Fix legalization of G_UMULO, G_SMULO still doesn't work.

jacobly0 · Mar 13, 2022 · 6d5b492 · 6d5b492
1 parent b165d7b
commit 6d5b492
Show file tree

Hide file tree

Showing 5 changed files with 74 additions and 15 deletions.
diff --git a/llvm/lib/CodeGen/GlobalISel/CSEMIRBuilder.cpp b/llvm/lib/CodeGen/GlobalISel/CSEMIRBuilder.cpp
@@ -185,7 +185,11 @@ MachineInstrBuilder CSEMIRBuilder::buildInstr(unsigned Opc,
   case TargetOpcode::G_UDIV:
   case TargetOpcode::G_SDIV:
   case TargetOpcode::G_UREM:
-  case TargetOpcode::G_SREM: {
+  case TargetOpcode::G_SREM:
+  case TargetOpcode::G_SMIN:
+  case TargetOpcode::G_SMAX:
+  case TargetOpcode::G_UMIN:
+  case TargetOpcode::G_UMAX: {
     // Try to constant fold these.
     assert(SrcOps.size() == 2 && "Invalid sources");
     assert(DstOps.size() == 1 && "Invalid dsts");

diff --git a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
@@ -3444,7 +3444,6 @@ LegalizerHelper::lower(MachineInstr &MI, unsigned TypeIdx, LLT LowerHintTy) {
     Observer.changedInstr(MI);
 
     auto HiPart = MIRBuilder.buildInstr(Opcode, {Ty}, {LHS, RHS});
-    auto Zero = MIRBuilder.buildConstant(Ty, 0);
 
     // Move insert point forward so we can use the Res register if needed.
     MIRBuilder.setInsertPt(MIRBuilder.getMBB(), ++MIRBuilder.getInsertPt());
@@ -3456,6 +3455,7 @@ LegalizerHelper::lower(MachineInstr &MI, unsigned TypeIdx, LLT LowerHintTy) {
       auto Shifted = MIRBuilder.buildAShr(Ty, Res, ShiftAmt);
       MIRBuilder.buildICmp(CmpInst::ICMP_NE, Overflow, HiPart, Shifted);
     } else {
+      auto Zero = MIRBuilder.buildConstant(Ty, 0);
       MIRBuilder.buildICmp(CmpInst::ICMP_NE, Overflow, HiPart, Zero);
     }
     return Legalized;

diff --git a/llvm/lib/CodeGen/GlobalISel/Utils.cpp b/llvm/lib/CodeGen/GlobalISel/Utils.cpp
@@ -526,21 +526,29 @@ Optional<APInt> llvm::ConstantFoldBinOp(unsigned Opcode, const Register Op1,
   case TargetOpcode::G_XOR:
     return C1 ^ C2;
   case TargetOpcode::G_UDIV:
-    if (!C2.getBoolValue())
+    if (C2.isZero())
       break;
     return C1.udiv(C2);
   case TargetOpcode::G_SDIV:
-    if (!C2.getBoolValue())
+    if (C2.isZero())
       break;
     return C1.sdiv(C2);
   case TargetOpcode::G_UREM:
-    if (!C2.getBoolValue())
+    if (C2.isZero())
       break;
     return C1.urem(C2);
   case TargetOpcode::G_SREM:
-    if (!C2.getBoolValue())
+    if (C2.isZero())
       break;
     return C1.srem(C2);
+  case TargetOpcode::G_SMIN:
+    return APIntOps::smin(C1, C2);
+  case TargetOpcode::G_SMAX:
+    return APIntOps::smax(C1, C2);
+  case TargetOpcode::G_UMIN:
+    return APIntOps::umin(C1, C2);
+  case TargetOpcode::G_UMAX:
+    return APIntOps::umax(C1, C2);
   }
 
   return None;

diff --git a/llvm/lib/Target/Z80/GISel/Z80LegalizerInfo.cpp b/llvm/lib/Target/Z80/GISel/Z80LegalizerInfo.cpp
@@ -135,7 +135,8 @@ Z80LegalizerInfo::Z80LegalizerInfo(const Z80Subtarget &STI,
       .clampScalar(1, s8, s8)
       .clampScalar(0, s8, s64);
 
-  getActionDefinitionsBuilder({G_FSHL, G_FSHR, G_MEMCPY, G_MEMMOVE, G_MEMSET})
+  getActionDefinitionsBuilder(
+      {G_FSHL, G_FSHR, G_UMULO, G_MEMCPY, G_MEMMOVE, G_MEMSET})
       .custom();
 
   getActionDefinitionsBuilder({G_INTRINSIC_TRUNC,
@@ -179,8 +180,6 @@ Z80LegalizerInfo::Z80LegalizerInfo(const Z80Subtarget &STI,
   getActionDefinitionsBuilder(G_INTRINSIC_LRINT)
       .libcallFor({{s32, s32}, {s32, s64}});
 
