[RISCV] Prune unnecessary vector pseudo instructions. NFC

[craig.topper]

Summary

For .vf instructions, we don't need MF8 pseudos for f16. We don't
need MF8 or MF4 pseudos for f32. Or MF8, MF4, MF2 for f64.

[khchen]

LGTM.

llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp

static bool isScalarMoveInstr(const MachineInstr &MI) {
   case RISCV::PseudoVMV_S_X_MF4:
   case RISCV::PseudoVMV_S_X_MF8:
   case RISCV::PseudoVFMV_S_F16_M1:
   case RISCV::PseudoVFMV_S_F16_M2:
   case RISCV::PseudoVFMV_S_F16_M4:
   case RISCV::PseudoVFMV_S_F16_M8:
   case RISCV::PseudoVFMV_S_F16_MF2:
   case RISCV::PseudoVFMV_S_F16_MF4:
-  case RISCV::PseudoVFMV_S_F16_MF8:
   case RISCV::PseudoVFMV_S_F32_M1:
   case RISCV::PseudoVFMV_S_F32_M2:
   case RISCV::PseudoVFMV_S_F32_M4:
   case RISCV::PseudoVFMV_S_F32_M8:
   case RISCV::PseudoVFMV_S_F32_MF2:
-  case RISCV::PseudoVFMV_S_F32_MF4:
-  case RISCV::PseudoVFMV_S_F32_MF8:
   case RISCV::PseudoVFMV_S_F64_M1:
   case RISCV::PseudoVFMV_S_F64_M2:
   case RISCV::PseudoVFMV_S_F64_M4:
   case RISCV::PseudoVFMV_S_F64_M8:
-  case RISCV::PseudoVFMV_S_F64_MF2:
-  case RISCV::PseudoVFMV_S_F64_MF4:
-  case RISCV::PseudoVFMV_S_F64_MF8:
     return true;
   }
 }
 
 static VSETVLIInfo computeInfoForInstr(const MachineInstr &MI, uint64_t TSFlags,
                                        const MachineRegisterInfo *MRI) {
   VSETVLIInfo InstrInfo;
   unsigned NumOperands = MI.getNumExplicitOperands();

llvm/lib/Target/RISCV/RISCVInstrInfo.cpp

It = MBB.insert(It,
                                         RISCVII::MO_CALL));
   return It;
 }
 
 // clang-format off
 #define CASE_VFMA_OPCODE_COMMON(OP, TYPE, LMUL)                                \
   RISCV::PseudoV##OP##_##TYPE##_##LMUL
 
-#define CASE_VFMA_OPCODE_LMULS(OP, TYPE)                                       \
-  CASE_VFMA_OPCODE_COMMON(OP, TYPE, MF8):                                      \
-  case CASE_VFMA_OPCODE_COMMON(OP, TYPE, MF4):                                 \
-  case CASE_VFMA_OPCODE_COMMON(OP, TYPE, MF2):                                 \
-  case CASE_VFMA_OPCODE_COMMON(OP, TYPE, M1):                                  \
+#define CASE_VFMA_OPCODE_LMULS_M1(OP, TYPE)                                    \
+  CASE_VFMA_OPCODE_COMMON(OP, TYPE, M1):                                       \
   case CASE_VFMA_OPCODE_COMMON(OP, TYPE, M2):                                  \
   case CASE_VFMA_OPCODE_COMMON(OP, TYPE, M4):                                  \
   case CASE_VFMA_OPCODE_COMMON(OP, TYPE, M8)
 
