[llvm] [MIPS] Add Scheduling model for MIPS i6400 and i6500 CPUs (PR #132704)
Mallikarjuna Gouda via llvm-commits
llvm-commits at lists.llvm.org
Tue Mar 25 03:51:07 PDT 2025
https://github.com/mgoudar updated https://github.com/llvm/llvm-project/pull/132704
>From e2e9366f1fe78d1dd7d6707b12a7c470d7e85b5b Mon Sep 17 00:00:00 2001
From: Mallikarjuna Gouda <mgouda at mips.com>
Date: Mon, 24 Mar 2025 14:35:32 +0530
Subject: [PATCH] [MIPS] Add Scheduling model for MIPS i6400 and i6500 CPUs
Add scheduling model for the MIPS i6400 and i6500, an in-order
MIPS64R6 processor.
i6400 and i6500 share same instruction latencies.
CPU has following pipelines
- Two ALUs
- Multiply and Divide unit (MDU)
- Branch Unit (CTU)
- Load/Store Unit (LSU)
- Short Floating-point Unit and
- Long Floating-point Unit
---
llvm/lib/Target/Mips/Mips.td | 5 +-
llvm/lib/Target/Mips/Mips32r6InstrInfo.td | 11 +-
llvm/lib/Target/Mips/MipsScheduleI6400.td | 452 ++++++++++++++++++++++
llvm/test/tools/llvm-mca/Mips/i6400.s | 105 +++++
4 files changed, 568 insertions(+), 5 deletions(-)
create mode 100644 llvm/lib/Target/Mips/MipsScheduleI6400.td
create mode 100644 llvm/test/tools/llvm-mca/Mips/i6400.s
diff --git a/llvm/lib/Target/Mips/Mips.td b/llvm/lib/Target/Mips/Mips.td
index 9159d11fd486f..4990d28c4fcd7 100644
--- a/llvm/lib/Target/Mips/Mips.td
+++ b/llvm/lib/Target/Mips/Mips.td
@@ -224,6 +224,7 @@ include "MipsRegisterBanks.td"
include "MipsCombine.td"
// Avoid forward declaration issues.
+include "MipsScheduleI6400.td"
include "MipsScheduleP5600.td"
include "MipsScheduleGeneric.td"
@@ -267,8 +268,8 @@ def : Proc<"mips64r6", [FeatureMips64r6]>;
def : Proc<"octeon", [FeatureMips64r2, FeatureCnMips]>;
def : Proc<"octeon+", [FeatureMips64r2, FeatureCnMips, FeatureCnMipsP]>;
def : ProcessorModel<"p5600", MipsP5600Model, [ImplP5600]>;
-def : ProcessorModel<"i6400", NoSchedModel, [ImplI6400]>;
-def : ProcessorModel<"i6500", NoSchedModel, [ImplI6400]>;
+def : ProcessorModel<"i6400", MipsI6400Model, [ImplI6400]>;
+def : ProcessorModel<"i6500", MipsI6400Model, [ImplI6400]>;
def MipsAsmParser : AsmParser {
let ShouldEmitMatchRegisterName = 0;
diff --git a/llvm/lib/Target/Mips/Mips32r6InstrInfo.td b/llvm/lib/Target/Mips/Mips32r6InstrInfo.td
index 27b9ce60ba826..cfa92f57383fb 100644
--- a/llvm/lib/Target/Mips/Mips32r6InstrInfo.td
+++ b/llvm/lib/Target/Mips/Mips32r6InstrInfo.td
@@ -983,7 +983,10 @@ let AdditionalPredicates = [NotInMicroMips] in {
def SEL_S : R6MMR6Rel, SEL_S_ENC, SEL_S_DESC, ISA_MIPS32R6, HARDFLOAT;
def SDC2_R6 : SDC2_R6_ENC, SDC2_R6_DESC, ISA_MIPS32R6;
def SWC2_R6 : SWC2_R6_ENC, SWC2_R6_DESC, ISA_MIPS32R6;
- def SIGRIE : SIGRIE_ENC, SIGRIE_DESC, ISA_MIPS32R6;
+
+ let hasNoSchedulingInfo = 1 in {
+ def SIGRIE : SIGRIE_ENC, SIGRIE_DESC, ISA_MIPS32R6;
+ }
}
let AdditionalPredicates = [NotInMicroMips] in {
@@ -996,8 +999,10 @@ let AdditionalPredicates = [NotInMicroMips] in {
}
let AdditionalPredicates = [NotInMicroMips] in {
- def GINVI : R6MMR6Rel, GINVI_ENC, GINVI_DESC, ISA_MIPS32R6, ASE_GINV;
- def GINVT : R6MMR6Rel, GINVT_ENC, GINVT_DESC, ISA_MIPS32R6, ASE_GINV;
+ let hasNoSchedulingInfo = 1 in {
+ def GINVI : R6MMR6Rel, GINVI_ENC, GINVI_DESC, ISA_MIPS32R6, ASE_GINV;
+ def GINVT : R6MMR6Rel, GINVT_ENC, GINVT_DESC, ISA_MIPS32R6, ASE_GINV;
