[MIPS GlobalISel] NarrowScalar G_UMULH

[Petar.Avramovic]

Summary

NarrowScalar G_UMULH in LegalizerHelper using MultiplyRegisters helper function.
NarrowScalar G_UMULH for MIPS32.

[Petar.Avramovic]

Rebased.

[atanasyan]

LGTM

lib/CodeGen/GlobalISel/LegalizerHelper.cpp

for (int i = 1; i < NumParts; ++i) {
       DstRegs.push_back(DstReg);
       BorrowIn = BorrowOut;
     }
     MIRBuilder.buildMerge(MI.getOperand(0).getReg(), DstRegs);
     MI.eraseFromParent();
     return Legalized;
   }
   case TargetOpcode::G_MUL:
+  case TargetOpcode::G_UMULH:
     return narrowScalarMul(MI, NarrowTy);
   case TargetOpcode::G_EXTRACT:
     return narrowScalarExtract(MI, TypeIdx, NarrowTy);
   case TargetOpcode::G_INSERT:
     return narrowScalarInsert(MI, TypeIdx, NarrowTy);
   case TargetOpcode::G_LOAD: {
     const auto &MMO = **MI.memoperands_begin();
     unsigned DstReg = MI.getOperand(0).getReg();
LegalizerHelper::narrowScalarMul(MachineInstr &MI, LLT NarrowTy) {
   unsigned SrcSize = MRI.getType(Src1).getSizeInBits();
   unsigned DstSize = Ty.getSizeInBits();
   unsigned NarrowSize = NarrowTy.getSizeInBits();
   if (DstSize % NarrowSize != 0 || SrcSize % NarrowSize != 0)
     return UnableToLegalize;
 
   unsigned NumDstParts = DstSize / NarrowSize;
   unsigned NumSrcParts = SrcSize / NarrowSize;
+  bool IsMulHigh = MI.getOpcode() == TargetOpcode::G_UMULH;
+  unsigned DstTmpParts = NumDstParts * (IsMulHigh ? 2 : 1);
 
-  SmallVector<unsigned, 2> Src1Parts, Src2Parts, DstRegs;
+  SmallVector<unsigned, 2> Src1Parts, Src2Parts, DstTmpRegs;
   extractParts(Src1, NarrowTy, NumSrcParts, Src1Parts);
   extractParts(Src2, NarrowTy, NumSrcParts, Src2Parts);
-  DstRegs.resize(NumDstParts);
-  MultiplyRegisters(DstRegs, Src1Parts, Src2Parts, NarrowTy);
+  DstTmpRegs.resize(DstTmpParts);
+  MultiplyRegisters(DstTmpRegs, Src1Parts, Src2Parts, NarrowTy);
 
+  // Take only high half of registers if this is high mul.
+  ArrayRef<unsigned> DstRegs(
+      IsMulHigh ? &DstTmpRegs[DstTmpParts / 2] : &DstTmpRegs[0], NumDstParts);
   MIRBuilder.buildMerge(DstReg, DstRegs);
+
   MI.eraseFromParent();
   return Legalized;
 }
 
 LegalizerHelper::LegalizeResult
 LegalizerHelper::narrowScalarExtract(MachineInstr &MI, unsigned TypeIdx,
                                      LLT NarrowTy) {
   if (TypeIdx != 1)

lib/Target/Mips/MipsLegalizerInfo.cpp

MipsLegalizerInfo::MipsLegalizerInfo(const MipsSubtarget &ST) {
   getActionDefinitionsBuilder({G_ADD, G_SUB, G_MUL})
       .legalFor({s32})
       .clampScalar(0, s32, s32);
 
   getActionDefinitionsBuilder({G_UADDO, G_UADDE, G_USUBO, G_USUBE, G_UMULO})
       .lowerFor({{s32, s1}});
 
   getActionDefinitionsBuilder(G_UMULH)
-      .legalFor({s32});
+      .legalFor({s32})
+      .maxScalar(0, s32);
 
   getActionDefinitionsBuilder({G_LOAD, G_STORE})
       .legalForTypesWithMemDesc({{s32, p0, 8, 8},
                                  {s32, p0, 16, 8},
                                  {s32, p0, 32, 8},
                                  {p0, p0, 32, 8}})
       .minScalar(0, s32);
 

test/CodeGen/Mips/GlobalISel/legalizer/mul.mir

 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
 # RUN: llc -O0 -mtriple=mipsel-linux-gnu -run-pass=legalizer -verify-machineinstrs %s -o - | FileCheck %s -check-prefixes=MIPS32
 --- |
 
   define void @mul_i32() {entry: ret void}
   define void @mul_i8_sext() {entry: ret void}
   define void @mul_i8_zext() {entry: ret void}
   define void @mul_i8_aext() {entry: ret void}
   define void @mul_i16_sext() {entry: ret void}
   define void @mul_i16_zext() {entry: ret void}
   define void @mul_i16_aext() {entry: ret void}
   define void @mul_i64() {entry: ret void}
   define void @mul_i128() {entry: ret void}
+  define void @umulh_i64() {entry: ret void}
   define void @umul_with_overflow(i32 %lhs, i32 %rhs, i32* %pmul, i1* %pcarry_flag) { ret void }
 
 ...
 ---
 name:            mul_i32
 alignment:       2
 tracksRegLiveness: true
 body:             |
bb.1.entry:
     $v0 = COPY %15(s32)
     $v1 = COPY %16(s32)
     $a0 = COPY %17(s32)
     $a1 = COPY %18(s32)
     RetRA implicit $v0, implicit $v1, implicit $a0, implicit $a1
 