+#define CASE_VFMA_OPCODE_LMULS_MF2(OP, TYPE)                                   \
+  CASE_VFMA_OPCODE_COMMON(OP, TYPE, MF2):                                      \
+  case CASE_VFMA_OPCODE_LMULS_M1(OP, TYPE)
+
+#define CASE_VFMA_OPCODE_LMULS_MF4(OP, TYPE)                                   \
+  CASE_VFMA_OPCODE_COMMON(OP, TYPE, MF4):                                      \
+  case CASE_VFMA_OPCODE_LMULS_MF2(OP, TYPE)
+
+#define CASE_VFMA_OPCODE_LMULS(OP, TYPE)                                       \
+  CASE_VFMA_OPCODE_COMMON(OP, TYPE, MF8):                                      \
+  case CASE_VFMA_OPCODE_LMULS_MF4(OP, TYPE)
+
 #define CASE_VFMA_SPLATS(OP)                                                   \
-  CASE_VFMA_OPCODE_LMULS(OP, VF16):                                            \
-  case CASE_VFMA_OPCODE_LMULS(OP, VF32):                                       \
-  case CASE_VFMA_OPCODE_LMULS(OP, VF64)
+  CASE_VFMA_OPCODE_LMULS_MF4(OP, VF16):                                        \
+  case CASE_VFMA_OPCODE_LMULS_MF2(OP, VF32):                                   \
+  case CASE_VFMA_OPCODE_LMULS_M1(OP, VF64)
 // clang-format on
 
 bool RISCVInstrInfo::findCommutedOpIndices(const MachineInstr &MI,
                                            unsigned &SrcOpIdx1,
                                            unsigned &SrcOpIdx2) const {
   const MCInstrDesc &Desc = MI.getDesc();
   if (!Desc.isCommutable())
     return false;
bool RISCVInstrInfo::findCommutedOpIndices(const MachineInstr &MI,
   return TargetInstrInfo::findCommutedOpIndices(MI, SrcOpIdx1, SrcOpIdx2);
 }
 
 #define CASE_VFMA_CHANGE_OPCODE_COMMON(OLDOP, NEWOP, TYPE, LMUL)               \
   case RISCV::PseudoV##OLDOP##_##TYPE##_##LMUL:                                \
     Opc = RISCV::PseudoV##NEWOP##_##TYPE##_##LMUL;                             \
     break;
 
-#define CASE_VFMA_CHANGE_OPCODE_LMULS(OLDOP, NEWOP, TYPE)                      \
-  CASE_VFMA_CHANGE_OPCODE_COMMON(OLDOP, NEWOP, TYPE, MF8)                      \
-  CASE_VFMA_CHANGE_OPCODE_COMMON(OLDOP, NEWOP, TYPE, MF4)                      \
-  CASE_VFMA_CHANGE_OPCODE_COMMON(OLDOP, NEWOP, TYPE, MF2)                      \
+#define CASE_VFMA_CHANGE_OPCODE_LMULS_M1(OLDOP, NEWOP, TYPE)                   \
   CASE_VFMA_CHANGE_OPCODE_COMMON(OLDOP, NEWOP, TYPE, M1)                       \
   CASE_VFMA_CHANGE_OPCODE_COMMON(OLDOP, NEWOP, TYPE, M2)                       \
   CASE_VFMA_CHANGE_OPCODE_COMMON(OLDOP, NEWOP, TYPE, M4)                       \
   CASE_VFMA_CHANGE_OPCODE_COMMON(OLDOP, NEWOP, TYPE, M8)
 
+#define CASE_VFMA_CHANGE_OPCODE_LMULS_MF2(OLDOP, NEWOP, TYPE)                  \
+  CASE_VFMA_CHANGE_OPCODE_COMMON(OLDOP, NEWOP, TYPE, MF2)                      \
+  CASE_VFMA_CHANGE_OPCODE_LMULS_M1(OLDOP, NEWOP, TYPE)
+
+#define CASE_VFMA_CHANGE_OPCODE_LMULS_MF4(OLDOP, NEWOP, TYPE)                  \
+  CASE_VFMA_CHANGE_OPCODE_COMMON(OLDOP, NEWOP, TYPE, MF4)                      \
+  CASE_VFMA_CHANGE_OPCODE_LMULS_MF2(OLDOP, NEWOP, TYPE)
+
+#define CASE_VFMA_CHANGE_OPCODE_LMULS(OLDOP, NEWOP, TYPE)                      \
+  CASE_VFMA_CHANGE_OPCODE_COMMON(OLDOP, NEWOP, TYPE, MF8)                      \
+  CASE_VFMA_CHANGE_OPCODE_LMULS_MF4(OLDOP, NEWOP, TYPE)
+
 #define CASE_VFMA_CHANGE_OPCODE_SPLATS(OLDOP, NEWOP)                           \
-  CASE_VFMA_CHANGE_OPCODE_LMULS(OLDOP, NEWOP, VF16)                            \
-  CASE_VFMA_CHANGE_OPCODE_LMULS(OLDOP, NEWOP, VF32)                            \
-  CASE_VFMA_CHANGE_OPCODE_LMULS(OLDOP, NEWOP, VF64)
+  CASE_VFMA_CHANGE_OPCODE_LMULS_MF4(OLDOP, NEWOP, VF16)                        \
+  CASE_VFMA_CHANGE_OPCODE_LMULS_MF2(OLDOP, NEWOP, VF32)                        \
+  CASE_VFMA_CHANGE_OPCODE_LMULS_M1(OLDOP, NEWOP, VF64)
 
