[RISCV] Scheduler description for Rocket Core

[HsiangKai]

Summary

[RISCV] Pipeline scheduler model for RISCV Rocket micro-architecture using the MIScheduler interface. Support for 32 and 64-bit Rocket cores is implemented.

[kito-cheng]

llvm/lib/Target/RISCV/RISCVInstrInfoM.td

35

Predicates are removed by accident?

[lenary]

A few nits

llvm/lib/Target/RISCV/RISCVSchedRocket32.td

9

Hmmmm :)

llvm/lib/Target/RISCV/RISCVSchedRocket64.td

9

Hmmmm :)

[lenary]

llvm/lib/Target/RISCV/RISCVSchedRocket32.td

1

This should match the filename.

llvm/lib/Target/RISCV/RISCVSchedRocket64.td

1

This should match the filename.

[luismarques]

llvm/lib/Target/RISCV/RISCVSchedRocket32.td

79

typo

[rogfer01]

@javedabsar (or @javed.absar) I seem to recall you have experience with schedulers. If you could give us a hand here that'd be great! :)

[javed.absar]

In D68685#1705570, @rogfer01 wrote:

@javedabsar (or @javed.absar) I seem to recall you have experience with schedulers. If you could give us a hand here that'd be great! :)

Sure no problem Roger :)

Could you please point me to some doc which describes the pipeline model of RISCVRocket64 - i.e. what kind of processing units are available, how each instruction flows through the pipeline (fully pipelined or partially, latencies, resource dependences)?

That would be my starting point to match against the schedules defined in schedule*.td.

[javed.absar]

Is there a forwarding mechanism in Rocket? If so, that can be added to make the model more accurate.

llvm/lib/Target/RISCV/RISCVSchedRocket32.td

73

I guess its input dependent. Probably better to take average case 17.
Is 'Rocket32UnitIMul]' intentional (i.e. Div using Mul instead of UnitDiv) ?
If DIV is not pipelined then please put ResourceCycle = ... , as well.

llvm/lib/Target/RISCV/RISCVSchedRocket64.td

1

What are the main differences in terms of schedule between Rocket63 and . Looks like to me that a lot of code can be factored out into a common td

92

You could use let Latency = 2 in { ... } to avoid repeating

[javed.absar]

Hi Michael:

By the way, thanks for the link - https://www.lowrisc.org/docs/tagged-memory-v0.1/rocket-core/ . But the waybacklink doesnt work for me. Is there an alternate way to get the arch doc. Thanks

[apazos]

llvm/lib/Target/RISCV/RISCVSchedRocket32.td

16

Can you add to the comments a link to the processor model documentation used in the schedule definitions (latencies, function units types, etc.)?

Have you been able to run utils/schedcover.py on Rocket32Model and Rocket64Model?

[HsiangKai]

Some suggestions inlined.

llvm/lib/Target/RISCV/RISCVInstrInfo.td

323

Specify SchedReadWrite for $rs1.

374

If all BranchCC_rri have the same Sched<> list, it could be specified in BranchCC_rri class.

380

I supposed you should specify each operand with SchedReadWrite type. It will be the interface between instructions and scheduling model. So, I think we should preserve the place holder for input operands. It will look like

def LB : Load_ri<0b000, "lb">, Sched<[WriteLDB, ReadLDB]>;

All the following instructions should specify the SchedReadWrite types for each operand, no matter output operand or input operand.

llvm/lib/Target/RISCV/RISCVSchedule.td

105

They are the only two SchedRead definitions. However, there is no place using them. I think you should specify more SchedRead types and associate these SchedRead to input operands in instruction definitions.

[jrtc27]

If you have let UnsupportedFeatures = [IsRV32] (or similarly for RV64) in your model definition, you can avoid needing to list a bunch of scheduler resources with let Unsupported = 1;.

[jrtc27]

I wonder whether, instead of putting all the scheduling resource information as part of the instruction definition, we should be doing something like we do with patterns, ie declaring them separately (either in each RISCVInstrInfoX.td, or in RISCVSchedule.td to keep scheduling completely separate from encoding and codegen). For example (formatting aside):

def : InstRW<[WriteFALU32], (instrs FADD_S, FSUB_S, FSGNJ_S, FSGNJN_S, FSGNJX_S, FMIN_S, FMAX_S)>;

[jrtc27]

Do we need separate schedule information for compressed and uncompressed instructions? I would assume that any sensible RVC implementation would decode the two to exactly the same control logic (other than for PC incrementing and the original instruction for mtval).

[HsiangKai]

In D68685#1714378, @jrtc27 wrote:

I wonder whether, instead of putting all the scheduling resource information as part of the instruction definition, we should be doing something like we do with patterns, ie declaring them separately (either in each RISCVInstrInfoX.td, or in RISCVSchedule.td to keep scheduling completely separate from encoding and codegen). For example (formatting aside):

def : InstRW<[WriteFALU32], (instrs FADD_S, FSUB_S, FSGNJ_S, FSGNJN_S, FSGNJX_S, FMIN_S, FMAX_S)>;

In general, InstRW is used to describe subtarget-specific scheduling behavior for specific instructions. If we use InstRW to describe the whole pipeline model, we need to create the mappings between scheduling types (SchedReadWrite) to instructions for every subtarget. It is a tedious work when adding a new subtarget for the target. I think it is more reasonable to associate SchedReadWrite list to instructions for target and specify latencies and cycles to these SchedReadWrite in subtarget. We only need to associate SchedReadWrite list to instructions only once and specify latencies and cycles for every subtarget. It is more modular and it is easier for subtarget.

[javed.absar]

In D68685#1716040, @HsiangKai wrote:

In D68685#1714378, @jrtc27 wrote:

I wonder whether, instead of putting all the scheduling resource information as part of the instruction definition, we should be doing something like we do with patterns, ie declaring them separately (either in each RISCVInstrInfoX.td, or in RISCVSchedule.td to keep scheduling completely separate from encoding and codegen). For example (formatting aside):

def : InstRW<[WriteFALU32], (instrs FADD_S, FSUB_S, FSGNJ_S, FSGNJN_S, FSGNJX_S, FMIN_S, FMAX_S)>;

In general, InstRW is used to describe subtarget-specific scheduling behavior for specific instructions. If we use InstRW to describe the whole pipeline model, we need to create the mappings between scheduling types (SchedReadWrite) to instructions for every subtarget. It is a tedious work when adding a new subtarget for the target. I think it is more reasonable to associate SchedReadWrite list to instructions for target and specify latencies and cycles to these SchedReadWrite in subtarget. We only need to associate SchedReadWrite list to instructions only once and specify latencies and cycles for every subtarget. It is more modular and it is easier for subtarget.

I too think that defining SchedReadWrite and associating them with Instruction classes is better (rather than wholly relying on InstRW). In fact there was a period I rewrote large part of ARM schedule.tds to move away from InstRW approach.

Having said that, it looks like we might be defining too many SchedReadWrite categories to differentiate instruction types that may actually have identical latency and resource numbers.

[compilerguy]

Addressed review comments.

[HsiangKai]

In D68685#1710474, @javed.absar wrote:

Hi Michael:

By the way, thanks for the link - https://www.lowrisc.org/docs/tagged-memory-v0.1/rocket-core/ . But the waybacklink doesnt work for me. Is there an alternate way to get the arch doc. Thanks

Hi @javed.absar,

Thanks for your review and comments. You could find the pipeline model in http://www-inst.eecs.berkeley.edu/~cs250/fa13/handouts/lab2-riscv.pdf, page 13. In addition, SiFive U54 is based on Rocket microarchitecture, so you could also refer to https://sifive.cdn.prismic.io/sifive%2Fac48c4fd-af85-46be-9dd9-22aa34ba6977_u54mc-core-complex-manual-v19.05.pdf for pipeline information. I know the information is limited in the documents. If there is any question that you can not find answer from documents, welcome to raise your questions. Thanks.

[HsiangKai]

In D68685#1714539, @jrtc27 wrote:

Do we need separate schedule information for compressed and uncompressed instructions? I would assume that any sensible RVC implementation would decode the two to exactly the same control logic (other than for PC incrementing and the original instruction for mtval).

We merged the schedule information for c-ext instructions. Thanks.

[HsiangKai]

llvm/lib/Target/RISCV/RISCVSchedRocket64.td

1

We also think that it should be factored out for the common part. However, we need to provide SchedMachineModel attribute to WriteRes and ReadAdvance and the model is different for Rocket32 and Rocket64. We have no idea how to factor out the common parts. Do you have any suggestions?

[khchen]

BTW, I used this patch and D71124 to run the CoreMark on HiFive unleashed with -O3 -mllvm -enable-misched -mllvm -enable-post-misched -mcpu=rocket-rv64,
the result is better than -O3 -mllvm -enable-misched -mllvm -enable-post-misched and get the performance 3.5% speedup (use ld.bfd and libgcc).

[Jim]

A few nits.
LLVM prefers spaces to tabs.

llvm/lib/Target/RISCV/RISCV.td

86

Blank line should not be deleted.

llvm/lib/Target/RISCV/RISCVInstrInfo.td

429

indentation with tabs.

436

indentation with tabs.

449

indentation with tabs.

459

indentation with tabs.

925

indentation with tabs.

930

indentation with tabs.

945

indentation with tabs.

[HsiangKai]

Fix format typo.

[evandro]

Overall, it looks nearly good enough. I don't know how the scheduler will behave is interesting cases when there is no ShedRead for each argument in the ins list of the instructions. However, it shouldn't be too difficult to add them in the instruction classes.

llvm/lib/Target/RISCV/RISCVInstrInfo.td

374

Agreed.

llvm/lib/Target/RISCV/RISCVSchedRocket32.td

32

You might consider setting BufferSize = 1 for this non-pipelined units.

37

You might consider setting BufferSize = 1 for this non-pipelined units.

61

Typically, for most values, the latency is near the worst case. I suggest using the worst case here. And, as @javed.absar said, it also needs ResourceCycle = <worst case>; here as well.

llvm/lib/Target/RISCV/RISCVSchedRocket64.td

31

You might consider setting BufferSize = 1 for this non-pipelined units.

36

You might consider setting BufferSize = 1 for this non-pipelined units.

60

Again, I suggest using the worst case here and below.

101

Per the comment above, is it 5 or 6?

126

Outline the Latency to avoid repetitions.

129

Ditto.

llvm/lib/Target/RISCV/RISCVSchedule.td

23

I don't think that nop needs a scheduling class.

28

s/WriteLDDW/WriteLDD/

33

s/WriteLSTDW/WriteSTD/

37

s/WriteAtomicLDDW/WriteAtomicLDD/

39

s/WriteAtomicSTDW/WriteAtomicSTD/

58

Swap this line with the one above it.

84

s/WriteFLd32/WriteFLD32/

85

s/WriteFLd64/WriteFLD64/

86

s/WriteFSt32s/WriteFST32/

87

s/WriteFSt64/WriteFST64/

[evandro]

llvm/lib/Target/RISCV/RISCVSchedule.td

23

Oh, never mind me. A SchedWrite is necessary to capture its µop, assuming it's not subsumed by the fetch engine.

[shiva0217]

llvm/lib/Target/RISCV/RISCVSchedule.td

105

Will the SchedRead for input operands be added, so they could be used by ReadAdvance to describe forwarding rules?

[HsiangKai]

llvm/lib/Target/RISCV/RISCVSchedule.td

105

I also think so. I will add SchedRead for input operands.

[HsiangKai]

Address Evandro's comments.

[lenary]

llvm/lib/Target/RISCV/RISCVSchedule.td

105

I presume this patch is not yet ready to land because it is still missing SchedRead for most input operands? Do you expect to have those ready before the LLVM 10.0 branch?

[HsiangKai]

llvm/lib/Target/RISCV/RISCVSchedule.td

105

I will add SchedRead for input operands for these two days.

[HsiangKai]

Add SchedRead for input operands.

[evandro]

Except for a handful of gnats, this is really close to LGTM.

• The stores also need a SchedRead for register being stored, besides one for the base address register.
• nop, both I and C, should have a ShedWrite to capture their µop, as some µarchitectures might subsume the µop in the front end.
• The Pseudo class definition in RISCVInstrFormats.td should set the SchedRW to an empty list [] so that you wouldn't have to include any SchedReadWrite class for pseudo instructions.

llvm/lib/Target/RISCV/RISCVInstrInfo.td

382–384

Stores also read the base address in a register.

479

Read base address in register.

llvm/lib/Target/RISCV/RISCVInstrInfoC.td

330

Again, the base address is in an input register.

364

This should use the same SchedReadWrite class as for the I nop.

527

Input register needs a SchedRead.

586

Same ShedReadWrite class as for the I nop.

llvm/lib/Target/RISCV/RISCVInstrInfoD.td

79

SchedReads missing.

llvm/lib/Target/RISCV/RISCVSchedule.td

127

A single SchedRead for the base address register should suffice for RV32 and RV64 loads and stores.

[HsiangKai]

Address Evandro's comments.

[evandro]

It LGTM, but let's wait till tomorrow, when the patch will be 3 business days old, for any input.

[evandro]

Going thrice!

[lenary]

I am happy with the changes, though unfamiliar with Schedulers specifically (except what I have tried to learn when reading this patch).

I am happy that the reviews have been thorough enough for this patch to land.

llvm/lib/Target/RISCV/RISCV.td

 //===----------------------------------------------------------------------===//
 
 include "RISCVSystemOperands.td"
 
 //===----------------------------------------------------------------------===//
 // Registers, calling conventions, instruction descriptions.
 //===----------------------------------------------------------------------===//
 
+include "RISCVSchedule.td"
 include "RISCVRegisterInfo.td"
 include "RISCVCallingConv.td"
 include "RISCVInstrInfo.td"
 include "RISCVRegisterBanks.td"
-
+include "RISCVSchedRocket32.td"
+include "RISCVSchedRocket64.td"
 //===----------------------------------------------------------------------===//
 // RISC-V processors supported.
 //===----------------------------------------------------------------------===//
 
 def : ProcessorModel<"generic-rv32", NoSchedModel, [FeatureRVCHints]>;
 
 def : ProcessorModel<"generic-rv64", NoSchedModel, [Feature64Bit,
                      FeatureRVCHints]>;
 
+def : ProcessorModel<"rocket-rv32", Rocket32Model, [FeatureRVCHints]>;
+
+def : ProcessorModel<"rocket-rv64", Rocket64Model, [Feature64Bit,
+                     FeatureRVCHints]>;
+
+
 //===----------------------------------------------------------------------===//
 // Define the RISC-V target.
 //===----------------------------------------------------------------------===//
 
 def RISCVInstrInfo : InstrInfo {
   let guessInstructionProperties = 0;
 }
 

llvm/lib/Target/RISCV/RISCVInstrInfo.td

 let hasSideEffects = 0, mayLoad = 0, mayStore = 0 in
 class ALU_rr<bits<7> funct7, bits<3> funct3, string opcodestr>
     : RVInstR<funct7, funct3, OPC_OP, (outs GPR:$rd), (ins GPR:$rs1, GPR:$rs2),
               opcodestr, "$rd, $rs1, $rs2">;
 
 let hasSideEffects = 1, mayLoad = 0, mayStore = 0 in
 class CSR_ir<bits<3> funct3, string opcodestr>
     : RVInstI<funct3, OPC_SYSTEM, (outs GPR:$rd), (ins csr_sysreg:$imm12, GPR:$rs1),
-              opcodestr, "$rd, $imm12, $rs1">;
+              opcodestr, "$rd, $imm12, $rs1">, Sched<[WriteCSR]>;
 
 let hasSideEffects = 1, mayLoad = 0, mayStore = 0 in
 class CSR_ii<bits<3> funct3, string opcodestr>
     : RVInstI<funct3, OPC_SYSTEM, (outs GPR:$rd),
               (ins csr_sysreg:$imm12, uimm5:$rs1),
-              opcodestr, "$rd, $imm12, $rs1">;
+              opcodestr, "$rd, $imm12, $rs1">, Sched<[WriteCSR]>;
 
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0 in
 class ShiftW_ri<bit arithshift, bits<3> funct3, string opcodestr>
     : RVInstIShiftW<arithshift, funct3, OPC_OP_IMM_32, (outs GPR:$rd),
                     (ins GPR:$rs1, uimm5:$shamt), opcodestr,
                     "$rd, $rs1, $shamt">;
 
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0 in
 class ALUW_rr<bits<7> funct7, bits<3> funct3, string opcodestr>
     : RVInstR<funct7, funct3, OPC_OP_32, (outs GPR:$rd),
               (ins GPR:$rs1, GPR:$rs2), opcodestr, "$rd, $rs1, $rs2">;
 
 let hasSideEffects = 1, mayLoad = 0, mayStore = 0 in
 class Priv<string opcodestr, bits<7> funct7>
     : RVInstR<funct7, 0b000, OPC_SYSTEM, (outs), (ins GPR:$rs1, GPR:$rs2),
               opcodestr, "">;
 
 //===----------------------------------------------------------------------===//
 // Instructions
 //===----------------------------------------------------------------------===//
 
