Gary Benson
2008-Jun-30 13:10 UTC
[LLVMdev] Implementing llvm.atomic.cmp.swap.i32 on PowerPC
Chris Lattner wrote:> On Jun 27, 2008, at 8:27 AM, Gary Benson wrote: > > def CMP_UNRESw : Pseudo<(outs), (ins GPRC:$rA, GPRC:$rB, i32imm: > > $label), > > "cmpw $rA, $rB\n\tbne- La${label}_exit", > > [(PPCcmp_unres GPRC:$rA, GPRC:$rB, imm: > > $label)]>; > > } > > > > ...and I can't figure out the syntax for that. Any suggestions? > > You have to write custom encoding logic in C++ for this. This > should go in PPCCodeEmitter::emitBasicBlock. I admit this isn't > very elegant. A better solution would be to fix the lowering to > produce two instructions instead of this pseudo instruction.Hi Chris, I'd prefer to fix the lowering if possible; the pseudo instructions are only used in three places, so it shouldn't be a huge change. I need to generate labels in PPCTargetLowering::LowerAtomicCMP_SWAP however: how do I do that? FWIW the code it needs to emit is: ; inputs: ptr, oldval, newval loop: lwarx $tmp, 0, $ptr cmpw $oldval, $tmp bne- exit stwcx. $newval, 0, $ptr bne- loop exit: ... Cheers, Gary -- http://gbenson.net/
Evan Cheng
2008-Jun-30 21:14 UTC
[LLVMdev] Implementing llvm.atomic.cmp.swap.i32 on PowerPC
You need to insert new basic blocks and update CFG to accomplish this. There is a hackish way to do this right now. Add a pseudo instruction to represent this operation and mark it usesCustomDAGSchedInserter. This means the intrinsic is mapped to a single (pseudo) node. But it is then expanded into instructions that can span multiple basic blocks. See PPCTargetLowering::EmitInstrWithCustomInserter(). Evan On Jun 30, 2008, at 6:10 AM, Gary Benson wrote:> Chris Lattner wrote: >> On Jun 27, 2008, at 8:27 AM, Gary Benson wrote: >>> def CMP_UNRESw : Pseudo<(outs), (ins GPRC:$rA, GPRC:$rB, i32imm: >>> $label), >>> "cmpw $rA, $rB\n\tbne- La${label}_exit", >>> [(PPCcmp_unres GPRC:$rA, GPRC:$rB, imm: >>> $label)]>; >>> } >>> >>> ...and I can't figure out the syntax for that. Any suggestions? >> >> You have to write custom encoding logic in C++ for this. This >> should go in PPCCodeEmitter::emitBasicBlock. I admit this isn't >> very elegant. A better solution would be to fix the lowering to >> produce two instructions instead of this pseudo instruction. > > Hi Chris, > > I'd prefer to fix the lowering if possible; the pseudo instructions > are only used in three places, so it shouldn't be a huge change. > I need to generate labels in PPCTargetLowering::LowerAtomicCMP_SWAP > however: how do I do that? FWIW the code it needs to emit is: > > ; inputs: ptr, oldval, newval > loop: > lwarx $tmp, 0, $ptr > cmpw $oldval, $tmp > bne- exit > stwcx. $newval, 0, $ptr > bne- loop > exit: > ... > > > Cheers, > Gary > > -- > http://gbenson.net/ > _______________________________________________ > LLVM Developers mailing list > LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
Gary Benson
2008-Jul-02 15:29 UTC
[LLVMdev] Implementing llvm.atomic.cmp.swap.i32 on PowerPC