-  getActionDefinitionsBuilder(G_FPOWI).lower();
-
   getActionDefinitionsBuilder(G_FCOPYSIGN)
       .libcallFor({{s32, s32}, {s64, s64}});
 
@@ -225,12 +224,26 @@ Z80LegalizerInfo::Z80LegalizerInfo(const Z80Subtarget &STI,
       .legalForCartesianProduct(LegalTypes, {s1})
       .clampScalar(0, s8, sMax);
 
-  getActionDefinitionsBuilder({G_ABS, G_DYN_STACKALLOC, G_SEXT_INREG,
-                               G_CTLZ_ZERO_UNDEF, G_CTTZ_ZERO_UNDEF,
-                               G_CTLZ, G_CTTZ, G_BSWAP,
-                               G_SMULO, G_UMULO, G_SMULH, G_UMULH,
-                               G_SMIN, G_SMAX, G_UMIN, G_UMAX,
-                               G_UADDSAT, G_SADDSAT, G_USUBSAT, G_SSUBSAT})
+  getActionDefinitionsBuilder({G_ABS,
+                               G_DYN_STACKALLOC,
+                               G_SEXT_INREG,
+                               G_CTLZ_ZERO_UNDEF,
+                               G_CTTZ_ZERO_UNDEF,
+                               G_CTLZ,
+                               G_CTTZ,
+                               G_BSWAP,
+                               G_SMULO,
+                               G_SMULH,
+                               G_UMULH,
+                               G_SMIN,
+                               G_SMAX,
+                               G_UMIN,
+                               G_UMAX,
+                               G_UADDSAT,
+                               G_SADDSAT,
+                               G_USUBSAT,
+                               G_SSUBSAT,
+                               G_FPOWI})
       .lower();
 
   getActionDefinitionsBuilder(G_CTPOP)
@@ -276,6 +289,8 @@ LegalizerHelper::LegalizeResult Z80LegalizerInfo::legalizeCustomMaybeLegal(
   case G_ICMP:
   case G_FCMP:
     return legalizeCompare(Helper, MI);
+  case G_UMULO:
+    return legalizeMultiplyWithOverflow(Helper, MI);
   case G_MEMCPY:
   case G_MEMMOVE:
   case G_MEMSET:
@@ -554,6 +569,36 @@ Z80LegalizerInfo::legalizeCompare(LegalizerHelper &Helper,
   return LegalizerHelper::Legalized;
 }
 
+LegalizerHelper::LegalizeResult
+Z80LegalizerInfo::legalizeMultiplyWithOverflow(LegalizerHelper &Helper,
+                                               MachineInstr &MI) const {
+  MachineIRBuilder &MIRBuilder = Helper.MIRBuilder;
+  MIRBuilder.setInstrAndDebugLoc(MI);
+  MachineRegisterInfo &MRI = *MIRBuilder.getMRI();
+
+  assert(MI.getOpcode() == G_UMULO && "Unexpected opcode");
+
+  Register MulReg = MI.getOperand(0).getReg();
+  Register OverflowReg = MI.getOperand(1).getReg();
+  LLT Ty = MRI.getType(MulReg);
+
+  Register LHSReg = MI.getOperand(2).getReg();
+  Register RHSReg = MI.getOperand(3).getReg();
+
+  MIRBuilder.buildMul(MulReg, LHSReg, RHSReg);
+
+  auto One = MIRBuilder.buildConstant(Ty, 1);
+  auto Max =
+      MIRBuilder.buildConstant(Ty, APInt::getMaxValue(Ty.getSizeInBits()));
+  MIRBuilder.buildICmp(
+      CmpInst::ICMP_UGT, OverflowReg, LHSReg,
+      MIRBuilder.buildInstr(G_UDIV, {Ty},
+                            {Max, MIRBuilder.buildUMax(Ty, RHSReg, One)}));
+
+  MI.eraseFromParent();
+  return LegalizerHelper::Legalized;
+}
+
 LegalizerHelper::LegalizeResult Z80LegalizerInfo::legalizeMemIntrinsic(
     LegalizerHelper &Helper, MachineInstr &MI,
     LostDebugLocObserver &LocObserver) const {

diff --git a/llvm/lib/Target/Z80/GISel/Z80LegalizerInfo.h b/llvm/lib/Target/Z80/GISel/Z80LegalizerInfo.h
@@ -59,6 +59,8 @@ class Z80LegalizerInfo : public LegalizerInfo {
   LegalizerHelper::LegalizeResult legalizeCompare(LegalizerHelper &Helper,
                                                   MachineInstr &MI) const;
   LegalizerHelper::LegalizeResult
+  legalizeMultiplyWithOverflow(LegalizerHelper &Helper, MachineInstr &MI) const;
+  LegalizerHelper::LegalizeResult
   legalizeMemIntrinsic(LegalizerHelper &Helper, MachineInstr &MI,
                        LostDebugLocObserver &LocObserver) const;
 };