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0 in {
 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
 def LUI : RVInstU<OPC_LUI, (outs GPR:$rd), (ins uimm20_lui:$imm20),
-                  "lui", "$rd, $imm20">;
+                  "lui", "$rd, $imm20">, Sched<[WriteIALU]>;
 
 def AUIPC : RVInstU<OPC_AUIPC, (outs GPR:$rd), (ins uimm20_auipc:$imm20),
-                    "auipc", "$rd, $imm20">;
+                    "auipc", "$rd, $imm20">, Sched<[WriteIALU]>;
 
 let isCall = 1 in
 def JAL : RVInstJ<OPC_JAL, (outs GPR:$rd), (ins simm21_lsb0_jal:$imm20),
-                  "jal", "$rd, $imm20">;
+                  "jal", "$rd, $imm20">, Sched<[WriteJal]>;
 
 let isCall = 1 in
 def JALR : RVInstI<0b000, OPC_JALR, (outs GPR:$rd),
                    (ins GPR:$rs1, simm12:$imm12),
-                   "jalr", "$rd, ${imm12}(${rs1})">;
+                   "jalr", "$rd, ${imm12}(${rs1})">, Sched<[WriteJalr]>;
 } // hasSideEffects = 0, mayLoad = 0, mayStore = 0
 
-def BEQ  : BranchCC_rri<0b000, "beq">;
-def BNE  : BranchCC_rri<0b001, "bne">;
-def BLT  : BranchCC_rri<0b100, "blt">;
-def BGE  : BranchCC_rri<0b101, "bge">;
-def BLTU : BranchCC_rri<0b110, "bltu">;
-def BGEU : BranchCC_rri<0b111, "bgeu">;
-
-def LB  : Load_ri<0b000, "lb">;
-def LH  : Load_ri<0b001, "lh">;
-def LW  : Load_ri<0b010, "lw">;
-def LBU : Load_ri<0b100, "lbu">;
-def LHU : Load_ri<0b101, "lhu">;
-
-def SB : Store_rri<0b000, "sb">;
-def SH : Store_rri<0b001, "sh">;
-def SW : Store_rri<0b010, "sw">;
+def BEQ  : BranchCC_rri<0b000, "beq">, Sched<[WriteJmp]>;
+def BNE  : BranchCC_rri<0b001, "bne">, Sched<[WriteJmp]>;
+def BLT  : BranchCC_rri<0b100, "blt">, Sched<[WriteJmp]>;
+def BGE  : BranchCC_rri<0b101, "bge">, Sched<[WriteJmp]>;
+def BLTU : BranchCC_rri<0b110, "bltu">, Sched<[WriteJmp]>;
+def BGEU : BranchCC_rri<0b111, "bgeu">, Sched<[WriteJmp]>;
+
+def LB  : Load_ri<0b000, "lb">, Sched<[WriteLDB]>;
+def LH  : Load_ri<0b001, "lh">, Sched<[WriteLDH]>;
+def LW  : Load_ri<0b010, "lw">, Sched<[WriteLDW]>;
+def LBU : Load_ri<0b100, "lbu">, Sched<[WriteLDB]>;
+def LHU : Load_ri<0b101, "lhu">, Sched<[WriteLDH]>;
+
+def SB : Store_rri<0b000, "sb">, Sched<[WriteSTB]>;
+def SH : Store_rri<0b001, "sh">, Sched<[WriteSTH]>;
+def SW : Store_rri<0b010, "sw">, Sched<[WriteSTW]>;
 
 // ADDI isn't always rematerializable, but isReMaterializable will be used as
 // a hint which is verified in isReallyTriviallyReMaterializable.
 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
-def ADDI  : ALU_ri<0b000, "addi">;
+def ADDI  : ALU_ri<0b000, "addi">, Sched<[WriteIALU]>;
 
-def SLTI  : ALU_ri<0b010, "slti">;
-def SLTIU : ALU_ri<0b011, "sltiu">;
+def SLTI  : ALU_ri<0b010, "slti">, Sched<[WriteIALU]>;
+def SLTIU : ALU_ri<0b011, "sltiu">, Sched<[WriteIALU]>;
 
 let isReMaterializable = 1, isAsCheapAsAMove = 1 in {
-def XORI  : ALU_ri<0b100, "xori">;
-def ORI   : ALU_ri<0b110, "ori">;
+def XORI  : ALU_ri<0b100, "xori">, Sched<[WriteIALU]>;
+def ORI   : ALU_ri<0b110, "ori">, Sched<[WriteIALU]>;
 }
 
-def ANDI  : ALU_ri<0b111, "andi">;
+def ANDI  : ALU_ri<0b111, "andi">, Sched<[WriteIALU]>;
 
-def SLLI : Shift_ri<0, 0b001, "slli">;
-def SRLI : Shift_ri<0, 0b101, "srli">;
-def SRAI : Shift_ri<1, 0b101, "srai">;
-
-def ADD  : ALU_rr<0b0000000, 0b000, "add">;
-def SUB  : ALU_rr<0b0100000, 0b000, "sub">;
-def SLL  : ALU_rr<0b0000000, 0b001, "sll">;
-def SLT  : ALU_rr<0b0000000, 0b010, "slt">;
-def SLTU : ALU_rr<0b0000000, 0b011, "sltu">;
-def XOR  : ALU_rr<0b0000000, 0b100, "xor">;
-def SRL  : ALU_rr<0b0000000, 0b101, "srl">;
-def SRA  : ALU_rr<0b0100000, 0b101, "sra">;
-def OR   : ALU_rr<0b0000000, 0b110, "or">;
-def AND  : ALU_rr<0b0000000, 0b111, "and">;
+def SLLI : Shift_ri<0, 0b001, "slli">, Sched<[WriteShift]>;
+def SRLI : Shift_ri<0, 0b101, "srli">, Sched<[WriteShift]>;
+def SRAI : Shift_ri<1, 0b101, "srai">, Sched<[WriteShift]>;
+
+def ADD  : ALU_rr<0b0000000, 0b000, "add">, Sched<[WriteIALU]>;
+def SUB  : ALU_rr<0b0100000, 0b000, "sub">, Sched<[WriteIALU]>;
+def SLL  : ALU_rr<0b0000000, 0b001, "sll">, Sched<[WriteIALU]>;
+def SLT  : ALU_rr<0b0000000, 0b010, "slt">, Sched<[WriteIALU]>;
+def SLTU : ALU_rr<0b0000000, 0b011, "sltu">, Sched<[WriteIALU]>;
+def XOR  : ALU_rr<0b0000000, 0b100, "xor">, Sched<[WriteIALU]>;
+def SRL  : ALU_rr<0b0000000, 0b101, "srl">, Sched<[WriteIALU]>;
+def SRA  : ALU_rr<0b0100000, 0b101, "sra">, Sched<[WriteIALU]>;
+def OR   : ALU_rr<0b0000000, 0b110, "or">, Sched<[WriteIALU]>;
+def AND  : ALU_rr<0b0000000, 0b111, "and">, Sched<[WriteIALU]>;
 
 let hasSideEffects = 1, mayLoad = 0, mayStore = 0 in {
 def FENCE : RVInstI<0b000, OPC_MISC_MEM, (outs),
                     (ins fencearg:$pred, fencearg:$succ),
-                    "fence", "$pred, $succ"> {
+                    "fence", "$pred, $succ">, Sched<[]> {
   bits<4> pred;
   bits<4> succ;
 
   let rs1 = 0;
   let rd = 0;
   let imm12 = {0b0000,pred,succ};
 }
 
-def FENCE_TSO : RVInstI<0b000, OPC_MISC_MEM, (outs), (ins), "fence.tso", ""> {
+def FENCE_TSO : RVInstI<0b000, OPC_MISC_MEM, (outs), (ins), "fence.tso", "">,
+              Sched<[WriteFence]> {
   let rs1 = 0;
   let rd = 0;
   let imm12 = {0b1000,0b0011,0b0011};
 }
 
-def FENCE_I : RVInstI<0b001, OPC_MISC_MEM, (outs), (ins), "fence.i", ""> {
+def FENCE_I : RVInstI<0b001, OPC_MISC_MEM, (outs), (ins), "fence.i", "">,
+            Sched<[WriteFence]> {
   let rs1 = 0;
   let rd = 0;
   let imm12 = 0;
 }
 
-def ECALL : RVInstI<0b000, OPC_SYSTEM, (outs), (ins), "ecall", ""> {
+def ECALL : RVInstI<0b000, OPC_SYSTEM, (outs), (ins), "ecall", "">, Sched<[WriteJmp]> {
   let rs1 = 0;
   let rd = 0;
   let imm12 = 0;
 }
 
-def EBREAK : RVInstI<0b000, OPC_SYSTEM, (outs), (ins), "ebreak", ""> {
+def EBREAK : RVInstI<0b000, OPC_SYSTEM, (outs), (ins), "ebreak", "">,
+           Sched<[]> {
   let rs1 = 0;
   let rd = 0;
   let imm12 = 1;
 }
 
 // This is a de facto standard (as set by GNU binutils) 32-bit unimplemented
 // instruction (i.e., it should always trap, if your implementation has invalid
 // instruction traps).
-def UNIMP : RVInstI<0b001, OPC_SYSTEM, (outs), (ins), "unimp", ""> {
+def UNIMP : RVInstI<0b001, OPC_SYSTEM, (outs), (ins), "unimp", "">,
+          Sched<[]> {
   let rs1 = 0;
   let rd = 0;
   let imm12 = 0b110000000000;
 }
 } // hasSideEffects = 1, mayLoad = 0, mayStore = 0
 
 def CSRRW : CSR_ir<0b001, "csrrw">;
 def CSRRS : CSR_ir<0b010, "csrrs">;
 def CSRRC : CSR_ir<0b011, "csrrc">;
 
 def CSRRWI : CSR_ii<0b101, "csrrwi">;
 def CSRRSI : CSR_ii<0b110, "csrrsi">;
 def CSRRCI : CSR_ii<0b111, "csrrci">;
 
 /// RV64I instructions
 
 let Predicates = [IsRV64] in {
-def LWU   : Load_ri<0b110, "lwu">;
-def LD    : Load_ri<0b011, "ld">;
-def SD    : Store_rri<0b011, "sd">;
+def LWU   : Load_ri<0b110, "lwu">, Sched<[WriteLDWU]>;
+def LD    : Load_ri<0b011, "ld">, Sched<[WriteLDDW]>;
+def SD    : Store_rri<0b011, "sd">, Sched<[WriteSTDW]>;
 
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0 in
 def ADDIW : RVInstI<0b000, OPC_OP_IMM_32, (outs GPR:$rd),
                     (ins GPR:$rs1, simm12:$imm12),
-                    "addiw", "$rd, $rs1, $imm12">;
+                    "addiw", "$rd, $rs1, $imm12">, Sched<[WriteIALU32]>;
 
-def SLLIW : ShiftW_ri<0, 0b001, "slliw">;
-def SRLIW : ShiftW_ri<0, 0b101, "srliw">;
-def SRAIW : ShiftW_ri<1, 0b101, "sraiw">;
-
-def ADDW  : ALUW_rr<0b0000000, 0b000, "addw">;
-def SUBW  : ALUW_rr<0b0100000, 0b000, "subw">;
-def SLLW  : ALUW_rr<0b0000000, 0b001, "sllw">;
-def SRLW  : ALUW_rr<0b0000000, 0b101, "srlw">;
-def SRAW  : ALUW_rr<0b0100000, 0b101, "sraw">;
+def SLLIW : ShiftW_ri<0, 0b001, "slliw">, Sched<[WriteShift32]>;
+def SRLIW : ShiftW_ri<0, 0b101, "srliw">, Sched<[WriteShift32]>;
+def SRAIW : ShiftW_ri<1, 0b101, "sraiw">, Sched<[WriteShift32]>;
+
+def ADDW  : ALUW_rr<0b0000000, 0b000, "addw">, Sched<[WriteIALU32]>;
+def SUBW  : ALUW_rr<0b0100000, 0b000, "subw">, Sched<[WriteIALU32]>;
+def SLLW  : ALUW_rr<0b0000000, 0b001, "sllw">, Sched<[WriteIALU32]>;
+def SRLW  : ALUW_rr<0b0000000, 0b101, "srlw">, Sched<[WriteIALU32]>;
+def SRAW  : ALUW_rr<0b0100000, 0b101, "sraw">, Sched<[WriteIALU32]>;
 } // Predicates = [IsRV64]
 
 //===----------------------------------------------------------------------===//
 // Privileged instructions
 //===----------------------------------------------------------------------===//
 
 let isBarrier = 1, isReturn = 1, isTerminator = 1 in {
-def URET : Priv<"uret", 0b0000000> {
+def URET : Priv<"uret", 0b0000000>, Sched<[WriteTrapRet]> {
   let rd = 0;
   let rs1 = 0;
   let rs2 = 0b00010;
 }
 
-def SRET : Priv<"sret", 0b0001000> {
+def SRET : Priv<"sret", 0b0001000>, Sched<[WriteTrapRet]> {
   let rd = 0;
   let rs1 = 0;
   let rs2 = 0b00010;
 }
 
-def MRET : Priv<"mret", 0b0011000> {
+def MRET : Priv<"mret", 0b0011000>, Sched<[WriteTrapRet]> {
   let rd = 0;
   let rs1 = 0;
   let rs2 = 0b00010;
 }
 } // isBarrier = 1, isReturn = 1, isTerminator = 1
 
-def WFI : Priv<"wfi", 0b0001000> {
+def WFI : Priv<"wfi", 0b0001000>, Sched<[]> {
   let rd = 0;
   let rs1 = 0;
   let rs2 = 0b00101;
 }
 
 let hasSideEffects = 1, mayLoad = 0, mayStore = 0 in
 def SFENCE_VMA : RVInstR<0b0001001, 0b000, OPC_SYSTEM, (outs),
                          (ins GPR:$rs1, GPR:$rs2),
-                         "sfence.vma", "$rs1, $rs2"> {
+                         "sfence.vma", "$rs1, $rs2">, Sched<[WriteFence]> {
   let rd = 0;
 }
 
 //===----------------------------------------------------------------------===//
 // Assembler Pseudo Instructions (User-Level ISA, Version 2.2, Chapter 20)
 //===----------------------------------------------------------------------===//
 
 def : InstAlias<"nop",           (ADDI      X0,      X0,       0)>;
 
 // Note that the size is 32 because up to 8 32-bit instructions are needed to
 // generate an arbitrary 64-bit immediate. However, the size does not really
 // matter since PseudoLI is currently only used in the AsmParser where it gets
 // expanded to real instructions immediately.
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0, Size = 32,
     isCodeGenOnly = 0, isAsmParserOnly = 1 in
 def PseudoLI : Pseudo<(outs GPR:$rd), (ins ixlenimm_li:$imm), [],
-                      "li", "$rd, $imm">;
+                      "li", "$rd, $imm">, Sched<[]>;
 
-def PseudoLB  : PseudoLoad<"lb">;
-def PseudoLBU : PseudoLoad<"lbu">;
-def PseudoLH  : PseudoLoad<"lh">;
-def PseudoLHU : PseudoLoad<"lhu">;
-def PseudoLW  : PseudoLoad<"lw">;
-
-def PseudoSB  : PseudoStore<"sb">;
-def PseudoSH  : PseudoStore<"sh">;
-def PseudoSW  : PseudoStore<"sw">;
+def PseudoLB  : PseudoLoad<"lb">, Sched<[]>;
+def PseudoLBU : PseudoLoad<"lbu">, Sched<[]>;
+def PseudoLH  : PseudoLoad<"lh">, Sched<[]>;
+def PseudoLHU : PseudoLoad<"lhu">, Sched<[]>;
+def PseudoLW  : PseudoLoad<"lw">, Sched<[]>;
+
+def PseudoSB  : PseudoStore<"sb">, Sched<[]>;
+def PseudoSH  : PseudoStore<"sh">, Sched<[]>;
+def PseudoSW  : PseudoStore<"sw">, Sched<[]>;
 
 let Predicates = [IsRV64] in {
-def PseudoLWU : PseudoLoad<"lwu">;
-def PseudoLD  : PseudoLoad<"ld">;
-def PseudoSD  : PseudoStore<"sd">;
+def PseudoLWU : PseudoLoad<"lwu">, Sched<[]>;
+def PseudoLD  : PseudoLoad<"ld">, Sched<[]>;
+def PseudoSD  : PseudoStore<"sd">, Sched<[]>;
 } // Predicates = [IsRV64]
 
 def : InstAlias<"mv $rd, $rs",   (ADDI GPR:$rd, GPR:$rs,       0)>;
 def : InstAlias<"not $rd, $rs",  (XORI GPR:$rd, GPR:$rs,      -1)>;
 def : InstAlias<"neg $rd, $rs",  (SUB  GPR:$rd,      X0, GPR:$rs)>;
 
 let Predicates = [IsRV64] in {
 def : InstAlias<"negw $rd, $rs",   (SUBW  GPR:$rd,      X0, GPR:$rs)>;

 
 // This is a special case of the ADD instruction used to facilitate the use of a
 // fourth operand to emit a relocation on a symbol relating to this instruction.
 // The relocation does not affect any bits of the instruction itself but is used
 // as a hint to the linker.
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0, isCodeGenOnly = 0 in
 def PseudoAddTPRel : Pseudo<(outs GPR:$rd),
                             (ins GPR:$rs1, GPR:$rs2, tprel_add_symbol:$src), [],
-                            "add", "$rd, $rs1, $rs2, $src">;
+                            "add", "$rd, $rs1, $rs2, $src">, Sched<[]>;
 
