThumbImmValShift - Try to handle Imm with a 8-bit immediate followed - /// by a left shift. Returns the shift amount to use. - static inline unsigned getThumbImmValShift(unsigned Imm) { - // 8-bit (or less) immediates are trivially immediate operand with a shift - // of zero. - if ((Imm & ~255U) == 0) return 0; - - // Use CTZ to compute the shift amount. - return CountTrailingZeros_32(Imm); - } - - /// isThumbImmShiftedVal - Return true if the specified value can be obtained - /// by left shifting a 8-bit immediate. - static inline bool isThumbImmShiftedVal(unsigned V) { - // If this can be handled with - V = (~255U << getThumbImmValShift(V)) & V; - return V == 0; - } - - /// getThumbImm16ValShift - Try to handle Imm with a 16-bit immediate followed - /// by a left shift. Returns the shift amount to use. - static inline unsigned getThumbImm16ValShift(unsigned Imm) { - // 16-bit (or less) immediates are trivially immediate operand with a shift - // of zero. - if ((Imm & ~65535U) == 0) return 0; - - // Use CTZ to compute the shift amount. - return CountTrailingZeros_32(Imm); - } - - /// isThumbImm16ShiftedVal - Return true if the specified value can be - /// obtained by left shifting a 16-bit immediate. - static inline bool isThumbImm16ShiftedVal(unsigned V) { - // If this can be handled with - V = (~65535U << getThumbImm16ValShift(V)) & V; - return V == 0; - } - - /// getThumbImmNonShiftedVal - If V is a value that satisfies - /// isThumbImmShiftedVal, return the non-shiftd value. - static inline unsigned getThumbImmNonShiftedVal(unsigned V) { - return V >> getThumbImmValShift(V); - } - - - /// getT2SOImmValSplat - Return the 12-bit encoded representation - /// if the specified value can be obtained by splatting the low 8 bits - /// into every other byte or every byte of a 32-bit value. i.e., - /// 00000000 00000000 00000000 abcdefgh control = 0 - /// 00000000 abcdefgh 00000000 abcdefgh control = 1 - /// abcdefgh 00000000 abcdefgh 00000000 control = 2 - /// abcdefgh abcdefgh abcdefgh abcdefgh control = 3 - /// Return -1 if none of the above apply. - /// See ARM Reference Manual A6.3.2. - static inline int getT2SOImmValSplatVal(unsigned V) { - unsigned u, Vs, Imm; - // control = 0 - if ((V & 0xffffff00) == 0) - return V; - - // If the value is zeroes in the first byte, just shift those off - Vs = ((V & 0xff) == 0) ? V >> 8 : V; - // Any passing value only has 8 bits of payload, splatted across the word - Imm = Vs & 0xff; - // Likewise, any passing values have the payload splatted into the 3rd byte - u = Imm | (Imm << 16); - - // control = 1 or 2 - if (Vs == u) - return (((Vs == V) ? 1 : 2) << 8) | Imm; - - // control = 3 - if (Vs == (u | (u << 8))) - return (3 << 8) | Imm; - - return -1; - } - - /// getT2SOImmValRotateVal - Return the 12-bit encoded representation if the - /// specified value is a rotated 8-bit value. Return -1 if no rotation - /// encoding is possible. - /// See ARM Reference Manual A6.3.2. - static inline int getT2SOImmValRotateVal(unsigned V) { - unsigned RotAmt = CountLeadingZeros_32(V); - if (RotAmt >= 24) - return -1; - - // If 'Arg' can be handled with a single shifter_op return the value. - if ((rotr32(0xff000000U, RotAmt) & V) == V) - return (rotr32(V, 24 - RotAmt) & 0x7f) | ((RotAmt + 8) << 7); - - return -1; - } - - /// getT2SOImmVal - Given a 32-bit immediate, if it is something that can fit - /// into a Thumb-2 shifter_operand immediate operand, return the 12-bit - /// encoding for it. If not, return -1. - /// See ARM Reference Manual A6.3.2. - static inline int getT2SOImmVal(unsigned Arg) { - // If 'Arg' is an 8-bit splat, then get the encoded value. - int Splat = getT2SOImmValSplatVal(Arg); - if (Splat != -1) - return Splat; - - // If 'Arg' can be handled with a single shifter_op return the value. - int Rot = getT2SOImmValRotateVal(Arg); - if (Rot != -1) - return Rot; - - return -1; - } - - static inline unsigned getT2SOImmValRotate(unsigned V) { - if ((V & ~255U) == 0) return 0; - // Use CTZ to compute the rotate amount. - unsigned RotAmt = CountTrailingZeros_32(V); - return (32 - RotAmt) & 31; - } - - static inline bool isT2SOImmTwoPartVal (unsigned Imm) { - unsigned V = Imm; - // Passing values can be any combination of splat values and shifter - // values. If this can be handled with a single shifter or splat, bail - // out. Those should be handled directly, not with a two-part val. - if (getT2SOImmValSplatVal(V) != -1) - return false; - V = rotr32 (~255U, getT2SOImmValRotate(V)) & V; - if (V == 0) - return false; - - // If this can be handled as an immediate, accept. - if (getT2SOImmVal(V) != -1) return true; - - // Likewise, try masking out a splat value first. - V = Imm; - if (getT2SOImmValSplatVal(V & 0xff00ff00U) != -1) - V &= ~0xff00ff00U; - else if (getT2SOImmValSplatVal(V & 0x00ff00ffU) != -1) - V &= ~0x00ff00ffU; - // If what's left can be handled as an immediate, accept. - if (getT2SOImmVal(V) != -1) return true; - - // Otherwise, do not accept. - return false; - } - - static inline unsigned getT2SOImmTwoPartFirst(unsigned Imm) { - assert (isT2SOImmTwoPartVal(Imm) && - "Immedate cannot be encoded as two part immediate!"); - // Try a shifter operand as one part - unsigned V = rotr32 (~255, getT2SOImmValRotate(Imm)) & Imm; - // If the rest is encodable as an immediate, then return it. - if (getT2SOImmVal(V) != -1) return V; - - // Try masking out a splat value first. - if (getT2SOImmValSplatVal(Imm & 0xff00ff00U) != -1) - return Imm & 0xff00ff00U; - - // The other splat is all that's left as an option. - assert (getT2SOImmValSplatVal(Imm & 0x00ff00ffU) != -1); - return Imm & 0x00ff00ffU; - } - - static inline unsigned getT2SOImmTwoPartSecond(unsigned Imm) { - // Mask out the first hunk - Imm ^= getT2SOImmTwoPartFirst(Imm); - // Return what's left - assert (getT2SOImmVal(Imm) != -1 && - "Unable to encode second part of T2 two part SO immediate"); - return Imm; - } - - - //===--------------------------------------------------------------------===// - // Addressing Mode #2 - //===--------------------------------------------------------------------===// - // - // This is used for most simple load/store instructions. - // - // addrmode2 := reg +/- reg shop imm - // addrmode2 := reg +/- imm12 - // - // The first operand is always a Reg. The second operand is a reg if in - // reg/reg form, otherwise it's reg#0. The third field encodes the operation - // in bit 12, the immediate in bits 0-11, and the shift op in 13-15. The - // fourth operand 16-17 encodes the index mode. - // - // If this addressing mode is a frame index (before prolog/epilog insertion - // and code rewriting), this operand will have the form: FI#, reg0, - // with no shift amount for the frame offset. - // - static inline unsigned getAM2Opc(AddrOpc Opc, unsigned Imm12, ShiftOpc SO, - unsigned IdxMode = 0) { - assert(Imm12 < (1 << 12) && "Imm too large!"); - bool isSub = Opc == sub; - return Imm12 | ((int)isSub << 12) | (SO << 13) | (IdxMode << 16) ; - } - static inline unsigned getAM2Offset(unsigned AM2Opc) { - return AM2Opc & ((1 << 12)-1); - } - static inline AddrOpc getAM2Op(unsigned AM2Opc) { - return ((AM2Opc >> 12) & 1) ? sub : add; - } - static inline ShiftOpc getAM2ShiftOpc(unsigned AM2Opc) { - return (ShiftOpc)((AM2Opc >> 13) & 7); - } - static inline unsigned getAM2IdxMode(unsigned AM2Opc) { - return (AM2Opc >> 16); - } - - - //===--------------------------------------------------------------------===// - // Addressing Mode #3 - //===--------------------------------------------------------------------===// - // - // This is used for sign-extending loads, and load/store-pair instructions. - // - // addrmode3 := reg +/- reg - // addrmode3 := reg +/- imm8 - // - // The first operand is always a Reg. The second operand is a reg if in - // reg/reg form, otherwise it's reg#0. The third field encodes the operation - // in bit 8, the immediate in bits 0-7. The fourth operand 9-10 encodes the - // index mode. - - /// getAM3Opc - This function encodes the addrmode3 opc field. - static inline unsigned getAM3Opc(AddrOpc Opc, unsigned char Offset, - unsigned IdxMode = 0) { - bool isSub = Opc == sub; - return ((int)isSub << 8) | Offset | (IdxMode << 9); - } - static inline unsigned char getAM3Offset(unsigned AM3Opc) { - return AM3Opc & 0xFF; - } - static inline AddrOpc getAM3Op(unsigned AM3Opc) { - return ((AM3Opc >> 8) & 1) ? sub : add; - } - static inline unsigned getAM3IdxMode(unsigned AM3Opc) { - return (AM3Opc >> 9); - } - - //===--------------------------------------------------------------------===// - // Addressing Mode #4 - //===--------------------------------------------------------------------===// - // - // This is used for load / store multiple instructions. - // - // addrmode4 := reg, - // - // The four modes are: - // IA - Increment after - // IB - Increment before - // DA - Decrement after - // DB - Decrement before - // For VFP instructions, only the IA and DB modes are valid. - - static inline AMSubMode getAM4SubMode(unsigned Mode) { - return (AMSubMode)(Mode & 0x7); - } - - static inline unsigned getAM4ModeImm(AMSubMode SubMode) { - return (int)SubMode; - } - - //===--------------------------------------------------------------------===// - // Addressing Mode #5 - //===--------------------------------------------------------------------===// - // - // This is used for coprocessor instructions, such as FP load/stores. - // - // addrmode5 := reg +/- imm8*4 - // - // The first operand is always a Reg. The second operand encodes the - // operation in bit 8 and the immediate in bits 0-7. - - /// getAM5Opc - This function encodes the addrmode5 opc field. - static inline unsigned getAM5Opc(AddrOpc Opc, unsigned char Offset) { - bool isSub = Opc == sub; - return ((int)isSub << 8) | Offset; - } - static inline unsigned char getAM5Offset(unsigned AM5Opc) { - return AM5Opc & 0xFF; - } - static inline AddrOpc getAM5Op(unsigned AM5Opc) { - return ((AM5Opc >> 8) & 1) ? sub : add; - } - - //===--------------------------------------------------------------------===// - // Addressing Mode #6 - //===--------------------------------------------------------------------===// - // - // This is used for NEON load / store instructions. - // - // addrmode6 := reg with optional alignment - // - // This is stored in two operands [regaddr, align]. The first is the - // address register. The second operand is the value of the alignment - // specifier in bytes or zero if no explicit alignment. - // Valid alignments depend on the specific instruction. - - //===--------------------------------------------------------------------===// - // NEON Modified Immediates - //===--------------------------------------------------------------------===// - // - // Several NEON instructions (e.g., VMOV) take a "modified immediate" - // vector operand, where a small immediate encoded in the instruction - // specifies a full NEON vector value. These modified immediates are - // represented here as encoded integers. The low 8 bits hold the immediate - // value; bit 12 holds the "Op" field of the instruction, and bits 11-8 hold - // the "Cmode" field of the instruction. The interfaces below treat the - // Op and Cmode values as a single 5-bit value. - - static inline unsigned createNEONModImm(unsigned OpCmode, unsigned Val) { - return (OpCmode << 8) | Val; - } - static inline unsigned getNEONModImmOpCmode(unsigned ModImm) { - return (ModImm >> 8) & 0x1f; - } - static inline unsigned getNEONModImmVal(unsigned ModImm) { - return ModImm & 0xff; - } - - /// decodeNEONModImm - Decode a NEON modified immediate value into the - /// element value and the element size in bits. (If the element size is - /// smaller than the vector, it is splatted into all the elements.) - static inline uint64_t decodeNEONModImm(unsigned ModImm, unsigned &EltBits) { - unsigned OpCmode = getNEONModImmOpCmode(ModImm); - unsigned Imm8 = getNEONModImmVal(ModImm); - uint64_t Val = 0; - - if (OpCmode == 0xe) { - // 8-bit vector elements - Val = Imm8; - EltBits = 8; - } else if ((OpCmode & 0xc) == 0x8) { - // 16-bit vector elements - unsigned ByteNum = (OpCmode & 0x6) >> 1; - Val = Imm8 << (8 * ByteNum); - EltBits = 16; - } else if ((OpCmode & 0x8) == 0) { - // 32-bit vector elements, zero with one byte set - unsigned ByteNum = (OpCmode & 0x6) >> 1; - Val = Imm8 << (8 * ByteNum); - EltBits = 32; - } else if ((OpCmode & 0xe) == 0xc) { - // 32-bit vector elements, one byte with low bits set - unsigned ByteNum = 1 + (OpCmode & 0x1); - Val = (Imm8 << (8 * ByteNum)) | (0xffff >> (8 * (2 - ByteNum))); - EltBits = 32; - } else if (OpCmode == 0x1e) { - // 64-bit vector elements - for (unsigned ByteNum = 0; ByteNum < 8; ++ByteNum) { - if ((ModImm >> ByteNum) & 1) - Val |= (uint64_t)0xff << (8 * ByteNum); - } - EltBits = 64; - } else { - assert(false && "Unsupported NEON immediate"); - } - return Val; - } - - AMSubMode getLoadStoreMultipleSubMode(int Opcode); - -} // end namespace ARM_AM -} // end namespace llvm - -#endif - diff --git a/lib/Target/ARM/ARMAsmBackend.cpp b/lib/Target/ARM/ARMAsmBackend.cpp deleted file mode 100644 index 5e438a9767324..0000000000000 --- a/lib/Target/ARM/ARMAsmBackend.cpp +++ /dev/null @@ -1,516 +0,0 @@ -//===-- ARMAsmBackend.cpp - ARM Assembler Backend -------------------------===// -// -// The LLVM Compiler Infrastructure -// -// This file is distributed under the University of Illinois Open Source -// License. See LICENSE.TXT for details. -// -//===----------------------------------------------------------------------===// - -#include "ARM.h" -#include "ARMAddressingModes.h" -#include "ARMFixupKinds.h" -#include "llvm/ADT/Twine.h" -#include "llvm/MC/MCAssembler.h" -#include "llvm/MC/MCDirectives.h" -#include "llvm/MC/MCELFObjectWriter.h" -#include "llvm/MC/MCExpr.h" -#include "llvm/MC/MCMachObjectWriter.h" -#include "llvm/MC/MCObjectWriter.h" -#include "llvm/MC/MCSectionELF.h" -#include "llvm/MC/MCSectionMachO.h" -#include "llvm/Object/MachOFormat.h" -#include "llvm/Support/ELF.h" -#include "llvm/Support/ErrorHandling.h" -#include "llvm/Support/raw_ostream.h" -#include "llvm/Target/TargetAsmBackend.h" -#include "llvm/Target/TargetRegistry.h" -using namespace llvm; - -namespace { -class ARMELFObjectWriter : public MCELFObjectTargetWriter { -public: - ARMELFObjectWriter(Triple::OSType OSType) - : MCELFObjectTargetWriter(/*Is64Bit*/ false, OSType, ELF::EM_ARM, - /*HasRelocationAddend*/ false) {} -}; - -class ARMAsmBackend : public TargetAsmBackend { - bool isThumbMode; // Currently emitting Thumb code. -public: - ARMAsmBackend(const Target &T) : TargetAsmBackend(), isThumbMode(false) {} - - unsigned getNumFixupKinds() const { return ARM::NumTargetFixupKinds; } - - const MCFixupKindInfo &getFixupKindInfo(MCFixupKind Kind) const { - const static MCFixupKindInfo Infos[ARM::NumTargetFixupKinds] = { -// This table *must* be in the order that the fixup_* kinds are defined in -// ARMFixupKinds.h. -// -// Name Offset (bits) Size (bits) Flags -{ "fixup_arm_ldst_pcrel_12", 1, 24, MCFixupKindInfo::FKF_IsPCRel }, -{ "fixup_t2_ldst_pcrel_12", 0, 32, MCFixupKindInfo::FKF_IsPCRel | - MCFixupKindInfo::FKF_IsAlignedDownTo32Bits}, -{ "fixup_arm_pcrel_10", 1, 24, MCFixupKindInfo::FKF_IsPCRel }, -{ "fixup_t2_pcrel_10", 0, 32, MCFixupKindInfo::FKF_IsPCRel | - MCFixupKindInfo::FKF_IsAlignedDownTo32Bits}, -{ "fixup_thumb_adr_pcrel_10",0, 8, MCFixupKindInfo::FKF_IsPCRel | - MCFixupKindInfo::FKF_IsAlignedDownTo32Bits}, -{ "fixup_arm_adr_pcrel_12", 1, 24, MCFixupKindInfo::FKF_IsPCRel }, -{ "fixup_t2_adr_pcrel_12", 0, 32, MCFixupKindInfo::FKF_IsPCRel | - MCFixupKindInfo::FKF_IsAlignedDownTo32Bits}, -{ "fixup_arm_condbranch", 0, 24, MCFixupKindInfo::FKF_IsPCRel }, -{ "fixup_arm_uncondbranch", 0, 24, MCFixupKindInfo::FKF_IsPCRel }, -{ "fixup_t2_condbranch", 0, 32, MCFixupKindInfo::FKF_IsPCRel }, -{ "fixup_t2_uncondbranch", 0, 32, MCFixupKindInfo::FKF_IsPCRel }, -{ "fixup_arm_thumb_br", 0, 16, MCFixupKindInfo::FKF_IsPCRel }, -{ "fixup_arm_thumb_bl", 0, 32, MCFixupKindInfo::FKF_IsPCRel }, -{ "fixup_arm_thumb_blx", 7, 21, MCFixupKindInfo::FKF_IsPCRel }, -{ "fixup_arm_thumb_cb", 0, 16, MCFixupKindInfo::FKF_IsPCRel }, -{ "fixup_arm_thumb_cp", 1, 8, MCFixupKindInfo::FKF_IsPCRel }, -{ "fixup_arm_thumb_bcc", 0, 8, MCFixupKindInfo::FKF_IsPCRel }, -// movw / movt: 16-bits immediate but scattered into two chunks 0 - 12, 16 - 19. -{ "fixup_arm_movt_hi16", 0, 20, 0 }, -{ "fixup_arm_movw_lo16", 0, 20, 0 }, -{ "fixup_t2_movt_hi16", 0, 20, 0 }, -{ "fixup_t2_movw_lo16", 0, 20, 0 }, -{ "fixup_arm_movt_hi16_pcrel", 0, 20, MCFixupKindInfo::FKF_IsPCRel }, -{ "fixup_arm_movw_lo16_pcrel", 0, 20, MCFixupKindInfo::FKF_IsPCRel }, -{ "fixup_t2_movt_hi16_pcrel", 0, 20, MCFixupKindInfo::FKF_IsPCRel }, -{ "fixup_t2_movw_lo16_pcrel", 0, 20, MCFixupKindInfo::FKF_IsPCRel }, - }; - - if (Kind < FirstTargetFixupKind) - return TargetAsmBackend::getFixupKindInfo(Kind); - - assert(unsigned(Kind - FirstTargetFixupKind) < getNumFixupKinds() && - "Invalid kind!"); - return Infos[Kind - FirstTargetFixupKind]; - } - - bool MayNeedRelaxation(const MCInst &Inst) const; - - void RelaxInstruction(const MCInst &Inst, MCInst &Res) const; - - bool WriteNopData(uint64_t Count, MCObjectWriter *OW) const; - - void HandleAssemblerFlag(MCAssemblerFlag Flag) { - switch (Flag) { - default: break; - case MCAF_Code16: - setIsThumb(true); - break; - case MCAF_Code32: - setIsThumb(false); - break; - } - } - - unsigned getPointerSize() const { return 4; } - bool isThumb() const { return isThumbMode; } - void setIsThumb(bool it) { isThumbMode = it; } -}; -} // end anonymous namespace - -bool ARMAsmBackend::MayNeedRelaxation(const MCInst &Inst) const { - // FIXME: Thumb targets, different move constant targets.. - return false; -} - -void ARMAsmBackend::RelaxInstruction(const MCInst &Inst, MCInst &Res) const { - assert(0 && "ARMAsmBackend::RelaxInstruction() unimplemented"); - return; -} - -bool ARMAsmBackend::WriteNopData(uint64_t Count, MCObjectWriter *OW) const { - if (isThumb()) { - // FIXME: 0xbf00 is the ARMv7 value. For v6 and before, we'll need to - // use 0x46c0 (which is a 'mov r8, r8' insn). - uint64_t NumNops = Count / 2; - for (uint64_t i = 0; i != NumNops; ++i) - OW->Write16(0xbf00); - if (Count & 1) - OW->Write8(0); - return true; - } - // ARM mode - uint64_t NumNops = Count / 4; - for (uint64_t i = 0; i != NumNops; ++i) - OW->Write32(0xe1a00000); - switch (Count % 4) { - default: break; // No leftover bytes to write - case 1: OW->Write8(0); break; - case 2: OW->Write16(0); break; - case 3: OW->Write16(0); OW->Write8(0xa0); break; - } - - return true; -} - -static unsigned adjustFixupValue(unsigned Kind, uint64_t Value) { - switch (Kind) { - default: - llvm_unreachable("Unknown fixup kind!"); - case FK_Data_1: - case FK_Data_2: - case FK_Data_4: - return Value; - case ARM::fixup_arm_movt_hi16: - Value >>= 16; - // Fallthrough - case ARM::fixup_arm_movw_lo16: - case ARM::fixup_arm_movt_hi16_pcrel: - case ARM::fixup_arm_movw_lo16_pcrel: { - unsigned Hi4 = (Value & 0xF000) >> 12; - unsigned Lo12 = Value & 0x0FFF; - assert ((((int64_t)Value) >= -0x8000) && (((int64_t)Value) <= 0x7fff) && - "Out of range pc-relative fixup value!"); - // inst{19-16} = Hi4; - // inst{11-0} = Lo12; - Value = (Hi4 << 16) | (Lo12); - return Value; - } - case ARM::fixup_t2_movt_hi16: - Value >>= 16; - // Fallthrough - case ARM::fixup_t2_movw_lo16: - case ARM::fixup_t2_movt_hi16_pcrel: //FIXME: Shouldn't this be shifted like - // the other hi16 fixup? - case ARM::fixup_t2_movw_lo16_pcrel: { - unsigned Hi4 = (Value & 0xF000) >> 12; - unsigned i = (Value & 0x800) >> 11; - unsigned Mid3 = (Value & 0x700) >> 8; - unsigned Lo8 = Value & 0x0FF; - // inst{19-16} = Hi4; - // inst{26} = i; - // inst{14-12} = Mid3; - // inst{7-0} = Lo8; - // The value comes in as the whole thing, not just the portion required - // for this fixup, so we need to mask off the bits not handled by this - // portion (lo vs. hi). - Value &= 0xffff; - Value = (Hi4 << 16) | (i << 26) | (Mid3 << 12) | (Lo8); - uint64_t swapped = (Value & 0xFFFF0000) >> 16; - swapped |= (Value & 0x0000FFFF) << 16; - return swapped; - } - case ARM::fixup_arm_ldst_pcrel_12: - // ARM PC-relative values are offset by 8. - Value -= 4; - // FALLTHROUGH - case ARM::fixup_t2_ldst_pcrel_12: { - // Offset by 4, adjusted by two due to the half-word ordering of thumb. - Value -= 4; - bool isAdd = true; - if ((int64_t)Value < 0) { - Value = -Value; - isAdd = false; - } - assert ((Value < 4096) && "Out of range pc-relative fixup value!"); - Value |= isAdd << 23; - - // Same addressing mode as fixup_arm_pcrel_10, - // but with 16-bit halfwords swapped. - if (Kind == ARM::fixup_t2_ldst_pcrel_12) { - uint64_t swapped = (Value & 0xFFFF0000) >> 16; - swapped |= (Value & 0x0000FFFF) << 16; - return swapped; - } - - return Value; - } - case ARM::fixup_thumb_adr_pcrel_10: - return ((Value - 4) >> 2) & 0xff; - case ARM::fixup_arm_adr_pcrel_12: { - // ARM PC-relative values are offset by 8. - Value -= 8; - unsigned opc = 4; // bits {24-21}. Default to add: 0b0100 - if ((int64_t)Value < 0) { - Value = -Value; - opc = 2; // 0b0010 - } - assert(ARM_AM::getSOImmVal(Value) != -1 && - "Out of range pc-relative fixup value!"); - // Encode the immediate and shift the opcode into place. - return ARM_AM::getSOImmVal(Value) | (opc << 21); - } - - case ARM::fixup_t2_adr_pcrel_12: { - Value -= 4; - unsigned opc = 0; - if ((int64_t)Value < 0) { - Value = -Value; - opc = 5; - } - - uint32_t out = (opc << 21); - out |= (Value & 0x800) << 15; - out |= (Value & 0x700) << 4; - out |= (Value & 0x0FF); - - uint64_t swapped = (out & 0xFFFF0000) >> 16; - swapped |= (out & 0x0000FFFF) << 16; - return swapped; - } - - case ARM::fixup_arm_condbranch: - case ARM::fixup_arm_uncondbranch: - // These values don't encode the low two bits since they're always zero. - // Offset by 8 just as above. - return 0xffffff & ((Value - 8) >> 2); - case ARM::fixup_t2_uncondbranch: { - Value = Value - 4; - Value >>= 1; // Low bit is not encoded. - - uint32_t out = 0; - bool I = Value & 0x800000; - bool J1 = Value & 0x400000; - bool J2 = Value & 0x200000; - J1 ^= I; - J2 ^= I; - - out |= I << 26; // S bit - out |= !J1 << 13; // J1 bit - out |= !J2 << 11; // J2 bit - out |= (Value & 0x1FF800) << 5; // imm6 field - out |= (Value & 0x0007FF); // imm11 field - - uint64_t swapped = (out & 0xFFFF0000) >> 16; - swapped |= (out & 0x0000FFFF) << 16; - return swapped; - } - case ARM::fixup_t2_condbranch: { - Value = Value - 4; - Value >>= 1; // Low bit is not encoded. - - uint64_t out = 0; - out |= (Value & 0x80000) << 7; // S bit - out |= (Value & 0x40000) >> 7; // J2 bit - out |= (Value & 0x20000) >> 4; // J1 bit - out |= (Value & 0x1F800) << 5; // imm6 field - out |= (Value & 0x007FF); // imm11 field - - uint32_t swapped = (out & 0xFFFF0000) >> 16; - swapped |= (out & 0x0000FFFF) << 16; - return swapped; - } - case ARM::fixup_arm_thumb_bl: { - // The value doesn't encode the low bit (always zero) and is offset by - // four. The value is encoded into disjoint bit positions in the destination - // opcode. x = unchanged, I = immediate value bit, S = sign extension bit - // - // BL: xxxxxSIIIIIIIIII xxxxxIIIIIIIIIII - // - // Note that the halfwords are stored high first, low second; so we need - // to transpose the fixup value here to map properly. - unsigned isNeg = (int64_t(Value - 4) < 0) ? 1 : 0; - uint32_t Binary = 0; - Value = 0x3fffff & ((Value - 4) >> 1); - Binary = (Value & 0x7ff) << 16; // Low imm11 value. - Binary |= (Value & 0x1ffc00) >> 11; // High imm10 value. - Binary |= isNeg << 10; // Sign bit. - return Binary; - } - case ARM::fixup_arm_thumb_blx: { - // The value doesn't encode the low two bits (always zero) and is offset by - // four (see fixup_arm_thumb_cp). The value is encoded into disjoint bit - // positions in the destination opcode. x = unchanged, I = immediate value - // bit, S = sign extension bit, 0 = zero. - // - // BLX: xxxxxSIIIIIIIIII xxxxxIIIIIIIIII0 - // - // Note that the halfwords are stored high first, low second; so we need - // to transpose the fixup value here to map properly. - unsigned isNeg = (int64_t(Value-4) < 0) ? 1 : 0; - uint32_t Binary = 0; - Value = 0xfffff & ((Value - 2) >> 2); - Binary = (Value & 0x3ff) << 17; // Low imm10L value. - Binary |= (Value & 0xffc00) >> 10; // High imm10H value. - Binary |= isNeg << 10; // Sign bit. - return Binary; - } - case ARM::fixup_arm_thumb_cp: - // Offset by 4, and don't encode the low two bits. Two bytes of that - // 'off by 4' is implicitly handled by the half-word ordering of the - // Thumb encoding, so we only need to adjust by 2 here. - return ((Value - 2) >> 2) & 0xff; - case ARM::fixup_arm_thumb_cb: { - // Offset by 4 and don't encode the lower bit, which is always 0. - uint32_t Binary = (Value - 4) >> 1; - return ((Binary & 0x20) << 4) | ((Binary & 0x1f) << 3); - } - case ARM::fixup_arm_thumb_br: - // Offset by 4 and don't encode the lower bit, which is always 0. - return ((Value - 4) >> 1) & 0x7ff; - case ARM::fixup_arm_thumb_bcc: - // Offset by 4 and don't encode the lower bit, which is always 0. - return ((Value - 4) >> 1) & 0xff; - case ARM::fixup_arm_pcrel_10: - Value = Value - 4; // ARM fixups offset by an additional word and don't - // need to adjust for the half-word ordering. - // Fall through. - case ARM::fixup_t2_pcrel_10: { - // Offset by 4, adjusted by two due to the half-word ordering of thumb. - Value = Value - 4; - bool isAdd = true; - if ((int64_t)Value < 0) { - Value = -Value; - isAdd = false; - } - // These values don't encode the low two bits since they're always zero. - Value >>= 2; - assert ((Value < 256) && "Out of range pc-relative fixup value!"); - Value |= isAdd << 23; - - // Same addressing mode as fixup_arm_pcrel_10, - // but with 16-bit halfwords swapped. - if (Kind == ARM::fixup_t2_pcrel_10) { - uint32_t swapped = (Value & 0xFFFF0000) >> 16; - swapped |= (Value & 0x0000FFFF) << 16; - return swapped; - } - - return Value; - } - } -} - -namespace { - -// FIXME: This should be in a separate file. -// ELF is an ELF of course... -class ELFARMAsmBackend : public ARMAsmBackend { -public: - Triple::OSType OSType; - ELFARMAsmBackend(const Target &T, Triple::OSType _OSType) - : ARMAsmBackend(T), OSType(_OSType) { } - - void ApplyFixup(const MCFixup &Fixup, char *Data, unsigned DataSize, - uint64_t Value) const; - - MCObjectWriter *createObjectWriter(raw_ostream &OS) const { - return createELFObjectWriter(new ARMELFObjectWriter(OSType), OS, - /*IsLittleEndian*/ true); - } -}; - -// FIXME: Raise this to share code between Darwin and ELF. -void ELFARMAsmBackend::ApplyFixup(const MCFixup &Fixup, char *Data, - unsigned DataSize, uint64_t Value) const { - unsigned NumBytes = 4; // FIXME: 2 for Thumb - Value = adjustFixupValue(Fixup.getKind(), Value); - if (!Value) return; // Doesn't change encoding. - - unsigned Offset = Fixup.getOffset(); - - // For each byte of the fragment that the fixup touches, mask in the bits from - // the fixup value. The Value has been "split up" into the appropriate - // bitfields above. - for (unsigned i = 0; i != NumBytes; ++i) - Data[Offset + i] |= uint8_t((Value >> (i * 8)) & 0xff); -} - -// FIXME: This should be in a separate file. -class DarwinARMAsmBackend : public ARMAsmBackend { -public: - const object::mach::CPUSubtypeARM Subtype; - DarwinARMAsmBackend(const Target &T, object::mach::CPUSubtypeARM st) - : ARMAsmBackend(T), Subtype(st) { } - - MCObjectWriter *createObjectWriter(raw_ostream &OS) const { - return createARMMachObjectWriter(OS, /*Is64Bit=*/false, - object::mach::CTM_ARM, - Subtype); - } - - void ApplyFixup(const MCFixup &Fixup, char *Data, unsigned DataSize, - uint64_t Value) const; - - virtual bool doesSectionRequireSymbols(const MCSection &Section) const { - return false; - } -}; - -/// getFixupKindNumBytes - The number of bytes the fixup may change. -static unsigned getFixupKindNumBytes(unsigned Kind) { - switch (Kind) { - default: - llvm_unreachable("Unknown fixup kind!"); - - case FK_Data_1: - case ARM::fixup_arm_thumb_bcc: - case ARM::fixup_arm_thumb_cp: - case ARM::fixup_thumb_adr_pcrel_10: - return 1; - - case FK_Data_2: - case ARM::fixup_arm_thumb_br: - case ARM::fixup_arm_thumb_cb: - return 2; - - case ARM::fixup_arm_ldst_pcrel_12: - case ARM::fixup_arm_pcrel_10: - case ARM::fixup_arm_adr_pcrel_12: - case ARM::fixup_arm_condbranch: - case ARM::fixup_arm_uncondbranch: - return 3; - - case FK_Data_4: - case ARM::fixup_t2_ldst_pcrel_12: - case ARM::fixup_t2_condbranch: - case ARM::fixup_t2_uncondbranch: - case ARM::fixup_t2_pcrel_10: - case ARM::fixup_t2_adr_pcrel_12: - case ARM::fixup_arm_thumb_bl: - case ARM::fixup_arm_thumb_blx: - case ARM::fixup_arm_movt_hi16: - case ARM::fixup_arm_movw_lo16: - case ARM::fixup_arm_movt_hi16_pcrel: - case ARM::fixup_arm_movw_lo16_pcrel: - case ARM::fixup_t2_movt_hi16: - case ARM::fixup_t2_movw_lo16: - case ARM::fixup_t2_movt_hi16_pcrel: - case ARM::fixup_t2_movw_lo16_pcrel: - return 4; - } -} - -void DarwinARMAsmBackend::ApplyFixup(const MCFixup &Fixup, char *Data, - unsigned DataSize, uint64_t Value) const { - unsigned NumBytes = getFixupKindNumBytes(Fixup.getKind()); - Value = adjustFixupValue(Fixup.getKind(), Value); - if (!Value) return; // Doesn't change encoding. - - unsigned Offset = Fixup.getOffset(); - assert(Offset + NumBytes <= DataSize && "Invalid fixup offset!"); - - // For each byte of the fragment that the fixup touches, mask in the - // bits from the fixup value. - for (unsigned i = 0; i != NumBytes; ++i) - Data[Offset + i] |= uint8_t((Value >> (i * 8)) & 0xff); -} - -} // end anonymous namespace - -TargetAsmBackend *llvm::createARMAsmBackend(const Target &T, - const std::string &TT) { - Triple TheTriple(TT); - - if (TheTriple.isOSDarwin()) { - if (TheTriple.getArchName() == "armv4t" || - TheTriple.getArchName() == "thumbv4t") - return new DarwinARMAsmBackend(T, object::mach::CSARM_V4T); - else if (TheTriple.getArchName() == "armv5e" || - TheTriple.getArchName() == "thumbv5e") - return new DarwinARMAsmBackend(T, object::mach::CSARM_V5TEJ); - else if (TheTriple.getArchName() == "armv6" || - TheTriple.getArchName() == "thumbv6") - return new DarwinARMAsmBackend(T, object::mach::CSARM_V6); - return new DarwinARMAsmBackend(T, object::mach::CSARM_V7); - } - - if (TheTriple.isOSWindows()) - assert(0 && "Windows not supported on ARM"); - - return new ELFARMAsmBackend(T, Triple(TT).getOS()); -} diff --git a/lib/Target/ARM/ARMAsmPrinter.cpp b/lib/Target/ARM/ARMAsmPrinter.cpp index dbc3ee41f3da8..ea3319fb0e03b 100644 --- a/lib/Target/ARM/ARMAsmPrinter.cpp +++ b/lib/Target/ARM/ARMAsmPrinter.cpp @@ -15,15 +15,15 @@ #define DEBUG_TYPE "asm-printer" #include "ARM.h" #include "ARMAsmPrinter.h" -#include "ARMAddressingModes.h" #include "ARMBuildAttrs.h" #include "ARMBaseRegisterInfo.h" #include "ARMConstantPoolValue.h" #include "ARMMachineFunctionInfo.h" -#include "ARMMCExpr.h" #include "ARMTargetMachine.h" #include "ARMTargetObjectFile.h" #include "InstPrinter/ARMInstPrinter.h" +#include "MCTargetDesc/ARMAddressingModes.h" +#include "MCTargetDesc/ARMMCExpr.h" #include "llvm/Analysis/DebugInfo.h" #include "llvm/Constants.h" #include "llvm/Module.h" @@ -45,13 +45,13 @@ #include "llvm/Target/TargetData.h" #include "llvm/Target/TargetMachine.h" #include "llvm/Target/TargetOptions.h" -#include "llvm/Target/TargetRegistry.h" #include "llvm/ADT/SmallPtrSet.h" #include "llvm/ADT/SmallString.h" #include "llvm/ADT/StringExtras.h" #include "llvm/Support/CommandLine.h" #include "llvm/Support/Debug.h" #include "llvm/Support/ErrorHandling.h" +#include "llvm/Support/TargetRegistry.h" #include "llvm/Support/raw_ostream.h" #include using namespace llvm; @@ -92,7 +92,7 @@ namespace { case ARMBuildAttrs::Advanced_SIMD_arch: case ARMBuildAttrs::VFP_arch: Streamer.EmitRawText(StringRef("\t.fpu ") + LowercaseString(String)); - break; + break; default: assert(0 && "Unsupported Text attribute in ASM Mode"); break; } } @@ -100,13 +100,41 @@ namespace { }; class ObjectAttributeEmitter : public AttributeEmitter { + // This structure holds all attributes, accounting for + // their string/numeric value, so we can later emmit them + // in declaration order, keeping all in the same vector + struct AttributeItemType { + enum { + HiddenAttribute = 0, + NumericAttribute, + TextAttribute + } Type; + unsigned Tag; + unsigned IntValue; + StringRef StringValue; + } AttributeItem; + MCObjectStreamer &Streamer; StringRef CurrentVendor; - SmallString<64> Contents; + SmallVector Contents; + + // Account for the ULEB/String size of each item, + // not just the number of items + size_t ContentsSize; + // FIXME: this should be in a more generic place, but + // getULEBSize() is in MCAsmInfo and will be moved to MCDwarf + size_t getULEBSize(int Value) { + size_t Size = 0; + do { + Value >>= 7; + Size += sizeof(int8_t); // Is this really necessary? + } while (Value); + return Size; + } public: ObjectAttributeEmitter(MCObjectStreamer &Streamer_) : - Streamer(Streamer_), CurrentVendor("") { } + Streamer(Streamer_), CurrentVendor(""), ContentsSize(0) { } void MaybeSwitchVendor(StringRef Vendor) { assert(!Vendor.empty() && "Vendor cannot be empty."); @@ -124,20 +152,32 @@ namespace { } void EmitAttribute(unsigned Attribute, unsigned Value) { - // FIXME: should be ULEB - Contents += Attribute; - Contents += Value; + AttributeItemType attr = { + AttributeItemType::NumericAttribute, + Attribute, + Value, + StringRef("") + }; + ContentsSize += getULEBSize(Attribute); + ContentsSize += getULEBSize(Value); + Contents.push_back(attr); } void EmitTextAttribute(unsigned Attribute, StringRef String) { - Contents += Attribute; - Contents += UppercaseString(String); - Contents += 0; + AttributeItemType attr = { + AttributeItemType::TextAttribute, + Attribute, + 0, + String + }; + ContentsSize += getULEBSize(Attribute); + // String + \0 + ContentsSize += String.size()+1; + + Contents.push_back(attr); } void Finish() { - const size_t ContentsSize = Contents.size(); - // Vendor size + Vendor name + '\0' const size_t VendorHeaderSize = 4 + CurrentVendor.size() + 1; @@ -151,7 +191,23 @@ namespace { Streamer.EmitIntValue(ARMBuildAttrs::File, 1); Streamer.EmitIntValue(TagHeaderSize + ContentsSize, 4); - Streamer.EmitBytes(Contents, 0); + // Size should have been accounted for already, now + // emit each field as its type (ULEB or String) + for (unsigned int i=0; i>1)) DW_OP_bit_piece(32, 0) // S[2x+1] = DW_OP_regx(256 + (x>>1)) DW_OP_bit_piece(32, 32) - + unsigned SReg = Reg - ARM::S0; bool odd = SReg & 0x1; unsigned Rx = 256 + (SReg >> 1); @@ -209,12 +265,13 @@ void ARMAsmPrinter::EmitDwarfRegOp(const MachineLocation &MLoc) const { } else if (Reg >= ARM::Q0 && Reg <= ARM::Q15) { assert(ARM::Q0 + 15 == ARM::Q15 && "Unexpected ARM Q register numbering"); // Q registers Q0-Q15 are described by composing two D registers together. - // Qx = DW_OP_regx(256+2x) DW_OP_piece(8) DW_OP_regx(256+2x+1) DW_OP_piece(8) + // Qx = DW_OP_regx(256+2x) DW_OP_piece(8) DW_OP_regx(256+2x+1) + // DW_OP_piece(8) unsigned QReg = Reg - ARM::Q0; unsigned D1 = 256 + 2 * QReg; unsigned D2 = D1 + 1; - + OutStreamer.AddComment("DW_OP_regx for Q register: D1"); EmitInt8(dwarf::DW_OP_regx); EmitULEB128(D1); @@ -233,6 +290,8 @@ void ARMAsmPrinter::EmitDwarfRegOp(const MachineLocation &MLoc) const { } void ARMAsmPrinter::EmitFunctionEntryLabel() { + OutStreamer.ForceCodeRegion(); + if (AFI->isThumbFunction()) { OutStreamer.EmitAssemblerFlag(MCAF_Code16); OutStreamer.EmitThumbFunc(CurrentFnSym); @@ -395,16 +454,16 @@ bool ARMAsmPrinter::PrintAsmOperand(const MachineInstr *MI, unsigned OpNum, // This takes advantage of the 2 operand-ness of ldm/stm and that we've // already got the operands in registers that are operands to the // inline asm statement. - + O << "{" << ARMInstPrinter::getRegisterName(RegBegin); - + // FIXME: The register allocator not only may not have given us the // registers in sequence, but may not be in ascending registers. This // will require changes in the register allocator that'll need to be // propagated down here if the operands change. unsigned RegOps = OpNum + 1; while (MI->getOperand(RegOps).isReg()) { - O << ", " + O << ", " << ARMInstPrinter::getRegisterName(MI->getOperand(RegOps).getReg()); RegOps++; } @@ -413,14 +472,34 @@ bool ARMAsmPrinter::PrintAsmOperand(const MachineInstr *MI, unsigned OpNum, return false; } + case 'R': // The most significant register of a pair. + case 'Q': { // The least significant register of a pair. + if (OpNum == 0) + return true; + const MachineOperand &FlagsOP = MI->getOperand(OpNum - 1); + if (!FlagsOP.isImm()) + return true; + unsigned Flags = FlagsOP.getImm(); + unsigned NumVals = InlineAsm::getNumOperandRegisters(Flags); + if (NumVals != 2) + return true; + unsigned RegOp = ExtraCode[0] == 'Q' ? OpNum : OpNum + 1; + if (RegOp >= MI->getNumOperands()) + return true; + const MachineOperand &MO = MI->getOperand(RegOp); + if (!MO.isReg()) + return true; + unsigned Reg = MO.getReg(); + O << ARMInstPrinter::getRegisterName(Reg); + return false; + } + // These modifiers are not yet supported. case 'p': // The high single-precision register of a VFP double-precision // register. case 'e': // The low doubleword register of a NEON quad register. case 'f': // The high doubleword register of a NEON quad register. case 'h': // A range of VFP/NEON registers suitable for VLD1/VST1. - case 'Q': // The least significant register of a pair. - case 'R': // The most significant register of a pair. case 'H': // The highest-numbered register of a pair. return true; } @@ -437,7 +516,7 @@ bool ARMAsmPrinter::PrintAsmMemoryOperand(const MachineInstr *MI, // Does this asm operand have a single letter operand modifier? if (ExtraCode && ExtraCode[0]) { if (ExtraCode[1] != 0) return true; // Unknown modifier. - + switch (ExtraCode[0]) { case 'A': // A memory operand for a VLD1/VST1 instruction. default: return true; // Unknown modifier. @@ -448,7 +527,7 @@ bool ARMAsmPrinter::PrintAsmMemoryOperand(const MachineInstr *MI, return false; } } - + const MachineOperand &MO = MI->getOperand(OpNum); assert(MO.isReg() && "unexpected inline asm memory operand"); O << "[" << ARMInstPrinter::getRegisterName(MO.getReg()) << "]"; @@ -772,13 +851,19 @@ EmitMachineConstantPoolValue(MachineConstantPoolValue *MCPV) { OS << MAI->getPrivateGlobalPrefix() << "_LSDA_" << getFunctionNumber(); MCSym = OutContext.GetOrCreateSymbol(OS.str()); } else if (ACPV->isBlockAddress()) { - MCSym = GetBlockAddressSymbol(ACPV->getBlockAddress()); + const BlockAddress *BA = + cast(ACPV)->getBlockAddress(); + MCSym = GetBlockAddressSymbol(BA); } else if (ACPV->isGlobalValue()) { - const GlobalValue *GV = ACPV->getGV(); + const GlobalValue *GV = cast(ACPV)->getGV(); MCSym = GetARMGVSymbol(GV); + } else if (ACPV->isMachineBasicBlock()) { + const MachineBasicBlock *MBB = cast(ACPV)->getMBB(); + MCSym = MBB->getSymbol(); } else { assert(ACPV->isExtSymbol() && "unrecognized constant pool value"); - MCSym = GetExternalSymbolSymbol(ACPV->getSymbol()); + const char *Sym = cast(ACPV)->getSymbol(); + MCSym = GetExternalSymbolSymbol(Sym); } // Create an MCSymbol for the reference. @@ -822,6 +907,9 @@ void ARMAsmPrinter::EmitJumpTable(const MachineInstr *MI) { const MachineOperand &MO2 = MI->getOperand(OpNum+1); // Unique Id unsigned JTI = MO1.getIndex(); + // Tag the jump table appropriately for precise disassembly. + OutStreamer.EmitJumpTable32Region(); + // Emit a label for the jump table. MCSymbol *JTISymbol = GetARMJTIPICJumpTableLabel2(JTI, MO2.getImm()); OutStreamer.EmitLabel(JTISymbol); @@ -847,6 +935,11 @@ void ARMAsmPrinter::EmitJumpTable(const MachineInstr *MI) { Expr = MCBinaryExpr::CreateSub(Expr, MCSymbolRefExpr::Create(JTISymbol, OutContext), OutContext); + // If we're generating a table of Thumb addresses in static relocation + // model, we need to add one to keep interworking correctly. + else if (AFI->isThumbFunction()) + Expr = MCBinaryExpr::CreateAdd(Expr, MCConstantExpr::Create(1,OutContext), + OutContext); OutStreamer.EmitValue(Expr, 4); } } @@ -859,6 +952,14 @@ void ARMAsmPrinter::EmitJump2Table(const MachineInstr *MI) { unsigned JTI = MO1.getIndex(); // Emit a label for the jump table. + if (MI->getOpcode() == ARM::t2TBB_JT) { + OutStreamer.EmitJumpTable8Region(); + } else if (MI->getOpcode() == ARM::t2TBH_JT) { + OutStreamer.EmitJumpTable16Region(); + } else { + OutStreamer.EmitJumpTable32Region(); + } + MCSymbol *JTISymbol = GetARMJTIPICJumpTableLabel2(JTI, MO2.getImm()); OutStreamer.EmitLabel(JTISymbol); @@ -881,6 +982,8 @@ void ARMAsmPrinter::EmitJump2Table(const MachineInstr *MI) { MCInst BrInst; BrInst.setOpcode(ARM::t2B); BrInst.addOperand(MCOperand::CreateExpr(MBBSymbolExpr)); + BrInst.addOperand(MCOperand::CreateImm(ARMCC::AL)); + BrInst.addOperand(MCOperand::CreateReg(0)); OutStreamer.EmitInstruction(BrInst); continue; } @@ -994,7 +1097,8 @@ void ARMAsmPrinter::EmitUnwindingInstruction(const MachineInstr *MI) { i != NumOps; ++i) RegList.push_back(MI->getOperand(i).getReg()); break; - case ARM::STR_PRE: + case ARM::STR_PRE_IMM: + case ARM::STR_PRE_REG: assert(MI->getOperand(2).getReg() == ARM::SP && "Only stack pointer as a source reg is supported"); RegList.push_back(SrcReg); @@ -1074,10 +1178,20 @@ extern cl::opt EnableARMEHABI; #include "ARMGenMCPseudoLowering.inc" void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) { + if (MI->getOpcode() != ARM::CONSTPOOL_ENTRY) + OutStreamer.EmitCodeRegion(); + + // Emit unwinding stuff for frame-related instructions + if (EnableARMEHABI && MI->getFlag(MachineInstr::FrameSetup)) + EmitUnwindingInstruction(MI); + // Do any auto-generated pseudo lowerings. if (emitPseudoExpansionLowering(OutStreamer, MI)) return; + assert(!convertAddSubFlagsOpcode(MI->getOpcode()) && + "Pseudo flag setting opcode should be expanded early"); + // Check for manual lowerings. unsigned Opc = MI->getOpcode(); switch (Opc) { @@ -1372,6 +1486,10 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) { unsigned CPIdx = (unsigned)MI->getOperand(1).getIndex(); EmitAlignment(2); + + // Mark the constant pool entry as data if we're not already in a data + // region. + OutStreamer.EmitDataRegion(); OutStreamer.EmitLabel(GetCPISymbol(LabelId)); const MachineConstantPoolEntry &MCPE = MCP->getConstants()[CPIdx]; @@ -1379,7 +1497,6 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) { EmitMachineConstantPoolValue(MCPE.Val.MachineCPVal); else EmitGlobalConstant(MCPE.Val.ConstVal); - return; } case ARM::t2BR_JT: { @@ -1590,6 +1707,8 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) { MCInst TmpInst; TmpInst.setOpcode(ARM::tB); TmpInst.addOperand(MCOperand::CreateExpr(SymbolExpr)); + TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL)); + TmpInst.addOperand(MCOperand::CreateReg(0)); OutStreamer.EmitInstruction(TmpInst); } { @@ -1804,10 +1923,6 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) { MCInst TmpInst; LowerARMMachineInstrToMCInst(MI, TmpInst, *this); - // Emit unwinding stuff for frame-related instructions - if (EnableARMEHABI && MI->getFlag(MachineInstr::FrameSetup)) - EmitUnwindingInstruction(MI); - OutStreamer.EmitInstruction(TmpInst); } @@ -1815,20 +1930,9 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) { // Target Registry Stuff //===----------------------------------------------------------------------===// -static MCInstPrinter *createARMMCInstPrinter(const Target &T, - unsigned SyntaxVariant, - const MCAsmInfo &MAI) { - if (SyntaxVariant == 0) - return new ARMInstPrinter(MAI); - return 0; -} - // Force static initialization. extern "C" void LLVMInitializeARMAsmPrinter() { RegisterAsmPrinter X(TheARMTarget); RegisterAsmPrinter Y(TheThumbTarget); - - TargetRegistry::RegisterMCInstPrinter(TheARMTarget, createARMMCInstPrinter); - TargetRegistry::RegisterMCInstPrinter(TheThumbTarget, createARMMCInstPrinter); } diff --git a/lib/Target/ARM/ARMBaseInfo.h b/lib/Target/ARM/ARMBaseInfo.h deleted file mode 100644 index 458f7dd1f784e..0000000000000 --- a/lib/Target/ARM/ARMBaseInfo.h +++ /dev/null @@ -1,294 +0,0 @@ -//===-- ARMBaseInfo.h - Top level definitions for ARM -------- --*- C++ -*-===// -// -// The LLVM Compiler Infrastructure -// -// This file is distributed under the University of Illinois Open Source -// License. See LICENSE.TXT for details. -// -//===----------------------------------------------------------------------===// -// -// This file contains small standalone helper functions and enum definitions for -// the ARM target useful for the compiler back-end and the MC libraries. -// As such, it deliberately does not include references to LLVM core -// code gen types, passes, etc.. -// -//===----------------------------------------------------------------------===// - -#ifndef ARMBASEINFO_H -#define ARMBASEINFO_H - -#include "MCTargetDesc/ARMMCTargetDesc.h" -#include "llvm/Support/ErrorHandling.h" - -// Note that the following auto-generated files only defined enum types, and -// so are safe to include here. - -namespace llvm { - -// Enums corresponding to ARM condition codes -namespace ARMCC { - // The CondCodes constants map directly to the 4-bit encoding of the - // condition field for predicated instructions. - enum CondCodes { // Meaning (integer) Meaning (floating-point) - EQ, // Equal Equal - NE, // Not equal Not equal, or unordered - HS, // Carry set >, ==, or unordered - LO, // Carry clear Less than - MI, // Minus, negative Less than - PL, // Plus, positive or zero >, ==, or unordered - VS, // Overflow Unordered - VC, // No overflow Not unordered - HI, // Unsigned higher Greater than, or unordered - LS, // Unsigned lower or same Less than or equal - GE, // Greater than or equal Greater than or equal - LT, // Less than Less than, or unordered - GT, // Greater than Greater than - LE, // Less than or equal <, ==, or unordered - AL // Always (unconditional) Always (unconditional) - }; - - inline static CondCodes getOppositeCondition(CondCodes CC) { - switch (CC) { - default: llvm_unreachable("Unknown condition code"); - case EQ: return NE; - case NE: return EQ; - case HS: return LO; - case LO: return HS; - case MI: return PL; - case PL: return MI; - case VS: return VC; - case VC: return VS; - case HI: return LS; - case LS: return HI; - case GE: return LT; - case LT: return GE; - case GT: return LE; - case LE: return GT; - } - } -} // namespace ARMCC - -inline static const char *ARMCondCodeToString(ARMCC::CondCodes CC) { - switch (CC) { - default: llvm_unreachable("Unknown condition code"); - case ARMCC::EQ: return "eq"; - case ARMCC::NE: return "ne"; - case ARMCC::HS: return "hs"; - case ARMCC::LO: return "lo"; - case ARMCC::MI: return "mi"; - case ARMCC::PL: return "pl"; - case ARMCC::VS: return "vs"; - case ARMCC::VC: return "vc"; - case ARMCC::HI: return "hi"; - case ARMCC::LS: return "ls"; - case ARMCC::GE: return "ge"; - case ARMCC::LT: return "lt"; - case ARMCC::GT: return "gt"; - case ARMCC::LE: return "le"; - case ARMCC::AL: return "al"; - } -} - -namespace ARM_PROC { - enum IMod { - IE = 2, - ID = 3 - }; - - enum IFlags { - F = 1, - I = 2, - A = 4 - }; - - inline static const char *IFlagsToString(unsigned val) { - switch (val) { - default: llvm_unreachable("Unknown iflags operand"); - case F: return "f"; - case I: return "i"; - case A: return "a"; - } - } - - inline static const char *IModToString(unsigned val) { - switch (val) { - default: llvm_unreachable("Unknown imod operand"); - case IE: return "ie"; - case ID: return "id"; - } - } -} - -namespace ARM_MB { - // The Memory Barrier Option constants map directly to the 4-bit encoding of - // the option field for memory barrier operations. - enum MemBOpt { - SY = 15, - ST = 14, - ISH = 11, - ISHST = 10, - NSH = 7, - NSHST = 6, - OSH = 3, - OSHST = 2 - }; - - inline static const char *MemBOptToString(unsigned val) { - switch (val) { - default: llvm_unreachable("Unknown memory operation"); - case SY: return "sy"; - case ST: return "st"; - case ISH: return "ish"; - case ISHST: return "ishst"; - case NSH: return "nsh"; - case NSHST: return "nshst"; - case OSH: return "osh"; - case OSHST: return "oshst"; - } - } -} // namespace ARM_MB - -/// getARMRegisterNumbering - Given the enum value for some register, e.g. -/// ARM::LR, return the number that it corresponds to (e.g. 14). -inline static unsigned getARMRegisterNumbering(unsigned Reg) { - using namespace ARM; - switch (Reg) { - default: - llvm_unreachable("Unknown ARM register!"); - case R0: case S0: case D0: case Q0: return 0; - case R1: case S1: case D1: case Q1: return 1; - case R2: case S2: case D2: case Q2: return 2; - case R3: case S3: case D3: case Q3: return 3; - case R4: case S4: case D4: case Q4: return 4; - case R5: case S5: case D5: case Q5: return 5; - case R6: case S6: case D6: case Q6: return 6; - case R7: case S7: case D7: case Q7: return 7; - case R8: case S8: case D8: case Q8: return 8; - case R9: case S9: case D9: case Q9: return 9; - case R10: case S10: case D10: case Q10: return 10; - case R11: case S11: case D11: case Q11: return 11; - case R12: case S12: case D12: case Q12: return 12; - case SP: case S13: case D13: case Q13: return 13; - case LR: case S14: case D14: case Q14: return 14; - case PC: case S15: case D15: case Q15: return 15; - - case S16: case D16: return 16; - case S17: case D17: return 17; - case S18: case D18: return 18; - case S19: case D19: return 19; - case S20: case D20: return 20; - case S21: case D21: return 21; - case S22: case D22: return 22; - case S23: case D23: return 23; - case S24: case D24: return 24; - case S25: case D25: return 25; - case S26: case D26: return 26; - case S27: case D27: return 27; - case S28: case D28: return 28; - case S29: case D29: return 29; - case S30: case D30: return 30; - case S31: case D31: return 31; - } -} - -namespace ARMII { - - /// ARM Index Modes - enum IndexMode { - IndexModeNone = 0, - IndexModePre = 1, - IndexModePost = 2, - IndexModeUpd = 3 - }; - - /// ARM Addressing Modes - enum AddrMode { - AddrModeNone = 0, - AddrMode1 = 1, - AddrMode2 = 2, - AddrMode3 = 3, - AddrMode4 = 4, - AddrMode5 = 5, - AddrMode6 = 6, - AddrModeT1_1 = 7, - AddrModeT1_2 = 8, - AddrModeT1_4 = 9, - AddrModeT1_s = 10, // i8 * 4 for pc and sp relative data - AddrModeT2_i12 = 11, - AddrModeT2_i8 = 12, - AddrModeT2_so = 13, - AddrModeT2_pc = 14, // +/- i12 for pc relative data - AddrModeT2_i8s4 = 15, // i8 * 4 - AddrMode_i12 = 16 - }; - - inline static const char *AddrModeToString(AddrMode addrmode) { - switch (addrmode) { - default: llvm_unreachable("Unknown memory operation"); - case AddrModeNone: return "AddrModeNone"; - case AddrMode1: return "AddrMode1"; - case AddrMode2: return "AddrMode2"; - case AddrMode3: return "AddrMode3"; - case AddrMode4: return "AddrMode4"; - case AddrMode5: return "AddrMode5"; - case AddrMode6: return "AddrMode6"; - case AddrModeT1_1: return "AddrModeT1_1"; - case AddrModeT1_2: return "AddrModeT1_2"; - case AddrModeT1_4: return "AddrModeT1_4"; - case AddrModeT1_s: return "AddrModeT1_s"; - case AddrModeT2_i12: return "AddrModeT2_i12"; - case AddrModeT2_i8: return "AddrModeT2_i8"; - case AddrModeT2_so: return "AddrModeT2_so"; - case AddrModeT2_pc: return "AddrModeT2_pc"; - case AddrModeT2_i8s4: return "AddrModeT2_i8s4"; - case AddrMode_i12: return "AddrMode_i12"; - } - } - - /// Target Operand Flag enum. - enum TOF { - //===------------------------------------------------------------------===// - // ARM Specific MachineOperand flags. - - MO_NO_FLAG, - - /// MO_LO16 - On a symbol operand, this represents a relocation containing - /// lower 16 bit of the address. Used only via movw instruction. - MO_LO16, - - /// MO_HI16 - On a symbol operand, this represents a relocation containing - /// higher 16 bit of the address. Used only via movt instruction. - MO_HI16, - - /// MO_LO16_NONLAZY - On a symbol operand "FOO", this represents a - /// relocation containing lower 16 bit of the non-lazy-ptr indirect symbol, - /// i.e. "FOO$non_lazy_ptr". - /// Used only via movw instruction. - MO_LO16_NONLAZY, - - /// MO_HI16_NONLAZY - On a symbol operand "FOO", this represents a - /// relocation containing lower 16 bit of the non-lazy-ptr indirect symbol, - /// i.e. "FOO$non_lazy_ptr". Used only via movt instruction. - MO_HI16_NONLAZY, - - /// MO_LO16_NONLAZY_PIC - On a symbol operand "FOO", this represents a - /// relocation containing