+ }
}
//===----------------------------------------------------------------------===//
diff --git a/llvm/lib/Target/Mips/MipsScheduleI6400.td b/llvm/lib/Target/Mips/MipsScheduleI6400.td
new file mode 100644
index 0000000000000..ff12a2d5a057a
--- /dev/null
+++ b/llvm/lib/Target/Mips/MipsScheduleI6400.td
@@ -0,0 +1,452 @@
+//==- MipsScheduleI6400.td - I6400 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 MipsI6400Model : SchedMachineModel {
+ int IssueWidth = 2; // 2x dispatched per cycle
+ int MicroOpBufferSize = 48; // min(48, 48, 64)
+ int LoadLatency = 3;
+ int MispredictPenalty = 8;
+
+ let CompleteModel = 1;
+ let FullInstRWOverlapCheck = 1;
+
+ list<Predicate> UnsupportedFeatures = [HasMips64r5, HasMips32r5,
+ InMicroMips, InMips16Mode, HasCnMips,
+ HasCnMipsP, HasDSP, HasDSPR2,
+ HasMips3D, HasMT, HasCRC,
+ NotMips32r6, NotMips64r6, HasEVA];
+}
+
+let SchedModel = MipsI6400Model in {
+
+ // AGEN Pipelines
+ // ==============
+ def I6400AGEN : ProcResource<1> { let BufferSize = 16; }
+ def I6400IssueLSU : ProcResource<1> { let Super = I6400AGEN; }
+ def I6400IssueALU1 : ProcResource<1> { let Super = I6400AGEN; }
+
+ def I6400WriteLSUStore : SchedWriteRes<[I6400IssueLSU]> { let Latency = 1; }
+ def I6400WriteLSUStore2 : SchedWriteRes<[I6400IssueLSU]> {
+ let Latency = 8;
+ let ReleaseAtCycles = [5];
+ }
+ def I6400WriteLSULoad : SchedWriteRes<[I6400IssueLSU]> { let Latency = 3; }
+ def I6400WriteLSUPref : SchedWriteRes<[I6400IssueLSU]> { let Latency = 1; }
+ def I6400WriteLSUOther : SchedWriteRes<[I6400IssueLSU]> {
+ let Latency = 6;
+ let ReleaseAtCycles = [5];
+ }
+
+ // LSU pipelines
+ // =============
+ def : InstRW<[I6400WriteLSUStore], (instrs SB, SD, SH, SW, SDC1, SDC164, SWC1, SWC2_R6,
+ SDC2_R6, SDC3)>;
+ def : InstRW<[I6400WriteLSUStore2], (instrs SC_R6, SCD_R6, SYNCI, TLBP, TLBR,
+ TLBWI, TLBWR, TLBINV, TLBINVF,
+ CACHE_R6, SC64_R6)>;
+ def : InstRW<[I6400WriteLSULoad],
+ (instrs LB, LBu, LBu64, LD, LH, LHu, LHu64, LW, LWu, LDC1,
+ LDC164, LWC1, LD_F16, ST_F16, LDC2_R6, LDC3, LWC2_R6,
+ LLD_R6, LL_R6, LWPC, LWUPC, LDPC, ST_B, ST_H, ST_W, ST_D,
+ LB64, LH64, LW64, LWL64, LWR64, SB64, SH64, SW64, SWL64,
+ SWR64, LL64_R6)>;
+ def : InstRW<[I6400WriteLSUPref], (instrs PREF, PREF_R6, PAUSE)>;
+ def : InstRW<[I6400WriteLSUOther], (instrs SYNC)>;
+
+ // CONTROL Pipelines
+ // =================
+ def I6400CTRL : ProcResource<1> { let BufferSize = 16; }
+ def I6400IssueCTU : ProcResource<1> { let Super = I6400CTRL; }
+ def I6400IssueALU0 : ProcResource<1> { let Super = I6400CTRL; }
+
+ def I6400WriteALU0 : SchedWriteRes<[I6400IssueALU0]> { let Latency = 1; }
+ def I6400WriteALU1 : SchedWriteRes<[I6400IssueALU1]> { let Latency = 1; }
+ def I6400WriteCTU : SchedWriteRes<[I6400IssueCTU]> { let Latency = 1; }
+
+ // CTU pipelines
+ // =============
+ def : InstRW<[I6400WriteCTU],
+ (instrs J, JAL, JALR, B, BEQ, BNE, BGEZ, BGTZ, BLEZ, BLTZ,
+ JALR64, JALR64Pseudo, JIALC, JIALC64, JIC, JIC64, JR64,
+ JR_HB_R6,
+ JR_HB64_R6, NAL, SDBBP_R6, SYSCALL, BEQC64, BEQZC64, BGEC64,
+ BGEUC64, BGTZC64, BLEZC64, BLTC64, BLTUC64, BNEC64, BNEZC64,
+ PseudoIndirectBranchR6, BC, BALC, BEQZC, BNEZC, BLEZC, BGEZC,
+ BGTZC, BLTZC, BEQZALC, BNEZALC, BLEZALC, BGEZALC, BGTZALC,