 /// FrameIndex calculations
 
 def : Pat<(add (i32 AddrFI:$Rs), simm12:$imm12),
           (ADDI (i32 AddrFI:$Rs), simm12:$imm12)>;
 def : Pat<(IsOrAdd (i32 AddrFI:$Rs), simm12:$imm12),
           (ADDI (i32 AddrFI:$Rs), simm12:$imm12)>;
 

 def : Pat<(setle GPR:$rs1, GPR:$rs2), (XORI (SLT GPR:$rs2, GPR:$rs1), 1)>;
 
 let usesCustomInserter = 1 in
 class SelectCC_rrirr<RegisterClass valty, RegisterClass cmpty>
     : Pseudo<(outs valty:$dst),
              (ins cmpty:$lhs, cmpty:$rhs, ixlenimm:$imm,
               valty:$truev, valty:$falsev),
              [(set valty:$dst, (riscv_selectcc cmpty:$lhs, cmpty:$rhs,
-              (XLenVT imm:$imm), valty:$truev, valty:$falsev))]>;
+              (XLenVT imm:$imm), valty:$truev, valty:$falsev))]>, Sched<[]>;
 
 def Select_GPR_Using_CC_GPR : SelectCC_rrirr<GPR, GPR>;
 
 /// Branches and jumps
 
 // Match `(brcond (CondOp ..), ..)` and lower to the appropriate RISC-V branch
 // instruction.
 class BccPat<PatFrag CondOp, RVInstB Inst>

 def : BccSwapPat<setule, BGEU>;
 
 // An extra pattern is needed for a brcond without a setcc (i.e. where the
 // condition was calculated elsewhere).
 def : Pat<(brcond GPR:$cond, bb:$imm12), (BNE GPR:$cond, X0, bb:$imm12)>;
 
 let isBarrier = 1, isBranch = 1, isTerminator = 1 in
 def PseudoBR : Pseudo<(outs), (ins simm21_lsb0_jal:$imm20), [(br bb:$imm20)]>,
-               PseudoInstExpansion<(JAL X0, simm21_lsb0_jal:$imm20)>;
+               PseudoInstExpansion<(JAL X0, simm21_lsb0_jal:$imm20)>, Sched<[WriteJal]>;
 
 let isCall = 1, Defs=[X1] in
 let isBarrier = 1, isBranch = 1, isIndirectBranch = 1, isTerminator = 1 in
 def PseudoBRIND : Pseudo<(outs), (ins GPR:$rs1, simm12:$imm12), []>,
-                  PseudoInstExpansion<(JALR X0, GPR:$rs1, simm12:$imm12)>;
+                  PseudoInstExpansion<(JALR X0, GPR:$rs1, simm12:$imm12)>, Sched<[WriteJmpReg]>;
 
 def : Pat<(brind GPR:$rs1), (PseudoBRIND GPR:$rs1, 0)>;
 def : Pat<(brind (add GPR:$rs1, simm12:$imm12)),
           (PseudoBRIND GPR:$rs1, simm12:$imm12)>;
 
 // PsuedoCALLReg is a generic pseudo instruction for calls which will eventually
 // expand to auipc and jalr while encoding, with any given register used as the
 // destination.
 // Define AsmString to print "call" when compile with -S flag.
 // Define isCodeGenOnly = 0 to support parsing assembly "call" instruction.
 let isCall = 1, isBarrier = 1, isCodeGenOnly = 0, hasSideEffects = 0,
     mayStore = 0, mayLoad = 0 in
-def PseudoCALLReg : Pseudo<(outs GPR:$rd), (ins call_symbol:$func), []> {
+def PseudoCALLReg : Pseudo<(outs GPR:$rd), (ins call_symbol:$func), []>, Sched<[]> {
   let AsmString = "call\t$rd, $func";
 }
 
 // PseudoCALL is a pseudo instruction which will eventually expand to auipc
 // and jalr while encoding. This is desirable, as an auipc+jalr pair with
 // R_RISCV_CALL and R_RISCV_RELAX relocations can be be relaxed by the linker
 // if the offset fits in a signed 21-bit immediate.
 // Define AsmString to print "call" when compile with -S flag.
 // Define isCodeGenOnly = 0 to support parsing assembly "call" instruction.
 let isCall = 1, Defs = [X1], isCodeGenOnly = 0 in
-def PseudoCALL : Pseudo<(outs), (ins call_symbol:$func), []> {
+def PseudoCALL : Pseudo<(outs), (ins call_symbol:$func), []>, Sched<[]> {
   let AsmString = "call\t$func";
 }
 
 def : Pat<(riscv_call tglobaladdr:$func), (PseudoCALL tglobaladdr:$func)>;
 def : Pat<(riscv_call texternalsym:$func), (PseudoCALL texternalsym:$func)>;
 
 def : Pat<(riscv_uret_flag), (URET X0, X0)>;
 def : Pat<(riscv_sret_flag), (SRET X0, X0)>;
 def : Pat<(riscv_mret_flag), (MRET X0, X0)>;
 
 let isCall = 1, Defs = [X1] in
 def PseudoCALLIndirect : Pseudo<(outs), (ins GPR:$rs1),
                                 [(riscv_call GPR:$rs1)]>,
-                         PseudoInstExpansion<(JALR X1, GPR:$rs1, 0)>;
+                         PseudoInstExpansion<(JALR X1, GPR:$rs1, 0)>,
+       Sched<[WriteJalr]>;
 
 let isBarrier = 1, isReturn = 1, isTerminator = 1 in
 def PseudoRET : Pseudo<(outs), (ins), [(riscv_ret_flag)]>,
-                PseudoInstExpansion<(JALR X0, X1, 0)>;
+                PseudoInstExpansion<(JALR X0, X1, 0)>,
+    Sched<[WriteJalr]>;
 
 // PseudoTAIL is a pseudo instruction similar to PseudoCALL and will eventually
 // expand to auipc and jalr while encoding.
 // Define AsmString to print "tail" when compile with -S flag.
 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [X2],
     isCodeGenOnly = 0 in
-def PseudoTAIL : Pseudo<(outs), (ins call_symbol:$dst), []> {
+def PseudoTAIL : Pseudo<(outs), (ins call_symbol:$dst), []>, Sched<[]> {
   let AsmString = "tail\t$dst";
 }
 
 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [X2] in
 def PseudoTAILIndirect : Pseudo<(outs), (ins GPRTC:$rs1),
                                 [(riscv_tail GPRTC:$rs1)]>,
-                         PseudoInstExpansion<(JALR X0, GPR:$rs1, 0)>;
+                         PseudoInstExpansion<(JALR X0, GPR:$rs1, 0)>,
+       Sched<[WriteJalr]>;
 
 def : Pat<(riscv_tail (iPTR tglobaladdr:$dst)),
           (PseudoTAIL texternalsym:$dst)>;
 def : Pat<(riscv_tail (iPTR texternalsym:$dst)),
           (PseudoTAIL texternalsym:$dst)>;
 
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0, isCodeGenOnly = 0,
     isAsmParserOnly = 1 in
 def PseudoLLA : Pseudo<(outs GPR:$dst), (ins bare_symbol:$src), [],
-                       "lla", "$dst, $src">;
+                       "lla", "$dst, $src">, Sched<[]>;
 
 let hasSideEffects = 0, mayLoad = 1, mayStore = 0, isCodeGenOnly = 0,
     isAsmParserOnly = 1 in
 def PseudoLA : Pseudo<(outs GPR:$dst), (ins bare_symbol:$src), [],
-                      "la", "$dst, $src">;
+                      "la", "$dst, $src">, Sched<[]>;
 
 let hasSideEffects = 0, mayLoad = 1, mayStore = 0, isCodeGenOnly = 0,
     isAsmParserOnly = 1 in
 def PseudoLA_TLS_IE : Pseudo<(outs GPR:$dst), (ins bare_symbol:$src), [],
-                             "la.tls.ie", "$dst, $src">;
+                             "la.tls.ie", "$dst, $src">, Sched<[]>;
 
 let hasSideEffects = 0, mayLoad = 1, mayStore = 0, isCodeGenOnly = 0,
     isAsmParserOnly = 1 in
 def PseudoLA_TLS_GD : Pseudo<(outs GPR:$dst), (ins bare_symbol:$src), [],
-                             "la.tls.gd", "$dst, $src">;
+                             "la.tls.gd", "$dst, $src">, Sched<[]>;
 
 /// Loads
 
 multiclass LdPat<PatFrag LoadOp, RVInst Inst> {
   def : Pat<(LoadOp GPR:$rs1), (Inst GPR:$rs1, 0)>;
   def : Pat<(LoadOp AddrFI:$rs1), (Inst AddrFI:$rs1, 0)>;
   def : Pat<(LoadOp (add GPR:$rs1, simm12:$imm12)),
             (Inst GPR:$rs1, simm12:$imm12)>;

 // present. This is necessary as it isn't valid to mix __atomic_* libcalls
 // with inline atomic operations for the same object.
 
 /// Other pseudo-instructions
 
 // Pessimistically assume the stack pointer will be clobbered
 let Defs = [X2], Uses = [X2] in {
 def ADJCALLSTACKDOWN : Pseudo<(outs), (ins i32imm:$amt1, i32imm:$amt2),
-                              [(callseq_start timm:$amt1, timm:$amt2)]>;
+                              [(callseq_start timm:$amt1, timm:$amt2)]>, Sched<[]>;
 def ADJCALLSTACKUP   : Pseudo<(outs), (ins i32imm:$amt1, i32imm:$amt2),
-                              [(callseq_end timm:$amt1, timm:$amt2)]>;
+                              [(callseq_end timm:$amt1, timm:$amt2)]>, Sched<[]>;
 } // Defs = [X2], Uses = [X2]
 
 /// RV64 patterns
 
 let Predicates = [IsRV64] in {
 
 /// sext and zext
 

llvm/lib/Target/RISCV/RISCVInstrInfoA.td

def : Pat<(StoreOp (IsOrAdd AddrFI:$rs1, simm12:$imm12), StTy:$rs2),
             (Inst StTy:$rs2, AddrFI:$rs1, simm12:$imm12)>;
 }
 
 //===----------------------------------------------------------------------===//
 // Instructions
 //===----------------------------------------------------------------------===//
 
 let Predicates = [HasStdExtA] in {
-defm LR_W       : LR_r_aq_rl<0b010, "lr.w">;
-defm SC_W       : AMO_rr_aq_rl<0b00011, 0b010, "sc.w">;
-defm AMOSWAP_W  : AMO_rr_aq_rl<0b00001, 0b010, "amoswap.w">;
-defm AMOADD_W   : AMO_rr_aq_rl<0b00000, 0b010, "amoadd.w">;
-defm AMOXOR_W   : AMO_rr_aq_rl<0b00100, 0b010, "amoxor.w">;
-defm AMOAND_W   : AMO_rr_aq_rl<0b01100, 0b010, "amoand.w">;
-defm AMOOR_W    : AMO_rr_aq_rl<0b01000, 0b010, "amoor.w">;
-defm AMOMIN_W   : AMO_rr_aq_rl<0b10000, 0b010, "amomin.w">;
-defm AMOMAX_W   : AMO_rr_aq_rl<0b10100, 0b010, "amomax.w">;
-defm AMOMINU_W  : AMO_rr_aq_rl<0b11000, 0b010, "amominu.w">;
-defm AMOMAXU_W  : AMO_rr_aq_rl<0b11100, 0b010, "amomaxu.w">;
+defm LR_W       : LR_r_aq_rl<0b010, "lr.w">, Sched<[WriteAtomicLDW]>;
+defm SC_W       : AMO_rr_aq_rl<0b00011, 0b010, "sc.w">, Sched<[WriteAtomicSTW]>;
+defm AMOSWAP_W  : AMO_rr_aq_rl<0b00001, 0b010, "amoswap.w">, Sched<[WriteAtomicW]>;
+defm AMOADD_W   : AMO_rr_aq_rl<0b00000, 0b010, "amoadd.w">, Sched<[WriteAtomicW]>;
+defm AMOXOR_W   : AMO_rr_aq_rl<0b00100, 0b010, "amoxor.w">, Sched<[WriteAtomicW]>;
+defm AMOAND_W   : AMO_rr_aq_rl<0b01100, 0b010, "amoand.w">, Sched<[WriteAtomicW]>;
+defm AMOOR_W    : AMO_rr_aq_rl<0b01000, 0b010, "amoor.w">, Sched<[WriteAtomicW]>;
+defm AMOMIN_W   : AMO_rr_aq_rl<0b10000, 0b010, "amomin.w">, Sched<[WriteAtomicW]>;
+defm AMOMAX_W   : AMO_rr_aq_rl<0b10100, 0b010, "amomax.w">, Sched<[WriteAtomicW]>;
+defm AMOMINU_W  : AMO_rr_aq_rl<0b11000, 0b010, "amominu.w">, Sched<[WriteAtomicW]>;
+defm AMOMAXU_W  : AMO_rr_aq_rl<0b11100, 0b010, "amomaxu.w">, Sched<[WriteAtomicW]>;
 } // Predicates = [HasStdExtA]
 
 let Predicates = [HasStdExtA, IsRV64] in {
-defm LR_D       : LR_r_aq_rl<0b011, "lr.d">;
-defm SC_D       : AMO_rr_aq_rl<0b00011, 0b011, "sc.d">;
-defm AMOSWAP_D  : AMO_rr_aq_rl<0b00001, 0b011, "amoswap.d">;
-defm AMOADD_D   : AMO_rr_aq_rl<0b00000, 0b011, "amoadd.d">;
-defm AMOXOR_D   : AMO_rr_aq_rl<0b00100, 0b011, "amoxor.d">;
-defm AMOAND_D   : AMO_rr_aq_rl<0b01100, 0b011, "amoand.d">;
-defm AMOOR_D    : AMO_rr_aq_rl<0b01000, 0b011, "amoor.d">;
-defm AMOMIN_D   : AMO_rr_aq_rl<0b10000, 0b011, "amomin.d">;
-defm AMOMAX_D   : AMO_rr_aq_rl<0b10100, 0b011, "amomax.d">;
-defm AMOMINU_D  : AMO_rr_aq_rl<0b11000, 0b011, "amominu.d">;
-defm AMOMAXU_D  : AMO_rr_aq_rl<0b11100, 0b011, "amomaxu.d">;
+defm LR_D       : LR_r_aq_rl<0b011, "lr.d">, Sched<[WriteAtomicLDDW]>;
+defm SC_D       : AMO_rr_aq_rl<0b00011, 0b011, "sc.d">, Sched<[WriteAtomicSTDW]>;
+defm AMOSWAP_D  : AMO_rr_aq_rl<0b00001, 0b011, "amoswap.d">, Sched<[WriteAtomicDW]>;
+defm AMOADD_D   : AMO_rr_aq_rl<0b00000, 0b011, "amoadd.d">, Sched<[WriteAtomicDW]>;
+defm AMOXOR_D   : AMO_rr_aq_rl<0b00100, 0b011, "amoxor.d">, Sched<[WriteAtomicDW]>;
+defm AMOAND_D   : AMO_rr_aq_rl<0b01100, 0b011, "amoand.d">, Sched<[WriteAtomicDW]>;
+defm AMOOR_D    : AMO_rr_aq_rl<0b01000, 0b011, "amoor.d">, Sched<[WriteAtomicDW]>;
+defm AMOMIN_D   : AMO_rr_aq_rl<0b10000, 0b011, "amomin.d">, Sched<[WriteAtomicDW]>;
+defm AMOMAX_D   : AMO_rr_aq_rl<0b10100, 0b011, "amomax.d">, Sched<[WriteAtomicDW]>;
+defm AMOMINU_D  : AMO_rr_aq_rl<0b11000, 0b011, "amominu.d">, Sched<[WriteAtomicDW]>;
+defm AMOMAXU_D  : AMO_rr_aq_rl<0b11100, 0b011, "amomaxu.d">, Sched<[WriteAtomicDW]>;
 } // Predicates = [HasStdExtA, IsRV64]
 
 //===----------------------------------------------------------------------===//
 // Pseudo-instructions and codegen patterns
 //===----------------------------------------------------------------------===//
 
 let Predicates = [HasStdExtA] in {
 

 class PseudoAMO : Pseudo<(outs GPR:$res, GPR:$scratch),
                          (ins GPR:$addr, GPR:$incr, ixlenimm:$ordering), []> {
   let Constraints = "@earlyclobber $res,@earlyclobber $scratch";
   let mayLoad = 1;
   let mayStore = 1;
   let hasSideEffects = 0;
 }
 
-def PseudoAtomicLoadNand32 : PseudoAMO;
+def PseudoAtomicLoadNand32 : PseudoAMO, Sched<[]>;
 // Ordering constants must be kept in sync with the AtomicOrdering enum in
 // AtomicOrdering.h.
 def : Pat<(atomic_load_nand_32_monotonic GPR:$addr, GPR:$incr),
           (PseudoAtomicLoadNand32 GPR:$addr, GPR:$incr, 2)>;
 def : Pat<(atomic_load_nand_32_acquire GPR:$addr, GPR:$incr),
           (PseudoAtomicLoadNand32 GPR:$addr, GPR:$incr, 4)>;
 def : Pat<(atomic_load_nand_32_release GPR:$addr, GPR:$incr),
           (PseudoAtomicLoadNand32 GPR:$addr, GPR:$incr, 5)>;
: Pat<(intrin GPR:$addr, GPR:$incr, GPR:$mask, timm:$ordering),
           (AMOInst GPR:$addr, GPR:$incr, GPR:$mask, imm:$ordering)>;
 
 class PseudoMaskedAMOMinMaxPat<Intrinsic intrin, Pseudo AMOInst>
     : Pat<(intrin GPR:$addr, GPR:$incr, GPR:$mask, GPR:$shiftamt,
            timm:$ordering),
           (AMOInst GPR:$addr, GPR:$incr, GPR:$mask, GPR:$shiftamt,
            imm:$ordering)>;
 
