[RISCV] Optimize immediate materialisation with SH*ADD

Use SH1ADD/SH2ADD/SH3ADD along with LUI+ADDI to compose int32*3,
int32*5 and int32*9.

Reviewed By: craig.topper, luismarques

Differential Revision: https://reviews.llvm.org/D111484

llvm/lib/Target/RISCV/AsmParser/RISCVAsmParser.cpp

@@ -2269,6 +2269,11 @@ void RISCVAsmParser::emitLoadImm(MCRegister DestReg, int64_t Value,
                               .addReg(DestReg)
                               .addReg(SrcReg)
                               .addReg(RISCV::X0));
+    } else if (Inst.Opc == RISCV::SH1ADD || Inst.Opc == RISCV::SH2ADD ||
+               Inst.Opc == RISCV::SH3ADD) {
+      emitToStreamer(
+          Out, MCInstBuilder(Inst.Opc).addReg(DestReg).addReg(SrcReg).addReg(
+                   SrcReg));
     } else {
       emitToStreamer(
           Out, MCInstBuilder(Inst.Opc).addReg(DestReg).addReg(SrcReg).addImm(

llvm/lib/Target/RISCV/MCTargetDesc/RISCVMatInt.cpp

@@ -250,6 +250,33 @@ InstSeq generateInstSeq(int64_t Val, const FeatureBitset &ActiveFeatures) {
     }
   }
 
+  // Perform optimization with SH*ADD in the Zba extension.
+  if (Res.size() > 2 && ActiveFeatures[RISCV::FeatureStdExtZba]) {
+    assert(ActiveFeatures[RISCV::Feature64Bit] &&
+           "Expected RV32 to only need 2 instructions");
+    int64_t Div = 0;
+    unsigned Opc = 0;
+    RISCVMatInt::InstSeq TmpSeq;
+    // Select the opcode and divisor.
+    if ((Val % 3) == 0 && isInt<32>(Val / 3)) {
+      Div = 3;
+      Opc = RISCV::SH1ADD;
+    } else if ((Val % 5) == 0 && isInt<32>(Val / 5)) {
+      Div = 5;
+      Opc = RISCV::SH2ADD;
+    } else if ((Val % 9) == 0 && isInt<32>(Val / 9)) {
+      Div = 9;
+      Opc = RISCV::SH3ADD;
+    }
+    // Build the new instruction sequence.
+    if (Div > 0) {
+      generateInstSeqImpl(Val / Div, ActiveFeatures, TmpSeq);
+      TmpSeq.push_back(RISCVMatInt::Inst(Opc, 0));
+      if (TmpSeq.size() < Res.size())
+        Res = TmpSeq;
+    }
+  }
+
   return Res;
 }
 

llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp

@@ -140,6 +140,9 @@ static SDNode *selectImm(SelectionDAG *CurDAG, const SDLoc &DL, int64_t Imm,
     else if (Inst.Opc == RISCV::ADDUW)
       Result = CurDAG->getMachineNode(RISCV::ADDUW, DL, XLenVT, SrcReg,
                                       CurDAG->getRegister(RISCV::X0, XLenVT));
+    else if (Inst.Opc == RISCV::SH1ADD || Inst.Opc == RISCV::SH2ADD ||
+             Inst.Opc == RISCV::SH3ADD)
+      Result = CurDAG->getMachineNode(Inst.Opc, DL, XLenVT, SrcReg, SrcReg);
     else
       Result = CurDAG->getMachineNode(Inst.Opc, DL, XLenVT, SrcReg, SDImm);
 

llvm/lib/Target/RISCV/RISCVInstrInfo.cpp

@@ -458,6 +458,12 @@ void RISCVInstrInfo::movImm(MachineBasicBlock &MBB,
           .addReg(SrcReg, RegState::Kill)
           .addReg(RISCV::X0)
           .setMIFlag(Flag);
+    } else if (Inst.Opc == RISCV::SH1ADD || Inst.Opc == RISCV::SH2ADD ||
+               Inst.Opc == RISCV::SH3ADD) {
+      BuildMI(MBB, MBBI, DL, get(Inst.Opc), Result)
+          .addReg(SrcReg, RegState::Kill)
+          .addReg(SrcReg, RegState::Kill)
+          .setMIFlag(Flag);
     } else {
       BuildMI(MBB, MBBI, DL, get(Inst.Opc), Result)
           .addReg(SrcReg, RegState::Kill)