+ BLTZALC, BEQC, BNEC, BGEC, BLTC, BGEUC, BLTUC, BAL, BOVC,
+ BNVC, BC1EQZ, BC1NEZ, BREAK, ERET, ERETNC, BAL_BR, DERET,
+ JALRHBPseudo, JALRPseudo, JALR_HB, JALR_HB64, TAILCALL,
+ PseudoIndirectBranch64R6, TAILCALL64R6REG,
+ TAILCALLR6REG, PseudoIndrectHazardBranch64R6,
+ PseudoIndrectHazardBranchR6, TAILCALLHB64R6REG,
+ TAILCALLHBR6REG, PseudoReturn, PseudoReturn64, ERet, RetRA,
+ BC2EQZ, BC2NEZ, TLT, TLTU, TNE, WAIT, DI, TRAP, EI,
+ BEQ64, BGEZ64, BGEZC64, BGTZ64, BGTZ64, BLEZ64, BLTZ64,
+ BLTZC64, BNE64, JALRHB64Pseudo)>;
+
+ // Either ALU0 or ALU1 pipelines
+ // =============================
+ def I6400IssueEitherALU : ProcResGroup<[I6400IssueALU0, I6400IssueALU1]>;
+ def I6400WriteEitherALU : SchedWriteRes<[I6400IssueEitherALU]> {
+ let Latency = 1;
+ }
+
+ def : InstRW<[I6400WriteEitherALU],
+ (instrs ADD, ADDiu, ADDIUPC, ADDu, ALIGN, ALUIPC, AND, ANDi, AUI,
+ AUIPC, BITSWAP, CFC1, CLO_R6, CLZ_R6, CTC1, DADD, DADDiu,
+ DADDu, DAHI, DALIGN, DATI, DAUI, DBITSWAP, DCLO_R6, DCLZ_R6,
+ DEXT, DEXT64_32, DEXTM, DEXTU, DINS, DINSM, DINSU, DLSA_R6,
+ DMFC1, DMTC1, DROTR, DROTR32, DROTRV, DSBH, DSHD, DSLL,
+ DSLL32, DSLLV, DSRA, DSRA32, DSRAV, DSRL, DSRL32, DSRLV,
+ DSUB, DSUBu, EXT, INS, LSA, LSA_R6, LUi, MFC1, MFC1_D64,
+ MFC0, MFC2, MTC0, MTC2,
+ MFHC1_D32, MFHC1_D64, MTC1, MTC1_D64, MTHC1_D32, MTHC1_D64,
+ NOP, NOR, OR, ORi, ROTR, ROTRV, SEB, SEB64, SEH, SEH64,
+ SELEQZ, SELEQZ64, SELNEZ, SELNEZ64, SLL, SLLV, SLT, SLTi, SLTiu, SLTu, SRA, SRAV,
+ SRL, SRLV, SSNOP, SUB, SUBu, WSBH, XOR, XORi, SLT64, SLTu64,
+ AND64, OR64, XOR64, NOR64, SLTi64, SLTiu64, ANDi64, ORi64,
+ XORi64, LUi64, DSLL64_32, SLL64_32, SLL64_64,
+ LONG_BRANCH_LUi2Op_64, LONG_BRANCH_DADDiu2Op,
+ LONG_BRANCH_DADDiu, DLSA,
+ TEQ, TGE, TGEU, COPY,
+ BuildPairF64, BuildPairF64_64,
+ ExtractElementF64, ExtractElementF64_64,
+ SELNEZ_D, SELNEZ_S, SELEQZ_D, SELEQZ_S,
+ SEL_D, SEL_S, EHB, RDHWR, RDHWR64, EVP, DVP,
+ DMFC0, DMFC2, DMTC0, DMTC2)>;
+
+ // MDU pipelines
+ // =============
+ def I6400MDU : ProcResource<1>;
+ def I6400GPMUL : SchedWriteRes<[I6400MDU]> { let Latency = 4; }
+ def : InstRW<[I6400GPMUL], (instrs MUL_R6, MULU, MUH, MUHU, DMUL_R6, DMUH,
+ DMULU, DMUHU)>;
+
+ def I6400GPDIV : SchedWriteRes<[I6400MDU]> { let Latency = 32; }
+ def : InstRW<[I6400GPDIV], (instrs DIV, DIVU, MOD, MODU, DDIV, DMOD, DDIVU,
+ DMODU)>;
+ def : InstRW<[I6400GPDIV], (instregex "^MOD_(S|U)_[BHWD]$")>;
+
+ // FPU pipelines
+ // =============
+ def I6400FPU : ProcResource<3> { let BufferSize = 16; }
+ def I6400FPUShort : ProcResource<1> { let Super = I6400FPU; }
+ def I6400FPULong : ProcResource<1> { let Super = I6400FPU; }
+ def I6400FPUApu : ProcResource<1>;
+ def I6400FPUFloatL : ProcResource<1>;
+
+ def I6400FPUFabs : SchedWriteRes<[I6400FPUShort]> { let Latency = 1; }
+ def : InstRW<[I6400FPUFabs], (instrs FABS_S, FNEG_S, FMOV_S,
+ FMOV_D32, FMOV_D64,
+ FNEG_D32, FNEG_D64, CLASS_S, CLASS_D)>;
+
+ def I6400FPUFadd : SchedWriteRes<[I6400FPULong]> {
+ let Latency = 4;
+ let ReleaseAtCycles = [1];
+ }
+ def : InstRW<[I6400FPUFadd], (instrs FADD_S, FSUB_S, FSUB_D32, FSUB_D64)>;
+
+ def I6400FPUFmul : SchedWriteRes<[I6400FPULong, I6400FPUApu]> {
+ let Latency = 5;
+ let ReleaseAtCycles = [1, 4];
+ }
+ def : InstRW<[I6400FPUFmul], (instrs FMUL_S, FMUL_D32, FMUL_D64)>;