-def PseudoMaskedAtomicSwap32 : PseudoMaskedAMO;
+def PseudoMaskedAtomicSwap32 : PseudoMaskedAMO, Sched<[]>;
 def : PseudoMaskedAMOPat<int_riscv_masked_atomicrmw_xchg_i32,
                          PseudoMaskedAtomicSwap32>;
-def PseudoMaskedAtomicLoadAdd32 : PseudoMaskedAMO;
+def PseudoMaskedAtomicLoadAdd32 : PseudoMaskedAMO, Sched<[]>;
 def : PseudoMaskedAMOPat<int_riscv_masked_atomicrmw_add_i32,
                          PseudoMaskedAtomicLoadAdd32>;
-def PseudoMaskedAtomicLoadSub32 : PseudoMaskedAMO;
+def PseudoMaskedAtomicLoadSub32 : PseudoMaskedAMO, Sched<[]>;
 def : PseudoMaskedAMOPat<int_riscv_masked_atomicrmw_sub_i32,
                          PseudoMaskedAtomicLoadSub32>;
-def PseudoMaskedAtomicLoadNand32 : PseudoMaskedAMO;
+def PseudoMaskedAtomicLoadNand32 : PseudoMaskedAMO, Sched<[]>;
 def : PseudoMaskedAMOPat<int_riscv_masked_atomicrmw_nand_i32,
                          PseudoMaskedAtomicLoadNand32>;
-def PseudoMaskedAtomicLoadMax32 : PseudoMaskedAMOMinMax;
+def PseudoMaskedAtomicLoadMax32 : PseudoMaskedAMOMinMax, Sched<[]>;
 def : PseudoMaskedAMOMinMaxPat<int_riscv_masked_atomicrmw_max_i32,
                                PseudoMaskedAtomicLoadMax32>;
-def PseudoMaskedAtomicLoadMin32 : PseudoMaskedAMOMinMax;
+def PseudoMaskedAtomicLoadMin32 : PseudoMaskedAMOMinMax, Sched<[]>;
 def : PseudoMaskedAMOMinMaxPat<int_riscv_masked_atomicrmw_min_i32,
                                PseudoMaskedAtomicLoadMin32>;
-def PseudoMaskedAtomicLoadUMax32 : PseudoMaskedAMOUMinUMax;
+def PseudoMaskedAtomicLoadUMax32 : PseudoMaskedAMOUMinUMax, Sched<[]>;
 def : PseudoMaskedAMOPat<int_riscv_masked_atomicrmw_umax_i32,
                          PseudoMaskedAtomicLoadUMax32>;
-def PseudoMaskedAtomicLoadUMin32 : PseudoMaskedAMOUMinUMax;
+def PseudoMaskedAtomicLoadUMin32 : PseudoMaskedAMOUMinUMax, Sched<[]>;
 def : PseudoMaskedAMOPat<int_riscv_masked_atomicrmw_umin_i32,
                          PseudoMaskedAtomicLoadUMin32>;
 
 /// Compare and exchange
 
 class PseudoCmpXchg
     : Pseudo<(outs GPR:$res, GPR:$scratch),
              (ins GPR:$addr, GPR:$cmpval, GPR:$newval, ixlenimm:$ordering), []> {
multiclass PseudoCmpXchgPat<string Op, Pseudo CmpXchgInst> {
   def : Pat<(!cast<PatFrag>(Op#"_release") GPR:$addr, GPR:$cmp, GPR:$new),
             (CmpXchgInst GPR:$addr, GPR:$cmp, GPR:$new, 5)>;
   def : Pat<(!cast<PatFrag>(Op#"_acq_rel") GPR:$addr, GPR:$cmp, GPR:$new),
             (CmpXchgInst GPR:$addr, GPR:$cmp, GPR:$new, 6)>;
   def : Pat<(!cast<PatFrag>(Op#"_seq_cst") GPR:$addr, GPR:$cmp, GPR:$new),
             (CmpXchgInst GPR:$addr, GPR:$cmp, GPR:$new, 7)>;
 }
 
-def PseudoCmpXchg32 : PseudoCmpXchg;
+def PseudoCmpXchg32 : PseudoCmpXchg, Sched<[]>;
 defm : PseudoCmpXchgPat<"atomic_cmp_swap_32", PseudoCmpXchg32>;
 
 def PseudoMaskedCmpXchg32
     : Pseudo<(outs GPR:$res, GPR:$scratch),
              (ins GPR:$addr, GPR:$cmpval, GPR:$newval, GPR:$mask,
-              ixlenimm:$ordering), []> {
+              ixlenimm:$ordering), []>,
+        Sched<[]> {
   let Constraints = "@earlyclobber $res,@earlyclobber $scratch";
   let mayLoad = 1;
   let mayStore = 1;
   let hasSideEffects = 0;
 }
 
 def : Pat<(int_riscv_masked_cmpxchg_i32
             GPR:$addr, GPR:$cmpval, GPR:$newval, GPR:$mask, timm:$ordering),
def : Pat<(atomic_load_sub_64_release GPR:$addr, GPR:$incr),
           (AMOADD_D_RL GPR:$addr, (SUB X0, GPR:$incr))>;
 def : Pat<(atomic_load_sub_64_acq_rel GPR:$addr, GPR:$incr),
           (AMOADD_D_AQ_RL GPR:$addr, (SUB X0, GPR:$incr))>;
 def : Pat<(atomic_load_sub_64_seq_cst GPR:$addr, GPR:$incr),
           (AMOADD_D_AQ_RL GPR:$addr, (SUB X0, GPR:$incr))>;
 
 /// 64-bit pseudo AMOs
 
-def PseudoAtomicLoadNand64 : PseudoAMO;
+def PseudoAtomicLoadNand64 : PseudoAMO, Sched<[]>;
 // Ordering constants must be kept in sync with the AtomicOrdering enum in
 // AtomicOrdering.h.
 def : Pat<(atomic_load_nand_64_monotonic GPR:$addr, GPR:$incr),
           (PseudoAtomicLoadNand64 GPR:$addr, GPR:$incr, 2)>;
 def : Pat<(atomic_load_nand_64_acquire GPR:$addr, GPR:$incr),
           (PseudoAtomicLoadNand64 GPR:$addr, GPR:$incr, 4)>;
 def : Pat<(atomic_load_nand_64_release GPR:$addr, GPR:$incr),
           (PseudoAtomicLoadNand64 GPR:$addr, GPR:$incr, 5)>;
def : PseudoMaskedAMOMinMaxPat<int_riscv_masked_atomicrmw_min_i64,
                                PseudoMaskedAtomicLoadMin32>;
 def : PseudoMaskedAMOPat<int_riscv_masked_atomicrmw_umax_i64,
                          PseudoMaskedAtomicLoadUMax32>;
 def : PseudoMaskedAMOPat<int_riscv_masked_atomicrmw_umin_i64,
                          PseudoMaskedAtomicLoadUMin32>;
 
 /// 64-bit compare and exchange
 
-def PseudoCmpXchg64 : PseudoCmpXchg;
+def PseudoCmpXchg64 : PseudoCmpXchg, Sched<[]>;
 defm : PseudoCmpXchgPat<"atomic_cmp_swap_64", PseudoCmpXchg64>;
 
 def : Pat<(int_riscv_masked_cmpxchg_i64
             GPR:$addr, GPR:$cmpval, GPR:$newval, GPR:$mask, timm:$ordering),
           (PseudoMaskedCmpXchg32
             GPR:$addr, GPR:$cmpval, GPR:$newval, GPR:$mask, imm:$ordering)>;
 } // Predicates = [HasStdExtA, IsRV64]

llvm/lib/Target/RISCV/RISCVInstrInfoC.td

 // Instructions
 //===----------------------------------------------------------------------===//
 
 let Predicates = [HasStdExtC] in {
 
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0, Uses = [X2] in
 def C_ADDI4SPN : RVInst16CIW<0b000, 0b00, (outs GPRC:$rd),
                              (ins SP:$rs1, uimm10_lsb00nonzero:$imm),
-                             "c.addi4spn", "$rd, $rs1, $imm"> {
+                             "c.addi4spn", "$rd, $rs1, $imm">,
+           Sched<[WriteIALUC]> {
   bits<5> rs1;
   let Inst{12-11} = imm{5-4};
   let Inst{10-7} = imm{9-6};
   let Inst{6} = imm{2};
   let Inst{5} = imm{3};
 }
 
 let Predicates = [HasStdExtC, HasStdExtD] in
-def C_FLD  : CLoad_ri<0b001, "c.fld", FPR64C, uimm8_lsb000> {
+def C_FLD  : CLoad_ri<0b001, "c.fld", FPR64C, uimm8_lsb000>,
+           Sched<[WriteFLd64C]> {
   bits<8> imm;
   let Inst{12-10} = imm{5-3};
   let Inst{6-5} = imm{7-6};
 }
 
-def C_LW : CLoad_ri<0b010, "c.lw", GPRC, uimm7_lsb00> {
+def C_LW : CLoad_ri<0b010, "c.lw", GPRC, uimm7_lsb00>,
+         Sched<[WriteLDWC]> {
   bits<7> imm;
   let Inst{12-10} = imm{5-3};
   let Inst{6} = imm{2};
   let Inst{5} = imm{6};
 }
 
 let DecoderNamespace = "RISCV32Only_",
     Predicates = [HasStdExtC, HasStdExtF, IsRV32] in
-def C_FLW  : CLoad_ri<0b011, "c.flw", FPR32C, uimm7_lsb00> {
+def C_FLW  : CLoad_ri<0b011, "c.flw", FPR32C, uimm7_lsb00>,
+           Sched<[WriteFLd32C]> {
   bits<7> imm;
   let Inst{12-10} = imm{5-3};
   let Inst{6} = imm{2};
   let Inst{5} = imm{6};
 }
 
 let Predicates = [HasStdExtC, IsRV64] in
-def C_LD : CLoad_ri<0b011, "c.ld", GPRC, uimm8_lsb000> {
+def C_LD : CLoad_ri<0b011, "c.ld", GPRC, uimm8_lsb000>,
+         Sched<[WriteLDDWC]> {
   bits<8> imm;
   let Inst{12-10} = imm{5-3};
   let Inst{6-5} = imm{7-6};
 }
 
 let Predicates = [HasStdExtC, HasStdExtD] in
-def C_FSD  : CStore_rri<0b101, "c.fsd", FPR64C, uimm8_lsb000> {
+def C_FSD  : CStore_rri<0b101, "c.fsd", FPR64C, uimm8_lsb000>,
+           Sched<[WriteFSt64C]> {
   bits<8> imm;
   let Inst{12-10} = imm{5-3};
   let Inst{6-5} = imm{7-6};
 }
 
-def C_SW : CStore_rri<0b110, "c.sw", GPRC, uimm7_lsb00> {
+def C_SW : CStore_rri<0b110, "c.sw", GPRC, uimm7_lsb00>,
+         Sched<[WriteSTWC]> {
   bits<7> imm;
   let Inst{12-10} = imm{5-3};
   let Inst{6} = imm{2};
   let Inst{5} = imm{6};
 }
 
 let DecoderNamespace = "RISCV32Only_",
     Predicates = [HasStdExtC, HasStdExtF, IsRV32]  in
-def C_FSW  : CStore_rri<0b111, "c.fsw", FPR32C, uimm7_lsb00> {
+def C_FSW  : CStore_rri<0b111, "c.fsw", FPR32C, uimm7_lsb00>,
+           Sched<[WriteFSt32C]> {
   bits<7> imm;
   let Inst{12-10} = imm{5-3};
   let Inst{6} = imm{2};
   let Inst{5} = imm{6};
 }
 
 let Predicates = [HasStdExtC, IsRV64] in
-def C_SD : CStore_rri<0b111, "c.sd", GPRC, uimm8_lsb000> {
+def C_SD : CStore_rri<0b111, "c.sd", GPRC, uimm8_lsb000>,
+         Sched<[WriteSTDWC]> {
   bits<8> imm;
   let Inst{12-10} = imm{5-3};
   let Inst{6-5} = imm{7-6};
 }
 
 let rd = 0, imm = 0, hasSideEffects = 0, mayLoad = 0, mayStore = 0 in
-def C_NOP : RVInst16CI<0b000, 0b01, (outs), (ins), "c.nop", "">
+def C_NOP : RVInst16CI<0b000, 0b01, (outs), (ins), "c.nop", "">,
+          Sched<[WriteNop]>
 {
   let Inst{6-2} = 0;
 }
 
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0 in
 def C_ADDI : RVInst16CI<0b000, 0b01, (outs GPRNoX0:$rd_wb),
                         (ins GPRNoX0:$rd, simm6nonzero:$imm),
-                        "c.addi", "$rd, $imm"> {
+                        "c.addi", "$rd, $imm">,
+      Sched<[WriteIALUC]> {
   let Constraints = "$rd = $rd_wb";
   let Inst{6-2} = imm{4-0};
 }
 
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0 in
 def C_ADDI_NOP : RVInst16CI<0b000, 0b01, (outs GPRX0:$rd_wb),
                             (ins GPRX0:$rd, immzero:$imm),
-                            "c.addi", "$rd, $imm"> {
+                            "c.addi", "$rd, $imm">,
+          Sched<[WriteIALUC]> {
   let Constraints = "$rd = $rd_wb";
   let Inst{6-2} = 0;
   let isAsmParserOnly = 1;
 }
 
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0, isCall = 1,
     DecoderNamespace = "RISCV32Only_", Defs = [X1],
     Predicates = [HasStdExtC, IsRV32]  in
 def C_JAL : RVInst16CJ<0b001, 0b01, (outs), (ins simm12_lsb0:$offset),
-                       "c.jal", "$offset">;
+                       "c.jal", "$offset">, Sched<[WriteJalC]>;
 
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0,
     Predicates = [HasStdExtC, IsRV64] in
 def C_ADDIW : RVInst16CI<0b001, 0b01, (outs GPRNoX0:$rd_wb),
                          (ins GPRNoX0:$rd, simm6:$imm),
-                         "c.addiw", "$rd, $imm"> {
+                         "c.addiw", "$rd, $imm">,
+       Sched<[WriteIALU32C]> {
   let Constraints = "$rd = $rd_wb";
   let Inst{6-2} = imm{4-0};
 }
 
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0 in
 def C_LI : RVInst16CI<0b010, 0b01, (outs GPRNoX0:$rd), (ins simm6:$imm),
-                      "c.li", "$rd, $imm"> {
+                      "c.li", "$rd, $imm">,
+          Sched<[WriteIALUC]> {
   let Inst{6-2} = imm{4-0};
 }
 
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0 in
 def C_ADDI16SP : RVInst16CI<0b011, 0b01, (outs SP:$rd_wb),
                             (ins SP:$rd, simm10_lsb0000nonzero:$imm),
-                            "c.addi16sp", "$rd, $imm"> {
+                            "c.addi16sp", "$rd, $imm">,
+          Sched<[WriteIALUC]> {
   let Constraints = "$rd = $rd_wb";
   let Inst{12} = imm{9};
   let Inst{11-7} = 2;
   let Inst{6} = imm{4};
   let Inst{5} = imm{6};
   let Inst{4-3} = imm{8-7};
   let Inst{2} = imm{5};
 }
 
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0 in
 def C_LUI : RVInst16CI<0b011, 0b01, (outs GPRNoX0X2:$rd),
                        (ins c_lui_imm:$imm),
-                       "c.lui", "$rd, $imm"> {
+                       "c.lui", "$rd, $imm">,
+           Sched<[WriteIALUC]> {
   let Inst{6-2} = imm{4-0};
 }
 
-def C_SRLI : Shift_right<0b00, "c.srli", GPRC, uimmlog2xlennonzero>;
-def C_SRAI : Shift_right<0b01, "c.srai", GPRC, uimmlog2xlennonzero>;
+def C_SRLI : Shift_right<0b00, "c.srli", GPRC, uimmlog2xlennonzero>,
+           Sched<[WriteShiftC]>;
+def C_SRAI : Shift_right<0b01, "c.srai", GPRC, uimmlog2xlennonzero>,
+           Sched<[WriteShiftC]>;
 
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0 in
 def C_ANDI : RVInst16CB<0b100, 0b01, (outs GPRC:$rs1_wb), (ins GPRC:$rs1, simm6:$imm),
-                        "c.andi", "$rs1, $imm"> {
+                        "c.andi", "$rs1, $imm">,
+      Sched<[WriteIALUC]> {
   let Constraints = "$rs1 = $rs1_wb";
   let Inst{12} = imm{5};
   let Inst{11-10} = 0b10;
   let Inst{6-2} = imm{4-0};
 }
 
-def C_SUB  : CS_ALU<0b100011, 0b00, "c.sub", GPRC>;
-def C_XOR  : CS_ALU<0b100011, 0b01, "c.xor", GPRC>;
-def C_OR   : CS_ALU<0b100011, 0b10, "c.or" , GPRC>;
-def C_AND  : CS_ALU<0b100011, 0b11, "c.and", GPRC>;
+def C_SUB  : CS_ALU<0b100011, 0b00, "c.sub", GPRC>, Sched<[WriteIALUC]>;
+def C_XOR  : CS_ALU<0b100011, 0b01, "c.xor", GPRC>, Sched<[WriteIALUC]>;
+def C_OR   : CS_ALU<0b100011, 0b10, "c.or" , GPRC>, Sched<[WriteIALUC]>;
+def C_AND  : CS_ALU<0b100011, 0b11, "c.and", GPRC>, Sched<[WriteIALUC]>;
 
 let Predicates = [HasStdExtC, IsRV64] in {
-def C_SUBW : CS_ALU<0b100111, 0b00, "c.subw", GPRC>;
-def C_ADDW : CS_ALU<0b100111, 0b01, "c.addw", GPRC>;
+def C_SUBW : CS_ALU<0b100111, 0b00, "c.subw", GPRC>, Sched<[WriteIALU32C]>;
+def C_ADDW : CS_ALU<0b100111, 0b01, "c.addw", GPRC>, Sched<[WriteIALU32C]>;
 }
 