Evan Cheng wrote:> You need to insert new basic blocks and update CFG to accomplish this. > There is a hackish way to do this right now. Add a pseudo instruction > to represent this operation and mark it usesCustomDAGSchedInserter. > This means the intrinsic is mapped to a single (pseudo) node. But it > is then expanded into instructions that can span multiple basic > blocks. See PPCTargetLowering::EmitInstrWithCustomInserter().How does this look? It's a big patch, but it basically does this: - Adds ATOMIC_LOAD_ADD, ATOMIC_CMP_SWAP and ATOMIC_SWAP nodes, and ATOMIC_LOAD_ADD_I{32,64}, ATOMIC_CMP_SWAP_I{32,64} and ATOMIC_SWAP_I{32,64} pseudo-instructions with custom inserters. - Replaces L[WD]ARX and ST[WD]CX pseudo-instructions with the actual PPC instructions they represent. - Removes CMP_UNRESERVE nodes and CMP_UNRES[wd]{,i} pseudo- instructions. Cheers, Gary -- http://gbenson.net/ -------------- next part -------------- Index: lib/Target/PowerPC/PPCISelLowering.h ==================================================================--- lib/Target/PowerPC/PPCISelLowering.h (revision 52957) +++ lib/Target/PowerPC/PPCISelLowering.h (working copy) @@ -152,6 +152,11 @@ /// MTFSF = F8RC, INFLAG - This moves the register into the FPSCR. MTFSF, + /// ATOMIC_LOAD_ADD, ATOMIC_CMP_SWAP, ATOMIC_SWAP - These + /// correspond to the llvm.atomic.load.add, llvm.atomic.cmp.swap + /// and llvm.atomic.swap intrinsics. + ATOMIC_LOAD_ADD, ATOMIC_CMP_SWAP, ATOMIC_SWAP, + /// LARX = This corresponds to PPC l{w|d}arx instrcution: load and /// reserve indexed. This is used to implement atomic operations. LARX, @@ -160,10 +165,6 @@ /// indexed. This is used to implement atomic operations. STCX, - /// CMP_UNRESERVE = Test for equality and "unreserve" if not true. This - /// is used to implement atomic operations. - CMP_UNRESERVE, - /// TAILCALL - Indicates a tail call should be taken. TAILCALL, /// TC_RETURN - A tail call return. @@ -325,10 +326,6 @@ SelectionDAG &DAG) const; private: - /// PPCAtomicLabelIndex - Keep track the number of PPC atomic labels. - /// - unsigned PPCAtomicLabelIndex; - SDOperand getFramePointerFrameIndex(SelectionDAG & DAG) const; SDOperand getReturnAddrFrameIndex(SelectionDAG & DAG) const; Index: lib/Target/PowerPC/PPCInstr64Bit.td ==================================================================--- lib/Target/PowerPC/PPCInstr64Bit.td (revision 52957) +++ lib/Target/PowerPC/PPCInstr64Bit.td (working copy) @@ -116,23 +116,35 @@ def : Pat<(PPCcall_ELF (i64 texternalsym:$dst)), (BL8_ELF texternalsym:$dst)>; -// Atomic operations. -def LDARX : Pseudo<(outs G8RC:$rD), (ins memrr:$ptr, i32imm:$label), - "\nLa${label}_entry:\n\tldarx $rD, $ptr", - [(set G8RC:$rD, (PPClarx xoaddr:$ptr, imm:$label))]>; +// Atomic operations +let usesCustomDAGSchedInserter = 1 in { + let Uses = [CR0] in { + let Uses = [R0] in + def ATOMIC_LOAD_ADD_I64 : Pseudo< + (outs G8RC:$dst), (ins memrr:$ptr, G8RC:$incr), + "${:comment} ATOMIC_LOAD_ADD_I64 PSEUDO!", + [(set G8RC:$dst, (PPCatomic_load_add xoaddr:$ptr, G8RC:$incr))]>; + def ATOMIC_CMP_SWAP_I64 : Pseudo< + (outs G8RC:$dst), (ins memrr:$ptr, G8RC:$old, G8RC:$new), + "${:comment} ATOMIC_CMP_SWAP_I64 PSEUDO!", + [(set G8RC:$dst, (PPCatomic_cmp_swap xoaddr:$ptr, G8RC:$old, G8RC:$new))]>; + def ATOMIC_SWAP_I64 : Pseudo< + (outs G8RC:$dst), (ins memrr:$ptr, G8RC:$new), + "${:comment} ATOMIC_SWAP_I64 PSEUDO!", + [(set G8RC:$dst, (PPCatomic_swap xoaddr:$ptr, G8RC:$new))]>; + } +} -let Defs = [CR0] in { -def STDCX : Pseudo<(outs), (ins G8RC:$rS, memrr:$dst, i32imm:$label), - "stdcx. $rS, $dst\n\tbne- La${label}_entry\nLa${label}_exit:", - [(PPCstcx G8RC:$rS, xoaddr:$dst, imm:$label)]>; +// Instructions to support atomic operations +def LDARX : XForm_1<31, 84, (outs G8RC:$rD), (ins memrr:$ptr), + "ldarx $rD, $ptr", LdStLDARX, + [(set G8RC:$rD, (PPClarx xoaddr:$ptr))]>; -def CMP_UNRESd : Pseudo<(outs), (ins G8RC:$rA, G8RC:$rB, i32imm:$label), - "cmpd $rA, $rB\n\tbne- La${label}_exit", - [(PPCcmp_unres G8RC:$rA, G8RC:$rB, imm:$label)]>; -def