 MachineInstr *RISCVInstrInfo::commuteInstructionImpl(MachineInstr &MI,
                                                      bool NewMI,
                                                      unsigned OpIdx1,
                                                      unsigned OpIdx2) const {
   auto cloneIfNew = [NewMI](MachineInstr &MI) -> MachineInstr & {
     if (NewMI)
       return *MI.getParent()->getParent()->CloneMachineInstr(&MI);

llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td

 defvar MxListVF2 = [V_MF4, V_MF2, V_M1, V_M2, V_M4, V_M8];
 
 // Use for zext/sext.vf4
 defvar MxListVF4 = [V_MF2, V_M1, V_M2, V_M4, V_M8];
 
 // Use for zext/sext.vf8
 defvar MxListVF8 = [V_M1, V_M2, V_M4, V_M8];
 
-class FPR_Info<RegisterClass regclass, string fx> {
+class MxSet<int eew> {
+  list<LMULInfo> m = !cond(!eq(eew, 8) : [V_MF8, V_MF4, V_MF2, V_M1, V_M2, V_M4, V_M8],
+                           !eq(eew, 16) : [V_MF4, V_MF2, V_M1, V_M2, V_M4, V_M8],
+                           !eq(eew, 32) : [V_MF2, V_M1, V_M2, V_M4, V_M8],
+                           !eq(eew, 64) : [V_M1, V_M2, V_M4, V_M8]);
+}
+
+class FPR_Info<RegisterClass regclass, string fx, list<LMULInfo> mxlist> {
   RegisterClass fprclass = regclass;
   string FX = fx;
+  list<LMULInfo> MxList = mxlist;
 }
 
-def SCALAR_F16 : FPR_Info<FPR16, "F16">;
-def SCALAR_F32 : FPR_Info<FPR32, "F32">;
-def SCALAR_F64 : FPR_Info<FPR64, "F64">;
+def SCALAR_F16 : FPR_Info<FPR16, "F16", MxSet<16>.m>;
+def SCALAR_F32 : FPR_Info<FPR32, "F32", MxSet<32>.m>;
+def SCALAR_F64 : FPR_Info<FPR64, "F64", MxSet<64>.m>;
 
 defvar FPList = [SCALAR_F16, SCALAR_F32, SCALAR_F64];
 
 // Used for widening instructions. It excludes F64.
 defvar FPListW = [SCALAR_F16, SCALAR_F32];
 
-class MxSet<int eew> {
-  list<LMULInfo> m = !cond(!eq(eew, 8) : [V_MF8, V_MF4, V_MF2, V_M1, V_M2, V_M4, V_M8],
-                           !eq(eew, 16) : [V_MF4, V_MF2, V_M1, V_M2, V_M4, V_M8],
-                           !eq(eew, 32) : [V_MF2, V_M1, V_M2, V_M4, V_M8],
-                           !eq(eew, 64) : [V_M1, V_M2, V_M4, V_M8]);
-}
-
 class NFSet<LMULInfo m> {
   list<int> L = !cond(!eq(m.value, V_M8.value): [],
                       !eq(m.value, V_M4.value): [2],
                       !eq(m.value, V_M2.value): [2, 3, 4],
                       true: [2, 3, 4, 5, 6, 7, 8]);
 }
 