+
+ def I6400FPUFdivS : SchedWriteRes<[I6400FPULong, I6400FPUApu]> {
+ let Latency = 22;
+ let ReleaseAtCycles = [1, 21];
+ }
+ def : InstRW<[I6400FPUFdivS], (instrs FDIV_S, FSQRT_S, FDIV_D32, FDIV_D64,
+ FSQRT_D, FSQRT_D32, FSQRT_D64,
+ FSQRT_W, RECIP_S, RECIP_D32, RECIP_D64,
+ FRSQRT_D, FRSQRT_W, RSQRT_S, RSQRT_D32,
+ RSQRT_D64, FRCP_D, FRCP_W)>;
+ def : InstRW<[I6400FPUFdivS], (instregex "^DIV_S_(B|D|H|W)$")>;
+ def : InstRW<[I6400FPUFdivS], (instregex "^DIV_U_(B|D|H|W)$")>;
+
+ // MSA pipelines
+ // =============
+ // Short pipe
+ def I6400MSAAdd : SchedWriteRes<[I6400FPUShort]> {
+ let Latency = 2;
+ let ReleaseAtCycles = [2];
+ }
+ def : InstRW<[I6400MSAAdd], (instrs ADDV_B, ADDV_H, ADDV_W, ADDV_D, ADDVI_B,
+ ADDVI_H, ADDVI_W, ADDVI_D, SUBV_B, SUBV_H,
+ SUBV_W, SUBV_D, SUBVI_B, SUBVI_H, SUBVI_W,
+ SUBVI_D)>;
+ def : InstRW<[I6400MSAAdd], (instregex "^ASUB_S_(B|D|H|W)$")>;
+ def : InstRW<[I6400MSAAdd], (instregex "^ASUB_U_(B|D|H|W)$")>;
+ def : InstRW<[I6400MSAAdd], (instregex "^SUBSUS_U_(B|D|H|W)$")>;
+
+ def I6400MSAMaxMin : SchedWriteRes<[I6400FPUShort]> {
+ let Latency = 2;
+ let ReleaseAtCycles = [1];
+ }
+ def : InstRW<[I6400MSAMaxMin], (instrs MAX_D, MAX_S, MAXA_D, MAXA_S,
+ MIN_D, MIN_S, MINA_D, MINA_S)>;
+
+ def I6400MSAIntAdd : SchedWriteRes<[I6400FPUShort]> {
+ let Latency = 2;
+ let ReleaseAtCycles = [2];
+ }
+ def : InstRW<[I6400MSAIntAdd],
+ (instrs ADD_A_B, ADD_A_H, ADD_A_W, ADD_A_D, ADDS_A_B, ADDS_A_H,
+ ADDS_A_W, ADDS_A_D, ADDS_S_B, ADDS_S_H, ADDS_S_W, ADDS_S_D,
+ ADDS_U_B, ADDS_U_H, ADDS_U_W, ADDS_U_D, HADD_S_H, HADD_S_W,
+ HADD_S_D, HADD_U_H, HADD_U_W, HADD_U_D, SUBS_S_B, SUBS_S_H,
+ SUBS_S_W, SUBS_S_D, SUBS_U_B, SUBS_U_H, SUBS_U_W, SUBS_U_D,
+ SUBSUU_S_B, SUBSUU_S_H, SUBSUU_S_W, SUBSUU_S_D, HSUB_S_H,
+ HSUB_S_W, HSUB_S_D, HSUB_U_H, HSUB_U_W, HSUB_U_D, AVE_S_B,
+ AVE_S_H, AVE_S_W, AVE_S_D, AVE_U_B, AVE_U_H, AVE_U_W,
+ AVE_U_D, AVER_S_B, AVER_S_H, AVER_S_W, AVER_S_D, AVER_U_B,
+ AVER_U_H, AVER_U_W, AVER_U_D, MIN_A_B, MIN_A_H, MIN_A_W,
+ MIN_A_D, MIN_S_B, MIN_S_H, MIN_S_W, MIN_S_D, MIN_U_B,
+ MIN_U_H, MIN_U_W, MIN_U_D, MINI_S_B, MINI_S_H, MINI_S_W,
+ MINI_S_D, MINI_U_B, MINI_U_H, MINI_U_W, MINI_U_D, MAX_A_B,
+ MAX_A_H, MAX_A_W, MAX_A_D, MAX_S_B, MAX_S_H, MAX_S_W,
+ MAX_S_D, MAX_U_B, MAX_U_H, MAX_U_W, MAX_U_D, MAXI_S_B,
+ MAXI_S_H, MAXI_S_W, MAXI_S_D, MAXI_U_B, MAXI_U_H, MAXI_U_W,
+ MAXI_U_D, CEQ_B, CEQ_H, CEQ_W, CEQ_D, CEQI_B, CEQI_H, CEQI_W,
+ CEQI_D, CLE_S_B, CLE_S_H, CLE_S_W, CLE_S_D, CLE_U_B, CLE_U_H,
+ CLE_U_W, CLE_U_D, CLEI_S_B, CLEI_S_H, CLEI_S_W, CLEI_S_D,
+ CLEI_U_B, CLEI_U_H, CLEI_U_W, CLEI_U_D, CLT_S_B, CLT_S_H,
+ CLT_S_W, CLT_S_D, CLT_U_B, CLT_U_H, CLT_U_W, CLT_U_D,
+ CLTI_S_B, CLTI_S_H, CLTI_S_W, CLTI_S_D, CLTI_U_B, CLTI_U_H,
+ CLTI_U_W, CLTI_U_D)>;
+
+ def I6400MSAShortLogic3 : SchedWriteRes<[I6400FPUShort]> {
+ let Latency = 3;
+ let ReleaseAtCycles = [2];
+ }
+ def : InstRW<[I6400MSAShortLogic3],
+ (instrs SAT_S_B, SAT_S_H, SAT_S_W, SAT_S_D, SAT_U_B, SAT_U_H,
+ SAT_U_W, SAT_U_D, PCNT_B, PCNT_H, PCNT_W, PCNT_D)>;
+
+ def I6400MSAShortLogic2 : SchedWriteRes<[I6400FPUShort]> {
+ let Latency = 2;
+ let ReleaseAtCycles = [2];
+ }
+ def : InstRW<[I6400MSAShortLogic2],
+ (instrs SLL_B, SLL_H, SLL_W, SLL_D, SLLI_B, SLLI_H, SLLI_W,
+ SLLI_D, SRA_B, SRA_H, SRA_W, SRA_D, SRAI_B, SRAI_H, SRAI_W,