lower 16 bit of the PC relative address of the - /// non-lazy-ptr indirect symbol, i.e. "FOO$non_lazy_ptr - LABEL". - /// Used only via movw instruction. - MO_LO16_NONLAZY_PIC, - - /// MO_HI16_NONLAZY_PIC - On a symbol operand "FOO", this represents a - /// relocation containing lower 16 bit of the PC relative address of the - /// non-lazy-ptr indirect symbol, i.e. "FOO$non_lazy_ptr - LABEL". - /// Used only via movt instruction. - MO_HI16_NONLAZY_PIC, - - /// MO_PLT - On a symbol operand, this represents an ELF PLT reference on a - /// call operand. - MO_PLT - }; -} // end namespace ARMII - -} // end namespace llvm; - -#endif diff --git a/lib/Target/ARM/ARMBaseInstrInfo.cpp b/lib/Target/ARM/ARMBaseInstrInfo.cpp index 649bd7d5ce3f9..408edfc20d4cc 100644 --- a/lib/Target/ARM/ARMBaseInstrInfo.cpp +++ b/lib/Target/ARM/ARMBaseInstrInfo.cpp @@ -13,11 +13,11 @@ #include "ARMBaseInstrInfo.h" #include "ARM.h" -#include "ARMAddressingModes.h" #include "ARMConstantPoolValue.h" #include "ARMHazardRecognizer.h" #include "ARMMachineFunctionInfo.h" #include "ARMRegisterInfo.h" +#include "MCTargetDesc/ARMAddressingModes.h" #include "llvm/Constants.h" #include "llvm/Function.h" #include "llvm/GlobalValue.h" @@ -29,6 +29,7 @@ #include "llvm/CodeGen/MachineMemOperand.h" #include "llvm/CodeGen/MachineRegisterInfo.h" #include "llvm/CodeGen/PseudoSourceValue.h" +#include "llvm/CodeGen/SelectionDAGNodes.h" #include "llvm/MC/MCAsmInfo.h" #include "llvm/Support/BranchProbability.h" #include "llvm/Support/CommandLine.h" @@ -45,6 +46,10 @@ static cl::opt EnableARM3Addr("enable-arm-3-addr-conv", cl::Hidden, cl::desc("Enable ARM 2-addr to 3-addr conv")); +static cl::opt +WidenVMOVS("widen-vmovs", cl::Hidden, + cl::desc("Widen ARM vmovs to vmovd when possible")); + /// ARM_MLxEntry - Record information about MLA / MLS instructions. struct ARM_MLxEntry { unsigned MLxOpc; // MLA / MLS opcode @@ -171,7 +176,7 @@ ARMBaseInstrInfo::convertToThreeAddress(MachineFunction::iterator &MFI, ARM_AM::ShiftOpc ShOpc = ARM_AM::getAM2ShiftOpc(OffImm); unsigned SOOpc = ARM_AM::getSORegOpc(ShOpc, Amt); UpdateMI = BuildMI(MF, MI->getDebugLoc(), - get(isSub ? ARM::SUBrs : ARM::ADDrs), WBReg) + get(isSub ? ARM::SUBrsi : ARM::ADDrsi), WBReg) .addReg(BaseReg).addReg(OffReg).addReg(0).addImm(SOOpc) .addImm(Pred).addReg(0).addReg(0); } else @@ -399,6 +404,7 @@ ARMBaseInstrInfo::InsertBranch(MachineBasicBlock &MBB, MachineBasicBlock *TBB, ? ARM::B : (AFI->isThumb2Function() ? ARM::t2B : ARM::tB); int BccOpc = !AFI->isThumbFunction() ? ARM::Bcc : (AFI->isThumb2Function() ? ARM::t2Bcc : ARM::tBcc); + bool isThumb = AFI->isThumbFunction() || AFI->isThumb2Function(); // Shouldn't be a fall through. assert(TBB && "InsertBranch must not be told to insert a fallthrough"); @@ -406,9 +412,12 @@ ARMBaseInstrInfo::InsertBranch(MachineBasicBlock &MBB, MachineBasicBlock *TBB, "ARM branch conditions have two components!"); if (FBB == 0) { - if (Cond.empty()) // Unconditional branch? - BuildMI(&MBB, DL, get(BOpc)).addMBB(TBB); - else + if (Cond.empty()) { // Unconditional branch? + if (isThumb) + BuildMI(&MBB, DL, get(BOpc)).addMBB(TBB).addImm(ARMCC::AL).addReg(0); + else + BuildMI(&MBB, DL, get(BOpc)).addMBB(TBB); + } else BuildMI(&MBB, DL, get(BccOpc)).addMBB(TBB) .addImm(Cond[0].getImm()).addReg(Cond[1].getReg()); return 1; @@ -417,7 +426,10 @@ ARMBaseInstrInfo::InsertBranch(MachineBasicBlock &MBB, MachineBasicBlock *TBB, // Two-way conditional branch. BuildMI(&MBB, DL, get(BccOpc)).addMBB(TBB) .addImm(Cond[0].getImm()).addReg(Cond[1].getReg()); - BuildMI(&MBB, DL, get(BOpc)).addMBB(FBB); + if (isThumb) + BuildMI(&MBB, DL, get(BOpc)).addMBB(FBB).addImm(ARMCC::AL).addReg(0); + else + BuildMI(&MBB, DL, get(BOpc)).addMBB(FBB); return 2; } @@ -627,7 +639,7 @@ void ARMBaseInstrInfo::copyPhysReg(MachineBasicBlock &MBB, bool SPRDest = ARM::SPRRegClass.contains(DestReg); bool SPRSrc = ARM::SPRRegClass.contains(SrcReg); - unsigned Opc; + unsigned Opc = 0; if (SPRDest && SPRSrc) Opc = ARM::VMOVS; else if (GPRDest && SPRSrc) @@ -638,19 +650,40 @@ void ARMBaseInstrInfo::copyPhysReg(MachineBasicBlock &MBB, Opc = ARM::VMOVD; else if (ARM::QPRRegClass.contains(DestReg, SrcReg)) Opc = ARM::VORRq; - else if (ARM::QQPRRegClass.contains(DestReg, SrcReg)) - Opc = ARM::VMOVQQ; - else if (ARM::QQQQPRRegClass.contains(DestReg, SrcReg)) - Opc = ARM::VMOVQQQQ; - else - llvm_unreachable("Impossible reg-to-reg copy"); - MachineInstrBuilder MIB = BuildMI(MBB, I, DL, get(Opc), DestReg); - MIB.addReg(SrcReg, getKillRegState(KillSrc)); - if (Opc == ARM::VORRq) + if (Opc) { + MachineInstrBuilder MIB = BuildMI(MBB, I, DL, get(Opc), DestReg); MIB.addReg(SrcReg, getKillRegState(KillSrc)); - if (Opc != ARM::VMOVQQ && Opc != ARM::VMOVQQQQ) + if (Opc == ARM::VORRq) + MIB.addReg(SrcReg, getKillRegState(KillSrc)); AddDefaultPred(MIB); + return; + } + + // Generate instructions for VMOVQQ and VMOVQQQQ pseudos in place. + if (ARM::QQPRRegClass.contains(DestReg, SrcReg) || + ARM::QQQQPRRegClass.contains(DestReg, SrcReg)) { + const TargetRegisterInfo *TRI = &getRegisterInfo(); + assert(ARM::qsub_0 + 3 == ARM::qsub_3 && "Expected contiguous enum."); + unsigned EndSubReg = ARM::QQPRRegClass.contains(DestReg, SrcReg) ? + ARM::qsub_1 : ARM::qsub_3; + for (unsigned i = ARM::qsub_0, e = EndSubReg + 1; i != e; ++i) { + unsigned Dst = TRI->getSubReg(DestReg, i); + unsigned Src = TRI->getSubReg(SrcReg, i); + MachineInstrBuilder Mov = + AddDefaultPred(BuildMI(MBB, I, I->getDebugLoc(), get(ARM::VORRq)) + .addReg(Dst, RegState::Define) + .addReg(Src, getKillRegState(KillSrc)) + .addReg(Src, getKillRegState(KillSrc))); + if (i == EndSubReg) { + Mov->addRegisterDefined(DestReg, TRI); + if (KillSrc) + Mov->addRegisterKilled(SrcReg, TRI); + } + } + return; + } + llvm_unreachable("Impossible reg-to-reg copy"); } static const @@ -683,82 +716,84 @@ storeRegToStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, MFI.getObjectSize(FI), Align); - // tGPR is used sometimes in ARM instructions that need to avoid using - // certain registers. Just treat it as GPR here. Likewise, rGPR. - if (RC == ARM::tGPRRegisterClass || RC == ARM::tcGPRRegisterClass - || RC == ARM::rGPRRegisterClass) - RC = ARM::GPRRegisterClass; - - switch (RC->getID()) { - case ARM::GPRRegClassID: - AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::STRi12)) + switch (RC->getSize()) { + case 4: + if (ARM::GPRRegClass.hasSubClassEq(RC)) { + AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::STRi12)) .addReg(SrcReg, getKillRegState(isKill)) .addFrameIndex(FI).addImm(0).addMemOperand(MMO)); - break; - case ARM::SPRRegClassID: - AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VSTRS)) + } else if (ARM::SPRRegClass.hasSubClassEq(RC)) { + AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VSTRS)) .addReg(SrcReg, getKillRegState(isKill)) .addFrameIndex(FI).addImm(0).addMemOperand(MMO)); - break; - case ARM::DPRRegClassID: - case ARM::DPR_VFP2RegClassID: - case ARM::DPR_8RegClassID: - AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VSTRD)) + } else + llvm_unreachable("Unknown reg class!"); + break; + case 8: + if (ARM::DPRRegClass.hasSubClassEq(RC)) { + AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VSTRD)) .addReg(SrcReg, getKillRegState(isKill)) .addFrameIndex(FI).addImm(0).addMemOperand(MMO)); - break; - case ARM::QPRRegClassID: - case ARM::QPR_VFP2RegClassID: - case ARM::QPR_8RegClassID: - if (Align >= 16 && getRegisterInfo().needsStackRealignment(MF)) { - AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VST1q64Pseudo)) + } else + llvm_unreachable("Unknown reg class!"); + break; + case 16: + if (ARM::QPRRegClass.hasSubClassEq(RC)) { + if (Align >= 16 && getRegisterInfo().needsStackRealignment(MF)) { + AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VST1q64Pseudo)) .addFrameIndex(FI).addImm(16) .addReg(SrcReg, getKillRegState(isKill)) .addMemOperand(MMO)); - } else { - AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VSTMQIA)) + } else { + AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VSTMQIA)) .addReg(SrcReg, getKillRegState(isKill)) .addFrameIndex(FI) .addMemOperand(MMO)); - } - break; - case ARM::QQPRRegClassID: - case ARM::QQPR_VFP2RegClassID: - if (Align >= 16 && getRegisterInfo().canRealignStack(MF)) { - // FIXME: It's possible to only store part of the QQ register if the - // spilled def has a sub-register index. - AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VST1d64QPseudo)) + } + } else + llvm_unreachable("Unknown reg class!"); + break; + case 32: + if (ARM::QQPRRegClass.hasSubClassEq(RC)) { + if (Align >= 16 && getRegisterInfo().canRealignStack(MF)) { + // FIXME: It's possible to only store part of the QQ register if the + // spilled def has a sub-register index. + AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VST1d64QPseudo)) .addFrameIndex(FI).addImm(16) .addReg(SrcReg, getKillRegState(isKill)) .addMemOperand(MMO)); - } else { - MachineInstrBuilder MIB = - AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VSTMDIA)) + } else { + MachineInstrBuilder MIB = + AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VSTMDIA)) .addFrameIndex(FI)) - .addMemOperand(MMO); - MIB = AddDReg(MIB, SrcReg, ARM::dsub_0, getKillRegState(isKill), TRI); - MIB = AddDReg(MIB, SrcReg, ARM::dsub_1, 0, TRI); - MIB = AddDReg(MIB, SrcReg, ARM::dsub_2, 0, TRI); - AddDReg(MIB, SrcReg, ARM::dsub_3, 0, TRI); - } - break; - case ARM::QQQQPRRegClassID: { - MachineInstrBuilder MIB = - AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VSTMDIA)) - .addFrameIndex(FI)) - .addMemOperand(MMO); - MIB = AddDReg(MIB, SrcReg, ARM::dsub_0, getKillRegState(isKill), TRI); - MIB = AddDReg(MIB, SrcReg, ARM::dsub_1, 0, TRI); - MIB = AddDReg(MIB, SrcReg, ARM::dsub_2, 0, TRI); - MIB = AddDReg(MIB, SrcReg, ARM::dsub_3, 0, TRI); - MIB = AddDReg(MIB, SrcReg, ARM::dsub_4, 0, TRI); - MIB = AddDReg(MIB, SrcReg, ARM::dsub_5, 0, TRI); - MIB = AddDReg(MIB, SrcReg, ARM::dsub_6, 0, TRI); - AddDReg(MIB, SrcReg, ARM::dsub_7, 0, TRI); - break; - } - default: - llvm_unreachable("Unknown regclass!"); + .addMemOperand(MMO); + MIB = AddDReg(MIB, SrcReg, ARM::dsub_0, getKillRegState(isKill), TRI); + MIB = AddDReg(MIB, SrcReg, ARM::dsub_1, 0, TRI); + MIB = AddDReg(MIB, SrcReg, ARM::dsub_2, 0, TRI); + AddDReg(MIB, SrcReg, ARM::dsub_3, 0, TRI); + } + } else + llvm_unreachable("Unknown reg class!"); + break; + case 64: + if (ARM::QQQQPRRegClass.hasSubClassEq(RC)) { + MachineInstrBuilder MIB = + AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VSTMDIA)) + .addFrameIndex(FI)) + .addMemOperand(MMO); + MIB = AddDReg(MIB, SrcReg, ARM::dsub_0, getKillRegState(isKill), TRI); + MIB = AddDReg(MIB, SrcReg, ARM::dsub_1, 0, TRI); + MIB = AddDReg(MIB, SrcReg, ARM::dsub_2, 0, TRI); + MIB = AddDReg(MIB, SrcReg, ARM::dsub_3, 0, TRI); + MIB = AddDReg(MIB, SrcReg, ARM::dsub_4, 0, TRI); + MIB = AddDReg(MIB, SrcReg, ARM::dsub_5, 0, TRI); + MIB = AddDReg(MIB, SrcReg, ARM::dsub_6, 0, TRI); + AddDReg(MIB, SrcReg, ARM::dsub_7, 0, TRI); + } else + llvm_unreachable("Unknown reg class!"); + break; + default: + llvm_unreachable("Unknown reg class!"); } } @@ -809,6 +844,12 @@ ARMBaseInstrInfo::isStoreToStackSlot(const MachineInstr *MI, return 0; } +unsigned ARMBaseInstrInfo::isStoreToStackSlotPostFE(const MachineInstr *MI, + int &FrameIndex) const { + const MachineMemOperand *Dummy; + return MI->getDesc().mayStore() && hasStoreToStackSlot(MI, Dummy, FrameIndex); +} + void ARMBaseInstrInfo:: loadRegFromStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, unsigned DestReg, int FI, @@ -826,72 +867,77 @@ loadRegFromStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, MFI.getObjectSize(FI), Align); - // tGPR is used sometimes in ARM instructions that need to avoid using - // certain registers. Just treat it as GPR here. - if (RC == ARM::tGPRRegisterClass || RC == ARM::tcGPRRegisterClass - || RC == ARM::rGPRRegisterClass) - RC = ARM::GPRRegisterClass; - - switch (RC->getID()) { - case ARM::GPRRegClassID: - AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::LDRi12), DestReg) + switch (RC->getSize()) { + case 4: + if (ARM::GPRRegClass.hasSubClassEq(RC)) { + AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::LDRi12), DestReg) .addFrameIndex(FI).addImm(0).addMemOperand(MMO)); - break; - case ARM::SPRRegClassID: - AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VLDRS), DestReg) + + } else if (ARM::SPRRegClass.hasSubClassEq(RC)) { + AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VLDRS), DestReg) .addFrameIndex(FI).addImm(0).addMemOperand(MMO)); + } else + llvm_unreachable("Unknown reg class!"); break; - case ARM::DPRRegClassID: - case ARM::DPR_VFP2RegClassID: - case ARM::DPR_8RegClassID: - AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VLDRD), DestReg) + case 8: + if (ARM::DPRRegClass.hasSubClassEq(RC)) { + AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VLDRD), DestReg) .addFrameIndex(FI).addImm(0).addMemOperand(MMO)); + } else + llvm_unreachable("Unknown reg class!"); break; - case ARM::QPRRegClassID: - case ARM::QPR_VFP2RegClassID: - case ARM::QPR_8RegClassID: - if (Align >= 16 && getRegisterInfo().needsStackRealignment(MF)) { - AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VLD1q64Pseudo), DestReg) + case 16: + if (ARM::QPRRegClass.hasSubClassEq(RC)) { + if (Align >= 16 && getRegisterInfo().needsStackRealignment(MF)) { + AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VLD1q64Pseudo), DestReg) .addFrameIndex(FI).addImm(16) .addMemOperand(MMO)); - } else { - AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VLDMQIA), DestReg) - .addFrameIndex(FI) - .addMemOperand(MMO)); - } + } else { + AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VLDMQIA), DestReg) + .addFrameIndex(FI) + .addMemOperand(MMO)); + } + } else + llvm_unreachable("Unknown reg class!"); break; - case ARM::QQPRRegClassID: - case ARM::QQPR_VFP2RegClassID: - if (Align >= 16 && getRegisterInfo().canRealignStack(MF)) { - AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VLD1d64QPseudo), DestReg) + case 32: + if (ARM::QQPRRegClass.hasSubClassEq(RC)) { + if (Align >= 16 && getRegisterInfo().canRealignStack(MF)) { + AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VLD1d64QPseudo), DestReg) .addFrameIndex(FI).addImm(16) .addMemOperand(MMO)); - } else { - MachineInstrBuilder MIB = + } else { + MachineInstrBuilder MIB = AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VLDMDIA)) .addFrameIndex(FI)) - .addMemOperand(MMO); - MIB = AddDReg(MIB, DestReg, ARM::dsub_0, RegState::Define, TRI); - MIB = AddDReg(MIB, DestReg, ARM::dsub_1, RegState::Define, TRI); - MIB = AddDReg(MIB, DestReg, ARM::dsub_2, RegState::Define, TRI); - AddDReg(MIB, DestReg, ARM::dsub_3, RegState::Define, TRI); - } + .addMemOperand(MMO); + MIB = AddDReg(MIB, DestReg, ARM::dsub_0, RegState::Define, TRI); + MIB = AddDReg(MIB, DestReg, ARM::dsub_1, RegState::Define, TRI); + MIB = AddDReg(MIB, DestReg, ARM::dsub_2, RegState::Define, TRI); + MIB = AddDReg(MIB, DestReg, ARM::dsub_3, RegState::Define, TRI); + MIB.addReg(DestReg, RegState::Define | RegState::Implicit); + } + } else + llvm_unreachable("Unknown reg class!"); break; - case ARM::QQQQPRRegClassID: { - MachineInstrBuilder MIB = + case 64: + if (ARM::QQQQPRRegClass.hasSubClassEq(RC)) { + MachineInstrBuilder MIB = AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VLDMDIA)) .addFrameIndex(FI)) - .addMemOperand(MMO); - MIB = AddDReg(MIB, DestReg, ARM::dsub_0, RegState::Define, TRI); - MIB = AddDReg(MIB, DestReg, ARM::dsub_1, RegState::Define, TRI); - MIB = AddDReg(MIB, DestReg, ARM::dsub_2, RegState::Define, TRI); - MIB = AddDReg(MIB, DestReg, ARM::dsub_3, RegState::Define, TRI); - MIB = AddDReg(MIB, DestReg, ARM::dsub_4, RegState::Define, TRI); - MIB = AddDReg(MIB, DestReg, ARM::dsub_5, RegState::Define, TRI); - MIB = AddDReg(MIB, DestReg, ARM::dsub_6, RegState::Define, TRI); - AddDReg(MIB, DestReg, ARM::dsub_7, RegState::Define, TRI); + .addMemOperand(MMO); + MIB = AddDReg(MIB, DestReg, ARM::dsub_0, RegState::Define, TRI); + MIB = AddDReg(MIB, DestReg, ARM::dsub_1, RegState::Define, TRI); + MIB = AddDReg(MIB, DestReg, ARM::dsub_2, RegState::Define, TRI); + MIB = AddDReg(MIB, DestReg, ARM::dsub_3, RegState::Define, TRI); + MIB = AddDReg(MIB, DestReg, ARM::dsub_4, RegState::Define, TRI); + MIB = AddDReg(MIB, DestReg, ARM::dsub_5, RegState::Define, TRI); + MIB = AddDReg(MIB, DestReg, ARM::dsub_6, RegState::Define, TRI); + MIB = AddDReg(MIB, DestReg, ARM::dsub_7, RegState::Define, TRI); + MIB.addReg(DestReg, RegState::Define | RegState::Implicit); + } else + llvm_unreachable("Unknown reg class!"); break; - } default: llvm_unreachable("Unknown regclass!"); } @@ -944,6 +990,78 @@ ARMBaseInstrInfo::isLoadFromStackSlot(const MachineInstr *MI, return 0; } +unsigned ARMBaseInstrInfo::isLoadFromStackSlotPostFE(const MachineInstr *MI, + int &FrameIndex) const { + const MachineMemOperand *Dummy; + return MI->getDesc().mayLoad() && hasLoadFromStackSlot(MI, Dummy, FrameIndex); +} + +bool ARMBaseInstrInfo::expandPostRAPseudo(MachineBasicBlock::iterator MI) const{ + // This hook gets to expand COPY instructions before they become + // copyPhysReg() calls. Look for VMOVS instructions that can legally be + // widened to VMOVD. We prefer the VMOVD when possible because it may be + // changed into a VORR that can go down the NEON pipeline. + if (!WidenVMOVS || !MI->isCopy()) + return false; + + // Look for a copy between even S-registers. That is where we keep floats + // when using NEON v2f32 instructions for f32 arithmetic. + unsigned DstRegS = MI->getOperand(0).getReg(); + unsigned SrcRegS = MI->getOperand(1).getReg(); + if (!ARM::SPRRegClass.contains(DstRegS, SrcRegS)) + return false; + + const TargetRegisterInfo *TRI = &getRegisterInfo(); + unsigned DstRegD = TRI->getMatchingSuperReg(DstRegS, ARM::ssub_0, + &ARM::DPRRegClass); + unsigned SrcRegD = TRI->getMatchingSuperReg(SrcRegS, ARM::ssub_0, + &ARM::DPRRegClass); + if (!DstRegD || !SrcRegD) + return false; + + // We want to widen this into a DstRegD = VMOVD SrcRegD copy. This is only + // legal if the COPY already defines the full DstRegD, and it isn't a + // sub-register insertion. + if (!MI->definesRegister(DstRegD, TRI) || MI->readsRegister(DstRegD, TRI)) + return false; + + // A dead copy shouldn't show up here, but reject it just in case. + if (MI->getOperand(0).isDead()) + return false; + + // All clear, widen the COPY. + DEBUG(dbgs() << "widening: " << *MI); + + // Get rid of the old of DstRegD. Leave it if it defines a Q-reg + // or some other super-register. + int ImpDefIdx = MI->findRegisterDefOperandIdx(DstRegD); + if (ImpDefIdx != -1) + MI->RemoveOperand(ImpDefIdx); + + // Change the opcode and operands. + MI->setDesc(get(ARM::VMOVD)); + MI->getOperand(0).setReg(DstRegD); + MI->getOperand(1).setReg(SrcRegD); + AddDefaultPred(MachineInstrBuilder(MI)); + + // We are now reading SrcRegD instead of SrcRegS. This may upset the + // register scavenger and machine verifier, so we need to indicate that we + // are reading an undefined value from SrcRegD, but a proper value from + // SrcRegS. + MI->getOperand(1).setIsUndef(); + MachineInstrBuilder(MI).addReg(SrcRegS, RegState::Implicit); + + // SrcRegD may actually contain an unrelated value in the ssub_1 + // sub-register. Don't kill it. Only kill the ssub_0 sub-register. + if (MI->getOperand(1).isKill()) { + MI->getOperand(1).setIsKill(false); + MI->addRegisterKilled(SrcRegS, TRI, true); + } + + DEBUG(dbgs() << "replaced by: " << *MI); + return true; +} + MachineInstr* ARMBaseInstrInfo::emitFrameIndexDebugValue(MachineFunction &MF, int FrameIx, uint64_t Offset, @@ -974,17 +1092,24 @@ static unsigned duplicateCPV(MachineFunction &MF, unsigned &CPI) { // instructions, so that's probably OK, but is PIC always correct when // we get here? if (ACPV->isGlobalValue()) - NewCPV = new ARMConstantPoolValue(ACPV->getGV(), PCLabelId, - ARMCP::CPValue, 4); + NewCPV = ARMConstantPoolConstant:: + Create(cast(ACPV)->getGV(), PCLabelId, + ARMCP::CPValue, 4); else if (ACPV->isExtSymbol()) - NewCPV = new ARMConstantPoolValue(MF.getFunction()->getContext(), - ACPV->getSymbol(), PCLabelId, 4); + NewCPV = ARMConstantPoolSymbol:: + Create(MF.getFunction()->getContext(), + cast(ACPV)->getSymbol(), PCLabelId, 4); else if (ACPV->isBlockAddress()) - NewCPV = new ARMConstantPoolValue(ACPV->getBlockAddress(), PCLabelId, - ARMCP::CPBlockAddress, 4); + NewCPV = ARMConstantPoolConstant:: + Create(cast(ACPV)->getBlockAddress(), PCLabelId, + ARMCP::CPBlockAddress, 4); else if (ACPV->isLSDA()) - NewCPV = new ARMConstantPoolValue(MF.getFunction(), PCLabelId, - ARMCP::CPLSDA, 4); + NewCPV = ARMConstantPoolConstant::Create(MF.getFunction(), PCLabelId, + ARMCP::CPLSDA, 4); + else if (ACPV->isMachineBasicBlock()) + NewCPV = ARMConstantPoolMBB:: + Create(MF.getFunction()->getContext(), + cast(ACPV)->getMBB(), PCLabelId, 4); else llvm_unreachable("Unexpected ARM constantpool value type!!"); CPI = MCP->getConstantPoolIndex(NewCPV, MCPE.getAlignment()); @@ -1289,7 +1414,7 @@ isProfitableToIfCvt(MachineBasicBlock &TMBB, // Attempt to estimate the relative costs of predication versus branching. unsigned TUnpredCost = Probability.getNumerator() * TCycles; TUnpredCost /= Probability.getDenominator(); - + uint32_t Comp = Probability.getDenominator() - Probability.getNumerator(); unsigned FUnpredCost = Comp * FCycles; FUnpredCost /= Probability.getDenominator(); @@ -1330,6 +1455,57 @@ int llvm::getMatchingCondBranchOpcode(int Opc) { } +/// Map pseudo instructions that imply an 'S' bit onto real opcodes. Whether the +/// instruction is encoded with an 'S' bit is determined by the optional CPSR +/// def operand. +/// +/// This will go away once we can teach tblgen how to set the optional CPSR def +/// operand itself. +struct AddSubFlagsOpcodePair { + unsigned PseudoOpc; + unsigned MachineOpc; +}; + +static AddSubFlagsOpcodePair AddSubFlagsOpcodeMap[] = { + {ARM::ADDSri, ARM::ADDri}, + {ARM::ADDSrr, ARM::ADDrr}, + {ARM::ADDSrsi, ARM::ADDrsi}, + {ARM::ADDSrsr, ARM::ADDrsr}, + + {ARM::SUBSri, ARM::SUBri}, + {ARM::SUBSrr, ARM::SUBrr}, + {ARM::SUBSrsi, ARM::SUBrsi}, + {ARM::SUBSrsr, ARM::SUBrsr}, + + {ARM::RSBSri, ARM::RSBri}, + {ARM::RSBSrr, ARM::RSBrr}, + {ARM::RSBSrsi, ARM::RSBrsi}, + {ARM::RSBSrsr, ARM::RSBrsr}, + + {ARM::t2ADDSri, ARM::t2ADDri}, + {ARM::t2ADDSrr, ARM::t2ADDrr}, + {ARM::t2ADDSrs, ARM::t2ADDrs}, + + {ARM::t2SUBSri, ARM::t2SUBri}, + {ARM::t2SUBSrr, ARM::t2SUBrr}, + {ARM::t2SUBSrs, ARM::t2SUBrs}, + + {ARM::t2RSBSri, ARM::t2RSBri}, + {ARM::t2RSBSrs, ARM::t2RSBrs}, +}; + +unsigned llvm::convertAddSubFlagsOpcode(unsigned OldOpc) { + static const int NPairs = + sizeof(AddSubFlagsOpcodeMap) / sizeof(AddSubFlagsOpcodePair); + for (AddSubFlagsOpcodePair *OpcPair = &AddSubFlagsOpcodeMap[0], + *End = &AddSubFlagsOpcodeMap[NPairs]; OpcPair != End; ++OpcPair) { + if (OldOpc == OpcPair->PseudoOpc) { + return OpcPair->MachineOpc; + } + } + return 0; +} + void llvm::emitARMRegPlusImmediate(MachineBasicBlock &MBB, MachineBasicBlock::iterator &MBBI, DebugLoc dl, unsigned DestReg, unsigned BaseReg, int NumBytes, @@ -1862,7 +2038,6 @@ ARMBaseInstrInfo::getNumMicroOps(const InstrItineraryData *ItinData, case ARM::STMIB_UPD: case ARM::tLDMIA: case ARM::tLDMIA_UPD: - case ARM::tSTMIA: case ARM::tSTMIA_UPD: case ARM::tPOP_RET: case ARM::tPOP: @@ -2128,7 +2303,6 @@ ARMBaseInstrInfo::getOperandLatency(const InstrItineraryData *ItinData, case ARM::STMDA_UPD: case ARM::STMDB_UPD: case ARM::STMIB_UPD: - case ARM::tSTMIA: case ARM::tSTMIA_UPD: case ARM::tPOP_RET: case ARM::tPOP: @@ -2567,6 +2741,15 @@ hasLowDefLatency(const InstrItineraryData *ItinData, return false; } +bool ARMBaseInstrInfo::verifyInstruction(const MachineInstr *MI, + StringRef &ErrInfo) const { + if (convertAddSubFlagsOpcode(MI->getOpcode())) { + ErrInfo = "Pseudo flag setting opcodes only exist in Selection DAG"; + return false; + } + return true; +} + bool ARMBaseInstrInfo::isFpMLxInstruction(unsigned Opcode, unsigned &MulOpc, unsigned &AddSubOpc, @@ -2582,3 +2765,66 @@ ARMBaseInstrInfo::isFpMLxInstruction(unsigned Opcode, unsigned &MulOpc, HasLane = Entry.HasLane; return true; } + +//===----------------------------------------------------------------------===// +// Execution domains. +//===----------------------------------------------------------------------===// +// +// Some instructions go down the NEON pipeline, some go down the VFP pipeline, +// and some can go down both. The vmov instructions go down the VFP pipeline, +// but they can be changed to vorr equivalents that are executed by the NEON +// pipeline. +// +// We use the following execution domain numbering: +// +enum ARMExeDomain { + ExeGeneric = 0, + ExeVFP = 1, + ExeNEON = 2 +}; +// +// Also see ARMInstrFormats.td and Domain* enums in ARMBaseInfo.h +// +std::pair +ARMBaseInstrInfo::getExecutionDomain(const MachineInstr *MI) const { + // VMOVD is a VFP instruction, but can be changed to NEON if it isn't + // predicated. + if (MI->getOpcode() == ARM::VMOVD && !isPredicated(MI)) + return std::make_pair(ExeVFP, (1<getDesc().TSFlags & ARMII::DomainMask; + + if (Domain & ARMII::DomainNEON) + return std::make_pair(ExeNEON, 0); + + // Certain instructions can go either way on Cortex-A8. + // Treat them as NEON instructions. + if ((Domain & ARMII::DomainNEONA8) && Subtarget.isCortexA8()) + return std::make_pair(ExeNEON, 0); + + if (Domain & ARMII::DomainVFP) + return std::make_pair(ExeVFP, 0); + + return std::make_pair(ExeGeneric, 0); +} + +void +ARMBaseInstrInfo::setExecutionDomain(MachineInstr *MI, unsigned Domain) const { + // We only know how to change VMOVD into VORR. + assert(MI->getOpcode() == ARM::VMOVD && "Can only swizzle VMOVD"); + if (Domain != ExeNEON) + return; + + // Zap the predicate operands. + assert(!isPredicated(MI) && "Cannot predicate a VORRd"); + MI->RemoveOperand(3); + MI->RemoveOperand(2); + + // Change to a VORRd which requires two identical use operands. + MI->setDesc(get(ARM::VORRd)); + + // Add the extra source operand and new predicates. + // This will go before any implicit ops. + AddDefaultPred(MachineInstrBuilder(MI).addOperand(MI->getOperand(1))); +} diff --git a/lib/Target/ARM/ARMBaseInstrInfo.h b/lib/Target/ARM/ARMBaseInstrInfo.h index 507e8974bf7b2..0f9f32179a310 100644 --- a/lib/Target/ARM/ARMBaseInstrInfo.h +++ b/lib/Target/ARM/ARMBaseInstrInfo.h @@ -27,146 +27,6 @@ namespace llvm { class ARMSubtarget; class ARMBaseRegisterInfo; -/// ARMII - This namespace holds all of the target specific flags that -/// instruction info tracks. -/// -namespace ARMII { - enum { - //===------------------------------------------------------------------===// - // Instruction Flags. - - //===------------------------------------------------------------------===// - // This four-bit field describes the addressing mode used. - AddrModeMask = 0x1f, // The AddrMode enums are declared in ARMBaseInfo.h - - // IndexMode - Unindex, pre-indexed, or post-indexed are valid for load - // and store ops only. Generic "updating" flag is used for ld/st multiple. - // The index mode enums are declared in ARMBaseInfo.h - IndexModeShift = 5, - IndexModeMask = 3 << IndexModeShift, - - //===------------------------------------------------------------------===// - // Instruction encoding formats. - // - FormShift = 7, - FormMask = 0x3f << FormShift, - - // Pseudo instructions - Pseudo = 0 << FormShift, - - // Multiply instructions - MulFrm = 1 << FormShift, - - // Branch instructions - BrFrm = 2 << FormShift, - BrMiscFrm = 3 << FormShift, - - // Data Processing instructions - DPFrm = 4 << FormShift, - DPSoRegFrm = 5 << FormShift, - - // Load and Store - LdFrm = 6 << FormShift, - StFrm = 7 << FormShift, - LdMiscFrm = 8 << FormShift, - StMiscFrm = 9 << FormShift, - LdStMulFrm = 10 << FormShift, - - LdStExFrm = 11 << FormShift, - - // Miscellaneous arithmetic instructions - ArithMiscFrm = 12 << FormShift, - SatFrm = 13 << FormShift, - - // Extend instructions - ExtFrm = 14 << FormShift, - - // VFP formats - VFPUnaryFrm = 15 << FormShift, - VFPBinaryFrm = 16 << FormShift, - VFPConv1Frm = 17 << FormShift, - VFPConv2Frm = 18 << FormShift, - VFPConv3Frm = 19 << FormShift, - VFPConv4Frm = 20 << FormShift, - VFPConv5Frm = 21 << FormShift, - VFPLdStFrm = 22 << FormShift, - VFPLdStMulFrm = 23 << FormShift, - VFPMiscFrm = 24 << FormShift, - - // Thumb format - ThumbFrm = 25 << FormShift, - - // Miscelleaneous format - MiscFrm = 26 << FormShift, - - // NEON formats - NGetLnFrm = 27 << FormShift, - NSetLnFrm = 28 << FormShift, - NDupFrm = 29 << FormShift, - NLdStFrm = 30 << FormShift, - N1RegModImmFrm= 31 << FormShift, - N2RegFrm = 32 << FormShift, - NVCVTFrm = 33 << FormShift, - NVDupLnFrm = 34 << FormShift, - N2RegVShLFrm = 35 << FormShift, - N2RegVShRFrm = 36 << FormShift, - N3RegFrm = 37 << FormShift, - N3RegVShFrm = 38 << FormShift, - NVExtFrm = 39 << FormShift, - NVMulSLFrm = 40 << FormShift, - NVTBLFrm = 41 << FormShift, - - //===------------------------------------------------------------------===// - // Misc flags. - - // UnaryDP - Indicates this is a unary data processing instruction, i.e. - // it doesn't have a Rn operand. - UnaryDP = 1 << 13, - - // Xform16Bit - Indicates this Thumb2 instruction may be transformed into - // a 16-bit Thumb instruction if certain conditions are met. - Xform16Bit = 1 << 14, - - //===------------------------------------------------------------------===// - // Code domain. - DomainShift = 15, - DomainMask = 7 << DomainShift, - DomainGeneral = 0 << DomainShift, - DomainVFP = 1 << DomainShift, - DomainNEON = 2 << DomainShift, - DomainNEONA8 = 4 << DomainShift, - - //===------------------------------------------------------------------===// - // Field shifts - such shifts are used to set field while generating - // machine instructions. - // - // FIXME: This list will need adjusting/fixing as the MC code emitter - // takes shape and the ARMCodeEmitter.cpp bits go away. - ShiftTypeShift = 4, - - M_BitShift = 5, - ShiftImmShift = 5, - ShiftShift = 7, - N_BitShift = 7, - ImmHiShift = 8, - SoRotImmShift = 8, - RegRsShift = 8, - ExtRotImmShift = 10, - RegRdLoShift = 12, - RegRdShift = 12, - RegRdHiShift = 16, - RegRnShift = 16, - S_BitShift = 20, - W_BitShift = 21, - AM3_I_BitShift = 22, - D_BitShift = 22, - U_BitShift = 23, - P_BitShift = 24, - I_BitShift = 25, - CondShift = 28 - }; -} - class ARMBaseInstrInfo : public ARMGenInstrInfo { const ARMSubtarget &Subtarget; @@ -241,6 +101,10 @@ public: int &FrameIndex) const; virtual unsigned isStoreToStackSlot(const MachineInstr *MI, int &FrameIndex) const; + virtual unsigned isLoadFromStackSlotPostFE(const MachineInstr *MI, + int &FrameIndex) const; + virtual unsigned isStoreToStackSlotPostFE(const MachineInstr *MI, + int &FrameIndex) const; virtual void copyPhysReg(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, DebugLoc DL, @@ -259,6 +123,8 @@ public: const TargetRegisterClass *RC, const TargetRegisterInfo *TRI) const; + virtual bool expandPostRAPseudo(MachineBasicBlock::iterator MI) const; + virtual MachineInstr *emitFrameIndexDebugValue(MachineFunction &MF, int FrameIx, uint64_t Offset, @@ -346,6 +212,12 @@ public: int getOperandLatency(const InstrItineraryData *ItinData, SDNode *DefNode, unsigned DefIdx, SDNode *UseNode, unsigned UseIdx) const; + + /// VFP/NEON execution domains. + std::pair + getExecutionDomain(const MachineInstr *MI) const; + void setExecutionDomain(MachineInstr *MI, unsigned Domain) const; + private: int getVLDMDefCycle(const InstrItineraryData *ItinData, const MCInstrDesc &DefMCID, @@ -382,6 +254,9 @@ private: bool hasLowDefLatency(const InstrItineraryData *ItinData, const MachineInstr *DefMI, unsigned DefIdx) const; + /// verifyInstruction - Perform target specific instruction verification. + bool verifyInstruction(const MachineInstr *MI, StringRef &ErrInfo) const; + private: /// Modeling special VFP / NEON fp MLA / MLS hazards. @@ -464,6 +339,12 @@ ARMCC::CondCodes getInstrPredicate(const MachineInstr *MI, unsigned &PredReg); int getMatchingCondBranchOpcode(int Opc); + +/// Map pseudo instructions that imply an 'S' bit onto real opcodes. Whether +/// the instruction is encoded with an 'S' bit is determined by the optional +/// CPSR def operand. +unsigned convertAddSubFlagsOpcode(unsigned OldOpc); + /// emitARMRegPlusImmediate / emitT2RegPlusImmediate - Emits a series of /// instructions to materializea destreg = basereg + immediate in ARM / Thumb2 /// code. diff --git a/lib/Target/ARM/ARMBaseRegisterInfo.cpp b/lib/Target/ARM/ARMBaseRegisterInfo.cpp index ba422952ac1a4..7c42342229a26 100644 --- a/lib/Target/ARM/ARMBaseRegisterInfo.cpp +++ b/lib/Target/ARM/ARMBaseRegisterInfo.cpp @@ -12,13 +12,13 @@ //===----------------------------------------------------------------------===// #include "ARM.h" -#include "ARMAddressingModes.h" #include "ARMBaseInstrInfo.h" #include "ARMBaseRegisterInfo.h" #include "ARMFrameLowering.h" #include "ARMInstrInfo.h" #include "ARMMachineFunctionInfo.h" #include "ARMSubtarget.h" +#include "MCTargetDesc/ARMAddressingModes.h" #include "llvm/Constants.h" #include "llvm/DerivedTypes.h" #include "llvm/Function.h" @@ -27,7 +27,6 @@ #include "llvm/CodeGen/MachineFrameInfo.h" #include "llvm/CodeGen/MachineFunction.h" #include "llvm/CodeGen/MachineInstrBuilder.h" -#include "llvm/CodeGen/MachineLocation.h" #include "llvm/CodeGen/MachineRegisterInfo.h" #include "llvm/CodeGen/RegisterScavenging.h" #include "llvm/Support/Debug.h" @@ -57,7 +56,7 @@ EnableBasePointer("arm-use-base-pointer", cl::Hidden, cl::init(true), ARMBaseRegisterInfo::ARMBaseRegisterInfo(const ARMBaseInstrInfo &tii, const ARMSubtarget &sti) - : ARMGenRegisterInfo(), TII(tii), STI(sti), + : ARMGenRegisterInfo(ARM::LR), TII(tii), STI(sti), FramePtr((STI.isTargetDarwin() || STI.isThumb()) ? ARM::R7 : ARM::R11), BasePtr(ARM::R6) { } @@ -354,7 +353,7 @@ const TargetRegisterClass* ARMBaseRegisterInfo::getLargestLegalSuperClass(const TargetRegisterClass *RC) const { const TargetRegisterClass *Super = RC; - TargetRegisterClass::sc_iterator I = RC->superclasses_begin(); + TargetRegisterClass::sc_iterator I = RC->getSuperClasses(); do { switch (Super->getID()) { case ARM::GPRRegClassID: @@ -375,6 +374,13 @@ ARMBaseRegisterInfo::getPointerRegClass(unsigned Kind) const { return ARM::GPRRegisterClass; } +const TargetRegisterClass * +ARMBaseRegisterInfo::getCrossCopyRegClass(const TargetRegisterClass *RC) const { + if (RC == &ARM::CCRRegClass) + return 0; // Can't copy CCR registers. + return RC; +} + unsigned ARMBaseRegisterInfo::getRegPressureLimit(const TargetRegisterClass *RC, MachineFunction &MF) const { @@ -487,19 +493,19 @@ ARMBaseRegisterInfo::getRawAllocationOrder(const TargetRegisterClass *RC, if (!TFI->hasFP(MF)) { if (!STI.isR9Reserved()) - return ArrayRef(GPREven1); + return makeArrayRef(GPREven1); else - return ArrayRef(GPREven4); + return makeArrayRef(GPREven4); } else if (FramePtr == ARM::R7) { if (!STI.isR9Reserved()) - return ArrayRef(GPREven2); + return makeArrayRef(GPREven2); else - return ArrayRef(GPREven5); + return makeArrayRef(GPREven5); } else { // FramePtr == ARM::R11 if (!STI.isR9Reserved()) - return ArrayRef(GPREven3); + return makeArrayRef(GPREven3); else - return ArrayRef(GPREven6); + return makeArrayRef(GPREven6); } } else if (HintType == ARMRI::RegPairOdd) { if (isPhysicalRegister(HintReg) && getRegisterPairOdd(HintReg, MF) == 0) @@ -509,19 +515,19 @@ ARMBaseRegisterInfo::getRawAllocationOrder(const TargetRegisterClass *RC, if (!TFI->hasFP(MF)) { if (!STI.isR9Reserved()) - return ArrayRef(GPROdd1); + return makeArrayRef(GPROdd1); else - return ArrayRef(GPROdd4); + return makeArrayRef(GPROdd4); } else if (FramePtr == ARM::R7) { if (!STI.isR9Reserved()) - return ArrayRef(GPROdd2); + return makeArrayRef(GPROdd2); else - return ArrayRef(GPROdd5); + return makeArrayRef(GPROdd5); } else { // FramePtr == ARM::R11 if (!STI.isR9Reserved()) - return ArrayRef(GPROdd3); + return makeArrayRef(GPROdd3); else - return ArrayRef(GPROdd6); + return makeArrayRef(GPROdd6); } } return RC->getRawAllocationOrder(MF); @@ -649,10 +655,6 @@ cannotEliminateFrame(const MachineFunction &MF) const { || needsStackRealignment(MF); } -unsigned ARMBaseRegisterInfo::getRARegister() const { - return ARM::LR; -} - unsigned ARMBaseRegisterInfo::getFrameRegister(const MachineFunction &MF) const { const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering(); @@ -672,99 +674,54 @@ unsigned ARMBaseRegisterInfo::getEHHandlerRegister() const { return 0; } -int ARMBaseRegisterInfo::getDwarfRegNum(unsigned RegNum, bool isEH) const { - return ARMGenRegisterInfo::getDwarfRegNumFull(RegNum, 0); -} - -int ARMBaseRegisterInfo::getLLVMRegNum(unsigned DwarfRegNo, bool isEH) const { - return ARMGenRegisterInfo::getLLVMRegNumFull(DwarfRegNo,0); -} - unsigned ARMBaseRegisterInfo::getRegisterPairEven(unsigned Reg, const MachineFunction &MF) const { switch (Reg) { default: break; // Return 0 if either register of the pair is a special register. // So no R12, etc. - case ARM::R1: - return ARM::R0; - case ARM::R3: - return ARM::R2; - case ARM::R5: - return ARM::R4; + case ARM::R1: return ARM::R0; + case ARM::R3: return ARM::R2; + case ARM::R5: return ARM::R4; case ARM::R7: return (isReservedReg(MF, ARM::R7) || isReservedReg(MF, ARM::R6)) ? 0 : ARM::R6; - case ARM::R9: - return isReservedReg(MF, ARM::R9) ? 0 :ARM::R8; - case ARM::R11: - return isReservedReg(MF, ARM::R11) ? 0 : ARM::R10; - - case ARM::S1: - return ARM::S0; - case ARM::S3: - return ARM::S2; - case ARM::S5: - return ARM::S4; - case ARM::S7: - return ARM::S6; - case ARM::S9: - return ARM::S8; - case ARM::S11: - return ARM::S10; - case ARM::S13: - return ARM::S12; - case ARM::S15: - return ARM::S14; - case ARM::S17: - return ARM::S16; - case ARM::S19: - return ARM::S18; - case ARM::S21: - return ARM::S20; - case ARM::S23: - return ARM::S22; - case ARM::S25: - return ARM::S24; - case ARM::S27: - return ARM::S26; - case ARM::S29: - return ARM::S28; - case ARM::S31: - return ARM::S30; - - case ARM::D1: - return ARM::D0; - case ARM::D3: - return ARM::D2; - case ARM::D5: - return ARM::D4; - case ARM::D7: - return ARM::D6; - case ARM::D9: - return ARM::D8; - case ARM::D11: - return ARM::D10; - case ARM::D13: - return ARM::D12; - case ARM::D15: - return ARM::D14; - case ARM::D17: - return ARM::D16; - case ARM::D19: - return ARM::D18; - case ARM::D21: - return ARM::D20; - case ARM::D23: - return ARM::D22; - case ARM::D25: - return ARM::D24; - case ARM::D27: - return ARM::D26; - case ARM::D29: - return ARM::D28; - case ARM::D31: - return ARM::D30; + case ARM::R9: return isReservedReg(MF, ARM::R9) ? 0 :ARM::R8; + case ARM::R11: return isReservedReg(MF, ARM::R11) ? 0 : ARM::R10; + + case ARM::S1: return ARM::S0; + case ARM::S3: return ARM::S2; + case ARM::S5: return ARM::S4; + case ARM::S7: return ARM::S6; + case ARM::S9: return ARM::S8; + case ARM::S11: return ARM::S10; + case ARM::S13: return ARM::S12; + case ARM::S15: return ARM::S14; + case ARM::S17: return ARM::S16; + case ARM::S19: return ARM::S18; + case ARM::S21: return ARM::S20; + case ARM::S23: return ARM::S22; + case ARM::S25: return ARM::S24; + case ARM::S27: return ARM::S26; + case ARM::S29: return ARM::S28; + case ARM::S31: return ARM::S30; + + case ARM::D1: return ARM::D0; + case ARM::D3: return ARM::D2; + case ARM::D5: return ARM::D4; + case ARM::D7: return ARM::D6; + case ARM::D9: return ARM::D8; + case ARM::D11: return ARM::D10; + case ARM::D13: return ARM::D12; + case ARM::D15: return ARM::D14; + case ARM::D17: return ARM::D16; + case ARM::D19: return ARM::D18; + case ARM::D21: return ARM::D20; + case ARM::D23: return ARM::D22; + case ARM::D25: return ARM::D24; + case ARM::D27: return ARM::D26; + case ARM::D29: return ARM::D28; + case ARM::D31: return ARM::D30; } return 0; @@ -776,85 +733,48 @@ unsigned ARMBaseRegisterInfo::getRegisterPairOdd(unsigned Reg, default: break; // Return 0 if either register of the pair is a special register. // So no R12, etc. - case ARM::R0: - return ARM::R1; - case ARM::R2: - return ARM::R3; - case ARM::R4: - return ARM::R5; + case ARM::R0: return ARM::R1; + case ARM::R2: return ARM::R3; + case ARM::R4: return ARM::R5; case ARM::R6: return (isReservedReg(MF, ARM::R7) || isReservedReg(MF, ARM::R6)) ? 0 : ARM::R7; - case ARM::R8: - return isReservedReg(MF, ARM::R9) ? 0 :ARM::R9; - case ARM::R10: - return isReservedReg(MF, ARM::R11) ? 0 : ARM::R11; - - case ARM::S0: - return ARM::S1; - case ARM::S2: - return ARM::S3; - case ARM::S4: - return ARM::S5; - case ARM::S6: - return ARM::S7; - case ARM::S8: - return ARM::S9; - case ARM::S10: - return ARM::S11; - case ARM::S12: - return ARM::S13; - case ARM::S14: - return ARM::S15; - case ARM::S16: - return ARM::S17; - case ARM::S18: - return ARM::S19; - case ARM::S20: - return ARM::S21; - case ARM::S22: - return ARM::S23; - case ARM::S24: - return ARM::S25; - case ARM::S26: - return ARM::S27; - case ARM::S28: - return ARM::S29; - case ARM::S30: - return ARM::S31; - - case ARM::D0: - return ARM::D1; - case ARM::D2: - return ARM::D3; - case ARM::D4: - return ARM::D5; - case ARM::D6: - return ARM::D7; - case ARM::D8: - return ARM::D9; - case ARM::D10: - return ARM::D11; - case ARM::D12: - return ARM::D13; - case ARM::D14: - return ARM::D15; - case ARM::D16: - return ARM::D17; - case ARM::D18: - return ARM::D19; - case ARM::D20: - return ARM::D21; - case ARM::D22: - return ARM::D23; - case ARM::D24: - return ARM::D25; - case ARM::D26: - return ARM::D27; - case ARM::D28: - return ARM::D29; - case ARM::D30: - return ARM::D31; + case ARM::R8: return isReservedReg(MF, ARM::R9) ? 0 :ARM::R9; + case ARM::R10: return isReservedReg(MF, ARM::R11) ? 0 : ARM::R11; + + case ARM::S0: return ARM::S1; + case ARM::S2: return ARM::S3; + case ARM::S4: return ARM::S5; + case ARM::S6: return ARM::S7; + case ARM::S8: return ARM::S9; + case ARM::S10: return ARM::S11; + case ARM::S12: return ARM::S13; + case ARM::S14: return ARM::S15; + case ARM::S16: return ARM::S17; + case ARM::S18: return ARM::S19; + case ARM::S20: return ARM::S21; + case ARM::S22: return ARM::S23; + case ARM::S24: return ARM::S25; + case ARM::S26: return ARM::S27; + case ARM::S28: return ARM::S29; + case ARM::S30: return ARM::S31; + + case ARM::D0: return ARM::D1; + case ARM::D2: return ARM::D3; + case ARM::D4: return ARM::D5; + case ARM::D6: return ARM::D7; + case ARM::D8: return ARM::D9; + case ARM::D10: return ARM::D11; + case ARM::D12: return ARM::D13; + case ARM::D14: return ARM::D15; + case ARM::D16: return ARM::D17; + case ARM::D18: return ARM::D19; + case ARM::D20: return ARM::D21; + case ARM::D22: return ARM::D23; + case ARM::D24: return ARM::D25; + case ARM::D26: return ARM::D27; + case ARM::D28: return ARM::D29; + case ARM::D30: return ARM::D31; } return 0; @@ -1111,11 +1031,11 @@ materializeFrameBaseRegister(MachineBasicBlock *MBB, MachineRegisterInfo &MRI = MBB->getParent()->getRegInfo(); MRI.constrainRegClass(BaseReg, TII.getRegClass(MCID, 0, this)); - MachineInstrBuilder MIB = BuildMI(*MBB, Ins, DL, MCID, BaseReg) - .addFrameIndex(FrameIdx).addImm(Offset); + MachineInstrBuilder MIB = AddDefaultPred(BuildMI(*MBB, Ins, DL, MCID, BaseReg) + .addFrameIndex(FrameIdx).addImm(Offset)); if (!AFI->isThumb1OnlyFunction()) - AddDefaultCC(AddDefaultPred(MIB)); + AddDefaultCC(MIB); } void @@ -1143,6 +1063,7 @@ ARMBaseRegisterInfo::resolveFrameIndex(MachineBasicBlock::iterator I, Done = rewriteT2FrameIndex(MI, i, BaseReg, Off, TII); } assert (Done && "Unable to resolve frame index!"); + (void)Done; } bool ARMBaseRegisterInfo::isFrameOffsetLegal(const MachineInstr *MI, diff --git a/lib/Target/ARM/ARMBaseRegisterInfo.h b/lib/Target/ARM/ARMBaseRegisterInfo.h index b4b4059e73611..fee17ff3c1ca1 100644 --- a/lib/Target/ARM/ARMBaseRegisterInfo.h +++ b/lib/Target/ARM/ARMBaseRegisterInfo.h @@ -33,19 +33,6 @@ namespace ARMRI { }; } -/// isARMLowRegister - Returns true if the register is low register r0-r7. -/// -static inline bool isARMLowRegister(unsigned Reg) { - using namespace ARM; - switch (Reg) { - case R0: case R1: case R2: case R3: - case R4: case R5: case R6: case R7: - return true; - default: - return false; - } -} - /// isARMArea1Register - Returns true if the register is a low register (r0-r7) /// or a stack/pc register that we should push/pop. static inline bool isARMArea1Register(unsigned Reg, bool isDarwin) { @@ -129,6 +116,8 @@ public: unsigned &NewSubIdx) const; const TargetRegisterClass *getPointerRegClass(unsigned Kind = 0) const; + const TargetRegisterClass* + getCrossCopyRegClass(const TargetRegisterClass *RC) const; const TargetRegisterClass* getLargestLegalSuperClass(const TargetRegisterClass *RC) const; @@ -164,7 +153,6 @@ public: bool cannotEliminateFrame(const MachineFunction &MF) const; // Debug information queries. - unsigned getRARegister() const; unsigned getFrameRegister(const MachineFunction &MF) const; unsigned getBaseRegister() const { return BasePtr; } @@ -172,9 +160,6 @@ public: unsigned getEHExceptionRegister() const; unsigned getEHHandlerRegister() const; - int getDwarfRegNum(unsigned RegNum, bool isEH) const; - int getLLVMRegNum(unsigned RegNum, bool isEH) const; - bool isLowRegister(unsigned Reg) const; diff --git a/lib/Target/ARM/ARMCodeEmitter.cpp b/lib/Target/ARM/ARMCodeEmitter.cpp index d6fca62775010..4148d4ab10e97 100644 --- a/lib/Target/ARM/ARMCodeEmitter.cpp +++ b/lib/Target/ARM/ARMCodeEmitter.cpp @@ -14,12 +14,12 @@ #define DEBUG_TYPE "jit" #include "ARM.h" -#include "ARMAddressingModes.h" #include "ARMConstantPoolValue.h" #include "ARMInstrInfo.h" #include "ARMRelocations.h" #include "ARMSubtarget.h" #include "ARMTargetMachine.h" +#include "MCTargetDesc/ARMAddressingModes.h" #include "llvm/Constants.h" #include "llvm/DerivedTypes.h" #include "llvm/Function.h" @@ -161,11 +161,11 @@ namespace { // are already handled elsewhere. They are placeholders to allow this // encoder to continue to function until the MC encoder is sufficiently // far along that this one can be eliminated entirely. - unsigned NEONThumb2DataIPostEncoder(const MachineInstr &MI, unsigned Val) + unsigned NEONThumb2DataIPostEncoder(const MachineInstr &MI, unsigned Val) const { return 0; } - unsigned NEONThumb2LoadStorePostEncoder(const MachineInstr &MI,unsigned Val) + unsigned NEONThumb2LoadStorePostEncoder(const MachineInstr &MI,unsigned Val) const { return 0; } - unsigned NEONThumb2DupPostEncoder(const MachineInstr &MI,unsigned Val) + unsigned NEONThumb2DupPostEncoder(const MachineInstr &MI,unsigned Val) const { return 0; } unsigned VFPThumb2PostEncoder(const MachineInstr&MI, unsigned Val) const { return 0; } @@ -189,13 +189,17 @@ namespace { unsigned Op) const { return 0; } unsigned getARMBranchTargetOpValue(const MachineInstr &MI, unsigned Op) const { return 0; } + unsigned getARMBLXTargetOpValue(const MachineInstr &MI, unsigned Op) + const { return 0; } unsigned getCCOutOpValue(const MachineInstr &MI, unsigned Op) const { return 0; } unsigned getSOImmOpValue(const MachineInstr &MI, unsigned Op) const { return 0; } unsigned getT2SOImmOpValue(const MachineInstr &MI, unsigned Op) const { return 0; } - unsigned getSORegOpValue(const MachineInstr &MI, unsigned Op) + unsigned getSORegRegOpValue(const MachineInstr &MI, unsigned Op) + const { return 0; } + unsigned getSORegImmOpValue(const MachineInstr &MI, unsigned Op) const { return 0; } unsigned getThumbAddrModeRegRegOpValue(const MachineInstr &MI, unsigned Op) const { return 0; } @@ -203,8 +207,12 @@ namespace { const { return 0; } unsigned getT2AddrModeImm8OpValue(const MachineInstr &MI, unsigned Op) const { return 0; } + unsigned getT2Imm8s4OpValue(const MachineInstr &MI, unsigned Op) + const { return 0; } unsigned getT2AddrModeImm8s4OpValue(const MachineInstr &MI, unsigned Op) const { return 0; } + unsigned getT2AddrModeImm0_1020s4OpValue(const MachineInstr &MI,unsigned Op) + const { return 0; } unsigned getT2AddrModeImm8OffsetOpValue(const MachineInstr &MI, unsigned Op) const { return 0; } unsigned getT2AddrModeImm12OffsetOpValue(const MachineInstr &MI,unsigned Op) @@ -213,10 +221,6 @@ namespace { const { return 0; } unsigned getT2SORegOpValue(const MachineInstr &MI, unsigned Op) const { return 0; } - unsigned getRotImmOpValue(const MachineInstr &MI, unsigned Op) - const { return 0; } - unsigned getImmMinusOneOpValue(const MachineInstr &MI, unsigned Op) - const { return 0; } unsigned getT2AdrLabelOpValue(const MachineInstr &MI, unsigned Op) const { return 0; } unsigned getAddrMode6AddressOpValue(const MachineInstr &MI, unsigned Op) @@ -230,8 +234,6 @@ namespace { const { return 0; } unsigned getBitfieldInvertedMaskOpValue(const MachineInstr &MI, unsigned Op) const { return 0; } - unsigned getMsbOpValue(const MachineInstr &MI, - unsigned Op) const { return 0; } unsigned getSsatBitPosValue(const MachineInstr &MI, unsigned Op) const { return 0; } uint32_t getLdStmModeOpValue(const MachineInstr &MI, unsigned OpIdx) @@ -268,6 +270,8 @@ namespace { const { return 0;} uint32_t getAddrMode2OffsetOpValue(const MachineInstr &MI, unsigned OpIdx) const { return 0;} + uint32_t getPostIdxRegOpValue(const MachineInstr &MI, unsigned OpIdx) + const { return 0;} uint32_t getAddrMode3OffsetOpValue(const MachineInstr &MI, unsigned OpIdx) const { return 0;} uint32_t getAddrMode3OpValue(const MachineInstr &MI, unsigned Op) @@ -632,15 +636,16 @@ void ARMCodeEmitter::emitConstPoolInstruction(const MachineInstr &MI) { << (void*)MCE.getCurrentPCValue() << " " << *ACPV << '\n'); assert(ACPV->isGlobalValue() && "unsupported constant pool value"); - const GlobalValue *GV = ACPV->getGV(); + const GlobalValue *GV = cast(ACPV)->getGV(); if (GV) { Reloc::Model RelocM = TM.getRelocationModel(); emitGlobalAddress(GV, ARM::reloc_arm_machine_cp_entry, isa(GV), Subtarget->GVIsIndirectSymbol(GV, RelocM), (intptr_t)ACPV); - } else { - emitExternalSymbolAddress(ACPV->getSymbol(), ARM::reloc_arm_absolute); + } else { + const char *Sym = cast(ACPV)->getSymbol(); + emitExternalSymbolAddress(Sym, ARM::reloc_arm_absolute); } emitWordLE(0); } else { @@ -983,7 +988,7 @@ unsigned ARMCodeEmitter::getMachineSoImmOpValue(unsigned SoImm) { unsigned ARMCodeEmitter::getAddrModeSBit(const MachineInstr &MI, const MCInstrDesc &MCID) const { - for (unsigned i = MI.getNumOperands(), e = MCID.getNumOperands(); i >= e; --i){ + for (unsigned i = MI.getNumOperands(), e = MCID.getNumOperands(); i >= e;--i){ const MachineOperand &MO = MI.getOperand(i-1); if (MO.isReg() && MO.isDef() && MO.getReg() == ARM::CPSR) return 1 << ARMII::S_BitShift; diff --git a/lib/Target/ARM/ARMConstantIslandPass.cpp b/lib/Target/ARM/ARMConstantIslandPass.cpp index f45ebdc535002..3e3a4134c7044 100644 --- a/lib/Target/ARM/ARMConstantIslandPass.cpp +++ b/lib/Target/ARM/ARMConstantIslandPass.cpp @@ -15,10 +15,10 @@ #define DEBUG_TYPE "arm-cp-islands" #include "ARM.h" -#include "ARMAddressingModes.h" #include "ARMMachineFunctionInfo.h" #include "ARMInstrInfo.h" #include "Thumb2InstrInfo.h" +#include "MCTargetDesc/ARMAddressingModes.h" #include "llvm/CodeGen/MachineConstantPool.h" #include "llvm/CodeGen/MachineFunctionPass.h" #include "llvm/CodeGen/MachineJumpTableInfo.h" @@ -739,7 +739,11 @@ MachineBasicBlock *ARMConstantIslands::SplitBlockBeforeInstr(MachineInstr *MI) { // There doesn't seem to be meaningful DebugInfo available; this doesn't // correspond to anything in the source. unsigned Opc = isThumb ? (isThumb2 ? ARM::t2B : ARM::tB) : ARM::B; - BuildMI(OrigBB, DebugLoc(), TII->get(Opc)).addMBB(NewBB); + if (!isThumb) + BuildMI(OrigBB, DebugLoc(), TII->get(Opc)).addMBB(NewBB); + else + BuildMI(OrigBB, DebugLoc(), TII->get(Opc)).addMBB(NewBB) + .addImm(ARMCC::AL).addReg(0); ++NumSplit; // Update the CFG. All succs of OrigBB are now succs of NewBB. @@ -1151,7 +1155,11 @@ void ARMConstantIslands::CreateNewWater(unsigned CPUserIndex, // targets will be exchanged, and the altered branch may be out of // range, so the machinery has to know about it. int UncondBr = isThumb ? ((isThumb2) ? ARM::t2B : ARM::tB) : ARM::B; - BuildMI(UserMBB, DebugLoc(), TII->get(UncondBr)).addMBB(NewMBB); + if (!isThumb) + BuildMI(UserMBB, DebugLoc(), TII->get(UncondBr)).addMBB(NewMBB); + else + BuildMI(UserMBB, DebugLoc(), TII->get(UncondBr)).addMBB(NewMBB) + .addImm(ARMCC::AL).addReg(0); unsigned MaxDisp = getUnconditionalBrDisp(UncondBr); ImmBranches.push_back(ImmBranch(&UserMBB->back(), MaxDisp, false, UncondBr)); @@ -1512,7 +1520,11 @@ ARMConstantIslands::FixUpConditionalBr(MachineFunction &MF, ImmBranch &Br) { .addMBB(NextBB).addImm(CC).addReg(CCReg); Br.MI = &MBB->back(); BBSizes[MBB->getNumber()] += TII->GetInstSizeInBytes(&MBB->back()); - BuildMI(MBB, DebugLoc(), TII->get(Br.UncondBr)).addMBB(DestBB); + if (isThumb) + BuildMI(MBB, DebugLoc(), TII->get(Br.UncondBr)).addMBB(DestBB) + .addImm(ARMCC::AL).addReg(0); + else + BuildMI(MBB, DebugLoc(), TII->get(Br.UncondBr)).addMBB(DestBB); BBSizes[MBB->getNumber()] += TII->GetInstSizeInBytes(&MBB->back()); unsigned MaxDisp = getUnconditionalBrDisp(Br.UncondBr); ImmBranches.push_back(ImmBranch(&MBB->back(), MaxDisp, false, Br.UncondBr)); @@ -1891,7 +1903,8 @@ AdjustJTTargetBlockForward(MachineBasicBlock *BB, MachineBasicBlock *JTBB) // There doesn't seem to be meaningful DebugInfo available; this doesn't // correspond directly to anything in the source. assert (isThumb2 && "Adjusting for TB[BH] but not in Thumb2?"); - BuildMI(NewBB, DebugLoc(), TII->get(ARM::t2B)).addMBB(BB); + BuildMI(NewBB, DebugLoc(), TII->get(ARM::t2B)).addMBB(BB) + .addImm(ARMCC::AL).addReg(0); // Update internal data structures to account for the newly inserted MBB. MF.RenumberBlocks(NewBB); diff --git a/lib/Target/ARM/ARMConstantPoolValue.cpp b/lib/Target/ARM/ARMConstantPoolValue.cpp index 165a1d849ad52..aadfd4779db36 100644 --- a/lib/Target/ARM/ARMConstantPoolValue.cpp +++ b/lib/Target/ARM/ARMConstantPoolValue.cpp @@ -17,79 +17,57 @@ #include "llvm/Constants.h" #include "llvm/GlobalValue.h" #include "llvm/Type.h" +#include "llvm/CodeGen/MachineBasicBlock.h" #include "llvm/Support/raw_ostream.h" #include using namespace llvm; -ARMConstantPoolValue::ARMConstantPoolValue(const Constant *cval, unsigned id, - ARMCP::ARMCPKind K, +//===----------------------------------------------------------------------===// +// ARMConstantPoolValue +//===----------------------------------------------------------------------===// + +ARMConstantPoolValue::ARMConstantPoolValue(Type *Ty, unsigned id, + ARMCP::ARMCPKind kind, unsigned char PCAdj, - ARMCP::ARMCPModifier Modif, - bool AddCA) - : MachineConstantPoolValue((const Type*)cval->getType()), - CVal(cval), S(NULL), LabelId(id), Kind(K), PCAdjust(PCAdj), - Modifier(Modif), AddCurrentAddress(AddCA) {} - -ARMConstantPoolValue::ARMConstantPoolValue(LLVMContext &C, - const char *s, unsigned id, + ARMCP::ARMCPModifier modifier, + bool addCurrentAddress) + : MachineConstantPoolValue(Ty), LabelId(id), Kind(kind), + PCAdjust(PCAdj), Modifier(modifier), + AddCurrentAddress(addCurrentAddress) {} + +ARMConstantPoolValue::ARMConstantPoolValue(LLVMContext &C, unsigned id, + ARMCP::ARMCPKind kind, unsigned char PCAdj, - ARMCP::ARMCPModifier Modif, - bool AddCA) - : MachineConstantPoolValue((const Type*)Type::getInt32Ty(C)), - CVal(NULL), S(strdup(s)), LabelId(id), Kind(ARMCP::CPExtSymbol), - PCAdjust(PCAdj), Modifier(Modif), AddCurrentAddress(AddCA) {} - -ARMConstantPoolValue::ARMConstantPoolValue(const GlobalValue *gv, - ARMCP::ARMCPModifier Modif) - : MachineConstantPoolValue((const Type*)Type::getInt32Ty(gv->getContext())), - CVal(gv), S(NULL), LabelId(0), Kind(ARMCP::CPValue), PCAdjust(0), - Modifier(Modif), AddCurrentAddress(false) {} - -const GlobalValue *ARMConstantPoolValue::getGV() const { - return dyn_cast_or_null(CVal); -} + ARMCP::ARMCPModifier modifier, + bool addCurrentAddress) + : MachineConstantPoolValue((Type*)Type::getInt32Ty(C)), + LabelId(id), Kind(kind), PCAdjust(PCAdj), Modifier(modifier), + AddCurrentAddress(addCurrentAddress) {} -const BlockAddress *ARMConstantPoolValue::getBlockAddress() const { - return dyn_cast_or_null(CVal); -} +ARMConstantPoolValue::~ARMConstantPoolValue() {} -static bool CPV_streq(const char *S1, const char *S2) { - if (S1 == S2) - return true; - if (S1 && S2 && strcmp(S1, S2) == 0) - return true; - return false; +const char *ARMConstantPoolValue::getModifierText() const { + switch (Modifier) { + default: llvm_unreachable("Unknown modifier!"); + // FIXME: Are these case sensitive? It'd be nice to lower-case all the + // strings if that's legal. + case ARMCP::no_modifier: return "none"; + case ARMCP::TLSGD: return "tlsgd"; + case ARMCP::GOT: return "GOT"; + case ARMCP::GOTOFF: return "GOTOFF"; + case ARMCP::GOTTPOFF: return "gottpoff"; + case ARMCP::TPOFF: return "tpoff"; + } } int ARMConstantPoolValue::getExistingMachineCPValue(MachineConstantPool *CP, unsigned Alignment) { - unsigned AlignMask = Alignment - 1; - const std::vector Constants = CP->getConstants(); - for (unsigned i = 0, e = Constants.size(); i != e; ++i) { - if (Constants[i].isMachineConstantPoolEntry() && - (Constants[i].getAlignment() & AlignMask) == 0) { - ARMConstantPoolValue *CPV = - (ARMConstantPoolValue *)Constants[i].Val.MachineCPVal; - if (CPV->CVal == CVal && - CPV->LabelId == LabelId && - CPV->PCAdjust == PCAdjust && - CPV_streq(CPV->S, S) && - CPV->Modifier == Modifier) - return i; - } - } - + assert(false && "Shouldn't be calling this directly!"); return -1; } -ARMConstantPoolValue::~ARMConstantPoolValue() { - free((void*)S); -} - void -ARMConstantPoolValue::AddSelectionDAGCSEId(FoldingSetNodeID &ID) { - ID.AddPointer(CVal); - ID.AddPointer(S); +ARMConstantPoolValue::addSelectionDAGCSEId(FoldingSetNodeID &ID) { ID.AddInteger(LabelId); ID.AddInteger(PCAdjust); } @@ -97,9 +75,7 @@ ARMConstantPoolValue::AddSelectionDAGCSEId(FoldingSetNodeID &ID) { bool ARMConstantPoolValue::hasSameValue(ARMConstantPoolValue *ACPV) { if (ACPV->Kind == Kind && - ACPV->CVal == CVal && ACPV->PCAdjust == PCAdjust && - CPV_streq(ACPV->S, S) && ACPV->Modifier == Modifier) { if (ACPV->LabelId == LabelId) return true; @@ -115,12 +91,7 @@ void ARMConstantPoolValue::dump() const { errs() << " " << *this; } - void ARMConstantPoolValue::print(raw_ostream &O) const { - if (CVal) - O << CVal->getName(); - else - O << S; if (Modifier) O << "(" << getModifierText() << ")"; if (PCAdjust != 0) { O << "-(LPC" << LabelId << "+" << (unsigned)PCAdjust; @@ -128,3 +99,221 @@ void ARMConstantPoolValue::print(raw_ostream &O) const { O << ")"; } } + +//===----------------------------------------------------------------------===// +// ARMConstantPoolConstant +//===----------------------------------------------------------------------===// + +ARMConstantPoolConstant::ARMConstantPoolConstant(Type *Ty, + const Constant *C, + unsigned ID, + ARMCP::ARMCPKind Kind, + unsigned char PCAdj, + ARMCP::ARMCPModifier Modifier, + bool AddCurrentAddress) + : ARMConstantPoolValue(Ty, ID, Kind, PCAdj, Modifier, AddCurrentAddress), + CVal(C) {} + +ARMConstantPoolConstant::ARMConstantPoolConstant(const Constant *C, + unsigned ID, + ARMCP::ARMCPKind Kind, + unsigned char PCAdj, + ARMCP::ARMCPModifier Modifier, + bool AddCurrentAddress) + : ARMConstantPoolValue((Type*)C->getType(), ID, Kind, PCAdj, Modifier, + AddCurrentAddress), + CVal(C) {} + +ARMConstantPoolConstant * +ARMConstantPoolConstant::Create(const Constant *C, unsigned ID) { + return new ARMConstantPoolConstant(C, ID, ARMCP::CPValue, 0, + ARMCP::no_modifier, false); +} + +ARMConstantPoolConstant * +ARMConstantPoolConstant::Create(const GlobalValue *GV, + ARMCP::ARMCPModifier Modifier) { + return new ARMConstantPoolConstant((Type*)Type::getInt32Ty(GV->getContext()), + GV, 0, ARMCP::CPValue, 0, + Modifier, false); +} + +ARMConstantPoolConstant * +ARMConstantPoolConstant::Create(const Constant *C, unsigned ID, + ARMCP::ARMCPKind Kind, unsigned char PCAdj) { + return new ARMConstantPoolConstant(C, ID, Kind, PCAdj, + ARMCP::no_modifier, false); +} + +ARMConstantPoolConstant * +ARMConstantPoolConstant::Create(const Constant *C, unsigned ID, + ARMCP::ARMCPKind Kind, unsigned char PCAdj, + ARMCP::ARMCPModifier Modifier, + bool AddCurrentAddress) { + return new ARMConstantPoolConstant(C, ID, Kind, PCAdj, Modifier, + AddCurrentAddress); +} + +const GlobalValue *ARMConstantPoolConstant::getGV() const { + return dyn_cast_or_null(CVal); +} + +const BlockAddress *ARMConstantPoolConstant::getBlockAddress() const { + return dyn_cast_or_null(CVal); +} + +int ARMConstantPoolConstant::getExistingMachineCPValue(MachineConstantPool *CP, + unsigned Alignment) { + unsigned AlignMask = Alignment - 1; + const std::vector Constants = CP->getConstants(); + for (unsigned i = 0, e = Constants.size(); i != e; ++i) { + if (Constants[i].isMachineConstantPoolEntry() && + (Constants[i].getAlignment() & AlignMask) == 0) { + ARMConstantPoolValue *CPV = + (ARMConstantPoolValue *)Constants[i].Val.MachineCPVal; + ARMConstantPoolConstant *APC = dyn_cast(CPV); + if (!APC) continue; + if (APC->CVal == CVal && equals(APC)) + return i; + } + } + + return -1; +} + +bool ARMConstantPoolConstant::hasSameValue(ARMConstantPoolValue *ACPV) { + const ARMConstantPoolConstant *ACPC = dyn_cast(ACPV); + return ACPC && ACPC->CVal == CVal && ARMConstantPoolValue::hasSameValue(ACPV); +} + +void ARMConstantPoolConstant::addSelectionDAGCSEId(FoldingSetNodeID &ID) { + ID.AddPointer(CVal); + ARMConstantPoolValue::addSelectionDAGCSEId(ID); +} + +void ARMConstantPoolConstant::print(raw_ostream &O) const { + O << CVal->getName(); + ARMConstantPoolValue::print(O); +} + +//===----------------------------------------------------------------------===// +// ARMConstantPoolSymbol +//===----------------------------------------------------------------------===// + +ARMConstantPoolSymbol::ARMConstantPoolSymbol(LLVMContext &C, const char *s, + unsigned id, + unsigned char PCAdj, + ARMCP::ARMCPModifier Modifier, + bool AddCurrentAddress) + : ARMConstantPoolValue(C, id, ARMCP::CPExtSymbol, PCAdj, Modifier, + AddCurrentAddress), + S(strdup(s)) {} + +ARMConstantPoolSymbol::~ARMConstantPoolSymbol() { + free((void*)S); +} + +ARMConstantPoolSymbol * +ARMConstantPoolSymbol::Create(LLVMContext &C, const char *s, + unsigned ID, unsigned char PCAdj) { + return new ARMConstantPoolSymbol(C, s, ID, PCAdj, ARMCP::no_modifier, false); +} + +static bool CPV_streq(const char *S1, const char *S2) { + if (S1 == S2) + return true; + if (S1 && S2 && strcmp(S1, S2) == 0) + return true; + return false; +} + +int ARMConstantPoolSymbol::getExistingMachineCPValue(MachineConstantPool *CP, + unsigned Alignment) { + unsigned AlignMask = Alignment - 1; + const std::vector Constants = CP->getConstants(); + for (unsigned i = 0, e = Constants.size(); i != e; ++i) { + if (Constants[i].isMachineConstantPoolEntry() && + (Constants[i].getAlignment() & AlignMask) == 0) { + ARMConstantPoolValue *CPV = + (ARMConstantPoolValue *)Constants[i].Val.MachineCPVal; + ARMConstantPoolSymbol *APS = dyn_cast(CPV); + if (!APS) continue; + + if (CPV_streq(APS->S, S) && equals(APS)) + return i; + } + } + + return -1; +} + +bool ARMConstantPoolSymbol::hasSameValue(ARMConstantPoolValue *ACPV) { + const ARMConstantPoolSymbol *ACPS = dyn_cast(ACPV); + return ACPS && CPV_streq(ACPS->S, S) && + ARMConstantPoolValue::hasSameValue(ACPV); +} + +void ARMConstantPoolSymbol::addSelectionDAGCSEId(FoldingSetNodeID &ID) { + ID.AddPointer(S); + ARMConstantPoolValue::addSelectionDAGCSEId(ID); +} + +void ARMConstantPoolSymbol::print(raw_ostream &O) const { + O << S; + ARMConstantPoolValue::print(O); +} + +//===----------------------------------------------------------------------===// +// ARMConstantPoolMBB +//===----------------------------------------------------------------------===// + +ARMConstantPoolMBB::ARMConstantPoolMBB(LLVMContext &C, + const MachineBasicBlock *mbb, + unsigned id, unsigned char PCAdj, + ARMCP::ARMCPModifier Modifier, + bool AddCurrentAddress) + : ARMConstantPoolValue(C, id, ARMCP::CPMachineBasicBlock, PCAdj, + Modifier, AddCurrentAddress), + MBB(mbb) {} + +ARMConstantPoolMBB *ARMConstantPoolMBB::Create(LLVMContext &C, + const MachineBasicBlock *mbb, + unsigned ID, + unsigned char PCAdj) { + return new ARMConstantPoolMBB(C, mbb, ID, PCAdj, ARMCP::no_modifier, false); +} + +int ARMConstantPoolMBB::getExistingMachineCPValue(MachineConstantPool *CP, + unsigned Alignment) { + unsigned AlignMask = Alignment - 1; + const std::vector Constants = CP->getConstants(); + for (unsigned i = 0, e = Constants.size(); i != e; ++i) { + if (Constants[i].isMachineConstantPoolEntry() && + (Constants[i].getAlignment() & AlignMask) == 0) { + ARMConstantPoolValue *CPV = + (ARMConstantPoolValue *)Constants[i].Val.MachineCPVal; + ARMConstantPoolMBB *APMBB = dyn_cast(CPV); + if (!APMBB) continue; + + if (APMBB->MBB == MBB && equals(APMBB)) + return i; + } + } + + return -1; +} + +bool ARMConstantPoolMBB::hasSameValue(ARMConstantPoolValue *ACPV) { + const ARMConstantPoolMBB *ACPMBB = dyn_cast(ACPV); + return ACPMBB && ACPMBB->MBB == MBB && + ARMConstantPoolValue::hasSameValue(ACPV); +} + +void ARMConstantPoolMBB::addSelectionDAGCSEId(FoldingSetNodeID &ID) { + ID.AddPointer(MBB); + ARMConstantPoolValue::addSelectionDAGCSEId(ID); +} + +void ARMConstantPoolMBB::print(raw_ostream &O) const { + ARMConstantPoolValue::print(O); +} diff --git a/lib/Target/ARM/ARMConstantPoolValue.h b/lib/Target/ARM/ARMConstantPoolValue.h index d008811c40e4e..0d0def32b7d8a 100644 --- a/lib/Target/ARM/ARMConstantPoolValue.h +++ b/lib/Target/ARM/ARMConstantPoolValue.h @@ -20,17 +20,19 @@ namespace llvm { -class Constant; class BlockAddress; +class Constant; class GlobalValue; class LLVMContext; +class MachineBasicBlock; namespace ARMCP { enum ARMCPKind { CPValue, CPExtSymbol, CPBlockAddress, - CPLSDA + CPLSDA, + CPMachineBasicBlock }; enum ARMCPModifier { @@ -47,8 +49,6 @@ namespace ARMCP { /// represent PC-relative displacement between the address of the load /// instruction and the constant being loaded, i.e. (&GV-(LPIC+8)). class ARMConstantPoolValue : public MachineConstantPoolValue { - const Constant *CVal; // Constant being loaded. - const char *S; // ExtSymbol being loaded. unsigned LabelId; // Label id of the load. ARMCP::ARMCPKind Kind; // Kind of constant. unsigned char PCAdjust; // Extra adjustment if constantpool is pc-relative. @@ -56,60 +56,54 @@ class ARMConstantPoolValue : public MachineConstantPoolValue { ARMCP::ARMCPModifier Modifier; // GV modifier i.e. (&GV(modifier)-(LPIC+8)) bool AddCurrentAddress; +protected: + ARMConstantPoolValue(Type *Ty, unsigned id, ARMCP::ARMCPKind Kind, + unsigned char PCAdj, ARMCP::ARMCPModifier Modifier, + bool AddCurrentAddress); + + ARMConstantPoolValue(LLVMContext &C, unsigned id, ARMCP::ARMCPKind Kind, + unsigned char PCAdj, ARMCP::ARMCPModifier Modifier, + bool AddCurrentAddress); public: - ARMConstantPoolValue(const Constant *cval, unsigned id, - ARMCP::ARMCPKind Kind = ARMCP::CPValue, - unsigned char PCAdj = 0, - ARMCP::ARMCPModifier Modifier = ARMCP::no_modifier, - bool AddCurrentAddress = false); - ARMConstantPoolValue(LLVMContext &C, const char *s, unsigned id, - unsigned char PCAdj = 0, - ARMCP::ARMCPModifier Modifier = ARMCP::no_modifier, - bool AddCurrentAddress = false); - ARMConstantPoolValue(const GlobalValue *GV, ARMCP::ARMCPModifier Modifier); - ARMConstantPoolValue(); - ~ARMConstantPoolValue(); + virtual ~ARMConstantPoolValue(); - const GlobalValue *getGV() const; - const char *getSymbol() const { return S; } - const BlockAddress *getBlockAddress() const; ARMCP::ARMCPModifier getModifier() const { return Modifier; } - const char *getModifierText() const { - switch (Modifier) { - default: llvm_unreachable("Unknown modifier!"); - // FIXME: Are these case sensitive? It'd be nice to lower-case all the - // strings if that's legal. - case ARMCP::no_modifier: return "none"; - case ARMCP::TLSGD: return "tlsgd"; - case ARMCP::GOT: return "GOT"; - case ARMCP::GOTOFF: return "GOTOFF"; - case ARMCP::GOTTPOFF: return "gottpoff"; - case ARMCP::TPOFF: return "tpoff"; - } - } + const char *getModifierText() const; bool hasModifier() const { return Modifier != ARMCP::no_modifier; } + bool mustAddCurrentAddress() const { return AddCurrentAddress; } + unsigned getLabelId() const { return LabelId; } unsigned char getPCAdjustment() const { return PCAdjust; } + bool isGlobalValue() const { return Kind == ARMCP::CPValue; } bool isExtSymbol() const { return Kind == ARMCP::CPExtSymbol; } - bool isBlockAddress() { return Kind == ARMCP::CPBlockAddress; } - bool isLSDA() { return Kind == ARMCP::CPLSDA; } + bool isBlockAddress() const { return Kind == ARMCP::CPBlockAddress; } + bool isLSDA() const { return Kind == ARMCP::CPLSDA; } + bool isMachineBasicBlock() const{ return Kind == ARMCP::CPMachineBasicBlock; } virtual unsigned getRelocationInfo() const { return 2; } virtual int getExistingMachineCPValue(MachineConstantPool *CP, unsigned Alignment); - virtual void AddSelectionDAGCSEId(FoldingSetNodeID &ID); + virtual void addSelectionDAGCSEId(FoldingSetNodeID &ID); - /// hasSameValue - Return true if this ARM constpool value - /// can share the same constantpool entry as another ARM constpool value. - bool hasSameValue(ARMConstantPoolValue *ACPV); + /// hasSameValue - Return true if this ARM constpool value can share the same + /// constantpool entry as another ARM constpool value. + virtual bool hasSameValue(ARMConstantPoolValue *ACPV); + + bool equals(const ARMConstantPoolValue *A) const { + return this->LabelId == A->LabelId && + this->PCAdjust == A->PCAdjust && + this->Modifier == A->Modifier; + } + virtual void print(raw_ostream &O) const; void print(raw_ostream *O) const { if (O) print(*O); } - void print(raw_ostream &O) const; void dump() const; + + static bool classof(const ARMConstantPoolValue *) { return true; } }; inline raw_ostream &operator<<(raw_ostream &O, const ARMConstantPoolValue &V) { @@ -117,6 +111,123 @@ inline raw_ostream &operator<<(raw_ostream &O, const ARMConstantPoolValue &V) { return O; } +/// ARMConstantPoolConstant - ARM-specific constant pool values for Constants, +/// Functions, and BlockAddresses. +class ARMConstantPoolConstant : public ARMConstantPoolValue { + const Constant *CVal; // Constant being loaded. + + ARMConstantPoolConstant(const Constant *C, + unsigned ID, + ARMCP::ARMCPKind Kind, + unsigned char PCAdj, + ARMCP::ARMCPModifier Modifier, + bool AddCurrentAddress); + ARMConstantPoolConstant(Type *Ty, const Constant *C, + unsigned ID, + ARMCP::ARMCPKind Kind, + unsigned char PCAdj, + ARMCP::ARMCPModifier Modifier, + bool AddCurrentAddress); + +public: + static ARMConstantPoolConstant *Create(const Constant *C, unsigned ID); + static ARMConstantPoolConstant *Create(const GlobalValue *GV, + ARMCP::ARMCPModifier Modifier); + static ARMConstantPoolConstant *Create(const Constant *C, unsigned ID, + ARMCP::ARMCPKind Kind, + unsigned char PCAdj); + static ARMConstantPoolConstant *Create(const Constant *C, unsigned ID, + ARMCP::ARMCPKind Kind, + unsigned char PCAdj, + ARMCP::ARMCPModifier Modifier, + bool AddCurrentAddress); + + const GlobalValue *getGV() const; + const BlockAddress *getBlockAddress() const; + + virtual int getExistingMachineCPValue(MachineConstantPool *CP, + unsigned Alignment); + + /// hasSameValue - Return true if this ARM constpool value can share the same + /// constantpool entry as another ARM constpool value. + virtual bool hasSameValue(ARMConstantPoolValue *ACPV); + + virtual void addSelectionDAGCSEId(FoldingSetNodeID &ID); + + virtual void print(raw_ostream &O) const; + static bool classof(const ARMConstantPoolValue *APV) { + return APV->isGlobalValue() || APV->isBlockAddress() || APV->isLSDA(); + } + static bool classof(const ARMConstantPoolConstant *) { return true; } +}; + +/// ARMConstantPoolSymbol - ARM-specific constantpool values for external +/// symbols. +class ARMConstantPoolSymbol : public ARMConstantPoolValue { + const char *S; // ExtSymbol being loaded. + + ARMConstantPoolSymbol(LLVMContext &C, const char *s, unsigned id, + unsigned char PCAdj, ARMCP::ARMCPModifier Modifier, + bool AddCurrentAddress); + +public: + ~ARMConstantPoolSymbol(); + + static ARMConstantPoolSymbol *Create(LLVMContext &C, const char *s, + unsigned ID, unsigned char PCAdj); + + const char *getSymbol() const { return S; } + + virtual int getExistingMachineCPValue(MachineConstantPool *CP, + unsigned Alignment); + + virtual void addSelectionDAGCSEId(FoldingSetNodeID &ID); + + /// hasSameValue - Return true if this ARM constpool value can share the same + /// constantpool entry as another ARM constpool value. + virtual bool hasSameValue(ARMConstantPoolValue *ACPV); + + virtual void print(raw_ostream &O) const; + + static bool classof(const ARMConstantPoolValue *ACPV) { + return ACPV->isExtSymbol(); + } + static bool classof(const ARMConstantPoolSymbol *) { return true; } +}; + +/// ARMConstantPoolMBB - ARM-specific constantpool value of a machine basic +/// block. +class ARMConstantPoolMBB : public ARMConstantPoolValue { + const MachineBasicBlock *MBB; // Machine basic block. + + ARMConstantPoolMBB(LLVMContext &C, const MachineBasicBlock *mbb, unsigned id, + unsigned char PCAdj, ARMCP::ARMCPModifier Modifier, + bool AddCurrentAddress); + +public: + static ARMConstantPoolMBB *Create(LLVMContext &C, + const MachineBasicBlock *mbb, + unsigned ID, unsigned char PCAdj); + + const MachineBasicBlock *getMBB() const { return MBB; } + + virtual int getExistingMachineCPValue(MachineConstantPool *CP, + unsigned Alignment); + + virtual void addSelectionDAGCSEId(FoldingSetNodeID &ID); + + /// hasSameValue - Return true if this ARM constpool value can share the same + /// constantpool entry as another ARM constpool value. + virtual bool hasSameValue(ARMConstantPoolValue *ACPV); + + virtual void print(raw_ostream &O) const; + + static bool classof(const ARMConstantPoolValue *ACPV) { + return ACPV->isMachineBasicBlock(); + } + static bool classof(const ARMConstantPoolMBB *) { return true; } +}; + } // End llvm namespace #endif diff --git a/lib/Target/ARM/ARMExpandPseudoInsts.cpp b/lib/Target/ARM/ARMExpandPseudoInsts.cpp index 94b72fdb9a7e2..7872cb90f4e7a 100644 --- a/lib/Target/ARM/ARMExpandPseudoInsts.cpp +++ b/lib/Target/ARM/ARMExpandPseudoInsts.cpp @@ -16,19 +16,24 @@ #define DEBUG_TYPE "arm-pseudo" #include "ARM.h" -#include "ARMAddressingModes.h" #include "ARMBaseInstrInfo.h" #include "ARMBaseRegisterInfo.h" #include "ARMMachineFunctionInfo.h" #include "ARMRegisterInfo.h" +#include "MCTargetDesc/ARMAddressingModes.h" #include "llvm/CodeGen/MachineFrameInfo.h" #include "llvm/CodeGen/MachineFunctionPass.h" #include "llvm/CodeGen/MachineInstrBuilder.h" #include "llvm/Target/TargetFrameLowering.h" #include "llvm/Target/TargetRegisterInfo.h" +#include "llvm/Support/CommandLine.h" #include "llvm/Support/raw_ostream.h" // FIXME: for debug only. remove! using namespace llvm; +static cl::opt +VerifyARMPseudo("verify-arm-pseudo-expand", cl::Hidden, + cl::desc("Verify machine code after expanding ARM pseudos")); + namespace { class ARMExpandPseudo : public MachineFunctionPass { public: @@ -741,8 +746,22 @@ bool ARMExpandPseudo::ExpandMI(MachineBasicBlock &MBB, MI.eraseFromParent(); return true; } - case ARM::MOVCCs: { - BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVs), + case ARM::MOVCCsi: { + BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVsi), + (MI.getOperand(1).getReg())) + .addReg(MI.getOperand(2).getReg(), + getKillRegState(MI.getOperand(2).isKill())) + .addImm(MI.getOperand(3).getImm()) + .addImm(MI.getOperand(4).getImm()) // 'pred' + .addReg(MI.getOperand(5).getReg()) + .addReg(0); // 's' bit + + MI.eraseFromParent(); + return true; + } + + case ARM::MOVCCsr: { + BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVsr), (MI.getOperand(1).getReg())) .addReg(MI.getOperand(2).getReg(), getKillRegState(MI.getOperand(2).isKill())) @@ -837,10 +856,9 @@ bool ARMExpandPseudo::ExpandMI(MachineBasicBlock &MBB, case ARM::MOVsrl_flag: case ARM::MOVsra_flag: { // These are just fancy MOVs insructions. - AddDefaultPred(BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVs), + AddDefaultPred(BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVsi), MI.getOperand(0).getReg()) .addOperand(MI.getOperand(1)) - .addReg(0) .addImm(ARM_AM::getSORegOpc((Opcode == ARM::MOVsrl_flag ? ARM_AM::lsr : ARM_AM::asr), 1))) @@ -851,10 +869,9 @@ bool ARMExpandPseudo::ExpandMI(MachineBasicBlock &MBB, case ARM::RRX: { // This encodes as "MOVs Rd, Rm, rrx MachineInstrBuilder MIB = - AddDefaultPred(BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVs), + AddDefaultPred(BuildMI(MBB, MBBI, MI.getDebugLoc(),TII->get(ARM::MOVsi), MI.getOperand(0).getReg()) .addOperand(MI.getOperand(1)) - .addOperand(MI.getOperand(1)) .addImm(ARM_AM::getSORegOpc(ARM_AM::rrx, 0))) .addReg(0); TransferImpOps(MI, MIB, MIB); @@ -953,34 +970,6 @@ bool ARMExpandPseudo::ExpandMI(MachineBasicBlock &MBB, ExpandMOV32BitImm(MBB, MBBI); return true; - case ARM::VMOVQQ: { - unsigned DstReg = MI.getOperand(0).getReg(); - bool DstIsDead = MI.getOperand(0).isDead(); - unsigned EvenDst = TRI->getSubReg(DstReg, ARM::qsub_0); - unsigned OddDst = TRI->getSubReg(DstReg, ARM::qsub_1); - unsigned SrcReg = MI.getOperand(1).getReg(); - bool SrcIsKill = MI.getOperand(1).isKill(); - unsigned EvenSrc = TRI->getSubReg(SrcReg, ARM::qsub_0); - unsigned OddSrc = TRI->getSubReg(SrcReg, ARM::qsub_1); - MachineInstrBuilder Even = - AddDefaultPred(BuildMI(MBB, MBBI, MI.getDebugLoc(), - TII->get(ARM::VORRq)) - .addReg(EvenDst, - RegState::Define | getDeadRegState(DstIsDead)) - .addReg(EvenSrc, getKillRegState(SrcIsKill)) - .addReg(EvenSrc, getKillRegState(SrcIsKill))); - MachineInstrBuilder Odd = - AddDefaultPred(BuildMI(MBB, MBBI, MI.getDebugLoc(), - TII->get(ARM::VORRq)) - .addReg(OddDst, - RegState::Define | getDeadRegState(DstIsDead)) - .addReg(OddSrc, getKillRegState(SrcIsKill)) - .addReg(OddSrc, getKillRegState(SrcIsKill))); - TransferImpOps(MI, Even, Odd); - MI.eraseFromParent(); - return true; - } - case ARM::VLDMQIA: { unsigned NewOpc = ARM::VLDMDIA; MachineInstrBuilder MIB = @@ -1316,6 +1305,8 @@ bool ARMExpandPseudo::runOnMachineFunction(MachineFunction &MF) { for (MachineFunction::iterator MFI = MF.begin(), E = MF.end(); MFI != E; ++MFI) Modified |= ExpandMBB(*MFI); + if (VerifyARMPseudo) + MF.verify(this, "After expanding ARM pseudo instructions."); return Modified; } diff --git a/lib/Target/ARM/ARMFastISel.cpp b/lib/Target/ARM/ARMFastISel.cpp index f469d7efe11a2..9bc7ef21d8a34 100644 --- a/lib/Target/ARM/ARMFastISel.cpp +++ b/lib/Target/ARM/ARMFastISel.cpp @@ -14,13 +14,13 @@ //===----------------------------------------------------------------------===// #include "ARM.h" -#include "ARMAddressingModes.h" #include "ARMBaseInstrInfo.h" #include "ARMCallingConv.h" #include "ARMRegisterInfo.h" #include "ARMTargetMachine.h" #include "ARMSubtarget.h" #include "ARMConstantPoolValue.h" +#include "MCTargetDesc/ARMAddressingModes.h" #include "llvm/CallingConv.h" #include "llvm/DerivedTypes.h" #include "llvm/GlobalVariable.h" @@ -171,8 +171,8 @@ class ARMFastISel : public FastISel { // Utility routines. private: - bool isTypeLegal(const Type *Ty, MVT &VT); - bool isLoadTypeLegal(const Type *Ty, MVT &VT); + bool isTypeLegal(Type *Ty, MVT &VT); + bool isLoadTypeLegal(Type *Ty, MVT &VT); bool ARMEmitLoad(EVT VT, unsigned &ResultReg, Address &Addr); bool ARMEmitStore(EVT VT, unsigned SrcReg, Address &Addr); bool ARMComputeAddress(const Value *Obj, Address &Addr); @@ -502,11 +502,19 @@ unsigned ARMFastISel::ARMMaterializeFP(const ConstantFP *CFP, EVT VT) { // This checks to see if we can use VFP3 instructions to materialize // a constant, otherwise we have to go through the constant pool. if (TLI.isFPImmLegal(Val, VT)) { - unsigned Opc = is64bit ? ARM::FCONSTD : ARM::FCONSTS; + int Imm; + unsigned Opc; + if (is64bit) { + Imm = ARM_AM::getFP64Imm(Val); + Opc = ARM::FCONSTD; + } else { + Imm = ARM_AM::getFP32Imm(Val); + Opc = ARM::FCONSTS; + } unsigned DestReg = createResultReg(TLI.getRegClassFor(VT)); AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(Opc), DestReg) - .addFPImm(CFP)); + .addImm(Imm)); return DestReg; } @@ -590,8 +598,9 @@ unsigned ARMFastISel::ARMMaterializeGV(const GlobalValue *GV, EVT VT) { // Grab index. unsigned PCAdj = (RelocM != Reloc::PIC_) ? 0 : (Subtarget->isThumb() ? 4 : 8); unsigned Id = AFI->createPICLabelUId(); - ARMConstantPoolValue *CPV = new ARMConstantPoolValue(GV, Id, - ARMCP::CPValue, PCAdj); + ARMConstantPoolValue *CPV = ARMConstantPoolConstant::Create(GV, Id, + ARMCP::CPValue, + PCAdj); unsigned Idx = MCP.getConstantPoolIndex(CPV, Align); // Load value. @@ -615,8 +624,8 @@ unsigned ARMFastISel::ARMMaterializeGV(const GlobalValue *GV, EVT VT) { if (Subtarget->GVIsIndirectSymbol(GV, RelocM)) { unsigned NewDestReg = createResultReg(TLI.getRegClassFor(VT)); if (isThumb) - MIB = BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(ARM::t2LDRi12), - NewDestReg) + MIB = BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, + TII.get(ARM::t2LDRi12), NewDestReg) .addReg(DestReg) .addImm(0); else @@ -673,7 +682,7 @@ unsigned ARMFastISel::TargetMaterializeAlloca(const AllocaInst *AI) { return 0; } -bool ARMFastISel::isTypeLegal(const Type *Ty, MVT &VT) { +bool ARMFastISel::isTypeLegal(Type *Ty, MVT &VT) { EVT evt = TLI.getValueType(Ty, true); // Only handle simple types. @@ -685,7 +694,7 @@ bool ARMFastISel::isTypeLegal(const Type *Ty, MVT &VT) { return TLI.isTypeLegal(VT); } -bool ARMFastISel::isLoadTypeLegal(const Type *Ty, MVT &VT) { +bool ARMFastISel::isLoadTypeLegal(Type *Ty, MVT &VT) { if (isTypeLegal(Ty, VT)) return true; // If this is a type than can be sign or zero-extended to a basic operation @@ -714,7 +723,7 @@ bool ARMFastISel::ARMComputeAddress(const Value *Obj, Address &Addr) { U = C; } - if (const PointerType *Ty = dyn_cast(Obj->getType())) + if (PointerType *Ty = dyn_cast(Obj->getType())) if (Ty->getAddressSpace() > 255) // Fast instruction selection doesn't support the special // address spaces. @@ -749,7 +758,7 @@ bool ARMFastISel::ARMComputeAddress(const Value *Obj, Address &Addr) { for (User::const_op_iterator i = U->op_begin() + 1, e = U->op_end(); i != e; ++i, ++GTI) { const Value *Op = *i; - if (const StructType *STy = dyn_cast(*GTI)) { + if (StructType *STy = dyn_cast(*GTI)) { const StructLayout *SL = TD.getStructLayout(STy); unsigned Idx = cast(Op)->getZExtValue(); TmpOffset += SL->getElementOffset(Idx); @@ -946,6 +955,10 @@ bool ARMFastISel::ARMEmitLoad(EVT VT, unsigned &ResultReg, Address &Addr) { } bool ARMFastISel::SelectLoad(const Instruction *I) { + // Atomic loads need special handling. + if (cast(I)->isAtomic()) + return false; + // Verify we have a legal type before going any further. MVT VT; if (!isLoadTypeLegal(I->getType(), VT)) @@ -1008,6 +1021,10 @@ bool ARMFastISel::SelectStore(const Instruction *I) { Value *Op0 = I->getOperand(0); unsigned SrcReg = 0; + // Atomic stores need special handling. + if (cast(I)->isAtomic()) + return false; + // Verify we have a legal type before going any further. MVT VT; if (!isLoadTypeLegal(I->getOperand(0)->getType(), VT)) @@ -1085,7 +1102,7 @@ bool ARMFastISel::SelectBranch(const Instruction *I) { // TODO: Factor this out. if (const CmpInst *CI = dyn_cast(BI->getCondition())) { MVT SourceVT; - const Type *Ty = CI->getOperand(0)->getType(); + Type *Ty = CI->getOperand(0)->getType(); if (CI->hasOneUse() && (CI->getParent() == I->getParent()) && isTypeLegal(Ty, SourceVT)) { bool isFloat = (Ty->isDoubleTy() || Ty->isFloatTy()); @@ -1201,7 +1218,7 @@ bool ARMFastISel::SelectCmp(const Instruction *I) { const CmpInst *CI = cast(I); MVT VT; - const Type *Ty = CI->getOperand(0)->getType(); + Type *Ty = CI->getOperand(0)->getType(); if (!isTypeLegal(Ty, VT)) return false; @@ -1309,7 +1326,7 @@ bool ARMFastISel::SelectSIToFP(const Instruction *I) { if (!Subtarget->hasVFP2()) return false; MVT DstVT; - const Type *Ty = I->getType(); + Type *Ty = I->getType(); if (!isTypeLegal(Ty, DstVT)) return false; @@ -1328,7 +1345,7 @@ bool ARMFastISel::SelectSIToFP(const Instruction *I) { unsigned Opc; if (Ty->isFloatTy()) Opc = ARM::VSITOS; else if (Ty->isDoubleTy()) Opc = ARM::VSITOD; - else return 0; + else return false; unsigned ResultReg = createResultReg(TLI.getRegClassFor(DstVT)); AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(Opc), @@ -1343,7 +1360,7 @@ bool ARMFastISel::SelectFPToSI(const Instruction *I) { if (!Subtarget->hasVFP2()) return false; MVT DstVT; - const Type *RetTy = I->getType(); + Type *RetTy = I->getType(); if (!isTypeLegal(RetTy, DstVT)) return false; @@ -1351,10 +1368,10 @@ bool ARMFastISel::SelectFPToSI(const Instruction *I) { if (Op == 0) return false; unsigned Opc; - const Type *OpTy = I->getOperand(0)->getType(); + Type *OpTy = I->getOperand(0)->getType(); if (OpTy->isFloatTy()) Opc = ARM::VTOSIZS; else if (OpTy->isDoubleTy()) Opc = ARM::VTOSIZD; - else return 0; + else return false; // f64->s32 or f32->s32 both need an intermediate f32 reg. unsigned ResultReg = createResultReg(TLI.getRegClassFor(MVT::f32)); @@ -1401,7 +1418,7 @@ bool ARMFastISel::SelectSelect(const Instruction *I) { bool ARMFastISel::SelectSDiv(const Instruction *I) { MVT VT; - const Type *Ty = I->getType(); + Type *Ty = I->getType(); if (!isTypeLegal(Ty, VT)) return false; @@ -1429,7 +1446,7 @@ bool ARMFastISel::SelectSDiv(const Instruction *I) { bool ARMFastISel::SelectSRem(const Instruction *I) { MVT VT; - const Type *Ty = I->getType(); + Type *Ty = I->getType(); if (!isTypeLegal(Ty, VT)) return false; @@ -1456,7 +1473,7 @@ bool ARMFastISel::SelectBinaryOp(const Instruction *I, unsigned ISDOpcode) { // operations, but can't figure out how to. Just use the vfp instructions // if we have them. // FIXME: It'd be nice to use NEON instructions. - const Type *Ty = I->getType(); + Type *Ty = I->getType(); bool isFloat = (Ty->isDoubleTy() || Ty->isFloatTy()); if (isFloat && !Subtarget->hasVFP2()) return false; @@ -1711,7 +1728,7 @@ bool ARMFastISel::SelectRet(const Instruction *I) { // Analyze operands of the call, assigning locations to each operand. SmallVector ValLocs; - CCState CCInfo(CC, F.isVarArg(), *FuncInfo.MF, TM, ValLocs, I->getContext()); + CCState CCInfo(CC, F.isVarArg(), *FuncInfo.MF, TM, ValLocs,I->getContext()); CCInfo.AnalyzeReturn(Outs, CCAssignFnForCall(CC, true /* is Ret */)); const Value *RV = Ret->getOperand(0); @@ -1778,7 +1795,7 @@ bool ARMFastISel::ARMEmitLibcall(const Instruction *I, RTLIB::Libcall Call) { CallingConv::ID CC = TLI.getLibcallCallingConv(Call); // Handle *simple* calls for now. - const Type *RetTy = I->getType(); + Type *RetTy = I->getType(); MVT RetVT; if (RetTy->isVoidTy()) RetVT = MVT::isVoid; @@ -1802,7 +1819,7 @@ bool ARMFastISel::ARMEmitLibcall(const Instruction *I, RTLIB::Libcall Call) { unsigned Arg = getRegForValue(Op); if (Arg == 0) return false; - const Type *ArgTy = Op->getType(); + Type *ArgTy = Op->getType(); MVT ArgVT; if (!isTypeLegal(ArgTy, ArgVT)) return false; @@ -1870,13 +1887,13 @@ bool ARMFastISel::SelectCall(const Instruction *I) { // TODO: Avoid some calling conventions? // Let SDISel handle vararg functions. - const PointerType *PT = cast(CS.getCalledValue()->getType()); - const FunctionType *FTy = cast(PT->getElementType()); + PointerType *PT = cast(CS.getCalledValue()->getType()); + FunctionType *FTy = cast(PT->getElementType()); if (FTy->isVarArg()) return false; // Handle *simple* calls for now. - const Type *RetTy = I->getType(); + Type *RetTy = I->getType(); MVT RetVT; if (RetTy->isVoidTy()) RetVT = MVT::isVoid; @@ -1915,7 +1932,7 @@ bool ARMFastISel::SelectCall(const Instruction *I) { CS.paramHasAttr(AttrInd, Attribute::ByVal)) return false; - const Type *ArgTy = (*i)->getType(); + Type *ArgTy = (*i)->getType(); MVT ArgVT; if (!isTypeLegal(ArgTy, ArgVT)) return false; @@ -1969,9 +1986,9 @@ bool ARMFastISel::SelectIntCast(const Instruction *I) { // On ARM, in general, integer casts don't involve legal types; this code // handles promotable integers. The high bits for a type smaller than // the register size are assumed to be undefined. - const Type *DestTy = I->getType(); + Type *DestTy = I->getType(); Value *Op = I->getOperand(0); - const Type *SrcTy = Op->getType(); + Type *SrcTy = Op->getType(); EVT SrcVT, DestVT; SrcVT = TLI.getValueType(SrcTy, true); @@ -2002,16 +2019,18 @@ bool ARMFastISel::SelectIntCast(const Instruction *I) { switch (SrcVT.getSimpleVT().SimpleTy) { default: return false; case MVT::i16: + if (!Subtarget->hasV6Ops()) return false; if (isZext) - Opc = isThumb ? ARM::t2UXTHr : ARM::UXTHr; + Opc = isThumb ? ARM::t2UXTH : ARM::UXTH; else - Opc = isThumb ? ARM::t2SXTHr : ARM::SXTHr; + Opc = isThumb ? ARM::t2SXTH : ARM::SXTH; break; case MVT::i8: + if (!Subtarget->hasV6Ops()) return false; if (isZext) - Opc = isThumb ? ARM::t2UXTBr : ARM::UXTBr; + Opc = isThumb ? ARM::t2UXTB : ARM::UXTB; else - Opc = isThumb ? ARM::t2SXTBr : ARM::SXTBr; + Opc = isThumb ? ARM::t2SXTB : ARM::SXTB; break; case MVT::i1: if (isZext) { @@ -2033,6 +2052,8 @@ bool ARMFastISel::SelectIntCast(const Instruction *I) { .addReg(SrcReg); if (isBoolZext) MIB.addImm(1); + else + MIB.addImm(0); AddOptionalDefs(MIB); UpdateValueMap(I, DestReg); return true; diff --git a/lib/Target/ARM/ARMFixupKinds.h b/lib/Target/ARM/ARMFixupKinds.h deleted file mode 100644 index 350c92decdce4..0000000000000 --- a/lib/Target/ARM/ARMFixupKinds.h +++ /dev/null @@ -1,97 +0,0 @@ -//===-- ARM/ARMFixupKinds.h - ARM Specific Fixup Entries --------*- C++ -*-===// -// -// The LLVM Compiler Infrastructure -// -// This file is distributed under the University of Illinois Open Source -// License. See LICENSE.TXT for details. -// -//===----------------------------------------------------------------------===// - -#ifndef LLVM_ARM_ARMFIXUPKINDS_H -#define LLVM_ARM_ARMFIXUPKINDS_H - -#include "llvm/MC/MCFixup.h" - -namespace llvm { -namespace ARM { -enum Fixups { - // fixup_arm_ldst_pcrel_12 - 12-bit PC relative relocation for symbol - // addresses - fixup_arm_ldst_pcrel_12 = FirstTargetFixupKind, - - // fixup_t2_ldst_pcrel_12 - Equivalent to fixup_arm_ldst_pcrel_12, with - // the 16-bit halfwords reordered. - fixup_t2_ldst_pcrel_12, - - // fixup_arm_pcrel_10 - 10-bit PC relative relocation for symbol addresses - // used in VFP instructions where the lower 2 bits are not encoded - // (so it's encoded as an 8-bit immediate). - fixup_arm_pcrel_10, - // fixup_t2_pcrel_10 - Equivalent to fixup_arm_pcrel_10, accounting for - // the short-swapped encoding of Thumb2 instructions. - fixup_t2_pcrel_10, - // fixup_thumb_adr_pcrel_10 - 10-bit PC relative relocation for symbol - // addresses where the lower 2 bits are not encoded (so it's encoded as an - // 8-bit immediate). - fixup_thumb_adr_pcrel_10, - // fixup_arm_adr_pcrel_12 - 12-bit PC relative relocation for the ADR - // instruction. - fixup_arm_adr_pcrel_12, - // fixup_t2_adr_pcrel_12 - 12-bit PC relative relocation for the ADR - // instruction. - fixup_t2_adr_pcrel_12, - // fixup_arm_condbranch - 24-bit PC relative relocation for conditional branch - // instructions. - fixup_arm_condbranch, - // fixup_arm_uncondbranch - 24-bit PC relative relocation for - // branch instructions. (unconditional) - fixup_arm_uncondbranch, - // fixup_t2_condbranch - 20-bit PC relative relocation for Thumb2 direct - // uconditional branch instructions. - fixup_t2_condbranch, - // fixup_t2_uncondbranch - 20-bit PC relative relocation for Thumb2 direct - // branch unconditional branch instructions. - fixup_t2_uncondbranch, - - // fixup_arm_thumb_br - 12-bit fixup for Thumb B instructions. - fixup_arm_thumb_br, - - // fixup_arm_thumb_bl - Fixup for Thumb BL instructions. - fixup_arm_thumb_bl, - - // fixup_arm_thumb_blx - Fixup for Thumb BLX instructions. - fixup_arm_thumb_blx, - - // fixup_arm_thumb_cb - Fixup for Thumb branch instructions. - fixup_arm_thumb_cb, - - // fixup_arm_thumb_cp - Fixup for Thumb load/store from constant pool instrs. - fixup_arm_thumb_cp, - - // fixup_arm_thumb_bcc - Fixup for Thumb conditional branching instructions. - fixup_arm_thumb_bcc, - - // The next two are for the movt/movw pair - // the 16bit imm field are split into imm{15-12} and imm{11-0} - fixup_arm_movt_hi16, // :upper16: - fixup_arm_movw_lo16, // :lower16: - fixup_t2_movt_hi16, // :upper16: - fixup_t2_movw_lo16, // :lower16: - - // It is possible to create an "immediate" that happens to be pcrel. - // movw r0, :lower16:Foo-(Bar+8) and movt r0, :upper16:Foo-(Bar+8) - // result in different reloc tags than the above two. - // Needed to support ELF::R_ARM_MOVT_PREL and ELF::R_ARM_MOVW_PREL_NC - fixup_arm_movt_hi16_pcrel, // :upper16: - fixup_arm_movw_lo16_pcrel, // :lower16: - fixup_t2_movt_hi16_pcrel, // :upper16: - fixup_t2_movw_lo16_pcrel, // :lower16: - - // Marker - LastTargetFixupKind, - NumTargetFixupKinds = LastTargetFixupKind - FirstTargetFixupKind -}; -} -} - -#endif diff --git a/lib/Target/ARM/ARMFrameLowering.cpp b/lib/Target/ARM/ARMFrameLowering.cpp index 381b404519e22..2d1de6fe8e9d1 100644 --- a/lib/Target/ARM/ARMFrameLowering.cpp +++ b/lib/Target/ARM/ARMFrameLowering.cpp @@ -12,10 +12,10 @@ //===----------------------------------------------------------------------===// #include "ARMFrameLowering.h" -#include "ARMAddressingModes.h" #include "ARMBaseInstrInfo.h" #include "ARMBaseRegisterInfo.h" #include "ARMMachineFunctionInfo.h" +#include "MCTargetDesc/ARMAddressingModes.h" #include "llvm/CodeGen/MachineFrameInfo.h" #include "llvm/CodeGen/MachineFunction.h" #include "llvm/CodeGen/MachineInstrBuilder.h" @@ -93,7 +93,8 @@ static bool isCSRestore(MachineInstr *MI, return false; return true; } - if ((MI->getOpcode() == ARM::LDR_POST || + if ((MI->getOpcode() == ARM::LDR_POST_IMM || + MI->getOpcode() == ARM::LDR_POST_REG || MI->getOpcode() == ARM::t2LDR_POST) && isCalleeSavedRegister(MI->getOperand(0).getReg(), CSRegs) && MI->getOperand(1).getReg() == ARM::SP) @@ -413,6 +414,9 @@ void ARMFrameLowering::emitEpilogue(MachineFunction &MF, MIB.addExternalSymbol(JumpTarget.getSymbolName(), JumpTarget.getTargetFlags()); } + + // Add the default predicate in Thumb mode. + if (STI.isThumb()) MIB.addImm(ARMCC::AL).addReg(0); } else if (RetOpcode == ARM::TCRETURNri) { BuildMI(MBB, MBBI, dl, TII.get(STI.isThumb() ? ARM::tTAILJMPr : ARM::TAILJMPr)). @@ -502,7 +506,7 @@ ARMFrameLowering::ResolveFrameIndexReference(const MachineFunction &MF, } } } else if (AFI->isThumb2Function()) { - // Use add , sp, # + // Use add , sp, # // ldr , [sp, #] // if at all possible to save space. if (Offset >= 0 && (Offset & 3) == 0 && Offset <= 1020) @@ -587,14 +591,8 @@ void ARMFrameLowering::emitPushInst(MachineBasicBlock &MBB, MachineInstrBuilder MIB = BuildMI(MBB, MI, DL, TII.get(StrOpc), ARM::SP) .addReg(Regs[0].first, getKillRegState(Regs[0].second)) - .addReg(ARM::SP).setMIFlags(MIFlags); - // ARM mode needs an extra reg0 here due to addrmode2. Will go away once - // that refactoring is complete (eventually). - if (StrOpc == ARM::STR_PRE) { - MIB.addReg(0); - MIB.addImm(ARM_AM::getAM2Opc(ARM_AM::sub, 4, ARM_AM::no_shift)); - } else - MIB.addImm(-4); + .addReg(ARM::SP).setMIFlags(MIFlags) + .addImm(-4); AddDefaultPred(MIB); } Regs.clear(); @@ -651,8 +649,10 @@ void ARMFrameLowering::emitPopInst(MachineBasicBlock &MBB, .addReg(ARM::SP)); for (unsigned i = 0, e = Regs.size(); i < e; ++i) MIB.addReg(Regs[i], getDefRegState(true)); - if (DeleteRet) + if (DeleteRet) { + MIB->copyImplicitOps(&*MI); MI->eraseFromParent(); + } MI = MIB; } else if (Regs.size() == 1) { // If we adjusted the reg to PC from LR above, switch it back here. We @@ -665,7 +665,7 @@ void ARMFrameLowering::emitPopInst(MachineBasicBlock &MBB, .addReg(ARM::SP); // ARM mode needs an extra reg0 here due to addrmode2. Will go away once // that refactoring is complete (eventually). - if (LdrOpc == ARM::LDR_POST) { + if (LdrOpc == ARM::LDR_POST_REG || LdrOpc == ARM::LDR_POST_IMM) { MIB.addReg(0); MIB.addImm(ARM_AM::getAM2Opc(ARM_AM::add, 4, ARM_AM::no_shift)); } else @@ -687,7 +687,8 @@ bool ARMFrameLowering::spillCalleeSavedRegisters(MachineBasicBlock &MBB, ARMFunctionInfo *AFI = MF.getInfo(); unsigned PushOpc = AFI->isThumbFunction() ? ARM::t2STMDB_UPD : ARM::STMDB_UPD; - unsigned PushOneOpc = AFI->isThumbFunction() ? ARM::t2STR_PRE : ARM::STR_PRE; + unsigned PushOneOpc = AFI->isThumbFunction() ? + ARM::t2STR_PRE : ARM::STR_PRE_IMM; unsigned FltOpc = ARM::VSTMDDB_UPD; emitPushInst(MBB, MI, CSI, PushOpc, PushOneOpc, false, &isARMArea1Register, MachineInstr::FrameSetup); @@ -711,7 +712,7 @@ bool ARMFrameLowering::restoreCalleeSavedRegisters(MachineBasicBlock &MBB, bool isVarArg = AFI->getVarArgsRegSaveSize() > 0; unsigned PopOpc = AFI->isThumbFunction() ? ARM::t2LDMIA_UPD : ARM::LDMIA_UPD; - unsigned LdrOpc = AFI->isThumbFunction() ? ARM::t2LDR_POST : ARM::LDR_POST; + unsigned LdrOpc = AFI->isThumbFunction() ? ARM::t2LDR_POST :ARM::LDR_POST_IMM; unsigned FltOpc = ARM::VLDMDIA_UPD; emitPopInst(MBB, MI, CSI, FltOpc, 0, isVarArg, true, &isARMArea3Register); emitPopInst(MBB, MI, CSI, PopOpc, LdrOpc, isVarArg, false, diff --git a/lib/Target/ARM/ARMGlobalMerge.cpp b/lib/Target/ARM/ARMGlobalMerge.cpp index 8d77b2d8383e1..5f863ea241ca8 100644 --- a/lib/Target/ARM/ARMGlobalMerge.cpp +++ b/lib/Target/ARM/ARMGlobalMerge.cpp @@ -100,8 +100,8 @@ namespace { GlobalCmp(const TargetData *td) : TD(td) { } bool operator()(const GlobalVariable *GV1, const GlobalVariable *GV2) { - const Type *Ty1 = cast(GV1->getType())->getElementType(); - const Type *Ty2 = cast(GV2->getType())->getElementType(); + Type *Ty1 = cast(GV1->getType())->getElementType(); + Type *Ty2 = cast(GV2->getType())->getElementType(); return (TD->getTypeAllocSize(Ty1) < TD->getTypeAllocSize(Ty2)); } @@ -123,7 +123,7 @@ bool ARMGlobalMerge::doMerge(SmallVectorImpl &Globals, // FIXME: Find better heuristics std::stable_sort(Globals.begin(), Globals.end(), GlobalCmp(TD)); - const Type *Int32Ty = Type::getInt32Ty(M.getContext()); + Type *Int32Ty = Type::getInt32Ty(M.getContext()); for (size_t i = 0, e = Globals.size(); i != e; ) { size_t j = 0; @@ -150,7 +150,7 @@ bool ARMGlobalMerge::doMerge(SmallVectorImpl &Globals, ConstantInt::get(Int32Ty, 0), ConstantInt::get(Int32Ty, k-i) }; - Constant *GEP = ConstantExpr::getInBoundsGetElementPtr(MergedGV, Idx, 2); + Constant *GEP = ConstantExpr::getInBoundsGetElementPtr(MergedGV, Idx); Globals[k]->replaceAllUsesWith(GEP); Globals[k]->eraseFromParent(); } @@ -176,7 +176,7 @@ bool ARMGlobalMerge::doInitialization(Module &M) { // Ignore fancy-aligned globals for now. unsigned Alignment = I->getAlignment(); - const Type *Ty = I->getType()->getElementType(); + Type *Ty = I->getType()->getElementType(); if (Alignment > TD->getABITypeAlignment(Ty)) continue; diff --git a/lib/Target/ARM/ARMISelDAGToDAG.cpp b/lib/Target/ARM/ARMISelDAGToDAG.cpp index 2c9481b86c557..5ee009c04c5b4 100644 --- a/lib/Target/ARM/ARMISelDAGToDAG.cpp +++ b/lib/Target/ARM/ARMISelDAGToDAG.cpp @@ -14,8 +14,8 @@ #define DEBUG_TYPE "arm-isel" #include "ARM.h" #include "ARMBaseInstrInfo.h" -#include "ARMAddressingModes.h" #include "ARMTargetMachine.h" +#include "MCTargetDesc/ARMAddressingModes.h" #include "llvm/CallingConv.h" #include "llvm/Constants.h" #include "llvm/DerivedTypes.h" @@ -47,6 +47,11 @@ CheckVMLxHazard("check-vmlx-hazard", cl::Hidden, cl::desc("Check fp vmla / vmls hazard at isel time"), cl::init(true)); +static cl::opt +DisableARMIntABS("disable-arm-int-abs", cl::Hidden, + cl::desc("Enable / disable ARM integer abs transform"), + cl::init(false)); + //===--------------------------------------------------------------------===// /// ARMDAGToDAGISel - ARM specific code to select ARM machine /// instructions for SelectionDAG operations. @@ -90,13 +95,20 @@ public: bool hasNoVMLxHazardUse(SDNode *N) const; bool isShifterOpProfitable(const SDValue &Shift, ARM_AM::ShiftOpc ShOpcVal, unsigned ShAmt); - bool SelectShifterOperandReg(SDValue N, SDValue &A, + bool SelectRegShifterOperand(SDValue N, SDValue &A, SDValue &B, SDValue &C, bool CheckProfitability = true); - bool SelectShiftShifterOperandReg(SDValue N, SDValue &A, + bool SelectImmShifterOperand(SDValue N, SDValue &A, + SDValue &B, bool CheckProfitability = true); + bool SelectShiftRegShifterOperand(SDValue N, SDValue &A, SDValue &B, SDValue &C) { // Don't apply the profitability check - return SelectShifterOperandReg(N, A, B, C, false); + return SelectRegShifterOperand(N, A, B, C, false); + } + bool SelectShiftImmShifterOperand(SDValue N, SDValue &A, + SDValue &B) { + // Don't apply the profitability check + return SelectImmShifterOperand(N, A, B, false); } bool SelectAddrModeImm12(SDValue N, SDValue &Base, SDValue &OffImm); @@ -122,8 +134,13 @@ public: return true; } - bool SelectAddrMode2Offset(SDNode *Op, SDValue N, + bool SelectAddrMode2OffsetReg(SDNode *Op, SDValue N, SDValue &Offset, SDValue &Opc); + bool SelectAddrMode2OffsetImm(SDNode *Op, SDValue N, + SDValue &Offset, SDValue &Opc); + bool SelectAddrMode2OffsetImmPre(SDNode *Op, SDValue N, + SDValue &Offset, SDValue &Opc); + bool SelectAddrOffsetNone(SDValue N, SDValue &Base); bool SelectAddrMode3(SDValue N, SDValue &Base, SDValue &Offset, SDValue &Opc); bool SelectAddrMode3Offset(SDNode *Op, SDValue N, @@ -240,8 +257,13 @@ private: ARMCC::CondCodes CCVal, SDValue CCR, SDValue InFlag); + // Select special operations if node forms integer ABS pattern + SDNode *SelectABSOp(SDNode *N); + SDNode *SelectConcatVector(SDNode *N); + SDNode *SelectAtomic64(SDNode *Node, unsigned Opc); + /// SelectInlineAsmMemoryOperand - Implement addressing mode selection for /// inline asm expressions. virtual bool SelectInlineAsmMemoryOperand(const SDValue &Op, @@ -291,10 +313,10 @@ static bool isOpcWithIntImmediate(SDNode *N, unsigned Opc, unsigned& Imm) { /// (N * Scale) where (N in [\arg RangeMin, \arg RangeMax). /// /// \param ScaledConstant [out] - On success, the pre-scaled constant value. -static bool isScaledConstantInRange(SDValue Node, unsigned Scale, +static bool isScaledConstantInRange(SDValue Node, int Scale, int RangeMin, int RangeMax, int &ScaledConstant) { - assert(Scale && "Invalid scale!"); + assert(Scale > 0 && "Invalid scale!"); // Check that this is a constant. const ConstantSDNode *C = dyn_cast(Node); @@ -365,7 +387,30 @@ bool ARMDAGToDAGISel::isShifterOpProfitable(const SDValue &Shift, return ShOpcVal == ARM_AM::lsl && ShAmt == 2; } -bool ARMDAGToDAGISel::SelectShifterOperandReg(SDValue N, +bool ARMDAGToDAGISel::SelectImmShifterOperand(SDValue N, + SDValue &BaseReg, + SDValue &Opc, + bool CheckProfitability) { + if (DisableShifterOp) + return false; + + ARM_AM::ShiftOpc ShOpcVal = ARM_AM::getShiftOpcForNode(N.getOpcode()); + + // Don't match base register only case. That is matched to a separate + // lower complexity pattern with explicit register operand. + if (ShOpcVal == ARM_AM::no_shift) return false; + + BaseReg = N.getOperand(0); + unsigned ShImmVal = 0; + ConstantSDNode *RHS = dyn_cast(N.getOperand(1)); + if (!RHS) return false; + ShImmVal = RHS->getZExtValue() & 31; + Opc = CurDAG->getTargetConstant(ARM_AM::getSORegOpc(ShOpcVal, ShImmVal), + MVT::i32); + return true; +} + +bool ARMDAGToDAGISel::SelectRegShifterOperand(SDValue N, SDValue &BaseReg, SDValue &ShReg, SDValue &Opc, @@ -373,7 +418,7 @@ bool ARMDAGToDAGISel::SelectShifterOperandReg(SDValue N, if (DisableShifterOp) return false; - ARM_AM::ShiftOpc ShOpcVal = ARM_AM::getShiftOpcForNode(N); + ARM_AM::ShiftOpc ShOpcVal = ARM_AM::getShiftOpcForNode(N.getOpcode()); // Don't match base register only case. That is matched to a separate // lower complexity pattern with explicit register operand. @@ -381,19 +426,18 @@ bool ARMDAGToDAGISel::SelectShifterOperandReg(SDValue N, BaseReg = N.getOperand(0); unsigned ShImmVal = 0; - if (ConstantSDNode *RHS = dyn_cast(N.getOperand(1))) { - ShReg = CurDAG->getRegister(0, MVT::i32); - ShImmVal = RHS->getZExtValue() & 31; - } else { - ShReg = N.getOperand(1); - if (CheckProfitability && !isShifterOpProfitable(N, ShOpcVal, ShImmVal)) - return false; - } + ConstantSDNode *RHS = dyn_cast(N.getOperand(1)); + if (RHS) return false; + + ShReg = N.getOperand(1); + if (CheckProfitability && !isShifterOpProfitable(N, ShOpcVal, ShImmVal)) + return false; Opc = CurDAG->getTargetConstant(ARM_AM::getSORegOpc(ShOpcVal, ShImmVal), MVT::i32); return true; } + bool ARMDAGToDAGISel::SelectAddrModeImm12(SDValue N, SDValue &Base, SDValue &OffImm) { @@ -483,13 +527,10 @@ bool ARMDAGToDAGISel::SelectLdStSOReg(SDValue N, SDValue &Base, SDValue &Offset, return false; } - if (Subtarget->isCortexA9() && !N.hasOneUse()) - // Compute R +/- (R << N) and reuse it. - return false; - // Otherwise this is R +/- [possibly shifted] R. ARM_AM::AddrOpc AddSub = N.getOpcode() == ISD::SUB ? ARM_AM::sub:ARM_AM::add; - ARM_AM::ShiftOpc ShOpcVal = ARM_AM::getShiftOpcForNode(N.getOperand(1)); + ARM_AM::ShiftOpc ShOpcVal = + ARM_AM::getShiftOpcForNode(N.getOperand(1).getOpcode()); unsigned ShAmt = 0; Base = N.getOperand(0); @@ -515,16 +556,14 @@ bool ARMDAGToDAGISel::SelectLdStSOReg(SDValue N, SDValue &Base, SDValue &Offset, // Try matching (R shl C) + (R). if (N.getOpcode() != ISD::SUB && ShOpcVal == ARM_AM::no_shift && !