 class log2<int num> {

 
 multiclass VPseudoVSLD1_VX<string Constraint = ""> {
   foreach m = MxList in
     defm "_VX" : VPseudoBinary<m.vrclass, m.vrclass, GPR, m, Constraint>,
                  Sched<[WriteVISlide1X, ReadVISlideV, ReadVISlideX, ReadVMask]>;
 }
 
 multiclass VPseudoBinaryV_VF<string Constraint = ""> {
-  foreach m = MxList in
-    foreach f = FPList in
+    foreach m = f.MxList in
       defm "_V" # f.FX : VPseudoBinary<m.vrclass, m.vrclass,
                                        f.fprclass, m, Constraint>;
 }
 
 multiclass VPseudoVSLD1_VF<string Constraint = ""> {
-  foreach m = MxList in
-    foreach f = FPList in
+    foreach m = f.MxList in
       defm "_V" # f.FX :
         VPseudoBinary<m.vrclass, m.vrclass, f.fprclass, m, Constraint>,
         Sched<[WriteVFSlide1F, ReadVFSlideV, ReadVFSlideF, ReadVMask]>;
 }
 
 multiclass VPseudoBinaryV_VI<Operand ImmType = simm5, string Constraint = ""> {
   foreach m = MxList in
     defm _VI : VPseudoBinary<m.vrclass, m.vrclass, ImmType, m, Constraint>;

 
 multiclass VPseudoBinaryW_VX {
   foreach m = MxListW in
     defm "_VX" : VPseudoBinary<m.wvrclass, m.vrclass, GPR, m,
                                "@earlyclobber $rd">;
 }
 
 multiclass VPseudoBinaryW_VF {
-  foreach m = MxListW in
-    foreach f = FPListW in
+    foreach m = f.MxList in
       defm "_V" # f.FX : VPseudoBinary<m.wvrclass, m.vrclass,
                                        f.fprclass, m,
                                        "@earlyclobber $rd">;
 }
 
 multiclass VPseudoBinaryW_WV {
   foreach m = MxListW in {
     defm _WV : VPseudoBinary<m.wvrclass, m.wvrclass, m.vrclass, m,
                              "@earlyclobber $rd">;
     defm _WV : VPseudoTiedBinary<m.wvrclass, m.vrclass, m,
                                  "@earlyclobber $rd">;
   }
 }
 
 multiclass VPseudoBinaryW_WX {
   foreach m = MxListW in
     defm "_WX" : VPseudoBinary<m.wvrclass, m.wvrclass, GPR, m>;
 }
 
 multiclass VPseudoBinaryW_WF {
-  foreach m = MxListW in
-    foreach f = FPListW in
+    foreach m = f.MxList in
       defm "_W" # f.FX : VPseudoBinary<m.wvrclass, m.wvrclass,
                                        f.fprclass, m>;
 }
 
 // Narrowing instructions like vnsrl/vnsra/vnclip(u) don't need @earlyclobber
 // if the source and destination have an LMUL<=1. This matches this overlap
 // exception from the spec.
 // "The destination EEW is smaller than the source EEW and the overlap is in the
foreach m = MxList in
     def "_VX" # !if(CarryIn, "M", "") # "_" # m.MX :
       VPseudoBinaryCarryIn<!if(CarryOut, VR,
                            !if(!and(CarryIn, !not(CarryOut)),
                                GetVRegNoV0<m.vrclass>.R, m.vrclass)),
                            m.vrclass, GPR, m, CarryIn, Constraint>;
 }
 
 multiclass VPseudoVMRG_FM {
-  foreach m = MxList in
-    foreach f = FPList in
+    foreach m = f.MxList in
       def "_V" # f.FX # "M_" # m.MX :
         VPseudoBinaryCarryIn<GetVRegNoV0<m.vrclass>.R,
                              m.vrclass, f.fprclass, m, /*CarryIn=*/1, "">,
         Sched<[WriteVFMergeV, ReadVFMergeV, ReadVFMergeF, ReadVMask]>;
 }
 
 multiclass VPseudoBinaryV_IM<bit CarryOut = 0, bit CarryIn = 1,
                              string Constraint = ""> {
let VLMul = m.value in {
                          Sched<[WriteVIMovX, ReadVIMovX]>;
       def "_I_" # m.MX : VPseudoUnaryNoDummyMask<m.vrclass, simm5>,
                          Sched<[WriteVIMovI]>;
     }
   }
 }
 
