mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-11-01 16:33:37 +01:00
6dc21c7358
TargetInstrInfo: Change produceSameValue() to take MachineRegisterInfo as an optional argument. When in SSA form, targets can use it to make more aggressive equality analysis. Machine LICM: 1. Eliminate isLoadFromConstantMemory, use MI.isInvariantLoad instead. 2. Fix a bug which prevent CSE of instructions which are not re-materializable. 3. Use improved form of produceSameValue. ARM: 1. Teach ARM produceSameValue to look pass some PIC labels. 2. Look for operands from different loads of different constant pool entries which have same values. 3. Re-implement PIC GA materialization using movw + movt. Combine the pair with a "add pc" or "ldr [pc]" to form pseudo instructions. This makes it possible to re-materialize the instruction, allow machine LICM to hoist the set of instructions out of the loop and make it possible to CSE them. It's a bit hacky, but it significantly improve code quality. 4. Some minor bug fixes as well. With the fixes, using movw + movt to materialize GAs significantly outperform the load from constantpool method. 186.crafty and 255.vortex improved > 20%, 254.gap and 176.gcc ~10%. llvm-svn: 123905
262 lines
9.6 KiB
TableGen
262 lines
9.6 KiB
TableGen
//===- ARMSchedule.td - ARM Scheduling Definitions ---------*- tablegen -*-===//
|
|
//
|
|
// The LLVM Compiler Infrastructure
|
|
//
|
|
// This file is distributed under the University of Illinois Open Source
|
|
// License. See LICENSE.TXT for details.
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// Instruction Itinerary classes used for ARM
|
|
//
|
|
def IIC_iALUx : InstrItinClass;
|
|
def IIC_iALUi : InstrItinClass;
|
|
def IIC_iALUr : InstrItinClass;
|
|
def IIC_iALUsi : InstrItinClass;
|
|
def IIC_iALUsir : InstrItinClass;
|
|
def IIC_iALUsr : InstrItinClass;
|
|
def IIC_iBITi : InstrItinClass;
|
|
def IIC_iBITr : InstrItinClass;
|
|
def IIC_iBITsi : InstrItinClass;
|
|
def IIC_iBITsr : InstrItinClass;
|
|
def IIC_iUNAr : InstrItinClass;
|
|
def IIC_iUNAsi : InstrItinClass;
|
|
def IIC_iEXTr : InstrItinClass;
|
|
def IIC_iEXTAr : InstrItinClass;
|
|
def IIC_iEXTAsr : InstrItinClass;
|
|
def IIC_iCMPi : InstrItinClass;
|
|
def IIC_iCMPr : InstrItinClass;
|
|
def IIC_iCMPsi : InstrItinClass;
|
|
def IIC_iCMPsr : InstrItinClass;
|
|
def IIC_iTSTi : InstrItinClass;
|
|
def IIC_iTSTr : InstrItinClass;
|
|
def IIC_iTSTsi : InstrItinClass;
|
|
def IIC_iTSTsr : InstrItinClass;
|
|
def IIC_iMOVi : InstrItinClass;
|
|
def IIC_iMOVr : InstrItinClass;
|
|
def IIC_iMOVsi : InstrItinClass;
|
|
def IIC_iMOVsr : InstrItinClass;
|
|
def IIC_iMOVix2 : InstrItinClass;
|
|
def IIC_iMOVix2addpc : InstrItinClass;
|
|
def IIC_iMOVix2ld : InstrItinClass;
|
|
def IIC_iMVNi : InstrItinClass;
|
|