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0 in
 def C_J : RVInst16CJ<0b101, 0b01, (outs), (ins simm12_lsb0:$offset),
-                     "c.j", "$offset"> {
+                     "c.j", "$offset">, Sched<[WriteJmpC]> {
   let isBranch = 1;
   let isTerminator=1;
   let isBarrier=1;
 }
 
-def C_BEQZ : Bcz<0b110, "c.beqz",  seteq, GPRC>;
-def C_BNEZ : Bcz<0b111, "c.bnez",  setne, GPRC>;
+def C_BEQZ : Bcz<0b110, "c.beqz",  seteq, GPRC>, Sched<[WriteJmpC]>;
+def C_BNEZ : Bcz<0b111, "c.bnez",  setne, GPRC>, Sched<[WriteJmpC]>;
 
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0 in
 def C_SLLI : RVInst16CI<0b000, 0b10, (outs GPRNoX0:$rd_wb),
                         (ins GPRNoX0:$rd, uimmlog2xlennonzero:$imm),
-                        "c.slli" ,"$rd, $imm"> {
+                        "c.slli" ,"$rd, $imm">,
+      Sched<[WriteShiftC]> {
   let Constraints = "$rd = $rd_wb";
   let Inst{6-2} = imm{4-0};
 }
 
 let Predicates = [HasStdExtC, HasStdExtD] in
-def C_FLDSP  : CStackLoad<0b001, "c.fldsp", FPR64, uimm9_lsb000> {
+def C_FLDSP  : CStackLoad<0b001, "c.fldsp", FPR64, uimm9_lsb000>,
+             Sched<[WriteFLd64C]> {
   let Inst{6-5} = imm{4-3};
   let Inst{4-2} = imm{8-6};
 }
 
-def C_LWSP : CStackLoad<0b010, "c.lwsp", GPRNoX0, uimm8_lsb00> {
+def C_LWSP : CStackLoad<0b010, "c.lwsp", GPRNoX0, uimm8_lsb00>,
+           Sched<[WriteLDWC]> {
   let Inst{6-4} = imm{4-2};
   let Inst{3-2} = imm{7-6};
 }
 
 let DecoderNamespace = "RISCV32Only_",
     Predicates = [HasStdExtC, HasStdExtF, IsRV32] in
-def C_FLWSP  : CStackLoad<0b011, "c.flwsp", FPR32, uimm8_lsb00> {
+def C_FLWSP  : CStackLoad<0b011, "c.flwsp", FPR32, uimm8_lsb00>,
+             Sched<[WriteFLd32C]> {
   let Inst{6-4} = imm{4-2};
   let Inst{3-2} = imm{7-6};
 }
 
 let Predicates = [HasStdExtC, IsRV64] in
-def C_LDSP : CStackLoad<0b011, "c.ldsp", GPRNoX0, uimm9_lsb000> {
+def C_LDSP : CStackLoad<0b011, "c.ldsp", GPRNoX0, uimm9_lsb000>,
+           Sched<[WriteLDDWC]> {
   let Inst{6-5} = imm{4-3};
   let Inst{4-2} = imm{8-6};
 }
 
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0 in
 def C_JR : RVInst16CR<0b1000, 0b10, (outs), (ins GPRNoX0:$rs1),
-                      "c.jr", "$rs1"> {
+                      "c.jr", "$rs1">, Sched<[WriteJmpRegC]> {
   let isBranch = 1;
   let isBarrier = 1;
   let isTerminator = 1;
   let isIndirectBranch = 1;
   let rs2 = 0;
 }
 
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0 in
 def C_MV : RVInst16CR<0b1000, 0b10, (outs GPRNoX0:$rs1), (ins GPRNoX0:$rs2),
-                      "c.mv", "$rs1, $rs2">;
+                      "c.mv", "$rs1, $rs2">,
+          Sched<[WriteIALUC]>;
 
 let rs1 = 0, rs2 = 0, hasSideEffects = 1, mayLoad = 0, mayStore = 0 in
-def C_EBREAK : RVInst16CR<0b1001, 0b10, (outs), (ins), "c.ebreak", "">;
+def C_EBREAK : RVInst16CR<0b1001, 0b10, (outs), (ins), "c.ebreak", "">, Sched<[]>;
 
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0,
     isCall=1, Defs=[X1], rs2 = 0 in
 def C_JALR : RVInst16CR<0b1001, 0b10, (outs), (ins GPRNoX0:$rs1),
-                        "c.jalr", "$rs1">;
+                        "c.jalr", "$rs1">, Sched<[WriteJalrC]>;
 
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0 in
 def C_ADD : RVInst16CR<0b1001, 0b10, (outs GPRNoX0:$rs1_wb),
                        (ins GPRNoX0:$rs1, GPRNoX0:$rs2),
-                       "c.add", "$rs1, $rs2"> {
+                       "c.add", "$rs1, $rs2">,
+           Sched<[WriteIALUC]> {
   let Constraints = "$rs1 = $rs1_wb";
 }
 
 let Predicates = [HasStdExtC, HasStdExtD] in
-def C_FSDSP  : CStackStore<0b101, "c.fsdsp", FPR64, uimm9_lsb000> {
+def C_FSDSP  : CStackStore<0b101, "c.fsdsp", FPR64, uimm9_lsb000>,
+             Sched<[WriteFSt64C]> {
   let Inst{12-10} = imm{5-3};
   let Inst{9-7}   = imm{8-6};
 }
 
-def C_SWSP : CStackStore<0b110, "c.swsp", GPR, uimm8_lsb00> {
+def C_SWSP : CStackStore<0b110, "c.swsp", GPR, uimm8_lsb00>,
+           Sched<[WriteSTWC]> {
   let Inst{12-9} = imm{5-2};
   let Inst{8-7}  = imm{7-6};
 }
 
 let DecoderNamespace = "RISCV32Only_",
     Predicates = [HasStdExtC, HasStdExtF, IsRV32] in
-def C_FSWSP  : CStackStore<0b111, "c.fswsp", FPR32, uimm8_lsb00> {
+def C_FSWSP  : CStackStore<0b111, "c.fswsp", FPR32, uimm8_lsb00>,
+             Sched<[WriteFSt32C]> {
   let Inst{12-9} = imm{5-2};
   let Inst{8-7}  = imm{7-6};
+  
 }
 
 let Predicates = [HasStdExtC, IsRV64] in
-def C_SDSP : CStackStore<0b111, "c.sdsp", GPR, uimm9_lsb000> {
+def C_SDSP : CStackStore<0b111, "c.sdsp", GPR, uimm9_lsb000>,
+           Sched<[WriteSTDWC]> {
   let Inst{12-10} = imm{5-3};
   let Inst{9-7}   = imm{8-6};
 }
 
 // The all zeros pattern isn't a valid RISC-V instruction. It's used by GNU
 // binutils as 16-bit instruction known to be unimplemented (i.e., trapping).
 let hasSideEffects = 1, mayLoad = 0, mayStore = 0 in
-def C_UNIMP : RVInst16<(outs), (ins), "c.unimp", "", [], InstFormatOther> {
+def C_UNIMP : RVInst16<(outs), (ins), "c.unimp", "", [], InstFormatOther>,
+            Sched<[]> {
   let Inst{15-0} = 0;
 }
 
 } // Predicates = [HasStdExtC]
 
 //===----------------------------------------------------------------------===//
 // HINT Instructions
 //===----------------------------------------------------------------------===//
 
 let Predicates = [HasStdExtC, HasRVCHints], hasSideEffects = 0, mayLoad = 0,
     mayStore = 0 in
 {
 
 let rd = 0 in
 def C_NOP_HINT : RVInst16CI<0b000, 0b01, (outs), (ins simm6nonzero:$imm),
-                            "c.nop", "$imm"> {
+                            "c.nop", "$imm">, Sched<[WriteNop]> {
   let Inst{6-2} = imm{4-0};
   let DecoderMethod = "decodeRVCInstrSImm";
 }
 
 // Just a different syntax for the c.nop hint: c.addi x0, simm6 vs c.nop simm6.
 def C_ADDI_HINT_X0 : RVInst16CI<0b000, 0b01, (outs GPRX0:$rd_wb),
                                 (ins GPRX0:$rd, simm6nonzero:$imm),
-                                "c.addi", "$rd, $imm"> {
+                                "c.addi", "$rd, $imm">,
+        Sched<[WriteIALUC]> {
   let Constraints = "$rd = $rd_wb";
   let Inst{6-2} = imm{4-0};
   let isAsmParserOnly = 1;
 }
 
 def C_ADDI_HINT_IMM_ZERO : RVInst16CI<0b000, 0b01, (outs GPRNoX0:$rd_wb),
                                       (ins GPRNoX0:$rd, immzero:$imm),
-                                      "c.addi", "$rd, $imm"> {
+                                      "c.addi", "$rd, $imm">,
+              Sched<[WriteIALUC]> {
   let Constraints = "$rd = $rd_wb";
   let Inst{6-2} = 0;
   let isAsmParserOnly = 1;
 }
 
 def C_LI_HINT : RVInst16CI<0b010, 0b01, (outs GPRX0:$rd), (ins simm6:$imm),
-                           "c.li", "$rd, $imm"> {
+                           "c.li", "$rd, $imm">,
+         Sched<[WriteIALUC]> {
   let Inst{6-2} = imm{4-0};
   let Inst{11-7} = 0;
   let DecoderMethod = "decodeRVCInstrRdSImm";
 }
 
 def C_LUI_HINT : RVInst16CI<0b011, 0b01, (outs GPRX0:$rd),
                             (ins c_lui_imm:$imm),
-                            "c.lui", "$rd, $imm"> {
+                            "c.lui", "$rd, $imm">,
+          Sched<[WriteIALUC]> {
   let Inst{6-2} = imm{4-0};
   let Inst{11-7} = 0;
   let DecoderMethod = "decodeRVCInstrRdSImm";
 }
 
 def C_MV_HINT : RVInst16CR<0b1000, 0b10, (outs GPRX0:$rs1), (ins GPRNoX0:$rs2),
-                           "c.mv", "$rs1, $rs2">
+                           "c.mv", "$rs1, $rs2">, Sched<[WriteIALUC]>
 {
   let Inst{11-7} = 0;
   let DecoderMethod = "decodeRVCInstrRdRs2";
 }
 
 def C_ADD_HINT : RVInst16CR<0b1001, 0b10, (outs GPRX0:$rs1_wb),
                             (ins GPRX0:$rs1, GPRNoX0:$rs2),
-                            "c.add", "$rs1, $rs2"> {
+                            "c.add", "$rs1, $rs2">,
+          Sched<[WriteIALUC]> {
   let Constraints = "$rs1 = $rs1_wb";
   let Inst{11-7} = 0;
   let DecoderMethod = "decodeRVCInstrRdRs1Rs2";
 }
 
 def C_SLLI_HINT : RVInst16CI<0b000, 0b10, (outs GPRX0:$rd_wb),
                              (ins GPRX0:$rd, uimmlog2xlennonzero:$imm),
-                             "c.slli" ,"$rd, $imm"> {
+                             "c.slli" ,"$rd, $imm">,
+           Sched<[WriteShiftC]> {
   let Constraints = "$rd = $rd_wb";
   let Inst{6-2} = imm{4-0};
   let Inst{11-7} = 0;
   let DecoderMethod = "decodeRVCInstrRdRs1UImm";
 }
 
 def C_SLLI64_HINT : RVInst16CI<0b000, 0b10, (outs GPR:$rd_wb), (ins GPR:$rd),
-                               "c.slli64" ,"$rd"> {
+                               "c.slli64" ,"$rd">,
+             Sched<[WriteShiftC]> {
   let Constraints = "$rd = $rd_wb";
   let Inst{6-2} = 0;
   let Inst{12} = 0;
 }
 
 def C_SRLI64_HINT : RVInst16CI<0b100, 0b01, (outs GPRC:$rd_wb),
                                (ins GPRC:$rd),
-                               "c.srli64", "$rd"> {
+                               "c.srli64", "$rd">,
+             Sched<[WriteShiftC]> {
   let Constraints = "$rd = $rd_wb";
   let Inst{6-2} = 0;
   let Inst{11-10} = 0;
   let Inst{12} = 0;
 }
 
 def C_SRAI64_HINT : RVInst16CI<0b100, 0b01, (outs GPRC:$rd_wb),
                                (ins GPRC:$rd),
-                               "c.srai64", "$rd"> {
+                               "c.srai64", "$rd">,
+             Sched<[WriteShiftC]> {
   let Constraints = "$rd = $rd_wb";
   let Inst{6-2} = 0;
   let Inst{11-10} = 1;
   let Inst{12} = 0;
 }
 
 } // Predicates = [HasStdExtC, HasRVCHints], hasSideEffects = 0, mayLoad = 0,
   // mayStore = 0

llvm/lib/Target/RISCV/RISCVInstrInfoD.td

 // Instructions
 //===----------------------------------------------------------------------===//
 
 let Predicates = [HasStdExtD] in {
 
 let hasSideEffects = 0, mayLoad = 1, mayStore = 0 in
 def FLD : RVInstI<0b011, OPC_LOAD_FP, (outs FPR64:$rd),
                   (ins GPR:$rs1, simm12:$imm12),
-                  "fld", "$rd, ${imm12}(${rs1})">;
+                  "fld", "$rd, ${imm12}(${rs1})">, Sched<[WriteFLd64]>;
 
 // Operands for stores are in the order srcreg, base, offset rather than
 // reflecting the order these fields are specified in the instruction
 // encoding.
 let hasSideEffects = 0, mayLoad = 0, mayStore = 1 in
 def FSD : RVInstS<0b011, OPC_STORE_FP, (outs),
                   (ins FPR64:$rs2, GPR:$rs1, simm12:$imm12),
-                   "fsd", "$rs2, ${imm12}(${rs1})">;
+                   "fsd", "$rs2, ${imm12}(${rs1})">, Sched<[WriteFSt64]>;
 
-def FMADD_D  : FPFMAD_rrr_frm<OPC_MADD, "fmadd.d">;
+def FMADD_D  : FPFMAD_rrr_frm<OPC_MADD, "fmadd.d">, Sched<[WriteFMulAdd64]>;
 def          : FPFMADDynFrmAlias<FMADD_D, "fmadd.d">;
-def FMSUB_D  : FPFMAD_rrr_frm<OPC_MSUB, "fmsub.d">;
+def FMSUB_D  : FPFMAD_rrr_frm<OPC_MSUB, "fmsub.d">, Sched<[WriteFMulSub64]>;
 def          : FPFMADDynFrmAlias<FMSUB_D, "fmsub.d">;
-def FNMSUB_D : FPFMAD_rrr_frm<OPC_NMSUB, "fnmsub.d">;
+def FNMSUB_D : FPFMAD_rrr_frm<OPC_NMSUB, "fnmsub.d">, Sched<[WriteFMulSub64]>;
 def          : FPFMADDynFrmAlias<FNMSUB_D, "fnmsub.d">;
-def FNMADD_D : FPFMAD_rrr_frm<OPC_NMADD, "fnmadd.d">;
+def FNMADD_D : FPFMAD_rrr_frm<OPC_NMADD, "fnmadd.d">, Sched<[WriteFMulAdd64]>;
 def          : FPFMADDynFrmAlias<FNMADD_D, "fnmadd.d">;
 
-def FADD_D : FPALUD_rr_frm<0b0000001, "fadd.d">;
+def FADD_D : FPALUD_rr_frm<0b0000001, "fadd.d">, Sched<[WriteFALU64]>;
 def        : FPALUDDynFrmAlias<FADD_D, "fadd.d">;
-def FSUB_D : FPALUD_rr_frm<0b0000101, "fsub.d">;
+def FSUB_D : FPALUD_rr_frm<0b0000101, "fsub.d">, Sched<[WriteFALU64]>;
 def        : FPALUDDynFrmAlias<FSUB_D, "fsub.d">;
-def FMUL_D : FPALUD_rr_frm<0b0001001, "fmul.d">;
+def FMUL_D : FPALUD_rr_frm<0b0001001, "fmul.d">, Sched<[WriteFALU64]>;
 def        : FPALUDDynFrmAlias<FMUL_D, "fmul.d">;
-def FDIV_D : FPALUD_rr_frm<0b0001101, "fdiv.d">;
+def FDIV_D : FPALUD_rr_frm<0b0001101, "fdiv.d">, Sched<[WriteFALU64]>;
 def        : FPALUDDynFrmAlias<FDIV_D, "fdiv.d">;
 
-def FSQRT_D : FPUnaryOp_r_frm<0b0101101, FPR64, FPR64, "fsqrt.d"> {
+def FSQRT_D : FPUnaryOp_r_frm<0b0101101, FPR64, FPR64, "fsqrt.d">,
+            Sched<[WriteFSqrt32]> {
   let rs2 = 0b00000;
 }
 def         : FPUnaryOpDynFrmAlias<FSQRT_D, "fsqrt.d", FPR64, FPR64>;
 