(Subtarget->isCortexA9() || N.getOperand(0).hasOneUse())) { - ShOpcVal = ARM_AM::getShiftOpcForNode(N.getOperand(0)); + ShOpcVal = ARM_AM::getShiftOpcForNode(N.getOperand(0).getOpcode()); if (ShOpcVal != ARM_AM::no_shift) { // Check to see if the RHS of the shift is a constant, if not, we can't // fold it. if (ConstantSDNode *Sh = dyn_cast(N.getOperand(0).getOperand(1))) { ShAmt = Sh->getZExtValue(); - if (!Subtarget->isCortexA9() || - (N.hasOneUse() && - isShifterOpProfitable(N.getOperand(0), ShOpcVal, ShAmt))) { + if (isShifterOpProfitable(N.getOperand(0), ShOpcVal, ShAmt)) { Offset = N.getOperand(0).getOperand(0); Base = N.getOperand(1); } else { @@ -630,7 +669,8 @@ AddrMode2Type ARMDAGToDAGISel::SelectAddrMode2Worker(SDValue N, // Otherwise this is R +/- [possibly shifted] R. ARM_AM::AddrOpc AddSub = N.getOpcode() != ISD::SUB ? ARM_AM::add:ARM_AM::sub; - ARM_AM::ShiftOpc ShOpcVal = ARM_AM::getShiftOpcForNode(N.getOperand(1)); + ARM_AM::ShiftOpc ShOpcVal = + ARM_AM::getShiftOpcForNode(N.getOperand(1).getOpcode()); unsigned ShAmt = 0; Base = N.getOperand(0); @@ -656,16 +696,14 @@ AddrMode2Type ARMDAGToDAGISel::SelectAddrMode2Worker(SDValue N, // Try matching (R shl C) + (R). if (N.getOpcode() != ISD::SUB && ShOpcVal == ARM_AM::no_shift && !(Subtarget->isCortexA9() || N.getOperand(0).hasOneUse())) { - ShOpcVal = ARM_AM::getShiftOpcForNode(N.getOperand(0)); + ShOpcVal = ARM_AM::getShiftOpcForNode(N.getOperand(0).getOpcode()); if (ShOpcVal != ARM_AM::no_shift) { // Check to see if the RHS of the shift is a constant, if not, we can't // fold it. if (ConstantSDNode *Sh = dyn_cast(N.getOperand(0).getOperand(1))) { ShAmt = Sh->getZExtValue(); - if (!Subtarget->isCortexA9() || - (N.hasOneUse() && - isShifterOpProfitable(N.getOperand(0), ShOpcVal, ShAmt))) { + if (isShifterOpProfitable(N.getOperand(0), ShOpcVal, ShAmt)) { Offset = N.getOperand(0).getOperand(0); Base = N.getOperand(1); } else { @@ -683,7 +721,7 @@ AddrMode2Type ARMDAGToDAGISel::SelectAddrMode2Worker(SDValue N, return AM2_SHOP; } -bool ARMDAGToDAGISel::SelectAddrMode2Offset(SDNode *Op, SDValue N, +bool ARMDAGToDAGISel::SelectAddrMode2OffsetReg(SDNode *Op, SDValue N, SDValue &Offset, SDValue &Opc) { unsigned Opcode = Op->getOpcode(); ISD::MemIndexedMode AM = (Opcode == ISD::LOAD) @@ -692,16 +730,11 @@ bool ARMDAGToDAGISel::SelectAddrMode2Offset(SDNode *Op, SDValue N, ARM_AM::AddrOpc AddSub = (AM == ISD::PRE_INC || AM == ISD::POST_INC) ? ARM_AM::add : ARM_AM::sub; int Val; - if (isScaledConstantInRange(N, /*Scale=*/1, 0, 0x1000, Val)) { // 12 bits. - Offset = CurDAG->getRegister(0, MVT::i32); - Opc = CurDAG->getTargetConstant(ARM_AM::getAM2Opc(AddSub, Val, - ARM_AM::no_shift), - MVT::i32); - return true; - } + if (isScaledConstantInRange(N, /*Scale=*/1, 0, 0x1000, Val)) + return false; Offset = N; - ARM_AM::ShiftOpc ShOpcVal = ARM_AM::getShiftOpcForNode(N); + ARM_AM::ShiftOpc ShOpcVal = ARM_AM::getShiftOpcForNode(N.getOpcode()); unsigned ShAmt = 0; if (ShOpcVal != ARM_AM::no_shift) { // Check to see if the RHS of the shift is a constant, if not, we can't fold @@ -724,6 +757,50 @@ bool ARMDAGToDAGISel::SelectAddrMode2Offset(SDNode *Op, SDValue N, return true; } +bool ARMDAGToDAGISel::SelectAddrMode2OffsetImmPre(SDNode *Op, SDValue N, + SDValue &Offset, SDValue &Opc) { + unsigned Opcode = Op->getOpcode(); + ISD::MemIndexedMode AM = (Opcode == ISD::LOAD) + ? cast(Op)->getAddressingMode() + : cast(Op)->getAddressingMode(); + ARM_AM::AddrOpc AddSub = (AM == ISD::PRE_INC || AM == ISD::POST_INC) + ? ARM_AM::add : ARM_AM::sub; + int Val; + if (isScaledConstantInRange(N, /*Scale=*/1, 0, 0x1000, Val)) { // 12 bits. + if (AddSub == ARM_AM::sub) Val *= -1; + Offset = CurDAG->getRegister(0, MVT::i32); + Opc = CurDAG->getTargetConstant(Val, MVT::i32); + return true; + } + + return false; +} + + +bool ARMDAGToDAGISel::SelectAddrMode2OffsetImm(SDNode *Op, SDValue N, + SDValue &Offset, SDValue &Opc) { + unsigned Opcode = Op->getOpcode(); + ISD::MemIndexedMode AM = (Opcode == ISD::LOAD) + ? cast(Op)->getAddressingMode() + : cast(Op)->getAddressingMode(); + ARM_AM::AddrOpc AddSub = (AM == ISD::PRE_INC || AM == ISD::POST_INC) + ? ARM_AM::add : ARM_AM::sub; + int Val; + if (isScaledConstantInRange(N, /*Scale=*/1, 0, 0x1000, Val)) { // 12 bits. + Offset = CurDAG->getRegister(0, MVT::i32); + Opc = CurDAG->getTargetConstant(ARM_AM::getAM2Opc(AddSub, Val, + ARM_AM::no_shift), + MVT::i32); + return true; + } + + return false; +} + +bool ARMDAGToDAGISel::SelectAddrOffsetNone(SDValue N, SDValue &Base) { + Base = N; + return true; +} bool ARMDAGToDAGISel::SelectAddrMode3(SDValue N, SDValue &Base, SDValue &Offset, @@ -1079,7 +1156,7 @@ bool ARMDAGToDAGISel::SelectT2ShifterOperandReg(SDValue N, SDValue &BaseReg, if (DisableShifterOp) return false; - ARM_AM::ShiftOpc ShOpcVal = ARM_AM::getShiftOpcForNode(N); + ARM_AM::ShiftOpc ShOpcVal = ARM_AM::getShiftOpcForNode(N.getOpcode()); // Don't match base register only case. That is matched to a separate // lower complexity pattern with explicit register operand. @@ -1208,21 +1285,15 @@ bool ARMDAGToDAGISel::SelectT2AddrModeSoReg(SDValue N, return false; } - if (Subtarget->isCortexA9() && !N.hasOneUse()) { - // Compute R + (R << [1,2,3]) and reuse it. - Base = N; - return false; - } - // Look for (R + R) or (R + (R << [1,2,3])). unsigned ShAmt = 0; Base = N.getOperand(0); OffReg = N.getOperand(1); // Swap if it is ((R << c) + R). - ARM_AM::ShiftOpc ShOpcVal = ARM_AM::getShiftOpcForNode(OffReg); + ARM_AM::ShiftOpc ShOpcVal = ARM_AM::getShiftOpcForNode(OffReg.getOpcode()); if (ShOpcVal != ARM_AM::lsl) { - ShOpcVal = ARM_AM::getShiftOpcForNode(Base); + ShOpcVal = ARM_AM::getShiftOpcForNode(Base.getOpcode()); if (ShOpcVal == ARM_AM::lsl) std::swap(Base, OffReg); } @@ -1266,10 +1337,19 @@ SDNode *ARMDAGToDAGISel::SelectARMIndexedLoad(SDNode *N) { bool isPre = (AM == ISD::PRE_INC) || (AM == ISD::PRE_DEC); unsigned Opcode = 0; bool Match = false; - if (LoadedVT == MVT::i32 && - SelectAddrMode2Offset(N, LD->getOffset(), Offset, AMOpc)) { - Opcode = isPre ? ARM::LDR_PRE : ARM::LDR_POST; + if (LoadedVT == MVT::i32 && isPre && + SelectAddrMode2OffsetImmPre(N, LD->getOffset(), Offset, AMOpc)) { + Opcode = ARM::LDR_PRE_IMM; + Match = true; + } else if (LoadedVT == MVT::i32 && !isPre && + SelectAddrMode2OffsetImm(N, LD->getOffset(), Offset, AMOpc)) { + Opcode = ARM::LDR_POST_IMM; Match = true; + } else if (LoadedVT == MVT::i32 && + SelectAddrMode2OffsetReg(N, LD->getOffset(), Offset, AMOpc)) { + Opcode = isPre ? ARM::LDR_PRE_REG : ARM::LDR_POST_REG; + Match = true; + } else if (LoadedVT == MVT::i16 && SelectAddrMode3Offset(N, LD->getOffset(), Offset, AMOpc)) { Match = true; @@ -1283,20 +1363,37 @@ SDNode *ARMDAGToDAGISel::SelectARMIndexedLoad(SDNode *N) { Opcode = isPre ? ARM::LDRSB_PRE : ARM::LDRSB_POST; } } else { - if (SelectAddrMode2Offset(N, LD->getOffset(), Offset, AMOpc)) { + if (isPre && + SelectAddrMode2OffsetImmPre(N, LD->getOffset(), Offset, AMOpc)) { + Match = true; + Opcode = ARM::LDRB_PRE_IMM; + } else if (!isPre && + SelectAddrMode2OffsetImm(N, LD->getOffset(), Offset, AMOpc)) { + Match = true; + Opcode = ARM::LDRB_POST_IMM; + } else if (SelectAddrMode2OffsetReg(N, LD->getOffset(), Offset, AMOpc)) { Match = true; - Opcode = isPre ? ARM::LDRB_PRE : ARM::LDRB_POST; + Opcode = isPre ? ARM::LDRB_PRE_REG : ARM::LDRB_POST_REG; } } } if (Match) { - SDValue Chain = LD->getChain(); - SDValue Base = LD->getBasePtr(); - SDValue Ops[]= { Base, Offset, AMOpc, getAL(CurDAG), - CurDAG->getRegister(0, MVT::i32), Chain }; - return CurDAG->getMachineNode(Opcode, N->getDebugLoc(), MVT::i32, MVT::i32, - MVT::Other, Ops, 6); + if (Opcode == ARM::LDR_PRE_IMM || Opcode == ARM::LDRB_PRE_IMM) { + SDValue Chain = LD->getChain(); + SDValue Base = LD->getBasePtr(); + SDValue Ops[]= { Base, AMOpc, getAL(CurDAG), + CurDAG->getRegister(0, MVT::i32), Chain }; + return CurDAG->getMachineNode(Opcode, N->getDebugLoc(), MVT::i32, + MVT::i32, MVT::Other, Ops, 5); + } else { + SDValue Chain = LD->getChain(); + SDValue Base = LD->getBasePtr(); + SDValue Ops[]= { Base, Offset, AMOpc, getAL(CurDAG), + CurDAG->getRegister(0, MVT::i32), Chain }; + return CurDAG->getMachineNode(Opcode, N->getDebugLoc(), MVT::i32, + MVT::i32, MVT::Other, Ops, 6); + } } return NULL; @@ -1966,7 +2063,8 @@ SDNode *ARMDAGToDAGISel::SelectV6T2BitfieldExtractOp(SDNode *N, Srl_imm)) { assert(Srl_imm > 0 && Srl_imm < 32 && "bad amount in shift node!"); - unsigned Width = CountTrailingOnes_32(And_imm); + // Note: The width operand is encoded as width-1. + unsigned Width = CountTrailingOnes_32(And_imm) - 1; unsigned LSB = Srl_imm; SDValue Reg0 = CurDAG->getRegister(0, MVT::i32); SDValue Ops[] = { N->getOperand(0).getOperand(0), @@ -1986,7 +2084,8 @@ SDNode *ARMDAGToDAGISel::SelectV6T2BitfieldExtractOp(SDNode *N, unsigned Srl_imm = 0; if (isInt32Immediate(N->getOperand(1), Srl_imm)) { assert(Srl_imm > 0 && Srl_imm < 32 && "bad amount in shift node!"); - unsigned Width = 32 - Srl_imm; + // Note: The width operand is encoded as width-1. + unsigned Width = 32 - Srl_imm - 1; int LSB = Srl_imm - Shl_imm; if (LSB < 0) return NULL; @@ -2034,10 +2133,16 @@ SelectARMCMOVShiftOp(SDNode *N, SDValue FalseVal, SDValue TrueVal, SDValue CPTmp0; SDValue CPTmp1; SDValue CPTmp2; - if (SelectShifterOperandReg(TrueVal, CPTmp0, CPTmp1, CPTmp2)) { + if (SelectImmShifterOperand(TrueVal, CPTmp0, CPTmp2)) { + SDValue CC = CurDAG->getTargetConstant(CCVal, MVT::i32); + SDValue Ops[] = { FalseVal, CPTmp0, CPTmp2, CC, CCR, InFlag }; + return CurDAG->SelectNodeTo(N, ARM::MOVCCsi, MVT::i32, Ops, 6); + } + + if (SelectRegShifterOperand(TrueVal, CPTmp0, CPTmp1, CPTmp2)) { SDValue CC = CurDAG->getTargetConstant(CCVal, MVT::i32); SDValue Ops[] = { FalseVal, CPTmp0, CPTmp1, CPTmp2, CC, CCR, InFlag }; - return CurDAG->SelectNodeTo(N, ARM::MOVCCs, MVT::i32, Ops, 7); + return CurDAG->SelectNodeTo(N, ARM::MOVCCsr, MVT::i32, Ops, 7); } return 0; } @@ -2198,6 +2303,56 @@ SDNode *ARMDAGToDAGISel::SelectCMOVOp(SDNode *N) { return CurDAG->SelectNodeTo(N, Opc, VT, Ops, 5); } +/// Target-specific DAG combining for ISD::XOR. +/// Target-independent combining lowers SELECT_CC nodes of the form +/// select_cc setg[ge] X, 0, X, -X +/// select_cc setgt X, -1, X, -X +/// select_cc setl[te] X, 0, -X, X +/// select_cc setlt X, 1, -X, X +/// which represent Integer ABS into: +/// Y = sra (X, size(X)-1); xor (add (X, Y), Y) +/// ARM instruction selection detects the latter and matches it to +/// ARM::ABS or ARM::t2ABS machine node. +SDNode *ARMDAGToDAGISel::SelectABSOp(SDNode *N){ + SDValue XORSrc0 = N->getOperand(0); + SDValue XORSrc1 = N->getOperand(1); + DebugLoc DL = N->getDebugLoc(); + EVT VT = N->getValueType(0); + + if (DisableARMIntABS) + return NULL; + + if (Subtarget->isThumb1Only()) + return NULL; + + if (XORSrc0.getOpcode() != ISD::ADD || + XORSrc1.getOpcode() != ISD::SRA) + return NULL; + + SDValue ADDSrc0 = XORSrc0.getOperand(0); + SDValue ADDSrc1 = XORSrc0.getOperand(1); + SDValue SRASrc0 = XORSrc1.getOperand(0); + SDValue SRASrc1 = XORSrc1.getOperand(1); + ConstantSDNode *SRAConstant = dyn_cast(SRASrc1); + EVT XType = SRASrc0.getValueType(); + unsigned Size = XType.getSizeInBits() - 1; + + if (ADDSrc1 == XORSrc1 && + ADDSrc0 == SRASrc0 && + XType.isInteger() && + SRAConstant != NULL && + Size == SRAConstant->getZExtValue()) { + + unsigned Opcode = ARM::ABS; + if (Subtarget->isThumb2()) + Opcode = ARM::t2ABS; + + return CurDAG->SelectNodeTo(N, Opcode, VT, ADDSrc0); + } + + return NULL; +} + SDNode *ARMDAGToDAGISel::SelectConcatVector(SDNode *N) { // The only time a CONCAT_VECTORS operation can have legal types is when // two 64-bit vectors are concatenated to a 128-bit vector. @@ -2207,6 +2362,25 @@ SDNode *ARMDAGToDAGISel::SelectConcatVector(SDNode *N) { return PairDRegs(VT, N->getOperand(0), N->getOperand(1)); } +SDNode *ARMDAGToDAGISel::SelectAtomic64(SDNode *Node, unsigned Opc) { + SmallVector Ops; + Ops.push_back(Node->getOperand(1)); // Ptr + Ops.push_back(Node->getOperand(2)); // Low part of Val1 + Ops.push_back(Node->getOperand(3)); // High part of Val1 + if (Opc == ARM::ATOMCMPXCHG6432) { + Ops.push_back(Node->getOperand(4)); // Low part of Val2 + Ops.push_back(Node->getOperand(5)); // High part of Val2 + } + Ops.push_back(Node->getOperand(0)); // Chain + MachineSDNode::mmo_iterator MemOp = MF->allocateMemRefsArray(1); + MemOp[0] = cast(Node)->getMemOperand(); + SDNode *ResNode = CurDAG->getMachineNode(Opc, Node->getDebugLoc(), + MVT::i32, MVT::i32, MVT::Other, + Ops.data() ,Ops.size()); + cast(ResNode)->setMemRefs(MemOp, MemOp + 1); + return ResNode; +} + SDNode *ARMDAGToDAGISel::Select(SDNode *N) { DebugLoc dl = N->getDebugLoc(); @@ -2215,6 +2389,14 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) { switch (N->getOpcode()) { default: break; + case ISD::XOR: { + // Select special operations if XOR node forms integer ABS pattern + SDNode *ResNode = SelectABSOp(N); + if (ResNode) + return ResNode; + // Other cases are autogenerated. + break; + } case ISD::Constant: { unsigned Val = cast(N)->getZExtValue(); bool UseCP = true; @@ -2269,8 +2451,9 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) { int FI = cast(N)->getIndex(); SDValue TFI = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy()); if (Subtarget->isThumb1Only()) { - return CurDAG->SelectNodeTo(N, ARM::tADDrSPi, MVT::i32, TFI, - CurDAG->getTargetConstant(0, MVT::i32)); + SDValue Ops[] = { TFI, CurDAG->getTargetConstant(0, MVT::i32), + getAL(CurDAG), CurDAG->getRegister(0, MVT::i32) }; + return CurDAG->SelectNodeTo(N, ARM::tADDrSPi, MVT::i32, Ops, 4); } else { unsigned Opc = ((Subtarget->isThumb() && Subtarget->hasThumb2()) ? ARM::t2ADDri : ARM::ADDri); @@ -2307,7 +2490,7 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) { return CurDAG->SelectNodeTo(N, ARM::t2ADDrs, MVT::i32, Ops, 6); } else { SDValue Ops[] = { V, V, Reg0, ShImmOp, getAL(CurDAG), Reg0, Reg0 }; - return CurDAG->SelectNodeTo(N, ARM::ADDrs, MVT::i32, Ops, 7); + return CurDAG->SelectNodeTo(N, ARM::ADDrsi, MVT::i32, Ops, 7); } } if (isPowerOf2_32(RHSV+1)) { // 2^n-1? @@ -2323,7 +2506,7 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) { return CurDAG->SelectNodeTo(N, ARM::t2RSBrs, MVT::i32, Ops, 6); } else { SDValue Ops[] = { V, V, Reg0, ShImmOp, getAL(CurDAG), Reg0, Reg0 }; - return CurDAG->SelectNodeTo(N, ARM::RSBrs, MVT::i32, Ops, 7); + return CurDAG->SelectNodeTo(N, ARM::RSBrsi, MVT::i32, Ops, 7); } } } @@ -2986,6 +3169,23 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) { case ISD::CONCAT_VECTORS: return SelectConcatVector(N); + + case ARMISD::ATOMOR64_DAG: + return SelectAtomic64(N, ARM::ATOMOR6432); + case ARMISD::ATOMXOR64_DAG: + return SelectAtomic64(N, ARM::ATOMXOR6432); + case ARMISD::ATOMADD64_DAG: + return SelectAtomic64(N, ARM::ATOMADD6432); + case ARMISD::ATOMSUB64_DAG: + return SelectAtomic64(N, ARM::ATOMSUB6432); + case ARMISD::ATOMNAND64_DAG: + return SelectAtomic64(N, ARM::ATOMNAND6432); + case ARMISD::ATOMAND64_DAG: + return SelectAtomic64(N, ARM::ATOMAND6432); + case ARMISD::ATOMSWAP64_DAG: + return SelectAtomic64(N, ARM::ATOMSWAP6432); + case ARMISD::ATOMCMPXCHG64_DAG: + return SelectAtomic64(N, ARM::ATOMCMPXCHG6432); } return SelectCode(N); diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp index cf8c5baa8e7d7..e44e35673aebb 100644 --- a/lib/Target/ARM/ARMISelLowering.cpp +++ b/lib/Target/ARM/ARMISelLowering.cpp @@ -14,7 +14,6 @@ #define DEBUG_TYPE "arm-isel" #include "ARM.h" -#include "ARMAddressingModes.h" #include "ARMCallingConv.h" #include "ARMConstantPoolValue.h" #include "ARMISelLowering.h" @@ -24,6 +23,7 @@ #include "ARMSubtarget.h" #include "ARMTargetMachine.h" #include "ARMTargetObjectFile.h" +#include "MCTargetDesc/ARMAddressingModes.h" #include "llvm/CallingConv.h" #include "llvm/Constants.h" #include "llvm/Function.h" @@ -38,6 +38,7 @@ #include "llvm/CodeGen/MachineFrameInfo.h" #include "llvm/CodeGen/MachineFunction.h" #include "llvm/CodeGen/MachineInstrBuilder.h" +#include "llvm/CodeGen/MachineModuleInfo.h" #include "llvm/CodeGen/MachineRegisterInfo.h" #include "llvm/CodeGen/PseudoSourceValue.h" #include "llvm/CodeGen/SelectionDAG.h" @@ -106,7 +107,7 @@ void ARMTargetLowering::addTypeForNEON(EVT VT, EVT PromotedLdStVT, EVT ElemTy = VT.getVectorElementType(); if (ElemTy != MVT::i64 && ElemTy != MVT::f64) - setOperationAction(ISD::VSETCC, VT.getSimpleVT(), Custom); + setOperationAction(ISD::SETCC, VT.getSimpleVT(), Custom); setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT.getSimpleVT(), Custom); if (ElemTy != MVT::i32) { setOperationAction(ISD::SINT_TO_FP, VT.getSimpleVT(), Expand); @@ -178,6 +179,8 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM) RegInfo = TM.getRegisterInfo(); Itins = TM.getInstrItineraryData(); + setBooleanVectorContents(ZeroOrNegativeOneBooleanContent); + if (Subtarget->isTargetDarwin()) { // Uses VFP for Thumb libfuncs if available. if (Subtarget->isThumb() && Subtarget->hasVFP2()) { @@ -419,6 +422,13 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM) setLibcallName(RTLIB::MEMSET, "__aeabi_memset"); } + // Use divmod compiler-rt calls for iOS 5.0 and later. + if (Subtarget->getTargetTriple().getOS() == Triple::IOS && + !Subtarget->getTargetTriple().isOSVersionLT(5, 0)) { + setLibcallName(RTLIB::SDIVREM_I32, "__divmodsi4"); + setLibcallName(RTLIB::UDIVREM_I32, "__udivmodsi4"); + } + if (Subtarget->isThumb1Only()) addRegisterClass(MVT::i32, ARM::tGPRRegisterClass); else @@ -453,7 +463,7 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM) setOperationAction(ISD::FDIV, MVT::v2f64, Expand); setOperationAction(ISD::FREM, MVT::v2f64, Expand); setOperationAction(ISD::FCOPYSIGN, MVT::v2f64, Expand); - setOperationAction(ISD::VSETCC, MVT::v2f64, Expand); + setOperationAction(ISD::SETCC, MVT::v2f64, Expand); setOperationAction(ISD::FNEG, MVT::v2f64, Expand); setOperationAction(ISD::FABS, MVT::v2f64, Expand); setOperationAction(ISD::FSQRT, MVT::v2f64, Expand); @@ -485,8 +495,8 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM) setOperationAction(ISD::SDIV, MVT::v8i8, Custom); setOperationAction(ISD::UDIV, MVT::v4i16, Custom); setOperationAction(ISD::UDIV, MVT::v8i8, Custom); - setOperationAction(ISD::VSETCC, MVT::v1i64, Expand); - setOperationAction(ISD::VSETCC, MVT::v2i64, Expand); + setOperationAction(ISD::SETCC, MVT::v1i64, Expand); + setOperationAction(ISD::SETCC, MVT::v2i64, Expand); // Neon does not have single instruction SINT_TO_FP and UINT_TO_FP with // a destination type that is wider than the source. setOperationAction(ISD::SINT_TO_FP, MVT::v4i16, Custom); @@ -551,6 +561,14 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM) setOperationAction(ISD::SRL, MVT::i64, Custom); setOperationAction(ISD::SRA, MVT::i64, Custom); + if (!Subtarget->isThumb1Only()) { + // FIXME: We should do this for Thumb1 as well. + setOperationAction(ISD::ADDC, MVT::i32, Custom); + setOperationAction(ISD::ADDE, MVT::i32, Custom); + setOperationAction(ISD::SUBC, MVT::i32, Custom); + setOperationAction(ISD::SUBE, MVT::i32, Custom); + } + // ARM does not have ROTL. setOperationAction(ISD::ROTL, MVT::i32, Expand); setOperationAction(ISD::CTTZ, MVT::i32, Custom); @@ -596,62 +614,46 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM) setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Expand); // ARMv6 Thumb1 (except for CPUs that support dmb / dsb) and earlier use // the default expansion. + // FIXME: This should be checking for v6k, not just v6. if (Subtarget->hasDataBarrier() || (Subtarget->hasV6Ops() && !Subtarget->isThumb())) { // membarrier needs custom lowering; the rest are legal and handled // normally. setOperationAction(ISD::MEMBARRIER, MVT::Other, Custom); + setOperationAction(ISD::ATOMIC_FENCE, MVT::Other, Custom); + // Custom lowering for 64-bit ops + setOperationAction(ISD::ATOMIC_LOAD_ADD, MVT::i64, Custom); + setOperationAction(ISD::ATOMIC_LOAD_SUB, MVT::i64, Custom); + setOperationAction(ISD::ATOMIC_LOAD_AND, MVT::i64, Custom); + setOperationAction(ISD::ATOMIC_LOAD_OR, MVT::i64, Custom); + setOperationAction(ISD::ATOMIC_LOAD_XOR, MVT::i64, Custom); + setOperationAction(ISD::ATOMIC_SWAP, MVT::i64, Custom); + setOperationAction(ISD::ATOMIC_CMP_SWAP, MVT::i64, Custom); + // Automatically insert fences (dmb ist) around ATOMIC_SWAP etc. + setInsertFencesForAtomic(true); } else { // Set them all for expansion, which will force libcalls. setOperationAction(ISD::MEMBARRIER, MVT::Other, Expand); - setOperationAction(ISD::ATOMIC_CMP_SWAP, MVT::i8, Expand); - setOperationAction(ISD::ATOMIC_CMP_SWAP, MVT::i16, Expand); + setOperationAction(ISD::ATOMIC_FENCE, MVT::Other, Expand); setOperationAction(ISD::ATOMIC_CMP_SWAP, MVT::i32, Expand); - setOperationAction(ISD::ATOMIC_SWAP, MVT::i8, Expand); - setOperationAction(ISD::ATOMIC_SWAP, MVT::i16, Expand); setOperationAction(ISD::ATOMIC_SWAP, MVT::i32, Expand); - setOperationAction(ISD::ATOMIC_LOAD_ADD, MVT::i8, Expand); - setOperationAction(ISD::ATOMIC_LOAD_ADD, MVT::i16, Expand); setOperationAction(ISD::ATOMIC_LOAD_ADD, MVT::i32, Expand); - setOperationAction(ISD::ATOMIC_LOAD_SUB, MVT::i8, Expand); - setOperationAction(ISD::ATOMIC_LOAD_SUB, MVT::i16, Expand); setOperationAction(ISD::ATOMIC_LOAD_SUB, MVT::i32, Expand); - setOperationAction(ISD::ATOMIC_LOAD_AND, MVT::i8, Expand); - setOperationAction(ISD::ATOMIC_LOAD_AND, MVT::i16, Expand); setOperationAction(ISD::ATOMIC_LOAD_AND, MVT::i32, Expand); - setOperationAction(ISD::ATOMIC_LOAD_OR, MVT::i8, Expand); - setOperationAction(ISD::ATOMIC_LOAD_OR, MVT::i16, Expand); setOperationAction(ISD::ATOMIC_LOAD_OR, MVT::i32, Expand); - setOperationAction(ISD::ATOMIC_LOAD_XOR, MVT::i8, Expand); - setOperationAction(ISD::ATOMIC_LOAD_XOR, MVT::i16, Expand); setOperationAction(ISD::ATOMIC_LOAD_XOR, MVT::i32, Expand); - setOperationAction(ISD::ATOMIC_LOAD_NAND, MVT::i8, Expand); - setOperationAction(ISD::ATOMIC_LOAD_NAND, MVT::i16, Expand); setOperationAction(ISD::ATOMIC_LOAD_NAND, MVT::i32, Expand); - setOperationAction(ISD::ATOMIC_LOAD_MIN, MVT::i8, Expand); - setOperationAction(ISD::ATOMIC_LOAD_MIN, MVT::i16, Expand); setOperationAction(ISD::ATOMIC_LOAD_MIN, MVT::i32, Expand); - setOperationAction(ISD::ATOMIC_LOAD_MAX, MVT::i8, Expand); - setOperationAction(ISD::ATOMIC_LOAD_MAX, MVT::i16, Expand); setOperationAction(ISD::ATOMIC_LOAD_MAX, MVT::i32, Expand); - setOperationAction(ISD::ATOMIC_LOAD_UMIN, MVT::i8, Expand); - setOperationAction(ISD::ATOMIC_LOAD_UMIN, MVT::i16, Expand); setOperationAction(ISD::ATOMIC_LOAD_UMIN, MVT::i32, Expand); - setOperationAction(ISD::ATOMIC_LOAD_UMAX, MVT::i8, Expand); - setOperationAction(ISD::ATOMIC_LOAD_UMAX, MVT::i16, Expand); setOperationAction(ISD::ATOMIC_LOAD_UMAX, MVT::i32, Expand); + // Mark ATOMIC_LOAD and ATOMIC_STORE custom so we can handle the + // Unordered/Monotonic case. + setOperationAction(ISD::ATOMIC_LOAD, MVT::i32, Custom); + setOperationAction(ISD::ATOMIC_STORE, MVT::i32, Custom); // Since the libcalls include locking, fold in the fences setShouldFoldAtomicFences(true); } - // 64-bit versions are always libcalls (for now) - setOperationAction(ISD::ATOMIC_CMP_SWAP, MVT::i64, Expand); - setOperationAction(ISD::ATOMIC_SWAP, MVT::i64, Expand); - setOperationAction(ISD::ATOMIC_LOAD_ADD, MVT::i64, Expand); - setOperationAction(ISD::ATOMIC_LOAD_SUB, MVT::i64, Expand); - setOperationAction(ISD::ATOMIC_LOAD_AND, MVT::i64, Expand); - setOperationAction(ISD::ATOMIC_LOAD_OR, MVT::i64, Expand); - setOperationAction(ISD::ATOMIC_LOAD_XOR, MVT::i64, Expand); - setOperationAction(ISD::ATOMIC_LOAD_NAND, MVT::i64, Expand); setOperationAction(ISD::PREFETCH, MVT::Other, Custom); @@ -839,6 +841,11 @@ const char *ARMTargetLowering::getTargetNodeName(unsigned Opcode) const { case ARMISD::SRA_FLAG: return "ARMISD::SRA_FLAG"; case ARMISD::RRX: return "ARMISD::RRX"; + case ARMISD::ADDC: return "ARMISD::ADDC"; + case ARMISD::ADDE: return "ARMISD::ADDE"; + case ARMISD::SUBC: return "ARMISD::SUBC"; + case ARMISD::SUBE: return "ARMISD::SUBE"; + case ARMISD::VMOVRRD: return "ARMISD::VMOVRRD"; case ARMISD::VMOVDRR: return "ARMISD::VMOVDRR"; @@ -935,6 +942,11 @@ const char *ARMTargetLowering::getTargetNodeName(unsigned Opcode) const { } } +EVT ARMTargetLowering::getSetCCResultType(EVT VT) const { + if (!VT.isVector()) return getPointerTy(); + return VT.changeVectorElementTypeToInteger(); +} + /// getRegClassFor - Return the register class that should be used for the /// specified value type. TargetRegisterClass *ARMTargetLowering::getRegClassFor(EVT VT) const { @@ -1210,8 +1222,8 @@ ARMTargetLowering::LowerCall(SDValue Chain, SDValue Callee, MachineFunction &MF = DAG.getMachineFunction(); bool IsStructRet = (Outs.empty()) ? false : Outs[0].Flags.isSRet(); bool IsSibCall = false; - // Temporarily disable tail calls so things don't break. - if (!EnableARMTailCalls) + // Disable tail calls if they're not supported. + if (!EnableARMTailCalls && !Subtarget->supportsTailCall()) isTailCall = false; if (isTailCall) { // Check if it's really possible to do a tail call. @@ -1336,10 +1348,12 @@ ARMTargetLowering::LowerCall(SDValue Chain, SDValue Callee, SDValue Src = DAG.getNode(ISD::ADD, dl, getPointerTy(), Arg, SrcOffset); SDValue SizeNode = DAG.getConstant(Flags.getByValSize() - 4*offset, MVT::i32); + // TODO: Disable AlwaysInline when it becomes possible + // to emit a nested call sequence. MemOpChains.push_back(DAG.getMemcpy(Chain, dl, Dst, Src, SizeNode, Flags.getByValAlign(), /*isVolatile=*/false, - /*AlwaysInline=*/false, + /*AlwaysInline=*/true, MachinePointerInfo(0), MachinePointerInfo(0))); @@ -1404,9 +1418,9 @@ ARMTargetLowering::LowerCall(SDValue Chain, SDValue Callee, const GlobalValue *GV = G->getGlobal(); // Create a constant pool entry for the callee address unsigned ARMPCLabelIndex = AFI->createPICLabelUId(); - ARMConstantPoolValue *CPV = new ARMConstantPoolValue(GV, - ARMPCLabelIndex, - ARMCP::CPValue, 0); + ARMConstantPoolValue *CPV = + ARMConstantPoolConstant::Create(GV, ARMPCLabelIndex, ARMCP::CPValue, 0); + // Get the address of the callee into a register SDValue CPAddr = DAG.getTargetConstantPool(CPV, getPointerTy(), 4); CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr); @@ -1419,8 +1433,9 @@ ARMTargetLowering::LowerCall(SDValue Chain, SDValue Callee, // Create a constant pool entry for the callee address unsigned ARMPCLabelIndex = AFI->createPICLabelUId(); - ARMConstantPoolValue *CPV = new ARMConstantPoolValue(*DAG.getContext(), - Sym, ARMPCLabelIndex, 0); + ARMConstantPoolValue *CPV = + ARMConstantPoolSymbol::Create(*DAG.getContext(), Sym, + ARMPCLabelIndex, 0); // Get the address of the callee into a register SDValue CPAddr = DAG.getTargetConstantPool(CPV, getPointerTy(), 4); CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr); @@ -1441,9 +1456,8 @@ ARMTargetLowering::LowerCall(SDValue Chain, SDValue Callee, // tBX takes a register source operand. if (isARMFunc && Subtarget->isThumb1Only() && !Subtarget->hasV5TOps()) { unsigned ARMPCLabelIndex = AFI->createPICLabelUId(); - ARMConstantPoolValue *CPV = new ARMConstantPoolValue(GV, - ARMPCLabelIndex, - ARMCP::CPValue, 4); + ARMConstantPoolValue *CPV = + ARMConstantPoolConstant::Create(GV, ARMPCLabelIndex, ARMCP::CPValue, 4); SDValue CPAddr = DAG.getTargetConstantPool(CPV, getPointerTy(), 4); CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr); Callee = DAG.getLoad(getPointerTy(), dl, @@ -1470,8 +1484,9 @@ ARMTargetLowering::LowerCall(SDValue Chain, SDValue Callee, const char *Sym = S->getSymbol(); if (isARMFunc && Subtarget->isThumb1Only() && !Subtarget->hasV5TOps()) { unsigned ARMPCLabelIndex = AFI->createPICLabelUId(); - ARMConstantPoolValue *CPV = new ARMConstantPoolValue(*DAG.getContext(), - Sym, ARMPCLabelIndex, 4); + ARMConstantPoolValue *CPV = + ARMConstantPoolSymbol::Create(*DAG.getContext(), Sym, + ARMPCLabelIndex, 4); SDValue CPAddr = DAG.getTargetConstantPool(CPV, getPointerTy(), 4); CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr); Callee = DAG.getLoad(getPointerTy(), dl, @@ -1940,9 +1955,9 @@ SDValue ARMTargetLowering::LowerBlockAddress(SDValue Op, } else { unsigned PCAdj = Subtarget->isThumb() ? 4 : 8; ARMPCLabelIndex = AFI->createPICLabelUId(); - ARMConstantPoolValue *CPV = new ARMConstantPoolValue(BA, ARMPCLabelIndex, - ARMCP::CPBlockAddress, - PCAdj); + ARMConstantPoolValue *CPV = + ARMConstantPoolConstant::Create(BA, ARMPCLabelIndex, + ARMCP::CPBlockAddress, PCAdj); CPAddr = DAG.getTargetConstantPool(CPV, PtrVT, 4); } CPAddr = DAG.getNode(ARMISD::Wrapper, DL, PtrVT, CPAddr); @@ -1966,8 +1981,8 @@ ARMTargetLowering::LowerToTLSGeneralDynamicModel(GlobalAddressSDNode *GA, ARMFunctionInfo *AFI = MF.getInfo(); unsigned ARMPCLabelIndex = AFI->createPICLabelUId(); ARMConstantPoolValue *CPV = - new ARMConstantPoolValue(GA->getGlobal(), ARMPCLabelIndex, - ARMCP::CPValue, PCAdj, ARMCP::TLSGD, true); + ARMConstantPoolConstant::Create(GA->getGlobal(), ARMPCLabelIndex, + ARMCP::CPValue, PCAdj, ARMCP::TLSGD, true); SDValue Argument = DAG.getTargetConstantPool(CPV, PtrVT, 4); Argument = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, Argument); Argument = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), Argument, @@ -1982,11 +1997,11 @@ ARMTargetLowering::LowerToTLSGeneralDynamicModel(GlobalAddressSDNode *GA, ArgListTy Args; ArgListEntry Entry; Entry.Node = Argument; - Entry.Ty = (const Type *) Type::getInt32Ty(*DAG.getContext()); + Entry.Ty = (Type *) Type::getInt32Ty(*DAG.getContext()); Args.push_back(Entry); // FIXME: is there useful debug info available here? std::pair CallResult = - LowerCallTo(Chain, (const Type *) Type::getInt32Ty(*DAG.getContext()), + LowerCallTo(Chain, (Type *) Type::getInt32Ty(*DAG.getContext()), false, false, false, false, 0, CallingConv::C, false, /*isReturnValueUsed=*/true, DAG.getExternalSymbol("__tls_get_addr", PtrVT), Args, DAG, dl); @@ -2013,8 +2028,9 @@ ARMTargetLowering::LowerToTLSExecModels(GlobalAddressSDNode *GA, // Initial exec model. unsigned char PCAdj = Subtarget->isThumb() ? 4 : 8; ARMConstantPoolValue *CPV = - new ARMConstantPoolValue(GA->getGlobal(), ARMPCLabelIndex, - ARMCP::CPValue, PCAdj, ARMCP::GOTTPOFF, true); + ARMConstantPoolConstant::Create(GA->getGlobal(), ARMPCLabelIndex, + ARMCP::CPValue, PCAdj, ARMCP::GOTTPOFF, + true); Offset = DAG.getTargetConstantPool(CPV, PtrVT, 4); Offset = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, Offset); Offset = DAG.getLoad(PtrVT, dl, Chain, Offset, @@ -2030,7 +2046,8 @@ ARMTargetLowering::LowerToTLSExecModels(GlobalAddressSDNode *GA, false, false, 0); } else { // local exec model - ARMConstantPoolValue *CPV = new ARMConstantPoolValue(GV, ARMCP::TPOFF); + ARMConstantPoolValue *CPV = + ARMConstantPoolConstant::Create(GV, ARMCP::TPOFF); Offset = DAG.getTargetConstantPool(CPV, PtrVT, 4); Offset = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, Offset); Offset = DAG.getLoad(PtrVT, dl, Chain, Offset, @@ -2066,7 +2083,8 @@ SDValue ARMTargetLowering::LowerGlobalAddressELF(SDValue Op, if (RelocM == Reloc::PIC_) { bool UseGOTOFF = GV->hasLocalLinkage() || GV->hasHiddenVisibility(); ARMConstantPoolValue *CPV = - new ARMConstantPoolValue(GV, UseGOTOFF ? ARMCP::GOTOFF : ARMCP::GOT); + ARMConstantPoolConstant::Create(GV, + UseGOTOFF ? ARMCP::GOTOFF : ARMCP::GOT); SDValue CPAddr = DAG.getTargetConstantPool(CPV, PtrVT, 4); CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr); SDValue Result = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), @@ -2135,7 +2153,8 @@ SDValue ARMTargetLowering::LowerGlobalAddressDarwin(SDValue Op, ARMPCLabelIndex = AFI->createPICLabelUId(); unsigned PCAdj = (RelocM != Reloc::PIC_) ? 0 : (Subtarget->isThumb()?4:8); ARMConstantPoolValue *CPV = - new ARMConstantPoolValue(GV, ARMPCLabelIndex, ARMCP::CPValue, PCAdj); + ARMConstantPoolConstant::Create(GV, ARMPCLabelIndex, ARMCP::CPValue, + PCAdj); CPAddr = DAG.getTargetConstantPool(CPV, PtrVT, 4); } CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr); @@ -2167,9 +2186,9 @@ SDValue ARMTargetLowering::LowerGLOBAL_OFFSET_TABLE(SDValue Op, EVT PtrVT = getPointerTy(); DebugLoc dl = Op.getDebugLoc(); unsigned PCAdj = Subtarget->isThumb() ? 4 : 8; - ARMConstantPoolValue *CPV = new ARMConstantPoolValue(*DAG.getContext(), - "_GLOBAL_OFFSET_TABLE_", - ARMPCLabelIndex, PCAdj); + ARMConstantPoolValue *CPV = + ARMConstantPoolSymbol::Create(*DAG.getContext(), "_GLOBAL_OFFSET_TABLE_", + ARMPCLabelIndex, PCAdj); SDValue CPAddr = DAG.getTargetConstantPool(CPV, PtrVT, 4); CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr); SDValue Result = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), CPAddr, @@ -2191,7 +2210,8 @@ SDValue ARMTargetLowering::LowerEH_SJLJ_SETJMP(SDValue Op, SelectionDAG &DAG) const { DebugLoc dl = Op.getDebugLoc(); SDValue Val = DAG.getConstant(0, MVT::i32); - return DAG.getNode(ARMISD::EH_SJLJ_SETJMP, dl, MVT::i32, Op.getOperand(0), + return DAG.getNode(ARMISD::EH_SJLJ_SETJMP, dl, + DAG.getVTList(MVT::i32, MVT::Other), Op.getOperand(0), Op.getOperand(1), Val); } @@ -2224,8 +2244,8 @@ ARMTargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG, unsigned PCAdj = (RelocM != Reloc::PIC_) ? 0 : (Subtarget->isThumb() ? 4 : 8); ARMConstantPoolValue *CPV = - new ARMConstantPoolValue(MF.getFunction(), ARMPCLabelIndex, - ARMCP::CPLSDA, PCAdj); + ARMConstantPoolConstant::Create(MF.getFunction(), ARMPCLabelIndex, + ARMCP::CPLSDA, PCAdj); CPAddr = DAG.getTargetConstantPool(CPV, PtrVT, 4); CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr); SDValue Result = @@ -2277,6 +2297,25 @@ static SDValue LowerMEMBARRIER(SDValue Op, SelectionDAG &DAG, DAG.getConstant(DMBOpt, MVT::i32)); } + +static SDValue LowerATOMIC_FENCE(SDValue Op, SelectionDAG &DAG, + const ARMSubtarget *Subtarget) { + // FIXME: handle "fence singlethread" more efficiently. + DebugLoc dl = Op.getDebugLoc(); + if (!Subtarget->hasDataBarrier()) { + // Some ARMv6 cpus can support data barriers with an mcr instruction. + // Thumb1 and pre-v6 ARM mode use a libcall instead and should never get + // here. + assert(Subtarget->hasV6Ops() && !Subtarget->isThumb() && + "Unexpected ISD::MEMBARRIER encountered. Should be libcall!"); + return DAG.getNode(ARMISD::MEMBARRIER_MCR, dl, MVT::Other, Op.getOperand(0), + DAG.getConstant(0, MVT::i32)); + } + + return DAG.getNode(ARMISD::MEMBARRIER, dl, MVT::Other, Op.getOperand(0), + DAG.getConstant(ARM_MB::ISH, MVT::i32)); +} + static SDValue LowerPREFETCH(SDValue Op, SelectionDAG &DAG, const ARMSubtarget *Subtarget) { // ARM pre v5TE and Thumb1 does not have preload instructions. @@ -2754,7 +2793,7 @@ SDValue ARMTargetLowering::LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const { SDValue ARMcc; SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32); SDValue Cmp = getARMCmp(LHS, RHS, CC, ARMcc, DAG, dl); - return DAG.getNode(ARMISD::CMOV, dl, VT, FalseVal, TrueVal, ARMcc, CCR, Cmp); + return DAG.getNode(ARMISD::CMOV, dl, VT, FalseVal, TrueVal, ARMcc, CCR,Cmp); } ARMCC::CondCodes CondCode, CondCode2; @@ -2993,8 +3032,8 @@ static SDValue LowerVectorINT_TO_FP(SDValue Op, SelectionDAG &DAG) { EVT VT = Op.getValueType(); DebugLoc dl = Op.getDebugLoc(); - EVT OperandVT = Op.getOperand(0).getValueType(); - assert(OperandVT == MVT::v4i16 && "Invalid type for custom lowering!"); + assert(Op.getOperand(0).getValueType() == MVT::v4i16 && + "Invalid type for custom lowering!"); if (VT != MVT::v4f32) return DAG.UnrollVectorOp(Op.getNode()); @@ -3905,8 +3944,7 @@ SDValue ARMTargetLowering::LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG, } // Try an immediate VMVN. - uint64_t NegatedImm = (SplatBits.getZExtValue() ^ - ((1LL << SplatBitSize) - 1)); + uint64_t NegatedImm = (~SplatBits).getZExtValue(); Val = isNEONModifiedImm(NegatedImm, SplatUndef.getZExtValue(), SplatBitSize, DAG, VmovVT, VT.is128BitVector(), @@ -4019,6 +4057,14 @@ SDValue ARMTargetLowering::ReconstructShuffle(SDValue Op, // A shuffle can only come from building a vector from various // elements of other vectors. return SDValue(); + } else if (V.getOperand(0).getValueType().getVectorElementType() != + VT.getVectorElementType()) { + // This code doesn't know how to handle shuffles where the vector + // element types do not match (this happens because type legalization + // promotes the return type of EXTRACT_VECTOR_ELT). + // FIXME: It might be appropriate to extend this code to handle + // mismatched types. + return SDValue(); } // Record this extraction against the appropriate vector if possible... @@ -4819,6 +4865,71 @@ static SDValue LowerUDIV(SDValue Op, SelectionDAG &DAG) { return N0; } +static SDValue LowerADDC_ADDE_SUBC_SUBE(SDValue Op, SelectionDAG &DAG) { + EVT VT = Op.getNode()->getValueType(0); + SDVTList VTs = DAG.getVTList(VT, MVT::i32); + + unsigned Opc; + bool ExtraOp = false; + switch (Op.getOpcode()) { + default: assert(0 && "Invalid code"); + case ISD::ADDC: Opc = ARMISD::ADDC; break; + case ISD::ADDE: Opc = ARMISD::ADDE; ExtraOp = true; break; + case ISD::SUBC: Opc = ARMISD::SUBC; break; + case ISD::SUBE: Opc = ARMISD::SUBE; ExtraOp = true; break; + } + + if (!ExtraOp) + return DAG.getNode(Opc, Op->getDebugLoc(), VTs, Op.getOperand(0), + Op.getOperand(1)); + return DAG.getNode(Opc, Op->getDebugLoc(), VTs, Op.getOperand(0), + Op.getOperand(1), Op.getOperand(2)); +} + +static SDValue LowerAtomicLoadStore(SDValue Op, SelectionDAG &DAG) { + // Monotonic load/store is legal for all targets + if (cast(Op)->getOrdering() <= Monotonic) + return Op; + + // Aquire/Release load/store is not legal for targets without a + // dmb or equivalent available. + return SDValue(); +} + + +static void +ReplaceATOMIC_OP_64(SDNode *Node, SmallVectorImpl& Results, + SelectionDAG &DAG, unsigned NewOp) { + DebugLoc dl = Node->getDebugLoc(); + assert (Node->getValueType(0) == MVT::i64 && + "Only know how to expand i64 atomics"); + + SmallVector Ops; + Ops.push_back(Node->getOperand(0)); // Chain + Ops.push_back(Node->getOperand(1)); // Ptr + // Low part of Val1 + Ops.push_back(DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::i32, + Node->getOperand(2), DAG.getIntPtrConstant(0))); + // High part of Val1 + Ops.push_back(DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::i32, + Node->getOperand(2), DAG.getIntPtrConstant(1))); + if (NewOp == ARMISD::ATOMCMPXCHG64_DAG) { + // High part of Val1 + Ops.push_back(DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::i32, + Node->getOperand(3), DAG.getIntPtrConstant(0))); + // High part of Val2 + Ops.push_back(DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::i32, + Node->getOperand(3), DAG.getIntPtrConstant(1))); + } + SDVTList Tys = DAG.getVTList(MVT::i32, MVT::i32, MVT::Other); + SDValue Result = + DAG.getMemIntrinsicNode(NewOp, dl, Tys, Ops.data(), Ops.size(), MVT::i64, + cast(Node)->getMemOperand()); + SDValue OpsF[] = { Result.getValue(0), Result.getValue(1) }; + Results.push_back(DAG.getNode(ISD::BUILD_PAIR, dl, MVT::i64, OpsF, 2)); + Results.push_back(Result.getValue(2)); +} + SDValue ARMTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const { switch (Op.getOpcode()) { default: llvm_unreachable("Don't know how to custom lower this!"); @@ -4834,6 +4945,7 @@ SDValue ARMTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const { case ISD::BR_JT: return LowerBR_JT(Op, DAG); case ISD::VASTART: return LowerVASTART(Op, DAG); case ISD::MEMBARRIER: return LowerMEMBARRIER(Op, DAG, Subtarget); + case ISD::ATOMIC_FENCE: return LowerATOMIC_FENCE(Op, DAG, Subtarget); case ISD::PREFETCH: return LowerPREFETCH(Op, DAG, Subtarget); case ISD::SINT_TO_FP: case ISD::UINT_TO_FP: return LowerINT_TO_FP(Op, DAG); @@ -4856,7 +4968,7 @@ SDValue ARMTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const { case ISD::SRL_PARTS: case ISD::SRA_PARTS: return LowerShiftRightParts(Op, DAG); case ISD::CTTZ: return LowerCTTZ(Op.getNode(), DAG, Subtarget); - case ISD::VSETCC: return LowerVSETCC(Op, DAG); + case ISD::SETCC: return LowerVSETCC(Op, DAG); case ISD::BUILD_VECTOR: return LowerBUILD_VECTOR(Op, DAG, Subtarget); case ISD::VECTOR_SHUFFLE: return LowerVECTOR_SHUFFLE(Op, DAG); case ISD::EXTRACT_VECTOR_ELT: return LowerEXTRACT_VECTOR_ELT(Op, DAG); @@ -4865,6 +4977,12 @@ SDValue ARMTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const { case ISD::MUL: return LowerMUL(Op, DAG); case ISD::SDIV: return LowerSDIV(Op, DAG); case ISD::UDIV: return LowerUDIV(Op, DAG); + case ISD::ADDC: + case ISD::ADDE: + case ISD::SUBC: + case ISD::SUBE: return LowerADDC_ADDE_SUBC_SUBE(Op, DAG); + case ISD::ATOMIC_LOAD: + case ISD::ATOMIC_STORE: return LowerAtomicLoadStore(Op, DAG); } return SDValue(); } @@ -4886,6 +5004,30 @@ void ARMTargetLowering::ReplaceNodeResults(SDNode *N, case ISD::SRA: Res = Expand64BitShift(N, DAG, Subtarget); break; + case ISD::ATOMIC_LOAD_ADD: + ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMADD64_DAG); + return; + case ISD::ATOMIC_LOAD_AND: + ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMAND64_DAG); + return; + case ISD::ATOMIC_LOAD_NAND: + ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMNAND64_DAG); + return; + case ISD::ATOMIC_LOAD_OR: + ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMOR64_DAG); + return; + case ISD::ATOMIC_LOAD_SUB: + ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMSUB64_DAG); + return; + case ISD::ATOMIC_LOAD_XOR: + ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMXOR64_DAG); + return; + case ISD::ATOMIC_SWAP: + ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMSWAP64_DAG); + return; + case ISD::ATOMIC_CMP_SWAP: + ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMCMPXCHG64_DAG); + return; } if (Res.getNode()) Results.push_back(Res); @@ -4963,7 +5105,10 @@ ARMTargetLowering::EmitAtomicCmpSwap(MachineInstr *MI, // cmp dest, oldval // bne exitMBB BB = loop1MBB; - AddDefaultPred(BuildMI(BB, dl, TII->get(ldrOpc), dest).addReg(ptr)); + MachineInstrBuilder MIB = BuildMI(BB, dl, TII->get(ldrOpc), dest).addReg(ptr); + if (ldrOpc == ARM::t2LDREX) + MIB.addImm(0); + AddDefaultPred(MIB); AddDefaultPred(BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2CMPrr : ARM::CMPrr)) .addReg(dest).addReg(oldval)); BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc)) @@ -4976,8 +5121,10 @@ ARMTargetLowering::EmitAtomicCmpSwap(MachineInstr *MI, // cmp scratch, #0 // bne loop1MBB BB = loop2MBB; - AddDefaultPred(BuildMI(BB, dl, TII->get(strOpc), scratch).addReg(newval) - .addReg(ptr)); + MIB = BuildMI(BB, dl, TII->get(strOpc), scratch).addReg(newval).addReg(ptr); + if (strOpc == ARM::t2STREX) + MIB.addImm(0); + AddDefaultPred(MIB); AddDefaultPred(BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2CMPri : ARM::CMPri)) .addReg(scratch).addImm(0)); BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc)) @@ -5063,7 +5210,10 @@ ARMTargetLowering::EmitAtomicBinary(MachineInstr *MI, MachineBasicBlock *BB, // bne- loopMBB // fallthrough --> exitMBB BB = loopMBB; - AddDefaultPred(BuildMI(BB, dl, TII->get(ldrOpc), dest).addReg(ptr)); + MachineInstrBuilder MIB = BuildMI(BB, dl, TII->get(ldrOpc), dest).addReg(ptr); + if (ldrOpc == ARM::t2LDREX) + MIB.addImm(0); + AddDefaultPred(MIB); if (BinOpcode) { // operand order needs to go the other way for NAND if (BinOpcode == ARM::BICrr || BinOpcode == ARM::t2BICrr) @@ -5074,8 +5224,10 @@ ARMTargetLowering::EmitAtomicBinary(MachineInstr *MI, MachineBasicBlock *BB, addReg(dest).addReg(incr)).addReg(0); } - AddDefaultPred(BuildMI(BB, dl, TII->get(strOpc), scratch).addReg(scratch2) - .addReg(ptr)); + MIB = BuildMI(BB, dl, TII->get(strOpc), scratch).addReg(scratch2).addReg(ptr); + if (strOpc == ARM::t2STREX) + MIB.addImm(0); + AddDefaultPred(MIB); AddDefaultPred(BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2CMPri : ARM::CMPri)) .addReg(scratch).addImm(0)); BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc)) @@ -5125,12 +5277,12 @@ ARMTargetLowering::EmitAtomicBinaryMinMax(MachineInstr *MI, case 1: ldrOpc = isThumb2 ? ARM::t2LDREXB : ARM::LDREXB; strOpc = isThumb2 ? ARM::t2STREXB : ARM::STREXB; - extendOpc = isThumb2 ? ARM::t2SXTBr : ARM::SXTBr; + extendOpc = isThumb2 ? ARM::t2SXTB : ARM::SXTB; break; case 2: ldrOpc = isThumb2 ? ARM::t2LDREXH : ARM::LDREXH; strOpc = isThumb2 ? ARM::t2STREXH : ARM::STREXH; - extendOpc = isThumb2 ? ARM::t2SXTHr : ARM::SXTHr; + extendOpc = isThumb2 ? ARM::t2SXTH : ARM::SXTH; break; case 4: ldrOpc = isThumb2 ? ARM::t2LDREX : ARM::LDREX; @@ -5170,12 +5322,17 @@ ARMTargetLowering::EmitAtomicBinaryMinMax(MachineInstr *MI, // bne- loopMBB // fallthrough --> exitMBB BB = loopMBB; - AddDefaultPred(BuildMI(BB, dl, TII->get(ldrOpc), dest).addReg(ptr)); + MachineInstrBuilder MIB = BuildMI(BB, dl, TII->get(ldrOpc), dest).addReg(ptr); + if (ldrOpc == ARM::t2LDREX) + MIB.addImm(0); + AddDefaultPred(MIB); // Sign extend the value, if necessary. if (signExtend && extendOpc) { oldval = MRI.createVirtualRegister(ARM::GPRRegisterClass); - AddDefaultPred(BuildMI(BB, dl, TII->get(extendOpc), oldval).addReg(dest)); + AddDefaultPred(BuildMI(BB, dl, TII->get(extendOpc), oldval) + .addReg(dest) + .addImm(0)); } // Build compare and cmov instructions. @@ -5184,8 +5341,10 @@ ARMTargetLowering::EmitAtomicBinaryMinMax(MachineInstr *MI, BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2MOVCCr : ARM::MOVCCr), scratch2) .addReg(oldval).addReg(incr).addImm(Cond).addReg(ARM::CPSR); - AddDefaultPred(BuildMI(BB, dl, TII->get(strOpc), scratch).addReg(scratch2) - .addReg(ptr)); + MIB = BuildMI(BB, dl, TII->get(strOpc), scratch).addReg(scratch2).addReg(ptr); + if (strOpc == ARM::t2STREX) + MIB.addImm(0); + AddDefaultPred(MIB); AddDefaultPred(BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2CMPri : ARM::CMPri)) .addReg(scratch).addImm(0)); BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc)) @@ -5203,79 +5362,596 @@ ARMTargetLowering::EmitAtomicBinaryMinMax(MachineInstr *MI, return BB; } -static -MachineBasicBlock *OtherSucc(MachineBasicBlock *MBB, MachineBasicBlock *Succ) { - for (MachineBasicBlock::succ_iterator I = MBB->succ_begin(), - E = MBB->succ_end(); I != E; ++I) - if (*I != Succ) - return *I; - llvm_unreachable("Expecting a BB with two successors!"); -} +MachineBasicBlock * +ARMTargetLowering::EmitAtomicBinary64(MachineInstr *MI, MachineBasicBlock *BB, + unsigned Op1, unsigned Op2, + bool NeedsCarry, bool IsCmpxchg) const { + // This also handles ATOMIC_SWAP, indicated by Op1==0. + const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); -// FIXME: This opcode table should obviously be expressed in the target -// description. We probably just need a "machine opcode" value in the pseudo -// instruction. But the ideal solution maybe to simply remove the "S" version -// of the opcode altogether. -struct AddSubFlagsOpcodePair { - unsigned PseudoOpc; - unsigned MachineOpc; -}; + const BasicBlock *LLVM_BB = BB->getBasicBlock(); + MachineFunction *MF = BB->getParent(); + MachineFunction::iterator It = BB; + ++It; -static AddSubFlagsOpcodePair AddSubFlagsOpcodeMap[] = { - {ARM::ADCSri, ARM::ADCri}, - {ARM::ADCSrr, ARM::ADCrr}, - {ARM::ADCSrs, ARM::ADCrs}, - {ARM::SBCSri, ARM::SBCri}, - {ARM::SBCSrr, ARM::SBCrr}, - {ARM::SBCSrs, ARM::SBCrs}, - {ARM::RSBSri, ARM::RSBri}, - {ARM::RSBSrr, ARM::RSBrr}, - {ARM::RSBSrs, ARM::RSBrs}, - {ARM::RSCSri, ARM::RSCri}, - {ARM::RSCSrs, ARM::RSCrs}, - {ARM::t2ADCSri, ARM::t2ADCri}, - {ARM::t2ADCSrr, ARM::t2ADCrr}, - {ARM::t2ADCSrs, ARM::t2ADCrs}, - {ARM::t2SBCSri, ARM::t2SBCri}, - {ARM::t2SBCSrr, ARM::t2SBCrr}, - {ARM::t2SBCSrs, ARM::t2SBCrs}, - {ARM::t2RSBSri, ARM::t2RSBri}, - {ARM::t2RSBSrs, ARM::t2RSBrs}, -}; + unsigned destlo = MI->getOperand(0).getReg(); + unsigned desthi = MI->getOperand(1).getReg(); + unsigned ptr = MI->getOperand(2).getReg(); + unsigned vallo = MI->getOperand(3).getReg(); + unsigned valhi = MI->getOperand(4).getReg(); + DebugLoc dl = MI->getDebugLoc(); + bool isThumb2 = Subtarget->isThumb2(); -// Convert and Add or Subtract with Carry and Flags to a generic opcode with -// CPSR operand. e.g. ADCS (...) -> ADC (... CPSR). -// -// FIXME: Somewhere we should assert that CPSR is in the correct -// position to be recognized by the target descrition as the 'S' bit. -bool ARMTargetLowering::RemapAddSubWithFlags(MachineInstr *MI, - MachineBasicBlock *BB) const { - unsigned OldOpc = MI->getOpcode(); - unsigned NewOpc = 0; + MachineRegisterInfo &MRI = BB->getParent()->getRegInfo(); + if (isThumb2) { + MRI.constrainRegClass(destlo, ARM::rGPRRegisterClass); + MRI.constrainRegClass(desthi, ARM::rGPRRegisterClass); + MRI.constrainRegClass(ptr, ARM::rGPRRegisterClass); + } - // This is only called for instructions that need remapping, so iterating over - // the tiny opcode table is not costly. - static const int NPairs = - sizeof(AddSubFlagsOpcodeMap) / sizeof(AddSubFlagsOpcodePair); - for (AddSubFlagsOpcodePair *Pair = &AddSubFlagsOpcodeMap[0], - *End = &AddSubFlagsOpcodeMap[NPairs]; Pair != End; ++Pair) { - if (OldOpc == Pair->PseudoOpc) { - NewOpc = Pair->MachineOpc; - break; + unsigned ldrOpc = isThumb2 ? ARM::t2LDREXD : ARM::LDREXD; + unsigned strOpc = isThumb2 ? ARM::t2STREXD : ARM::STREXD; + + MachineBasicBlock *loopMBB = MF->CreateMachineBasicBlock(LLVM_BB); + MachineBasicBlock *contBB = 0, *cont2BB = 0; + if (IsCmpxchg) { + contBB = MF->CreateMachineBasicBlock(LLVM_BB); + cont2BB = MF->CreateMachineBasicBlock(LLVM_BB); + } + MachineBasicBlock *exitMBB = MF->CreateMachineBasicBlock(LLVM_BB); + MF->insert(It, loopMBB); + if (IsCmpxchg) { + MF->insert(It, contBB); + MF->insert(It, cont2BB); + } + MF->insert(It, exitMBB); + + // Transfer the remainder of BB and its successor edges to exitMBB. + exitMBB->splice(exitMBB->begin(), BB, + llvm::next(MachineBasicBlock::iterator(MI)), + BB->end()); + exitMBB->transferSuccessorsAndUpdatePHIs(BB); + + TargetRegisterClass *TRC = + isThumb2 ? ARM::tGPRRegisterClass : ARM::GPRRegisterClass; + unsigned storesuccess = MRI.createVirtualRegister(TRC); + + // thisMBB: + // ... + // fallthrough --> loopMBB + BB->addSuccessor(loopMBB); + + // loopMBB: + // ldrexd r2, r3, ptr + // r0, r2, incr + // r1, r3, incr + // strexd storesuccess, r0, r1, ptr + // cmp storesuccess, #0 + // bne- loopMBB + // fallthrough --> exitMBB + // + // Note that the registers are explicitly specified because there is not any + // way to force the register allocator to allocate a register pair. + // + // FIXME: The hardcoded registers are not necessary for Thumb2, but we + // need to properly enforce the restriction that the two output registers + // for ldrexd must be different. + BB = loopMBB; + // Load + AddDefaultPred(BuildMI(BB, dl, TII->get(ldrOpc)) + .addReg(ARM::R2, RegState::Define) + .addReg(ARM::R3, RegState::Define).addReg(ptr)); + // Copy r2/r3 into dest. (This copy will normally be coalesced.) + BuildMI(BB, dl, TII->get(TargetOpcode::COPY), destlo).addReg(ARM::R2); + BuildMI(BB, dl, TII->get(TargetOpcode::COPY), desthi).addReg(ARM::R3); + + if (IsCmpxchg) { + // Add early exit + for (unsigned i = 0; i < 2; i++) { + AddDefaultPred(BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2CMPrr : + ARM::CMPrr)) + .addReg(i == 0 ? destlo : desthi) + .addReg(i == 0 ? vallo : valhi)); + BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc)) + .addMBB(exitMBB).addImm(ARMCC::NE).addReg(ARM::CPSR); + BB->addSuccessor(exitMBB); + BB->addSuccessor(i == 0 ? contBB : cont2BB); + BB = (i == 0 ? contBB : cont2BB); } + + // Copy to physregs for strexd + unsigned setlo = MI->getOperand(5).getReg(); + unsigned sethi = MI->getOperand(6).getReg(); + BuildMI(BB, dl, TII->get(TargetOpcode::COPY), ARM::R0).addReg(setlo); + BuildMI(BB, dl, TII->get(TargetOpcode::COPY), ARM::R1).addReg(sethi); + } else if (Op1) { + // Perform binary operation + AddDefaultPred(BuildMI(BB, dl, TII->get(Op1), ARM::R0) + .addReg(destlo).addReg(vallo)) + .addReg(NeedsCarry ? ARM::CPSR : 0, getDefRegState(NeedsCarry)); + AddDefaultPred(BuildMI(BB, dl, TII->get(Op2), ARM::R1) + .addReg(desthi).addReg(valhi)).addReg(0); + } else { + // Copy to physregs for strexd + BuildMI(BB, dl, TII->get(TargetOpcode::COPY), ARM::R0).addReg(vallo); + BuildMI(BB, dl, TII->get(TargetOpcode::COPY), ARM::R1).addReg(valhi); } - if (!NewOpc) - return false; + // Store + AddDefaultPred(BuildMI(BB, dl, TII->get(strOpc), storesuccess) + .addReg(ARM::R0).addReg(ARM::R1).addReg(ptr)); + // Cmp+jump + AddDefaultPred(BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2CMPri : ARM::CMPri)) + .addReg(storesuccess).addImm(0)); + BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc)) + .addMBB(loopMBB).addImm(ARMCC::NE).addReg(ARM::CPSR); + + BB->addSuccessor(loopMBB); + BB->addSuccessor(exitMBB); + + // exitMBB: + // ... + BB = exitMBB; + + MI->eraseFromParent(); // The instruction is gone now. + + return BB; +} + +/// EmitBasePointerRecalculation - For functions using a base pointer, we +/// rematerialize it (via the frame pointer). +void ARMTargetLowering:: +EmitBasePointerRecalculation(MachineInstr *MI, MachineBasicBlock *MBB, + MachineBasicBlock *DispatchBB) const { + const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); + const ARMBaseInstrInfo *AII = static_cast(TII); + MachineFunction &MF = *MI->getParent()->getParent(); + ARMFunctionInfo *AFI = MF.getInfo(); + const ARMBaseRegisterInfo &RI = AII->getRegisterInfo(); + + if (!RI.hasBasePointer(MF)) return; + + MachineBasicBlock::iterator MBBI = MI; + + int32_t NumBytes = AFI->getFramePtrSpillOffset(); + unsigned FramePtr = RI.getFrameRegister(MF); + assert(MF.getTarget().getFrameLowering()->hasFP(MF) && + "Base pointer without frame pointer?"); + + if (AFI->isThumb2Function()) + llvm::emitT2RegPlusImmediate(*MBB, MBBI, MI->getDebugLoc(), ARM::R6, + FramePtr, -NumBytes, ARMCC::AL, 0, *AII); + else if (AFI->isThumbFunction()) + llvm::emitThumbRegPlusImmediate(*MBB, MBBI, MI->getDebugLoc(), ARM::R6, + FramePtr, -NumBytes, *AII, RI); + else + llvm::emitARMRegPlusImmediate(*MBB, MBBI, MI->getDebugLoc(), ARM::R6, + FramePtr, -NumBytes, ARMCC::AL, 0, *AII); + + if (!RI.needsStackRealignment(MF)) return; + + // If there's dynamic realignment, adjust for it. + MachineFrameInfo *MFI = MF.getFrameInfo(); + unsigned MaxAlign = MFI->getMaxAlignment(); + assert(!AFI->isThumb1OnlyFunction()); + + // Emit bic r6, r6, MaxAlign + unsigned bicOpc = AFI->isThumbFunction() ? ARM::t2BICri : ARM::BICri; + AddDefaultCC( + AddDefaultPred( + BuildMI(*MBB, MBBI, MI->getDebugLoc(), TII->get(bicOpc), ARM::R6) + .addReg(ARM::R6, RegState::Kill) + .addImm(MaxAlign - 1))); +} + +/// SetupEntryBlockForSjLj - Insert code into the entry block that creates and +/// registers the function context. +void ARMTargetLowering:: +SetupEntryBlockForSjLj(MachineInstr *MI, MachineBasicBlock *MBB, + MachineBasicBlock *DispatchBB, int FI) const { const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); DebugLoc dl = MI->getDebugLoc(); - MachineInstrBuilder MIB = BuildMI(*BB, MI, dl, TII->get(NewOpc)); - for (unsigned i = 0; i < MI->getNumOperands(); ++i) - MIB.addOperand(MI->getOperand(i)); - AddDefaultPred(MIB); - MIB.addReg(ARM::CPSR, RegState::Define); // S bit + MachineFunction *MF = MBB->getParent(); + MachineRegisterInfo *MRI = &MF->getRegInfo(); + MachineConstantPool *MCP = MF->getConstantPool(); + ARMFunctionInfo *AFI = MF->getInfo(); + const Function *F = MF->getFunction(); + + bool isThumb = Subtarget->isThumb(); + bool isThumb2 = Subtarget->isThumb2(); + + unsigned PCLabelId = AFI->createPICLabelUId(); + unsigned PCAdj = (isThumb || isThumb2) ? 