CMP_UNRESdi : Pseudo<(outs), (ins G8RC:$rA, s16imm64:$imm, i32imm:$label), - "cmpdi $rA, $imm\n\tbne- La${label}_exit", - [(PPCcmp_unres G8RC:$rA, immSExt16:$imm, imm:$label)]>; -} +let Defs = [CR0] in +def STDCX : XForm_1<31, 214, (outs), (ins G8RC:$rS, memrr:$dst), + "stdcx. $rS, $dst", LdStSTDCX, + [(PPCstcx G8RC:$rS, xoaddr:$dst)]>, + isDOT; let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1 in def TCRETURNdi8 :Pseudo< (outs), Index: lib/Target/PowerPC/PPCInstrInfo.td ==================================================================--- lib/Target/PowerPC/PPCInstrInfo.td (revision 52957) +++ lib/Target/PowerPC/PPCInstrInfo.td (working copy) @@ -42,17 +42,23 @@ SDTCisVT<0, i32>, SDTCisPtrTy<1>, SDTCisVT<2, OtherVT>, SDTCisVT<3, OtherVT> ]>; - -def SDT_PPClarx : SDTypeProfile<1, 2, [ - SDTCisInt<0>, SDTCisPtrTy<1>, SDTCisVT<2, i32> +def SDT_PPCatomic_load_add : SDTypeProfile<1, 2, [ + SDTCisInt<0>, SDTCisPtrTy<1>, SDTCisInt<2> ]>; -def SDT_PPCstcx : SDTypeProfile<0, 3, [ - SDTCisInt<0>, SDTCisPtrTy<1>, SDTCisVT<2, i32> +def SDT_PPCatomic_cmp_swap : SDTypeProfile<1, 3, [ + SDTCisInt<0>, SDTCisPtrTy<1>, SDTCisInt<2>, SDTCisInt<3> ]>; -def SDT_PPCcmp_unres : SDTypeProfile<0, 3, [ - SDTCisSameAs<0, 1>, SDTCisInt<1>, SDTCisVT<2, i32> +def SDT_PPCatomic_swap : SDTypeProfile<1, 2, [ + SDTCisInt<0>, SDTCisPtrTy<1>, SDTCisInt<2> ]>; +def SDT_PPClarx : SDTypeProfile<1, 1, [ + SDTCisInt<0>, SDTCisPtrTy<1> +]>; +def SDT_PPCstcx : SDTypeProfile<0, 2, [ + SDTCisInt<0>, SDTCisPtrTy<1> +]>; + def SDT_PPCTC_ret : SDTypeProfile<0, 2, [ SDTCisPtrTy<0>, SDTCisVT<1, i32> ]>; @@ -143,12 +149,22 @@ def PPCstbrx : SDNode<"PPCISD::STBRX", SDT_PPCstbrx, [SDNPHasChain, SDNPMayStore]>; +// Atomic operations +def PPCatomic_load_add : SDNode<"PPCISD::ATOMIC_LOAD_ADD", + SDT_PPCatomic_load_add, + [SDNPHasChain, SDNPMayLoad, SDNPMayStore]>; +def PPCatomic_cmp_swap : SDNode<"PPCISD::ATOMIC_CMP_SWAP", + SDT_PPCatomic_cmp_swap, + [SDNPHasChain, SDNPMayLoad, SDNPMayStore]>; +def PPCatomic_swap : SDNode<"PPCISD::ATOMIC_SWAP", + SDT_PPCatomic_swap, + [SDNPHasChain, SDNPMayLoad, SDNPMayStore]>; + +// Instructions to support atomic operations def PPClarx : SDNode<"PPCISD::LARX", SDT_PPClarx, [SDNPHasChain, SDNPMayLoad]>; def PPCstcx : SDNode<"PPCISD::STCX", SDT_PPCstcx, [SDNPHasChain, SDNPMayStore]>; -def PPCcmp_unres : SDNode<"PPCISD::CMP_UNRESERVE", SDT_PPCcmp_unres, - [SDNPHasChain]>; // Instructions to support dynamic alloca. def SDTDynOp : SDTypeProfile<1, 2, []>; @@ -530,23 +546,35 @@ "dcbzl $dst", LdStDCBF, [(int_ppc_dcbzl xoaddr:$dst)]>, PPC970_DGroup_Single; -// Atomic operations. -def LWARX : XForm_1<31, 20, (outs GPRC:$rD), (ins memrr:$ptr, i32imm:$label), - "\nLa${label}_entry:\n\tlwarx $rD, $ptr", LdStLWARX, - [(set GPRC:$rD, (PPClarx xoaddr:$ptr, imm:$label))]>; +// Atomic operations +let usesCustomDAGSchedInserter = 1 in { + let Uses = [CR0] in { + let Uses = [R0] in + def ATOMIC_LOAD_ADD_I32 : Pseudo< + (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$incr), + "${:comment} ATOMIC_LOAD_ADD_I32 PSEUDO!", + [(set GPRC:$dst, (PPCatomic_load_add xoaddr:$ptr, GPRC:$incr))]>; + def ATOMIC_CMP_SWAP_I32 : Pseudo< + (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$old, GPRC:$new), + "${:comment} ATOMIC_CMP_SWAP_I32 PSEUDO!", + [(set GPRC:$dst, (PPCatomic_cmp_swap xoaddr:$ptr, GPRC:$old, GPRC:$new))]>; + def ATOMIC_SWAP_I32 : Pseudo< + (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$new), + "${:comment} ATOMIC_SWAP_I32 PSEUDO!", + [(set