+ SRAI_D, SRAR_B, SRAR_H, SRAR_W, SRAR_D, SRARI_B, SRARI_H,
+ SRARI_W, SRARI_D, SRL_B, SRL_H, SRL_W, SRL_D, SRLI_B, SRLI_H,
+ SRLI_W, SRLI_D, SRLR_B, SRLR_H, SRLR_W, SRLR_D, SRLRI_B,
+ SRLRI_H, SRLRI_W, SRLRI_D, NLOC_B, NLOC_H, NLOC_W, NLOC_D,
+ NLZC_B, NLZC_H, NLZC_W, NLZC_D, BNEG_B, BNEG_H, BNEG_W,
+ BNEG_D, BNEGI_B, BNEGI_H, BNEGI_W, BNEGI_D, BCLR_B, BCLR_H,
+ BCLR_W, BCLR_D, BCLRI_B, BCLRI_H, BCLRI_W, BCLRI_D, SHF_B,
+ SHF_H, SHF_W)>;
+
+ def I6400MSAShortLogic : SchedWriteRes<[I6400FPUShort]> {
+ let Latency = 1;
+ let ReleaseAtCycles = [2];
+ }
+ def : InstRW<[I6400MSAShortLogic],
+ (instrs AND_V, ANDI_B, OR_V, ORI_B, XOR_V, XORI_B,
+ NOR_V, NORI_B)>;
+
+ def : InstRW<[I6400MSAShortLogic], (instregex "^NOR_V_(H|W|D)_PSEUDO$")>;
+ def : InstRW<[I6400MSAShortLogic], (instregex "^OR_V_(H|W|D)_PSEUDO$")>;
+ def : InstRW<[I6400MSAShortLogic], (instregex "^XOR_V_(H|W|D)_PSEUDO$")>;
+ def : InstRW<[I6400MSAShortLogic], (instregex "^AND_V_(H|W|D)_PSEUDO$")>;
+ def : InstRW<[I6400MSAShortLogic], (instregex "^ILVEV_(B|H|W|D)$")>;
+ def : InstRW<[I6400MSAShortLogic], (instregex "^ILVL_(B|H|W|D)$")>;
+ def : InstRW<[I6400MSAShortLogic], (instregex "^ILVOD_(B|H|W|D)$")>;
+ def : InstRW<[I6400MSAShortLogic], (instregex "^ILVR_(B|H|W|D)$")>;
+ def : InstRW<[I6400MSAShortLogic], (instregex "^PCKEV_(B|H|W|D)$")>;
+ def : InstRW<[I6400MSAShortLogic], (instregex "^PCKOD_(B|H|W|D)$")>;
+ def : InstRW<[I6400MSAShortLogic], (instregex "^FILL_(B|H|W|D)$")>;
+ def : InstRW<[I6400MSAShortLogic], (instregex "^FILL_F(D|W)_PSEUDO$")>;
+ def : InstRW<[I6400MSAShortLogic], (instregex "^INSERT_F(D|W)_PSEUDO$")>;
+ def : InstRW<[I6400MSAShortLogic], (instregex "^SPLAT_(B|H|W|D)$")>;
+ def : InstRW<[I6400MSAShortLogic], (instregex "^SPLATI_(B|H|W|D)$")>;
+
+ def I6400MSAShortLogic4 : SchedWriteRes<[I6400FPUShort]> { let Latency = 1; }
+ def : InstRW<[I6400MSAShortLogic4],
+ (instrs CTCMSA, CFCMSA, COPY_S_B, COPY_S_H, COPY_S_W, COPY_S_D,
+ COPY_U_B, COPY_U_H, COPY_U_W, BNZ_B, BNZ_H, BNZ_W, BNZ_D,
+ BNZ_V, BZ_B, BZ_H, BZ_W, BZ_D, BZ_V)>;
+
+ def I6400MSAMove : SchedWriteRes<[I6400FPUShort]> {
+ let Latency = 3;
+ let ReleaseAtCycles = [2];
+ }
+ def : InstRW<[I6400MSAMove], (instrs LD_B, LD_H, LD_W, LD_D)>;
+
+ def I6400MSAMove2 : SchedWriteRes<[I6400FPUShort]> {
+ let Latency = 1;
+ let ReleaseAtCycles = [2];
+ }
+ def : InstRW<[I6400MSAMove2], (instrs LDI_B, LDI_H, LDI_W, LDI_D, MOVE_V)>;
+
+ def I6400MSACmp : SchedWriteRes<[I6400FPUShort]> {
+ let Latency = 2;
+ let ReleaseAtCycles = [2];
+ }
+ def : InstRW<[I6400MSACmp],
+ (instrs FCAF_W, FCAF_D, FCUN_W, FCUN_D, FCOR_W, FCOR_D, FCEQ_W,
+ FCEQ_D, FCUNE_W, FCUNE_D, FCUEQ_W, FCUEQ_D, FCNE_W, FCNE_D,
+ FCLT_W, FCLT_D, FCULT_W, FCULT_D, FCLE_W, FCLE_D, FCULE_W,
+ FCULE_D, FSAF_W, FSAF_D, FSUN_W, FSUN_D, FSOR_W, FSOR_D,
+ FSEQ_W, FSEQ_D, FSUNE_W, FSUNE_D, FSUEQ_W, FSUEQ_D, FSNE_W,
+ FSNE_D, FSLT_W, FSLT_D, FSULT_W, FSULT_D, FSLE_W, FSLE_D,
+ FSULE_W, FSULE_D)>;
+
+ def I6400MSAShortFloat : SchedWriteRes<[I6400FPUShort]> {
+ let Latency = 2;
+ let ReleaseAtCycles = [2];
+ }
+ def : InstRW<[I6400MSAShortFloat], (instrs FMAX_W, FMAX_D, FMAX_A_W, FMAX_A_D,
+ FMIN_W, FMIN_D, FMIN_A_W, FMIN_A_D,
+ FCLASS_W, FCLASS_D, FABS_D, FABS_W,
+ FABS_D32, FABS_D64)>;