-def FSGNJ_D  : FPALUD_rr<0b0010001, 0b000, "fsgnj.d">;
-def FSGNJN_D : FPALUD_rr<0b0010001, 0b001, "fsgnjn.d">;
-def FSGNJX_D : FPALUD_rr<0b0010001, 0b010, "fsgnjx.d">;
-def FMIN_D   : FPALUD_rr<0b0010101, 0b000, "fmin.d">;
-def FMAX_D   : FPALUD_rr<0b0010101, 0b001, "fmax.d">;
+def FSGNJ_D  : FPALUD_rr<0b0010001, 0b000, "fsgnj.d">, Sched<[WriteFALU64]>;
+def FSGNJN_D : FPALUD_rr<0b0010001, 0b001, "fsgnjn.d">, Sched<[WriteFALU64]>;
+def FSGNJX_D : FPALUD_rr<0b0010001, 0b010, "fsgnjx.d">, Sched<[WriteFALU64]>;
+def FMIN_D   : FPALUD_rr<0b0010101, 0b000, "fmin.d">, Sched<[WriteFALU64]>;
+def FMAX_D   : FPALUD_rr<0b0010101, 0b001, "fmax.d">, Sched<[WriteFALU64]>;
 
-def FCVT_S_D : FPUnaryOp_r_frm<0b0100000, FPR32, FPR64, "fcvt.s.d"> {
+def FCVT_S_D : FPUnaryOp_r_frm<0b0100000, FPR32, FPR64, "fcvt.s.d">,
+             Sched<[WriteFCvtF64ToF32]> {
   let rs2 = 0b00001;
 }
 def          : FPUnaryOpDynFrmAlias<FCVT_S_D, "fcvt.s.d", FPR32, FPR64>;
 
-def FCVT_D_S : FPUnaryOp_r<0b0100001, 0b000, FPR64, FPR32, "fcvt.d.s"> {
+def FCVT_D_S : FPUnaryOp_r<0b0100001, 0b000, FPR64, FPR32, "fcvt.d.s">,
+             Sched<[WriteFCvtF32ToF64]> {
   let rs2 = 0b00000;
 }
 
-def FEQ_D : FPCmpD_rr<0b010, "feq.d">;
-def FLT_D : FPCmpD_rr<0b001, "flt.d">;
-def FLE_D : FPCmpD_rr<0b000, "fle.d">;
+def FEQ_D : FPCmpD_rr<0b010, "feq.d">, Sched<[WriteFCmp64]>;
+def FLT_D : FPCmpD_rr<0b001, "flt.d">, Sched<[WriteFCmp64]>;
+def FLE_D : FPCmpD_rr<0b000, "fle.d">, Sched<[WriteFCmp64]>;
 
-def FCLASS_D : FPUnaryOp_r<0b1110001, 0b001, GPR, FPR64, "fclass.d"> {
+def FCLASS_D : FPUnaryOp_r<0b1110001, 0b001, GPR, FPR64, "fclass.d">,
+             Sched<[WriteFClass64]> {
   let rs2 = 0b00000;
 }
 
-def FCVT_W_D : FPUnaryOp_r_frm<0b1100001, GPR, FPR64, "fcvt.w.d"> {
+def FCVT_W_D : FPUnaryOp_r_frm<0b1100001, GPR, FPR64, "fcvt.w.d">,
+             Sched<[WriteFCvtF64ToI32]> {
   let rs2 = 0b00000;
 }
 def          : FPUnaryOpDynFrmAlias<FCVT_W_D, "fcvt.w.d", GPR, FPR64>;
 
-def FCVT_WU_D : FPUnaryOp_r_frm<0b1100001, GPR, FPR64, "fcvt.wu.d"> {
+def FCVT_WU_D : FPUnaryOp_r_frm<0b1100001, GPR, FPR64, "fcvt.wu.d">,
+              Sched<[WriteFCvtF64ToI32]> {
   let rs2 = 0b00001;
 }
 def           : FPUnaryOpDynFrmAlias<FCVT_WU_D, "fcvt.wu.d", GPR, FPR64>;
 
-def FCVT_D_W : FPUnaryOp_r<0b1101001, 0b000, FPR64, GPR, "fcvt.d.w"> {
+def FCVT_D_W : FPUnaryOp_r<0b1101001, 0b000, FPR64, GPR, "fcvt.d.w">,
+             Sched<[WriteFCvtI32ToF64]> {
   let rs2 = 0b00000;
 }
 
-def FCVT_D_WU : FPUnaryOp_r<0b1101001, 0b000, FPR64, GPR, "fcvt.d.wu"> {
+def FCVT_D_WU : FPUnaryOp_r<0b1101001, 0b000, FPR64, GPR, "fcvt.d.wu">,
+              Sched<[WriteFCvtI32ToF64]> {
   let rs2 = 0b00001;
 }
 } // Predicates = [HasStdExtD]
 
 let Predicates = [HasStdExtD, IsRV64] in {
-def FCVT_L_D : FPUnaryOp_r_frm<0b1100001, GPR, FPR64, "fcvt.l.d"> {
+def FCVT_L_D : FPUnaryOp_r_frm<0b1100001, GPR, FPR64, "fcvt.l.d">,
+             Sched<[WriteFCvtF64ToI64]> {
   let rs2 = 0b00010;
 }
 def          : FPUnaryOpDynFrmAlias<FCVT_L_D, "fcvt.l.d", GPR, FPR64>;
 
-def FCVT_LU_D : FPUnaryOp_r_frm<0b1100001, GPR, FPR64, "fcvt.lu.d"> {
+def FCVT_LU_D : FPUnaryOp_r_frm<0b1100001, GPR, FPR64, "fcvt.lu.d">,
+              Sched<[WriteFCvtF64ToI64]> {
   let rs2 = 0b00011;
 }
 def           : FPUnaryOpDynFrmAlias<FCVT_LU_D, "fcvt.lu.d", GPR, FPR64>;
 
-def FMV_X_D : FPUnaryOp_r<0b1110001, 0b000, GPR, FPR64, "fmv.x.d"> {
+def FMV_X_D : FPUnaryOp_r<0b1110001, 0b000, GPR, FPR64, "fmv.x.d">,
+            Sched<[WriteFMovF64ToI64]> {
   let rs2 = 0b00000;
 }
 
-def FCVT_D_L : FPUnaryOp_r_frm<0b1101001, FPR64, GPR, "fcvt.d.l"> {
+def FCVT_D_L : FPUnaryOp_r_frm<0b1101001, FPR64, GPR, "fcvt.d.l">,
+             Sched<[WriteFCvtI64ToF64]> {
   let rs2 = 0b00010;
 }
 def          : FPUnaryOpDynFrmAlias<FCVT_D_L, "fcvt.d.l", FPR64, GPR>;
 
-def FCVT_D_LU : FPUnaryOp_r_frm<0b1101001, FPR64, GPR, "fcvt.d.lu"> {
+def FCVT_D_LU : FPUnaryOp_r_frm<0b1101001, FPR64, GPR, "fcvt.d.lu">,
+              Sched<[WriteFCvtI64ToF64]> {
   let rs2 = 0b00011;
 }
 def           : FPUnaryOpDynFrmAlias<FCVT_D_LU, "fcvt.d.lu", FPR64, GPR>;
 
-def FMV_D_X : FPUnaryOp_r<0b1111001, 0b000, FPR64, GPR, "fmv.d.x"> {
+def FMV_D_X : FPUnaryOp_r<0b1111001, 0b000, FPR64, GPR, "fmv.d.x">,
+            Sched<[WriteFMovI64ToF64]> {
   let rs2 = 0b00000;
 }
 } // Predicates = [HasStdExtD, IsRV64]
 
 //===----------------------------------------------------------------------===//
 // Assembler Pseudo Instructions (User-Level ISA, Version 2.2, Chapter 20)
 //===----------------------------------------------------------------------===//
 
 let Predicates = [HasStdExtD] in {
 def : InstAlias<"fld $rd, (${rs1})",  (FLD FPR64:$rd,  GPR:$rs1, 0), 0>;
 def : InstAlias<"fsd $rs2, (${rs1})", (FSD FPR64:$rs2, GPR:$rs1, 0), 0>;
 
 def : InstAlias<"fmv.d $rd, $rs",  (FSGNJ_D  FPR64:$rd, FPR64:$rs, FPR64:$rs)>;
 def : InstAlias<"fabs.d $rd, $rs", (FSGNJX_D FPR64:$rd, FPR64:$rs, FPR64:$rs)>;
 def : InstAlias<"fneg.d $rd, $rs", (FSGNJN_D FPR64:$rd, FPR64:$rs, FPR64:$rs)>;
 
 // fgt.d/fge.d are recognised by the GNU assembler but the canonical
 // flt.d/fle.d forms will always be printed. Therefore, set a zero weight.
 def : InstAlias<"fgt.d $rd, $rs, $rt",
                 (FLT_D GPR:$rd, FPR64:$rt, FPR64:$rs), 0>;
 def : InstAlias<"fge.d $rd, $rs, $rt",
                 (FLE_D GPR:$rd, FPR64:$rt, FPR64:$rs), 0>;
 
-def PseudoFLD  : PseudoFloatLoad<"fld", FPR64>;
-def PseudoFSD  : PseudoStore<"fsd", FPR64>;
+def PseudoFLD  : PseudoFloatLoad<"fld", FPR64>, Sched<[]>;
+def PseudoFSD  : PseudoStore<"fsd", FPR64>, Sched<[]>;
 } // Predicates = [HasStdExtD]
 
 //===----------------------------------------------------------------------===//
 // Pseudo-instructions and codegen patterns
 //===----------------------------------------------------------------------===//
 
 class PatFpr64Fpr64<SDPatternOperator OpNode, RVInstR Inst>
     : Pat<(OpNode FPR64:$rs1, FPR64:$rs2), (Inst $rs1, $rs2)>;

 defm : StPat<store, FSD, FPR64>;
 
 /// Pseudo-instructions needed for the soft-float ABI with RV32D
 
 // Moves two GPRs to an FPR.
 let usesCustomInserter = 1 in
 def BuildPairF64Pseudo
     : Pseudo<(outs FPR64:$dst), (ins GPR:$src1, GPR:$src2),
-             [(set FPR64:$dst, (RISCVBuildPairF64 GPR:$src1, GPR:$src2))]>;
+             [(set FPR64:$dst, (RISCVBuildPairF64 GPR:$src1, GPR:$src2))]>,
+       Sched<[]>;
 
 // Moves an FPR to two GPRs.
 let usesCustomInserter = 1 in
 def SplitF64Pseudo
     : Pseudo<(outs GPR:$dst1, GPR:$dst2), (ins FPR64:$src),
-             [(set GPR:$dst1, GPR:$dst2, (RISCVSplitF64 FPR64:$src))]>;
+             [(set GPR:$dst1, GPR:$dst2, (RISCVSplitF64 FPR64:$src))]>,
+       Sched<[]>;
 
 } // Predicates = [HasStdExtD]
 
 let Predicates = [HasStdExtD, IsRV32] in {
 // double->[u]int. Round-to-zero must be used.
 def : Pat<(fp_to_sint FPR64:$rs1), (FCVT_W_D FPR64:$rs1, 0b001)>;
 def : Pat<(fp_to_uint FPR64:$rs1), (FCVT_WU_D FPR64:$rs1, 0b001)>;
 

llvm/lib/Target/RISCV/RISCVInstrInfoF.td

 //===----------------------------------------------------------------------===//
 // Instructions
 //===----------------------------------------------------------------------===//
 
 let Predicates = [HasStdExtF] in {
 let hasSideEffects = 0, mayLoad = 1, mayStore = 0 in
 def FLW : RVInstI<0b010, OPC_LOAD_FP, (outs FPR32:$rd),
                   (ins GPR:$rs1, simm12:$imm12),
-                   "flw", "$rd, ${imm12}(${rs1})">;
+                   "flw", "$rd, ${imm12}(${rs1})">, Sched<[WriteFLd32]>;
 
 // Operands for stores are in the order srcreg, base, offset rather than
 // reflecting the order these fields are specified in the instruction
 // encoding.
 let hasSideEffects = 0, mayLoad = 0, mayStore = 1 in
 def FSW : RVInstS<0b010, OPC_STORE_FP, (outs),
                   (ins FPR32:$rs2, GPR:$rs1, simm12:$imm12),
-                   "fsw", "$rs2, ${imm12}(${rs1})">;
+                   "fsw", "$rs2, ${imm12}(${rs1})">, Sched<[WriteFSt32]>;
 
-def FMADD_S  : FPFMAS_rrr_frm<OPC_MADD, "fmadd.s">;
+def FMADD_S  : FPFMAS_rrr_frm<OPC_MADD, "fmadd.s">, Sched<[WriteFMulAdd32]>;
 def          : FPFMASDynFrmAlias<FMADD_S, "fmadd.s">;
-def FMSUB_S  : FPFMAS_rrr_frm<OPC_MSUB, "fmsub.s">;
+def FMSUB_S  : FPFMAS_rrr_frm<OPC_MSUB, "fmsub.s">, Sched<[WriteFMulSub32]>;
 def          : FPFMASDynFrmAlias<FMSUB_S, "fmsub.s">;
-def FNMSUB_S : FPFMAS_rrr_frm<OPC_NMSUB, "fnmsub.s">;
+def FNMSUB_S : FPFMAS_rrr_frm<OPC_NMSUB, "fnmsub.s">, Sched<[WriteFMulSub32]>;
 def          : FPFMASDynFrmAlias<FNMSUB_S, "fnmsub.s">;
-def FNMADD_S : FPFMAS_rrr_frm<OPC_NMADD, "fnmadd.s">;
+def FNMADD_S : FPFMAS_rrr_frm<OPC_NMADD, "fnmadd.s">, Sched<[WriteFMulAdd32]>;
 def          : FPFMASDynFrmAlias<FNMADD_S, "fnmadd.s">;
 
-def FADD_S : FPALUS_rr_frm<0b0000000, "fadd.s">;
+def FADD_S : FPALUS_rr_frm<0b0000000, "fadd.s">, Sched<[WriteFALU32]>;
 def        : FPALUSDynFrmAlias<FADD_S, "fadd.s">;
-def FSUB_S : FPALUS_rr_frm<0b0000100, "fsub.s">;
+def FSUB_S : FPALUS_rr_frm<0b0000100, "fsub.s">, Sched<[WriteFALU32]>;
 def        : FPALUSDynFrmAlias<FSUB_S, "fsub.s">;
-def FMUL_S : FPALUS_rr_frm<0b0001000, "fmul.s">;
+def FMUL_S : FPALUS_rr_frm<0b0001000, "fmul.s">, Sched<[WriteFMul32]>;
 def        : FPALUSDynFrmAlias<FMUL_S, "fmul.s">;
-def FDIV_S : FPALUS_rr_frm<0b0001100, "fdiv.s">;
+def FDIV_S : FPALUS_rr_frm<0b0001100, "fdiv.s">, Sched<[WriteFDiv32]>;
 def        : FPALUSDynFrmAlias<FDIV_S, "fdiv.s">;
 
-def FSQRT_S : FPUnaryOp_r_frm<0b0101100, FPR32, FPR32, "fsqrt.s"> {
+def FSQRT_S : FPUnaryOp_r_frm<0b0101100, FPR32, FPR32, "fsqrt.s">,
+            Sched<[WriteFSqrt32]> {
   let rs2 = 0b00000;
 }
 def         : FPUnaryOpDynFrmAlias<FSQRT_S, "fsqrt.s", FPR32, FPR32>;
 
-def FSGNJ_S  : FPALUS_rr<0b0010000, 0b000, "fsgnj.s">;
-def FSGNJN_S : FPALUS_rr<0b0010000, 0b001, "fsgnjn.s">;
-def FSGNJX_S : FPALUS_rr<0b0010000, 0b010, "fsgnjx.s">;
-def FMIN_S   : FPALUS_rr<0b0010100, 0b000, "fmin.s">;
-def FMAX_S   : FPALUS_rr<0b0010100, 0b001, "fmax.s">;
+def FSGNJ_S  : FPALUS_rr<0b0010000, 0b000, "fsgnj.s">, Sched<[WriteFALU32]>;
+def FSGNJN_S : FPALUS_rr<0b0010000, 0b001, "fsgnjn.s">, Sched<[WriteFALU32]>;
+def FSGNJX_S : FPALUS_rr<0b0010000, 0b010, "fsgnjx.s">, Sched<[WriteFALU32]>;
+def FMIN_S   : FPALUS_rr<0b0010100, 0b000, "fmin.s">, Sched<[WriteFALU32]>;
+def FMAX_S   : FPALUS_rr<0b0010100, 0b001, "fmax.s">, Sched<[WriteFALU32]>;
 
-def FCVT_W_S : FPUnaryOp_r_frm<0b1100000, GPR, FPR32, "fcvt.w.s"> {
+def FCVT_W_S : FPUnaryOp_r_frm<0b1100000, GPR, FPR32, "fcvt.w.s">,
+             Sched<[WriteFCvtF32ToI32]> {
   let rs2 = 0b00000;
 }
 def          : FPUnaryOpDynFrmAlias<FCVT_W_S, "fcvt.w.s", GPR, FPR32>;
 
-def FCVT_WU_S : FPUnaryOp_r_frm<0b1100000, GPR, FPR32, "fcvt.wu.s"> {
+def FCVT_WU_S : FPUnaryOp_r_frm<0b1100000, GPR, FPR32, "fcvt.wu.s">,
+              Sched<[WriteFCvtF32ToI32]> {
   let rs2 = 0b00001;
 }
 def           : FPUnaryOpDynFrmAlias<FCVT_WU_S, "fcvt.wu.s", GPR, FPR32>;
 