GPRC:$dst, (PPCatomic_swap xoaddr:$ptr, GPRC:$new))]>; + } +} -let Defs = [CR0] in { -def STWCX : Pseudo<(outs), (ins GPRC:$rS, memrr:$dst, i32imm:$label), - "stwcx. $rS, $dst\n\tbne- La${label}_entry\nLa${label}_exit:", - [(PPCstcx GPRC:$rS, xoaddr:$dst, imm:$label)]>; +// Instructions to support atomic operations +def LWARX : XForm_1<31, 20, (outs GPRC:$rD), (ins memrr:$src), + "lwarx $rD, $src", LdStLWARX, + [(set GPRC:$rD, (PPClarx xoaddr:$src))]>; -def CMP_UNRESw : Pseudo<(outs), (ins GPRC:$rA, GPRC:$rB, i32imm:$label), - "cmpw $rA, $rB\n\tbne- La${label}_exit", - [(PPCcmp_unres GPRC:$rA, GPRC:$rB, imm:$label)]>; -def CMP_UNRESwi : Pseudo<(outs), (ins GPRC:$rA, s16imm:$imm, i32imm:$label), - "cmpwi $rA, $imm\n\tbne- La${label}_exit", - [(PPCcmp_unres GPRC:$rA, immSExt16:$imm, imm:$label)]>; -} +let Defs = [CR0] in +def STWCX : XForm_1<31, 150, (outs), (ins GPRC:$rS, memrr:$dst), + "stwcx. $rS, $dst", LdStSTWCX, + [(PPCstcx GPRC:$rS, xoaddr:$dst)]>, + isDOT; //===----------------------------------------------------------------------===// // PPC32 Load Instructions. @@ -1327,9 +1355,5 @@ def : Pat<(extloadf32 xaddr:$src), (FMRSD (LFSX xaddr:$src))>; -// Atomic operations -def : Pat<(PPCcmp_unres immSExt16:$imm, GPRC:$rA, imm:$label), - (CMP_UNRESwi GPRC:$rA, immSExt16:$imm, imm:$label)>; - include "PPCInstrAltivec.td" include "PPCInstr64Bit.td" Index: lib/Target/PowerPC/PPCISelLowering.cpp ==================================================================--- lib/Target/PowerPC/PPCISelLowering.cpp (revision 52957) +++ lib/Target/PowerPC/PPCISelLowering.cpp (working copy) @@ -40,8 +40,7 @@ cl::Hidden); PPCTargetLowering::PPCTargetLowering(PPCTargetMachine &TM) - : TargetLowering(TM), PPCSubTarget(*TM.getSubtargetImpl()), - PPCAtomicLabelIndex(0) { + : TargetLowering(TM), PPCSubTarget(*TM.getSubtargetImpl()) { setPow2DivIsCheap(); @@ -378,45 +377,47 @@ const char *PPCTargetLowering::getTargetNodeName(unsigned Opcode) const { switch (Opcode) { default: return 0; - case PPCISD::FSEL: return "PPCISD::FSEL"; - case PPCISD::FCFID: return "PPCISD::FCFID"; - case PPCISD::FCTIDZ: return "PPCISD::FCTIDZ"; - case PPCISD::FCTIWZ: return "PPCISD::FCTIWZ"; - case PPCISD::STFIWX: return "PPCISD::STFIWX"; - case PPCISD::VMADDFP: return "PPCISD::VMADDFP"; - case PPCISD::VNMSUBFP: return "PPCISD::VNMSUBFP"; - case PPCISD::VPERM: return "PPCISD::VPERM"; - case PPCISD::Hi: return "PPCISD::Hi"; - case PPCISD::Lo: return "PPCISD::Lo"; - case PPCISD::DYNALLOC: return "PPCISD::DYNALLOC"; - case PPCISD::GlobalBaseReg: return "PPCISD::GlobalBaseReg"; - case PPCISD::SRL: return "PPCISD::SRL"; - case PPCISD::SRA: return "PPCISD::SRA"; - case PPCISD::SHL: return "PPCISD::SHL"; - case PPCISD::EXTSW_32: return "PPCISD::EXTSW_32"; - case PPCISD::STD_32: return "PPCISD::STD_32"; - case PPCISD::CALL_ELF: return "PPCISD::CALL_ELF"; - case PPCISD::CALL_Macho: return "PPCISD::CALL_Macho"; - case PPCISD::MTCTR: return "PPCISD::MTCTR"; - case PPCISD::BCTRL_Macho: return "PPCISD::BCTRL_Macho"; - case PPCISD::BCTRL_ELF: return "PPCISD::BCTRL_ELF"; - case PPCISD::RET_FLAG: return "PPCISD::RET_FLAG"; - case PPCISD::MFCR: return "PPCISD::MFCR"; - case PPCISD::VCMP: return "PPCISD::VCMP"; - case PPCISD::VCMPo: return "PPCISD::VCMPo"; - case PPCISD::LBRX: return "PPCISD::LBRX"; - case PPCISD::STBRX: return "PPCISD::STBRX"; - case PPCISD::LARX: return "PPCISD::LARX"; - case PPCISD::STCX: return "PPCISD::STCX"; - case PPCISD::CMP_UNRESERVE: return "PPCISD::CMP_UNRESERVE"; - case