+ def I6400MSAFloatCompare : SchedWriteRes<[I6400FPUShort]> {
+ let Latency = 2;
+ let ReleaseAtCycles = [1];
+ }
+ def : InstRW<[I6400MSAFloatCompare], (instregex "^CMP_UN_(S|D)$")>;
+ def : InstRW<[I6400MSAFloatCompare], (instregex "^CMP_UEQ_(S|D)$")>;
+ def : InstRW<[I6400MSAFloatCompare], (instregex "^CMP_EQ_(S|D)$")>;
+ def : InstRW<[I6400MSAFloatCompare], (instregex "^CMP_LT_(S|D)$")>;
+ def : InstRW<[I6400MSAFloatCompare], (instregex "^CMP_ULT_(S|D)$")>;
+ def : InstRW<[I6400MSAFloatCompare], (instregex "^CMP_LE_(S|D)$")>;
+ def : InstRW<[I6400MSAFloatCompare], (instregex "^CMP_ULE_(S|D)$")>;
+ def : InstRW<[I6400MSAFloatCompare], (instregex "^CMP_F_(D|S)$")>;
+ def : InstRW<[I6400MSAFloatCompare], (instregex "^CMP_SAF_(D|S)$")>;
+ def : InstRW<[I6400MSAFloatCompare], (instregex "^CMP_SEQ_(D|S)$")>;
+ def : InstRW<[I6400MSAFloatCompare], (instregex "^CMP_SLE_(D|S)$")>;
+ def : InstRW<[I6400MSAFloatCompare], (instregex "^CMP_SLT_(D|S)$")>;
+ def : InstRW<[I6400MSAFloatCompare], (instregex "^CMP_SUEQ_(D|S)$")>;
+ def : InstRW<[I6400MSAFloatCompare], (instregex "^CMP_SULE_(D|S)$")>;
+ def : InstRW<[I6400MSAFloatCompare], (instregex "^CMP_SULT_(D|S)$")>;
+ def : InstRW<[I6400MSAFloatCompare], (instregex "^CMP_SUN_(D|S)$")>;
+
+ def I6400FPULongFloat : SchedWriteRes<[I6400FPULong]> { let Latency = 4; }
+ def : InstRW<[I6400FPULongFloat],
+ (instrs TRUNC_W_S, TRUNC_L_S, TRUNC_L_D64,
+ TRUNC_W_D32, TRUNC_W_D64, PseudoTRUNC_W_D,
+ PseudoTRUNC_W_D32, PseudoTRUNC_W_S,
+ ROUND_W_S, ROUND_L_S,
+ ROUND_L_D64, FLOOR_W_S, FLOOR_L_S, FLOOR_L_D64, CVT_D32_S,
+ CVT_D32_W, CVT_D64_W, CVT_D64_S, CVT_D64_L, CVT_L_S,
+ CVT_L_D64, CVT_S_W, CVT_S_D32, CVT_S_PU64, CVT_S_PL64,
+ CVT_S_L, CVT_S_D64, CVT_W_S, CEIL_W_S, CEIL_L_S,
+ CEIL_L_D64, FLOOR_W_D32, FLOOR_W_D64, CVT_W_D64, CVT_W_D32,
+ RINT_D, RINT_S, ROUND_W_D32, ROUND_W_D64,
+ CEIL_W_D32, CEIL_W_D64)>;
+
+ def I6400MSALongLogic1 : SchedWriteRes<[I6400FPULong]> {
+ let Latency = 1;
+ let ReleaseAtCycles = [2];
+ }
+ def : InstRW<[I6400MSALongLogic1], (instrs BMZ_V, BMZI_B, BMNZ_V, BMNZI_B,
+ INSERT_B, INSERT_H,
+ INSERT_W, INSERT_D, INSVE_B, INSVE_H,
+ INSVE_W, INSVE_D)>;
+ def : InstRW<[I6400MSALongLogic1], (instregex "^(BSEL_V|BSELI_B)$")>;
+ def : InstRW<[I6400MSALongLogic1],
+ (instregex "^BSEL_(H|W|D|FW|FD)_PSEUDO$")>;
+
+ def I6400MSALongLogic2 : SchedWriteRes<[I6400FPULong]> {
+ let Latency = 2;
+ let ReleaseAtCycles = [2];
+ }
+ def : InstRW<[I6400MSALongLogic2],
+ (instrs BINSL_B, BINSL_H, BINSL_W, BINSL_D, BINSLI_B, BINSLI_H,
+ BINSLI_W, BINSLI_D, BINSR_B, BINSR_H, BINSR_W, BINSR_D,
+ BINSRI_B, BINSRI_H, BINSRI_W, BINSRI_D, VSHF_B, VSHF_H,
+ VSHF_W, VSHF_D, SLD_B, SLD_H, SLD_W, SLD_D, SLDI_B, SLDI_H,
+ SLDI_W, SLDI_D, BSET_B, BSET_H, BSET_W, BSET_D, BSETI_B,
+ BSETI_H, BSETI_W, BSETI_D)>;
+
+ def I6400MSAMult : SchedWriteRes<[I6400FPULong]> {
+ let Latency = 5;
+ let ReleaseAtCycles = [2];
+ }
+ def : InstRW<[I6400MSAMult],
+ (instrs MADDV_B, MADDV_H, MADDV_W, MADDV_D, MSUBV_B, MSUBV_H,
+ MSUBV_W, MSUBV_D, MULV_B, MULV_H, MULV_W, MULV_D, DOTP_S_H,
+ DOTP_S_W, DOTP_S_D, DOTP_U_H, DOTP_U_W, DOTP_U_D, MUL_Q_H,
+ MUL_Q_W, MULR_Q_H, MULR_Q_W, MSUB_Q_H, MSUB_Q_W, MSUBR_Q_H,
+ MSUBR_Q_W, MADD_Q_H, MADD_Q_W, MADDR_Q_H, MADDR_Q_W)>;
+