4 : 8; + ARMConstantPoolValue *CPV = + ARMConstantPoolMBB::Create(F->getContext(), DispatchBB, PCLabelId, PCAdj); + unsigned CPI = MCP->getConstantPoolIndex(CPV, 4); + + const TargetRegisterClass *TRC = + isThumb ? ARM::tGPRRegisterClass : ARM::GPRRegisterClass; + + // Grab constant pool and fixed stack memory operands. + MachineMemOperand *CPMMO = + MF->getMachineMemOperand(MachinePointerInfo::getConstantPool(), + MachineMemOperand::MOLoad, 4, 4); + + MachineMemOperand *FIMMOSt = + MF->getMachineMemOperand(MachinePointerInfo::getFixedStack(FI), + MachineMemOperand::MOStore, 4, 4); + + EmitBasePointerRecalculation(MI, MBB, DispatchBB); + + // Load the address of the dispatch MBB into the jump buffer. + if (isThumb2) { + // Incoming value: jbuf + // ldr.n r5, LCPI1_1 + // orr r5, r5, #1 + // add r5, pc + // str r5, [$jbuf, #+4] ; &jbuf[1] + unsigned NewVReg1 = MRI->createVirtualRegister(TRC); + AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::t2LDRpci), NewVReg1) + .addConstantPoolIndex(CPI) + .addMemOperand(CPMMO)); + // Set the low bit because of thumb mode. + unsigned NewVReg2 = MRI->createVirtualRegister(TRC); + AddDefaultCC( + AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::t2ORRri), NewVReg2) + .addReg(NewVReg1, RegState::Kill) + .addImm(0x01))); + unsigned NewVReg3 = MRI->createVirtualRegister(TRC); + BuildMI(*MBB, MI, dl, TII->get(ARM::tPICADD), NewVReg3) + .addReg(NewVReg2, RegState::Kill) + .addImm(PCLabelId); + AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::t2STRi12)) + .addReg(NewVReg3, RegState::Kill) + .addFrameIndex(FI) + .addImm(36) // &jbuf[1] :: pc + .addMemOperand(FIMMOSt)); + } else if (isThumb) { + // Incoming value: jbuf + // ldr.n r1, LCPI1_4 + // add r1, pc + // mov r2, #1 + // orrs r1, r2 + // add r2, $jbuf, #+4 ; &jbuf[1] + // str r1, [r2] + unsigned NewVReg1 = MRI->createVirtualRegister(TRC); + AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::tLDRpci), NewVReg1) + .addConstantPoolIndex(CPI) + .addMemOperand(CPMMO)); + unsigned NewVReg2 = MRI->createVirtualRegister(TRC); + BuildMI(*MBB, MI, dl, TII->get(ARM::tPICADD), NewVReg2) + .addReg(NewVReg1, RegState::Kill) + .addImm(PCLabelId); + // Set the low bit because of thumb mode. + unsigned NewVReg3 = MRI->createVirtualRegister(TRC); + AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::tMOVi8), NewVReg3) + .addReg(ARM::CPSR, RegState::Define) + .addImm(1)); + unsigned NewVReg4 = MRI->createVirtualRegister(TRC); + AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::tORR), NewVReg4) + .addReg(ARM::CPSR, RegState::Define) + .addReg(NewVReg2, RegState::Kill) + .addReg(NewVReg3, RegState::Kill)); + unsigned NewVReg5 = MRI->createVirtualRegister(TRC); + AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::tADDrSPi), NewVReg5) + .addFrameIndex(FI) + .addImm(36)); // &jbuf[1] :: pc + AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::tSTRi)) + .addReg(NewVReg4, RegState::Kill) + .addReg(NewVReg5, RegState::Kill) + .addImm(0) + .addMemOperand(FIMMOSt)); + } else { + // Incoming value: jbuf + // ldr r1, LCPI1_1 + // add r1, pc, r1 + // str r1, [$jbuf, #+4] ; &jbuf[1] + unsigned NewVReg1 = MRI->createVirtualRegister(TRC); + AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::LDRi12), NewVReg1) + .addConstantPoolIndex(CPI) + .addImm(0) + .addMemOperand(CPMMO)); + unsigned NewVReg2 = MRI->createVirtualRegister(TRC); + AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::PICADD), NewVReg2) + .addReg(NewVReg1, RegState::Kill) + .addImm(PCLabelId)); + AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::STRi12)) + .addReg(NewVReg2, RegState::Kill) + .addFrameIndex(FI) + .addImm(36) // &jbuf[1] :: pc + .addMemOperand(FIMMOSt)); + } +} + +MachineBasicBlock *ARMTargetLowering:: +EmitSjLjDispatchBlock(MachineInstr *MI, MachineBasicBlock *MBB) const { + const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); + DebugLoc dl = MI->getDebugLoc(); + MachineFunction *MF = MBB->getParent(); + MachineRegisterInfo *MRI = &MF->getRegInfo(); + ARMFunctionInfo *AFI = MF->getInfo(); + MachineFrameInfo *MFI = MF->getFrameInfo(); + int FI = MFI->getFunctionContextIndex(); + + const TargetRegisterClass *TRC = + Subtarget->isThumb() ? ARM::tGPRRegisterClass : ARM::GPRRegisterClass; + + // Get a mapping of the call site numbers to all of the landing pads they're + // associated with. + DenseMap > CallSiteNumToLPad; + unsigned MaxCSNum = 0; + MachineModuleInfo &MMI = MF->getMMI(); + for (MachineFunction::iterator BB = MF->begin(), E = MF->end(); BB != E; ++BB) { + if (!BB->isLandingPad()) continue; + + // FIXME: We should assert that the EH_LABEL is the first MI in the landing + // pad. + for (MachineBasicBlock::iterator + II = BB->begin(), IE = BB->end(); II != IE; ++II) { + if (!II->isEHLabel()) continue; + + MCSymbol *Sym = II->getOperand(0).getMCSymbol(); + if (!MMI.hasCallSiteLandingPad(Sym)) continue; + + SmallVectorImpl &CallSiteIdxs = MMI.getCallSiteLandingPad(Sym); + for (SmallVectorImpl::iterator + CSI = CallSiteIdxs.begin(), CSE = CallSiteIdxs.end(); + CSI != CSE; ++CSI) { + CallSiteNumToLPad[*CSI].push_back(BB); + MaxCSNum = std::max(MaxCSNum, *CSI); + } + break; + } + } + + // Get an ordered list of the machine basic blocks for the jump table. + std::vector LPadList; + SmallPtrSet InvokeBBs; + LPadList.reserve(CallSiteNumToLPad.size()); + for (unsigned I = 1; I <= MaxCSNum; ++I) { + SmallVectorImpl &MBBList = CallSiteNumToLPad[I]; + for (SmallVectorImpl::iterator + II = MBBList.begin(), IE = MBBList.end(); II != IE; ++II) { + LPadList.push_back(*II); + InvokeBBs.insert((*II)->pred_begin(), (*II)->pred_end()); + } + } + + assert(!LPadList.empty() && + "No landing pad destinations for the dispatch jump table!"); + + // Create the jump table and associated information. + MachineJumpTableInfo *JTI = + MF->getOrCreateJumpTableInfo(MachineJumpTableInfo::EK_Inline); + unsigned MJTI = JTI->createJumpTableIndex(LPadList); + unsigned UId = AFI->createJumpTableUId(); + + // Create the MBBs for the dispatch code. + + // Shove the dispatch's address into the return slot in the function context. + MachineBasicBlock *DispatchBB = MF->CreateMachineBasicBlock(); + DispatchBB->setIsLandingPad(); + + MachineBasicBlock *TrapBB = MF->CreateMachineBasicBlock(); + BuildMI(TrapBB, dl, TII->get(Subtarget->isThumb() ? ARM::tTRAP : ARM::TRAP)); + DispatchBB->addSuccessor(TrapBB); + + MachineBasicBlock *DispContBB = MF->CreateMachineBasicBlock(); + DispatchBB->addSuccessor(DispContBB); + + // Insert and renumber MBBs. + MachineBasicBlock *Last = &MF->back(); + MF->insert(MF->end(), DispatchBB); + MF->insert(MF->end(), DispContBB); + MF->insert(MF->end(), TrapBB); + MF->RenumberBlocks(Last); + + // Insert code into the entry block that creates and registers the function + // context. + SetupEntryBlockForSjLj(MI, MBB, DispatchBB, FI); + + MachineMemOperand *FIMMOLd = + MF->getMachineMemOperand(MachinePointerInfo::getFixedStack(FI), + MachineMemOperand::MOLoad | + MachineMemOperand::MOVolatile, 4, 4); + + if (Subtarget->isThumb2()) { + unsigned NewVReg1 = MRI->createVirtualRegister(TRC); + AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::t2LDRi12), NewVReg1) + .addFrameIndex(FI) + .addImm(4) + .addMemOperand(FIMMOLd)); + AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::t2CMPri)) + .addReg(NewVReg1) + .addImm(LPadList.size())); + BuildMI(DispatchBB, dl, TII->get(ARM::t2Bcc)) + .addMBB(TrapBB) + .addImm(ARMCC::HI) + .addReg(ARM::CPSR); + + unsigned NewVReg2 = MRI->createVirtualRegister(TRC); + AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::t2LEApcrelJT),NewVReg2) + .addJumpTableIndex(MJTI) + .addImm(UId)); + + unsigned NewVReg3 = MRI->createVirtualRegister(TRC); + AddDefaultCC( + AddDefaultPred( + BuildMI(DispContBB, dl, TII->get(ARM::t2ADDrs), NewVReg3) + .addReg(NewVReg2, RegState::Kill) + .addReg(NewVReg1) + .addImm(ARM_AM::getSORegOpc(ARM_AM::lsl, 2)))); + + BuildMI(DispContBB, dl, TII->get(ARM::t2BR_JT)) + .addReg(NewVReg3, RegState::Kill) + .addReg(NewVReg1) + .addJumpTableIndex(MJTI) + .addImm(UId); + } else if (Subtarget->isThumb()) { + unsigned NewVReg1 = MRI->createVirtualRegister(TRC); + AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::tLDRspi), NewVReg1) + .addFrameIndex(FI) + .addImm(1) + .addMemOperand(FIMMOLd)); + + AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::tCMPi8)) + .addReg(NewVReg1) + .addImm(LPadList.size())); + BuildMI(DispatchBB, dl, TII->get(ARM::tBcc)) + .addMBB(TrapBB) + .addImm(ARMCC::HI) + .addReg(ARM::CPSR); + + unsigned NewVReg2 = MRI->createVirtualRegister(TRC); + AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::tLSLri), NewVReg2) + .addReg(ARM::CPSR, RegState::Define) + .addReg(NewVReg1) + .addImm(2)); + + unsigned NewVReg3 = MRI->createVirtualRegister(TRC); + AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::tLEApcrelJT), NewVReg3) + .addJumpTableIndex(MJTI) + .addImm(UId)); + + unsigned NewVReg4 = MRI->createVirtualRegister(TRC); + AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::tADDrr), NewVReg4) + .addReg(ARM::CPSR, RegState::Define) + .addReg(NewVReg2, RegState::Kill) + .addReg(NewVReg3)); + + MachineMemOperand *JTMMOLd = + MF->getMachineMemOperand(MachinePointerInfo::getJumpTable(), + MachineMemOperand::MOLoad, 4, 4); + + unsigned NewVReg5 = MRI->createVirtualRegister(TRC); + AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::tLDRi), NewVReg5) + .addReg(NewVReg4, RegState::Kill) + .addImm(0) + .addMemOperand(JTMMOLd)); + + unsigned NewVReg6 = MRI->createVirtualRegister(TRC); + AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::tADDrr), NewVReg6) + .addReg(ARM::CPSR, RegState::Define) + .addReg(NewVReg5, RegState::Kill) + .addReg(NewVReg3)); + + BuildMI(DispContBB, dl, TII->get(ARM::tBR_JTr)) + .addReg(NewVReg6, RegState::Kill) + .addJumpTableIndex(MJTI) + .addImm(UId); + } else { + unsigned NewVReg1 = MRI->createVirtualRegister(TRC); + AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::LDRi12), NewVReg1) + .addFrameIndex(FI) + .addImm(4) + .addMemOperand(FIMMOLd)); + AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::CMPri)) + .addReg(NewVReg1) + .addImm(LPadList.size())); + BuildMI(DispatchBB, dl, TII->get(ARM::Bcc)) + .addMBB(TrapBB) + .addImm(ARMCC::HI) + .addReg(ARM::CPSR); + + unsigned NewVReg2 = MRI->createVirtualRegister(TRC); + AddDefaultCC( + AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::MOVsi), NewVReg2) + .addReg(NewVReg1) + .addImm(ARM_AM::getSORegOpc(ARM_AM::lsl, 2)))); + unsigned NewVReg3 = MRI->createVirtualRegister(TRC); + AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::LEApcrelJT), NewVReg3) + .addJumpTableIndex(MJTI) + .addImm(UId)); + + MachineMemOperand *JTMMOLd = + MF->getMachineMemOperand(MachinePointerInfo::getJumpTable(), + MachineMemOperand::MOLoad, 4, 4); + unsigned NewVReg4 = MRI->createVirtualRegister(TRC); + AddDefaultPred( + BuildMI(DispContBB, dl, TII->get(ARM::LDRrs), NewVReg4) + .addReg(NewVReg2, RegState::Kill) + .addReg(NewVReg3) + .addImm(0) + .addMemOperand(JTMMOLd)); + + BuildMI(DispContBB, dl, TII->get(ARM::BR_JTadd)) + .addReg(NewVReg4, RegState::Kill) + .addReg(NewVReg3) + .addJumpTableIndex(MJTI) + .addImm(UId); + } + + // Add the jump table entries as successors to the MBB. + MachineBasicBlock *PrevMBB = 0; + for (std::vector::iterator + I = LPadList.begin(), E = LPadList.end(); I != E; ++I) { + MachineBasicBlock *CurMBB = *I; + if (PrevMBB != CurMBB) + DispContBB->addSuccessor(CurMBB); + PrevMBB = CurMBB; + } + + const ARMBaseInstrInfo *AII = static_cast(TII); + const ARMBaseRegisterInfo &RI = AII->getRegisterInfo(); + const unsigned *SavedRegs = RI.getCalleeSavedRegs(MF); + for (SmallPtrSet::iterator + I = InvokeBBs.begin(), E = InvokeBBs.end(); I != E; ++I) { + MachineBasicBlock *BB = *I; + + // Remove the landing pad successor from the invoke block and replace it + // with the new dispatch block. + for (MachineBasicBlock::succ_iterator + SI = BB->succ_begin(), SE = BB->succ_end(); SI != SE; ++SI) { + MachineBasicBlock *SMBB = *SI; + if (SMBB->isLandingPad()) { + BB->removeSuccessor(SMBB); + SMBB->setIsLandingPad(false); + } + } + + BB->addSuccessor(DispatchBB); + + // Find the invoke call and mark all of the callee-saved registers as + // 'implicit defined' so that they're spilled. This prevents code from + // moving instructions to before the EH block, where they will never be + // executed. + for (MachineBasicBlock::reverse_iterator + II = BB->rbegin(), IE = BB->rend(); II != IE; ++II) { + if (!II->getDesc().isCall()) continue; + + DenseMap DefRegs; + for (MachineInstr::mop_iterator + OI = II->operands_begin(), OE = II->operands_end(); + OI != OE; ++OI) { + if (!OI->isReg()) continue; + DefRegs[OI->getReg()] = true; + } + + MachineInstrBuilder MIB(&*II); + + for (unsigned i = 0; SavedRegs[i] != 0; ++i) { + if (!TRC->contains(SavedRegs[i])) continue; + if (!DefRegs[SavedRegs[i]]) + MIB.addReg(SavedRegs[i], RegState::ImplicitDefine | RegState::Dead); + } + + break; + } + } + + // The instruction is gone now. MI->eraseFromParent(); - return true; + + return MBB; +} + +static +MachineBasicBlock *OtherSucc(MachineBasicBlock *MBB, MachineBasicBlock *Succ) { + for (MachineBasicBlock::succ_iterator I = MBB->succ_begin(), + E = MBB->succ_end(); I != E; ++I) + if (*I != Succ) + return *I; + llvm_unreachable("Expecting a BB with two successors!"); } MachineBasicBlock * @@ -5286,12 +5962,61 @@ ARMTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI, bool isThumb2 = Subtarget->isThumb2(); switch (MI->getOpcode()) { default: { - if (RemapAddSubWithFlags(MI, BB)) - return BB; - MI->dump(); llvm_unreachable("Unexpected instr type to insert"); } + // The Thumb2 pre-indexed stores have the same MI operands, they just + // define them differently in the .td files from the isel patterns, so + // they need pseudos. + case ARM::t2STR_preidx: + MI->setDesc(TII->get(ARM::t2STR_PRE)); + return BB; + case ARM::t2STRB_preidx: + MI->setDesc(TII->get(ARM::t2STRB_PRE)); + return BB; + case ARM::t2STRH_preidx: + MI->setDesc(TII->get(ARM::t2STRH_PRE)); + return BB; + + case ARM::STRi_preidx: + case ARM::STRBi_preidx: { + unsigned NewOpc = MI->getOpcode() == ARM::STRi_preidx ? + ARM::STR_PRE_IMM : ARM::STRB_PRE_IMM; + // Decode the offset. + unsigned Offset = MI->getOperand(4).getImm(); + bool isSub = ARM_AM::getAM2Op(Offset) == ARM_AM::sub; + Offset = ARM_AM::getAM2Offset(Offset); + if (isSub) + Offset = -Offset; + + MachineMemOperand *MMO = *MI->memoperands_begin(); + BuildMI(*BB, MI, dl, TII->get(NewOpc)) + .addOperand(MI->getOperand(0)) // Rn_wb + .addOperand(MI->getOperand(1)) // Rt + .addOperand(MI->getOperand(2)) // Rn + .addImm(Offset) // offset (skip GPR==zero_reg) + .addOperand(MI->getOperand(5)) // pred + .addOperand(MI->getOperand(6)) + .addMemOperand(MMO); + MI->eraseFromParent(); + return BB; + } + case ARM::STRr_preidx: + case ARM::STRBr_preidx: + case ARM::STRH_preidx: { + unsigned NewOpc; + switch (MI->getOpcode()) { + default: llvm_unreachable("unexpected opcode!"); + case ARM::STRr_preidx: NewOpc = ARM::STR_PRE_REG; break; + case ARM::STRBr_preidx: NewOpc = ARM::STRB_PRE_REG; break; + case ARM::STRH_preidx: NewOpc = ARM::STRH_PRE; break; + } + MachineInstrBuilder MIB = BuildMI(*BB, MI, dl, TII->get(NewOpc)); + for (unsigned i = 0; i < MI->getNumOperands(); ++i) + MIB.addOperand(MI->getOperand(i)); + MI->eraseFromParent(); + return BB; + } case ARM::ATOMIC_LOAD_ADD_I8: return EmitAtomicBinary(MI, BB, 1, isThumb2 ? ARM::t2ADDrr : ARM::ADDrr); case ARM::ATOMIC_LOAD_ADD_I16: @@ -5370,6 +6095,31 @@ ARMTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI, case ARM::ATOMIC_CMP_SWAP_I16: return EmitAtomicCmpSwap(MI, BB, 2); case ARM::ATOMIC_CMP_SWAP_I32: return EmitAtomicCmpSwap(MI, BB, 4); + + case ARM::ATOMADD6432: + return EmitAtomicBinary64(MI, BB, isThumb2 ? ARM::t2ADDrr : ARM::ADDrr, + isThumb2 ? ARM::t2ADCrr : ARM::ADCrr, + /*NeedsCarry*/ true); + case ARM::ATOMSUB6432: + return EmitAtomicBinary64(MI, BB, isThumb2 ? ARM::t2SUBrr : ARM::SUBrr, + isThumb2 ? ARM::t2SBCrr : ARM::SBCrr, + /*NeedsCarry*/ true); + case ARM::ATOMOR6432: + return EmitAtomicBinary64(MI, BB, isThumb2 ? ARM::t2ORRrr : ARM::ORRrr, + isThumb2 ? ARM::t2ORRrr : ARM::ORRrr); + case ARM::ATOMXOR6432: + return EmitAtomicBinary64(MI, BB, isThumb2 ? ARM::t2EORrr : ARM::EORrr, + isThumb2 ? ARM::t2EORrr : ARM::EORrr); + case ARM::ATOMAND6432: + return EmitAtomicBinary64(MI, BB, isThumb2 ? ARM::t2ANDrr : ARM::ANDrr, + isThumb2 ? ARM::t2ANDrr : ARM::ANDrr); + case ARM::ATOMSWAP6432: + return EmitAtomicBinary64(MI, BB, 0, 0, false); + case ARM::ATOMCMPXCHG6432: + return EmitAtomicBinary64(MI, BB, isThumb2 ? ARM::t2SUBrr : ARM::SUBrr, + isThumb2 ? ARM::t2SBCrr : ARM::SBCrr, + /*NeedsCarry*/ false, /*IsCmpxchg*/true); + case ARM::tMOVCCr_pseudo: { // To "insert" a SELECT_CC instruction, we actually have to insert the // diamond control-flow pattern. The incoming instruction knows the @@ -5461,13 +6211,159 @@ ARMTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI, BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc)) .addMBB(destMBB).addImm(ARMCC::EQ).addReg(ARM::CPSR); - BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2B : ARM::B)) - .addMBB(exitMBB); + if (isThumb2) + AddDefaultPred(BuildMI(BB, dl, TII->get(ARM::t2B)).addMBB(exitMBB)); + else + BuildMI(BB, dl, TII->get(ARM::B)) .addMBB(exitMBB); MI->eraseFromParent(); // The pseudo instruction is gone now. return BB; } + + case ARM::ABS: + case ARM::t2ABS: { + // To insert an ABS instruction, we have to insert the + // diamond control-flow pattern. The incoming instruction knows the + // source vreg to test against 0, the destination vreg to set, + // the condition code register to branch on, the + // true/false values to select between, and a branch opcode to use. + // It transforms + // V1 = ABS V0 + // into + // V2 = MOVS V0 + // BCC (branch to SinkBB if V0 >= 0) + // RSBBB: V3 = RSBri V2, 0 (compute ABS if V2 < 0) + // SinkBB: V1 = PHI(V2, V3) + const BasicBlock *LLVM_BB = BB->getBasicBlock(); + MachineFunction::iterator BBI = BB; + ++BBI; + MachineFunction *Fn = BB->getParent(); + MachineBasicBlock *RSBBB = Fn->CreateMachineBasicBlock(LLVM_BB); + MachineBasicBlock *SinkBB = Fn->CreateMachineBasicBlock(LLVM_BB); + Fn->insert(BBI, RSBBB); + Fn->insert(BBI, SinkBB); + + unsigned int ABSSrcReg = MI->getOperand(1).getReg(); + unsigned int ABSDstReg = MI->getOperand(0).getReg(); + bool isThumb2 = Subtarget->isThumb2(); + MachineRegisterInfo &MRI = Fn->getRegInfo(); + // In Thumb mode S must not be specified if source register is the SP or + // PC and if destination register is the SP, so restrict register class + unsigned NewMovDstReg = MRI.createVirtualRegister( + isThumb2 ? ARM::rGPRRegisterClass : ARM::GPRRegisterClass); + unsigned NewRsbDstReg = MRI.createVirtualRegister( + isThumb2 ? ARM::rGPRRegisterClass : ARM::GPRRegisterClass); + + // Transfer the remainder of BB and its successor edges to sinkMBB. + SinkBB->splice(SinkBB->begin(), BB, + llvm::next(MachineBasicBlock::iterator(MI)), + BB->end()); + SinkBB->transferSuccessorsAndUpdatePHIs(BB); + + BB->addSuccessor(RSBBB); + BB->addSuccessor(SinkBB); + + // fall through to SinkMBB + RSBBB->addSuccessor(SinkBB); + + // insert a movs at the end of BB + BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2MOVr : ARM::MOVr), + NewMovDstReg) + .addReg(ABSSrcReg, RegState::Kill) + .addImm((unsigned)ARMCC::AL).addReg(0) + .addReg(ARM::CPSR, RegState::Define); + + // insert a bcc with opposite CC to ARMCC::MI at the end of BB + BuildMI(BB, dl, + TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc)).addMBB(SinkBB) + .addImm(ARMCC::getOppositeCondition(ARMCC::MI)).addReg(ARM::CPSR); + + // insert rsbri in RSBBB + // Note: BCC and rsbri will be converted into predicated rsbmi + // by if-conversion pass + BuildMI(*RSBBB, RSBBB->begin(), dl, + TII->get(isThumb2 ? ARM::t2RSBri : ARM::RSBri), NewRsbDstReg) + .addReg(NewMovDstReg, RegState::Kill) + .addImm(0).addImm((unsigned)ARMCC::AL).addReg(0).addReg(0); + + // insert PHI in SinkBB, + // reuse ABSDstReg to not change uses of ABS instruction + BuildMI(*SinkBB, SinkBB->begin(), dl, + TII->get(ARM::PHI), ABSDstReg) + .addReg(NewRsbDstReg).addMBB(RSBBB) + .addReg(NewMovDstReg).addMBB(BB); + + // remove ABS instruction + MI->eraseFromParent(); + + // return last added BB + return SinkBB; + } + } +} + +void ARMTargetLowering::AdjustInstrPostInstrSelection(MachineInstr *MI, + SDNode *Node) const { + const MCInstrDesc &MCID = MI->getDesc(); + if (!MCID.hasPostISelHook()) { + assert(!convertAddSubFlagsOpcode(MI->getOpcode()) && + "Pseudo flag-setting opcodes must be marked with 'hasPostISelHook'"); + return; + } + + // Adjust potentially 's' setting instructions after isel, i.e. ADC, SBC, RSB, + // RSC. Coming out of isel, they have an implicit CPSR def, but the optional + // operand is still set to noreg. If needed, set the optional operand's + // register to CPSR, and remove the redundant implicit def. + // + // e.g. ADCS (...opt:%noreg, CPSR) -> ADC (... opt:CPSR). + + // Rename pseudo opcodes. + unsigned NewOpc = convertAddSubFlagsOpcode(MI->getOpcode()); + if (NewOpc) { + const ARMBaseInstrInfo *TII = + static_cast(getTargetMachine().getInstrInfo()); + MI->setDesc(TII->get(NewOpc)); + } + unsigned ccOutIdx = MCID.getNumOperands() - 1; + + // Any ARM instruction that sets the 's' bit should specify an optional + // "cc_out" operand in the last operand position. + if (!MCID.hasOptionalDef() || !MCID.OpInfo[ccOutIdx].isOptionalDef()) { + assert(!NewOpc && "Optional cc_out operand required"); + return; + } + // Look for an implicit def of CPSR added by MachineInstr ctor. Remove it + // since we already have an optional CPSR def. + bool definesCPSR = false; + bool deadCPSR = false; + for (unsigned i = MCID.getNumOperands(), e = MI->getNumOperands(); + i != e; ++i) { + const MachineOperand &MO = MI->getOperand(i); + if (MO.isReg() && MO.isDef() && MO.getReg() == ARM::CPSR) { + definesCPSR = true; + if (MO.isDead()) + deadCPSR = true; + MI->RemoveOperand(i); + break; + } + } + if (!definesCPSR) { + assert(!NewOpc && "Optional cc_out operand required"); + return; + } + assert(deadCPSR == !Node->hasAnyUseOfValue(1) && "inconsistent dead flag"); + if (deadCPSR) { + assert(!MI->getOperand(ccOutIdx).getReg() && + "expect uninitialized optional cc_out operand"); + return; } + + // If this instruction was defined with an optional CPSR def and its dag node + // had a live implicit CPSR def, then activate the optional CPSR def. + MachineOperand &MO = MI->getOperand(ccOutIdx); + MO.setReg(ARM::CPSR); + MO.setIsDef(true); } //===----------------------------------------------------------------------===// @@ -6975,7 +7871,8 @@ ARMTargetLowering::PerformCMOVCombine(SDNode *N, SelectionDAG &DAG) const { SDValue FalseVal = N->getOperand(0); SDValue TrueVal = N->getOperand(1); SDValue ARMcc = N->getOperand(2); - ARMCC::CondCodes CC = (ARMCC::CondCodes)cast(ARMcc)->getZExtValue(); + ARMCC::CondCodes CC = + (ARMCC::CondCodes)cast(ARMcc)->getZExtValue(); // Simplify // mov r1, r0 @@ -6995,7 +7892,7 @@ ARMTargetLowering::PerformCMOVCombine(SDNode *N, SelectionDAG &DAG) const { // movne r0, y /// FIXME: Turn this into a target neutral optimization? SDValue Res; - if (CC == ARMCC::NE && FalseVal == RHS) { + if (CC == ARMCC::NE && FalseVal == RHS && FalseVal != LHS) { Res = DAG.getNode(ARMISD::CMOV, dl, VT, LHS, TrueVal, ARMcc, N->getOperand(3), Cmp); } else if (CC == ARMCC::EQ && TrueVal == RHS) { @@ -7235,7 +8132,7 @@ bool ARMTargetLowering::isLegalT2ScaledAddressingMode(const AddrMode &AM, /// isLegalAddressingMode - Return true if the addressing mode represented /// by AM is legal for this target, for a load/store of the specified type. bool ARMTargetLowering::isLegalAddressingMode(const AddrMode &AM, - const Type *Ty) const { + Type *Ty) const { EVT VT = getValueType(Ty, true); if (!isLegalAddressImmediate(AM.BaseOffs, VT, Subtarget)) return false; @@ -7351,7 +8248,8 @@ static bool getARMIndexedAddressParts(SDNode *Ptr, EVT VT, if (Ptr->getOpcode() == ISD::ADD) { isInc = true; - ARM_AM::ShiftOpc ShOpcVal= ARM_AM::getShiftOpcForNode(Ptr->getOperand(0)); + ARM_AM::ShiftOpc ShOpcVal= + ARM_AM::getShiftOpcForNode(Ptr->getOperand(0).getOpcode()); if (ShOpcVal != ARM_AM::no_shift) { Base = Ptr->getOperand(1); Offset = Ptr->getOperand(0); @@ -7536,7 +8434,7 @@ bool ARMTargetLowering::ExpandInlineAsm(CallInst *CI) const { if (AsmPieces.size() == 3 && AsmPieces[0] == "rev" && AsmPieces[1] == "$0" && AsmPieces[2] == "$1" && IA->getConstraintString().compare(0, 4, "=l,l") == 0) { - const IntegerType *Ty = dyn_cast(CI->getType()); + IntegerType *Ty = dyn_cast(CI->getType()); if (Ty && Ty->getBitWidth() == 32) return IntrinsicLowering::LowerToByteSwap(CI); } @@ -7559,6 +8457,9 @@ ARMTargetLowering::getConstraintType(const std::string &Constraint) const { case 'x': return C_RegisterClass; case 't': return C_RegisterClass; case 'j': return C_Other; // Constant for movw. + // An address with a single base register. Due to the way we + // currently handle addresses it is the same as an 'r' memory constraint. + case 'Q': return C_Memory; } } else if (Constraint.size() == 2) { switch (Constraint[0]) { @@ -7582,7 +8483,7 @@ ARMTargetLowering::getSingleConstraintMatchWeight( // but allow it at the lowest weight. if (CallOperandVal == NULL) return CW_Default; - const Type *type = CallOperandVal->getType(); + Type *type = CallOperandVal->getType(); // Look at the constraint type. switch (*constraint) { default: @@ -7618,7 +8519,7 @@ ARMTargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint, return RCPair(0U, ARM::GPRRegisterClass); case 'h': // High regs or no regs. if (Subtarget->isThumb()) - return RCPair(0U, ARM::hGPRRegisterClass); + return RCPair(0U, ARM::hGPRRegisterClass); break; case 'r': return RCPair(0U, ARM::GPRRegisterClass); @@ -7632,15 +8533,15 @@ ARMTargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint, break; case 'x': if (VT == MVT::f32) - return RCPair(0U, ARM::SPR_8RegisterClass); + return RCPair(0U, ARM::SPR_8RegisterClass); if (VT.getSizeInBits() == 64) - return RCPair(0U, ARM::DPR_8RegisterClass); + return RCPair(0U, ARM::DPR_8RegisterClass); if (VT.getSizeInBits() == 128) - return RCPair(0U, ARM::QPR_8RegisterClass); + return RCPair(0U, ARM::QPR_8RegisterClass); break; case 't': if (VT == MVT::f32) - return RCPair(0U, ARM::SPRRegisterClass); + return RCPair(0U, ARM::SPRRegisterClass); break; } } @@ -7680,12 +8581,12 @@ void ARMTargetLowering::LowerAsmOperandForConstraint(SDValue Op, switch (ConstraintLetter) { case 'j': - // Constant suitable for movw, must be between 0 and - // 65535. - if (Subtarget->hasV6T2Ops()) - if (CVal >= 0 && CVal <= 65535) - break; - return; + // Constant suitable for movw, must be between 0 and + // 65535. + if (Subtarget->hasV6T2Ops()) + if (CVal >= 0 && CVal <= 65535) + break; + return; case 'I': if (Subtarget->isThumb1Only()) { // This must be a constant between 0 and 255, for ADD @@ -7823,50 +8724,6 @@ ARMTargetLowering::isOffsetFoldingLegal(const GlobalAddressSDNode *GA) const { return false; } -int ARM::getVFPf32Imm(const APFloat &FPImm) { - APInt Imm = FPImm.bitcastToAPInt(); - uint32_t Sign = Imm.lshr(31).getZExtValue() & 1; - int32_t Exp = (Imm.lshr(23).getSExtValue() & 0xff) - 127; // -126 to 127 - int64_t Mantissa = Imm.getZExtValue() & 0x7fffff; // 23 bits - - // We can handle 4 bits of mantissa. - // mantissa = (16+UInt(e:f:g:h))/16. - if (Mantissa & 0x7ffff) - return -1; - Mantissa >>= 19; - if ((Mantissa & 0xf) != Mantissa) - return -1; - - // We can handle 3 bits of exponent: exp == UInt(NOT(b):c:d)-3 - if (Exp < -3 || Exp > 4) - return -1; - Exp = ((Exp+3) & 0x7) ^ 4; - - return ((int)Sign << 7) | (Exp << 4) | Mantissa; -} - -int ARM::getVFPf64Imm(const APFloat &FPImm) { - APInt Imm = FPImm.bitcastToAPInt(); - uint64_t Sign = Imm.lshr(63).getZExtValue() & 1; - int64_t Exp = (Imm.lshr(52).getSExtValue() & 0x7ff) - 1023; // -1022 to 1023 - uint64_t Mantissa = Imm.getZExtValue() & 0xfffffffffffffLL; - - // We can handle 4 bits of mantissa. - // mantissa = (16+UInt(e:f:g:h))/16. - if (Mantissa & 0xffffffffffffLL) - return -1; - Mantissa >>= 48; - if ((Mantissa & 0xf) != Mantissa) - return -1; - - // We can handle 3 bits of exponent: exp == UInt(NOT(b):c:d)-3 - if (Exp < -3 || Exp > 4) - return -1; - Exp = ((Exp+3) & 0x7) ^ 4; - - return ((int)Sign << 7) | (Exp << 4) | Mantissa; -} - bool ARM::isBitFieldInvertedMask(unsigned v) { if (v == 0xffffffff) return 0; @@ -7889,9 +8746,9 @@ bool ARMTargetLowering::isFPImmLegal(const APFloat &Imm, EVT VT) const { if (!Subtarget->hasVFP3()) return false; if (VT == MVT::f32) - return ARM::getVFPf32Imm(Imm) != -1; + return ARM_AM::getFP32Imm(Imm) != -1; if (VT == MVT::f64) - return ARM::getVFPf64Imm(Imm) != -1; + return ARM_AM::getFP64Imm(Imm) != -1; return false; } @@ -7933,7 +8790,7 @@ bool ARMTargetLowering::getTgtMemIntrinsic(IntrinsicInfo &Info, // Conservatively set memVT to the entire set of vectors stored. unsigned NumElts = 0; for (unsigned ArgI = 1, ArgE = I.getNumArgOperands(); ArgI < ArgE; ++ArgI) { - const Type *ArgTy = I.getArgOperand(ArgI)->getType(); + Type *ArgTy = I.getArgOperand(ArgI)->getType(); if (!ArgTy->isVectorTy()) break; NumElts += getTargetData()->getTypeAllocSize(ArgTy) / 8; diff --git a/lib/Target/ARM/ARMISelLowering.h b/lib/Target/ARM/ARMISelLowering.h index 980fb404887e0..5da9b27fca68e 100644 --- a/lib/Target/ARM/ARMISelLowering.h +++ b/lib/Target/ARM/ARMISelLowering.h @@ -71,6 +71,11 @@ namespace llvm { SRA_FLAG, // V,Flag = sra_flag X -> sra X, 1 + save carry out. RRX, // V = RRX X, Flag -> srl X, 1 + shift in carry flag. + ADDC, // Add with carry + ADDE, // Add using carry + SUBC, // Sub with carry + SUBE, // Sub using carry + VMOVRRD, // double to two gprs. VMOVDRR, // Two gprs to double. @@ -206,18 +211,22 @@ namespace llvm { VST4_UPD, VST2LN_UPD, VST3LN_UPD, - VST4LN_UPD + VST4LN_UPD, + + // 64-bit atomic ops (value split into two registers) + ATOMADD64_DAG, + ATOMSUB64_DAG, + ATOMOR64_DAG, + ATOMXOR64_DAG, + ATOMAND64_DAG, + ATOMNAND64_DAG, + ATOMSWAP64_DAG, + ATOMCMPXCHG64_DAG }; } /// Define some predicates that are used for node matching. namespace ARM { - /// getVFPf32Imm / getVFPf64Imm - If the given fp immediate can be - /// materialized with a VMOV.f32 / VMOV.f64 (i.e. fconsts / fconstd) - /// instruction, returns its 8-bit integer representation. Otherwise, - /// returns -1. - int getVFPf32Imm(const APFloat &FPImm); - int getVFPf64Imm(const APFloat &FPImm); bool isBitFieldInvertedMask(unsigned v); } @@ -240,10 +249,16 @@ namespace llvm { virtual const char *getTargetNodeName(unsigned Opcode) const; + /// getSetCCResultType - Return the value type to use for ISD::SETCC. + virtual EVT getSetCCResultType(EVT VT) const; + virtual MachineBasicBlock * EmitInstrWithCustomInserter(MachineInstr *MI, MachineBasicBlock *MBB) const; + virtual void + AdjustInstrPostInstrSelection(MachineInstr *MI, SDNode *Node) const; + SDValue PerformCMOVCombine(SDNode *N, SelectionDAG &DAG) const; virtual SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const; @@ -256,7 +271,7 @@ namespace llvm { /// isLegalAddressingMode - Return true if the addressing mode represented /// by AM is legal for this target, for a load/store of the specified type. - virtual bool isLegalAddressingMode(const AddrMode &AM, const Type *Ty)const; + virtual bool isLegalAddressingMode(const AddrMode &AM, Type *Ty)const; bool isLegalT2ScaledAddressingMode(const AddrMode &AM, EVT VT) const; /// isLegalICmpImmediate - Return true if the specified immediate is legal @@ -485,12 +500,28 @@ namespace llvm { MachineBasicBlock *BB, unsigned Size, unsigned BinOpcode) const; + MachineBasicBlock *EmitAtomicBinary64(MachineInstr *MI, + MachineBasicBlock *BB, + unsigned Op1, + unsigned Op2, + bool NeedsCarry = false, + bool IsCmpxchg = false) const; MachineBasicBlock * EmitAtomicBinaryMinMax(MachineInstr *MI, MachineBasicBlock *BB, unsigned Size, bool signExtend, ARMCC::CondCodes Cond) const; + void EmitBasePointerRecalculation(MachineInstr *MI, MachineBasicBlock *MBB, + MachineBasicBlock *DispatchBB) const; + + void SetupEntryBlockForSjLj(MachineInstr *MI, + MachineBasicBlock *MBB, + MachineBasicBlock *DispatchBB, int FI) const; + + MachineBasicBlock *EmitSjLjDispatchBlock(MachineInstr *MI, + MachineBasicBlock *MBB) const; + bool RemapAddSubWithFlags(MachineInstr *MI, MachineBasicBlock *BB) const; }; diff --git a/lib/Target/ARM/ARMInstrFormats.td b/lib/Target/ARM/ARMInstrFormats.td index 3ccf22f80b7da..7cbc9111dec37 100644 --- a/lib/Target/ARM/ARMInstrFormats.td +++ b/lib/Target/ARM/ARMInstrFormats.td @@ -25,7 +25,7 @@ def BrFrm : Format<2>; def BrMiscFrm : Format<3>; def DPFrm : Format<4>; -def DPSoRegFrm : Format<5>; +def DPSoRegRegFrm : Format<5>; def LdFrm : Format<6>; def StFrm : Format<7>; @@ -68,6 +68,7 @@ def N3RegVShFrm : Format<38>; def NVExtFrm : Format<39>; def NVMulSLFrm : Format<40>; def NVTBLFrm : Format<41>; +def DPSoRegImmFrm : Format<42>; // Misc flags. @@ -130,39 +131,15 @@ def VFPNeonA8Domain : Domain<5>; // Instructions in VFP & Neon under A8 // ARM special operands. // -def CondCodeOperand : AsmOperandClass { - let Name = "CondCode"; - let SuperClasses = []; -} - -def CCOutOperand : AsmOperandClass { - let Name = "CCOut"; - let SuperClasses = []; -} - -def MemBarrierOptOperand : AsmOperandClass { - let Name = "MemBarrierOpt"; - let SuperClasses = []; - let ParserMethod = "tryParseMemBarrierOptOperand"; -} - -def ProcIFlagsOperand : AsmOperandClass { - let Name = "ProcIFlags"; - let SuperClasses = []; - let ParserMethod = "tryParseProcIFlagsOperand"; -} - -def MSRMaskOperand : AsmOperandClass { - let Name = "MSRMask"; - let SuperClasses = []; - let ParserMethod = "tryParseMSRMaskOperand"; -} - // ARM imod and iflag operands, used only by the CPS instruction. def imod_op : Operand { let PrintMethod = "printCPSIMod"; } +def ProcIFlagsOperand : AsmOperandClass { + let Name = "ProcIFlags"; + let ParserMethod = "parseProcIFlagsOperand"; +} def iflags_op : Operand { let PrintMethod = "printCPSIFlag"; let ParserMatchClass = ProcIFlagsOperand; @@ -170,17 +147,21 @@ def iflags_op : Operand { // ARM Predicate operand. Default to 14 = always (AL). Second part is CC // register whose default is 0 (no register). -def pred : PredicateOperand { let PrintMethod = "printPredicateOperand"; let ParserMatchClass = CondCodeOperand; + let DecoderMethod = "DecodePredicateOperand"; } // Conditional code result for instructions whose 's' bit is set, e.g. subs. +def CCOutOperand : AsmOperandClass { let Name = "CCOut"; } def cc_out : OptionalDefOperand { let EncoderMethod = "getCCOutOpValue"; let PrintMethod = "printSBitModifierOperand"; let ParserMatchClass = CCOutOperand; + let DecoderMethod = "DecodeCCOutOperand"; } // Same as cc_out except it defaults to setting CPSR. @@ -188,16 +169,27 @@ def s_cc_out : OptionalDefOperand { let EncoderMethod = "getCCOutOpValue"; let PrintMethod = "printSBitModifierOperand"; let ParserMatchClass = CCOutOperand; + let DecoderMethod = "DecodeCCOutOperand"; } // ARM special operands for disassembly only. // +def SetEndAsmOperand : AsmOperandClass { + let Name = "SetEndImm"; + let ParserMethod = "parseSetEndImm"; +} def setend_op : Operand { let PrintMethod = "printSetendOperand"; + let ParserMatchClass = SetEndAsmOperand; } +def MSRMaskOperand : AsmOperandClass { + let Name = "MSRMask"; + let ParserMethod = "parseMSRMaskOperand"; +} def msr_mask : Operand { let PrintMethod = "printMSRMaskOperand"; + let DecoderMethod = "DecodeMSRMask"; let ParserMatchClass = MSRMaskOperand; } @@ -211,21 +203,40 @@ def msr_mask : Operand { // 64 64 - is encoded in imm6<5:0> def shr_imm8 : Operand { let EncoderMethod = "getShiftRight8Imm"; + let DecoderMethod = "DecodeShiftRight8Imm"; } def shr_imm16 : Operand { let EncoderMethod = "getShiftRight16Imm"; + let DecoderMethod = "DecodeShiftRight16Imm"; } def shr_imm32 : Operand { let EncoderMethod = "getShiftRight32Imm"; + let DecoderMethod = "DecodeShiftRight32Imm"; } def shr_imm64 : Operand { let EncoderMethod = "getShiftRight64Imm"; + let DecoderMethod = "DecodeShiftRight64Imm"; } +//===----------------------------------------------------------------------===// +// ARM Assembler alias templates. +// +class ARMInstAlias + : InstAlias, Requires<[IsARM]>; +class tInstAlias + : InstAlias, Requires<[IsThumb]>; +class t2InstAlias + : InstAlias, Requires<[IsThumb2]>; +class VFP2InstAlias + : InstAlias, Requires<[HasVFP2]>; +class VFP3InstAlias + : InstAlias, Requires<[HasVFP3]>; + //===----------------------------------------------------------------------===// // ARM Instruction templates. // + class InstTemplate : Instruction { @@ -240,17 +251,22 @@ class InstTemplate(f), "Pseudo"); - // The layout of TSFlags should be kept in sync with ARMBaseInstrInfo.h. + // The layout of TSFlags should be kept in sync with ARMBaseInfo.h. let TSFlags{4-0} = AM.Value; let TSFlags{6-5} = IndexModeBits; let TSFlags{12-7} = Form; let TSFlags{13} = isUnaryDataProc; let TSFlags{14} = canXformTo16Bit; let TSFlags{17-15} = D.Value; + let TSFlags{18} = thumbArithFlagSetting; let Constraints = cstr; let Itinerary = itin; @@ -262,13 +278,17 @@ class Encoding { class InstARM - : InstTemplate, Encoding; + : InstTemplate, Encoding { + let DecoderNamespace = "ARM"; +} // This Encoding-less class is used by Thumb1 to specify the encoding bits later // on by adding flavors to specific instructions. class InstThumb - : InstTemplate; + : InstTemplate { + let DecoderNamespace = "Thumb"; +} class PseudoInst pattern> : InstTemplate opcod, dag oops, dag iops, InstrItinClass itin, : I { bits<4> Rt; - bits<4> Rn; + bits<4> addr; let Inst{27-23} = 0b00011; let Inst{22-21} = opcod; let Inst{20} = 1; - let Inst{19-16} = Rn; + let Inst{19-16} = addr; let Inst{15-12} = Rt; let Inst{11-0} = 0b111110011111; } @@ -450,14 +470,14 @@ class AIstrex opcod, dag oops, dag iops, InstrItinClass itin, let Inst{3-0} = Rt; } class AIswp pattern> - : AI { + : AI { bits<4> Rt; bits<4> Rt2; - bits<4> Rn; + bits<4> addr; let Inst{27-23} = 0b00010; let Inst{22} = b; let Inst{21-20} = 0b00; - let Inst{19-16} = Rn; + let Inst{19-16} = addr; let Inst{15-12} = Rt; let Inst{11-4} = 0b00001001; let Inst{3-0} = Rt2; @@ -515,22 +535,41 @@ class AI2ldstidx pattern> + : AI2ldstidx<0, isByte, isPre, oops, iops, im, f, itin, opc, asm, cstr, + pattern> { + // AM2 store w/ two operands: (GPR, am2offset) + // {12} isAdd + // {11-0} imm12/Rm + bits<14> offset; + bits<4> Rn; + let Inst{25} = 1; + let Inst{23} = offset{12}; + let Inst{19-16} = Rn; + let Inst{11-5} = offset{11-5}; + let Inst{4} = 0; + let Inst{3-0} = offset{3-0}; +} + +class AI2stridx_imm pattern> : AI2ldstidx<0, isByte, isPre, oops, iops, im, f, itin, opc, asm, cstr, pattern> { // AM2 store w/ two operands: (GPR, am2offset) - // {13} 1 == Rm, 0 == imm12 // {12} isAdd // {11-0} imm12/Rm bits<14> offset; bits<4> Rn; - let Inst{25} = offset{13}; + let Inst{25} = 0; let Inst{23} = offset{12}; let Inst{19-16} = Rn; let Inst{11-0} = offset{11-0}; } + + // FIXME: Merge with the above class when addrmode2 gets used for STR, STRB // but for now use this class for STRT and STRBT. class AI2stridxT op, bit op20, dag oops, dag iops, Format f, let Inst{11-8} = addr{7-4}; // imm7_4/zero let Inst{7-4} = op; let Inst{3-0} = addr{3-0}; // imm3_0/Rm + + let DecoderMethod = "DecodeAddrMode3Instruction"; } -class AI3ldstidx op, bit op20, bit isLd, bit isPre, dag oops, dag iops, +class AI3ldstidx op, bit op20, bit isPre, dag oops, dag iops, IndexMode im, Format f, InstrItinClass itin, string opc, string asm, string cstr, list pattern> : I op, bit op20, bit isLd, bit isPre, dag oops, dag iops, // FIXME: Merge with the above class when addrmode2 gets used for LDR, LDRB // but for now use this class for LDRSBT, LDRHT, LDSHT. -class AI3ldstidxT op, bit op20, bit isLd, bit isPre, dag oops, dag iops, +class AI3ldstidxT op, bit isLoad, dag oops, dag iops, IndexMode im, Format f, InstrItinClass itin, string opc, string asm, string cstr, list pattern> - : I { + : I { // {13} 1 == imm8, 0 == Rm // {12-9} Rn // {8} isAdd // {7-4} imm7_4/zero // {3-0} imm3_0/Rm - bits<14> addr; - bits<4> Rt; - let Inst{27-25} = 0b000; - let Inst{24} = isPre; // P bit - let Inst{23} = addr{8}; // U bit - let Inst{22} = addr{13}; // 1 == imm8, 0 == Rm - let Inst{20} = op20; // L bit - let Inst{19-16} = addr{12-9}; // Rn - let Inst{15-12} = Rt; // Rt - let Inst{11-8} = addr{7-4}; // imm7_4/zero - let Inst{7-4} = op; - let Inst{3-0} = addr{3-0}; // imm3_0/Rm - let AsmMatchConverter = "CvtLdWriteBackRegAddrMode3"; -} - -class AI3stridx op, bit isByte, bit isPre, dag oops, dag iops, - IndexMode im, Format f, InstrItinClass itin, string opc, - string asm, string cstr, list pattern> - : AI2ldstidx<0, isByte, isPre, oops, iops, im, f, itin, opc, asm, cstr, - pattern> { - // AM3 store w/ two operands: (GPR, am3offset) - bits<14> offset; + bits<4> addr; bits<4> Rt; - bits<4> Rn; let Inst{27-25} = 0b000; - let Inst{23} = offset{8}; - let Inst{22} = offset{9}; - let Inst{19-16} = Rn; + let Inst{24} = 0; // P bit + let Inst{21} = 1; + let Inst{20} = isLoad; // L bit + let Inst{19-16} = addr; // Rn let Inst{15-12} = Rt; // Rt - let Inst{11-8} = offset{7-4}; // imm7_4/zero let Inst{7-4} = op; - let Inst{3-0} = offset{3-0}; // imm3_0/Rm } // stores @@ -648,75 +665,7 @@ class AI3str op, dag oops, dag iops, Format f, InstrItinClass itin, let Inst{11-8} = addr{7-4}; // imm7_4/zero let Inst{7-4} = op; let Inst{3-0} = addr{3-0}; // imm3_0/Rm -} - -// Pre-indexed stores -class AI3sthpr pattern> - : I { - let Inst{4} = 1; - let Inst{5} = 1; // H bit - let Inst{6} = 0; // S bit - let Inst{7} = 1; - let Inst{20} = 0; // L bit - let Inst{21} = 1; // W bit - let Inst{24} = 1; // P bit - let Inst{27-25} = 0b000; -} -class AI3stdpr pattern> - : I { - let Inst{4} = 1; - let Inst{5} = 1; // H bit - let Inst{6} = 1; // S bit - let Inst{7} = 1; - let Inst{20} = 0; // L bit - let Inst{21} = 1; // W bit - let Inst{24} = 1; // P bit - let Inst{27-25} = 0b000; -} - -// Post-indexed stores -class AI3sthpo pattern> - : I { - // {13} 1 == imm8, 0 == Rm - // {12-9} Rn - // {8} isAdd - // {7-4} imm7_4/zero - // {3-0} imm3_0/Rm - bits<14> addr; - bits<4> Rt; - let Inst{3-0} = addr{3-0}; // imm3_0/Rm - let Inst{4} = 1; - let Inst{5} = 1; // H bit - let Inst{6} = 0; // S bit - let Inst{7} = 1; - let Inst{11-8} = addr{7-4}; // imm7_4/zero - let Inst{15-12} = Rt; // Rt - let Inst{19-16} = addr{12-9}; // Rn - let Inst{20} = 0; // L bit - let Inst{21} = 0; // W bit - let Inst{22} = addr{13}; // 1 == imm8, 0 == Rm - let Inst{23} = addr{8}; // U bit - let Inst{24} = 0; // P bit - let Inst{27-25} = 0b000; -} -class AI3stdpo pattern> - : I { - let Inst{4} = 1; - let Inst{5} = 1; // H bit - let Inst{6} = 1; // S bit - let Inst{7} = 1; - let Inst{20} = 0; // L bit - let Inst{21} = 0; // W bit - let Inst{24} = 0; // P bit - let Inst{27-25} = 0b000; + let DecoderMethod = "DecodeAddrMode3Instruction"; } // addrmode4 instructions @@ -843,6 +792,23 @@ class AMiscA1I opcod, bits<4> opc7_4, dag oops, dag iops, } // PKH instructions +def PKHLSLAsmOperand : AsmOperandClass { + let Name = "PKHLSLImm"; + let ParserMethod = "parsePKHLSLImm"; +} +def pkh_lsl_amt: Operand, ImmLeaf= 0 && Imm < 32; }]>{ + let PrintMethod = "printPKHLSLShiftImm"; + let ParserMatchClass = PKHLSLAsmOperand; +} +def PKHASRAsmOperand : AsmOperandClass { + let Name = "PKHASRImm"; + let ParserMethod = "parsePKHASRImm"; +} +def pkh_asr_amt: Operand, ImmLeaf 0 && Imm <= 32; }]>{ + let PrintMethod = "printPKHASRShiftImm"; + let ParserMatchClass = PKHASRAsmOperand; +} + class APKHI opcod, bit tb, dag oops, dag iops, InstrItinClass itin, string opc, string asm, list pattern> : I opcod, bit tb, dag oops, dag iops, InstrItinClass itin, bits<4> Rd; bits<4> Rn; bits<4> Rm; - bits<8> sh; + bits<5> sh; let Inst{27-20} = opcod; let Inst{19-16} = Rn; let Inst{15-12} = Rd; - let Inst{11-7} = sh{7-3}; + let Inst{11-7} = sh; let Inst{6} = tb; let Inst{5-4} = 0b01; let Inst{3-0} = Rm; @@ -949,7 +915,9 @@ class Thumb1sI Predicates = [IsThumb, IsThumb1Only]; + let DecoderNamespace = "ThumbSBit"; } class T1sI Predicates = [IsThumb2]; + let DecoderNamespace = "Thumb2"; } // Same as Thumb2I except it can optionally modify CPSR. Note it's modeled as an @@ -1091,6 +1060,7 @@ class Thumb2sI Predicates = [IsThumb2]; + let DecoderNamespace = "Thumb2"; } // Special cases @@ -1103,6 +1073,7 @@ class Thumb2XI Predicates = [IsThumb2]; + let DecoderNamespace = "Thumb2"; } class ThumbXI Predicates = [IsThumb, IsThumb1Only]; + let DecoderNamespace = "Thumb"; } class T2I pattern> : Thumb2I; class T2Ii8s4 pattern> - : Thumb2I pattern> + : Thumb2I { bits<4> Rt; bits<4> Rt2; @@ -1149,6 +1121,26 @@ class T2Ii8s4 pattern> + : Thumb2I { + bits<4> Rt; + bits<4> Rt2; + bits<4> addr; + bits<9> imm; + let Inst{31-25} = 0b1110100; + let Inst{24} = P; + let Inst{23} = imm{8}; + let Inst{22} = 1; + let Inst{21} = W; + let Inst{20} = isLoad; + let Inst{19-16} = addr; + let Inst{15-12} = Rt{3-0}; + let Inst{11-8} = Rt2{3-0}; + let Inst{7-0} = imm{7-0}; +} class T2sI pattern> @@ -1172,8 +1164,8 @@ class T2XIt pattern> : Thumb2XI; -// T2Iidxldst - Thumb2 indexed load / store instructions. -class T2Iidxldst opcod, bit load, bit pre, +// T2Ipreldst - Thumb2 pre-indexed load / store instructions. +class T2Ipreldst opcod, bit load, bit pre, dag oops, dag iops, AddrMode am, IndexMode im, InstrItinClass itin, string opc, string asm, string cstr, list pattern> @@ -1183,25 +1175,60 @@ class T2Iidxldst opcod, bit load, bit pre, let AsmString = !strconcat(opc, "${p}", asm); let Pattern = pattern; list Predicates = [IsThumb2]; + let DecoderNamespace = "Thumb2"; + + bits<4> Rt; + bits<13> addr; let Inst{31-27} = 0b11111; let Inst{26-25} = 0b00; let Inst{24} = signed; let Inst{23} = 0; let Inst{22-21} = opcod; let Inst{20} = load; + let Inst{19-16} = addr{12-9}; + let Inst{15-12} = Rt{3-0}; let Inst{11} = 1; // (P, W) = (1, 1) Pre-indexed or (0, 1) Post-indexed let Inst{10} = pre; // The P bit. + let Inst{9} = addr{8}; // Sign bit let Inst{8} = 1; // The W bit. + let Inst{7-0} = addr{7-0}; - bits<9> addr; - let Inst{7-0} = addr{7-0}; - let Inst{9} = addr{8}; // Sign bit + let DecoderMethod = "DecodeT2LdStPre"; +} + +// T2Ipostldst - Thumb2 post-indexed load / store instructions. +class T2Ipostldst opcod, bit load, bit pre, + dag oops, dag iops, + AddrMode am, IndexMode im, InstrItinClass itin, + string opc, string asm, string cstr, list pattern> + : InstARM { + let OutOperandList = oops; + let InOperandList = !con(iops, (ins pred:$p)); + let AsmString = !strconcat(opc, "${p}", asm); + let Pattern = pattern; + list Predicates = [IsThumb2]; + let DecoderNamespace = "Thumb2"; bits<4> Rt; bits<4> Rn; + bits<9> offset; + let Inst{31-27} = 0b11111; + let Inst{26-25} = 0b00; + let Inst{24} = signed; + let Inst{23} = 0; + let Inst{22-21} = opcod; + let Inst{20} = load; + let Inst{19-16} = Rn; let Inst{15-12} = Rt{3-0}; - let Inst{19-16} = Rn{3-0}; + let Inst{11} = 1; + // (P, W) = (1, 1) Pre-indexed or (0, 1) Post-indexed + let Inst{10} = pre; // The P bit. + let Inst{9} = offset{8}; // Sign bit + let Inst{8} = 1; // The W bit. + let Inst{7-0} = offset{7-0}; + + let DecoderMethod = "DecodeT2LdStPre"; } // Tv5Pat - Same as Pat<>, but requires V5T Thumb mode. @@ -1242,6 +1269,7 @@ class VFPI Predicates = [HasVFP2]; } @@ -1257,6 +1285,7 @@ class VFPXI Predicates = [HasVFP2]; } @@ -1574,6 +1603,7 @@ class NeonI Predicates = [HasNEON]; + let DecoderNamespace = "NEON"; } // Same as NeonI except it does not have a "data type" specifier. @@ -1586,6 +1616,7 @@ class NeonXI Predicates = [HasNEON]; + let DecoderNamespace = "NEON"; } class NLdSt op21_20, bits<4> op11_8, bits<4> op7_4, @@ -1600,6 +1631,7 @@ class NLdSt op21_20, bits<4> op11_8, bits<4> op7_4, let Inst{7-4} = op7_4; let PostEncoderMethod = "NEONThumb2LoadStorePostEncoder"; + let DecoderNamespace = "NEONLoadStore"; bits<5> Vd; bits<6> Rn; @@ -1643,6 +1675,7 @@ class NDataI { let Inst{31-25} = 0b1111001; let PostEncoderMethod = "NEONThumb2DataIPostEncoder"; + let DecoderNamespace = "NEONData"; } class NDataXI { let Inst{31-25} = 0b1111001; let PostEncoderMethod = "NEONThumb2DataIPostEncoder"; + let DecoderNamespace = "NEONData"; } // NEON "one register and a modified immediate" format. @@ -1677,6 +1711,7 @@ class N1ModImm op21_19, bits<4> op11_8, bit op7, bit op6, let Inst{24} = SIMM{7}; let Inst{18-16} = SIMM{6-4}; let Inst{3-0} = SIMM{3-0}; + let DecoderMethod = "DecodeNEONModImmInstruction"; } // NEON 2 vector register format. @@ -1874,6 +1909,7 @@ class NVLaneOp opcod1, bits<4> opcod2, bits<2> opcod3, list Predicates = [HasNEON]; let PostEncoderMethod = "NEONThumb2DupPostEncoder"; + let DecoderNamespace = "NEONDup"; bits<5> V; bits<4> R; @@ -1915,7 +1951,6 @@ class NVDupLane op19_16, bit op6, dag oops, dag iops, bits<5> Vd; bits<5> Vm; - bits<4> lane; let Inst{22} = Vd{4}; let Inst{15-12} = Vd{3-0}; diff --git a/lib/Target/ARM/ARMInstrInfo.cpp b/lib/Target/ARM/ARMInstrInfo.cpp index adcbf1806fe3d..48da03f63bb93 100644 --- a/lib/Target/ARM/ARMInstrInfo.cpp +++ b/lib/Target/ARM/ARMInstrInfo.cpp @@ -13,8 +13,8 @@ #include "ARMInstrInfo.h" #include "ARM.h" -#include "ARMAddressingModes.h" #include "ARMMachineFunctionInfo.h" +#include "MCTargetDesc/ARMAddressingModes.h" #include "llvm/ADT/STLExtras.h" #include "llvm/CodeGen/LiveVariables.h" #include "llvm/CodeGen/MachineFrameInfo.h" @@ -30,14 +30,18 @@ ARMInstrInfo::ARMInstrInfo(const ARMSubtarget &STI) unsigned ARMInstrInfo::getUnindexedOpcode(unsigned Opc) const { switch (Opc) { default: break; - case ARM::LDR_PRE: - case ARM::LDR_POST: + case ARM::LDR_PRE_IMM: + case ARM::LDR_PRE_REG: + case ARM::LDR_POST_IMM: + case ARM::LDR_POST_REG: return ARM::LDRi12; case ARM::LDRH_PRE: case ARM::LDRH_POST: return ARM::LDRH; - case ARM::LDRB_PRE: - case ARM::LDRB_POST: + case ARM::LDRB_PRE_IMM: + case ARM::LDRB_PRE_REG: + case ARM::LDRB_POST_IMM: + case ARM::LDRB_POST_REG: return ARM::LDRBi12; case ARM::LDRSH_PRE: case ARM::LDRSH_POST: @@ -45,14 +49,18 @@ unsigned ARMInstrInfo::getUnindexedOpcode(unsigned Opc) const { case ARM::LDRSB_PRE: case ARM::LDRSB_POST: return ARM::LDRSB; - case ARM::STR_PRE: - case ARM::STR_POST: + case ARM::STR_PRE_IMM: + case ARM::STR_PRE_REG: + case ARM::STR_POST_IMM: + case ARM::STR_POST_REG: return ARM::STRi12; case ARM::STRH_PRE: case ARM::STRH_POST: return ARM::STRH; - case ARM::STRB_PRE: - case ARM::STRB_POST: + case ARM::STRB_PRE_IMM: + case ARM::STRB_PRE_REG: + case ARM::STRB_POST_IMM: + case ARM::STRB_POST_REG: return ARM::STRBi12; } diff --git a/lib/Target/ARM/ARMInstrInfo.td b/lib/Target/ARM/ARMInstrInfo.td index a42dd1a54ec72..2cf0f09ffc64e 100644 --- a/lib/Target/ARM/ARMInstrInfo.td +++ b/lib/Target/ARM/ARMInstrInfo.td @@ -70,6 +70,18 @@ def SDT_ARMTCRET : SDTypeProfile<0, 1, [SDTCisPtrTy<0>]>; def SDT_ARMBFI : SDTypeProfile<1, 3, [SDTCisVT<0, i32>, SDTCisVT<1, i32>, SDTCisVT<2, i32>, SDTCisVT<3, i32>]>; +def SDTBinaryArithWithFlags : SDTypeProfile<2, 2, + [SDTCisSameAs<0, 2>, + SDTCisSameAs<0, 3>, + SDTCisInt<0>, SDTCisVT<1, i32>]>; + +// SDTBinaryArithWithFlagsInOut - RES1, CPSR = op LHS, RHS, CPSR +def SDTBinaryArithWithFlagsInOut : SDTypeProfile<2, 3, + [SDTCisSameAs<0, 2>, + SDTCisSameAs<0, 3>, + SDTCisInt<0>, + SDTCisVT<1, i32>, + SDTCisVT<4, i32>]>; // Node definitions. def ARMWrapper : SDNode<"ARMISD::Wrapper", SDTIntUnaryOp>; def ARMWrapperDYN : SDNode<"ARMISD::WrapperDYN", SDTIntUnaryOp>; @@ -120,6 +132,12 @@ def ARMsrl_flag : SDNode<"ARMISD::SRL_FLAG", SDTIntUnaryOp, [SDNPOutGlue]>; def ARMsra_flag : SDNode<"ARMISD::SRA_FLAG", SDTIntUnaryOp, [SDNPOutGlue]>; def ARMrrx : SDNode<"ARMISD::RRX" , SDTIntUnaryOp, [SDNPInGlue ]>; +def ARMaddc : SDNode<"ARMISD::ADDC", SDTBinaryArithWithFlags, + [SDNPCommutative]>; +def ARMsubc : SDNode<"ARMISD::SUBC", SDTBinaryArithWithFlags>; +def ARMadde : SDNode<"ARMISD::ADDE", SDTBinaryArithWithFlagsInOut>; +def ARMsube : SDNode<"ARMISD::SUBE", SDTBinaryArithWithFlagsInOut>; + def ARMthread_pointer: SDNode<"ARMISD::THREAD_POINTER", SDT_ARMThreadPointer>; def ARMeh_sjlj_setjmp: SDNode<"ARMISD::EH_SJLJ_SETJMP", SDT_ARMEH_SJLJ_Setjmp, [SDNPHasChain]>; @@ -187,10 +205,16 @@ def IsThumb : Predicate<"Subtarget->isThumb()">, def IsThumb1Only : Predicate<"Subtarget->isThumb1Only()">; def IsThumb2 : Predicate<"Subtarget->isThumb2()">, AssemblerPredicate<"ModeThumb,FeatureThumb2">; +def IsMClass : Predicate<"Subtarget->isMClass()">, + AssemblerPredicate<"FeatureMClass">; +def IsARClass : Predicate<"!Subtarget->isMClass()">, + AssemblerPredicate<"!FeatureMClass">; def IsARM : Predicate<"!Subtarget->isThumb()">, AssemblerPredicate<"!ModeThumb">; def IsDarwin : Predicate<"Subtarget->isTargetDarwin()">; def IsNotDarwin : Predicate<"!Subtarget->isTargetDarwin()">; +def IsNaCl : Predicate<"Subtarget->isTargetNaCl()">, + AssemblerPredicate<"ModeNaCl">; // FIXME: Eventually this will be just "hasV6T2Ops". def UseMovt : Predicate<"Subtarget->useMovt()">; @@ -263,24 +287,11 @@ def imm0_65535 : Operand, ImmLeaf : + PatFrag<(ops node:$LHS, node:$RHS, node:$FLAG), res>; class BinOpFrag : PatFrag<(ops node:$LHS, node:$RHS), res>; class UnOpFrag : PatFrag<(ops node:$Src), res>; -/// adde and sube predicates - True based on whether the carry flag output -/// will be needed or not. -def adde_dead_carry : - PatFrag<(ops node:$LHS, node:$RHS), (adde node:$LHS, node:$RHS), - [{return !N->hasAnyUseOfValue(1);}]>; -def sube_dead_carry : - PatFrag<(ops node:$LHS, node:$RHS), (sube node:$LHS, node:$RHS), - [{return !N->hasAnyUseOfValue(1);}]>; -def adde_live_carry : - PatFrag<(ops node:$LHS, node:$RHS), (adde node:$LHS, node:$RHS), - [{return N->hasAnyUseOfValue(1);}]>; -def sube_live_carry : - PatFrag<(ops node:$LHS, node:$RHS), (sube node:$LHS, node:$RHS), - [{return N->hasAnyUseOfValue(1);}]>; - // An 'and' node with a single use. def and_su : PatFrag<(ops node:$lhs, node:$rhs), (and node:$lhs, node:$rhs), [{ return N->hasOneUse(); @@ -315,6 +326,7 @@ def fsub_mlx : PatFrag<(ops node:$lhs, node:$rhs),(fsub node:$lhs, node:$rhs),[{ def brtarget : Operand { let EncoderMethod = "getBranchTargetOpValue"; let OperandType = "OPERAND_PCREL"; + let DecoderMethod = "DecodeT2BROperand"; } // FIXME: get rid of this one? @@ -345,39 +357,35 @@ def bl_target : Operand { let OperandType = "OPERAND_PCREL"; } - -// A list of registers separated by comma. Used by load/store multiple. -def RegListAsmOperand : AsmOperandClass { - let Name = "RegList"; - let SuperClasses = []; -} - -def DPRRegListAsmOperand : AsmOperandClass { - let Name = "DPRRegList"; - let SuperClasses = []; -} - -def SPRRegListAsmOperand : AsmOperandClass { - let Name = "SPRRegList"; - let SuperClasses = []; +def blx_target : Operand { + // Encoded the same as branch targets. + let EncoderMethod = "getARMBLXTargetOpValue"; + let OperandType = "OPERAND_PCREL"; } +// A list of registers separated by comma. Used by load/store multiple. +def RegListAsmOperand : AsmOperandClass { let Name = "RegList"; } def reglist : Operand { let EncoderMethod = "getRegisterListOpValue"; let ParserMatchClass = RegListAsmOperand; let PrintMethod = "printRegisterList"; + let DecoderMethod = "DecodeRegListOperand"; } +def DPRRegListAsmOperand : AsmOperandClass { let Name = "DPRRegList"; } def dpr_reglist : Operand { let EncoderMethod = "getRegisterListOpValue"; let ParserMatchClass = DPRRegListAsmOperand; let PrintMethod = "printRegisterList"; + let DecoderMethod = "DecodeDPRRegListOperand"; } +def SPRRegListAsmOperand : AsmOperandClass { let Name = "SPRRegList"; } def spr_reglist : Operand { let EncoderMethod = "getRegisterListOpValue"; let ParserMatchClass = SPRRegListAsmOperand; let PrintMethod = "printRegisterList"; + let DecoderMethod = "DecodeSPRRegListOperand"; } // An operand for the CONSTPOOL_ENTRY pseudo-instruction. @@ -397,56 +405,99 @@ def adrlabel : Operand { def neon_vcvt_imm32 : Operand { let EncoderMethod = "getNEONVcvtImm32OpValue"; + let DecoderMethod = "DecodeVCVTImmOperand"; } // rot_imm: An integer that encodes a rotate amount. Must be 8, 16, or 24. -def rot_imm : Operand, ImmLeaf { - let EncoderMethod = "getRotImmOpValue"; +def rot_imm_XFORM: SDNodeXFormgetZExtValue()){ + default: assert(0); + case 0: return CurDAG->getTargetConstant(0, MVT::i32); + case 8: return CurDAG->getTargetConstant(1, MVT::i32); + case 16: return CurDAG->getTargetConstant(2, MVT::i32); + case 24: return CurDAG->getTargetConstant(3, MVT::i32); + } +}]>; +def RotImmAsmOperand : AsmOperandClass { + let Name = "RotImm"; + let ParserMethod = "parseRotImm"; } - -def ShifterAsmOperand : AsmOperandClass { - let Name = "Shifter"; - let SuperClasses = []; +def rot_imm : Operand, PatLeaf<(i32 imm), [{ + int32_t v = N->getZExtValue(); + return v == 8 || v == 16 || v == 24; }], + rot_imm_XFORM> { + let PrintMethod = "printRotImmOperand"; + let ParserMatchClass = RotImmAsmOperand; } // shift_imm: An integer that encodes a shift amount and the type of shift -// (currently either asr or lsl) using the same encoding used for the -// immediates in so_reg operands. +// (asr or lsl). The 6-bit immediate encodes as: +// {5} 0 ==> lsl +// 1 asr +// {4-0} imm5 shift amount. +// asr #32 encoded as imm5 == 0. +def ShifterImmAsmOperand : AsmOperandClass { + let Name = "ShifterImm"; + let ParserMethod = "parseShifterImm"; +} def shift_imm : Operand { let PrintMethod = "printShiftImmOperand"; - let ParserMatchClass = ShifterAsmOperand; + let ParserMatchClass = ShifterImmAsmOperand; } -def ShiftedRegAsmOperand : AsmOperandClass { - let Name = "ShiftedReg"; +// shifter_operand operands: so_reg_reg, so_reg_imm, and so_imm. +def ShiftedRegAsmOperand : AsmOperandClass { let Name = "RegShiftedReg"; } +def so_reg_reg : Operand, // reg reg imm + ComplexPattern { + let EncoderMethod = "getSORegRegOpValue"; + let PrintMethod = "printSORegRegOperand"; + let DecoderMethod = "DecodeSORegRegOperand"; + let ParserMatchClass = ShiftedRegAsmOperand; + let MIOperandInfo = (ops GPRnopc, GPRnopc, i32imm); } -// shifter_operand operands: so_reg and so_imm. -def so_reg : Operand, // reg reg imm - ComplexPattern { - let EncoderMethod = "getSORegOpValue"; - let PrintMethod = "printSORegOperand"; - let ParserMatchClass = ShiftedRegAsmOperand; - let MIOperandInfo = (ops GPR, GPR, shift_imm); +def ShiftedImmAsmOperand : AsmOperandClass { let Name = "RegShiftedImm"; } +def so_reg_imm : Operand, // reg imm + ComplexPattern { + let EncoderMethod = "getSORegImmOpValue"; + let PrintMethod = "printSORegImmOperand"; + let DecoderMethod = "DecodeSORegImmOperand"; + let ParserMatchClass = ShiftedImmAsmOperand; + let MIOperandInfo = (ops GPR, i32imm); +} + +// FIXME: Does this need to be distinct from so_reg? +def shift_so_reg_reg : Operand, // reg reg imm + ComplexPattern { + let EncoderMethod = "getSORegRegOpValue"; + let PrintMethod = "printSORegRegOperand"; + let DecoderMethod = "DecodeSORegRegOperand"; + let MIOperandInfo = (ops GPR, GPR, i32imm); } + // FIXME: Does this need to be distinct from so_reg? -def shift_so_reg : Operand, // reg reg imm - ComplexPattern, // reg reg imm + ComplexPattern { - let EncoderMethod = "getSORegOpValue"; - let PrintMethod = "printSORegOperand"; - let MIOperandInfo = (ops GPR, GPR, shift_imm); + let EncoderMethod = "getSORegImmOpValue"; + let PrintMethod = "printSORegImmOperand"; + let DecoderMethod = "DecodeSORegImmOperand"; + let MIOperandInfo = (ops GPR, i32imm); } + // so_imm - Match a 32-bit shifter_operand immediate operand, which is an // 8-bit immediate rotated by an arbitrary number of bits. +def SOImmAsmOperand: AsmOperandClass { let Name = "ARMSOImm"; } def so_imm : Operand, ImmLeaf { let EncoderMethod = "getSOImmOpValue"; + let ParserMatchClass = SOImmAsmOperand; + let DecoderMethod = "DecodeSOImmOperand"; } // Break so_imm's up into two pieces. This handles immediates with up to 16 @@ -464,7 +515,7 @@ def arm_i32imm : PatLeaf<(imm), [{ return ARM_AM::isSOImmTwoPartVal((unsigned)N->getZExtValue()); }]>; -/// imm0_7 predicate - Immediate in the range [0,31]. +/// imm0_7 predicate - Immediate in the range [0,7]. def Imm0_7AsmOperand: AsmOperandClass { let Name = "Imm0_7"; } def imm0_7 : Operand, ImmLeaf= 0 && Imm < 8; @@ -472,7 +523,7 @@ def imm0_7 : Operand, ImmLeaf, ImmLeaf= 0 && Imm < 16; @@ -481,68 +532,83 @@ def imm0_15 : Operand, ImmLeaf, ImmLeaf= 0 && Imm < 32; -}]>; - -/// imm0_31_m1 - Matches and prints like imm0_31, but encodes as 'value - 1'. -def imm0_31_m1 : Operand, ImmLeaf= 0 && Imm < 32; }]> { - let EncoderMethod = "getImmMinusOneOpValue"; + let ParserMatchClass = Imm0_31AsmOperand; } -// i32imm_hilo16 - For movt/movw - sets the MC Encoder method. -// The imm is split into imm{15-12}, imm{11-0} +/// imm0_255 predicate - Immediate in the range [0,255]. +def Imm0_255AsmOperand : AsmOperandClass { let Name = "Imm0_255"; } +def imm0_255 : Operand, ImmLeaf= 0 && Imm < 256; }]> { + let ParserMatchClass = Imm0_255AsmOperand; +} + +// imm0_65535_expr - For movt/movw - 16-bit immediate that can also reference +// a relocatable expression. // -def i32imm_hilo16 : Operand { +// FIXME: This really needs a Thumb version separate from the ARM version. +// While the range is the same, and can thus use the same match class, +// the encoding is different so it should have a different encoder method. +def Imm0_65535ExprAsmOperand: AsmOperandClass { let Name = "Imm0_65535Expr"; } +def imm0_65535_expr : Operand { let EncoderMethod = "getHiLo16ImmOpValue"; + let ParserMatchClass = Imm0_65535ExprAsmOperand; +} + +/// imm24b - True if the 32-bit immediate is encodable in 24 bits. +def Imm24bitAsmOperand: AsmOperandClass { let Name = "Imm24bit"; } +def imm24b : Operand, ImmLeaf= 0 && Imm <= 0xffffff; +}]> { + let ParserMatchClass = Imm24bitAsmOperand; } + /// bf_inv_mask_imm predicate - An AND mask to clear an arbitrary width bitfield /// e.g., 0xf000ffff +def BitfieldAsmOperand : AsmOperandClass { + let Name = "Bitfield"; + let ParserMethod = "parseBitfield"; +} def bf_inv_mask_imm : Operand, PatLeaf<(imm), [{ return ARM::isBitFieldInvertedMask(N->getZExtValue()); }] > { let EncoderMethod = "getBitfieldInvertedMaskOpValue"; let PrintMethod = "printBitfieldInvMaskImmOperand"; + let DecoderMethod = "DecodeBitfieldMaskOperand"; + let ParserMatchClass = BitfieldAsmOperand; } -/// lsb_pos_imm - position of the lsb bit, used by BFI4p and t2BFI4p -def lsb_pos_imm : Operand, ImmLeaf(Imm); +def imm1_32_XFORM: SDNodeXFormgetTargetConstant((int)N->getZExtValue() - 1, MVT::i32); }]>; - -/// width_imm - number of bits to be copied, used by BFI4p and t2BFI4p -def width_imm : Operand, ImmLeaf 0 && Imm <= 32; -}] > { - let EncoderMethod = "getMsbOpValue"; -} - -def ssat_imm : Operand, ImmLeaf 0 && Imm <= 32; -}]> { - let EncoderMethod = "getSsatBitPosValue"; +def Imm1_32AsmOperand: AsmOperandClass { let Name = "Imm1_32"; } +def imm1_32 : Operand, PatLeaf<(imm), [{ + uint64_t Imm = N->getZExtValue(); + return Imm > 0 && Imm <= 32; + }], + imm1_32_XFORM> { + let PrintMethod = "printImmPlusOneOperand"; + let ParserMatchClass = Imm1_32AsmOperand; +} + +def imm1_16_XFORM: SDNodeXFormgetTargetConstant((int)N->getZExtValue() - 1, MVT::i32); +}]>; +def Imm1_16AsmOperand: AsmOperandClass { let Name = "Imm1_16"; } +def imm1_16 : Operand, PatLeaf<(imm), [{ return Imm > 0 && Imm <= 16; }], + imm1_16_XFORM> { + let PrintMethod = "printImmPlusOneOperand"; + let ParserMatchClass = Imm1_16AsmOperand; } // Define ARM specific addressing modes. - -def MemMode2AsmOperand : AsmOperandClass { - let Name = "MemMode2"; - let SuperClasses = []; - let ParserMethod = "tryParseMemMode2Operand"; -} - -def MemMode3AsmOperand : AsmOperandClass { - let Name = "MemMode3"; - let SuperClasses = []; - let ParserMethod = "tryParseMemMode3Operand"; -} - // addrmode_imm12 := reg +/- imm12 // +def MemImm12OffsetAsmOperand : AsmOperandClass { let Name = "MemImm12Offset"; } def addrmode_imm12 : Operand, ComplexPattern { // 12-bit immediate operand. Note that instructions using this encode @@ -551,53 +617,129 @@ def addrmode_imm12 : Operand, let EncoderMethod = "getAddrModeImm12OpValue"; let PrintMethod = "printAddrModeImm12Operand"; + let DecoderMethod = "DecodeAddrModeImm12Operand"; + let ParserMatchClass = MemImm12OffsetAsmOperand; let MIOperandInfo = (ops GPR:$base, i32imm:$offsimm); } // ldst_so_reg := reg +/- reg shop imm // +def MemRegOffsetAsmOperand : AsmOperandClass { let Name = "MemRegOffset"; } def ldst_so_reg : Operand, ComplexPattern { let EncoderMethod = "getLdStSORegOpValue"; // FIXME: Simplify the printer let PrintMethod = "printAddrMode2Operand"; - let MIOperandInfo = (ops GPR:$base, GPR:$offsreg, i32imm:$offsimm); + let DecoderMethod = "DecodeSORegMemOperand"; + let ParserMatchClass = MemRegOffsetAsmOperand; + let MIOperandInfo = (ops GPR:$base, GPRnopc:$offsreg, i32imm:$shift); } +// postidx_imm8 := +/- [0,255] +// +// 9 bit value: +// {8} 1 is imm8 is non-negative. 0 otherwise. +// {7-0} [0,255] imm8 value. +def PostIdxImm8AsmOperand : AsmOperandClass { let Name = "PostIdxImm8"; } +def postidx_imm8 : Operand { + let PrintMethod = "printPostIdxImm8Operand"; + let ParserMatchClass = PostIdxImm8AsmOperand; + let MIOperandInfo = (ops i32imm); +} + +// postidx_imm8s4 := +/- [0,1020] +// +// 9 bit value: +// {8} 1 is imm8 is non-negative. 0 otherwise. +// {7-0} [0,255] imm8 value, scaled by 4. +def PostIdxImm8s4AsmOperand : AsmOperandClass { let Name = "PostIdxImm8s4"; } +def postidx_imm8s4 : Operand { + let PrintMethod = "printPostIdxImm8s4Operand"; + let ParserMatchClass = PostIdxImm8s4AsmOperand; + let MIOperandInfo = (ops i32imm); +} + + +// postidx_reg := +/- reg +// +def PostIdxRegAsmOperand : AsmOperandClass { + let Name = "PostIdxReg"; + let ParserMethod = "parsePostIdxReg"; +} +def postidx_reg : Operand { + let EncoderMethod = "getPostIdxRegOpValue"; + let DecoderMethod = "DecodePostIdxReg"; + let PrintMethod = "printPostIdxRegOperand"; + let ParserMatchClass = PostIdxRegAsmOperand; + let MIOperandInfo = (ops GPR, i32imm); +} + + // addrmode2 := reg +/- imm12 // := reg +/- reg shop imm // +// FIXME: addrmode2 should be refactored the rest of the way to always +// use explicit imm vs. reg versions above (addrmode_imm12 and ldst_so_reg). +def AddrMode2AsmOperand : AsmOperandClass { let Name = "AddrMode2"; } def addrmode2 : Operand, ComplexPattern { let EncoderMethod = "getAddrMode2OpValue"; let PrintMethod = "printAddrMode2Operand"; - let ParserMatchClass = MemMode2AsmOperand; + let ParserMatchClass = AddrMode2AsmOperand; let MIOperandInfo = (ops GPR:$base, GPR:$offsreg, i32imm:$offsimm); } -def am2offset : Operand, - ComplexPattern, + ComplexPattern { + let EncoderMethod = "getAddrMode2OffsetOpValue"; + let PrintMethod = "printAddrMode2OffsetOperand"; + // When using this for assembly, it's always as a post-index offset. + let ParserMatchClass = PostIdxRegShiftedAsmOperand; + let MIOperandInfo = (ops GPR, i32imm); +} + +// FIXME: am2offset_imm should only need the immediate, not the GPR. Having +// the GPR is purely vestigal at this point. +def AM2OffsetImmAsmOperand : AsmOperandClass { let Name = "AM2OffsetImm"; } +def am2offset_imm : Operand, + ComplexPattern { let EncoderMethod = "getAddrMode2OffsetOpValue"; let PrintMethod = "printAddrMode2OffsetOperand"; + let ParserMatchClass = AM2OffsetImmAsmOperand; let MIOperandInfo = (ops GPR, i32imm); } + // addrmode3 := reg +/- reg // addrmode3 := reg +/- imm8 // +// FIXME: split into imm vs. reg versions. +def AddrMode3AsmOperand : AsmOperandClass { let Name = "AddrMode3"; } def addrmode3 : Operand, ComplexPattern { let EncoderMethod = "getAddrMode3OpValue"; let PrintMethod = "printAddrMode3Operand"; - let ParserMatchClass = MemMode3AsmOperand; + let ParserMatchClass = AddrMode3AsmOperand; let MIOperandInfo = (ops GPR:$base, GPR:$offsreg, i32imm:$offsimm); } +// FIXME: split into imm vs. reg versions. +// FIXME: parser method to handle +/- register. +def AM3OffsetAsmOperand : AsmOperandClass { + let Name = "AM3Offset"; + let ParserMethod = "parseAM3Offset"; +} def am3offset : Operand, ComplexPattern { let EncoderMethod = "getAddrMode3OffsetOpValue"; let PrintMethod = "printAddrMode3OffsetOperand"; + let ParserMatchClass = AM3OffsetAsmOperand; let MIOperandInfo = (ops GPR, i32imm); } @@ -608,28 +750,28 @@ def ldstm_mode : OptionalDefOperand { let PrintMethod = "printLdStmModeOperand"; } -def MemMode5AsmOperand : AsmOperandClass { - let Name = "MemMode5"; - let SuperClasses = []; -} - // addrmode5 := reg +/- imm8*4 // +def AddrMode5AsmOperand : AsmOperandClass { let Name = "AddrMode5"; } def addrmode5 : Operand, ComplexPattern { let PrintMethod = "printAddrMode5Operand"; - let MIOperandInfo = (ops GPR:$base, i32imm); - let ParserMatchClass = MemMode5AsmOperand; let EncoderMethod = "getAddrMode5OpValue"; + let DecoderMethod = "DecodeAddrMode5Operand"; + let ParserMatchClass = AddrMode5AsmOperand; + let MIOperandInfo = (ops GPR:$base, i32imm); } // addrmode6 := reg with optional alignment // +def AddrMode6AsmOperand : AsmOperandClass { let Name = "AlignedMemory"; } def addrmode6 : Operand, ComplexPattern{ let PrintMethod = "printAddrMode6Operand"; - let MIOperandInfo = (ops GPR:$addr, i32imm); + let MIOperandInfo = (ops GPR:$addr, i32imm:$align); let EncoderMethod = "getAddrMode6AddressOpValue"; + let DecoderMethod = "DecodeAddrMode6Operand"; + let ParserMatchClass = AddrMode6AsmOperand; } def am6offset : Operand, @@ -638,6 +780,7 @@ def am6offset : Operand, let PrintMethod = "printAddrMode6OffsetOperand"; let MIOperandInfo = (ops GPR); let EncoderMethod = "getAddrMode6OffsetOpValue"; + let DecoderMethod = "DecodeGPRRegisterClass"; } // Special version of addrmode6 to handle alignment encoding for VST1/VLD1 @@ -666,19 +809,15 @@ def addrmodepc : Operand, let MIOperandInfo = (ops GPR, i32imm); } -def MemMode7AsmOperand : AsmOperandClass { - let Name = "MemMode7"; - let SuperClasses = []; -} - -// addrmode7 := reg -// Used by load/store exclusive instructions. Useful to enable right assembly -// parsing and printing. Not used for any codegen matching. +// addr_offset_none := reg // -def addrmode7 : Operand { +def MemNoOffsetAsmOperand : AsmOperandClass { let Name = "MemNoOffset"; } +def addr_offset_none : Operand, + ComplexPattern { let PrintMethod = "printAddrMode7Operand"; - let MIOperandInfo = (ops GPR); - let ParserMatchClass = MemMode7AsmOperand; + let DecoderMethod = "DecodeAddrMode7Operand"; + let ParserMatchClass = MemNoOffsetAsmOperand; + let MIOperandInfo = (ops GPR:$base); } def nohash_imm : Operand { @@ -687,25 +826,30 @@ def nohash_imm : Operand { def CoprocNumAsmOperand : AsmOperandClass { let Name = "CoprocNum"; - let SuperClasses = []; - let ParserMethod = "tryParseCoprocNumOperand"; -} - -def CoprocRegAsmOperand : AsmOperandClass { - let Name = "CoprocReg"; - let SuperClasses = []; - let ParserMethod = "tryParseCoprocRegOperand"; + let ParserMethod = "parseCoprocNumOperand"; } - def p_imm : Operand { let PrintMethod = "printPImmediate"; let ParserMatchClass = CoprocNumAsmOperand; + let DecoderMethod = "DecodeCoprocessor"; } +def CoprocRegAsmOperand : AsmOperandClass { + let Name = "CoprocReg"; + let ParserMethod = "parseCoprocRegOperand"; +} def c_imm : Operand { let PrintMethod = "printCImmediate"; let ParserMatchClass = CoprocRegAsmOperand; } +def CoprocOptionAsmOperand : AsmOperandClass { + let Name = "CoprocOption"; + let ParserMethod = "parseCoprocOptionOperand"; +} +def coproc_option_imm : Operand { + let PrintMethod = "printCoprocOptionImm"; + let ParserMatchClass = CoprocOptionAsmOperand; +} //===----------------------------------------------------------------------===// @@ -748,16 +892,37 @@ multiclass AsI1_bin_irs opcod, string opc, let Inst{11-4} = 0b00000000; let Inst{3-0} = Rm; } - def rs : AsI1 { + bits<4> Rd; + bits<4> Rn; + bits<12> shift; + let Inst{25} = 0; + let Inst{19-16} = Rn; + let Inst{15-12} = Rd; + let Inst{11-5} = shift{11-5}; + let Inst{4} = 0; + let Inst{3-0} = shift{3-0}; + } + + def rsr : AsI1 { + [(set GPR:$Rd, (opnode GPR:$Rn, so_reg_reg:$shift))]> { bits<4> Rd; bits<4> Rn; bits<12> shift; let Inst{25} = 0; let Inst{19-16} = Rn; let Inst{15-12} = Rd; - let Inst{11-0} = shift; + let Inst{11-8} = shift{11-8}; + let Inst{7} = 0; + let Inst{6-5} = shift{6-5}; + let Inst{4} = 1; + let Inst{3-0} = shift{3-0}; } // Assembly aliases for optional destination operand when it's the same @@ -773,59 +938,175 @@ multiclass AsI1_bin_irs opcod, string opc, cc_out:$s)>, Requires<[IsARM]>; def : InstAlias(!strconcat(baseOpc, "rs")) GPR:$Rdn, GPR:$Rdn, - so_reg:$shift, pred:$p, + (!cast(!strconcat(baseOpc, "rsi")) GPR:$Rdn, GPR:$Rdn, + so_reg_imm:$shift, pred:$p, cc_out:$s)>, Requires<[IsARM]>; + def : InstAlias(!strconcat(baseOpc, "rsr")) GPR:$Rdn, GPR:$Rdn, + so_reg_reg:$shift, pred:$p, + cc_out:$s)>, + Requires<[IsARM]>; + } -/// AI1_bin_s_irs - Similar to AsI1_bin_irs except it sets the 's' bit so the -/// instruction modifies the CPSR register. -let isCodeGenOnly = 1, Defs = [CPSR] in { -multiclass AI1_bin_s_irs opcod, string opc, +/// AsI1_rbin_irs - Same as AsI1_bin_irs except the order of operands are +/// reversed. The 'rr' form is only defined for the disassembler; for codegen +/// it is equivalent to the AsI1_bin_irs counterpart. +multiclass AsI1_rbin_irs opcod, string opc, InstrItinClass iii, InstrItinClass iir, InstrItinClass iis, - PatFrag opnode, bit Commutable = 0> { - def ri : AI1 { + // The register-immediate version is re-materializable. This is useful + // in particular for taking the address of a local. + let isReMaterializable = 1 in { + def ri : AsI1 { + [(set GPR:$Rd, (opnode so_imm:$imm, GPR:$Rn))]> { bits<4> Rd; bits<4> Rn; bits<12> imm; let Inst{25} = 1; - let Inst{20} = 1; let Inst{19-16} = Rn; let Inst{15-12} = Rd; let Inst{11-0} = imm; } - def rr : AI1 { + [/* pattern left blank */]> { bits<4> Rd; bits<4> Rn; bits<4> Rm; - let isCommutable = Commutable; + let Inst{11-4} = 0b00000000; + let Inst{25} = 0; + let Inst{3-0} = Rm; + let Inst{15-12} = Rd; + let Inst{19-16} = Rn; + } + + def rsi : AsI1 { + bits<4> Rd; + bits<4> Rn; + bits<12> shift; let Inst{25} = 0; - let Inst{20} = 1; let Inst{19-16} = Rn; let Inst{15-12} = Rd; - let Inst{11-4} = 0b00000000; - let Inst{3-0} = Rm; + let Inst{11-5} = shift{11-5}; + let Inst{4} = 0; + let Inst{3-0} = shift{3-0}; } - def rs : AI1 { + [(set GPR:$Rd, (opnode so_reg_reg:$shift, GPR:$Rn))]> { + bits<4> Rd; + bits<4> Rn; + bits<12> shift; + let Inst{25} = 0; + let Inst{19-16} = Rn; + let Inst{15-12} = Rd; + let Inst{11-8} = shift{11-8}; + let Inst{7} = 0; + let Inst{6-5} = shift{6-5}; + let Inst{4} = 1; + let Inst{3-0} = shift{3-0}; + } + + // Assembly aliases for optional destination operand when it's the same + // as the source operand. + def : InstAlias(!strconcat(baseOpc, "ri")) GPR:$Rdn, GPR:$Rdn, + so_imm:$imm, pred:$p, + cc_out:$s)>, + Requires<[IsARM]>; + def : InstAlias(!strconcat(baseOpc, "rr")) GPR:$Rdn, GPR:$Rdn, + GPR:$Rm, pred:$p, + cc_out:$s)>, + Requires<[IsARM]>; + def : InstAlias(!strconcat(baseOpc, "rsi")) GPR:$Rdn, GPR:$Rdn, + so_reg_imm:$shift, pred:$p, + cc_out:$s)>, + Requires<[IsARM]>; + def : InstAlias(!strconcat(baseOpc, "rsr")) GPR:$Rdn, GPR:$Rdn, + so_reg_reg:$shift, pred:$p, + cc_out:$s)>, + Requires<[IsARM]>; + +} + +/// AsI1_rbin_s_is - Same as AsI1_rbin_s_is except it sets 's' bit by default. +/// +/// These opcodes will be converted to the real non-S opcodes by +/// AdjustInstrPostInstrSelection after giving then an optional CPSR operand. +let hasPostISelHook = 1, isCodeGenOnly = 1, isPseudo = 1, Defs = [CPSR] in { +multiclass AsI1_rbin_s_is opcod, string opc, + InstrItinClass iii, InstrItinClass iir, InstrItinClass iis, + PatFrag opnode, bit Commutable = 0> { + def ri : AsI1; + + def rr : AsI1; + + def rsi : AsI1; + + def rsr : AsI1 { bits<4> Rd; bits<4> Rn; bits<12> shift; let Inst{25} = 0; - let Inst{20} = 1; let Inst{19-16} = Rn; let Inst{15-12} = Rd; - let Inst{11-0} = shift; + let Inst{11-8} = shift{11-8}; + let Inst{7} = 0; + let Inst{6-5} = shift{6-5}; + let Inst{4} = 1; + let Inst{3-0} = shift{3-0}; } } } +/// AsI1_bin_s_irs - Same as AsI1_bin_irs except it sets the 's' bit by default. +/// +/// These opcodes will be converted to the real non-S opcodes by +/// AdjustInstrPostInstrSelection after giving then an optional CPSR operand. +let hasPostISelHook = 1, isCodeGenOnly = 1, isPseudo = 1, Defs = [CPSR] in { +multiclass AsI1_bin_s_irs opcod, string opc, + InstrItinClass iii, InstrItinClass iir, InstrItinClass iis, + PatFrag opnode, bit Commutable = 0> { + def ri : AsI1; + def rr : AsI1; + def rsi : AsI1; + + def rsr : AsI1; +} +} + /// AI1_cmp_irs - Defines a set of (op r, {so_imm|r|so_reg}) cmp / test /// patterns. Similar to AsI1_bin_irs except the instruction does not produce /// a explicit result, only implicitly set CPSR. @@ -857,128 +1138,190 @@ multiclass AI1_cmp_irs opcod, string opc, let Inst{11-4} = 0b00000000; let Inst{3-0} = Rm; } - def rs : AI1 { + [(opnode GPR:$Rn, so_reg_imm:$shift)]> { bits<4> Rn; bits<12> shift; let Inst{25} = 0; let Inst{20} = 1; let Inst{19-16} = Rn; let Inst{15-12} = 0b0000; - let Inst{11-0} = shift; + let Inst{11-5} = shift{11-5}; + let Inst{4} = 0; + let Inst{3-0} = shift{3-0}; } + def rsr : AI1 { + bits<4> Rn; + bits<12> shift; + let Inst{25} = 0; + let Inst{20} = 1; + let Inst{19-16} = Rn; + let Inst{15-12} = 0b0000; + let Inst{11-8} = shift{11-8}; + let Inst{7} = 0; + let Inst{6-5} = shift{6-5}; + let Inst{4} = 1; + let Inst{3-0} = shift{3-0}; + } + } } /// AI_ext_rrot - A unary operation with two forms: one whose operand is a /// register and one whose operand is a register rotated by 8/16/24. /// FIXME: Remove the 'r' variant. Its rot_imm is zero. -multiclass AI_ext_rrot opcod, string opc, PatFrag opnode> { - def r : AExtI, - Requires<[IsARM, HasV6]> { - bits<4> Rd; - bits<4> Rm; - let Inst{19-16} = 0b1111; - let Inst{15-12} = Rd; - let Inst{11-10} = 0b00; - let Inst{3-0} = Rm; - } - def r_rot : AExtI, - Requires<[IsARM, HasV6]> { - bits<4> Rd; - bits<4> Rm; - bits<2> rot; - let Inst{19-16} = 0b1111; - let Inst{15-12} = Rd; - let Inst{11-10} = rot; - let Inst{3-0} = Rm; - } +class AI_ext_rrot opcod, string opc, PatFrag opnode> + : AExtI, + Requires<[IsARM, HasV6]> { + bits<4> Rd; + bits<4> Rm; + bits<2> rot; + let Inst{19-16} = 0b1111; + let Inst{15-12} = Rd; + let Inst{11-10} = rot; + let Inst{3-0} = Rm; } -multiclass AI_ext_rrot_np opcod, string opc> { - def r : AExtI, - Requires<[IsARM, HasV6]> { - let Inst{19-16} = 0b1111; - let Inst{11-10} = 0b00; - } - def r_rot : AExtI, - Requires<[IsARM, HasV6]> { - bits<2> rot; - let Inst{19-16} = 0b1111; - let Inst{11-10} = rot; - } +class AI_ext_rrot_np opcod, string opc> + : AExtI, + Requires<[IsARM, HasV6]> { + bits<2> rot; + let Inst{19-16} = 0b1111; + let Inst{11-10} = rot; } /// AI_exta_rrot - A binary operation with two forms: one whose operand is a /// register and one whose operand is a register rotated by 8/16/24. -multiclass AI_exta_rrot opcod, string opc, PatFrag opnode> { - def rr : AExtI, - Requires<[IsARM, HasV6]> { +class AI_exta_rrot opcod, string opc, PatFrag opnode> + : AExtI, + Requires<[IsARM, HasV6]> { + bits<4> Rd; + bits<4> Rm; + bits<4> Rn; + bits<2> rot; + let Inst{19-16} = Rn; + let Inst{15-12} = Rd; + let Inst{11-10} = rot; + let Inst{9-4} = 0b000111; + let Inst{3-0} = Rm; +} + +class AI_exta_rrot_np opcod, string opc> + : AExtI, + Requires<[IsARM, HasV6]> { + bits<4> Rn; + bits<2> rot; + let Inst{19-16} = Rn; + let Inst{11-10} = rot; +} + +/// AI1_adde_sube_irs - Define instructions and patterns for adde and sube. +multiclass AI1_adde_sube_irs opcod, string opc, PatFrag opnode, + string baseOpc, bit Commutable = 0> { + let hasPostISelHook = 1, Defs = [CPSR], Uses = [CPSR] in { + def ri : AsI1, + Requires<[IsARM]> { bits<4> Rd; - bits<4> Rm; bits<4> Rn; - let Inst{19-16} = Rn; + bits<12> imm; + let Inst{25} = 1; let Inst{15-12} = Rd; - let Inst{11-10} = 0b00; - let Inst{9-4} = 0b000111; - let Inst{3-0} = Rm; - } - def rr_rot : AExtI, - Requires<[IsARM, HasV6]> { + let Inst{19-16} = Rn; + let Inst{11-0} = imm; + } + def rr : AsI1, + Requires<[IsARM]> { bits<4> Rd; + bits<4> Rn; bits<4> Rm; + let Inst{11-4} = 0b00000000; + let Inst{25} = 0; + let isCommutable = Commutable; + let Inst{3-0} = Rm; + let Inst{15-12} = Rd; + let Inst{19-16} = Rn; + } + def rsi : AsI1, + Requires<[IsARM]> { + bits<4> Rd; bits<4> Rn; - bits<2> rot; + bits<12> shift; + let Inst{25} = 0; let Inst{19-16} = Rn; let Inst{15-12} = Rd; - let Inst{11-10} = rot; - let Inst{9-4} = 0b000111; - let Inst{3-0} = Rm; - } -} - -// For disassembly only. -multiclass AI_exta_rrot_np opcod, string opc> { - def rr : AExtI, - Requires<[IsARM, HasV6]> { - let Inst{11-10} = 0b00; + let Inst{11-5} = shift{11-5}; + let Inst{4} = 0; + let Inst{3-0} = shift{3-0}; } - def rr_rot : AExtI, - Requires<[IsARM, HasV6]> { + def rsr : AsI1, + Requires<[IsARM]> { + bits<4> Rd; bits<4> Rn; - bits<2> rot; + bits<12> shift; + let Inst{25} = 0; let Inst{19-16} = Rn; - let Inst{11-10} = rot; + let Inst{15-12} = Rd; + let Inst{11-8} = shift{11-8}; + let Inst{7} = 0; + let Inst{6-5} = shift{6-5}; + let Inst{4} = 1; + let Inst{3-0} = shift{3-0}; + } } + + // Assembly aliases for optional destination operand when it's the same + // as the source operand. + def : InstAlias(!strconcat(baseOpc, "ri")) GPR:$Rdn, GPR:$Rdn, + so_imm:$imm, pred:$p, + cc_out:$s)>, + Requires<[IsARM]>; + def : InstAlias(!strconcat(baseOpc, "rr")) GPR:$Rdn, GPR:$Rdn, + GPR:$Rm, pred:$p, + cc_out:$s)>, + Requires<[IsARM]>; + def : InstAlias(!strconcat(baseOpc, "rsi")) GPR:$Rdn, GPR:$Rdn, + so_reg_imm:$shift, pred:$p, + cc_out:$s)>, + Requires<[IsARM]>; + def : InstAlias(!strconcat(baseOpc, "rsr")) GPR:$Rdn, GPR:$Rdn, + so_reg_reg:$shift, pred:$p, + cc_out:$s)>, + Requires<[IsARM]>; } -/// AI1_adde_sube_irs - Define instructions and patterns for adde and sube. -multiclass AI1_adde_sube_irs opcod, string opc, PatFrag opnode, - string baseOpc, bit Commutable = 0> { - let Uses = [CPSR] in { +/// AI1_rsc_irs - Define instructions and patterns for rsc +multiclass AI1_rsc_irs opcod, string opc, PatFrag opnode, + string baseOpc> { + let hasPostISelHook = 1, Defs = [CPSR], Uses = [CPSR] in { def ri : AsI1, + [(set GPR:$Rd, CPSR, (opnode so_imm:$imm, GPR:$Rn, CPSR))]>, Requires<[IsARM]> { bits<4> Rd; bits<4> Rn; @@ -990,31 +1333,48 @@ multiclass AI1_adde_sube_irs opcod, string opc, PatFrag opnode, } def rr : AsI1, - Requires<[IsARM]> { + [/* pattern left blank */]> { bits<4> Rd; bits<4> Rn; bits<4> Rm; let Inst{11-4} = 0b00000000; let Inst{25} = 0; - let isCommutable = Commutable; let Inst{3-0} = Rm; let Inst{15-12} = Rd; let Inst{19-16} = Rn; } - def rs : AsI1, + def rsi : AsI1, Requires<[IsARM]> { bits<4> Rd; bits<4> Rn; bits<12> shift; let Inst{25} = 0; - let Inst{11-0} = shift; + let Inst{19-16} = Rn; let Inst{15-12} = Rd; + let Inst{11-5} = shift{11-5}; + let Inst{4} = 0; + let Inst{3-0} = shift{3-0}; + } + def rsr : AsI1, + Requires<[IsARM]> { + bits<4> Rd; + bits<4> Rn; + bits<12> shift; + let Inst{25} = 0; let Inst{19-16} = Rn; + let Inst{15-12} = Rd; + let Inst{11-8} = shift{11-8}; + let Inst{7} = 0; + let Inst{6-5} = shift{6-5}; + let Inst{4} = 1; + let Inst{3-0} = shift{3-0}; } } + // Assembly aliases for optional destination operand when it's the same // as the source operand. def : InstAlias opcod, string opc, PatFrag opnode, cc_out:$s)>, Requires<[IsARM]>; def : InstAlias(!strconcat(baseOpc, "rs")) GPR:$Rdn, GPR:$Rdn, - so_reg:$shift, pred:$p, + (!cast(!strconcat(baseOpc, "rsi")) GPR:$Rdn, GPR:$Rdn, + so_reg_imm:$shift, pred:$p, + cc_out:$s)>, + Requires<[IsARM]>; + def : InstAlias(!strconcat(baseOpc, "rsr")) GPR:$Rdn, GPR:$Rdn, + so_reg_reg:$shift, pred:$p, cc_out:$s)>, Requires<[IsARM]>; -} - -// Carry setting variants -// NOTE: CPSR def omitted because it will be handled by the custom inserter. -let usesCustomInserter = 1 in { -multiclass AI1_adde_sube_s_irs { - def ri : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, so_imm:$imm), - 4, IIC_iALUi, - [(set GPR:$Rd, (opnode GPR:$Rn, so_imm:$imm))]>; - def rr : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm), - 4, IIC_iALUr, - [(set GPR:$Rd, (opnode GPR:$Rn, GPR:$Rm))]> { - let isCommutable = Commutable; - } - def rs : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, so_reg:$shift), - 4, IIC_iALUsr, - [(set GPR:$Rd, (opnode GPR:$Rn, so_reg:$shift))]>; -} } let canFoldAsLoad = 1, isReMaterializable = 1 in { @@ -1082,6 +1429,37 @@ multiclass AI_ldr1 { + // Note: We use the complex addrmode_imm12 rather than just an input + // GPR and a constrained immediate so that we can use this to match + // frame index references and avoid matching constant pool references. + def i12: AI2ldst<0b010, 1, isByte, (outs GPRnopc:$Rt), (ins addrmode_imm12:$addr), + AddrMode_i12, LdFrm, iii, opc, "\t$Rt, $addr", + [(set GPRnopc:$Rt, (opnode addrmode_imm12:$addr))]> { + bits<4> Rt; + bits<17> addr; + let Inst{23} = addr{12}; // U (add = ('U' == 1)) + let Inst{19-16} = addr{16-13}; // Rn + let Inst{15-12} = Rt; + let Inst{11-0} = addr{11-0}; // imm12 + } + def rs : AI2ldst<0b011, 1, isByte, (outs GPRnopc:$Rt), (ins ldst_so_reg:$shift), + AddrModeNone, LdFrm, iir, opc, "\t$Rt, $shift", + [(set GPRnopc:$Rt, (opnode ldst_so_reg:$shift))]> { + bits<4> Rt; + bits<17> shift; + let shift{4} = 0; // Inst{4} = 0 + let Inst{23} = shift{12}; // U (add = ('U' == 1)) + let Inst{19-16} = shift{16-13}; // Rn + let Inst{15-12} = Rt; + let Inst{11-0} = shift{11-0}; + } +} +} + + multiclass AI_str1 { // Note: We use the complex addrmode_imm12 rather than just an input @@ -1110,6 +1488,37 @@ multiclass AI_str1 { + // Note: We use the complex addrmode_imm12 rather than just an input + // GPR and a constrained immediate so that we can use this to match + // frame index references and avoid matching constant pool references. + def i12 : AI2ldst<0b010, 0, isByte, (outs), + (ins GPRnopc:$Rt, addrmode_imm12:$addr), + AddrMode_i12, StFrm, iii, opc, "\t$Rt, $addr", + [(opnode GPRnopc:$Rt, addrmode_imm12:$addr)]> { + bits<4> Rt; + bits<17> addr; + let Inst{23} = addr{12}; // U (add = ('U' == 1)) + let Inst{19-16} = addr{16-13}; // Rn + let Inst{15-12} = Rt; + let Inst{11-0} = addr{11-0}; // imm12 + } + def rs : AI2ldst<0b011, 0, isByte, (outs), (ins GPRnopc:$Rt, ldst_so_reg:$shift), + AddrModeNone, StFrm, iir, opc, "\t$Rt, $shift", + [(opnode GPRnopc:$Rt, ldst_so_reg:$shift)]> { + bits<4> Rt; + bits<17> shift; + let shift{4} = 0; // Inst{4} = 0 + let Inst{23} = shift{12}; // U (add = ('U' == 1)) + let Inst{19-16} = shift{16-13}; // Rn + let Inst{15-12} = Rt; + let Inst{11-0} = shift{11-0}; + } +} + + //===----------------------------------------------------------------------===// // Instructions //===----------------------------------------------------------------------===// @@ -1140,42 +1549,66 @@ PseudoInst<(outs), (ins i32imm:$amt, pred:$p), NoItinerary, [(ARMcallseq_start timm:$amt)]>; } -def NOP : AI<(outs), (ins), MiscFrm, NoItinerary, "nop", "", - [/* For disassembly only; pattern left blank */]>, +// Atomic pseudo-insts which will be lowered to ldrexd/strexd loops. +// (These psuedos use a hand-written selection code). +let usesCustomInserter = 1, Defs = [CPSR], mayLoad = 1, mayStore = 1 in { +def ATOMOR6432 : PseudoInst<(outs GPR:$dst1, GPR:$dst2), + (ins GPR:$addr, GPR:$src1, GPR:$src2), + NoItinerary, []>; +def ATOMXOR6432 : PseudoInst<(outs GPR:$dst1, GPR:$dst2), + (ins GPR:$addr, GPR:$src1, GPR:$src2), + NoItinerary, []>; +def ATOMADD6432 : PseudoInst<(outs GPR:$dst1, GPR:$dst2), + (ins GPR:$addr, GPR:$src1, GPR:$src2), + NoItinerary, []>; +def ATOMSUB6432 : PseudoInst<(outs GPR:$dst1, GPR:$dst2), + (ins GPR:$addr, GPR:$src1, GPR:$src2), + NoItinerary, []>; +def ATOMNAND6432 : PseudoInst<(outs GPR:$dst1, GPR:$dst2), + (ins GPR:$addr, GPR:$src1, GPR:$src2), + NoItinerary, []>; +def ATOMAND6432 : PseudoInst<(outs GPR:$dst1, GPR:$dst2), + (ins GPR:$addr, GPR:$src1, GPR:$src2), + NoItinerary, []>; +def ATOMSWAP6432 : PseudoInst<(outs GPR:$dst1, GPR:$dst2), + (ins GPR:$addr, GPR:$src1, GPR:$src2), + NoItinerary, []>; +def ATOMCMPXCHG6432 : PseudoInst<(outs GPR:$dst1, GPR:$dst2), + (ins GPR:$addr, GPR:$cmp1, GPR:$cmp2, + GPR:$set1, GPR:$set2), + NoItinerary, []>; +} + +def NOP : AI<(outs), (ins), MiscFrm, NoItinerary, "nop", "", []>, Requires<[IsARM, HasV6T2]> { let Inst{27-16} = 0b001100100000; let Inst{15-8} = 0b11110000; let Inst{7-0} = 0b00000000; } -def YIELD : AI<(outs), (ins), MiscFrm, NoItinerary, "yield", "", - [/* For disassembly only; pattern left blank */]>, +def YIELD : AI<(outs), (ins), MiscFrm, NoItinerary, "yield", "", []>, Requires<[IsARM, HasV6T2]> { let Inst{27-16} = 0b001100100000; let Inst{15-8} = 0b11110000; let Inst{7-0} = 0b00000001; } -def WFE : AI<(outs), (ins), MiscFrm, NoItinerary, "wfe", "", - [/* For disassembly only; pattern left blank */]>, +def WFE : AI<(outs), (ins), MiscFrm, NoItinerary, "wfe", "", []>, Requires<[IsARM, HasV6T2]> { let Inst{27-16} = 0b001100100000; let Inst{15-8} = 0b11110000; let Inst{7-0} = 0b00000010; } -def WFI : AI<(outs), (ins), MiscFrm, NoItinerary, "wfi", "", - [/* For disassembly only; pattern left blank */]>, +def WFI : AI<(outs), (ins), MiscFrm, NoItinerary, "wfi", "", []>, Requires<[IsARM, HasV6T2]> { let Inst{27-16} = 0b001100100000; let Inst{15-8} = 0b11110000; let Inst{7-0} = 0b00000011; } -def SEL : AI<(outs GPR:$dst), (ins GPR:$a, GPR:$b), DPFrm, NoItinerary, "sel", - "\t$dst, $a, $b", - [/* For disassembly only; pattern left blank */]>, - Requires<[IsARM, HasV6]> { +def SEL : AI<(outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm), DPFrm, NoItinerary, "sel", + "\t$Rd, $Rn, $Rm", []>, Requires<[IsARM, HasV6]> { bits<4> Rd; bits<4> Rn; bits<4> Rm; @@ -1188,8 +1621,7 @@ def SEL : AI<(outs GPR:$dst), (ins GPR:$a, GPR:$b), DPFrm, NoItinerary, "sel", } def SEV : AI<(outs), (ins), MiscFrm, NoItinerary, "sev", "", - [/* For disassembly only; pattern left blank */]>, - Requires<[IsARM, HasV6T2]> { + []>, Requires<[IsARM, HasV6T2]> { let Inst{27-16} = 0b001100100000; let Inst{15-8} = 0b11110000; let Inst{7-0} = 0b00000100; @@ -1206,14 +1638,11 @@ def BKPT : AI<(outs), (ins imm0_65535:$val), MiscFrm, NoItinerary, let Inst{7-4} = 0b0111; } -// Change Processor State is a system instruction -- for disassembly and -// parsing only. -// FIXME: Since the asm parser has currently no clean way to handle optional -// operands, create 3 versions of the same instruction. Once there's a clean -// framework to represent optional operands, change this behavior. +// Change Processor State +// FIXME: We should use InstAlias to handle the optional operands. class CPS : AXI<(outs), iops, MiscFrm, NoItinerary, !strconcat("cps", asm_ops), - [/* For disassembly only; pattern left blank */]>, Requires<[IsARM]> { + []>, Requires<[IsARM]> { bits<2> imod; bits<3> iflags; bits<5> mode; @@ -1229,17 +1658,18 @@ class CPS let Inst{4-0} = mode; } +let DecoderMethod = "DecodeCPSInstruction" in { let M = 1 in - def CPS3p : CPS<(ins imod_op:$imod, iflags_op:$iflags, i32imm:$mode), + def CPS3p : CPS<(ins imod_op:$imod, iflags_op:$iflags, imm0_31:$mode), "$imod\t$iflags, $mode">; let mode = 0, M = 0 in def CPS2p : CPS<(ins imod_op:$imod, iflags_op:$iflags), "$imod\t$iflags">; let imod = 0, iflags = 0, M = 1 in - def CPS1p : CPS<(ins i32imm:$mode), "\t$mode">; + def CPS1p : CPS<(ins imm0_31:$mode), "\t$mode">; +} // Preload signals the memory system of possible future data/instruction access. -// These are for disassembly only. multiclass APreLoad read, bits<1> data, string opc> { def i12 : AXI<(outs), (ins addrmode_imm12:$addr), MiscFrm, IIC_Preload, @@ -1271,6 +1701,7 @@ multiclass APreLoad read, bits<1> data, string opc> { let Inst{19-16} = shift{16-13}; // Rn let Inst{15-12} = 0b1111; let Inst{11-0} = shift{11-0}; + let Inst{4} = 0; } } @@ -1278,10 +1709,8 @@ defm PLD : APreLoad<1, 1, "pld">, Requires<[IsARM]>; defm PLDW : APreLoad<0, 1, "pldw">, Requires<[IsARM,HasV7,HasMP]>; defm PLI : APreLoad<1, 0, "pli">, Requires<[IsARM,HasV7]>; -def SETEND : AXI<(outs),(ins setend_op:$end), MiscFrm, NoItinerary, - "setend\t$end", - [/* For disassembly only; pattern left blank */]>, - Requires<[IsARM]> { +def SETEND : AXI<(outs), (ins setend_op:$end), MiscFrm, NoItinerary, + "setend\t$end", []>, Requires<[IsARM]> { bits<1> end; let Inst{31-10} = 0b1111000100000001000000; let Inst{9} = end; @@ -1351,14 +1780,17 @@ let neverHasSideEffects = 1, isReMaterializable = 1 in // the instruction. The {24-21} opcode bits are set by the fixup, as we don't // know until then which form of the instruction will be used. def ADR : AI1<{0,?,?,0}, (outs GPR:$Rd), (ins adrlabel:$label), - MiscFrm, IIC_iALUi, "adr", "\t$Rd, #$label", []> { + MiscFrm, IIC_iALUi, "adr", "\t$Rd, $label", []> { bits<4> Rd; - bits<12> label; + bits<14> label; let Inst{27-25} = 0b001; + let Inst{24} = 0; + let Inst{23-22} = label{13-12}; + let Inst{21} = 0; let Inst{20} = 0; let Inst{19-16} = 0b1111; let Inst{15-12} = Rd; - let Inst{11-0} = label; + let Inst{11-0} = label{11-0}; } def LEApcrel : ARMPseudoInst<(outs GPR:$Rd), (ins i32imm:$label, pred:$p), 4, IIC_iALUi, []>; @@ -1424,6 +1856,7 @@ let isCall = 1, let Inst{31-28} = 0b1110; bits<24> func; let Inst{23-0} = func; + let DecoderMethod = "DecodeBranchImmInstruction"; } def BL_pred : ABI<0b1011, (outs), (ins bl_target:$func, variable_ops), @@ -1432,6 +1865,7 @@ let isCall = 1, Requires<[IsARM, IsNotDarwin]> { bits<24> func; let Inst{23-0} = func; + let DecoderMethod = "DecodeBranchImmInstruction"; } // ARMv5T and above @@ -1516,6 +1950,7 @@ let isBranch = 1, isTerminator = 1 in { [/*(ARMbrcond bb:$target, imm:$cc, CCR:$ccr)*/]> { bits<24> target; let Inst{23-0} = target; + let DecoderMethod = "DecodeBranchImmInstruction"; } let isBarrier = 1 in { @@ -1549,9 +1984,9 @@ let isBranch = 1, isTerminator = 1 in { } -// BLX (immediate) -- for disassembly only -def BLXi : AXI<(outs), (ins br_target:$target), BrMiscFrm, NoItinerary, - "blx\t$target", [/* pattern left blank */]>, +// BLX (immediate) +def BLXi : AXI<(outs), (ins blx_target:$target), BrMiscFrm, NoItinerary, + "blx\t$target", []>, Requires<[IsARM, HasV5T]> { let Inst{31-25} = 0b1111101; bits<25> target; @@ -1614,64 +2049,100 @@ let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1 in { } } - - - - -// Secure Monitor Call is a system instruction -- for disassembly only -def SMC : ABI<0b0001, (outs), (ins i32imm:$opt), NoItinerary, "smc", "\t$opt", - [/* For disassembly only; pattern left blank */]> { +// Secure Monitor Call is a system instruction. +def SMC : ABI<0b0001, (outs), (ins imm0_15:$opt), NoItinerary, "smc", "\t$opt", + []> { bits<4> opt; let Inst{23-4} = 0b01100000000000000111; let Inst{3-0} = opt; } -// Supervisor Call (Software Interrupt) -- for disassembly only +// Supervisor Call (Software Interrupt) let isCall = 1, Uses = [SP] in { -def SVC : ABI<0b1111, (outs), (ins i32imm:$svc), IIC_Br, "svc", "\t$svc", - [/* For disassembly only; pattern left blank */]> { +def SVC : ABI<0b1111, (outs), (ins imm24b:$svc), IIC_Br, "svc", "\t$svc", []> { bits<24> svc; let Inst{23-0} = svc; } } -// Store Return State is a system instruction -- for disassembly only -let isCodeGenOnly = 1 in { // FIXME: This should not use submode! -def SRSW : ABXI<{1,0,0,?}, (outs), (ins ldstm_mode:$amode, i32imm:$mode), - NoItinerary, "srs${amode}\tsp!, $mode", - [/* For disassembly only; pattern left blank */]> { +// Store Return State +class SRSI + : XI<(outs), (ins imm0_31:$mode), AddrModeNone, 4, IndexModeNone, BrFrm, + NoItinerary, asm, "", []> { + bits<5> mode; let Inst{31-28} = 0b1111; - let Inst{22-20} = 0b110; // W = 1 - let Inst{19-8} = 0xd05; - let Inst{7-5} = 0b000; + let Inst{27-25} = 0b100; + let Inst{22} = 1; + let Inst{21} = wb; + let Inst{20} = 0; + let Inst{19-16} = 0b1101; // SP + let Inst{15-5} = 0b00000101000; + let Inst{4-0} = mode; } -def SRS : ABXI<{1,0,0,?}, (outs), (ins ldstm_mode:$amode, i32imm:$mode), - NoItinerary, "srs${amode}\tsp, $mode", - [/* For disassembly only; pattern left blank */]> { - let Inst{31-28} = 0b1111; - let Inst{22-20} = 0b100; // W = 0 - let Inst{19-8} = 0xd05; - let Inst{7-5} = 0b000; +def SRSDA : SRSI<0, "srsda\tsp, $mode"> { + let Inst{24-23} = 0; +} +def SRSDA_UPD : SRSI<1, "srsda\tsp!, $mode"> { + let Inst{24-23} = 0; +} +def SRSDB : SRSI<0, "srsdb\tsp, $mode"> { + let Inst{24-23} = 0b10; +} +def SRSDB_UPD : SRSI<1, "srsdb\tsp!, $mode"> { + let Inst{24-23} = 0b10; +} +def SRSIA : SRSI<0, "srsia\tsp, $mode"> { + let Inst{24-23} = 0b01; +} +def SRSIA_UPD : SRSI<1, "srsia\tsp!, $mode"> { + let Inst{24-23} = 0b01; +} +def SRSIB : SRSI<0, "srsib\tsp, $mode"> { + let Inst{24-23} = 0b11; +} +def SRSIB_UPD : SRSI<1, "srsib\tsp!, $mode"> { + let Inst{24-23} = 0b11; } -// Return From Exception is a system instruction -- for disassembly only -def RFEW : ABXI<{1,0,0,?}, (outs), (ins ldstm_mode:$amode, GPR:$base), - NoItinerary, "rfe${amode}\t$base!", - [/* For disassembly only; pattern left blank */]> { +// Return From Exception +class RFEI + : XI<(outs), (ins GPR:$Rn), AddrModeNone, 4, IndexModeNone, BrFrm, + NoItinerary, asm, "", []> { + bits<4> Rn; let Inst{31-28} = 0b1111; - let Inst{22-20} = 0b011; // W = 1 - let Inst{15-0} = 0x0a00; + let Inst{27-25} = 0b100; + let Inst{22} = 0; + let Inst{21} = wb; + let Inst{20} = 1; + let Inst{19-16} = Rn; + let Inst{15-0} = 0xa00; } -def RFE : ABXI<{1,0,0,?}, (outs), (ins ldstm_mode:$amode, GPR:$base), - NoItinerary, "rfe${amode}\t$base", - [/* For disassembly only; pattern left blank */]> { - let Inst{31-28} = 0b1111; - let Inst{22-20} = 0b001; // W = 0 - let Inst{15-0} = 0x0a00; +def RFEDA : RFEI<0, "rfeda\t$Rn"> { + let Inst{24-23} = 0; +} +def RFEDA_UPD : RFEI<1, "rfeda\t$Rn!"> { + let Inst{24-23} = 0; +} +def RFEDB : RFEI<0, "rfedb\t$Rn"> { + let Inst{24-23} = 0b10; +} +def RFEDB_UPD : RFEI<1, "rfedb\t$Rn!"> { + let Inst{24-23} = 0b10; +} +def RFEIA : RFEI<0, "rfeia\t$Rn"> { + let Inst{24-23} = 0b01; +} +def RFEIA_UPD : RFEI<1, "rfeia\t$Rn!"> { + let Inst{24-23} = 0b01; +} +def RFEIB : RFEI<0, "rfeib\t$Rn"> { + let Inst{24-23} = 0b11; +} +def RFEIB_UPD : RFEI<1, "rfeib\t$Rn!"> { + let Inst{24-23} = 0b11; } -} // isCodeGenOnly = 1 //===----------------------------------------------------------------------===// // Load / store Instructions. @@ -1682,16 +2153,16 @@ def RFE : ABXI<{1,0,0,?