PPCISD::COND_BRANCH: return "PPCISD::COND_BRANCH"; - case PPCISD::MFFS: return "PPCISD::MFFS"; - case PPCISD::MTFSB0: return "PPCISD::MTFSB0"; - case PPCISD::MTFSB1: return "PPCISD::MTFSB1"; - case PPCISD::FADDRTZ: return "PPCISD::FADDRTZ"; - case PPCISD::MTFSF: return "PPCISD::MTFSF"; - case PPCISD::TAILCALL: return "PPCISD::TAILCALL"; - case PPCISD::TC_RETURN: return "PPCISD::TC_RETURN"; + case PPCISD::FSEL: return "PPCISD::FSEL"; + case PPCISD::FCFID: return "PPCISD::FCFID"; + case PPCISD::FCTIDZ: return "PPCISD::FCTIDZ"; + case PPCISD::FCTIWZ: return "PPCISD::FCTIWZ"; + case PPCISD::STFIWX: return "PPCISD::STFIWX"; + case PPCISD::VMADDFP: return "PPCISD::VMADDFP"; + case PPCISD::VNMSUBFP: return "PPCISD::VNMSUBFP"; + case PPCISD::VPERM: return "PPCISD::VPERM"; + case PPCISD::Hi: return "PPCISD::Hi"; + case PPCISD::Lo: return "PPCISD::Lo"; + case PPCISD::DYNALLOC: return "PPCISD::DYNALLOC"; + case PPCISD::GlobalBaseReg: return "PPCISD::GlobalBaseReg"; + case PPCISD::SRL: return "PPCISD::SRL"; + case PPCISD::SRA: return "PPCISD::SRA"; + case PPCISD::SHL: return "PPCISD::SHL"; + case PPCISD::EXTSW_32: return "PPCISD::EXTSW_32"; + case PPCISD::STD_32: return "PPCISD::STD_32"; + case PPCISD::CALL_ELF: return "PPCISD::CALL_ELF"; + case PPCISD::CALL_Macho: return "PPCISD::CALL_Macho"; + case PPCISD::MTCTR: return "PPCISD::MTCTR"; + case PPCISD::BCTRL_Macho: return "PPCISD::BCTRL_Macho"; + case PPCISD::BCTRL_ELF: return "PPCISD::BCTRL_ELF"; + case PPCISD::RET_FLAG: return "PPCISD::RET_FLAG"; + case PPCISD::MFCR: return "PPCISD::MFCR"; + case PPCISD::VCMP: return "PPCISD::VCMP"; + case PPCISD::VCMPo: return "PPCISD::VCMPo"; + case PPCISD::LBRX: return "PPCISD::LBRX"; + case PPCISD::STBRX: return "PPCISD::STBRX"; + case PPCISD::ATOMIC_LOAD_ADD: return "PPCISD::ATOMIC_LOAD_ADD"; + case PPCISD::ATOMIC_CMP_SWAP: return "PPCISD::ATOMIC_CMP_SWAP"; + case PPCISD::ATOMIC_SWAP: return "PPCISD::ATOMIC_SWAP"; + case PPCISD::LARX: return "PPCISD::LARX"; + case PPCISD::STCX: return "PPCISD::STCX"; + case PPCISD::COND_BRANCH: return "PPCISD::COND_BRANCH"; + case PPCISD::MFFS: return "PPCISD::MFFS"; + case PPCISD::MTFSB0: return "PPCISD::MTFSB0"; + case PPCISD::MTFSB1: return "PPCISD::MTFSB1"; + case PPCISD::FADDRTZ: return "PPCISD::FADDRTZ"; + case PPCISD::MTFSF: return "PPCISD::MTFSF"; + case PPCISD::TAILCALL: return "PPCISD::TAILCALL"; + case PPCISD::TC_RETURN: return "PPCISD::TC_RETURN"; } } @@ -2726,33 +2727,13 @@ SDOperand Ptr = Op.getOperand(1); SDOperand Incr = Op.getOperand(2); - // Issue a "load and reserve". - std::vector<MVT> VTs; - VTs.push_back(VT); - VTs.push_back(MVT::Other); - - SDOperand Label = DAG.getConstant(PPCAtomicLabelIndex++, MVT::i32); + SDVTList VTs = DAG.getVTList(VT, MVT::Other); SDOperand Ops[] = { - Chain, // Chain - Ptr, // Ptr - Label, // Label + Chain, + Ptr, + Incr, }; - SDOperand Load = DAG.getNode(PPCISD::LARX, VTs, Ops, 3); - Chain = Load.getValue(1); - - // Compute new value. - SDOperand NewVal = DAG.getNode(ISD::ADD, VT, Load, Incr); - - // Issue a "store and check". - SDOperand Ops2[] = { - Chain, // Chain - NewVal, // Value - Ptr, // Ptr - Label, // Label - }; - SDOperand Store = DAG.getNode(PPCISD::STCX, MVT::Other, Ops2, 4); - SDOperand OutOps[] = { Load, Store }; - return DAG.getMergeValues(DAG.getVTList(VT, MVT::Other), OutOps, 2); + return DAG.getNode(PPCISD::ATOMIC_LOAD_ADD, VTs, Ops, 3); } SDOperand PPCTargetLowering::LowerAtomicCMP_SWAP(SDOperand Op, SelectionDAG &DAG) { @@ -2762,39 +2743,14 @@ SDOperand