+ def I6400MSALongFloat2 : SchedWriteRes<[I6400FPULong]> {
+ let Latency = 2;
+ let ReleaseAtCycles = [2];
+ }
+ def : InstRW<[I6400MSALongFloat2], (instrs FLOG2_W, FLOG2_D)>;
+
+ def I6400MSALongFloat4 : SchedWriteRes<[I6400FPULong]> {
+ let Latency = 4;
+ let ReleaseAtCycles = [2];
+ }
+ def : InstRW<[I6400MSALongFloat4],
+ (instrs FADD_W, FADD_D, FSUB_W, FSUB_D, FEXDO_H, FEXDO_W,
+ FEXUPL_W, FEXUPL_D, FEXUPR_W, FEXUPR_D, FFINT_S_W, FFINT_S_D,
+ FFINT_U_W, FFINT_U_D, FFQL_W, FFQL_D, FFQR_W, FFQR_D,
+ FTINT_S_W, FTINT_S_D, FTINT_U_W, FTINT_U_D, FTRUNC_S_W,
+ FTRUNC_S_D, FTRUNC_U_W, FTRUNC_U_D, FTQ_H, FTQ_W, FRINT_W,
+ FRINT_D, FADD_D32, FADD_D64)>;
+
+ def : InstRW<[I6400MSALongFloat4], (instregex "^PseudoCVT_(S|D32|D64)_(L|W)$")>;
+
+ def I6400MSALongFloat5 : SchedWriteRes<[I6400FPULong]> {
+ let Latency = 5;
+ let ReleaseAtCycles = [2];
+ }
+
+ def : InstRW<[I6400MSALongFloat5],
+ (instrs FMUL_W, FMUL_D, FEXP2_W, FEXP2_D, DPADD_S_H, DPADD_S_W,
+ DPADD_S_D, DPADD_U_H, DPADD_U_W, DPADD_U_D, DPSUB_S_H,
+ DPSUB_S_W, DPSUB_S_D, DPSUB_U_H, DPSUB_U_W, DPSUB_U_D)>;
+
+ def I6400MSALongFloat8 : SchedWriteRes<[I6400FPULong]> {
+ let Latency = 8;
+ let ReleaseAtCycles = [2];
+ }
+ def : InstRW<[I6400MSALongFloat8], (instrs FMADD_W, FMADD_D, FMSUB_W,
+ FMSUB_D)>;
+
+ def I6400MSALongFloatFuse : SchedWriteRes<[I6400FPULong]> {
+ let Latency = 8;
+ let ReleaseAtCycles = [1];
+ }
+ def : InstRW<[I6400MSALongFloatFuse], (instrs MSUBF_D, MSUBF_S,
+ MADDF_D, MADDF_S)>;
+
+ def I6400MSADivD : SchedWriteRes<[I6400FPULong, I6400FPUFloatL]> {
+ let Latency = 30;
+ let ReleaseAtCycles = [1, 29];
+ }
+ def : InstRW<[I6400MSADivD], (instrs FDIV_D)>;
+
+ def I6400MSADivW : SchedWriteRes<[I6400FPULong, I6400FPUFloatL]> {
+ let Latency = 22;
+ let ReleaseAtCycles = [1, 21];
+ }
+ def : InstRW<[I6400MSADivW], (instrs FDIV_W)>;
+
+ // Atomic instructions
+
+ // FIXME: Define `WriteAtomic` in the MipsSchedule.td and
+ // attach it to the Atomic2OpsPostRA, AtomicCmpSwapPostRA, ...
+ // classes. Then just define resources for the `WriteAtomic` in each
+ // machine models.
+ def I6400Atomic : ProcResource<1> { let BufferSize = 1; }
+ def I6400WriteAtomic : SchedWriteRes<[I6400Atomic]> { let Latency = 2; }
+
+ def : InstRW<[I6400WriteAtomic],
+ (instregex "^ATOMIC_SWAP_I(8|16|32|64)_POSTRA$")>;
+ def : InstRW<[I6400WriteAtomic],
+ (instregex "^ATOMIC_CMP_SWAP_I(8|16|32|64)_POSTRA$")>;
+ def : InstRW<[I6400WriteAtomic],
+ (instregex "^ATOMIC_LOAD_(ADD|SUB|AND|OR|XOR|NAND|MIN|MAX|UMIN|UMAX)"
+ "_I(8|16|32|64)_POSTRA$")>;
+}
diff --git a/llvm/test/tools/llvm-mca/Mips/i6400.s b/llvm/test/tools/llvm-mca/Mips/i6400.s
new file mode 100644
index 0000000000000..9b148a99a86ff
--- /dev/null
+++ b/llvm/test/tools/llvm-mca/Mips/i6400.s
@@ -0,0 +1,105 @@
+# NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
+# RUN: llvm-mca -mtriple=mips64el -mcpu=i6400 -timeline -iterations=1 < %s | FileCheck %s
+ .abicalls
+ .text
+ .globl i6400
+ .p2align 3
+ .type i6400, at function
+ .ent i6400
+i6400:
+ .set noat
+ lw $1, 0($sp)
+ lw $2, 8($sp)
+ move $3, $1
+ mul $3, $2, $3
+ sw $3, 8($sp)
+ div $2, $1
+ sw $2, 0($sp)
+ .end i6400
+
+# CHECK: Iterations: 1
+# CHECK-NEXT: Instructions: 7
+# CHECK-NEXT: Total Cycles: 41
+# CHECK-NEXT: Total uOps: 7
+