 ...
 ---
+name:            umulh_i64
+alignment:       2
+tracksRegLiveness: true
+body:             |
+  bb.1.entry:
+    liveins: $a0, $a1, $a2, $a3
+
+    ; MIPS32-LABEL: name: umulh_i64
+    ; MIPS32: liveins: $a0, $a1, $a2, $a3
+    ; MIPS32: [[COPY:%[0-9]+]]:_(s32) = COPY $a0
+    ; MIPS32: [[COPY1:%[0-9]+]]:_(s32) = COPY $a1
+    ; MIPS32: [[COPY2:%[0-9]+]]:_(s32) = COPY $a2
+    ; MIPS32: [[COPY3:%[0-9]+]]:_(s32) = COPY $a3
+    ; MIPS32: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[COPY3]], [[COPY]]
+    ; MIPS32: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[COPY2]], [[COPY1]]
+    ; MIPS32: [[UMULH:%[0-9]+]]:_(s32) = G_UMULH [[COPY2]], [[COPY]]
+    ; MIPS32: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[MUL]], [[MUL1]]
+    ; MIPS32: [[ICMP:%[0-9]+]]:_(s32) = G_ICMP intpred(ult), [[ADD]](s32), [[MUL1]]
+    ; MIPS32: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
+    ; MIPS32: [[COPY4:%[0-9]+]]:_(s32) = COPY [[ICMP]](s32)
+    ; MIPS32: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C]]
+    ; MIPS32: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[ADD]], [[UMULH]]
+    ; MIPS32: [[ICMP1:%[0-9]+]]:_(s32) = G_ICMP intpred(ult), [[ADD1]](s32), [[UMULH]]
+    ; MIPS32: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
+    ; MIPS32: [[COPY5:%[0-9]+]]:_(s32) = COPY [[ICMP1]](s32)
+    ; MIPS32: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
+    ; MIPS32: [[ADD2:%[0-9]+]]:_(s32) = G_ADD [[AND]], [[AND1]]
+    ; MIPS32: [[MUL2:%[0-9]+]]:_(s32) = G_MUL [[COPY3]], [[COPY1]]
+    ; MIPS32: [[UMULH1:%[0-9]+]]:_(s32) = G_UMULH [[COPY3]], [[COPY]]
+    ; MIPS32: [[UMULH2:%[0-9]+]]:_(s32) = G_UMULH [[COPY2]], [[COPY1]]
+    ; MIPS32: [[ADD3:%[0-9]+]]:_(s32) = G_ADD [[MUL2]], [[UMULH1]]
+    ; MIPS32: [[ICMP2:%[0-9]+]]:_(s32) = G_ICMP intpred(ult), [[ADD3]](s32), [[UMULH1]]
+    ; MIPS32: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
+    ; MIPS32: [[COPY6:%[0-9]+]]:_(s32) = COPY [[ICMP2]](s32)
+    ; MIPS32: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C2]]
+    ; MIPS32: [[ADD4:%[0-9]+]]:_(s32) = G_ADD [[ADD3]], [[UMULH2]]
+    ; MIPS32: [[ICMP3:%[0-9]+]]:_(s32) = G_ICMP intpred(ult), [[ADD4]](s32), [[UMULH2]]
+    ; MIPS32: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
+    ; MIPS32: [[COPY7:%[0-9]+]]:_(s32) = COPY [[ICMP3]](s32)
+    ; MIPS32: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
+    ; MIPS32: [[ADD5:%[0-9]+]]:_(s32) = G_ADD [[AND2]], [[AND3]]
+    ; MIPS32: [[ADD6:%[0-9]+]]:_(s32) = G_ADD [[ADD4]], [[ADD2]]
+    ; MIPS32: [[ICMP4:%[0-9]+]]:_(s32) = G_ICMP intpred(ult), [[ADD6]](s32), [[ADD2]]
+    ; MIPS32: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
+    ; MIPS32: [[COPY8:%[0-9]+]]:_(s32) = COPY [[ICMP4]](s32)
+    ; MIPS32: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C4]]
+    ; MIPS32: [[ADD7:%[0-9]+]]:_(s32) = G_ADD [[ADD5]], [[AND4]]
+    ; MIPS32: [[UMULH3:%[0-9]+]]:_(s32) = G_UMULH [[COPY3]], [[COPY1]]
+    ; MIPS32: [[ADD8:%[0-9]+]]:_(s32) = G_ADD [[UMULH3]], [[ADD7]]
+    ; MIPS32: $v0 = COPY [[ADD6]](s32)
+    ; MIPS32: $v1 = COPY [[ADD8]](s32)
+    ; MIPS32: RetRA implicit $v0, implicit $v1
+    %2:_(s32) = COPY $a0
+    %3:_(s32) = COPY $a1
+    %0:_(s64) = G_MERGE_VALUES %2(s32), %3(s32)
+    %4:_(s32) = COPY $a2
+    %5:_(s32) = COPY $a3
+    %1:_(s64) = G_MERGE_VALUES %4(s32), %5(s32)
+    %6:_(s64) = G_UMULH %1, %0
+    %7:_(s32), %8:_(s32) = G_UNMERGE_VALUES %6(s64)
+    $v0 = COPY %7(s32)
+    $v1 = COPY %8(s32)
+    RetRA implicit $v0, implicit $v1
+
+...
+---
 name:            umul_with_overflow
 alignment:       2
 tracksRegLiveness: true
 body:             |
   bb.1 (%ir-block.0):
     liveins: $a0, $a1, $a2, $a3
 
     ; MIPS32-LABEL: name: umul_with_overflow