llvm/test/CodeGen/RISCV/imm.ll

@@ -959,10 +959,9 @@ define i64 @imm_neg_5372288229() {
 ;
 ; RV64IZBA-LABEL: imm_neg_5372288229:
 ; RV64IZBA:       # %bb.0:
-; RV64IZBA-NEXT:    lui a0, 1048416
+; RV64IZBA-NEXT:    lui a0, 611378
-; RV64IZBA-NEXT:    addiw a0, a0, -437
+; RV64IZBA-NEXT:    addiw a0, a0, 265
-; RV64IZBA-NEXT:    slli a0, a0, 13
+; RV64IZBA-NEXT:    sh1add a0, a0, a0
-; RV64IZBA-NEXT:    addi a0, a0, 795
 ; RV64IZBA-NEXT:    ret
 ;
 ; RV64IZBS-LABEL: imm_neg_5372288229:
@@ -992,10 +991,9 @@ define i64 @imm_8953813715() {
 ;
 ; RV64IZBA-LABEL: imm_8953813715:
 ; RV64IZBA:       # %bb.0:
-; RV64IZBA-NEXT:    lui a0, 267
+; RV64IZBA-NEXT:    lui a0, 437198
-; RV64IZBA-NEXT:    addiw a0, a0, -637
+; RV64IZBA-NEXT:    addiw a0, a0, -265
-; RV64IZBA-NEXT:    slli a0, a0, 13
+; RV64IZBA-NEXT:    sh2add a0, a0, a0
-; RV64IZBA-NEXT:    addi a0, a0, -1325
 ; RV64IZBA-NEXT:    ret
 ;
 ; RV64IZBS-LABEL: imm_8953813715:
@@ -1025,10 +1023,9 @@ define i64 @imm_neg_8953813715() {
 ;
 ; RV64IZBA-LABEL: imm_neg_8953813715:
 ; RV64IZBA:       # %bb.0:
-; RV64IZBA-NEXT:    lui a0, 1048309
+; RV64IZBA-NEXT:    lui a0, 611378
-; RV64IZBA-NEXT:    addiw a0, a0, 637
+; RV64IZBA-NEXT:    addiw a0, a0, 265
-; RV64IZBA-NEXT:    slli a0, a0, 13
+; RV64IZBA-NEXT:    sh2add a0, a0, a0
-; RV64IZBA-NEXT:    addi a0, a0, 1325
 ; RV64IZBA-NEXT:    ret
 ;
 ; RV64IZBS-LABEL: imm_neg_8953813715:
@@ -1058,10 +1055,9 @@ define i64 @imm_16116864687() {
 ;
 ; RV64IZBA-LABEL: imm_16116864687:
 ; RV64IZBA:       # %bb.0:
-; RV64IZBA-NEXT:    lui a0, 961
+; RV64IZBA-NEXT:    lui a0, 437198
-; RV64IZBA-NEXT:    addiw a0, a0, -1475
+; RV64IZBA-NEXT:    addiw a0, a0, -265
-; RV64IZBA-NEXT:    slli a0, a0, 12
+; RV64IZBA-NEXT:    sh3add a0, a0, a0
-; RV64IZBA-NEXT:    addi a0, a0, 1711
 ; RV64IZBA-NEXT:    ret
 ;
 ; RV64IZBS-LABEL: imm_16116864687:
@@ -1092,10 +1088,9 @@ define i64 @imm_neg_16116864687() {
 ;
 ; RV64IZBA-LABEL: imm_neg_16116864687:
 ; RV64IZBA:       # %bb.0:
-; RV64IZBA-NEXT:    lui a0, 1047615
+; RV64IZBA-NEXT:    lui a0, 611378
-; RV64IZBA-NEXT:    addiw a0, a0, 1475
+; RV64IZBA-NEXT:    addiw a0, a0, 265
-; RV64IZBA-NEXT:    slli a0, a0, 12
+; RV64IZBA-NEXT:    sh3add a0, a0, a0
-; RV64IZBA-NEXT:    addi a0, a0, -1711
 ; RV64IZBA-NEXT:    ret
 ;
 ; RV64IZBS-LABEL: imm_neg_16116864687:
@@ -1447,10 +1442,9 @@ define i64 @imm_neg_2863311530() {
 ;
 ; RV64IZBA-LABEL: imm_neg_2863311530:
 ; RV64IZBA:       # %bb.0:
-; RV64IZBA-NEXT:    lui a0, 1048405
+; RV64IZBA-NEXT:    lui a0, 908766
-; RV64IZBA-NEXT:    addiw a0, a0, 1365
+; RV64IZBA-NEXT:    addiw a0, a0, -546
-; RV64IZBA-NEXT:    slli a0, a0, 12
+; RV64IZBA-NEXT:    sh2add a0, a0, a0
-; RV64IZBA-NEXT:    addi a0, a0, 1366
 ; RV64IZBA-NEXT:    ret
 ;
 ; RV64IZBS-LABEL: imm_neg_2863311530:

llvm/test/MC/RISCV/rv64zba-aliases-valid.s

@@ -46,3 +46,43 @@ li x5, 0xbbbbb0007bb
 # CHECK-S-OBJ: lui t0, 768955
 # CHECK-S-OBJ-NEXT: slli.uw t0, t0, 4
 li x5, 0xbbbbb0000
+
+# CHECK-S-OBJ-NOALIAS: lui t1, 611378
+# CHECK-S-OBJ-NOALIAS-NEXT: addiw t1, t1, 265
+# CHECK-S-OBJ-NOALIAS-NEXT: sh1add t1, t1, t1
+# CHECK-S-OBJ: lui t1, 611378
+# CHECK-S-OBJ-NEXT: addiw t1, t1, 265
+# CHECK-S-OBJ-NEXT: sh1add t1, t1, t1
+li x6, -5372288229
+
+# CHECK-S-OBJ-NOALIAS: lui t1, 437198
+# CHECK-S-OBJ-NOALIAS-NEXT: addiw t1, t1, -265
+# CHECK-S-OBJ-NOALIAS-NEXT: sh2add t1, t1, t1
+# CHECK-S-OBJ: lui t1, 437198
+# CHECK-S-OBJ-NEXT: addiw t1, t1, -265
+# CHECK-S-OBJ-NEXT: sh2add t1, t1, t1
+li x6, 8953813715
+
+# CHECK-S-OBJ-NOALIAS: lui t1, 611378
+# CHECK-S-OBJ-NOALIAS-NEXT: addiw t1, t1, 265
+# CHECK-S-OBJ-NOALIAS-NEXT: sh2add t1, t1, t1
+# CHECK-S-OBJ: lui t1, 611378
+# CHECK-S-OBJ-NEXT: addiw t1, t1, 265
+# CHECK-S-OBJ-NEXT: sh2add t1, t1, t1
+li x6, -8953813715
+
+# CHECK-S-OBJ-NOALIAS: lui t1, 437198
+# CHECK-S-OBJ-NOALIAS-NEXT: addiw t1, t1, -265
+# CHECK-S-OBJ-NOALIAS-NEXT: sh3add t1, t1, t1
+# CHECK-S-OBJ: lui t1, 437198
+# CHECK-S-OBJ-NEXT: addiw t1, t1, -265
+# CHECK-S-OBJ-NEXT: sh3add t1, t1, t1
+li x6, 16116864687
+
+# CHECK-S-OBJ-NOALIAS: lui t1, 611378
+# CHECK-S-OBJ-NOALIAS-NEXT: addiw t1, t1, 265
+# CHECK-S-OBJ-NOALIAS-NEXT: sh3add t1, t1, t1
+# CHECK-S-OBJ: lui t1, 611378
+# CHECK-S-OBJ-NEXT: addiw t1, t1, 265
+# CHECK-S-OBJ-NEXT: sh3add t1, t1, t1
+li x6, -16116864687