+# CHECK: Dispatch Width: 2
+# CHECK-NEXT: uOps Per Cycle: 0.17
+# CHECK-NEXT: IPC: 0.17
+# CHECK-NEXT: Block RThroughput: 4.0
+
+# CHECK: Instruction Info:
+# CHECK-NEXT: [1]: #uOps
+# CHECK-NEXT: [2]: Latency
+# CHECK-NEXT: [3]: RThroughput
+# CHECK-NEXT: [4]: MayLoad
+# CHECK-NEXT: [5]: MayStore
+# CHECK-NEXT: [6]: HasSideEffects (U)
+
+# CHECK: [1] [2] [3] [4] [5] [6] Instructions:
+# CHECK-NEXT: 1 3 1.00 * lw $1, 0($sp)
+# CHECK-NEXT: 1 3 1.00 * lw $2, 8($sp)
+# CHECK-NEXT: 1 1 0.50 move $3, $1
+# CHECK-NEXT: 1 4 1.00 mul $3, $2, $3
+# CHECK-NEXT: 1 1 1.00 * sw $3, 8($sp)
+# CHECK-NEXT: 1 32 1.00 div $2, $1
+# CHECK-NEXT: 1 1 1.00 * sw $2, 0($sp)
+
+# CHECK: Resources:
+# CHECK-NEXT: [0] - I6400AGEN
+# CHECK-NEXT: [1] - I6400Atomic
+# CHECK-NEXT: [2] - I6400CTRL
+# CHECK-NEXT: [3.0] - I6400FPU
+# CHECK-NEXT: [3.1] - I6400FPU
+# CHECK-NEXT: [3.2] - I6400FPU
+# CHECK-NEXT: [4] - I6400FPUApu
+# CHECK-NEXT: [5] - I6400FPUFloatL
+# CHECK-NEXT: [6] - I6400FPULong
+# CHECK-NEXT: [7] - I6400FPUShort
+# CHECK-NEXT: [8] - I6400IssueALU0
+# CHECK-NEXT: [9] - I6400IssueALU1
+# CHECK-NEXT: [10] - I6400IssueCTU
+# CHECK-NEXT: [11] - I6400IssueLSU
+# CHECK-NEXT: [12] - I6400MDU
+
+# CHECK: Resource pressure per iteration:
+# CHECK-NEXT: [0] [1] [2] [3.0] [3.1] [3.2] [4] [5] [6] [7] [8] [9] [10] [11] [12]
+# CHECK-NEXT: 4.00 - - - - - - - - - - 1.00 - 4.00 2.00
+
+# CHECK: Resource pressure by instruction:
+# CHECK-NEXT: [0] [1] [2] [3.0] [3.1] [3.2] [4] [5] [6] [7] [8] [9] [10] [11] [12] Instructions:
+# CHECK-NEXT: 1.00 - - - - - - - - - - - - 1.00 - lw $1, 0($sp)
+# CHECK-NEXT: 1.00 - - - - - - - - - - - - 1.00 - lw $2, 8($sp)
+# CHECK-NEXT: - - - - - - - - - - - 1.00 - - - move $3, $1
+# CHECK-NEXT: - - - - - - - - - - - - - - 1.00 mul $3, $2, $3
+# CHECK-NEXT: 1.00 - - - - - - - - - - - - 1.00 - sw $3, 8($sp)
+# CHECK-NEXT: - - - - - - - - - - - - - - 1.00 div $2, $1
+# CHECK-NEXT: 1.00 - - - - - - - - - - - - 1.00 - sw $2, 0($sp)
+
+# CHECK: Timeline view:
+# CHECK-NEXT: 0123456789 0123456789
+# CHECK-NEXT: Index 0123456789 0123456789 0
+
+# CHECK: [0,0] DeeeER . . . . . . . lw $1, 0($sp)
+# CHECK-NEXT: [0,1] D=eeeER . . . . . . . lw $2, 8($sp)
+# CHECK-NEXT: [0,2] .D==eER . . . . . . . move $3, $1
+# CHECK-NEXT: [0,3] .D===eeeeER . . . . . . mul $3, $2, $3
+# CHECK-NEXT: [0,4] . D======eER . . . . . . sw $3, 8($sp)
+# CHECK-NEXT: [0,5] . D===eeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeER. div $2, $1
+# CHECK-NEXT: [0,6] . D==================================eER sw $2, 0($sp)
+
+# CHECK: Average Wait times (based on the timeline view):
+# CHECK-NEXT: [0]: Executions
+# CHECK-NEXT: [1]: Average time spent waiting in a scheduler's queue
+# CHECK-NEXT: [2]: Average time spent waiting in a scheduler's queue while ready
+# CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
+
+# CHECK: [0] [1] [2] [3]
+# CHECK-NEXT: 0. 1 1.0 1.0 0.0 lw $1, 0($sp)
+# CHECK-NEXT: 1. 1 2.0 2.0 0.0 lw $2, 8($sp)
+# CHECK-NEXT: 2. 1 3.0 0.0 0.0 move $3, $1
+# CHECK-NEXT: 3. 1 4.0 0.0 0.0 mul $3, $2, $3
+# CHECK-NEXT: 4. 1 7.0 0.0 0.0 sw $3, 8($sp)
+# CHECK-NEXT: 5. 1 4.0 1.0 0.0 div $2, $1
+# CHECK-NEXT: 6. 1 35.0 0.0 0.0 sw $2, 0($sp)
+# CHECK-NEXT: 1 8.0 0.6 0.0 <total>
+
More information about the llvm-commits
mailing list