}, (outs), (ins ldstm_mode:$amode, GPR:$base), defm LDR : AI_ldr1<0, "ldr", IIC_iLoad_r, IIC_iLoad_si, UnOpFrag<(load node:$Src)>>; -defm LDRB : AI_ldr1<1, "ldrb", IIC_iLoad_bh_r, IIC_iLoad_bh_si, +defm LDRB : AI_ldr1nopc<1, "ldrb", IIC_iLoad_bh_r, IIC_iLoad_bh_si, UnOpFrag<(zextloadi8 node:$Src)>>; defm STR : AI_str1<0, "str", IIC_iStore_r, IIC_iStore_si, BinOpFrag<(store node:$LHS, node:$RHS)>>; -defm STRB : AI_str1<1, "strb", IIC_iStore_bh_r, IIC_iStore_bh_si, +defm STRB : AI_str1nopc<1, "strb", IIC_iStore_bh_r, IIC_iStore_bh_si, BinOpFrag<(truncstorei8 node:$LHS, node:$RHS)>>; // Special LDR for loads from non-pc-relative constpools. let canFoldAsLoad = 1, mayLoad = 1, neverHasSideEffects = 1, - isReMaterializable = 1 in + isReMaterializable = 1, isCodeGenOnly = 1 in def LDRcp : AI2ldst<0b010, 1, 0, (outs GPR:$Rt), (ins addrmode_imm12:$addr), AddrMode_i12, LdFrm, IIC_iLoad_r, "ldr", "\t$Rt, $addr", []> { @@ -1725,36 +2196,67 @@ def LDRD : AI3ld<0b1101, 0, (outs GPR:$Rd, GPR:$dst2), []>, Requires<[IsARM, HasV5TE]>; } -// Indexed loads -multiclass AI2_ldridx { - def _PRE : AI2ldstidx<1, isByte, 1, (outs GPR:$Rt, GPR:$Rn_wb), - (ins addrmode2:$addr), IndexModePre, LdFrm, itin, +// Indexed loads +multiclass AI2_ldridx { + def _PRE_IMM : AI2ldstidx<1, isByte, 1, (outs GPR:$Rt, GPR:$Rn_wb), + (ins addrmode_imm12:$addr), IndexModePre, LdFrm, itin, + opc, "\t$Rt, $addr!", "$addr.base = $Rn_wb", []> { + bits<17> addr; + let Inst{25} = 0; + let Inst{23} = addr{12}; + let Inst{19-16} = addr{16-13}; + let Inst{11-0} = addr{11-0}; + let DecoderMethod = "DecodeLDRPreImm"; + let AsmMatchConverter = "cvtLdWriteBackRegAddrModeImm12"; + } + + def _PRE_REG : AI2ldstidx<1, isByte, 1, (outs GPR:$Rt, GPR:$Rn_wb), + (ins ldst_so_reg:$addr), IndexModePre, LdFrm, itin, opc, "\t$Rt, $addr!", "$addr.base = $Rn_wb", []> { - // {17-14} Rn - // {13} 1 == Rm, 0 == imm12 - // {12} isAdd - // {11-0} imm12/Rm - bits<18> addr; - let Inst{25} = addr{13}; + bits<17> addr; + let Inst{25} = 1; let Inst{23} = addr{12}; - let Inst{19-16} = addr{17-14}; + let Inst{19-16} = addr{16-13}; let Inst{11-0} = addr{11-0}; - let AsmMatchConverter = "CvtLdWriteBackRegAddrMode2"; + let Inst{4} = 0; + let DecoderMethod = "DecodeLDRPreReg"; + let AsmMatchConverter = "cvtLdWriteBackRegAddrMode2"; } - def _POST : AI2ldstidx<1, isByte, 0, (outs GPR:$Rt, GPR:$Rn_wb), - (ins GPR:$Rn, am2offset:$offset), + + def _POST_REG : AI2ldstidx<1, isByte, 0, (outs GPR:$Rt, GPR:$Rn_wb), + (ins addr_offset_none:$addr, am2offset_reg:$offset), + IndexModePost, LdFrm, itin, + opc, "\t$Rt, $addr, $offset", + "$addr.base = $Rn_wb", []> { + // {12} isAdd + // {11-0} imm12/Rm + bits<14> offset; + bits<4> addr; + let Inst{25} = 1; + let Inst{23} = offset{12}; + let Inst{19-16} = addr; + let Inst{11-0} = offset{11-0}; + + let DecoderMethod = "DecodeAddrMode2IdxInstruction"; + } + + def _POST_IMM : AI2ldstidx<1, isByte, 0, (outs GPR:$Rt, GPR:$Rn_wb), + (ins addr_offset_none:$addr, am2offset_imm:$offset), IndexModePost, LdFrm, itin, - opc, "\t$Rt, [$Rn], $offset", "$Rn = $Rn_wb", []> { - // {13} 1 == Rm, 0 == imm12 + opc, "\t$Rt, $addr, $offset", + "$addr.base = $Rn_wb", []> { // {12} isAdd // {11-0} imm12/Rm bits<14> offset; - bits<4> Rn; - let Inst{25} = offset{13}; + bits<4> addr; + let Inst{25} = 0; let Inst{23} = offset{12}; - let Inst{19-16} = Rn; + let Inst{19-16} = addr; let Inst{11-0} = offset{11-0}; + + let DecoderMethod = "DecodeAddrMode2IdxInstruction"; } + } let mayLoad = 1, neverHasSideEffects = 1 in { @@ -1762,8 +2264,8 @@ defm LDR : AI2_ldridx<0, "ldr", IIC_iLoad_ru>; defm LDRB : AI2_ldridx<1, "ldrb", IIC_iLoad_bh_ru>; } -multiclass AI3_ldridx op, bit op20, string opc, InstrItinClass itin> { - def _PRE : AI3ldstidx op, string opc, InstrItinClass itin> { + def _PRE : AI3ldstidx { @@ -1773,27 +2275,31 @@ multiclass AI3_ldridx op, bit op20, string opc, InstrItinClass itin> { let Inst{19-16} = addr{12-9}; // Rn let Inst{11-8} = addr{7-4}; // imm7_4/zero let Inst{3-0} = addr{3-0}; // imm3_0/Rm + let AsmMatchConverter = "cvtLdWriteBackRegAddrMode3"; + let DecoderMethod = "DecodeAddrMode3Instruction"; } - def _POST : AI3ldstidx { + def _POST : AI3ldstidx { bits<10> offset; - bits<4> Rn; + bits<4> addr; let Inst{23} = offset{8}; // U bit let Inst{22} = offset{9}; // 1 == imm8, 0 == Rm - let Inst{19-16} = Rn; + let Inst{19-16} = addr; let Inst{11-8} = offset{7-4}; // imm7_4/zero let Inst{3-0} = offset{3-0}; // imm3_0/Rm + let DecoderMethod = "DecodeAddrMode3Instruction"; } } let mayLoad = 1, neverHasSideEffects = 1 in { -defm LDRH : AI3_ldridx<0b1011, 1, "ldrh", IIC_iLoad_bh_ru>; -defm LDRSH : AI3_ldridx<0b1111, 1, "ldrsh", IIC_iLoad_bh_ru>; -defm LDRSB : AI3_ldridx<0b1101, 1, "ldrsb", IIC_iLoad_bh_ru>; +defm LDRH : AI3_ldridx<0b1011, "ldrh", IIC_iLoad_bh_ru>; +defm LDRSH : AI3_ldridx<0b1111, "ldrsh", IIC_iLoad_bh_ru>; +defm LDRSB : AI3_ldridx<0b1101, "ldrsb", IIC_iLoad_bh_ru>; let hasExtraDefRegAllocReq = 1 in { -def LDRD_PRE : AI3ldstidx<0b1101, 0, 1, 1, (outs GPR:$Rt, GPR:$Rt2, GPR:$Rn_wb), +def LDRD_PRE : AI3ldstidx<0b1101, 0, 1, (outs GPR:$Rt, GPR:$Rt2, GPR:$Rn_wb), (ins addrmode3:$addr), IndexModePre, LdMiscFrm, IIC_iLoad_d_ru, "ldrd", "\t$Rt, $Rt2, $addr!", @@ -1804,70 +2310,128 @@ def LDRD_PRE : AI3ldstidx<0b1101, 0, 1, 1, (outs GPR:$Rt, GPR:$Rt2, GPR:$Rn_wb), let Inst{19-16} = addr{12-9}; // Rn let Inst{11-8} = addr{7-4}; // imm7_4/zero let Inst{3-0} = addr{3-0}; // imm3_0/Rm + let DecoderMethod = "DecodeAddrMode3Instruction"; + let AsmMatchConverter = "cvtLdrdPre"; } -def LDRD_POST: AI3ldstidx<0b1101, 0, 1, 0, (outs GPR:$Rt, GPR:$Rt2, GPR:$Rn_wb), - (ins GPR:$Rn, am3offset:$offset), IndexModePost, - LdMiscFrm, IIC_iLoad_d_ru, - "ldrd", "\t$Rt, $Rt2, [$Rn], $offset", - "$Rn = $Rn_wb", []> { +def LDRD_POST: AI3ldstidx<0b1101, 0, 0, (outs GPR:$Rt, GPR:$Rt2, GPR:$Rn_wb), + (ins addr_offset_none:$addr, am3offset:$offset), + IndexModePost, LdMiscFrm, IIC_iLoad_d_ru, + "ldrd", "\t$Rt, $Rt2, $addr, $offset", + "$addr.base = $Rn_wb", []> { bits<10> offset; - bits<4> Rn; + bits<4> addr; let Inst{23} = offset{8}; // U bit let Inst{22} = offset{9}; // 1 == imm8, 0 == Rm - let Inst{19-16} = Rn; + let Inst{19-16} = addr; let Inst{11-8} = offset{7-4}; // imm7_4/zero let Inst{3-0} = offset{3-0}; // imm3_0/Rm + let DecoderMethod = "DecodeAddrMode3Instruction"; } } // hasExtraDefRegAllocReq = 1 } // mayLoad = 1, neverHasSideEffects = 1 -// LDRT, LDRBT, LDRSBT, LDRHT, LDRSHT are for disassembly only. +// LDRT, LDRBT, LDRSBT, LDRHT, LDRSHT. let mayLoad = 1, neverHasSideEffects = 1 in { -def LDRT : AI2ldstidx<1, 0, 0, (outs GPR:$Rt, GPR:$base_wb), - (ins addrmode2:$addr), IndexModePost, LdFrm, IIC_iLoad_ru, - "ldrt", "\t$Rt, $addr", "$addr.base = $base_wb", []> { - // {17-14} Rn - // {13} 1 == Rm, 0 == imm12 +def LDRT_POST_REG : AI2ldstidx<1, 0, 0, (outs GPR:$Rt, GPR:$Rn_wb), + (ins addr_offset_none:$addr, am2offset_reg:$offset), + IndexModePost, LdFrm, IIC_iLoad_ru, + "ldrt", "\t$Rt, $addr, $offset", + "$addr.base = $Rn_wb", []> { // {12} isAdd // {11-0} imm12/Rm - bits<18> addr; - let Inst{25} = addr{13}; - let Inst{23} = addr{12}; + bits<14> offset; + bits<4> addr; + let Inst{25} = 1; + let Inst{23} = offset{12}; let Inst{21} = 1; // overwrite - let Inst{19-16} = addr{17-14}; - let Inst{11-0} = addr{11-0}; - let AsmMatchConverter = "CvtLdWriteBackRegAddrMode2"; -} -def LDRBT : AI2ldstidx<1, 1, 0, (outs GPR:$Rt, GPR:$base_wb), - (ins addrmode2:$addr), IndexModePost, LdFrm, IIC_iLoad_bh_ru, - "ldrbt", "\t$Rt, $addr", "$addr.base = $base_wb", []> { - // {17-14} Rn - // {13} 1 == Rm, 0 == imm12 + let Inst{19-16} = addr; + let Inst{11-5} = offset{11-5}; + let Inst{4} = 0; + let Inst{3-0} = offset{3-0}; + let DecoderMethod = "DecodeAddrMode2IdxInstruction"; +} + +def LDRT_POST_IMM : AI2ldstidx<1, 0, 0, (outs GPR:$Rt, GPR:$Rn_wb), + (ins addr_offset_none:$addr, am2offset_imm:$offset), + IndexModePost, LdFrm, IIC_iLoad_ru, + "ldrt", "\t$Rt, $addr, $offset", + "$addr.base = $Rn_wb", []> { // {12} isAdd // {11-0} imm12/Rm - bits<18> addr; - let Inst{25} = addr{13}; - let Inst{23} = addr{12}; - let Inst{21} = 1; // overwrite - let Inst{19-16} = addr{17-14}; - let Inst{11-0} = addr{11-0}; - let AsmMatchConverter = "CvtLdWriteBackRegAddrMode2"; -} -def LDRSBT : AI3ldstidxT<0b1101, 1, 1, 0, (outs GPR:$Rt, GPR:$base_wb), - (ins addrmode3:$addr), IndexModePost, LdMiscFrm, IIC_iLoad_bh_ru, - "ldrsbt", "\t$Rt, $addr", "$addr.base = $base_wb", []> { + bits<14> offset; + bits<4> addr; + let Inst{25} = 0; + let Inst{23} = offset{12}; let Inst{21} = 1; // overwrite + let Inst{19-16} = addr; + let Inst{11-0} = offset{11-0}; + let DecoderMethod = "DecodeAddrMode2IdxInstruction"; } -def LDRHT : AI3ldstidxT<0b1011, 1, 1, 0, (outs GPR:$Rt, GPR:$base_wb), - (ins addrmode3:$addr), IndexModePost, LdMiscFrm, IIC_iLoad_bh_ru, - "ldrht", "\t$Rt, $addr", "$addr.base = $base_wb", []> { + +def LDRBT_POST_REG : AI2ldstidx<1, 1, 0, (outs GPR:$Rt, GPR:$Rn_wb), + (ins addr_offset_none:$addr, am2offset_reg:$offset), + IndexModePost, LdFrm, IIC_iLoad_bh_ru, + "ldrbt", "\t$Rt, $addr, $offset", + "$addr.base = $Rn_wb", []> { + // {12} isAdd + // {11-0} imm12/Rm + bits<14> offset; + bits<4> addr; + let Inst{25} = 1; + let Inst{23} = offset{12}; let Inst{21} = 1; // overwrite -} -def LDRSHT : AI3ldstidxT<0b1111, 1, 1, 0, (outs GPR:$Rt, GPR:$base_wb), - (ins addrmode3:$addr), IndexModePost, LdMiscFrm, IIC_iLoad_bh_ru, - "ldrsht", "\t$Rt, $addr", "$addr.base = $base_wb", []> { + let Inst{19-16} = addr; + let Inst{11-5} = offset{11-5}; + let Inst{4} = 0; + let Inst{3-0} = offset{3-0}; + let DecoderMethod = "DecodeAddrMode2IdxInstruction"; +} + +def LDRBT_POST_IMM : AI2ldstidx<1, 1, 0, (outs GPR:$Rt, GPR:$Rn_wb), + (ins addr_offset_none:$addr, am2offset_imm:$offset), + IndexModePost, LdFrm, IIC_iLoad_bh_ru, + "ldrbt", "\t$Rt, $addr, $offset", + "$addr.base = $Rn_wb", []> { + // {12} isAdd + // {11-0} imm12/Rm + bits<14> offset; + bits<4> addr; + let Inst{25} = 0; + let Inst{23} = offset{12}; let Inst{21} = 1; // overwrite + let Inst{19-16} = addr; + let Inst{11-0} = offset{11-0}; + let DecoderMethod = "DecodeAddrMode2IdxInstruction"; +} + +multiclass AI3ldrT op, string opc> { + def i : AI3ldstidxT { + bits<9> offset; + let Inst{23} = offset{8}; + let Inst{22} = 1; + let Inst{11-8} = offset{7-4}; + let Inst{3-0} = offset{3-0}; + let AsmMatchConverter = "cvtLdExtTWriteBackImm"; + } + def r : AI3ldstidxT { + bits<5> Rm; + let Inst{23} = Rm{4}; + let Inst{22} = 0; + let Inst{11-8} = 0; + let Inst{3-0} = Rm{3-0}; + let AsmMatchConverter = "cvtLdExtTWriteBackReg"; + } } + +defm LDRSBT : AI3ldrT<0b1101, "ldrsbt">; +defm LDRHT : AI3ldrT<0b1011, "ldrht">; +defm LDRSHT : AI3ldrT<0b1111, "ldrsht">; } // Store @@ -1881,98 +2445,302 @@ def STRH : AI3str<0b1011, (outs), (ins GPR:$Rt, addrmode3:$addr), StMiscFrm, let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in def STRD : AI3str<0b1111, (outs), (ins GPR:$Rt, GPR:$src2, addrmode3:$addr), StMiscFrm, IIC_iStore_d_r, - "strd", "\t$Rt, $src2, $addr", []>, Requires<[IsARM, HasV5TE]>; + "strd", "\t$Rt, $src2, $addr", []>, + Requires<[IsARM, HasV5TE]> { + let Inst{21} = 0; +} // Indexed stores -def STR_PRE : AI2stridx<0, 1, (outs GPR:$Rn_wb), - (ins GPR:$Rt, GPR:$Rn, am2offset:$offset), - IndexModePre, StFrm, IIC_iStore_ru, - "str", "\t$Rt, [$Rn, $offset]!", - "$Rn = $Rn_wb,@earlyclobber $Rn_wb", - [(set GPR:$Rn_wb, - (pre_store GPR:$Rt, GPR:$Rn, am2offset:$offset))]>; - -def STR_POST : AI2stridx<0, 0, (outs GPR:$Rn_wb), - (ins GPR:$Rt, GPR:$Rn, am2offset:$offset), - IndexModePost, StFrm, IIC_iStore_ru, - "str", "\t$Rt, [$Rn], $offset", - "$Rn = $Rn_wb,@earlyclobber $Rn_wb", - [(set GPR:$Rn_wb, - (post_store GPR:$Rt, GPR:$Rn, am2offset:$offset))]>; - -def STRB_PRE : AI2stridx<1, 1, (outs GPR:$Rn_wb), - (ins GPR:$Rt, GPR:$Rn, am2offset:$offset), - IndexModePre, StFrm, IIC_iStore_bh_ru, - "strb", "\t$Rt, [$Rn, $offset]!", - "$Rn = $Rn_wb,@earlyclobber $Rn_wb", - [(set GPR:$Rn_wb, (pre_truncsti8 GPR:$Rt, - GPR:$Rn, am2offset:$offset))]>; -def STRB_POST: AI2stridx<1, 0, (outs GPR:$Rn_wb), - (ins GPR:$Rt, GPR:$Rn, am2offset:$offset), - IndexModePost, StFrm, IIC_iStore_bh_ru, - "strb", "\t$Rt, [$Rn], $offset", - "$Rn = $Rn_wb,@earlyclobber $Rn_wb", - [(set GPR:$Rn_wb, (post_truncsti8 GPR:$Rt, - GPR:$Rn, am2offset:$offset))]>; - -def STRH_PRE : AI3stridx<0b1011, 0, 1, (outs GPR:$Rn_wb), - (ins GPR:$Rt, GPR:$Rn, am3offset:$offset), - IndexModePre, StMiscFrm, IIC_iStore_ru, - "strh", "\t$Rt, [$Rn, $offset]!", - "$Rn = $Rn_wb,@earlyclobber $Rn_wb", - [(set GPR:$Rn_wb, - (pre_truncsti16 GPR:$Rt, GPR:$Rn, am3offset:$offset))]>; - -def STRH_POST: AI3stridx<0b1011, 0, 0, (outs GPR:$Rn_wb), - (ins GPR:$Rt, GPR:$Rn, am3offset:$offset), - IndexModePost, StMiscFrm, IIC_iStore_bh_ru, - "strh", "\t$Rt, [$Rn], $offset", - "$Rn = $Rn_wb,@earlyclobber $Rn_wb", - [(set GPR:$Rn_wb, (post_truncsti16 GPR:$Rt, - GPR:$Rn, am3offset:$offset))]>; - -// For disassembly only +multiclass AI2_stridx { + def _PRE_IMM : AI2ldstidx<0, isByte, 1, (outs GPR:$Rn_wb), + (ins GPR:$Rt, addrmode_imm12:$addr), IndexModePre, + StFrm, itin, + opc, "\t$Rt, $addr!", "$addr.base = $Rn_wb", []> { + bits<17> addr; + let Inst{25} = 0; + let Inst{23} = addr{12}; // U (add = ('U' == 1)) + let Inst{19-16} = addr{16-13}; // Rn + let Inst{11-0} = addr{11-0}; // imm12 + let AsmMatchConverter = "cvtStWriteBackRegAddrModeImm12"; + let DecoderMethod = "DecodeSTRPreImm"; + } + + def _PRE_REG : AI2ldstidx<0, isByte, 1, (outs GPR:$Rn_wb), + (ins GPR:$Rt, ldst_so_reg:$addr), + IndexModePre, StFrm, itin, + opc, "\t$Rt, $addr!", "$addr.base = $Rn_wb", []> { + bits<17> addr; + let Inst{25} = 1; + let Inst{23} = addr{12}; // U (add = ('U' == 1)) + let Inst{19-16} = addr{16-13}; // Rn + let Inst{11-0} = addr{11-0}; + let Inst{4} = 0; // Inst{4} = 0 + let AsmMatchConverter = "cvtStWriteBackRegAddrMode2"; + let DecoderMethod = "DecodeSTRPreReg"; + } + def _POST_REG : AI2ldstidx<0, isByte, 0, (outs GPR:$Rn_wb), + (ins GPR:$Rt, addr_offset_none:$addr, am2offset_reg:$offset), + IndexModePost, StFrm, itin, + opc, "\t$Rt, $addr, $offset", + "$addr.base = $Rn_wb", []> { + // {12} isAdd + // {11-0} imm12/Rm + bits<14> offset; + bits<4> addr; + let Inst{25} = 1; + let Inst{23} = offset{12}; + let Inst{19-16} = addr; + let Inst{11-0} = offset{11-0}; + + let DecoderMethod = "DecodeAddrMode2IdxInstruction"; + } + + def _POST_IMM : AI2ldstidx<0, isByte, 0, (outs GPR:$Rn_wb), + (ins GPR:$Rt, addr_offset_none:$addr, am2offset_imm:$offset), + IndexModePost, StFrm, itin, + opc, "\t$Rt, $addr, $offset", + "$addr.base = $Rn_wb", []> { + // {12} isAdd + // {11-0} imm12/Rm + bits<14> offset; + bits<4> addr; + let Inst{25} = 0; + let Inst{23} = offset{12}; + let Inst{19-16} = addr; + let Inst{11-0} = offset{11-0}; + + let DecoderMethod = "DecodeAddrMode2IdxInstruction"; + } +} + +let mayStore = 1, neverHasSideEffects = 1 in { +defm STR : AI2_stridx<0, "str", IIC_iStore_ru>; +defm STRB : AI2_stridx<1, "strb", IIC_iStore_bh_ru>; +} + +def : ARMPat<(post_store GPR:$Rt, addr_offset_none:$addr, + am2offset_reg:$offset), + (STR_POST_REG GPR:$Rt, addr_offset_none:$addr, + am2offset_reg:$offset)>; +def : ARMPat<(post_store GPR:$Rt, addr_offset_none:$addr, + am2offset_imm:$offset), + (STR_POST_IMM GPR:$Rt, addr_offset_none:$addr, + am2offset_imm:$offset)>; +def : ARMPat<(post_truncsti8 GPR:$Rt, addr_offset_none:$addr, + am2offset_reg:$offset), + (STRB_POST_REG GPR:$Rt, addr_offset_none:$addr, + am2offset_reg:$offset)>; +def : ARMPat<(post_truncsti8 GPR:$Rt, addr_offset_none:$addr, + am2offset_imm:$offset), + (STRB_POST_IMM GPR:$Rt, addr_offset_none:$addr, + am2offset_imm:$offset)>; + +// Pseudo-instructions for pattern matching the pre-indexed stores. We can't +// put the patterns on the instruction definitions directly as ISel wants +// the address base and offset to be separate operands, not a single +// complex operand like we represent the instructions themselves. The +// pseudos map between the two. +let usesCustomInserter = 1, + Constraints = "$Rn = $Rn_wb,@earlyclobber $Rn_wb" in { +def STRi_preidx: ARMPseudoInst<(outs GPR:$Rn_wb), + (ins GPR:$Rt, GPR:$Rn, am2offset_imm:$offset, pred:$p), + 4, IIC_iStore_ru, + [(set GPR:$Rn_wb, + (pre_store GPR:$Rt, GPR:$Rn, am2offset_imm:$offset))]>; +def STRr_preidx: ARMPseudoInst<(outs GPR:$Rn_wb), + (ins GPR:$Rt, GPR:$Rn, am2offset_reg:$offset, pred:$p), + 4, IIC_iStore_ru, + [(set GPR:$Rn_wb, + (pre_store GPR:$Rt, GPR:$Rn, am2offset_reg:$offset))]>; +def STRBi_preidx: ARMPseudoInst<(outs GPR:$Rn_wb), + (ins GPR:$Rt, GPR:$Rn, am2offset_imm:$offset, pred:$p), + 4, IIC_iStore_ru, + [(set GPR:$Rn_wb, + (pre_truncsti8 GPR:$Rt, GPR:$Rn, am2offset_imm:$offset))]>; +def STRBr_preidx: ARMPseudoInst<(outs GPR:$Rn_wb), + (ins GPR:$Rt, GPR:$Rn, am2offset_reg:$offset, pred:$p), + 4, IIC_iStore_ru, + [(set GPR:$Rn_wb, + (pre_truncsti8 GPR:$Rt, GPR:$Rn, am2offset_reg:$offset))]>; +def STRH_preidx: ARMPseudoInst<(outs GPR:$Rn_wb), + (ins GPR:$Rt, GPR:$Rn, am3offset:$offset, pred:$p), + 4, IIC_iStore_ru, + [(set GPR:$Rn_wb, + (pre_truncsti16 GPR:$Rt, GPR:$Rn, am3offset:$offset))]>; +} + + + +def STRH_PRE : AI3ldstidx<0b1011, 0, 1, (outs GPR:$Rn_wb), + (ins GPR:$Rt, addrmode3:$addr), IndexModePre, + StMiscFrm, IIC_iStore_bh_ru, + "strh", "\t$Rt, $addr!", "$addr.base = $Rn_wb", []> { + bits<14> addr; + let Inst{23} = addr{8}; // U bit + let Inst{22} = addr{13}; // 1 == imm8, 0 == Rm + let Inst{19-16} = addr{12-9}; // Rn + let Inst{11-8} = addr{7-4}; // imm7_4/zero + let Inst{3-0} = addr{3-0}; // imm3_0/Rm + let AsmMatchConverter = "cvtStWriteBackRegAddrMode3"; + let DecoderMethod = "DecodeAddrMode3Instruction"; +} + +def STRH_POST : AI3ldstidx<0b1011, 0, 0, (outs GPR:$Rn_wb), + (ins GPR:$Rt, addr_offset_none:$addr, am3offset:$offset), + IndexModePost, StMiscFrm, IIC_iStore_bh_ru, + "strh", "\t$Rt, $addr, $offset", "$addr.base = $Rn_wb", + [(set GPR:$Rn_wb, (post_truncsti16 GPR:$Rt, + addr_offset_none:$addr, + am3offset:$offset))]> { + bits<10> offset; + bits<4> addr; + let Inst{23} = offset{8}; // U bit + let Inst{22} = offset{9}; // 1 == imm8, 0 == Rm + let Inst{19-16} = addr; + let Inst{11-8} = offset{7-4}; // imm7_4/zero + let Inst{3-0} = offset{3-0}; // imm3_0/Rm + let DecoderMethod = "DecodeAddrMode3Instruction"; +} + let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in { -def STRD_PRE : AI3stdpr<(outs GPR:$base_wb), - (ins GPR:$src1, GPR:$src2, GPR:$base, am3offset:$offset), - StMiscFrm, IIC_iStore_d_ru, - "strd", "\t$src1, $src2, [$base, $offset]!", - "$base = $base_wb", []>; - -// For disassembly only -def STRD_POST: AI3stdpo<(outs GPR:$base_wb), - (ins GPR:$src1, GPR:$src2, GPR:$base, am3offset:$offset), - StMiscFrm, IIC_iStore_d_ru, - "strd", "\t$src1, $src2, [$base], $offset", - "$base = $base_wb", []>; +def STRD_PRE : AI3ldstidx<0b1111, 0, 1, (outs GPR:$Rn_wb), + (ins GPR:$Rt, GPR:$Rt2, addrmode3:$addr), + IndexModePre, StMiscFrm, IIC_iStore_d_ru, + "strd", "\t$Rt, $Rt2, $addr!", + "$addr.base = $Rn_wb", []> { + bits<14> addr; + let Inst{23} = addr{8}; // U bit + let Inst{22} = addr{13}; // 1 == imm8, 0 == Rm + let Inst{19-16} = addr{12-9}; // Rn + let Inst{11-8} = addr{7-4}; // imm7_4/zero + let Inst{3-0} = addr{3-0}; // imm3_0/Rm + let DecoderMethod = "DecodeAddrMode3Instruction"; + let AsmMatchConverter = "cvtStrdPre"; +} + +def STRD_POST: AI3ldstidx<0b1111, 0, 0, (outs GPR:$Rn_wb), + (ins GPR:$Rt, GPR:$Rt2, addr_offset_none:$addr, + am3offset:$offset), + IndexModePost, StMiscFrm, IIC_iStore_d_ru, + "strd", "\t$Rt, $Rt2, $addr, $offset", + "$addr.base = $Rn_wb", []> { + bits<10> offset; + bits<4> addr; + let Inst{23} = offset{8}; // U bit + let Inst{22} = offset{9}; // 1 == imm8, 0 == Rm + let Inst{19-16} = addr; + let Inst{11-8} = offset{7-4}; // imm7_4/zero + let Inst{3-0} = offset{3-0}; // imm3_0/Rm + let DecoderMethod = "DecodeAddrMode3Instruction"; +} } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 -// STRT, STRBT, and STRHT are for disassembly only. +// STRT, STRBT, and STRHT -def STRT : AI2stridxT<0, 0, (outs GPR:$Rn_wb), (ins GPR:$Rt, addrmode2:$addr), - IndexModePost, StFrm, IIC_iStore_ru, - "strt", "\t$Rt, $addr", "$addr.base = $Rn_wb", - [/* For disassembly only; pattern left blank */]> { +def STRBT_POST_REG : AI2ldstidx<0, 1, 0, (outs GPR:$Rn_wb), + (ins GPR:$Rt, addr_offset_none:$addr, am2offset_reg:$offset), + IndexModePost, StFrm, IIC_iStore_bh_ru, + "strbt", "\t$Rt, $addr, $offset", + "$addr.base = $Rn_wb", []> { + // {12} isAdd + // {11-0} imm12/Rm + bits<14> offset; + bits<4> addr; + let Inst{25} = 1; + let Inst{23} = offset{12}; + let Inst{21} = 1; // overwrite + let Inst{19-16} = addr; + let Inst{11-5} = offset{11-5}; + let Inst{4} = 0; + let Inst{3-0} = offset{3-0}; + let DecoderMethod = "DecodeAddrMode2IdxInstruction"; +} + +def STRBT_POST_IMM : AI2ldstidx<0, 1, 0, (outs GPR:$Rn_wb), + (ins GPR:$Rt, addr_offset_none:$addr, am2offset_imm:$offset), + IndexModePost, StFrm, IIC_iStore_bh_ru, + "strbt", "\t$Rt, $addr, $offset", + "$addr.base = $Rn_wb", []> { + // {12} isAdd + // {11-0} imm12/Rm + bits<14> offset; + bits<4> addr; + let Inst{25} = 0; + let Inst{23} = offset{12}; let Inst{21} = 1; // overwrite - let AsmMatchConverter = "CvtStWriteBackRegAddrMode2"; + let Inst{19-16} = addr; + let Inst{11-0} = offset{11-0}; + let DecoderMethod = "DecodeAddrMode2IdxInstruction"; } -def STRBT : AI2stridxT<1, 0, (outs GPR:$Rn_wb), (ins GPR:$Rt, addrmode2:$addr), - IndexModePost, StFrm, IIC_iStore_bh_ru, - "strbt", "\t$Rt, $addr", "$addr.base = $Rn_wb", - [/* For disassembly only; pattern left blank */]> { +let mayStore = 1, neverHasSideEffects = 1 in { +def STRT_POST_REG : AI2ldstidx<0, 0, 0, (outs GPR:$Rn_wb), + (ins GPR:$Rt, addr_offset_none:$addr, am2offset_reg:$offset), + IndexModePost, StFrm, IIC_iStore_ru, + "strt", "\t$Rt, $addr, $offset", + "$addr.base = $Rn_wb", []> { + // {12} isAdd + // {11-0} imm12/Rm + bits<14> offset; + bits<4> addr; + let Inst{25} = 1; + let Inst{23} = offset{12}; + let Inst{21} = 1; // overwrite + let Inst{19-16} = addr; + let Inst{11-5} = offset{11-5}; + let Inst{4} = 0; + let Inst{3-0} = offset{3-0}; + let DecoderMethod = "DecodeAddrMode2IdxInstruction"; +} + +def STRT_POST_IMM : AI2ldstidx<0, 0, 0, (outs GPR:$Rn_wb), + (ins GPR:$Rt, addr_offset_none:$addr, am2offset_imm:$offset), + IndexModePost, StFrm, IIC_iStore_ru, + "strt", "\t$Rt, $addr, $offset", + "$addr.base = $Rn_wb", []> { + // {12} isAdd + // {11-0} imm12/Rm + bits<14> offset; + bits<4> addr; + let Inst{25} = 0; + let Inst{23} = offset{12}; let Inst{21} = 1; // overwrite - let AsmMatchConverter = "CvtStWriteBackRegAddrMode2"; + let Inst{19-16} = addr; + let Inst{11-0} = offset{11-0}; + let DecoderMethod = "DecodeAddrMode2IdxInstruction"; +} } -def STRHT: AI3sthpo<(outs GPR:$base_wb), (ins GPR:$Rt, addrmode3:$addr), - StMiscFrm, IIC_iStore_bh_ru, - "strht", "\t$Rt, $addr", "$addr.base = $base_wb", - [/* For disassembly only; pattern left blank */]> { - let Inst{21} = 1; // overwrite - let AsmMatchConverter = "CvtStWriteBackRegAddrMode3"; + +multiclass AI3strT op, string opc> { + def i : AI3ldstidxT { + bits<9> offset; + let Inst{23} = offset{8}; + let Inst{22} = 1; + let Inst{11-8} = offset{7-4}; + let Inst{3-0} = offset{3-0}; + let AsmMatchConverter = "cvtStExtTWriteBackImm"; + } + def r : AI3ldstidxT { + bits<5> Rm; + let Inst{23} = Rm{4}; + let Inst{22} = 0; + let Inst{11-8} = 0; + let Inst{3-0} = Rm{3-0}; + let AsmMatchConverter = "cvtStExtTWriteBackReg"; + } } + +defm STRHT : AI3strT<0b1011, "strht">; + + //===----------------------------------------------------------------------===// // Load / store multiple Instructions. // @@ -1996,6 +2764,8 @@ multiclass arm_ldst_mult; + // A version for the smaller set of tail call registers. let neverHasSideEffects = 1 in def MOVr_TC : AsI1<0b1101, (outs tcGPR:$Rd), (ins tcGPR:$Rm), DPFrm, @@ -2097,15 +2876,33 @@ def MOVr_TC : AsI1<0b1101, (outs tcGPR:$Rd), (ins tcGPR:$Rm), DPFrm, let Inst{15-12} = Rd; } -def MOVs : AsI1<0b1101, (outs GPR:$Rd), (ins shift_so_reg:$src), - DPSoRegFrm, IIC_iMOVsr, - "mov", "\t$Rd, $src", [(set GPR:$Rd, shift_so_reg:$src)]>, +def MOVsr : AsI1<0b1101, (outs GPRnopc:$Rd), (ins shift_so_reg_reg:$src), + DPSoRegRegFrm, IIC_iMOVsr, + "mov", "\t$Rd, $src", + [(set GPRnopc:$Rd, shift_so_reg_reg:$src)]>, UnaryDP { + bits<4> Rd; + bits<12> src; + let Inst{15-12} = Rd; + let Inst{19-16} = 0b0000; + let Inst{11-8} = src{11-8}; + let Inst{7} = 0; + let Inst{6-5} = src{6-5}; + let Inst{4} = 1; + let Inst{3-0} = src{3-0}; + let Inst{25} = 0; +} + +def MOVsi : AsI1<0b1101, (outs GPR:$Rd), (ins shift_so_reg_imm:$src), + DPSoRegImmFrm, IIC_iMOVsr, + "mov", "\t$Rd, $src", [(set GPR:$Rd, shift_so_reg_imm:$src)]>, UnaryDP { bits<4> Rd; bits<12> src; let Inst{15-12} = Rd; let Inst{19-16} = 0b0000; - let Inst{11-0} = src; + let Inst{11-5} = src{11-5}; + let Inst{4} = 0; + let Inst{3-0} = src{3-0}; let Inst{25} = 0; } @@ -2121,7 +2918,7 @@ def MOVi : AsI1<0b1101, (outs GPR:$Rd), (ins so_imm:$imm), DPFrm, IIC_iMOVi, } let isReMaterializable = 1, isAsCheapAsAMove = 1, isMoveImm = 1 in -def MOVi16 : AI1<0b1000, (outs GPR:$Rd), (ins i32imm_hilo16:$imm), +def MOVi16 : AI1<0b1000, (outs GPR:$Rd), (ins imm0_65535_expr:$imm), DPFrm, IIC_iMOVi, "movw", "\t$Rd, $imm", [(set GPR:$Rd, imm0_65535:$imm)]>, @@ -2133,16 +2930,22 @@ def MOVi16 : AI1<0b1000, (outs GPR:$Rd), (ins i32imm_hilo16:$imm), let Inst{19-16} = imm{15-12}; let Inst{20} = 0; let Inst{25} = 1; + let DecoderMethod = "DecodeArmMOVTWInstruction"; } +def : InstAlias<"mov${p} $Rd, $imm", + (MOVi16 GPR:$Rd, imm0_65535_expr:$imm, pred:$p)>, + Requires<[IsARM]>; + def MOVi16_ga_pcrel : PseudoInst<(outs GPR:$Rd), (ins i32imm:$addr, pclabel:$id), IIC_iMOVi, []>; let Constraints = "$src = $Rd" in { -def MOVTi16 : AI1<0b1010, (outs GPR:$Rd), (ins GPR:$src, i32imm_hilo16:$imm), +def MOVTi16 : AI1<0b1010, (outs GPRnopc:$Rd), + (ins GPR:$src, imm0_65535_expr:$imm), DPFrm, IIC_iMOVi, "movt", "\t$Rd, $imm", - [(set GPR:$Rd, + [(set GPRnopc:$Rd, (or (and GPR:$src, 0xffff), lo16AllZero:$imm))]>, UnaryDP, Requires<[IsARM, HasV6T2]> { @@ -2153,6 +2956,7 @@ def MOVTi16 : AI1<0b1010, (outs GPR:$Rd), (ins GPR:$src, i32imm_hilo16:$imm), let Inst{19-16} = imm{15-12}; let Inst{20} = 0; let Inst{25} = 1; + let DecoderMethod = "DecodeArmMOVTWInstruction"; } def MOVTi16_ga_pcrel : PseudoInst<(outs GPR:$Rd), @@ -2186,30 +2990,28 @@ def MOVsra_flag : PseudoInst<(outs GPR:$dst), (ins GPR:$src), IIC_iMOVsi, // Sign extenders -defm SXTB : AI_ext_rrot<0b01101010, +def SXTB : AI_ext_rrot<0b01101010, "sxtb", UnOpFrag<(sext_inreg node:$Src, i8)>>; -defm SXTH : AI_ext_rrot<0b01101011, +def SXTH : AI_ext_rrot<0b01101011, "sxth", UnOpFrag<(sext_inreg node:$Src, i16)>>; -defm SXTAB : AI_exta_rrot<0b01101010, +def SXTAB : AI_exta_rrot<0b01101010, "sxtab", BinOpFrag<(add node:$LHS, (sext_inreg node:$RHS, i8))>>; -defm SXTAH : AI_exta_rrot<0b01101011, +def SXTAH : AI_exta_rrot<0b01101011, "sxtah", BinOpFrag<(add node:$LHS, (sext_inreg node:$RHS,i16))>>; -// For disassembly only -defm SXTB16 : AI_ext_rrot_np<0b01101000, "sxtb16">; +def SXTB16 : AI_ext_rrot_np<0b01101000, "sxtb16">; -// For disassembly only -defm SXTAB16 : AI_exta_rrot_np<0b01101000, "sxtab16">; +def SXTAB16 : AI_exta_rrot_np<0b01101000, "sxtab16">; // Zero extenders let AddedComplexity = 16 in { -defm UXTB : AI_ext_rrot<0b01101110, +def UXTB : AI_ext_rrot<0b01101110, "uxtb" , UnOpFrag<(and node:$Src, 0x000000FF)>>; -defm UXTH : AI_ext_rrot<0b01101111, +def UXTH : AI_ext_rrot<0b01101111, "uxth" , UnOpFrag<(and node:$Src, 0x0000FFFF)>>; -defm UXTB16 : AI_ext_rrot<0b01101100, +def UXTB16 : AI_ext_rrot<0b01101100, "uxtb16", UnOpFrag<(and node:$Src, 0x00FF00FF)>>; // FIXME: This pattern incorrectly assumes the shl operator is a rotate. @@ -2217,23 +3019,22 @@ defm UXTB16 : AI_ext_rrot<0b01101100, // instead so we can include a check for masking back in the upper // eight bits of the source into the lower eight bits of the result. //def : ARMV6Pat<(and (shl GPR:$Src, (i32 8)), 0xFF00FF), -// (UXTB16r_rot GPR:$Src, 24)>; +// (UXTB16r_rot GPR:$Src, 3)>; def : ARMV6Pat<(and (srl GPR:$Src, (i32 8)), 0xFF00FF), - (UXTB16r_rot GPR:$Src, 8)>; + (UXTB16 GPR:$Src, 1)>; -defm UXTAB : AI_exta_rrot<0b01101110, "uxtab", +def UXTAB : AI_exta_rrot<0b01101110, "uxtab", BinOpFrag<(add node:$LHS, (and node:$RHS, 0x00FF))>>; -defm UXTAH : AI_exta_rrot<0b01101111, "uxtah", +def UXTAH : AI_exta_rrot<0b01101111, "uxtah", BinOpFrag<(add node:$LHS, (and node:$RHS, 0xFFFF))>>; } // This isn't safe in general, the add is two 16-bit units, not a 32-bit add. -// For disassembly only -defm UXTAB16 : AI_exta_rrot_np<0b01101100, "uxtab16">; +def UXTAB16 : AI_exta_rrot_np<0b01101100, "uxtab16">; -def SBFX : I<(outs GPR:$Rd), - (ins GPR:$Rn, imm0_31:$lsb, imm0_31_m1:$width), +def SBFX : I<(outs GPRnopc:$Rd), + (ins GPRnopc:$Rn, imm0_31:$lsb, imm1_32:$width), AddrMode1, 4, IndexModeNone, DPFrm, IIC_iUNAsi, "sbfx", "\t$Rd, $Rn, $lsb, $width", "", []>, Requires<[IsARM, HasV6T2]> { @@ -2250,7 +3051,7 @@ def SBFX : I<(outs GPR:$Rd), } def UBFX : I<(outs GPR:$Rd), - (ins GPR:$Rn, imm0_31:$lsb, imm0_31_m1:$width), + (ins GPR:$Rn, imm0_31:$lsb, imm1_32:$width), AddrMode1, 4, IndexModeNone, DPFrm, IIC_iUNAsi, "ubfx", "\t$Rd, $Rn, $lsb, $width", "", []>, Requires<[IsARM, HasV6T2]> { @@ -2278,148 +3079,58 @@ defm SUB : AsI1_bin_irs<0b0010, "sub", BinOpFrag<(sub node:$LHS, node:$RHS)>, "SUB">; // ADD and SUB with 's' bit set. -defm ADDS : AI1_bin_s_irs<0b0100, "adds", +// +// Currently, t2ADDS/t2SUBS are pseudo opcodes that exist only in the +// selection DAG. They are "lowered" to real t2ADD/t2SUB opcodes by +// AdjustInstrPostInstrSelection where we determine whether or not to +// set the "s" bit based on CPSR liveness. +// +// FIXME: Eliminate t2ADDS/t2SUBS pseudo opcodes after adding tablegen +// support for an optional CPSR definition that corresponds to the DAG +// node's second value. We can then eliminate the implicit def of CPSR. +defm ADDS : AsI1_bin_s_irs<0b0100, "add", IIC_iALUi, IIC_iALUr, IIC_iALUsr, - BinOpFrag<(addc node:$LHS, node:$RHS)>, 1>; -defm SUBS : AI1_bin_s_irs<0b0010, "subs", + BinOpFrag<(ARMaddc node:$LHS, node:$RHS)>, 1>; +defm SUBS : AsI1_bin_s_irs<0b0010, "sub", IIC_iALUi, IIC_iALUr, IIC_iALUsr, - BinOpFrag<(subc node:$LHS, node:$RHS)>>; + BinOpFrag<(ARMsubc node:$LHS, node:$RHS)>>; defm ADC : AI1_adde_sube_irs<0b0101, "adc", - BinOpFrag<(adde_dead_carry node:$LHS, node:$RHS)>, + BinOpWithFlagFrag<(ARMadde node:$LHS, node:$RHS, node:$FLAG)>, "ADC", 1>; defm SBC : AI1_adde_sube_irs<0b0110, "sbc", - BinOpFrag<(sube_dead_carry node:$LHS, node:$RHS)>, + BinOpWithFlagFrag<(ARMsube node:$LHS, node:$RHS, node:$FLAG)>, "SBC">; -// ADC and SUBC with 's' bit set. -let usesCustomInserter = 1 in { -defm ADCS : AI1_adde_sube_s_irs< - BinOpFrag<(adde_live_carry node:$LHS, node:$RHS)>, 1>; -defm SBCS : AI1_adde_sube_s_irs< - BinOpFrag<(sube_live_carry node:$LHS, node:$RHS) >>; -} - -def RSBri : AsI1<0b0011, (outs GPR:$Rd), (ins GPR:$Rn, so_imm:$imm), DPFrm, - IIC_iALUi, "rsb", "\t$Rd, $Rn, $imm", - [(set GPR:$Rd, (sub so_imm:$imm, GPR:$Rn))]> { - bits<4> Rd; - bits<4> Rn; - bits<12> imm; - let Inst{25} = 1; - let Inst{15-12} = Rd; - let Inst{19-16} = Rn; - let Inst{11-0} = imm; -} - -// The reg/reg form is only defined for the disassembler; for codegen it is -// equivalent to SUBrr. -def RSBrr : AsI1<0b0011, (outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm), DPFrm, - IIC_iALUr, "rsb", "\t$Rd, $Rn, $Rm", - [/* For disassembly only; pattern left blank */]> { - bits<4> Rd; - bits<4> Rn; - bits<4> Rm; - let Inst{11-4} = 0b00000000; - let Inst{25} = 0; - let Inst{3-0} = Rm; - let Inst{15-12} = Rd; - let Inst{19-16} = Rn; -} - -def RSBrs : AsI1<0b0011, (outs GPR:$Rd), (ins GPR:$Rn, so_reg:$shift), - DPSoRegFrm, IIC_iALUsr, "rsb", "\t$Rd, $Rn, $shift", - [(set GPR:$Rd, (sub so_reg:$shift, GPR:$Rn))]> { - bits<4> Rd; - bits<4> Rn; - bits<12> shift; - let Inst{25} = 0; - let Inst{11-0} = shift; - let Inst{15-12} = Rd; - let Inst{19-16} = Rn; -} +defm RSB : AsI1_rbin_irs <0b0011, "rsb", + IIC_iALUi, IIC_iALUr, IIC_iALUsr, + BinOpFrag<(sub node:$LHS, node:$RHS)>, "RSB">; -// RSB with 's' bit set. -// NOTE: CPSR def omitted because it will be handled by the custom inserter. -let usesCustomInserter = 1 in { -def RSBSri : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, so_imm:$imm), - 4, IIC_iALUi, - [(set GPR:$Rd, (subc so_imm:$imm, GPR:$Rn))]>; -def RSBSrr : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm), - 4, IIC_iALUr, - [/* For disassembly only; pattern left blank */]>; -def RSBSrs : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, so_reg:$shift), - 4, IIC_iALUsr, - [(set GPR:$Rd, (subc so_reg:$shift, GPR:$Rn))]>; -} - -let Uses = [CPSR] in { -def RSCri : AsI1<0b0111, (outs GPR:$Rd), (ins GPR:$Rn, so_imm:$imm), - DPFrm, IIC_iALUi, "rsc", "\t$Rd, $Rn, $imm", - [(set GPR:$Rd, (sube_dead_carry so_imm:$imm, GPR:$Rn))]>, - Requires<[IsARM]> { - bits<4> Rd; - bits<4> Rn; - bits<12> imm; - let Inst{25} = 1; - let Inst{15-12} = Rd; - let Inst{19-16} = Rn; - let Inst{11-0} = imm; -} -// The reg/reg form is only defined for the disassembler; for codegen it is -// equivalent to SUBrr. -def RSCrr : AsI1<0b0111, (outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm), - DPFrm, IIC_iALUr, "rsc", "\t$Rd, $Rn, $Rm", - [/* For disassembly only; pattern left blank */]> { - bits<4> Rd; - bits<4> Rn; - bits<4> Rm; - let Inst{11-4} = 0b00000000; - let Inst{25} = 0; - let Inst{3-0} = Rm; - let Inst{15-12} = Rd; - let Inst{19-16} = Rn; -} -def RSCrs : AsI1<0b0111, (outs GPR:$Rd), (ins GPR:$Rn, so_reg:$shift), - DPSoRegFrm, IIC_iALUsr, "rsc", "\t$Rd, $Rn, $shift", - [(set GPR:$Rd, (sube_dead_carry so_reg:$shift, GPR:$Rn))]>, - Requires<[IsARM]> { - bits<4> Rd; - bits<4> Rn; - bits<12> shift; - let Inst{25} = 0; - let Inst{11-0} = shift; - let Inst{15-12} = Rd; - let Inst{19-16} = Rn; -} -} +// FIXME: Eliminate them if we can write def : Pat patterns which defines +// CPSR and the implicit def of CPSR is not needed. +defm RSBS : AsI1_rbin_s_is<0b0011, "rsb", + IIC_iALUi, IIC_iALUr, IIC_iALUsr, + BinOpFrag<(ARMsubc node:$LHS, node:$RHS)>>; -// NOTE: CPSR def omitted because it will be handled by the custom inserter. -let usesCustomInserter = 1, Uses = [CPSR] in { -def RSCSri : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, so_imm:$imm), - 4, IIC_iALUi, - [(set GPR:$Rd, (sube_dead_carry so_imm:$imm, GPR:$Rn))]>; -def RSCSrs : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, so_reg:$shift), - 4, IIC_iALUsr, - [(set GPR:$Rd, (sube_dead_carry so_reg:$shift, GPR:$Rn))]>; -} +defm RSC : AI1_rsc_irs<0b0111, "rsc", + BinOpWithFlagFrag<(ARMsube node:$LHS, node:$RHS, node:$FLAG)>, + "RSC">; // (sub X, imm) gets canonicalized to (add X, -imm). Match this form. // The assume-no-carry-in form uses the negation of the input since add/sub // assume opposite meanings of the carry flag (i.e., carry == !borrow). // See the definition of AddWithCarry() in the ARM ARM A2.2.1 for the gory // details. -def : ARMPat<(add GPR:$src, so_imm_neg:$imm), - (SUBri GPR:$src, so_imm_neg:$imm)>; -def : ARMPat<(addc GPR:$src, so_imm_neg:$imm), - (SUBSri GPR:$src, so_imm_neg:$imm)>; +def : ARMPat<(add GPR:$src, so_imm_neg:$imm), + (SUBri GPR:$src, so_imm_neg:$imm)>; +def : ARMPat<(ARMaddc GPR:$src, so_imm_neg:$imm), + (SUBSri GPR:$src, so_imm_neg:$imm)>; + // The with-carry-in form matches bitwise not instead of the negation. // Effectively, the inverse interpretation of the carry flag already accounts // for part of the negation. -def : ARMPat<(adde_dead_carry GPR:$src, so_imm_not:$imm), - (SBCri GPR:$src, so_imm_not:$imm)>; -def : ARMPat<(adde_live_carry GPR:$src, so_imm_not:$imm), - (SBCSri GPR:$src, so_imm_not:$imm)>; +def : ARMPat<(ARMadde GPR:$src, so_imm_not:$imm, CPSR), + (SBCri GPR:$src, so_imm_not:$imm)>; // Note: These are implemented in C++ code, because they have to generate // ADD/SUBrs instructions, which use a complex pattern that a xform function @@ -2427,12 +3138,13 @@ def : ARMPat<(adde_live_carry GPR:$src, so_imm_not:$imm), // (mul X, 2^n+1) -> (add (X << n), X) // (mul X, 2^n-1) -> (rsb X, (X << n)) -// ARM Arithmetic Instruction -- for disassembly only +// ARM Arithmetic Instruction // GPR:$dst = GPR:$a op GPR:$b class AAI op27_20, bits<8> op11_4, string opc, - list pattern = [/* For disassembly only; pattern left blank */], - dag iops = (ins GPR:$Rn, GPR:$Rm), string asm = "\t$Rd, $Rn, $Rm"> - : AI<(outs GPR:$Rd), iops, DPFrm, IIC_iALUr, opc, asm, pattern> { + list pattern = [], + dag iops = (ins GPRnopc:$Rn, GPRnopc:$Rm), + string asm = "\t$Rd, $Rn, $Rm"> + : AI<(outs GPRnopc:$Rd), iops, DPFrm, IIC_iALUr, opc, asm, pattern> { bits<4> Rn; bits<4> Rd; bits<4> Rm; @@ -2443,17 +3155,19 @@ class AAI op27_20, bits<8> op11_4, string opc, let Inst{3-0} = Rm; } -// Saturating add/subtract -- for disassembly only +// Saturating add/subtract def QADD : AAI<0b00010000, 0b00000101, "qadd", - [(set GPR:$Rd, (int_arm_qadd GPR:$Rm, GPR:$Rn))], - (ins GPR:$Rm, GPR:$Rn), "\t$Rd, $Rm, $Rn">; + [(set GPRnopc:$Rd, (int_arm_qadd GPRnopc:$Rm, GPRnopc:$Rn))], + (ins GPRnopc:$Rm, GPRnopc:$Rn), "\t$Rd, $Rm, $Rn">; def QSUB : AAI<0b00010010, 0b00000101, "qsub", - [(set GPR:$Rd, (int_arm_qsub GPR:$Rm, GPR:$Rn))], - (ins GPR:$Rm, GPR:$Rn), "\t$Rd, $Rm, $Rn">; -def QDADD : AAI<0b00010100, 0b00000101, "qdadd", [], (ins GPR:$Rm, GPR:$Rn), + [(set GPRnopc:$Rd, (int_arm_qsub GPRnopc:$Rm, GPRnopc:$Rn))], + (ins GPRnopc:$Rm, GPRnopc:$Rn), "\t$Rd, $Rm, $Rn">; +def QDADD : AAI<0b00010100, 0b00000101, "qdadd", [], + (ins GPRnopc:$Rm, GPRnopc:$Rn), "\t$Rd, $Rm, $Rn">; -def QDSUB : AAI<0b00010110, 0b00000101, "qdsub", [], (ins GPR:$Rm, GPR:$Rn), +def QDSUB : AAI<0b00010110, 0b00000101, "qdsub", [], + (ins GPRnopc:$Rm, GPRnopc:$Rn), "\t$Rd, $Rm, $Rn">; def QADD16 : AAI<0b01100010, 0b11110001, "qadd16">; @@ -2469,7 +3183,7 @@ def UQSAX : AAI<0b01100110, 0b11110101, "uqsax">; def UQSUB16 : AAI<0b01100110, 0b11110111, "uqsub16">; def UQSUB8 : AAI<0b01100110, 0b11111111, "uqsub8">; -// Signed/Unsigned add/subtract -- for disassembly only +// Signed/Unsigned add/subtract def SASX : AAI<0b01100001, 0b11110011, "sasx">; def SADD16 : AAI<0b01100001, 0b11110001, "sadd16">; @@ -2484,7 +3198,7 @@ def USAX : AAI<0b01100101, 0b11110101, "usax">; def USUB16 : AAI<0b01100101, 0b11110111, "usub16">; def USUB8 : AAI<0b01100101, 0b11111111, "usub8">; -// Signed/Unsigned halving add/subtract -- for disassembly only +// Signed/Unsigned halving add/subtract def SHASX : AAI<0b01100011, 0b11110011, "shasx">; def SHADD16 : AAI<0b01100011, 0b11110001, "shadd16">; @@ -2499,7 +3213,7 @@ def UHSAX : AAI<0b01100111, 0b11110101, "uhsax">; def UHSUB16 : AAI<0b01100111, 0b11110111, "uhsub16">; def UHSUB8 : AAI<0b01100111, 0b11111111, "uhsub8">; -// Unsigned Sum of Absolute Differences [and Accumulate] -- for disassembly only +// Unsigned Sum of Absolute Differences [and Accumulate]. def USAD8 : AI<(outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm), MulFrm /* for convenience */, NoItinerary, "usad8", @@ -2531,11 +3245,11 @@ def USADA8 : AI<(outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm, GPR:$Ra), let Inst{3-0} = Rn; } -// Signed/Unsigned saturate -- for disassembly only +// Signed/Unsigned saturate -def SSAT : AI<(outs GPR:$Rd), (ins ssat_imm:$sat_imm, GPR:$a, shift_imm:$sh), - SatFrm, NoItinerary, "ssat", "\t$Rd, $sat_imm, $a$sh", - [/* For disassembly only; pattern left blank */]> { +def SSAT : AI<(outs GPRnopc:$Rd), + (ins imm1_32:$sat_imm, GPRnopc:$Rn, shift_imm:$sh), + SatFrm, NoItinerary, "ssat", "\t$Rd, $sat_imm, $Rn$sh", []> { bits<4> Rd; bits<5> sat_imm; bits<4> Rn; @@ -2544,14 +3258,14 @@ def SSAT : AI<(outs GPR:$Rd), (ins ssat_imm:$sat_imm, GPR:$a, shift_imm:$sh), let Inst{5-4} = 0b01; let Inst{20-16} = sat_imm; let Inst{15-12} = Rd; - let Inst{11-7} = sh{7-3}; - let Inst{6} = sh{0}; + let Inst{11-7} = sh{4-0}; + let Inst{6} = sh{5}; let Inst{3-0} = Rn; } -def SSAT16 : AI<(outs GPR:$Rd), (ins ssat_imm:$sat_imm, GPR:$Rn), SatFrm, - NoItinerary, "ssat16", "\t$Rd, $sat_imm, $Rn", - [/* For disassembly only; pattern left blank */]> { +def SSAT16 : AI<(outs GPRnopc:$Rd), + (ins imm1_16:$sat_imm, GPRnopc:$Rn), SatFrm, + NoItinerary, "ssat16", "\t$Rd, $sat_imm, $Rn", []> { bits<4> Rd; bits<4> sat_imm; bits<4> Rn; @@ -2562,9 +3276,9 @@ def SSAT16 : AI<(outs GPR:$Rd), (ins ssat_imm:$sat_imm, GPR:$Rn), SatFrm, let Inst{3-0} = Rn; } -def USAT : AI<(outs GPR:$Rd), (ins i32imm:$sat_imm, GPR:$a, shift_imm:$sh), - SatFrm, NoItinerary, "usat", "\t$Rd, $sat_imm, $a$sh", - [/* For disassembly only; pattern left blank */]> { +def USAT : AI<(outs GPRnopc:$Rd), + (ins imm0_31:$sat_imm, GPRnopc:$Rn, shift_imm:$sh), + SatFrm, NoItinerary, "usat", "\t$Rd, $sat_imm, $Rn$sh", []> { bits<4> Rd; bits<5> sat_imm; bits<4> Rn; @@ -2572,15 +3286,15 @@ def USAT : AI<(outs GPR:$Rd), (ins i32imm:$sat_imm, GPR:$a, shift_imm:$sh), let Inst{27-21} = 0b0110111; let Inst{5-4} = 0b01; let Inst{15-12} = Rd; - let Inst{11-7} = sh{7-3}; - let Inst{6} = sh{0}; + let Inst{11-7} = sh{4-0}; + let Inst{6} = sh{5}; let Inst{20-16} = sat_imm; let Inst{3-0} = Rn; } -def USAT16 : AI<(outs GPR:$Rd), (ins i32imm:$sat_imm, GPR:$a), SatFrm, - NoItinerary, "usat16", "\t$Rd, $sat_imm, $a", - [/* For disassembly only; pattern left blank */]> { +def USAT16 : AI<(outs GPRnopc:$Rd), + (ins imm0_15:$sat_imm, GPRnopc:$Rn), SatFrm, + NoItinerary, "usat16", "\t$Rd, $sat_imm, $Rn", []> { bits<4> Rd; bits<4> sat_imm; bits<4> Rn; @@ -2591,8 +3305,10 @@ def USAT16 : AI<(outs GPR:$Rd), (ins i32imm:$sat_imm, GPR:$a), SatFrm, let Inst{3-0} = Rn; } -def : ARMV6Pat<(int_arm_ssat GPR:$a, imm:$pos), (SSAT imm:$pos, GPR:$a, 0)>; -def : ARMV6Pat<(int_arm_usat GPR:$a, imm:$pos), (USAT imm:$pos, GPR:$a, 0)>; +def : ARMV6Pat<(int_arm_ssat GPRnopc:$a, imm:$pos), + (SSAT imm:$pos, GPRnopc:$a, 0)>; +def : ARMV6Pat<(int_arm_usat GPRnopc:$a, imm:$pos), + (USAT imm:$pos, GPRnopc:$a, 0)>; //===----------------------------------------------------------------------===// // Bitwise Instructions. @@ -2611,6 +3327,10 @@ defm BIC : AsI1_bin_irs<0b1110, "bic", IIC_iBITi, IIC_iBITr, IIC_iBITsr, BinOpFrag<(and node:$LHS, (not node:$RHS))>, "BIC">; +// FIXME: bf_inv_mask_imm should be two operands, the lsb and the msb, just +// like in the actual instruction encoding. The complexity of mapping the mask +// to the lsb/msb pair should be handled by ISel, not encapsulated in the +// instruction description. def BFC : I<(outs GPR:$Rd), (ins GPR:$src, bf_inv_mask_imm:$imm), AddrMode1, 4, IndexModeNone, DPFrm, IIC_iUNAsi, "bfc", "\t$Rd, $imm", "$src = $Rd", @@ -2622,16 +3342,16 @@ def BFC : I<(outs GPR:$Rd), (ins GPR:$src, bf_inv_mask_imm:$imm), let Inst{6-0} = 0b0011111; let Inst{15-12} = Rd; let Inst{11-7} = imm{4-0}; // lsb - let Inst{20-16} = imm{9-5}; // width + let Inst{20-16} = imm{9-5}; // msb } // A8.6.18 BFI - Bitfield insert (Encoding A1) -def BFI : I<(outs GPR:$Rd), (ins GPR:$src, GPR:$Rn, bf_inv_mask_imm:$imm), - AddrMode1, 4, IndexModeNone, DPFrm, IIC_iUNAsi, - "bfi", "\t$Rd, $Rn, $imm", "$src = $Rd", - [(set GPR:$Rd, (ARMbfi GPR:$src, GPR:$Rn, - bf_inv_mask_imm:$imm))]>, - Requires<[IsARM, HasV6T2]> { +def BFI:I<(outs GPRnopc:$Rd), (ins GPRnopc:$src, GPR:$Rn, bf_inv_mask_imm:$imm), + AddrMode1, 4, IndexModeNone, DPFrm, IIC_iUNAsi, + "bfi", "\t$Rd, $Rn, $imm", "$src = $Rd", + [(set GPRnopc:$Rd, (ARMbfi GPRnopc:$src, GPR:$Rn, + bf_inv_mask_imm:$imm))]>, + Requires<[IsARM, HasV6T2]> { bits<4> Rd; bits<4> Rn; bits<10> imm; @@ -2643,25 +3363,6 @@ def BFI : I<(outs GPR:$Rd), (ins GPR:$src, GPR:$Rn, bf_inv_mask_imm:$imm), let Inst{3-0} = Rn; } -// GNU as only supports this form of bfi (w/ 4 arguments) -let isAsmParserOnly = 1 in -def BFI4p : I<(outs GPR:$Rd), (ins GPR:$src, GPR:$Rn, - lsb_pos_imm:$lsb, width_imm:$width), - AddrMode1, 4, IndexModeNone, DPFrm, IIC_iUNAsi, - "bfi", "\t$Rd, $Rn, $lsb, $width", "$src = $Rd", - []>, Requires<[IsARM, HasV6T2]> { - bits<4> Rd; - bits<4> Rn; - bits<5> lsb; - bits<5> width; - let Inst{27-21} = 0b0111110; - let Inst{6-4} = 0b001; // Rn: Inst{3-0} != 15 - let Inst{15-12} = Rd; - let Inst{11-7} = lsb; - let Inst{20-16} = width; // Custom encoder => lsb+width-1 - let Inst{3-0} = Rn; -} - def MVNr : AsI1<0b1111, (outs GPR:$Rd), (ins GPR:$Rm), DPFrm, IIC_iMVNr, "mvn", "\t$Rd, $Rm", [(set GPR:$Rd, (not GPR:$Rm))]>, UnaryDP { @@ -2673,15 +3374,31 @@ def MVNr : AsI1<0b1111, (outs GPR:$Rd), (ins GPR:$Rm), DPFrm, IIC_iMVNr, let Inst{15-12} = Rd; let Inst{3-0} = Rm; } -def MVNs : AsI1<0b1111, (outs GPR:$Rd), (ins so_reg:$shift), DPSoRegFrm, - IIC_iMVNsr, "mvn", "\t$Rd, $shift", - [(set GPR:$Rd, (not so_reg:$shift))]>, UnaryDP { +def MVNsi : AsI1<0b1111, (outs GPR:$Rd), (ins so_reg_imm:$shift), + DPSoRegImmFrm, IIC_iMVNsr, "mvn", "\t$Rd, $shift", + [(set GPR:$Rd, (not so_reg_imm:$shift))]>, UnaryDP { + bits<4> Rd; + bits<12> shift; + let Inst{25} = 0; + let Inst{19-16} = 0b0000; + let Inst{15-12} = Rd; + let Inst{11-5} = shift{11-5}; + let Inst{4} = 0; + let Inst{3-0} = shift{3-0}; +} +def MVNsr : AsI1<0b1111, (outs GPR:$Rd), (ins so_reg_reg:$shift), + DPSoRegRegFrm, IIC_iMVNsr, "mvn", "\t$Rd, $shift", + [(set GPR:$Rd, (not so_reg_reg:$shift))]>, UnaryDP { bits<4> Rd; bits<12> shift; let Inst{25} = 0; let Inst{19-16} = 0b0000; let Inst{15-12} = Rd; - let Inst{11-0} = shift; + let Inst{11-8} = shift{11-8}; + let Inst{7} = 0; + let Inst{6-5} = shift{6-5}; + let Inst{4} = 1; + let Inst{3-0} = shift{3-0}; } let isReMaterializable = 1, isAsCheapAsAMove = 1, isMoveImm = 1 in def MVNi : AsI1<0b1111, (outs GPR:$Rd), (ins so_imm:$imm), DPFrm, @@ -2820,8 +3537,8 @@ def UMAAL : AMul1I <0b0000010, (outs GPR:$RdLo, GPR:$RdHi), bits<4> RdHi; bits<4> Rm; bits<4> Rn; - let Inst{19-16} = RdLo; - let Inst{15-12} = RdHi; + let Inst{19-16} = RdHi; + let Inst{15-12} = RdLo; let Inst{11-8} = Rm; let Inst{3-0} = Rn; } @@ -2855,8 +3572,7 @@ def SMMUL : AMul2I <0b0111010, 0b0001, (outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm), } def SMMULR : AMul2I <0b0111010, 0b0011, (outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm), - IIC_iMUL32, "smmulr", "\t$Rd, $Rn, $Rm", - [/* For disassembly only; pattern left blank */]>, + IIC_iMUL32, "smmulr", "\t$Rd, $Rn, $Rm", []>, Requires<[IsARM, HasV6]> { let Inst{15-12} = 0b1111; } @@ -2869,8 +3585,7 @@ def SMMLA : AMul2Ia <0b0111010, 0b0001, (outs GPR:$Rd), def SMMLAR : AMul2Ia <0b0111010, 0b0011, (outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm, GPR:$Ra), - IIC_iMAC32, "smmlar", "\t$Rd, $Rn, $Rm, $Ra", - [/* For disassembly only; pattern left blank */]>, + IIC_iMAC32, "smmlar", "\t$Rd, $Rn, $Rm, $Ra", []>, Requires<[IsARM, HasV6]>; def SMMLS : AMul2Ia <0b0111010, 0b1101, (outs GPR:$Rd), @@ -2881,8 +3596,7 @@ def SMMLS : AMul2Ia <0b0111010, 0b1101, (outs GPR:$Rd), def SMMLSR : AMul2Ia <0b0111010, 0b1111, (outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm, GPR:$Ra), - IIC_iMAC32, "smmlsr", "\t$Rd, $Rn, $Rm, $Ra", - [/* For disassembly only; pattern left blank */]>, + IIC_iMAC32, "smmlsr", "\t$Rd, $Rn, $Rm, $Ra", []>, Requires<[IsARM, HasV6]>; multiclass AI_smul { @@ -2925,92 +3639,95 @@ multiclass AI_smul { multiclass AI_smla { - def BB : AMulxyIa<0b0001000, 0b00, (outs GPR:$Rd), - (ins GPR:$Rn, GPR:$Rm, GPR:$Ra), + let DecoderMethod = "DecodeSMLAInstruction" in { + def BB : AMulxyIa<0b0001000, 0b00, (outs GPRnopc:$Rd), + (ins GPRnopc:$Rn, GPRnopc:$Rm, GPR:$Ra), IIC_iMAC16, !strconcat(opc, "bb"), "\t$Rd, $Rn, $Rm, $Ra", - [(set GPR:$Rd, (add GPR:$Ra, - (opnode (sext_inreg GPR:$Rn, i16), - (sext_inreg GPR:$Rm, i16))))]>, + [(set GPRnopc:$Rd, (add GPR:$Ra, + (opnode (sext_inreg GPRnopc:$Rn, i16), + (sext_inreg GPRnopc:$Rm, i16))))]>, Requires<[IsARM, HasV5TE]>; - def BT : AMulxyIa<0b0001000, 0b10, (outs GPR:$Rd), - (ins GPR:$Rn, GPR:$Rm, GPR:$Ra), + def BT : AMulxyIa<0b0001000, 0b10, (outs GPRnopc:$Rd), + (ins GPRnopc:$Rn, GPRnopc:$Rm, GPR:$Ra), IIC_iMAC16, !strconcat(opc, "bt"), "\t$Rd, $Rn, $Rm, $Ra", - [(set GPR:$Rd, (add GPR:$Ra, (opnode (sext_inreg GPR:$Rn, i16), - (sra GPR:$Rm, (i32 16)))))]>, + [(set GPRnopc:$Rd, + (add GPR:$Ra, (opnode (sext_inreg GPRnopc:$Rn, i16), + (sra GPRnopc:$Rm, (i32 16)))))]>, Requires<[IsARM, HasV5TE]>; - def TB : AMulxyIa<0b0001000, 0b01, (outs GPR:$Rd), - (ins GPR:$Rn, GPR:$Rm, GPR:$Ra), + def TB : AMulxyIa<0b0001000, 0b01, (outs GPRnopc:$Rd), + (ins GPRnopc:$Rn, GPRnopc:$Rm, GPR:$Ra), IIC_iMAC16, !strconcat(opc, "tb"), "\t$Rd, $Rn, $Rm, $Ra", - [(set GPR:$Rd, (add GPR:$Ra, (opnode (sra GPR:$Rn, (i32 16)), - (sext_inreg GPR:$Rm, i16))))]>, + [(set GPRnopc:$Rd, + (add GPR:$Ra, (opnode (sra GPRnopc:$Rn, (i32 16)), + (sext_inreg GPRnopc:$Rm, i16))))]>, Requires<[IsARM, HasV5TE]>; - def TT : AMulxyIa<0b0001000, 0b11, (outs GPR:$Rd), - (ins GPR:$Rn, GPR:$Rm, GPR:$Ra), + def TT : AMulxyIa<0b0001000, 0b11, (outs GPRnopc:$Rd), + (ins GPRnopc:$Rn, GPRnopc:$Rm, GPR:$Ra), IIC_iMAC16, !strconcat(opc, "tt"), "\t$Rd, $Rn, $Rm, $Ra", - [(set GPR:$Rd, (add GPR:$Ra, (opnode (sra GPR:$Rn, (i32 16)), - (sra GPR:$Rm, (i32 16)))))]>, + [(set GPRnopc:$Rd, + (add GPR:$Ra, (opnode (sra GPRnopc:$Rn, (i32 16)), + (sra GPRnopc:$Rm, (i32 16)))))]>, Requires<[IsARM, HasV5TE]>; - def WB : AMulxyIa<0b0001001, 0b00, (outs GPR:$Rd), - (ins GPR:$Rn, GPR:$Rm, GPR:$Ra), + def WB : AMulxyIa<0b0001001, 0b00, (outs GPRnopc:$Rd), + (ins GPRnopc:$Rn, GPRnopc:$Rm, GPR:$Ra), IIC_iMAC16, !strconcat(opc, "wb"), "\t$Rd, $Rn, $Rm, $Ra", - [(set GPR:$Rd, (add GPR:$Ra, (sra (opnode GPR:$Rn, - (sext_inreg GPR:$Rm, i16)), (i32 16))))]>, + [(set GPRnopc:$Rd, + (add GPR:$Ra, (sra (opnode GPRnopc:$Rn, + (sext_inreg GPRnopc:$Rm, i16)), (i32 16))))]>, Requires<[IsARM, HasV5TE]>; - def WT : AMulxyIa<0b0001001, 0b10, (outs GPR:$Rd), - (ins GPR:$Rn, GPR:$Rm, GPR:$Ra), + def WT : AMulxyIa<0b0001001, 0b10, (outs GPRnopc:$Rd), + (ins GPRnopc:$Rn, GPRnopc:$Rm, GPR:$Ra), IIC_iMAC16, !strconcat(opc, "wt"), "\t$Rd, $Rn, $Rm, $Ra", - [(set GPR:$Rd, (add GPR:$Ra, (sra (opnode GPR:$Rn, - (sra GPR:$Rm, (i32 16))), (i32 16))))]>, + [(set GPRnopc:$Rd, + (add GPR:$Ra, (sra (opnode GPRnopc:$Rn, + (sra GPRnopc:$Rm, (i32 16))), (i32 16))))]>, Requires<[IsARM, HasV5TE]>; + } } defm SMUL : AI_smul<"smul", BinOpFrag<(mul node:$LHS, node:$RHS)>>; defm SMLA : AI_smla<"smla", BinOpFrag<(mul node:$LHS, node:$RHS)>>; -// Halfword multiply accumulate long: SMLAL -- for disassembly only -def SMLALBB : AMulxyI64<0b0001010, 0b00, (outs GPR:$RdLo, GPR:$RdHi), - (ins GPR:$Rn, GPR:$Rm), - IIC_iMAC64, "smlalbb", "\t$RdLo, $RdHi, $Rn, $Rm", - [/* For disassembly only; pattern left blank */]>, +// Halfword multiply accumulate long: SMLAL. +def SMLALBB : AMulxyI64<0b0001010, 0b00, (outs GPRnopc:$RdLo, GPRnopc:$RdHi), + (ins GPRnopc:$Rn, GPRnopc:$Rm), + IIC_iMAC64, "smlalbb", "\t$RdLo, $RdHi, $Rn, $Rm", []>, Requires<[IsARM, HasV5TE]>; -def SMLALBT : AMulxyI64<0b0001010, 0b10, (outs GPR:$RdLo, GPR:$RdHi), - (ins GPR:$Rn, GPR:$Rm), - IIC_iMAC64, "smlalbt", "\t$RdLo, $RdHi, $Rn, $Rm", - [/* For disassembly only; pattern left blank */]>, +def SMLALBT : AMulxyI64<0b0001010, 0b10, (outs GPRnopc:$RdLo, GPRnopc:$RdHi), + (ins GPRnopc:$Rn, GPRnopc:$Rm), + IIC_iMAC64, "smlalbt", "\t$RdLo, $RdHi, $Rn, $Rm", []>, Requires<[IsARM, HasV5TE]>; -def SMLALTB : AMulxyI64<0b0001010, 0b01, (outs GPR:$RdLo, GPR:$RdHi), - (ins GPR:$Rn, GPR:$Rm), - IIC_iMAC64, "smlaltb", "\t$RdLo, $RdHi, $Rn, $Rm", - [/* For disassembly only; pattern left blank */]>, +def SMLALTB : AMulxyI64<0b0001010, 0b01, (outs GPRnopc:$RdLo, GPRnopc:$RdHi), + (ins GPRnopc:$Rn, GPRnopc:$Rm), + IIC_iMAC64, "smlaltb", "\t$RdLo, $RdHi, $Rn, $Rm", []>, Requires<[IsARM, HasV5TE]>; -def SMLALTT : AMulxyI64<0b0001010, 0b11, (outs GPR:$RdLo, GPR:$RdHi), - (ins GPR:$Rn, GPR:$Rm), - IIC_iMAC64, "smlaltt", "\t$RdLo, $RdHi, $Rn, $Rm", - [/* For disassembly only; pattern left blank */]>, +def SMLALTT : AMulxyI64<0b0001010, 0b11, (outs GPRnopc:$RdLo, GPRnopc:$RdHi), + (ins GPRnopc:$Rn, GPRnopc:$Rm), + IIC_iMAC64, "smlaltt", "\t$RdLo, $RdHi, $Rn, $Rm", []>, Requires<[IsARM, HasV5TE]>; -// Helper class for AI_smld -- for disassembly only +// Helper class for AI_smld. class AMulDualIbase : AI, Requires<[IsARM, HasV6]> { bits<4> Rn; bits<4> Rm; - let Inst{4} = 1; - let Inst{5} = swap; - let Inst{6} = sub; - let Inst{7} = 0; - let Inst{21-20} = 0b00; - let Inst{22} = long; let Inst{27-23} = 0b01110; + let Inst{22} = long; + let Inst{21-20} = 0b00; let Inst{11-8} = Rm; + let Inst{7} = 0; + let Inst{6} = sub; + let Inst{5} = swap; + let Inst{4} = 1; let Inst{3-0} = Rn; } class AMulDualI : AMulDualIbase { bits<4> Ra; + bits<4> Rd; + let Inst{19-16} = Rd; let Inst{15-12} = Ra; } class AMulDualI64 { - def D : AMulDualIa<0, sub, 0, (outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm, GPR:$Ra), + def D : AMulDualIa<0, sub, 0, (outs GPRnopc:$Rd), + (ins GPRnopc:$Rn, GPRnopc:$Rm, GPR:$Ra), NoItinerary, !strconcat(opc, "d"), "\t$Rd, $Rn, $Rm, $Ra">; - def DX: AMulDualIa<0, sub, 1, (outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm, GPR:$Ra), + def DX: AMulDualIa<0, sub, 1, (outs GPRnopc:$Rd), + (ins GPRnopc:$Rn, GPRnopc:$Rm, GPR:$Ra), NoItinerary, !strconcat(opc, "dx"), "\t$Rd, $Rn, $Rm, $Ra">; - def LD: AMulDualI64<1, sub, 0, (outs GPR:$RdLo,GPR:$RdHi), - (ins GPR:$Rn, GPR:$Rm), NoItinerary, + def LD: AMulDualI64<1, sub, 0, (outs GPRnopc:$RdLo, GPRnopc:$RdHi), + (ins GPRnopc:$Rn, GPRnopc:$Rm), NoItinerary, !strconcat(opc, "ld"), "\t$RdLo, $RdHi, $Rn, $Rm">; - def LDX : AMulDualI64<1, sub, 1, (outs GPR:$RdLo,GPR:$RdHi), - (ins GPR:$Rn, GPR:$Rm), NoItinerary, + def LDX : AMulDualI64<1, sub, 1, (outs GPRnopc:$RdLo, GPRnopc:$RdHi), + (ins GPRnopc:$Rn, GPRnopc:$Rm), NoItinerary, !strconcat(opc, "ldx"),"\t$RdLo, $RdHi, $Rn, $Rm">; } @@ -3058,10 +3779,10 @@ defm SMLS : AI_smld<1, "smls">; multiclass AI_sdml { - def D : AMulDualI<0, sub, 0, (outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm), - NoItinerary, !strconcat(opc, "d"), "\t$Rd, $Rn, $Rm">; - def DX : AMulDualI<0, sub, 1, (outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm), - NoItinerary, !strconcat(opc, "dx"), "\t$Rd, $Rn, $Rm">; + def D:AMulDualI<0, sub, 0, (outs GPRnopc:$Rd), (ins GPRnopc:$Rn, GPRnopc:$Rm), + NoItinerary, !strconcat(opc, "d"), "\t$Rd, $Rn, $Rm">; + def DX:AMulDualI<0, sub, 1, (outs GPRnopc:$Rd),(ins GPRnopc:$Rn, GPRnopc:$Rm), + NoItinerary, !strconcat(opc, "dx"), "\t$Rd, $Rn, $Rm">; } defm SMUA : AI_sdml<0, "smua">; @@ -3100,55 +3821,38 @@ def : ARMV6Pat<(or (sra (shl GPR:$Rm, (i32 24)), (i32 16)), (and (srl GPR:$Rm, (i32 8)), 0xFF)), (REVSH GPR:$Rm)>; -def lsl_shift_imm : SDNodeXFormgetZExtValue()); - return CurDAG->getTargetConstant(Sh, MVT::i32); -}]>; - -def lsl_amt : ImmLeaf 0 && Imm < 32; -}], lsl_shift_imm>; - -def PKHBT : APKHI<0b01101000, 0, (outs GPR:$Rd), - (ins GPR:$Rn, GPR:$Rm, shift_imm:$sh), +def PKHBT : APKHI<0b01101000, 0, (outs GPRnopc:$Rd), + (ins GPRnopc:$Rn, GPRnopc:$Rm, pkh_lsl_amt:$sh), IIC_iALUsi, "pkhbt", "\t$Rd, $Rn, $Rm$sh", - [(set GPR:$Rd, (or (and GPR:$Rn, 0xFFFF), - (and (shl GPR:$Rm, lsl_amt:$sh), - 0xFFFF0000)))]>, + [(set GPRnopc:$Rd, (or (and GPRnopc:$Rn, 0xFFFF), + (and (shl GPRnopc:$Rm, pkh_lsl_amt:$sh), + 0xFFFF0000)))]>, Requires<[IsARM, HasV6]>; // Alternate cases for PKHBT where identities eliminate some nodes. -def : ARMV6Pat<(or (and GPR:$Rn, 0xFFFF), (and GPR:$Rm, 0xFFFF0000)), - (PKHBT GPR:$Rn, GPR:$Rm, 0)>; -def : ARMV6Pat<(or (and GPR:$Rn, 0xFFFF), (shl GPR:$Rm, imm16_31:$sh)), - (PKHBT GPR:$Rn, GPR:$Rm, (lsl_shift_imm imm16_31:$sh))>; - -def asr_shift_imm : SDNodeXFormgetZExtValue()); - return CurDAG->getTargetConstant(Sh, MVT::i32); -}]>; - -def asr_amt : ImmLeaf 0 && Imm <= 32; -}], asr_shift_imm>; +def : ARMV6Pat<(or (and GPRnopc:$Rn, 0xFFFF), (and GPRnopc:$Rm, 0xFFFF0000)), + (PKHBT GPRnopc:$Rn, GPRnopc:$Rm, 0)>; +def : ARMV6Pat<(or (and GPRnopc:$Rn, 0xFFFF), (shl GPRnopc:$Rm, imm16_31:$sh)), + (PKHBT GPRnopc:$Rn, GPRnopc:$Rm, imm16_31:$sh)>; // Note: Shifts of 1-15 bits will be transformed to srl instead of sra and // will match the pattern below. -def PKHTB : APKHI<0b01101000, 1, (outs GPR:$Rd), - (ins GPR:$Rn, GPR:$Rm, shift_imm:$sh), +def PKHTB : APKHI<0b01101000, 1, (outs GPRnopc:$Rd), + (ins GPRnopc:$Rn, GPRnopc:$Rm, pkh_asr_amt:$sh), IIC_iBITsi, "pkhtb", "\t$Rd, $Rn, $Rm$sh", - [(set GPR:$Rd, (or (and GPR:$Rn, 0xFFFF0000), - (and (sra GPR:$Rm, asr_amt:$sh), - 0xFFFF)))]>, + [(set GPRnopc:$Rd, (or (and GPRnopc:$Rn, 0xFFFF0000), + (and (sra GPRnopc:$Rm, pkh_asr_amt:$sh), + 0xFFFF)))]>, Requires<[IsARM, HasV6]>; // Alternate cases for PKHTB where identities eliminate some nodes. Note that // a shift amount of 0 is *not legal* here, it is PKHBT instead. -def : ARMV6Pat<(or (and GPR:$src1, 0xFFFF0000), (srl GPR:$src2, imm16_31:$sh)), - (PKHTB GPR:$src1, GPR:$src2, (asr_shift_imm imm16_31:$sh))>; -def : ARMV6Pat<(or (and GPR:$src1, 0xFFFF0000), - (and (srl GPR:$src2, imm1_15:$sh), 0xFFFF)), - (PKHTB GPR:$src1, GPR:$src2, (asr_shift_imm imm1_15:$sh))>; +def : ARMV6Pat<(or (and GPRnopc:$src1, 0xFFFF0000), + (srl GPRnopc:$src2, imm16_31:$sh)), + (PKHTB GPRnopc:$src1, GPRnopc:$src2, imm16_31:$sh)>; +def : ARMV6Pat<(or (and GPRnopc:$src1, 0xFFFF0000), + (and (srl GPRnopc:$src2, imm1_15:$sh), 0xFFFF)), + (PKHTB GPRnopc:$src1, GPRnopc:$src2, imm1_15:$sh)>; //===----------------------------------------------------------------------===// // Comparison Instructions... @@ -3163,8 +3867,10 @@ def : ARMPat<(ARMcmpZ GPR:$src, so_imm:$imm), (CMPri GPR:$src, so_imm:$imm)>; def : ARMPat<(ARMcmpZ GPR:$src, GPR:$rhs), (CMPrr GPR:$src, GPR:$rhs)>; -def : ARMPat<(ARMcmpZ GPR:$src, so_reg:$rhs), - (CMPrs GPR:$src, so_reg:$rhs)>; +def : ARMPat<(ARMcmpZ GPR:$src, so_reg_imm:$rhs), + (CMPrsi GPR:$src, so_reg_imm:$rhs)>; +def : ARMPat<(ARMcmpZ GPR:$src, so_reg_reg:$rhs), + (CMPrsr GPR:$src, so_reg_reg:$rhs)>; // FIXME: We have to be careful when using the CMN instruction and comparison // with 0. One would expect these two pieces of code should give identical @@ -3250,15 +3956,23 @@ def MOVCCr : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$false, GPR:$Rm, pred:$p), 4, IIC_iCMOVr, [/*(set GPR:$Rd, (ARMcmov GPR:$false, GPR:$Rm, imm:$cc, CCR:$ccr))*/]>, RegConstraint<"$false = $Rd">; -def MOVCCs : ARMPseudoInst<(outs GPR:$Rd), - (ins GPR:$false, so_reg:$shift, pred:$p), +def MOVCCsi : ARMPseudoInst<(outs GPR:$Rd), + (ins GPR:$false, so_reg_imm:$shift, pred:$p), + 4, IIC_iCMOVsr, + [/*(set GPR:$Rd, (ARMcmov GPR:$false, so_reg_imm:$shift, + imm:$cc, CCR:$ccr))*/]>, + RegConstraint<"$false = $Rd">; +def MOVCCsr : ARMPseudoInst<(outs GPR:$Rd), + (ins GPR:$false, so_reg_reg:$shift, pred:$p), 4, IIC_iCMOVsr, - [/*(set GPR:$Rd, (ARMcmov GPR:$false, so_reg:$shift, imm:$cc, CCR:$ccr))*/]>, + [/*(set GPR:$Rd, (ARMcmov GPR:$false, so_reg_reg:$shift, + imm:$cc, CCR:$ccr))*/]>, RegConstraint<"$false = $Rd">; + let isMoveImm = 1 in def MOVCCi16 : ARMPseudoInst<(outs GPR:$Rd), - (ins GPR:$false, i32imm_hilo16:$imm, pred:$p), + (ins GPR:$false, imm0_65535_expr:$imm, pred:$p), 4, IIC_iMOVi, []>, RegConstraint<"$false = $Rd">, Requires<[IsARM, HasV6T2]>; @@ -3288,9 +4002,14 @@ def MVNCCi : ARMPseudoInst<(outs GPR:$Rd), // Atomic operations intrinsics // +def MemBarrierOptOperand : AsmOperandClass { + let Name = "MemBarrierOpt"; + let ParserMethod = "parseMemBarrierOptOperand"; +} def memb_opt : Operand { let PrintMethod = "printMemBOption"; let ParserMatchClass = MemBarrierOptOperand; + let DecoderMethod = "DecodeMemBarrierOption"; } // memory barriers protect the atomic sequences @@ -3321,8 +4040,16 @@ def ISB : AInoP<(outs), (ins memb_opt:$opt), MiscFrm, NoItinerary, let Inst{3-0} = opt; } +// Pseudo isntruction that combines movs + predicated rsbmi +// to implement integer ABS +let usesCustomInserter = 1, Defs = [CPSR] in { +def ABS : ARMPseudoInst< + (outs GPR:$dst), (ins GPR:$src), + 8, NoItinerary, []>; +} + let usesCustomInserter = 1 in { - let Uses = [CPSR] in { + let Defs = [CPSR] in { def ATOMIC_LOAD_ADD_I8 : PseudoInst< (outs GPR:$dst), (ins GPR:$ptr, GPR:$incr), NoItinerary, [(set GPR:$dst, (atomic_load_add_8 GPR:$ptr, GPR:$incr))]>; @@ -3437,44 +4164,47 @@ let usesCustomInserter = 1 in { } let mayLoad = 1 in { -def LDREXB : AIldrex<0b10, (outs GPR:$Rt), (ins addrmode7:$addr), NoItinerary, +def LDREXB : AIldrex<0b10, (outs GPR:$Rt), (ins addr_offset_none:$addr), + NoItinerary, "ldrexb", "\t$Rt, $addr", []>; -def LDREXH : AIldrex<0b11, (outs GPR:$Rt), (ins addrmode7:$addr), NoItinerary, - "ldrexh", "\t$Rt, $addr", []>; -def LDREX : AIldrex<0b00, (outs GPR:$Rt), (ins addrmode7:$addr), NoItinerary, - "ldrex", "\t$Rt, $addr", []>; +def LDREXH : AIldrex<0b11, (outs GPR:$Rt), (ins addr_offset_none:$addr), + NoItinerary, "ldrexh", "\t$Rt, $addr", []>; +def LDREX : AIldrex<0b00, (outs GPR:$Rt), (ins addr_offset_none:$addr), + NoItinerary, "ldrex", "\t$Rt, $addr", []>; let hasExtraDefRegAllocReq = 1 in - def LDREXD : AIldrex<0b01, (outs GPR:$Rt, GPR:$Rt2), (ins addrmode7:$addr), - NoItinerary, "ldrexd", "\t$Rt, $Rt2, $addr", []>; +def LDREXD: AIldrex<0b01, (outs GPR:$Rt, GPR:$Rt2),(ins addr_offset_none:$addr), + NoItinerary, "ldrexd", "\t$Rt, $Rt2, $addr", []> { + let DecoderMethod = "DecodeDoubleRegLoad"; +} } let mayStore = 1, Constraints = "@earlyclobber $Rd" in { -def STREXB : AIstrex<0b10, (outs GPR:$Rd), (ins GPR:$Rt, addrmode7:$addr), +def STREXB: AIstrex<0b10, (outs GPR:$Rd), (ins GPR:$Rt, addr_offset_none:$addr), NoItinerary, "strexb", "\t$Rd, $Rt, $addr", []>; -def STREXH : AIstrex<0b11, (outs GPR:$Rd), (ins GPR:$Rt, addrmode7:$addr), +def STREXH: AIstrex<0b11, (outs GPR:$Rd), (ins GPR:$Rt, addr_offset_none:$addr), NoItinerary, "strexh", "\t$Rd, $Rt, $addr", []>; -def STREX : AIstrex<0b00, (outs GPR:$Rd), (ins GPR:$Rt, addrmode7:$addr), +def STREX : AIstrex<0b00, (outs GPR:$Rd), (ins GPR:$Rt, addr_offset_none:$addr), NoItinerary, "strex", "\t$Rd, $Rt, $addr", []>; } let hasExtraSrcRegAllocReq = 1, Constraints = "@earlyclobber $Rd" in def STREXD : AIstrex<0b01, (outs GPR:$Rd), - (ins GPR:$Rt, GPR:$Rt2, addrmode7:$addr), - NoItinerary, "strexd", "\t$Rd, $Rt, $Rt2, $addr", []>; + (ins GPR:$Rt, GPR:$Rt2, addr_offset_none:$addr), + NoItinerary, "strexd", "\t$Rd, $Rt, $Rt2, $addr", []> { + let DecoderMethod = "DecodeDoubleRegStore"; +} -// Clear-Exclusive is for disassembly only. -def CLREX : AXI<(outs), (ins), MiscFrm, NoItinerary, "clrex", - [/* For disassembly only; pattern left blank */]>, +def CLREX : AXI<(outs), (ins), MiscFrm, NoItinerary, "clrex", []>, Requires<[IsARM, HasV7]> { let Inst{31-0} = 0b11110101011111111111000000011111; } -// SWP/SWPB are deprecated in V6/V7 and for disassembly only. -let mayLoad = 1 in { -def SWP : AIswp<0, (outs GPR:$Rt), (ins GPR:$Rt2, GPR:$Rn), "swp", - [/* For disassembly only; pattern left blank */]>; -def SWPB : AIswp<1, (outs GPR:$Rt), (ins GPR:$Rt2, GPR:$Rn), "swpb", - [/* For disassembly only; pattern left blank */]>; +// SWP/SWPB are deprecated in V6/V7. +let mayLoad = 1, mayStore = 1 in { +def SWP : AIswp<0, (outs GPR:$Rt), (ins GPR:$Rt2, addr_offset_none:$addr), + "swp", []>; +def SWPB: AIswp<1, (outs GPR:$Rt), (ins GPR:$Rt2, addr_offset_none:$addr), + "swpb", []>; } //===----------------------------------------------------------------------===// @@ -3526,108 +4256,171 @@ def CDP2 : ABXI<0b1110, (outs), (ins p_imm:$cop, imm0_15:$opc1, class ACI + : I { + let Inst{27-25} = 0b110; +} +class ACInoP : InoP { + opc, asm, "", []> { + let Inst{31-28} = 0b1111; let Inst{27-25} = 0b110; } - -multiclass LdStCop op31_28, bit load, dag ops, string opc, string cond>{ - - def _OFFSET : ACI<(outs), - !con((ins nohash_imm:$cop, nohash_imm:$CRd, addrmode2:$addr), ops), - !strconcat(opc, cond), "\tp$cop, cr$CRd, $addr"> { - let Inst{31-28} = op31_28; +multiclass LdStCop { + def _OFFSET : ACI<(outs), (ins p_imm:$cop, c_imm:$CRd, addrmode5:$addr), + asm, "\t$cop, $CRd, $addr"> { + bits<13> addr; + bits<4> cop; + bits<4> CRd; let Inst{24} = 1; // P = 1 + let Inst{23} = addr{8}; + let Inst{22} = Dbit; let Inst{21} = 0; // W = 0 - let Inst{22} = 0; // D = 0 let Inst{20} = load; + let Inst{19-16} = addr{12-9}; + let Inst{15-12} = CRd; + let Inst{11-8} = cop; + let Inst{7-0} = addr{7-0}; + let DecoderMethod = "DecodeCopMemInstruction"; } - - def _PRE : ACI<(outs), - !con((ins nohash_imm:$cop, nohash_imm:$CRd, addrmode2:$addr), ops), - !strconcat(opc, cond), "\tp$cop, cr$CRd, $addr!", IndexModePre> { - let Inst{31-28} = op31_28; + def _PRE : ACI<(outs), (ins p_imm:$cop, c_imm:$CRd, addrmode5:$addr), + asm, "\t$cop, $CRd, $addr!", IndexModePre> { + bits<13> addr; + bits<4> cop; + bits<4> CRd; let Inst{24} = 1; // P = 1 + let Inst{23} = addr{8}; + let Inst{22} = Dbit; let Inst{21} = 1; // W = 1 - let Inst{22} = 0; // D = 0 let Inst{20} = load; + let Inst{19-16} = addr{12-9}; + let Inst{15-12} = CRd; + let Inst{11-8} = cop; + let Inst{7-0} = addr{7-0}; + let DecoderMethod = "DecodeCopMemInstruction"; } - - def _POST : ACI<(outs), - !con((ins nohash_imm:$cop, nohash_imm:$CRd, addrmode2:$addr), ops), - !strconcat(opc, cond), "\tp$cop, cr$CRd, $addr", IndexModePost> { - let Inst{31-28} = op31_28; + def _POST: ACI<(outs), (ins p_imm:$cop, c_imm:$CRd, addr_offset_none:$addr, + postidx_imm8s4:$offset), + asm, "\t$cop, $CRd, $addr, $offset", IndexModePost> { + bits<9> offset; + bits<4> addr; + bits<4> cop; + bits<4> CRd; let Inst{24} = 0; // P = 0 + let Inst{23} = offset{8}; + let Inst{22} = Dbit; let Inst{21} = 1; // W = 1 - let Inst{22} = 0; // D = 0 let Inst{20} = load; + let Inst{19-16} = addr; + let Inst{15-12} = CRd; + let Inst{11-8} = cop; + let Inst{7-0} = offset{7-0}; + let DecoderMethod = "DecodeCopMemInstruction"; } - def _OPTION : ACI<(outs), - !con((ins nohash_imm:$cop,nohash_imm:$CRd,GPR:$base, nohash_imm:$option), - ops), - !strconcat(opc, cond), "\tp$cop, cr$CRd, [$base], \\{$option\\}"> { - let Inst{31-28} = op31_28; + (ins p_imm:$cop, c_imm:$CRd, addr_offset_none:$addr, + coproc_option_imm:$option), + asm, "\t$cop, $CRd, $addr, $option"> { + bits<8> option; + bits<4> addr; + bits<4> cop; + bits<4> CRd; let Inst{24} = 0; // P = 0 let Inst{23} = 1; // U = 1 + let Inst{22} = Dbit; let Inst{21} = 0; // W = 0 - let Inst{22} = 0; // D = 0 let Inst{20} = load; + let Inst{19-16} = addr; + let Inst{15-12} = CRd; + let Inst{11-8} = cop; + let Inst{7-0} = option; + let DecoderMethod = "DecodeCopMemInstruction"; } - - def L_OFFSET : ACI<(outs), - !con((ins nohash_imm:$cop, nohash_imm:$CRd, addrmode2:$addr), ops), - !strconcat(!strconcat(opc, "l"), cond), "\tp$cop, cr$CRd, $addr"> { - let Inst{31-28} = op31_28; +} +multiclass LdSt2Cop { + def _OFFSET : ACInoP<(outs), (ins p_imm:$cop, c_imm:$CRd, addrmode5:$addr), + asm, "\t$cop, $CRd, $addr"> { + bits<13> addr; + bits<4> cop; + bits<4> CRd; let Inst{24} = 1; // P = 1 + let Inst{23} = addr{8}; + let Inst{22} = Dbit; let Inst{21} = 0; // W = 0 - let Inst{22} = 1; // D = 1 let Inst{20} = load; + let Inst{19-16} = addr{12-9}; + let Inst{15-12} = CRd; + let Inst{11-8} = cop; + let Inst{7-0} = addr{7-0}; + let DecoderMethod = "DecodeCopMemInstruction"; } - - def L_PRE : ACI<(outs), - !con((ins nohash_imm:$cop, nohash_imm:$CRd, addrmode2:$addr), ops), - !strconcat(!strconcat(opc, "l"), cond), "\tp$cop, cr$CRd, $addr!", - IndexModePre> { - let Inst{31-28} = op31_28; + def _PRE : ACInoP<(outs), (ins p_imm:$cop, c_imm:$CRd, addrmode5:$addr), + asm, "\t$cop, $CRd, $addr!", IndexModePre> { + bits<13> addr; + bits<4> cop; + bits<4> CRd; let Inst{24} = 1; // P = 1 + let Inst{23} = addr{8}; + let Inst{22} = Dbit; let Inst{21} = 1; // W = 1 - let Inst{22} = 1; // D = 1 let Inst{20} = load; + let Inst{19-16} = addr{12-9}; + let Inst{15-12} = CRd; + let Inst{11-8} = cop; + let Inst{7-0} = addr{7-0}; + let DecoderMethod = "DecodeCopMemInstruction"; } - - def L_POST : ACI<(outs), - !con((ins nohash_imm:$cop, nohash_imm:$CRd, addrmode2:$addr), ops), - !strconcat(!strconcat(opc, "l"), cond), "\tp$cop, cr$CRd, $addr", - IndexModePost> { - let Inst{31-28} = op31_28; + def _POST: ACInoP<(outs), (ins p_imm:$cop, c_imm:$CRd, addr_offset_none:$addr, + postidx_imm8s4:$offset), + asm, "\t$cop, $CRd, $addr, $offset", IndexModePost> { + bits<9> offset; + bits<4> addr; + bits<4> cop; + bits<4> CRd; let Inst{24} = 0; // P = 0 + let Inst{23} = offset{8}; + let Inst{22} = Dbit; let Inst{21} = 1; // W = 1 - let Inst{22} = 1; // D = 1 let Inst{20} = load; + let Inst{19-16} = addr; + let Inst{15-12} = CRd; + let Inst{11-8} = cop; + let Inst{7-0} = offset{7-0}; + let DecoderMethod = "DecodeCopMemInstruction"; } - - def L_OPTION : ACI<(outs), - !con((ins nohash_imm:$cop, nohash_imm:$CRd,GPR:$base,nohash_imm:$option), - ops), - !strconcat(!strconcat(opc, "l"), cond), - "\tp$cop, cr$CRd, [$base], \\{$option\\}"> { - let Inst{31-28} = op31_28; + def _OPTION : ACInoP<(outs), + (ins p_imm:$cop, c_imm:$CRd, addr_offset_none:$addr, + coproc_option_imm:$option), + asm, "\t$cop, $CRd, $addr, $option"> { + bits<8> option; + bits<4> addr; + bits<4> cop; + bits<4> CRd; let Inst{24} = 0; // P = 0 let Inst{23} = 1; // U = 1 + let Inst{22} = Dbit; let Inst{21} = 0; // W = 0 - let Inst{22} = 1; // D = 1 let Inst{20} = load; + let Inst{19-16} = addr; + let Inst{15-12} = CRd; + let Inst{11-8} = cop; + let Inst{7-0} = option; + let DecoderMethod = "DecodeCopMemInstruction"; } } -defm LDC : LdStCop<{?,?,?,?}, 1, (ins pred:$p), "ldc", "${p}">; -defm LDC2 : LdStCop<0b1111, 1, (ins), "ldc2", "">; -defm STC : LdStCop<{?,?,?,?}, 0, (ins pred:$p), "stc", "${p}">; -defm STC2 : LdStCop<0b1111, 0, (ins), "stc2", "">; +defm LDC : LdStCop <1, 0, "ldc">; +defm LDCL : LdStCop <1, 1, "ldcl">; +defm STC : LdStCop <0, 0, "stc">; +defm STCL : LdStCop <0, 1, "stcl">; +defm LDC2 : LdSt2Cop<1, 0, "ldc2">; +defm LDC2L : LdSt2Cop<1, 1, "ldc2l">; +defm STC2 : LdSt2Cop<0, 0, "stc2">; +defm STC2L : LdSt2Cop<0, 1, "stc2l">; //===----------------------------------------------------------------------===// -// Move between coprocessor and ARM core register -- for disassembly only +// Move between coprocessor and ARM core register. // class MovRCopro; def MRC : MovRCopro<"mrc", 1 /* from coprocessor to ARM core register */, (outs GPR:$Rt), - (ins p_imm:$cop, i32imm:$opc1, c_imm:$CRn, c_imm:$CRm, - i32imm:$opc2), []>; + (ins p_imm:$cop, imm0_7:$opc1, c_imm:$CRn, c_imm:$CRm, + imm0_7:$opc2), []>; def : ARMPat<(int_arm_mrc imm:$cop, imm:$opc1, imm:$CRn, imm:$CRm, imm:$opc2), (MRC imm:$cop, imm:$opc1, imm:$CRn, imm:$CRm, imm:$opc2)>; @@ -3697,15 +4490,14 @@ def MCR2 : MovRCopro2<"mcr2", 0 /* from ARM core register to coprocessor */, imm:$CRm, imm:$opc2)]>; def MRC2 : MovRCopro2<"mrc2", 1 /* from coprocessor to ARM core register */, (outs GPR:$Rt), - (ins p_imm:$cop, i32imm:$opc1, c_imm:$CRn, c_imm:$CRm, - i32imm:$opc2), []>; + (ins p_imm:$cop, imm0_7:$opc1, c_imm:$CRn, c_imm:$CRm, + imm0_7:$opc2), []>; def : ARMV5TPat<(int_arm_mrc2 imm:$cop, imm:$opc1, imm:$CRn, imm:$CRm, imm:$opc2), (MRC2 imm:$cop, imm:$opc1, imm:$CRn, imm:$CRm, imm:$opc2)>; -class MovRRCopro pattern = [/* For disassembly only */]> +class MovRRCopro pattern = []> : ABI<0b1100, (outs), (ins p_imm:$cop, imm0_15:$opc1, GPR:$Rt, GPR:$Rt2, c_imm:$CRm), NoItinerary, opc, "\t$cop, $opc1, $Rt, $Rt2, $CRm", pattern> { @@ -3730,8 +4522,7 @@ def MCRR : MovRRCopro<"mcrr", 0 /* from ARM core register to coprocessor */, imm:$CRm)]>; def MRRC : MovRRCopro<"mrrc", 1 /* from coprocessor to ARM core register */>; -class MovRRCopro2 pattern = [/* For disassembly only */]> +class MovRRCopro2 pattern = []> : ABXI<0b1100, (outs), (ins p_imm:$cop, imm0_15:$opc1, GPR:$Rt, GPR:$Rt2, c_imm:$CRm), NoItinerary, !strconcat(opc, "\t$cop, $opc1, $Rt, $Rt2, $CRm"), pattern> { @@ -3758,20 +4549,22 @@ def MCRR2 : MovRRCopro2<"mcrr2", 0 /* from ARM core register to coprocessor */, def MRRC2 : MovRRCopro2<"mrrc2", 1 /* from coprocessor to ARM core register */>; //===----------------------------------------------------------------------===// -// Move between special register and ARM core register -- for disassembly only +// Move between special register and ARM core register // // Move to ARM core register from Special Register -def MRS : ABI<0b0001, (outs GPR:$Rd), (ins), NoItinerary, "mrs", "\t$Rd, cpsr", - [/* For disassembly only; pattern left blank */]> { +def MRS : ABI<0b0001, (outs GPR:$Rd), (ins), NoItinerary, + "mrs", "\t$Rd, apsr", []> { bits<4> Rd; let Inst{23-16} = 0b00001111; let Inst{15-12} = Rd; let Inst{7-4} = 0b0000; } -def MRSsys : ABI<0b0001, (outs GPR:$Rd), (ins), NoItinerary,"mrs","\t$Rd, spsr", - [/* For disassembly only; pattern left blank */]> { +def : InstAlias<"mrs${p} $Rd, cpsr", (MRS GPR:$Rd, pred:$p)>, Requires<[IsARM]>; + +def MRSsys : ABI<0b0001, (outs GPR:$Rd), (ins), NoItinerary, + "mrs", "\t$Rd, spsr", []> { bits<4> Rd; let Inst{23-16} = 0b01001111; let Inst{15-12} = Rd; @@ -3785,8 +4578,7 @@ def MRSsys : ABI<0b0001, (outs GPR:$Rd), (ins), NoItinerary,"mrs","\t$Rd, spsr", // operand contains the special register (R Bit) in bit 4 and bits 3-0 contains // the mask with the fields to be accessed in the special register. def MSR : ABI<0b0001, (outs), (ins msr_mask:$mask, GPR:$Rn), NoItinerary, - "msr", "\t$mask, $Rn", - [/* For disassembly only; pattern left blank */]> { + "msr", "\t$mask, $Rn", []> { bits<5> mask; bits<4> Rn; @@ -3800,8 +4592,7 @@ def MSR : ABI<0b0001, (outs), (ins msr_mask:$mask, GPR:$Rn), NoItinerary, } def MSRi : ABI<0b0011, (outs), (ins msr_mask:$mask, so_imm:$a), NoItinerary, - "msr", "\t$mask, $a", - [/* For disassembly only; pattern left blank */]> { + "msr", "\t$mask, $a", []> { bits<5> mask; bits<12> a; @@ -4030,6 +4821,47 @@ def : ARMV5TEPat<(add GPR:$acc, def : ARMPat<(ARMMemBarrierMCR GPR:$zero), (MCR 15, 0, GPR:$zero, 7, 10, 5)>, Requires<[IsARM, HasV6]>; +// SXT/UXT with no rotate +let AddedComplexity = 16 in { +def : ARMV6Pat<(and GPR:$Src, 0x000000FF), (UXTB GPR:$Src, 0)>; +def : ARMV6Pat<(and GPR:$Src, 0x0000FFFF), (UXTH GPR:$Src, 0)>; +def : ARMV6Pat<(and GPR:$Src, 0x00FF00FF), (UXTB16 GPR:$Src, 0)>; +def : ARMV6Pat<(add GPR:$Rn, (and GPR:$Rm, 0x00FF)), + (UXTAB GPR:$Rn, GPR:$Rm, 0)>; +def : ARMV6Pat<(add GPR:$Rn, (and GPR:$Rm, 0xFFFF)), + (UXTAH GPR:$Rn, GPR:$Rm, 0)>; +} + +def : ARMV6Pat<(sext_inreg GPR:$Src, i8), (SXTB GPR:$Src, 0)>; +def : ARMV6Pat<(sext_inreg GPR:$Src, i16), (SXTH GPR:$Src, 0)>; + +def : ARMV6Pat<(add GPR:$Rn, (sext_inreg GPRnopc:$Rm, i8)), + (SXTAB GPR:$Rn, GPRnopc:$Rm, 0)>; +def : ARMV6Pat<(add GPR:$Rn, (sext_inreg GPRnopc:$Rm, i16)), + (SXTAH GPR:$Rn, GPRnopc:$Rm, 0)>; + +// Atomic load/store patterns +def : ARMPat<(atomic_load_8 ldst_so_reg:$src), + (LDRBrs ldst_so_reg:$src)>; +def : ARMPat<(atomic_load_8 addrmode_imm12:$src), + (LDRBi12 addrmode_imm12:$src)>; +def : ARMPat<(atomic_load_16 addrmode3:$src), + (LDRH addrmode3:$src)>; +def : ARMPat<(atomic_load_32 ldst_so_reg:$src), + (LDRrs ldst_so_reg:$src)>; +def : ARMPat<(atomic_load_32 addrmode_imm12:$src), + (LDRi12 addrmode_imm12:$src)>; +def : ARMPat<(atomic_store_8 ldst_so_reg:$ptr, GPR:$val), + (STRBrs GPR:$val, ldst_so_reg:$ptr)>; +def : ARMPat<(atomic_store_8 addrmode_imm12:$ptr, GPR:$val), + (STRBi12 GPR:$val, addrmode_imm12:$ptr)>; +def : ARMPat<(atomic_store_16 addrmode3:$ptr, GPR:$val), + (STRH GPR:$val, addrmode3:$ptr)>; +def : ARMPat<(atomic_store_32 ldst_so_reg:$ptr, GPR:$val), + (STRrs GPR:$val, ldst_so_reg:$ptr)>; +def : ARMPat<(atomic_store_32 addrmode_imm12:$ptr, GPR:$val), + (STRi12 GPR:$val, addrmode_imm12:$ptr)>; + //===----------------------------------------------------------------------===// // Thumb Support @@ -4070,7 +4902,103 @@ def : MnemonicAlias<"swi", "svc">; // Load / Store Multiple def : MnemonicAlias<"ldmfd", "ldm">; def : MnemonicAlias<"ldmia", "ldm">; +def : MnemonicAlias<"ldmea", "ldmdb">; def : MnemonicAlias<"stmfd", "stmdb">; def : MnemonicAlias<"stmia", "stm">; def : MnemonicAlias<"stmea", "stm">; +// PKHBT/PKHTB with default shift amount. PKHTB is equivalent to PKHBT when the +// shift amount is zero (i.e., unspecified). +def : InstAlias<"pkhbt${p} $Rd, $Rn, $Rm", + (PKHBT GPRnopc:$Rd, GPRnopc:$Rn, GPRnopc:$Rm, 0, pred:$p)>, + Requires<[IsARM, HasV6]>; +def : InstAlias<"pkhtb${p} $Rd, $Rn, $Rm", + (PKHBT GPRnopc:$Rd, GPRnopc:$Rn, GPRnopc:$Rm, 0, pred:$p)>, + Requires<[IsARM, HasV6]>; + +// PUSH/POP aliases for STM/LDM +def : ARMInstAlias<"push${p} $regs", (STMDB_UPD SP, pred:$p, reglist:$regs)>; +def : ARMInstAlias<"pop${p} $regs", (LDMIA_UPD SP, pred:$p, reglist:$regs)>; + +// SSAT/USAT optional shift operand. +def : ARMInstAlias<"ssat${p} $Rd, $sat_imm, $Rn", + (SSAT GPRnopc:$Rd, imm1_32:$sat_imm, GPRnopc:$Rn, 0, pred:$p)>; +def : ARMInstAlias<"usat${p} $Rd, $sat_imm, $Rn", + (USAT GPRnopc:$Rd, imm0_31:$sat_imm, GPRnopc:$Rn, 0, pred:$p)>; + + +// Extend instruction optional rotate operand. +def : ARMInstAlias<"sxtab${p} $Rd, $Rn, $Rm", + (SXTAB GPRnopc:$Rd, GPR:$Rn, GPRnopc:$Rm, 0, pred:$p)>; +def : ARMInstAlias<"sxtah${p} $Rd, $Rn, $Rm", + (SXTAH GPRnopc:$Rd, GPR:$Rn, GPRnopc:$Rm, 0, pred:$p)>; +def : ARMInstAlias<"sxtab16${p} $Rd, $Rn, $Rm", + (SXTAB16 GPRnopc:$Rd, GPR:$Rn, GPRnopc:$Rm, 0, pred:$p)>; +def : ARMInstAlias<"sxtb${p} $Rd, $Rm", + (SXTB GPRnopc:$Rd, GPRnopc:$Rm, 0, pred:$p)>; +def : ARMInstAlias<"sxtb16${p} $Rd, $Rm", + (SXTB16 GPRnopc:$Rd, GPRnopc:$Rm, 0, pred:$p)>; +def : ARMInstAlias<"sxth${p} $Rd, $Rm", + (SXTH GPRnopc:$Rd, GPRnopc:$Rm, 0, pred:$p)>; + +def : ARMInstAlias<"uxtab${p} $Rd, $Rn, $Rm", + (UXTAB GPRnopc:$Rd, GPR:$Rn, GPRnopc:$Rm, 0, pred:$p)>; +def : ARMInstAlias<"uxtah${p} $Rd, $Rn, $Rm", + (UXTAH GPRnopc:$Rd, GPR:$Rn, GPRnopc:$Rm, 0, pred:$p)>; +def : ARMInstAlias<"uxtab16${p} $Rd, $Rn, $Rm", + (UXTAB16 GPRnopc:$Rd, GPR:$Rn, GPRnopc:$Rm, 0, pred:$p)>; +def : ARMInstAlias<"uxtb${p} $Rd, $Rm", + (UXTB GPRnopc:$Rd, GPRnopc:$Rm, 0, pred:$p)>; +def : ARMInstAlias<"uxtb16${p} $Rd, $Rm", + (UXTB16 GPRnopc:$Rd, GPRnopc:$Rm, 0, pred:$p)>; +def : ARMInstAlias<"uxth${p} $Rd, $Rm", + (UXTH GPRnopc:$Rd, GPRnopc:$Rm, 0, pred:$p)>; + + +// RFE aliases +def : MnemonicAlias<"rfefa", "rfeda">; +def : MnemonicAlias<"rfeea", "rfedb">; +def : MnemonicAlias<"rfefd", "rfeia">; +def : MnemonicAlias<"rfeed", "rfeib">; +def : MnemonicAlias<"rfe", "rfeia">; + +// SRS aliases +def : MnemonicAlias<"srsfa", "srsda">; +def : MnemonicAlias<"srsea", "srsdb">; +def : MnemonicAlias<"srsfd", "srsia">; +def : MnemonicAlias<"srsed", "srsib">; +def : MnemonicAlias<"srs", "srsia">; + +// QSAX == QSUBADDX +def : MnemonicAlias<"qsubaddx", "qsax">; +// SASX == SADDSUBX +def : MnemonicAlias<"saddsubx", "sasx">; +// SHASX == SHADDSUBX +def : MnemonicAlias<"shaddsubx", "shasx">; +// SHSAX == SHSUBADDX +def : MnemonicAlias<"shsubaddx", "shsax">; +// SSAX == SSUBADDX +def : MnemonicAlias<"ssubaddx", "ssax">; +// UASX == UADDSUBX +def : MnemonicAlias<"uaddsubx", "uasx">; +// UHASX == UHADDSUBX +def : MnemonicAlias<"uhaddsubx", "uhasx">; +// UHSAX == UHSUBADDX +def : MnemonicAlias<"uhsubaddx", "uhsax">; +// UQASX == UQADDSUBX +def : MnemonicAlias<"uqaddsubx", "uqasx">; +// UQSAX == UQSUBADDX +def : MnemonicAlias<"uqsubaddx", "uqsax">; +// USAX == USUBADDX +def : MnemonicAlias<"usubaddx", "usax">; + +// LDRSBT/LDRHT/LDRSHT post-index offset if optional. +// Note that the write-back output register is a dummy operand for MC (it's +// only meaningful for codegen), so we just pass zero here. +// FIXME: tblgen not cooperating with argument conversions. +//def : InstAlias<"ldrsbt${p} $Rt, $addr", +// (LDRSBTi GPR:$Rt, GPR:$Rt, addr_offset_none:$addr, 0,pred:$p)>; +//def : InstAlias<"ldrht${p} $Rt, $addr", +// (LDRHTi GPR:$Rt, GPR:$Rt, addr_offset_none:$addr, 0, pred:$p)>; +//def : InstAlias<"ldrsht${p} $Rt, $addr", +// (LDRSHTi GPR:$Rt, GPR:$Rt, addr_offset_none:$addr, 0, pred:$p)>; diff --git a/lib/Target/ARM/ARMInstrNEON.td b/lib/Target/ARM/ARMInstrNEON.td index 0df62f4563436..7aad18695bb3b 100644 --- a/lib/Target/ARM/ARMInstrNEON.td +++ b/lib/Target/ARM/ARMInstrNEON.td @@ -11,6 +11,35 @@ // //===----------------------------------------------------------------------===// + +//===----------------------------------------------------------------------===// +// NEON-specific Operands. +//===----------------------------------------------------------------------===// +def VectorIndex8Operand : AsmOperandClass { let Name = "VectorIndex8"; } +def VectorIndex16Operand : AsmOperandClass { let Name = "VectorIndex16"; } +def VectorIndex32Operand : AsmOperandClass { let Name = "VectorIndex32"; } +def VectorIndex8 : Operand, ImmLeaf { + let ParserMatchClass = VectorIndex8Operand; + let PrintMethod = "printVectorIndex"; + let MIOperandInfo = (ops i32imm); +} +def VectorIndex16 : Operand, ImmLeaf { + let ParserMatchClass = VectorIndex16Operand; + let PrintMethod = "printVectorIndex"; + let MIOperandInfo = (ops i32imm); +} +def VectorIndex32 : Operand, ImmLeaf { + let ParserMatchClass = VectorIndex32Operand; + let PrintMethod = "printVectorIndex"; + let MIOperandInfo = (ops i32imm); +} + //===----------------------------------------------------------------------===// // NEON-specific DAG Nodes. //===----------------------------------------------------------------------===// @@ -175,7 +204,8 @@ class VLDQQWBPseudo (ins addrmode6:$addr, am6offset:$offset), itin, "$addr.addr = $wb">; class VLDQQQQPseudo - : PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src),itin,"">; + : PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src),itin, + "$src = $dst">; class VLDQQQQWBPseudo : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb), (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin, @@ -190,6 +220,7 @@ class VLD1D op7_4, string Dt> "vld1", Dt, "\\{$Vd\\}, $Rn", "", []> { let Rm = 0b1111; let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVLDInstruction"; } class VLD1Q op7_4, string Dt> : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$Vd, DPR:$dst2), @@ -197,6 +228,7 @@ class VLD1Q op7_4, string Dt> "vld1", Dt, "\\{$Vd, $dst2\\}, $Rn", "", []> { let Rm = 0b1111; let Inst{5-4} = Rn{5-4}; + let DecoderMethod = "DecodeVLDInstruction"; } def VLD1d8 : VLD1D<{0,0,0,?}, "8">; @@ -221,6 +253,7 @@ class VLD1DWB op7_4, string Dt> "vld1", Dt, "\\{$Vd\\}, $Rn$Rm", "$Rn.addr = $wb", []> { let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVLDInstruction"; } class VLD1QWB op7_4, string Dt> : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb), @@ -228,6 +261,7 @@ class VLD1QWB op7_4, string Dt> "vld1", Dt, "\\{$Vd, $dst2\\}, $Rn$Rm", "$Rn.addr = $wb", []> { let Inst{5-4} = Rn{5-4}; + let DecoderMethod = "DecodeVLDInstruction"; } def VLD1d8_UPD : VLD1DWB<{0,0,0,?}, "8">; @@ -252,12 +286,14 @@ class VLD1D3 op7_4, string Dt> "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> { let Rm = 0b1111; let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVLDInstruction"; } class VLD1D3WB op7_4, string Dt> : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb), (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x3u, "vld1", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm", "$Rn.addr = $wb", []> { let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVLDInstruction"; } def VLD1d8T : VLD1D3<{0,0,0,?}, "8">; @@ -280,6 +316,7 @@ class VLD1D4 op7_4, string Dt> "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> { let Rm = 0b1111; let Inst{5-4} = Rn{5-4}; + let DecoderMethod = "DecodeVLDInstruction"; } class VLD1D4WB op7_4, string Dt> : NLdSt<0,0b10,0b0010,op7_4, @@ -288,6 +325,7 @@ class VLD1D4WB op7_4, string Dt> "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm", "$Rn.addr = $wb", []> { let Inst{5-4} = Rn{5-4}; + let DecoderMethod = "DecodeVLDInstruction"; } def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">; @@ -310,6 +348,7 @@ class VLD2D op11_8, bits<4> op7_4, string Dt> "vld2", Dt, "\\{$Vd, $dst2\\}, $Rn", "", []> { let Rm = 0b1111; let Inst{5-4} = Rn{5-4}; + let DecoderMethod = "DecodeVLDInstruction"; } class VLD2Q op7_4, string Dt> : NLdSt<0, 0b10, 0b0011, op7_4, @@ -318,6 +357,7 @@ class VLD2Q op7_4, string Dt> "vld2", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> { let Rm = 0b1111; let Inst{5-4} = Rn{5-4}; + let DecoderMethod = "DecodeVLDInstruction"; } def VLD2d8 : VLD2D<0b1000, {0,0,?,?}, "8">; @@ -343,6 +383,7 @@ class VLD2DWB op11_8, bits<4> op7_4, string Dt> "vld2", Dt, "\\{$Vd, $dst2\\}, $Rn$Rm", "$Rn.addr = $wb", []> { let Inst{5-4} = Rn{5-4}; + let DecoderMethod = "DecodeVLDInstruction"; } class VLD2QWB op7_4, string Dt> : NLdSt<0, 0b10, 0b0011, op7_4, @@ -351,6 +392,7 @@ class VLD2QWB op7_4, string Dt> "vld2", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm", "$Rn.addr = $wb", []> { let Inst{5-4} = Rn{5-4}; + let DecoderMethod = "DecodeVLDInstruction"; } def VLD2d8_UPD : VLD2DWB<0b1000, {0,0,?,?}, "8">; @@ -384,6 +426,7 @@ class VLD3D op11_8, bits<4> op7_4, string Dt> "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> { let Rm = 0b1111; let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVLDInstruction"; } def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">; @@ -402,6 +445,7 @@ class VLD3DWB op11_8, bits<4> op7_4, string Dt> "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm", "$Rn.addr = $wb", []> { let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVLDInstruction"; } def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">; @@ -441,6 +485,7 @@ class VLD4D op11_8, bits<4> op7_4, string Dt> "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> { let Rm = 0b1111; let Inst{5-4} = Rn{5-4}; + let DecoderMethod = "DecodeVLDInstruction"; } def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">; @@ -459,6 +504,7 @@ class VLD4DWB op11_8, bits<4> op7_4, string Dt> "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm", "$Rn.addr = $wb", []> { let Inst{5-4} = Rn{5-4}; + let DecoderMethod = "DecodeVLDInstruction"; } def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">; @@ -530,6 +576,7 @@ class VLD1LN op11_8, bits<4> op7_4, string Dt, ValueType Ty, (i32 (LoadOp addrmode6:$Rn)), imm:$lane))]> { let Rm = 0b1111; + let DecoderMethod = "DecodeVLD1LN"; } class VLD1LN32 op11_8, bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp> @@ -541,6 +588,7 @@ class VLD1LN32 op11_8, bits<4> op7_4, string Dt, ValueType Ty, (i32 (LoadOp addrmode6oneL32:$Rn)), imm:$lane))]> { let Rm = 0b1111; + let DecoderMethod = "DecodeVLD1LN"; } class VLD1QLNPseudo : VLDQLNPseudo { let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src), @@ -580,7 +628,9 @@ class VLD1LNWB op11_8, bits<4> op7_4, string Dt> (ins addrmode6:$Rn, am6offset:$Rm, DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn$Rm", - "$src = $Vd, $Rn.addr = $wb", []>; + "$src = $Vd, $Rn.addr = $wb", []> { + let DecoderMethod = "DecodeVLD1LN"; +} def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> { let Inst{7-5} = lane{2-0}; @@ -607,6 +657,7 @@ class VLD2LN op11_8, bits<4> op7_4, string Dt> "$src1 = $Vd, $src2 = $dst2", []> { let Rm = 0b1111; let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVLD2LN"; } def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> { @@ -642,6 +693,7 @@ class VLD2LNWB op11_8, bits<4> op7_4, string Dt> "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm", "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> { let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVLD2LN"; } def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> { @@ -676,6 +728,7 @@ class VLD3LN op11_8, bits<4> op7_4, string Dt> "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn", "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> { let Rm = 0b1111; + let DecoderMethod = "DecodeVLD3LN"; } def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> { @@ -712,7 +765,9 @@ class VLD3LNWB op11_8, bits<4> op7_4, string Dt> IIC_VLD3lnu, "vld3", Dt, "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm", "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb", - []>; + []> { + let DecoderMethod = "DecodeVLD3LN"; +} def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> { let Inst{7-5} = lane{2-0}; @@ -748,6 +803,7 @@ class VLD4LN op11_8, bits<4> op7_4, string Dt> "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> { let Rm = 0b1111; let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVLD4LN"; } def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> { @@ -788,6 +844,7 @@ class VLD4LNWB op11_8, bits<4> op7_4, string Dt> "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb", []> { let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVLD4LN" ; } def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> { @@ -825,6 +882,7 @@ class VLD1DUP op7_4, string Dt, ValueType Ty, PatFrag LoadOp> [(set DPR:$Vd, (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> { let Rm = 0b1111; let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVLD1DupInstruction"; } class VLD1QDUPPseudo : VLDQPseudo { let Pattern = [(set QPR:$dst, @@ -852,6 +910,7 @@ class VLD1QDUP op7_4, string Dt> "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> { let Rm = 0b1111; let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVLD1DupInstruction"; } def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8">; @@ -864,12 +923,14 @@ class VLD1DUPWB op7_4, string Dt> (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD1dupu, "vld1", Dt, "\\{$Vd[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> { let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVLD1DupInstruction"; } class VLD1QDUPWB op7_4, string Dt> : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb), (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD1dupu, "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> { let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVLD1DupInstruction"; } def VLD1DUPd8_UPD : VLD1DUPWB<{0,0,0,0}, "8">; @@ -891,6 +952,7 @@ class VLD2DUP op7_4, string Dt> "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> { let Rm = 0b1111; let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVLD2DupInstruction"; } def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8">; @@ -912,6 +974,7 @@ class VLD2DUPWB op7_4, string Dt> (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD2dupu, "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> { let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVLD2DupInstruction"; } def VLD2DUPd8_UPD : VLD2DUPWB<{0,0,0,0}, "8">; @@ -932,7 +995,8 @@ class VLD3DUP op7_4, string Dt> (ins addrmode6dup:$Rn), IIC_VLD3dup, "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []> { let Rm = 0b1111; - let Inst{4} = Rn{4}; + let Inst{4} = 0; + let DecoderMethod = "DecodeVLD3DupInstruction"; } def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">; @@ -954,7 +1018,8 @@ class VLD3DUPWB op7_4, string Dt> (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD3dupu, "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> { - let Inst{4} = Rn{4}; + let Inst{4} = 0; + let DecoderMethod = "DecodeVLD3DupInstruction"; } def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8">; @@ -977,6 +1042,7 @@ class VLD4DUP op7_4, string Dt> "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> { let Rm = 0b1111; let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVLD4DupInstruction"; } def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">; @@ -1000,6 +1066,7 @@ class VLD4DUPWB op7_4, string Dt> "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> { let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVLD4DupInstruction"; } def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">; @@ -1045,6 +1112,7 @@ class VST1D op7_4, string Dt> IIC_VST1, "vst1", Dt, "\\{$Vd\\}, $Rn", "", []> { let Rm = 0b1111; let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVSTInstruction"; } class VST1Q op7_4, string Dt> : NLdSt<0,0b00,0b1010,op7_4, (outs), @@ -1052,6 +1120,7 @@ class VST1Q op7_4, string Dt> "vst1", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> { let Rm = 0b1111; let Inst{5-4} = Rn{5-4}; + let DecoderMethod = "DecodeVSTInstruction"; } def VST1d8 : VST1D<{0,0,0,?}, "8">; @@ -1075,6 +1144,7 @@ class VST1DWB op7_4, string Dt> (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd), IIC_VST1u, "vst1", Dt, "\\{$Vd\\}, $Rn$Rm", "$Rn.addr = $wb", []> { let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVSTInstruction"; } class VST1QWB op7_4, string Dt> : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb), @@ -1082,6 +1152,7 @@ class VST1QWB op7_4, string Dt> IIC_VST1x2u, "vst1", Dt, "\\{$Vd, $src2\\}, $Rn$Rm", "$Rn.addr = $wb", []> { let Inst{5-4} = Rn{5-4}; + let DecoderMethod = "DecodeVSTInstruction"; } def VST1d8_UPD : VST1DWB<{0,0,0,?}, "8">; @@ -1106,6 +1177,7 @@ class VST1D3 op7_4, string Dt> IIC_VST1x3, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> { let Rm = 0b1111; let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVSTInstruction"; } class VST1D3WB op7_4, string Dt> : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb), @@ -1114,6 +1186,7 @@ class VST1D3WB op7_4, string Dt> IIC_VST1x3u, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm", "$Rn.addr = $wb", []> { let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVSTInstruction"; } def VST1d8T : VST1D3<{0,0,0,?}, "8">; @@ -1137,6 +1210,7 @@ class VST1D4 op7_4, string Dt> []> { let Rm = 0b1111; let Inst{5-4} = Rn{5-4}; + let DecoderMethod = "DecodeVSTInstruction"; } class VST1D4WB op7_4, string Dt> : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb), @@ -1145,6 +1219,7 @@ class VST1D4WB op7_4, string Dt> "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm", "$Rn.addr = $wb", []> { let Inst{5-4} = Rn{5-4}; + let DecoderMethod = "DecodeVSTInstruction"; } def VST1d8Q : VST1D4<{0,0,?,?}, "8">; @@ -1167,6 +1242,7 @@ class VST2D op11_8, bits<4> op7_4, string Dt> IIC_VST2, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> { let Rm = 0b1111; let Inst{5-4} = Rn{5-4}; + let DecoderMethod = "DecodeVSTInstruction"; } class VST2Q op7_4, string Dt> : NLdSt<0, 0b00, 0b0011, op7_4, (outs), @@ -1175,6 +1251,7 @@ class VST2Q op7_4, string Dt> "", []> { let Rm = 0b1111; let Inst{5-4} = Rn{5-4}; + let DecoderMethod = "DecodeVSTInstruction"; } def VST2d8 : VST2D<0b1000, {0,0,?,?}, "8">; @@ -1200,6 +1277,7 @@ class VST2DWB op11_8, bits<4> op7_4, string Dt> IIC_VST2u, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn$Rm", "$Rn.addr = $wb", []> { let Inst{5-4} = Rn{5-4}; + let DecoderMethod = "DecodeVSTInstruction"; } class VST2QWB op7_4, string Dt> : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb), @@ -1208,6 +1286,7 @@ class VST2QWB op7_4, string Dt> "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm", "$Rn.addr = $wb", []> { let Inst{5-4} = Rn{5-4}; + let DecoderMethod = "DecodeVSTInstruction"; } def VST2d8_UPD : VST2DWB<0b1000, {0,0,?,?}, "8">; @@ -1241,6 +1320,7 @@ class VST3D op11_8, bits<4> op7_4, string Dt> "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> { let Rm = 0b1111; let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVSTInstruction"; } def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">; @@ -1259,6 +1339,7 @@ class VST3DWB op11_8, bits<4> op7_4, string Dt> "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm", "$Rn.addr = $wb", []> { let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVSTInstruction"; } def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">; @@ -1298,6 +1379,7 @@ class VST4D op11_8, bits<4> op7_4, string Dt> "", []> { let Rm = 0b1111; let Inst{5-4} = Rn{5-4}; + let DecoderMethod = "DecodeVSTInstruction"; } def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">; @@ -1316,6 +1398,7 @@ class VST4DWB op11_8, bits<4> op7_4, string Dt> "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm", "$Rn.addr = $wb", []> { let Inst{5-4} = Rn{5-4}; + let DecoderMethod = "DecodeVSTInstruction"; } def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">; @@ -1381,6 +1464,7 @@ class VST1LN op11_8, bits<4> op7_4, string Dt, ValueType Ty, IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "", [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6:$Rn)]> { let Rm = 0b1111; + let DecoderMethod = "DecodeVST1LN"; } class VST1LN32 op11_8, bits<4> op7_4, string Dt, ValueType Ty, PatFrag StoreOp, SDNode ExtractOp> @@ -1389,6 +1473,7 @@ class VST1LN32 op11_8, bits<4> op7_4, string Dt, ValueType Ty, IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "", [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6oneL32:$Rn)]>{ let Rm = 0b1111; + let DecoderMethod = "DecodeVST1LN"; } class VST1QLNPseudo : VSTQLNPseudo { @@ -1429,7 +1514,9 @@ class VST1LNWB op11_8, bits<4> op7_4, string Dt, ValueType Ty, "\\{$Vd[$lane]\\}, $Rn$Rm", "$Rn.addr = $wb", [(set GPR:$wb, (StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), - addrmode6:$Rn, am6offset:$Rm))]>; + addrmode6:$Rn, am6offset:$Rm))]> { + let DecoderMethod = "DecodeVST1LN"; +} class VST1QLNWBPseudo : VSTQLNWBPseudo { let Pattern = [(set GPR:$wb, (StoreOp (ExtractOp (Ty QPR:$src), imm:$lane), @@ -1465,6 +1552,7 @@ class VST2LN op11_8, bits<4> op7_4, string Dt> "", []> { let Rm = 0b1111; let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVST2LN"; } def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> { @@ -1502,6 +1590,7 @@ class VST2LNWB op11_8, bits<4> op7_4, string Dt> "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset", "$addr.addr = $wb", []> { let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVST2LN"; } def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> { @@ -1535,6 +1624,7 @@ class VST3LN op11_8, bits<4> op7_4, string Dt> nohash_imm:$lane), IIC_VST3ln, "vst3", Dt, "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> { let Rm = 0b1111; + let DecoderMethod = "DecodeVST3LN"; } def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> { @@ -1569,7 +1659,9 @@ class VST3LNWB op11_8, bits<4> op7_4, string Dt> DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane), IIC_VST3lnu, "vst3", Dt, "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm", - "$Rn.addr = $wb", []>; + "$Rn.addr = $wb", []> { + let DecoderMethod = "DecodeVST3LN"; +} def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> { let Inst{7-5} = lane{2-0}; @@ -1604,6 +1696,7 @@ class VST4LN op11_8, bits<4> op7_4, string Dt> "", []> { let Rm = 0b1111; let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVST4LN"; } def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> { @@ -1642,6 +1735,7 @@ class VST4LNWB op11_8, bits<4> op7_4, string Dt> "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm", "$Rn.addr = $wb", []> { let Inst{4} = Rn{4}; + let DecoderMethod = "DecodeVST4LN"; } def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> { @@ -4039,6 +4133,7 @@ class N2VLShMax op21_16, bits<4> op11_8, bit op7, : N2VLSh { let Inst{21-16} = op21_16; + let DecoderMethod = "DecodeVSHLMaxInstruction"; } def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8", v8i16, v8i8, NEONvshlli>; @@ -4219,16 +4314,6 @@ def : InstAlias<"vmov${p} $Vd, $Vm", def : InstAlias<"vmov${p} $Vd, $Vm", (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>; -let neverHasSideEffects = 1 in { -// Pseudo vector move instructions for QQ and QQQQ registers. This should -// be expanded after register allocation is completed. -def VMOVQQ : PseudoInst<(outs QQPR:$dst), (ins QQPR:$src), - NoItinerary, []>; - -def VMOVQQQQ : PseudoInst<(outs QQQQPR:$dst), (ins QQQQPR:$src), - NoItinerary, []>; -} // neverHasSideEffects - // VMOV : Vector Move (Immediate) let isReMaterializable = 1 in { @@ -4462,36 +4547,42 @@ def : Pat<(v4f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32q GPR:$R)>; // VDUP : Vector Duplicate Lane (from scalar to all elements) class VDUPLND op19_16, string OpcodeStr, string Dt, - ValueType Ty> - : NVDupLane + : NVDupLane; class VDUPLNQ op19_16, string OpcodeStr, string Dt, - ValueType ResTy, ValueType OpTy> - : NVDupLane + : NVDupLane; + VectorIndex32:$lane)))]>; // Inst{19-16} is partially specified depending on the element size. -def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8> { +def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8, VectorIndex8> { + bits<3> lane; let Inst{19-17} = lane{2-0}; } -def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16> { +def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16, VectorIndex16> { + bits<2> lane; let Inst{19-18} = lane{1-0}; } -def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32> { +def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32, VectorIndex32> { + bits<1> lane; let Inst{19} = lane{0}; } -def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8> { +def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8, VectorIndex8> { + bits<3> lane; let Inst{19-17} = lane{2-0}; } -def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16> { +def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16, VectorIndex16> { + bits<2> lane; let Inst{19-18} = lane{1-0}; } -def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32> { +def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32, VectorIndex32> { + bits<1> lane; let Inst{19} = lane{0}; } @@ -4753,6 +4844,7 @@ def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">; // Vector Table Lookup and Table Extension. // VTBL : Vector Table Lookup +let DecoderMethod = "DecodeTBLInstruction" in { def VTBL1 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1, @@ -4815,6 +4907,7 @@ def VTBX3Pseudo def VTBX4Pseudo : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src), IIC_VTBX4, "$orig = $dst", []>; +} // DecoderMethod = "DecodeTBLInstruction" //===----------------------------------------------------------------------===// // NEON instructions for single-precision FP math diff --git a/lib/Target/ARM/ARMInstrThumb.td b/lib/Target/ARM/ARMInstrThumb.td index bfe83eceb13fb..cedb54799db0d 100644 --- a/lib/Target/ARM/ARMInstrThumb.td +++ b/lib/Target/ARM/ARMInstrThumb.td @@ -19,6 +19,19 @@ def ARMtcall : SDNode<"ARMISD::tCALL", SDT_ARMcall, [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue, SDNPVariadic]>; +def imm_sr_XFORM: SDNodeXFormgetZExtValue(); + return CurDAG->getTargetConstant((Imm == 32 ? 0 : Imm), MVT::i32); +}]>; +def ThumbSRImmAsmOperand: AsmOperandClass { let Name = "ImmThumbSR"; } +def imm_sr : Operand, PatLeaf<(imm), [{ + uint64_t Imm = N->getZExtValue(); + return Imm > 0 && Imm <= 32; +}], imm_sr_XFORM> { + let PrintMethod = "printThumbSRImm"; + let ParserMatchClass = ThumbSRImmAsmOperand; +} + def imm_neg_XFORM : SDNodeXFormgetTargetConstant(-(int)N->getZExtValue(), MVT::i32); }]>; @@ -30,10 +43,6 @@ def imm0_7_neg : PatLeaf<(i32 imm), [{ return (uint32_t)-N->getZExtValue() < 8; }], imm_neg_XFORM>; -def imm0_255_asmoperand : AsmOperandClass { let Name = "Imm0_255"; } -def imm0_255 : Operand, ImmLeaf= 0 && Imm < 256; }]> { - let ParserMatchClass = imm0_255_asmoperand; -} def imm0_255_comp : PatLeaf<(i32 imm), [{ return ~((uint32_t)N->getZExtValue()) < 256; }]>; @@ -69,8 +78,17 @@ def t_adrlabel : Operand { } // Scaled 4 immediate. -def t_imm_s4 : Operand { +def t_imm0_1020s4_asmoperand: AsmOperandClass { let Name = "Imm0_1020s4"; } +def t_imm0_1020s4 : Operand { + let PrintMethod = "printThumbS4ImmOperand"; + let ParserMatchClass = t_imm0_1020s4_asmoperand; + let OperandType = "OPERAND_IMMEDIATE"; +} + +def t_imm0_508s4_asmoperand: AsmOperandClass { let Name = "Imm0_508s4"; } +def t_imm0_508s4 : Operand { let PrintMethod = "printThumbS4ImmOperand"; + let ParserMatchClass = t_imm0_508s4_asmoperand; let OperandType = "OPERAND_IMMEDIATE"; } @@ -79,113 +97,129 @@ def t_imm_s4 : Operand { let OperandType = "OPERAND_PCREL" in { def t_brtarget : Operand { let EncoderMethod = "getThumbBRTargetOpValue"; + let DecoderMethod = "DecodeThumbBROperand"; } def t_bcctarget : Operand { let EncoderMethod = "getThumbBCCTargetOpValue"; + let DecoderMethod = "DecodeThumbBCCTargetOperand"; } def t_cbtarget : Operand { let EncoderMethod = "getThumbCBTargetOpValue"; + let DecoderMethod = "DecodeThumbCmpBROperand"; } def t_bltarget : Operand { let EncoderMethod = "getThumbBLTargetOpValue"; + let DecoderMethod = "DecodeThumbBLTargetOperand"; } def t_blxtarget : Operand { let EncoderMethod = "getThumbBLXTargetOpValue"; + let DecoderMethod = "DecodeThumbBLXOffset"; } } -def MemModeRegThumbAsmOperand : AsmOperandClass { - let Name = "MemModeRegThumb"; - let SuperClasses = []; -} - -def MemModeImmThumbAsmOperand : AsmOperandClass { - let Name = "MemModeImmThumb"; - let SuperClasses = []; -} - // t_addrmode_rr := reg + reg // +def t_addrmode_rr_asm_operand : AsmOperandClass { let Name = "MemThumbRR"; } def t_addrmode_rr : Operand, ComplexPattern { let EncoderMethod = "getThumbAddrModeRegRegOpValue"; let PrintMethod = "printThumbAddrModeRROperand"; + let DecoderMethod = "DecodeThumbAddrModeRR"; + let ParserMatchClass = t_addrmode_rr_asm_operand; let MIOperandInfo = (ops tGPR:$base, tGPR:$offsreg); } // t_addrmode_rrs := reg + reg // +// We use separate scaled versions because the Select* functions need +// to explicitly check for a matching constant and return false here so that +// the reg+imm forms will match instead. This is a horrible way to do that, +// as it forces tight coupling between the methods, but it's how selectiondag +// currently works. def t_addrmode_rrs1 : Operand, ComplexPattern { let EncoderMethod = "getThumbAddrModeRegRegOpValue"; let PrintMethod = "printThumbAddrModeRROperand"; + let DecoderMethod = "DecodeThumbAddrModeRR"; + let ParserMatchClass = t_addrmode_rr_asm_operand; let MIOperandInfo = (ops tGPR:$base, tGPR:$offsreg); - let ParserMatchClass = MemModeRegThumbAsmOperand; } def t_addrmode_rrs2 : Operand, ComplexPattern { let EncoderMethod = "getThumbAddrModeRegRegOpValue"; + let DecoderMethod = "DecodeThumbAddrModeRR"; let PrintMethod = "printThumbAddrModeRROperand"; + let ParserMatchClass = t_addrmode_rr_asm_operand; let MIOperandInfo = (ops tGPR:$base, tGPR:$offsreg); - let ParserMatchClass = MemModeRegThumbAsmOperand; } def t_addrmode_rrs4 : Operand, ComplexPattern { let EncoderMethod = "getThumbAddrModeRegRegOpValue"; + let DecoderMethod = "DecodeThumbAddrModeRR"; let PrintMethod = "printThumbAddrModeRROperand"; + let ParserMatchClass = t_addrmode_rr_asm_operand; let MIOperandInfo = (ops tGPR:$base, tGPR:$offsreg); - let ParserMatchClass = MemModeRegThumbAsmOperand; } // t_addrmode_is4 := reg + imm5 * 4 // +def t_addrmode_is4_asm_operand : AsmOperandClass { let Name = "MemThumbRIs4"; } def t_addrmode_is4 : Operand, ComplexPattern { let EncoderMethod = "getAddrModeISOpValue"; + let DecoderMethod = "DecodeThumbAddrModeIS"; let PrintMethod = "printThumbAddrModeImm5S4Operand"; + let ParserMatchClass = t_addrmode_is4_asm_operand; let MIOperandInfo = (ops tGPR:$base, i32imm:$offsimm); - let ParserMatchClass = MemModeImmThumbAsmOperand; } // t_addrmode_is2 := reg + imm5 * 2 // +def t_addrmode_is2_asm_operand : AsmOperandClass { let Name = "MemThumbRIs2"; } def t_addrmode_is2 : Operand, ComplexPattern { let EncoderMethod = "getAddrModeISOpValue"; + let DecoderMethod = "DecodeThumbAddrModeIS"; let PrintMethod = "printThumbAddrModeImm5S2Operand"; + let ParserMatchClass = t_addrmode_is2_asm_operand; let MIOperandInfo = (ops tGPR:$base, i32imm:$offsimm); - let ParserMatchClass = MemModeImmThumbAsmOperand; } // t_addrmode_is1 := reg + imm5 // +def t_addrmode_is1_asm_operand : AsmOperandClass { let Name = "MemThumbRIs1"; } def t_addrmode_is1 : Operand, ComplexPattern { let EncoderMethod = "getAddrModeISOpValue"; + let DecoderMethod = "DecodeThumbAddrModeIS"; let PrintMethod = "printThumbAddrModeImm5S1Operand"; + let ParserMatchClass = t_addrmode_is1_asm_operand; let MIOperandInfo = (ops tGPR:$base, i32imm:$offsimm); - let ParserMatchClass = MemModeImmThumbAsmOperand; } // t_addrmode_sp := sp + imm8 * 4 // +// FIXME: This really shouldn't have an explicit SP operand at all. It should +// be implicit, just like in the instruction encoding itself. +def t_addrmode_sp_asm_operand : AsmOperandClass { let Name = "MemThumbSPI"; } def t_addrmode_sp : Operand, ComplexPattern { let EncoderMethod = "getAddrModeThumbSPOpValue"; + let DecoderMethod = "DecodeThumbAddrModeSP"; let PrintMethod = "printThumbAddrModeSPOperand"; + let ParserMatchClass = t_addrmode_sp_asm_operand; let MIOperandInfo = (ops GPR:$base, i32imm:$offsimm); - let ParserMatchClass = MemModeImmThumbAsmOperand; } // t_addrmode_pc :=