[GlobalISel] Import samesign flag

llvm/include/llvm/CodeGen/GlobalISel/GenericMachineInstrs.h

@@ -28,7 +28,7 @@ namespace llvm {
 class GenericMachineInstr : public MachineInstr {
   constexpr static unsigned PoisonFlags = NoUWrap | NoSWrap | NoUSWrap |
                                           IsExact | Disjoint | NonNeg |
-                                          FmNoNans | FmNoInfs;
+                                          FmNoNans | FmNoInfs | SameSign;
 
 public:
   GenericMachineInstr() = delete;

llvm/include/llvm/CodeGen/GlobalISel/MachineIRBuilder.h

@@ -1266,7 +1266,8 @@ class MachineIRBuilder {
   ///
   /// \return a MachineInstrBuilder for the newly created instruction.
   MachineInstrBuilder buildICmp(CmpInst::Predicate Pred, const DstOp &Res,
-                                const SrcOp &Op0, const SrcOp &Op1);
+                                const SrcOp &Op0, const SrcOp &Op1,
+                                std::optional<unsigned> Flgs = std::nullopt);
 
   /// Build and insert a \p Res = G_FCMP \p Pred\p Op0, \p Op1
   ///

llvm/include/llvm/CodeGen/MachineInstr.h

@@ -119,6 +119,7 @@ class MachineInstr
     Disjoint = 1 << 19,      // Each bit is zero in at least one of the inputs.
     NoUSWrap = 1 << 20,      // Instruction supports geps
                              // no unsigned signed wrap.
+    SameSign = 1 << 21       // Both operands have the same sign.
   };
 
 private:

llvm/lib/CodeGen/GlobalISel/IRTranslator.cpp

@@ -340,20 +340,17 @@ bool IRTranslator::translateCompare(const User &U,
   Register Op1 = getOrCreateVReg(*U.getOperand(1));
   Register Res = getOrCreateVReg(U);
   CmpInst::Predicate Pred = CI->getPredicate();
+  uint32_t Flags = MachineInstr::copyFlagsFromInstruction(*CI);
   if (CmpInst::isIntPredicate(Pred))
-    MIRBuilder.buildICmp(Pred, Res, Op0, Op1);
+    MIRBuilder.buildICmp(Pred, Res, Op0, Op1, Flags);
   else if (Pred == CmpInst::FCMP_FALSE)
     MIRBuilder.buildCopy(
         Res, getOrCreateVReg(*Constant::getNullValue(U.getType())));
   else if (Pred == CmpInst::FCMP_TRUE)
     MIRBuilder.buildCopy(
         Res, getOrCreateVReg(*Constant::getAllOnesValue(U.getType())));
-  else {
-    uint32_t Flags = 0;
-    if (CI)
-      Flags = MachineInstr::copyFlagsFromInstruction(*CI);
+  else
     MIRBuilder.buildFCmp(Pred, Res, Op0, Op1, Flags);
-  }
 
   return true;
 }

llvm/lib/CodeGen/GlobalISel/MachineIRBuilder.cpp

@@ -898,8 +898,9 @@ MachineIRBuilder::buildFPTrunc(const DstOp &Res, const SrcOp &Op,
 MachineInstrBuilder MachineIRBuilder::buildICmp(CmpInst::Predicate Pred,
                                                 const DstOp &Res,
                                                 const SrcOp &Op0,
-                                                const SrcOp &Op1) {
-  return buildInstr(TargetOpcode::G_ICMP, Res, {Pred, Op0, Op1});
+                                                const SrcOp &Op1,
+                                                std::optional<unsigned> Flags) {
+  return buildInstr(TargetOpcode::G_ICMP, Res, {Pred, Op0, Op1}, Flags);
 }
 