-def FMV_X_W : FPUnaryOp_r<0b1110000, 0b000, GPR, FPR32, "fmv.x.w"> {
+def FMV_X_W : FPUnaryOp_r<0b1110000, 0b000, GPR, FPR32, "fmv.x.w">,
+            Sched<[WriteFMovF32ToI32]> {
   let rs2 = 0b00000;
 }
 
-def FEQ_S : FPCmpS_rr<0b010, "feq.s">;
-def FLT_S : FPCmpS_rr<0b001, "flt.s">;
-def FLE_S : FPCmpS_rr<0b000, "fle.s">;
+def FEQ_S : FPCmpS_rr<0b010, "feq.s">, Sched<[WriteFCmp32]>;
+def FLT_S : FPCmpS_rr<0b001, "flt.s">, Sched<[WriteFCmp32]>;
+def FLE_S : FPCmpS_rr<0b000, "fle.s">, Sched<[WriteFCmp32]>;
 
-def FCLASS_S : FPUnaryOp_r<0b1110000, 0b001, GPR, FPR32, "fclass.s"> {
+def FCLASS_S : FPUnaryOp_r<0b1110000, 0b001, GPR, FPR32, "fclass.s">,
+             Sched<[WriteFClass32]> {
   let rs2 = 0b00000;
 }
 
-def FCVT_S_W : FPUnaryOp_r_frm<0b1101000, FPR32, GPR, "fcvt.s.w"> {
+def FCVT_S_W : FPUnaryOp_r_frm<0b1101000, FPR32, GPR, "fcvt.s.w">,
+             Sched<[WriteFCvtI32ToF32]> {
   let rs2 = 0b00000;
 }
 def          : FPUnaryOpDynFrmAlias<FCVT_S_W, "fcvt.s.w", FPR32, GPR>;
 
-def FCVT_S_WU : FPUnaryOp_r_frm<0b1101000, FPR32, GPR, "fcvt.s.wu"> {
+def FCVT_S_WU : FPUnaryOp_r_frm<0b1101000, FPR32, GPR, "fcvt.s.wu">,
+              Sched<[WriteFCvtI32ToF32]> {
   let rs2 = 0b00001;
 }
 def           : FPUnaryOpDynFrmAlias<FCVT_S_WU, "fcvt.s.wu", FPR32, GPR>;
 
-def FMV_W_X : FPUnaryOp_r<0b1111000, 0b000, FPR32, GPR, "fmv.w.x"> {
+def FMV_W_X : FPUnaryOp_r<0b1111000, 0b000, FPR32, GPR, "fmv.w.x">,
+            Sched<[WriteFMovI32ToF32]> {
   let rs2 = 0b00000;
 }
 } // Predicates = [HasStdExtF]
 
 let Predicates = [HasStdExtF, IsRV64] in {
-def FCVT_L_S  : FPUnaryOp_r_frm<0b1100000, GPR, FPR32, "fcvt.l.s"> {
+def FCVT_L_S  : FPUnaryOp_r_frm<0b1100000, GPR, FPR32, "fcvt.l.s">,
+              Sched<[WriteFCvtF32ToI64]> {
   let rs2 = 0b00010;
 }
 def           : FPUnaryOpDynFrmAlias<FCVT_L_S, "fcvt.l.s", GPR, FPR32>;
 
-def FCVT_LU_S  : FPUnaryOp_r_frm<0b1100000, GPR, FPR32, "fcvt.lu.s"> {
+def FCVT_LU_S  : FPUnaryOp_r_frm<0b1100000, GPR, FPR32, "fcvt.lu.s">,
+               Sched<[WriteFCvtF32ToI64]> {
   let rs2 = 0b00011;
 }
 def            : FPUnaryOpDynFrmAlias<FCVT_LU_S, "fcvt.lu.s", GPR, FPR32>;
 
-def FCVT_S_L : FPUnaryOp_r_frm<0b1101000, FPR32, GPR, "fcvt.s.l"> {
+def FCVT_S_L : FPUnaryOp_r_frm<0b1101000, FPR32, GPR, "fcvt.s.l">,
+             Sched<[WriteFCvtI64ToF32]> {
   let rs2 = 0b00010;
 }
 def          : FPUnaryOpDynFrmAlias<FCVT_S_L, "fcvt.s.l", FPR32, GPR>;
 
-def FCVT_S_LU : FPUnaryOp_r_frm<0b1101000, FPR32, GPR, "fcvt.s.lu"> {
+def FCVT_S_LU : FPUnaryOp_r_frm<0b1101000, FPR32, GPR, "fcvt.s.lu">,
+              Sched<[WriteFCvtI64ToF32]> {
   let rs2 = 0b00011;
 }
 def           : FPUnaryOpDynFrmAlias<FCVT_S_LU, "fcvt.s.lu", FPR32, GPR>;
 } // Predicates = [HasStdExtF, IsRV64]
 
 //===----------------------------------------------------------------------===//
 // Assembler Pseudo Instructions (User-Level ISA, Version 2.2, Chapter 20)
 //===----------------------------------------------------------------------===//

 def : InstAlias<"fsflagsi $rd, $imm", (CSRRWI GPR:$rd, FFLAGS.Encoding, uimm5:$imm)>;
 def : InstAlias<"fsflagsi $imm",      (CSRRWI      X0, FFLAGS.Encoding, uimm5:$imm), 2>;
 
 // fmv.w.x and fmv.x.w were previously known as fmv.s.x and fmv.x.s. Both
 // spellings should be supported by standard tools.
 def : MnemonicAlias<"fmv.s.x", "fmv.w.x">;
 def : MnemonicAlias<"fmv.x.s", "fmv.x.w">;
 
-def PseudoFLW  : PseudoFloatLoad<"flw", FPR32>;
-def PseudoFSW  : PseudoStore<"fsw", FPR32>;
+def PseudoFLW  : PseudoFloatLoad<"flw", FPR32>, Sched<[]>;
+def PseudoFSW  : PseudoStore<"fsw", FPR32>, Sched<[]>;
 } // Predicates = [HasStdExtF]
 
 //===----------------------------------------------------------------------===//
 // Pseudo-instructions and codegen patterns
 //===----------------------------------------------------------------------===//
 
 /// Generic pattern classes
 class PatFpr32Fpr32<SDPatternOperator OpNode, RVInstR Inst>

llvm/lib/Target/RISCV/RISCVInstrInfoM.td

 def riscv_divw  : SDNode<"RISCVISD::DIVW",  SDTIntBinOp>;
 def riscv_divuw : SDNode<"RISCVISD::DIVUW", SDTIntBinOp>;
 def riscv_remuw : SDNode<"RISCVISD::REMUW", SDTIntBinOp>;
 
 //===----------------------------------------------------------------------===//
 // Instructions
 //===----------------------------------------------------------------------===//
 
-let Predicates = [HasStdExtM] in {
-def MUL     : ALU_rr<0b0000001, 0b000, "mul">;
-def MULH    : ALU_rr<0b0000001, 0b001, "mulh">;
-def MULHSU  : ALU_rr<0b0000001, 0b010, "mulhsu">;
-def MULHU   : ALU_rr<0b0000001, 0b011, "mulhu">;
-def DIV     : ALU_rr<0b0000001, 0b100, "div">;
-def DIVU    : ALU_rr<0b0000001, 0b101, "divu">;
-def REM     : ALU_rr<0b0000001, 0b110, "rem">;
-def REMU    : ALU_rr<0b0000001, 0b111, "remu">;
-} // Predicates = [HasStdExtM]
+def MUL     : ALU_rr<0b0000001, 0b000, "mul">, Sched<[WriteIMul]>;
+def MULH    : ALU_rr<0b0000001, 0b001, "mulh">, Sched<[WriteIMul]>;
+def MULHSU  : ALU_rr<0b0000001, 0b010, "mulhsu">, Sched<[WriteIMul]>;
+def MULHU   : ALU_rr<0b0000001, 0b011, "mulhu">, Sched<[WriteIMul]>;
+def DIV     : ALU_rr<0b0000001, 0b100, "div">, Sched<[WriteIDiv]>;
+def DIVU    : ALU_rr<0b0000001, 0b101, "divu">, Sched<[WriteIDiv]>;
+def REM     : ALU_rr<0b0000001, 0b110, "rem">, Sched<[WriteIDiv]>;
+def REMU    : ALU_rr<0b0000001, 0b111, "remu">, Sched<[WriteIDiv]>;
 
 let Predicates = [HasStdExtM, IsRV64] in {
-def MULW    : ALUW_rr<0b0000001, 0b000, "mulw">;
-def DIVW    : ALUW_rr<0b0000001, 0b100, "divw">;
-def DIVUW   : ALUW_rr<0b0000001, 0b101, "divuw">;
-def REMW    : ALUW_rr<0b0000001, 0b110, "remw">;
-def REMUW   : ALUW_rr<0b0000001, 0b111, "remuw">;
+
+def MULW    : ALUW_rr<0b0000001, 0b000, "mulw">, Sched<[WriteIMul32]>;
+def DIVW    : ALUW_rr<0b0000001, 0b100, "divw">, Sched<[WriteIDiv32]>;
+def DIVUW   : ALUW_rr<0b0000001, 0b101, "divuw">, Sched<[WriteIDiv32]>;
+def REMW    : ALUW_rr<0b0000001, 0b110, "remw">, Sched<[WriteIDiv32]>;
+def REMUW   : ALUW_rr<0b0000001, 0b111, "remuw">, Sched<[WriteIDiv32]>;
 } // Predicates = [HasStdExtM, IsRV64]
 
 //===----------------------------------------------------------------------===//
 // Pseudo-instructions and codegen patterns
 //===----------------------------------------------------------------------===//
 
 let Predicates = [HasStdExtM] in {
 def : PatGprGpr<mul, MUL>;

llvm/lib/Target/RISCV/RISCVSchedRocket32.td

+//==- RISCVSchedRocket.td - Rocket Scheduling Definitions -*- tablegen -*-=//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines the itinerary class data for the ARM Cortex A53 processors.
+//
+//===----------------------------------------------------------------------===//
+
+// ===---------------------------------------------------------------------===//
+// The following definitions describe the simpler per-operand machine model.
+// This works with MachineScheduler. See MCSchedule.h for details.
+
+// Rocket machine model for scheduling and other instruction cost heuristics.
+def Rocket32Model : SchedMachineModel {
+  let MicroOpBufferSize = 0; // Explicitly set to zero since Rocket is in-order.
+  let IssueWidth = 1;        // 1 micro-ops are dispatched per cycle.
+  let LoadLatency = 3;
+  let MispredictPenalty = 3;
+  let CompleteModel = 1;
+}
+
+
+//===----------------------------------------------------------------------===//
+// Define each kind of processor resource and number available.
+
+// Modeling each pipeline as a ProcResource using the BufferSize = 0 since
+// Rocket is in-order.
+
+def Rocket32UnitALU    : ProcResource<1> { let BufferSize = 0; } // Int ALU
+def Rocket32UnitIMul   : ProcResource<1> { let BufferSize = 0; } // Int Multiply
+def Rocket32UnitIDiv   : ProcResource<1> { let BufferSize = 0; } // Int Division
+def Rocket32UnitMem    : ProcResource<1> { let BufferSize = 0; } // Load/Store
+def Rocket32UnitB      : ProcResource<1> { let BufferSize = 0; } // Branch
+
+def Rocket32UnitFPALU  : ProcResource<1> { let BufferSize = 0; } // FP ALU
+def Rocket32UnitFPDivSqrt  : ProcResource<1> { let BufferSize = 0; } // FP Divide/Sqrt
+
+//===----------------------------------------------------------------------===//
+// Subtarget-specific SchedWrite types which both map the ProcResources and
+// set the latency.
+
+let SchedModel = Rocket32Model in {
+
+def : WriteRes<WriteJmp, [Rocket32UnitB]>;
+def : WriteRes<WriteJal, [Rocket32UnitB]>;
+def : WriteRes<WriteJalr, [Rocket32UnitB]>;
+def : WriteRes<WriteJmpReg, [Rocket32UnitB]>;
+
+def : WriteRes<WriteJmpC, [Rocket32UnitB]>;
+def : WriteRes<WriteJalC, [Rocket32UnitB]>;
+def : WriteRes<WriteJalrC, [Rocket32UnitB]>;
+def : WriteRes<WriteJmpRegC, [Rocket32UnitB]>;
+
+def : WriteRes<WriteTrapRet, [Rocket32UnitB]>;
+
+def : WriteRes<WriteIALU, [Rocket32UnitALU]>;
+def : WriteRes<WriteIALU32, []>  { let Unsupported = 1; }
+def : WriteRes<WriteIALUC, [Rocket32UnitALU]>;
+def : WriteRes<WriteIALU32C, []>  { let Unsupported = 1; }
+def : WriteRes<WriteShift, [Rocket32UnitALU]>;
+def : WriteRes<WriteShiftC, [Rocket32UnitALU]>;
+def : WriteRes<WriteShift32, []>  { let Unsupported = 1; }
+
+// Multiplies on Rocket differ by implementation; placeholder until
+// we can determine how to read from command line
+def : WriteRes<WriteIMul, [Rocket32UnitIMul]> { let Latency = 4; }
+def : WriteRes<WriteIMul32, []>  { let Unsupported = 1; }
+
+// 32-bit divides have worse case latency of 34 cycle
+def : WriteRes<WriteIDiv, [Rocket32UnitIMul]> { let Latency = 34; }
+def : WriteRes<WriteIDiv32, []>  { let Unsupported = 1; }
+
+// Memory
+
+// What is the correct latecny to model for fence instructions?
+def : WriteRes<WriteFence, [Rocket32UnitMem]>;
+def : WriteRes<WriteSTB, [Rocket32UnitMem]>;
+def : WriteRes<WriteSTH, [Rocket32UnitMem]>;
+def : WriteRes<WriteSTW, [Rocket32UnitMem]>;
+def : WriteRes<WriteSTWC, [Rocket32UnitMem]>;
+def : WriteRes<WriteSTDW, []>  { let Unsupported = 1; }
+def : WriteRes<WriteSTDWC, []>  { let Unsupported = 1; }
+def : WriteRes<WriteFSt32, [Rocket32UnitMem]>;
+def : WriteRes<WriteFSt64, [Rocket32UnitMem]>;
+def : WriteRes<WriteFSt32C, [Rocket32UnitMem]>;
+def : WriteRes<WriteFSt64C, [Rocket32UnitMem]>;
+
+def : WriteRes<WriteLDB, [Rocket32UnitMem]> { let Latency = 3; }
+def : WriteRes<WriteLDH, [Rocket32UnitMem]> { let Latency = 3; }
+def : WriteRes<WriteLDW, [Rocket32UnitMem]> { let Latency = 2; }
+def : WriteRes<WriteLDWU, []>  { let Unsupported = 1; }
+def : WriteRes<WriteLDWC, [Rocket32UnitMem]> { let Latency = 2; }
+def : WriteRes<WriteLDDW, []>  { let Unsupported = 1; }
+def : WriteRes<WriteLDDWC, []>  { let Unsupported = 1; }
+def : WriteRes<WriteFLd32, [Rocket32UnitMem]> { let Latency = 2; }
+def : WriteRes<WriteFLd64, [Rocket32UnitMem]> { let Latency = 2; }
+def : WriteRes<WriteFLd32C, [Rocket32UnitMem]> { let Latency = 2; }
+def : WriteRes<WriteFLd64C, [Rocket32UnitMem]> { let Latency = 2; }
+
+def : WriteRes<WriteAtomicW, [Rocket32UnitMem]> { let Latency = 2; }
+def : WriteRes<WriteAtomicDW, []>  { let Unsupported = 1; }
+def : WriteRes<WriteAtomicLDW, [Rocket32UnitMem]> { let Latency = 2; }
+def : WriteRes<WriteAtomicLDDW, []>  { let Unsupported = 1; }
+def : WriteRes<WriteAtomicSTW, [Rocket32UnitMem]>;
+def : WriteRes<WriteAtomicSTDW, []>  { let Unsupported = 1; }
+
+def : WriteRes<WriteCSR, [Rocket32UnitALU]> { let Latency = 3; }
+
+// Most FP single precision operations are 4 cycles
+def : WriteRes<WriteFALU32, [Rocket32UnitFPALU]> { let Latency = 4; }
+
+// Most FP double precision operations are 6 cycles
+def : WriteRes<WriteFALU64, [Rocket32UnitFPALU]> { let Latency = 6; }
+
+def : WriteRes<WriteFCvtI32ToF32, [Rocket32UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFCvtI32ToF64, [Rocket32UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFCvtI64ToF32, []>  { let Unsupported = 1; }
+def : WriteRes<WriteFCvtI64ToF64, []>  { let Unsupported = 1; }
+def : WriteRes<WriteFCvtF64ToI64, []>  { let Unsupported = 1; }
+def : WriteRes<WriteFCvtF32ToI32, [Rocket32UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFCvtF32ToI64, []>  { let Unsupported = 1; }
+def : WriteRes<WriteFCvtF64ToI32, [Rocket32UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFCvtF32ToF64, [Rocket32UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFCvtF64ToF32, [Rocket32UnitFPALU]> { let Latency = 2; }
+
+def : WriteRes<WriteFClass32, [Rocket32UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFClass64, [Rocket32UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFCmp32, [Rocket32UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFCmp64, [Rocket32UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFMovF32ToI32, [Rocket32UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFMovI32ToF32, [Rocket32UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFMovI64ToF64, []> { let Unsupported = 1; }
+def : WriteRes<WriteFMovF64ToI64, []> { let Unsupported = 1; }
+
+def : WriteRes<WriteFMul32, [Rocket32UnitFPALU]> { let Latency = 5; }
+def : WriteRes<WriteFMulAdd32, [Rocket32UnitFPALU]> { let Latency = 5; }
+def : WriteRes<WriteFMulSub32, [Rocket32UnitFPALU]> { let Latency = 5; }
+def : WriteRes<WriteFMul64, [Rocket32UnitFPALU]> { let Latency = 7; }
+def : WriteRes<WriteFMulAdd64, [Rocket32UnitFPALU]> { let Latency = 7; }
+def : WriteRes<WriteFMulSub64, [Rocket32UnitFPALU]> { let Latency = 7; }
+
+// FP Divide unit on Rocket is not pipelined, so set resource cycles to latency
+def : WriteRes<WriteFDiv32, [Rocket32UnitFPDivSqrt]> { let Latency = 20;
+                           let ResourceCycles = [20];}
+def : WriteRes<WriteFDiv64, [Rocket32UnitFPDivSqrt]> { let Latency = 20;
+                           let ResourceCycles = [20];}
+
+// FP Sqrt unit on Rocket is not pipelined, so set resource cycles to latency
+def : WriteRes<WriteFSqrt32, [Rocket32UnitFPDivSqrt]> { let Latency = 20;
+                           let ResourceCycles = [20];}
+def : WriteRes<WriteFSqrt64, [Rocket32UnitFPDivSqrt]> { let Latency = 25;
+                           let ResourceCycles = [25];}
+
+def : WriteRes<WriteNop, []>;
+
+def : InstRW<[WriteIALU], (instrs COPY)>;
+}