NewVal = Op.getOperand(2); SDOperand OldVal = Op.getOperand(3); - // Issue a "load and reserve". - std::vector<MVT> VTs; - VTs.push_back(VT); - VTs.push_back(MVT::Other); - - SDOperand Label = DAG.getConstant(PPCAtomicLabelIndex++, MVT::i32); + SDVTList VTs = DAG.getVTList(VT, MVT::Other); SDOperand Ops[] = { - Chain, // Chain - Ptr, // Ptr - Label, // Label + Chain, + Ptr, + OldVal, + NewVal, }; - SDOperand Load = DAG.getNode(PPCISD::LARX, VTs, Ops, 3); - Chain = Load.getValue(1); - - // Compare and unreserve if not equal. - SDOperand Ops2[] = { - Chain, // Chain - OldVal, // Old value - Load, // Value in memory - Label, // Label - }; - Chain = DAG.getNode(PPCISD::CMP_UNRESERVE, MVT::Other, Ops2, 4); - - // Issue a "store and check". - SDOperand Ops3[] = { - Chain, // Chain - NewVal, // Value - Ptr, // Ptr - Label, // Label - }; - SDOperand Store = DAG.getNode(PPCISD::STCX, MVT::Other, Ops3, 4); - SDOperand OutOps[] = { Load, Store }; - return DAG.getMergeValues(DAG.getVTList(VT, MVT::Other), OutOps, 2); + return DAG.getNode(PPCISD::ATOMIC_CMP_SWAP, VTs, Ops, 4); } SDOperand PPCTargetLowering::LowerAtomicSWAP(SDOperand Op, SelectionDAG &DAG) { @@ -2803,30 +2759,13 @@ SDOperand Ptr = Op.getOperand(1); SDOperand NewVal = Op.getOperand(2); - // Issue a "load and reserve". - std::vector<MVT> VTs; - VTs.push_back(VT); - VTs.push_back(MVT::Other); - - SDOperand Label = DAG.getConstant(PPCAtomicLabelIndex++, MVT::i32); + SDVTList VTs = DAG.getVTList(VT, MVT::Other); SDOperand Ops[] = { - Chain, // Chain - Ptr, // Ptr - Label, // Label + Chain, + Ptr, + NewVal, }; - SDOperand Load = DAG.getNode(PPCISD::LARX, VTs, Ops, 3); - Chain = Load.getValue(1); - - // Issue a "store and check". - SDOperand Ops2[] = { - Chain, // Chain - NewVal, // Value - Ptr, // Ptr - Label, // Label - }; - SDOperand Store = DAG.getNode(PPCISD::STCX, MVT::Other, Ops2, 4); - SDOperand OutOps[] = { Load, Store }; - return DAG.getMergeValues(DAG.getVTList(VT, MVT::Other), OutOps, 2); + return DAG.getNode(PPCISD::ATOMIC_SWAP, VTs, Ops, 3); } /// LowerSELECT_CC - Lower floating point select_cc's into fsel instruction when @@ -3980,59 +3919,198 @@ PPCTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI, MachineBasicBlock *BB) { const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); - assert((MI->getOpcode() == PPC::SELECT_CC_I4 || - MI->getOpcode() == PPC::SELECT_CC_I8 || - MI->getOpcode() == PPC::SELECT_CC_F4 || - MI->getOpcode() == PPC::SELECT_CC_F8 || - MI->getOpcode() == PPC::SELECT_CC_VRRC) && - "Unexpected instr type to insert"); - - // To "insert" a SELECT_CC instruction, we actually have to insert the diamond - // control-flow pattern. The incoming instruction knows 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. + + // To "insert" these instructions we actually have to insert their + // control-flow patterns. const BasicBlock *LLVM_BB = BB->getBasicBlock(); ilist<MachineBasicBlock>::iterator It = BB; ++It; - - // thisMBB: - // ... - // TrueVal = ... - // cmpTY ccX, r1, r2 - // bCC copy1MBB - // fallthrough --> copy0MBB - MachineBasicBlock *thisMBB = BB; - MachineBasicBlock *copy0MBB = new MachineBasicBlock(LLVM_BB); - MachineBasicBlock *sinkMBB = new MachineBasicBlock(LLVM_BB); - unsigned SelectPred = MI->getOperand(4).getImm(); - BuildMI(BB, TII->get(PPC::BCC)) - .addImm(SelectPred).addReg(MI->getOperand(1).getReg()).addMBB(sinkMBB); - MachineFunction *F = BB->getParent(); - F->getBasicBlockList().insert(It, copy0MBB); - F->getBasicBlockList().insert(It, sinkMBB); - // Update