def IIC_iMVNr : InstrItinClass;
|
|
def IIC_iMVNsi : InstrItinClass;
|
|
def IIC_iMVNsr : InstrItinClass;
|
|
def IIC_iCMOVi : InstrItinClass;
|
|
def IIC_iCMOVr : InstrItinClass;
|
|
def IIC_iCMOVsi : InstrItinClass;
|
|
def IIC_iCMOVsr : InstrItinClass;
|
|
def IIC_iCMOVix2 : InstrItinClass;
|
|
def IIC_iMUL16 : InstrItinClass;
|
|
def IIC_iMAC16 : InstrItinClass;
|
|
def IIC_iMUL32 : InstrItinClass;
|
|
def IIC_iMAC32 : InstrItinClass;
|
|
def IIC_iMUL64 : InstrItinClass;
|
|
def IIC_iMAC64 : InstrItinClass;
|
|
def IIC_iLoad_i : InstrItinClass;
|
|
def IIC_iLoad_r : InstrItinClass;
|
|
def IIC_iLoad_si : InstrItinClass;
|
|
def IIC_iLoad_iu : InstrItinClass;
|
|
def IIC_iLoad_ru : InstrItinClass;
|
|
def IIC_iLoad_siu : InstrItinClass;
|
|
def IIC_iLoad_bh_i : InstrItinClass;
|
|
def IIC_iLoad_bh_r : InstrItinClass;
|
|
def IIC_iLoad_bh_si : InstrItinClass;
|
|
def IIC_iLoad_bh_iu : InstrItinClass;
|
|
def IIC_iLoad_bh_ru : InstrItinClass;
|
|
def IIC_iLoad_bh_siu : InstrItinClass;
|
|
def IIC_iLoad_d_i : InstrItinClass;
|
|
def IIC_iLoad_d_r : InstrItinClass;
|
|
def IIC_iLoad_d_ru : InstrItinClass;
|
|
def IIC_iLoad_m : InstrItinClass<0>; // micro-coded
|
|
def IIC_iLoad_mu : InstrItinClass<0>; // micro-coded
|
|
def IIC_iLoad_mBr : InstrItinClass<0>; // micro-coded
|
|
def IIC_iPop : InstrItinClass<0>; // micro-coded
|
|
def IIC_iPop_Br : InstrItinClass<0>; // micro-coded
|
|
def IIC_iLoadiALU : InstrItinClass;
|
|
def IIC_iStore_i : InstrItinClass;
|
|
def IIC_iStore_r : InstrItinClass;
|
|
def IIC_iStore_si : InstrItinClass;
|
|
def IIC_iStore_iu : InstrItinClass;
|
|
def IIC_iStore_ru : InstrItinClass;
|
|
def IIC_iStore_siu : InstrItinClass;
|
|
def IIC_iStore_bh_i : InstrItinClass;
|
|
def IIC_iStore_bh_r : InstrItinClass;
|
|
def IIC_iStore_bh_si : InstrItinClass;
|
|
def IIC_iStore_bh_iu : InstrItinClass;
|
|
def IIC_iStore_bh_ru : InstrItinClass;
|
|
def IIC_iStore_bh_siu : InstrItinClass;
|
|
def IIC_iStore_d_i : InstrItinClass;
|
|
def IIC_iStore_d_r : InstrItinClass;
|
|
def IIC_iStore_d_ru : InstrItinClass;
|
|
def IIC_iStore_m : InstrItinClass<0>; // micro-coded
|
|
def IIC_iStore_mu : InstrItinClass<0>; // micro-coded
|
|
def IIC_Preload : InstrItinClass;
|
|
def IIC_Br : InstrItinClass;
|
|
def IIC_fpSTAT : InstrItinClass;
|
|
def IIC_fpUNA32 : InstrItinClass;
|
|
def IIC_fpUNA64 : InstrItinClass;
|
|
def IIC_fpCMP32 : InstrItinClass;
|
|
def IIC_fpCMP64 : InstrItinClass;
|
|
def IIC_fpCVTSD : InstrItinClass;
|
|
def IIC_fpCVTDS : InstrItinClass;
|
|
def IIC_fpCVTSH : InstrItinClass;
|
|
def IIC_fpCVTHS : InstrItinClass;
|
|
def IIC_fpCVTIS : InstrItinClass;
|
|
def IIC_fpCVTID : InstrItinClass;
|
|
def IIC_fpCVTSI : InstrItinClass;
|
|
def IIC_fpCVTDI : InstrItinClass;
|
|
def IIC_fpMOVIS : InstrItinClass;
|
|