llvm/lib/Target/RISCV/RISCVSchedRocket64.td

+//==- RISCVSchedRocket.td - Rocket Scheduling Definitions -*- tablegen -*-=//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines the itinerary class data for the ARM Cortex A53 processors.
+//
+//===----------------------------------------------------------------------===//
+
+// ===---------------------------------------------------------------------===//
+// The following definitions describe the simpler per-operand machine model.
+// This works with MachineScheduler. See MCSchedule.h for details.
+
+// Rocket machine model for scheduling and other instruction cost heuristics.
+def Rocket64Model : SchedMachineModel {
+  let MicroOpBufferSize = 0; // Explicitly set to zero since Rocket is in-order.
+  let IssueWidth = 1;        // 1 micro-ops are dispatched per cycle.
+  let LoadLatency = 3;
+  let MispredictPenalty = 3;
+}
+
+
+//===----------------------------------------------------------------------===//
+// Define each kind of processor resource and number available.
+
+// Modeling each pipeline as a ProcResource using the BufferSize = 0 since
+// Rocket is in-order.
+
+def Rocket64UnitALU    : ProcResource<1> { let BufferSize = 0; } // Int ALU
+def Rocket64UnitIMul   : ProcResource<1> { let BufferSize = 0; } // Int Multiply
+def Rocket64UnitIDiv   : ProcResource<1> { let BufferSize = 0; } // Int Division
+def Rocket64UnitMem    : ProcResource<1> { let BufferSize = 0; } // Load/Store
+def Rocket64UnitB      : ProcResource<1> { let BufferSize = 0; } // Branch
+
+def Rocket64UnitFPALU  : ProcResource<1> { let BufferSize = 0; } // FP ALU
+def Rocket64UnitFPDivSqrt  : ProcResource<1> { let BufferSize = 0; } // FP Divide/Sqrt
+
+//===----------------------------------------------------------------------===//
+// Subtarget-specific SchedWrite types which both map the ProcResources and
+// set the latency.
+
+let SchedModel = Rocket64Model in {
+
+def : WriteRes<WriteJmp, [Rocket64UnitB]>;
+def : WriteRes<WriteJal, [Rocket64UnitB]>;
+def : WriteRes<WriteJalr, [Rocket64UnitB]>;
+def : WriteRes<WriteJmpReg, [Rocket64UnitB]>;
+
+// RV64I has has no c.jal isntruction
+
+def : WriteRes<WriteJmpC, [Rocket64UnitB]>;
+def : WriteRes<WriteJalrC, [Rocket64UnitB]>;
+def : WriteRes<WriteJmpRegC, [Rocket64UnitB]>;
+def : WriteRes<WriteJalC, []>  { let Unsupported = 1; }
+def : WriteRes<WriteTrapRet, [Rocket64UnitB]>;
+
+def : WriteRes<WriteIALU32, [Rocket64UnitALU]>;
+def : WriteRes<WriteIALU, [Rocket64UnitALU]>;
+def : WriteRes<WriteIALU32C, [Rocket64UnitALU]>;
+def : WriteRes<WriteIALUC, [Rocket64UnitALU]>;
+def : WriteRes<WriteShift32, [Rocket64UnitALU]>;
+def : WriteRes<WriteShift, [Rocket64UnitALU]>;
+def : WriteRes<WriteShiftC, [Rocket64UnitALU]>;
+
+// Multiplies on Rocket differ by implementation; placeholder until
+// we can determine how to read from command line
+def : WriteRes<WriteIMul, [Rocket64UnitIMul]> { let Latency = 4; }
+def : WriteRes<WriteIMul32, [Rocket64UnitIMul]> { let Latency = 4; }
+
+// Integer divide varies based on operand magnitude and sign; use worse case latency
+def : WriteRes<WriteIDiv32, [Rocket64UnitIMul]> { let Latency = 34; }
+def : WriteRes<WriteIDiv, [Rocket64UnitIMul]> { let Latency = 66; }
+
+// Memory
+
+// What is the correct latency to model for fence instructions?
+def : WriteRes<WriteFence, [Rocket64UnitMem]>;
+def : WriteRes<WriteSTB, [Rocket64UnitMem]>;
+def : WriteRes<WriteSTH, [Rocket64UnitMem]>;
+def : WriteRes<WriteSTW, [Rocket64UnitMem]>;
+def : WriteRes<WriteSTDW, [Rocket64UnitMem]>;
+def : WriteRes<WriteSTWC, [Rocket64UnitMem]>;
+def : WriteRes<WriteSTDWC, [Rocket64UnitMem]>;
+def : WriteRes<WriteFSt32, [Rocket64UnitMem]>;
+def : WriteRes<WriteFSt32C, []>  { let Unsupported = 1; }
+def : WriteRes<WriteFSt64, [Rocket64UnitMem]>;
+def : WriteRes<WriteFSt64C, [Rocket64UnitMem]>;
+
+def : WriteRes<WriteLDB, [Rocket64UnitMem]> { let Latency = 3; }
+def : WriteRes<WriteLDH, [Rocket64UnitMem]> { let Latency = 3; }
+def : WriteRes<WriteLDW, [Rocket64UnitMem]> { let Latency = 2; }
+def : WriteRes<WriteLDWU, [Rocket64UnitMem]> { let Latency = 2; }
+def : WriteRes<WriteLDDW, [Rocket64UnitMem]> { let Latency = 2; }
+def : WriteRes<WriteLDWC, [Rocket64UnitMem]> { let Latency = 2; }
+def : WriteRes<WriteLDDWC, [Rocket64UnitMem]> { let Latency = 2; }
+def : WriteRes<WriteFLd32, [Rocket64UnitMem]> { let Latency = 2; }
+def : WriteRes<WriteFLd32C, []>  { let Unsupported = 1; }
+def : WriteRes<WriteFLd64, [Rocket64UnitMem]> { let Latency = 2; }
+def : WriteRes<WriteFLd64C, [Rocket64UnitMem]> { let Latency = 2; }
+
+def : WriteRes<WriteAtomicW, [Rocket64UnitMem]> { let Latency = 2; }
+def : WriteRes<WriteAtomicDW, [Rocket64UnitMem]> { let Latency = 2; }
+
+def : WriteRes<WriteAtomicLDW, [Rocket64UnitMem]> { let Latency = 2; }
+def : WriteRes<WriteAtomicLDDW, [Rocket64UnitMem]> { let Latency = 2; }
+
+def : WriteRes<WriteAtomicSTW, [Rocket64UnitMem]>;
+def : WriteRes<WriteAtomicSTDW, [Rocket64UnitMem]>;
+
+def : WriteRes<WriteCSR, [Rocket64UnitALU]> { let Latency = 3; }
+
+// Most FP single precision operations are 4 cycles
+def : WriteRes<WriteFALU32, [Rocket64UnitFPALU]> { let Latency = 4; }
+
+// Most FP double precision operations are 5 cycles
+def : WriteRes<WriteFALU64, [Rocket64UnitFPALU]> { let Latency = 6; }
+
+// Conversion instructions
+def : WriteRes<WriteFCvtI32ToF32, [Rocket32UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFCvtI32ToF64, [Rocket32UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFCvtI64ToF32, [Rocket32UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFCvtI64ToF64, [Rocket32UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFCvtF32ToI32, [Rocket32UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFCvtF32ToI64, [Rocket32UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFCvtF64ToI32, [Rocket32UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFCvtF64ToI64, [Rocket32UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFCvtF32ToF64, [Rocket32UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFCvtF64ToF32, [Rocket32UnitFPALU]> { let Latency = 2; }
+
+def : WriteRes<WriteFClass32, [Rocket64UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFClass64, [Rocket64UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFCmp32, [Rocket64UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFCmp64, [Rocket64UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFMovF32ToI32, [Rocket64UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFMovI32ToF32, [Rocket64UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFMovF64ToI64, [Rocket64UnitFPALU]> { let Latency = 2; }
+def : WriteRes<WriteFMovI64ToF64, [Rocket64UnitFPALU]> { let Latency = 2; }
+
+def : WriteRes<WriteFMul32, [Rocket64UnitFPALU]> { let Latency = 5; }
+def : WriteRes<WriteFMulAdd32, [Rocket64UnitFPALU]> { let Latency = 5; }
+def : WriteRes<WriteFMulSub32, [Rocket64UnitFPALU]> { let Latency = 5; }
+def : WriteRes<WriteFMul64, [Rocket64UnitFPALU]> { let Latency = 7; }
+def : WriteRes<WriteFMulAdd64, [Rocket64UnitFPALU]> { let Latency = 7; }
+def : WriteRes<WriteFMulSub64, [Rocket64UnitFPALU]> { let Latency = 7; }
+
+// FP Divide unit on Rocket is not pipelined, so set resource cycles to latency
+def : WriteRes<WriteFDiv32, [Rocket64UnitFPDivSqrt]> { let Latency = 20;
+                           let ResourceCycles = [20];}
+def : WriteRes<WriteFDiv64, [Rocket64UnitFPDivSqrt]> { let Latency = 20;
+                           let ResourceCycles = [20];}
+
+// FP Sqrt unit on Rocket is not pipelined, so set resource cycles to latency
+def : WriteRes<WriteFSqrt32, [Rocket64UnitFPDivSqrt]> { let Latency = 20;
+                           let ResourceCycles = [20];}
+def : WriteRes<WriteFSqrt64, [Rocket64UnitFPDivSqrt]> { let Latency = 25;
+                           let ResourceCycles = [25];}
+
+def : WriteRes<WriteNop, []>;
+
+def : InstRW<[WriteIALU], (instrs COPY)>;
+}

llvm/lib/Target/RISCV/RISCVSchedule.td

+//===-- RISCVSchedule.td - RISCV Scheduling Definitions -------*- tablegen -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+def WriteIALU : SchedWrite; // 32 or 64-bit integer ALU operations
+def WriteIALU32 : SchedWrite; // 32-bit integer ALU operations on RV64I
+def WriteShift32 : SchedWrite;  // 32-bit shift operations on RV64Ix
+def WriteShift : SchedWrite;  // 32 or 64-bit shift operations
+def WriteIDiv : SchedWrite; // 32-bit or 64-bit divide and remainder
+def WriteIDiv32 : SchedWrite; // 32-bit divide and remainder on RV64I
+def WriteIMul : SchedWrite; // 32-bit or 64-bit multiply
+def WriteIMul32 : SchedWrite; // 32-bit multiply on RV64I
+def WriteJmp  : SchedWrite; // Jump
+def WriteJal  : SchedWrite; // Jump and link
+def WriteJalr : SchedWrite; // Jump and link register
+def WriteJmpReg : SchedWrite; // Jump register
+def WriteFence  : SchedWrite; // Fence instructions
+def WriteNop  : SchedWrite;
+def WriteLDB  : SchedWrite; // Load byte
+def WriteLDW  : SchedWrite; // Load word
+def WriteLDWU : SchedWrite; // Load word unsigned
+def WriteLDH  : SchedWrite; // Load half-word
+def WriteLDDW : SchedWrite; // Load double-word
+def WriteCSR  : SchedWrite; // CSR instructions
+def WriteSTB  : SchedWrite; // Store byte
+def WriteSTW  : SchedWrite; // Store word
+def WriteSTH  : SchedWrite; // Store half-word
+def WriteSTDW     : SchedWrite; // Store double-word
+def WriteAtomicW    : SchedWrite; //Atomic memory operation word size
+def WriteAtomicDW   : SchedWrite; //Atomic memory operation double word size
+def WriteAtomicLDW  : SchedWrite; // Atomic load word
+def WriteAtomicLDDW : SchedWrite; // Atomic load double word
+def WriteAtomicSTW  : SchedWrite; // Atomic store word
+def WriteAtomicSTDW : SchedWrite; // Atomic store double word
+def WriteTrapRet    : SchedWrite; // Trap return instructions
+def WriteFALU32     : SchedWrite; // FP 32-bit computation
+def WriteFALU64     : SchedWrite; // FP 64-bit computation
+def WriteFMul32     : SchedWrite; // 32-bit floating point multiply
+def WriteFMulAdd32  : SchedWrite; // 32-bit floating point multiply add
+def WriteFMulSub32  : SchedWrite; // 32-bit floating point multiply sub
+def WriteFMul64     : SchedWrite; // 64-bit floating point multiply
+def WriteFMulAdd64  : SchedWrite;   // 64-bit floating point multiply add
+def WriteFMulSub64  : SchedWrite; // 64-bit floating point multiply sub
+def WriteFDiv32     : SchedWrite; // 32-bit floating point divide
+def WriteFDiv64     : SchedWrite; // 64-bit floating point divide
+def WriteFSqrt32    : SchedWrite; // 32-bit floating point sqrt
+def WriteFSqrt64    : SchedWrite; // 64-bit floating point sqrt
+
+// Integer to float conversions
+def WriteFCvtI32ToF32  : SchedWrite;
+def WriteFCvtI32ToF64  : SchedWrite;
+def WriteFCvtI64ToF64  : SchedWrite;  // RV64I only
+def WriteFCvtI64ToF32  : SchedWrite;  // RV64I only
+
+//Float to integer conversions
+def WriteFCvtF32ToI32  : SchedWrite;
+def WriteFCvtF32ToI64  : SchedWrite;  // RV64I only
+def WriteFCvtF64ToI32  : SchedWrite;
+def WriteFCvtF64ToI64  : SchedWrite;  // RV64I only
+
+// Float to float conversions
+def WriteFCvtF32ToF64  : SchedWrite;
+def WriteFCvtF64ToF32  : SchedWrite;
+
+def WriteFConv32    : SchedWrite; // 32-bit floating point convert
+def WriteFConv64    : SchedWrite; // 64-bit floating point convert
+def WriteFClass32   : SchedWrite; // 32-bit floating point classify
+def WriteFClass64   : SchedWrite; // 64-bit floating point classify
+def WriteFCmp32     : SchedWrite; // 32-bit floating point compare
+def WriteFCmp64     : SchedWrite; // 64-bit floating point compare
+
+def WriteFMovF32ToI32   : SchedWrite;
+def WriteFMovI32ToF32   : SchedWrite;
+def WriteFMovF64ToI64   : SchedWrite; // RV64I only
+def WriteFMovI64ToF64   : SchedWrite; // RV64I only
+
+def WriteFMov32     : SchedWrite; // 32-bit floating point move
+def WriteFMov64     : SchedWrite; // 64-bit floating point move
+def WriteFLd32      : SchedWrite; // Floating point sp load
+def WriteFLd64      : SchedWrite; // Floating point dp load
+def WriteFSt32      : SchedWrite; // Floating point sp store
+def WriteFSt64      : SchedWrite; // Floating point dp store
+
+def WriteIALUC      : SchedWrite; // Compressed 32 or 64-bit integer ALU operations
+def WriteIALU32C    : SchedWrite; // Compressed 32-bit integer ALU operations on RV64Ix
+def WriteShiftC     : SchedWrite; // Compressed 32 or 64-bit shift operations
+def WriteLDWC     : SchedWrite; // Compressed load word
+def WriteLDDWC      : SchedWrite; // Compressed load double-word
+def WriteSTWC     : SchedWrite; // Compressed store word
+def WriteSTDWC      : SchedWrite; // Compressed store double-word
+def WriteJmpC     : SchedWrite; // Jump
+def WriteJalC     : SchedWrite; // Jump and link
+def WriteJalrC      : SchedWrite; // Jump and link register
+def WriteJmpRegC    : SchedWrite; // Jump register
+def WriteFLd32C     : SchedWrite; // Compressed floating point sp load
+def WriteFLd64C     : SchedWrite; // Compressed floating point dp load
+def WriteFSt32C     : SchedWrite; // Compressed floating point sp store
+def WriteFSt64C     : SchedWrite; // Compressed floating point dp store
+
+def ReadIM    : SchedRead;  // 32 & 64-bit multiply
+def ReadID    : SchedRead;  // 32 & 64-bit divide