machine-CFG edges by transferring all successors of the current - // block to the new block which will contain the Phi node for the select. - sinkMBB->transferSuccessors(BB); - // Next, add the true and fallthrough blocks as its successors. - BB->addSuccessor(copy0MBB); - BB->addSuccessor(sinkMBB); - - // copy0MBB: - // %FalseValue = ... - // # fallthrough to sinkMBB - BB = copy0MBB; - - // Update machine-CFG edges - BB->addSuccessor(sinkMBB); - - // sinkMBB: - // %Result = phi [ %FalseValue, copy0MBB ], [ %TrueValue, thisMBB ] - // ... - BB = sinkMBB; - BuildMI(BB, TII->get(PPC::PHI), MI->getOperand(0).getReg()) - .addReg(MI->getOperand(3).getReg()).addMBB(copy0MBB) - .addReg(MI->getOperand(2).getReg()).addMBB(thisMBB); + if (MI->getOpcode() == PPC::SELECT_CC_I4 || + MI->getOpcode() == PPC::SELECT_CC_I8 || + MI->getOpcode() == PPC::SELECT_CC_F4 || + MI->getOpcode() == PPC::SELECT_CC_F8 || + MI->getOpcode() == PPC::SELECT_CC_VRRC) { + + // The incoming instruction knows 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. + + // thisMBB: + // ... + // TrueVal = ... + // cmpTY ccX, r1, r2 + // bCC copy1MBB + // fallthrough --> copy0MBB + MachineBasicBlock *thisMBB = BB; + MachineBasicBlock *copy0MBB = new MachineBasicBlock(LLVM_BB); + MachineBasicBlock *sinkMBB = new MachineBasicBlock(LLVM_BB); + unsigned SelectPred = MI->getOperand(4).getImm(); + BuildMI(BB, TII->get(PPC::BCC)) + .addImm(SelectPred).addReg(MI->getOperand(1).getReg()).addMBB(sinkMBB); + MachineFunction *F = BB->getParent(); + F->getBasicBlockList().insert(It, copy0MBB); + F->getBasicBlockList().insert(It, sinkMBB); + // Update machine-CFG edges by transferring all successors of the current + // block to the new block which will contain the Phi node for the select. + sinkMBB->transferSuccessors(BB); + // Next, add the true and fallthrough blocks as its successors. + BB->addSuccessor(copy0MBB); + BB->addSuccessor(sinkMBB); + + // copy0MBB: + // %FalseValue = ... + // # fallthrough to sinkMBB + BB = copy0MBB; + + // Update machine-CFG edges + BB->addSuccessor(sinkMBB); + + // sinkMBB: + // %Result = phi [ %FalseValue, copy0MBB ], [ %TrueValue, thisMBB ] + // ... + BB = sinkMBB; + BuildMI(BB, TII->get(PPC::PHI), MI->getOperand(0).getReg()) + .addReg(MI->getOperand(3).getReg()).addMBB(copy0MBB) + .addReg(MI->getOperand(2).getReg()).addMBB(thisMBB); + } + else if (MI->getOpcode() == PPC::ATOMIC_LOAD_ADD_I32 || + MI->getOpcode() == PPC::ATOMIC_LOAD_ADD_I64) { + bool is64bit = MI->getOpcode() == PPC::ATOMIC_LOAD_ADD_I64; + + unsigned dest = MI->getOperand(0).getReg(); + unsigned ptrA = MI->getOperand(1).getReg(); + unsigned ptrB = MI->getOperand(2).getReg(); + unsigned incr = MI->getOperand(3).getReg(); + + MachineBasicBlock *loopMBB = new MachineBasicBlock(LLVM_BB); + MachineBasicBlock *exitMBB = new MachineBasicBlock(LLVM_BB); + + MachineFunction *F = BB->getParent(); + F->getBasicBlockList().insert(It, loopMBB); + F->getBasicBlockList().insert(It, exitMBB); + exitMBB->transferSuccessors(BB); + + // thisMBB: + // ... + // fallthrough --> loopMBB + BB->addSuccessor(loopMBB); + + // loopMBB: + // l[wd]arx dest, ptr + // add r0, dest, incr + // st[wd]cx. r0, ptr + // bne- loopMBB + // fallthrough --> exitMBB + BB = loopMBB; + BuildMI(BB, TII->get(is64bit ? PPC::LDARX : PPC::LWARX), dest) + .addReg(ptrA).addReg(ptrB); + BuildMI(BB, TII->get(is64bit ? PPC::ADD4 : PPC::ADD8), PPC::R0) + .addReg(incr).addReg(dest); + BuildMI(BB, TII->get(is64bit ? PPC::STDCX : PPC::STWCX)) + .addReg(PPC::R0).addReg(ptrA).addReg(ptrB); + BuildMI(BB, TII->get(PPC::BCC)) + .addImm(PPC::PRED_NE).addReg(PPC::CR0).addMBB(loopMBB); + BB->addSuccessor(loopMBB); + BB->addSuccessor(exitMBB); + + // exitMBB: + // ... + BB = exitMBB; + } + else if (MI->getOpcode() == PPC::ATOMIC_CMP_SWAP_I32 || + MI->getOpcode() == PPC::ATOMIC_CMP_SWAP_I64) { + bool is64bit = MI->getOpcode() == PPC::ATOMIC_CMP_SWAP_I64; + + unsigned dest = MI->getOperand(0).getReg(); + unsigned ptrA = MI->getOperand(1).getReg(); + unsigned ptrB = MI->getOperand(2).getReg(); + unsigned oldval = MI->getOperand(3).getReg(); + unsigned newval = MI->getOperand(4).getReg(); + + MachineBasicBlock *loopMBB = new MachineBasicBlock(LLVM_BB); + MachineBasicBlock *exitMBB = new MachineBasicBlock(LLVM_BB); + + MachineFunction *F = BB->getParent(); + F->getBasicBlockList().insert(It, loopMBB); + F->getBasicBlockList().insert(It, exitMBB); + exitMBB->transferSuccessors(BB); + + // thisMBB: + // ... + // fallthrough --> loopMBB + BB->addSuccessor(loopMBB); + + // loopMBB: + // l[wd]arx dest, ptr + // cmp[wd] dest, oldval + // bne- exitMBB + // st[wd]cx. newval, ptr + // bne- loopMBB + // fallthrough --> exitMBB + BB = loopMBB; + BuildMI(BB, TII->get(is64bit ? PPC::LDARX : PPC::LWARX), dest) + .addReg(ptrA).addReg(ptrB); + BuildMI(BB, TII->get(is64bit ? PPC::CMPD : PPC::CMPW), PPC::CR0) + .addReg(oldval).addReg(dest); + BuildMI(BB, TII->get(PPC::BCC)) + .addImm(PPC::PRED_NE).addReg(PPC::CR0).addMBB(exitMBB); + BuildMI(BB, TII->get(is64bit ? PPC::STDCX : PPC::STWCX)) + .addReg(newval).addReg(ptrA).addReg(ptrB); + BuildMI(BB, TII->get(PPC::BCC)) + .addImm(PPC::PRED_NE).addReg(PPC::CR0).addMBB(loopMBB); + BB->addSuccessor(loopMBB); + BB->addSuccessor(exitMBB); + + // exitMBB: + // ... + BB = exitMBB; + } + else if (MI->getOpcode() == PPC::ATOMIC_SWAP_I32 || + MI->getOpcode() == PPC::ATOMIC_SWAP_I64) { + bool is64bit = MI->getOpcode() == PPC::ATOMIC_SWAP_I64; + + unsigned dest = MI->getOperand(0).getReg(); + unsigned ptrA = MI->getOperand(1).getReg(); + unsigned ptrB = MI->getOperand(2).getReg(); + unsigned newval = MI->getOperand(3).getReg(); + + MachineBasicBlock *loopMBB = new MachineBasicBlock(LLVM_BB); + MachineBasicBlock *exitMBB = new MachineBasicBlock(LLVM_BB); + + MachineFunction *F = BB->getParent(); + F->getBasicBlockList().insert(It, loopMBB); + F->getBasicBlockList().insert(It, exitMBB); + exitMBB->transferSuccessors(BB); + + // thisMBB: + // ... + // fallthrough --> loopMBB + BB->addSuccessor(loopMBB); + + // loopMBB: + // l[wd]arx dest, ptr + // st[wd]cx. newval, ptr + // bne- loopMBB + // fallthrough --> exitMBB + BB = loopMBB; + BuildMI(BB, TII->get(is64bit ? PPC::LDARX : PPC::LWARX), dest) + .addReg(ptrA).addReg(ptrB); + BuildMI(BB, TII->get(is64bit ? PPC::STDCX : PPC::STWCX)) + .addReg(newval).addReg(ptrA).addReg(ptrB); + BuildMI(BB, TII->get(PPC::BCC)) + .addImm(PPC::PRED_NE).addReg(PPC::CR0).addMBB(loopMBB); + BB->addSuccessor(loopMBB); + BB->addSuccessor(exitMBB); + + // exitMBB: + // ... + BB = exitMBB; + } + else { + assert(0 && "Unexpected instr type to insert"); + } + delete MI; // The pseudo instruction is gone now. return BB; }
Apparently Analagous Threads
- [LLVMdev] Implementing llvm.atomic.cmp.swap.i32 on PowerPC
- [LLVMdev] Implementing llvm.atomic.cmp.swap.i32 on PowerPC
- [LLVMdev] Implementing llvm.atomic.cmp.swap.i32 on PowerPC
- [LLVMdev] Implementing llvm.atomic.cmp.swap.i32 on PowerPC
- [LLVMdev] Implementing llvm.atomic.cmp.swap.i32 on PowerPC