def IIC_fpMOVID : InstrItinClass;
|
|
def IIC_fpMOVSI : InstrItinClass;
|
|
def IIC_fpMOVDI : InstrItinClass;
|
|
def IIC_fpALU32 : InstrItinClass;
|
|
def IIC_fpALU64 : InstrItinClass;
|
|
def IIC_fpMUL32 : InstrItinClass;
|
|
def IIC_fpMUL64 : InstrItinClass;
|
|
def IIC_fpMAC32 : InstrItinClass;
|
|
def IIC_fpMAC64 : InstrItinClass;
|
|
def IIC_fpDIV32 : InstrItinClass;
|
|
def IIC_fpDIV64 : InstrItinClass;
|
|
def IIC_fpSQRT32 : InstrItinClass;
|
|
def IIC_fpSQRT64 : InstrItinClass;
|
|
def IIC_fpLoad32 : InstrItinClass;
|
|
def IIC_fpLoad64 : InstrItinClass;
|
|
def IIC_fpLoad_m : InstrItinClass<0>; // micro-coded
|
|
def IIC_fpLoad_mu : InstrItinClass<0>; // micro-coded
|
|
def IIC_fpStore32 : InstrItinClass;
|
|
def IIC_fpStore64 : InstrItinClass;
|
|
def IIC_fpStore_m : InstrItinClass<0>; // micro-coded
|
|
def IIC_fpStore_mu : InstrItinClass<0>; // micro-coded
|
|
def IIC_VLD1 : InstrItinClass;
|
|
def IIC_VLD1x2 : InstrItinClass;
|
|
def IIC_VLD1x3 : InstrItinClass;
|
|
def IIC_VLD1x4 : InstrItinClass;
|
|
def IIC_VLD1u : InstrItinClass;
|
|
def IIC_VLD1x2u : InstrItinClass;
|
|
def IIC_VLD1x3u : InstrItinClass;
|
|
def IIC_VLD1x4u : InstrItinClass;
|
|
def IIC_VLD1ln : InstrItinClass;
|
|
def IIC_VLD1lnu : InstrItinClass;
|
|
def IIC_VLD1dup : InstrItinClass;
|
|
def IIC_VLD1dupu : InstrItinClass;
|
|
def IIC_VLD2 : InstrItinClass;
|
|
def IIC_VLD2x2 : InstrItinClass;
|
|
def IIC_VLD2u : InstrItinClass;
|
|
def IIC_VLD2x2u : InstrItinClass;
|
|
def IIC_VLD2ln : InstrItinClass;
|
|
def IIC_VLD2lnu : InstrItinClass;
|
|
def IIC_VLD2dup : InstrItinClass;
|
|
def IIC_VLD2dupu : InstrItinClass;
|
|
def IIC_VLD3 : InstrItinClass;
|
|
def IIC_VLD3ln : InstrItinClass;
|
|
def IIC_VLD3u : InstrItinClass;
|
|
def IIC_VLD3lnu : InstrItinClass;
|
|
def IIC_VLD3dup : InstrItinClass;
|
|
def IIC_VLD3dupu : InstrItinClass;
|
|
def IIC_VLD4 : InstrItinClass;
|
|
def IIC_VLD4ln : InstrItinClass;
|
|
def IIC_VLD4u : InstrItinClass;
|
|
def IIC_VLD4lnu : InstrItinClass;
|
|
def IIC_VLD4dup : InstrItinClass;
|
|
def IIC_VLD4dupu : InstrItinClass;
|
|
def IIC_VST1 : InstrItinClass;
|
|
def IIC_VST1x2 : InstrItinClass;
|
|
def IIC_VST1x3 : InstrItinClass;
|
|
def IIC_VST1x4 : InstrItinClass;
|
|
def IIC_VST1u : InstrItinClass;
|
|
def IIC_VST1x2u : InstrItinClass;
|
|
def IIC_VST1x3u : InstrItinClass;
|
|
def IIC_VST1x4u : InstrItinClass;
|
|
def IIC_VST1ln : InstrItinClass;
|
|
def IIC_VST1lnu : InstrItinClass;
|
|
def IIC_VST2 : InstrItinClass;
|
|
def IIC_VST2x2 : InstrItinClass;
|
|
def IIC_VST2u : InstrItinClass;
|
|
def IIC_VST2x2u : InstrItinClass;
|
|
def IIC_VST2ln : InstrItinClass;
|
|
def IIC_VST2lnu : InstrItinClass;
|
|
def IIC_VST3 : InstrItinClass;
|
|
def IIC_VST3u : InstrItinClass;
|
|
def IIC_VST3ln : InstrItinClass;
|
|