 multiclass VPseudoVMV_F {
-  foreach m = MxList in {
-    foreach f = FPList in {
+    foreach m = f.MxList in {
       let VLMul = m.value in {
         def "_" # f.FX # "_" # m.MX :
           VPseudoUnaryNoDummyMask<m.vrclass, f.fprclass>,
           Sched<[WriteVFMovV, ReadVFMovF]>;
       }
     }
   }
 }

 multiclass VPseudoBinaryM_VX {
   foreach m = MxList in
     defm "_VX" :
       VPseudoBinaryM<VR, m.vrclass, GPR, m,
                      !if(!ge(m.octuple, 16), "@earlyclobber $rd", "")>;
 }
 
 multiclass VPseudoBinaryM_VF {
-  foreach m = MxList in
-    foreach f = FPList in
+    foreach m = f.MxList in
       defm "_V" # f.FX :
         VPseudoBinaryM<VR, m.vrclass, f.fprclass, m,
                        !if(!ge(m.octuple, 16), "@earlyclobber $rd", "")>;
 }
 
 multiclass VPseudoBinaryM_VI {
   foreach m = MxList in
     defm _VI : VPseudoBinaryM<VR, m.vrclass, simm5, m,

 
 multiclass VPseudoTernaryV_VX_AAXA<string Constraint = ""> {
   foreach m = MxList in
     defm "_VX" : VPseudoTernaryWithPolicy<m.vrclass, GPR, m.vrclass, m,
                                           Constraint, /*Commutable*/1>;
 }
 
 multiclass VPseudoTernaryV_VF_AAXA<string Constraint = ""> {
-  foreach m = MxList in
-    foreach f = FPList in
+    foreach m = f.MxList in
       defm "_V" # f.FX : VPseudoTernaryWithPolicy<m.vrclass, f.fprclass,
                                                   m.vrclass, m, Constraint,
                                                   /*Commutable*/1>;
 }
 
 multiclass VPseudoTernaryW_VV {
   defvar constraint = "@earlyclobber $rd";
   foreach m = MxListW in
     defm _VV : VPseudoTernaryWithPolicy<m.wvrclass, m.vrclass, m.vrclass, m,
                                         constraint>;
 }
 
 multiclass VPseudoTernaryW_VX {
   defvar constraint = "@earlyclobber $rd";
   foreach m = MxListW in
     defm "_VX" : VPseudoTernaryWithPolicy<m.wvrclass, GPR, m.vrclass, m,
                                           constraint>;
 }
 
 multiclass VPseudoTernaryW_VF {
   defvar constraint = "@earlyclobber $rd";
-  foreach m = MxListW in
-    foreach f = FPListW in
+    foreach m = f.MxList in
       defm "_V" # f.FX : VPseudoTernaryWithPolicy<m.wvrclass, f.fprclass,
                                                   m.vrclass, m, constraint>;
 }
 
 multiclass VPseudoTernaryV_VI<Operand ImmType = simm5, string Constraint = ""> {
   foreach m = MxList in
     defm _VI : VPseudoTernary<m.vrclass, m.vrclass, ImmType, m, Constraint>;
 }

 } // Predicates = [HasVInstructions]
 
 //===----------------------------------------------------------------------===//
 // 17.2. Floating-Point Scalar Move Instructions
 //===----------------------------------------------------------------------===//
 
 let Predicates = [HasVInstructionsAnyF] in {
 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
-  foreach m = MxList in {
-    foreach f = FPList in {
+    foreach m = f.MxList in {
       let VLMul = m.value in {
         let HasSEWOp = 1, BaseInstr = VFMV_F_S in
         def "PseudoVFMV_" # f.FX # "_S_" # m.MX :
           Pseudo<(outs f.fprclass:$rd),
                  (ins m.vrclass:$rs2, ixlenimm:$sew), []>,
           Sched<[WriteVFMovVF, ReadVFMovVF]>,
           RISCVVPseudo;
         let HasVLOp = 1, HasSEWOp = 1, BaseInstr = VFMV_S_F,