def IIC_VST3lnu : InstrItinClass;
|
|
def IIC_VST4 : InstrItinClass;
|
|
def IIC_VST4u : InstrItinClass;
|
|
def IIC_VST4ln : InstrItinClass;
|
|
def IIC_VST4lnu : InstrItinClass;
|
|
def IIC_VUNAD : InstrItinClass;
|
|
def IIC_VUNAQ : InstrItinClass;
|
|
def IIC_VBIND : InstrItinClass;
|
|
def IIC_VBINQ : InstrItinClass;
|
|
def IIC_VPBIND : InstrItinClass;
|
|
def IIC_VFMULD : InstrItinClass;
|
|
def IIC_VFMULQ : InstrItinClass;
|
|
def IIC_VMOV : InstrItinClass;
|
|
def IIC_VMOVImm : InstrItinClass;
|
|
def IIC_VMOVD : InstrItinClass;
|
|
def IIC_VMOVQ : InstrItinClass;
|
|
def IIC_VMOVIS : InstrItinClass;
|
|
def IIC_VMOVID : InstrItinClass;
|
|
def IIC_VMOVISL : InstrItinClass;
|
|
def IIC_VMOVSI : InstrItinClass;
|
|
def IIC_VMOVDI : InstrItinClass;
|
|
def IIC_VMOVN : InstrItinClass;
|
|
def IIC_VPERMD : InstrItinClass;
|
|
def IIC_VPERMQ : InstrItinClass;
|
|
def IIC_VPERMQ3 : InstrItinClass;
|
|
def IIC_VMACD : InstrItinClass;
|
|
def IIC_VMACQ : InstrItinClass;
|
|
def IIC_VRECSD : InstrItinClass;
|
|
def IIC_VRECSQ : InstrItinClass;
|
|
def IIC_VCNTiD : InstrItinClass;
|
|
def IIC_VCNTiQ : InstrItinClass;
|
|
def IIC_VUNAiD : InstrItinClass;
|
|
def IIC_VUNAiQ : InstrItinClass;
|
|
def IIC_VQUNAiD : InstrItinClass;
|
|
def IIC_VQUNAiQ : InstrItinClass;
|
|
def IIC_VBINiD : InstrItinClass;
|
|
def IIC_VBINiQ : InstrItinClass;
|
|
def IIC_VSUBiD : InstrItinClass;
|
|
def IIC_VSUBiQ : InstrItinClass;
|
|
def IIC_VBINi4D : InstrItinClass;
|
|
def IIC_VBINi4Q : InstrItinClass;
|
|
def IIC_VSUBi4D : InstrItinClass;
|
|
def IIC_VSUBi4Q : InstrItinClass;
|
|
def IIC_VABAD : InstrItinClass;
|
|
def IIC_VABAQ : InstrItinClass;
|
|
def IIC_VSHLiD : InstrItinClass;
|
|
def IIC_VSHLiQ : InstrItinClass;
|
|
def IIC_VSHLi4D : InstrItinClass;
|
|
def IIC_VSHLi4Q : InstrItinClass;
|
|
def IIC_VPALiD : InstrItinClass;
|
|
def IIC_VPALiQ : InstrItinClass;
|
|
def IIC_VMULi16D : InstrItinClass;
|
|
def IIC_VMULi32D : InstrItinClass;
|
|
def IIC_VMULi16Q : InstrItinClass;
|
|
def IIC_VMULi32Q : InstrItinClass;
|
|
def IIC_VMACi16D : InstrItinClass;
|
|
def IIC_VMACi32D : InstrItinClass;
|
|
def IIC_VMACi16Q : InstrItinClass;
|
|
def IIC_VMACi32Q : InstrItinClass;
|
|
def IIC_VEXTD : InstrItinClass;
|
|
def IIC_VEXTQ : InstrItinClass;
|
|
def IIC_VTB1 : InstrItinClass;
|
|
def IIC_VTB2 : InstrItinClass;
|
|
def IIC_VTB3 : InstrItinClass;
|
|
def IIC_VTB4 : InstrItinClass;
|
|
def IIC_VTBX1 : InstrItinClass;
|
|
def IIC_VTBX2 : InstrItinClass;
|
|
def IIC_VTBX3 : InstrItinClass;
|
|
def IIC_VTBX4 : InstrItinClass;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// Processor instruction itineraries.
|
|
|
|
def GenericItineraries : ProcessorItineraries<[], [], []>;
|
|
|
|
include "ARMScheduleV6.td"
|
|
include "ARMScheduleA8.td"
|
|
include "ARMScheduleA9.td"
|