 MachineInstrBuilder MachineIRBuilder::buildFCmp(CmpInst::Predicate Pred,

llvm/lib/CodeGen/MIRParser/MILexer.cpp

@@ -216,6 +216,7 @@ static MIToken::TokenKind getIdentifierKind(StringRef Identifier) {
       .Case("exact", MIToken::kw_exact)
       .Case("nneg", MIToken::kw_nneg)
       .Case("disjoint", MIToken::kw_disjoint)
+      .Case("samesign", MIToken::kw_samesign)
       .Case("nofpexcept", MIToken::kw_nofpexcept)
       .Case("unpredictable", MIToken::kw_unpredictable)
       .Case("debug-location", MIToken::kw_debug_location)

llvm/lib/CodeGen/MIRParser/MILexer.h

@@ -77,6 +77,7 @@ struct MIToken {
     kw_unpredictable,
     kw_nneg,
     kw_disjoint,
+    kw_samesign,
     kw_debug_location,
     kw_debug_instr_number,
     kw_dbg_instr_ref,

llvm/lib/CodeGen/MIRParser/MIParser.cpp

@@ -1476,7 +1476,8 @@ bool MIParser::parseInstruction(unsigned &OpCode, unsigned &Flags) {
          Token.is(MIToken::kw_noconvergent) ||
          Token.is(MIToken::kw_unpredictable) ||
          Token.is(MIToken::kw_nneg) ||
-         Token.is(MIToken::kw_disjoint)) {
+         Token.is(MIToken::kw_disjoint) ||
+         Token.is(MIToken::kw_samesign)) {
     // clang-format on
     // Mine frame and fast math flags
     if (Token.is(MIToken::kw_frame_setup))
@@ -1513,6 +1514,8 @@ bool MIParser::parseInstruction(unsigned &OpCode, unsigned &Flags) {
       Flags |= MachineInstr::NonNeg;
     if (Token.is(MIToken::kw_disjoint))
       Flags |= MachineInstr::Disjoint;
+    if (Token.is(MIToken::kw_samesign))
+      Flags |= MachineInstr::SameSign;
 
     lex();
   }

llvm/lib/CodeGen/MIRPrinter.cpp

@@ -837,6 +837,8 @@ void MIPrinter::print(const MachineInstr &MI) {
     OS << "disjoint ";
   if (MI.getFlag(MachineInstr::NoUSWrap))
     OS << "nusw ";
+  if (MI.getFlag(MachineInstr::SameSign))
+    OS << "samesign ";
 
   OS << TII->getName(MI.getOpcode());
   if (I < E)

llvm/lib/CodeGen/MachineInstr.cpp

@@ -596,6 +596,11 @@ uint32_t MachineInstr::copyFlagsFromInstruction(const Instruction &I) {
       MIFlags |= MachineInstr::MIFlag::Disjoint;
   }
 
+  // Copy the samesign flag.
+  if (const ICmpInst *ICmp = dyn_cast<ICmpInst>(&I))
+    if (ICmp->hasSameSign())
+      MIFlags |= MachineInstr::MIFlag::SameSign;
+
   // Copy the exact flag.
   if (const PossiblyExactOperator *PE = dyn_cast<PossiblyExactOperator>(&I))
     if (PE->isExact())
@@ -1770,6 +1775,8 @@ void MachineInstr::print(raw_ostream &OS, ModuleSlotTracker &MST,
     OS << "nneg ";
   if (getFlag(MachineInstr::Disjoint))
     OS << "disjoint ";
+  if (getFlag(MachineInstr::SameSign))
+    OS << "samesign ";
 
   // Print the opcode name.
   if (TII)

llvm/test/CodeGen/AArch64/GlobalISel/icmp-flags.mir

@@ -0,0 +1,45 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple aarch64 -run-pass=none -verify-machineinstrs %s -o - | FileCheck %s
+
+---
+name:            icmp_samesign
+body:             |
+  bb.0:
+    liveins: $w0, $w1
+    ; CHECK-LABEL: name: icmp_samesign
+    ; CHECK: liveins: $w0, $w1
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: %x:_(s32) = COPY $w0
+    ; CHECK-NEXT: %y:_(s32) = COPY $w1
+    ; CHECK-NEXT: %cmp:_(s1) = samesign G_ICMP intpred(eq), %y(s32), %y
+    ; CHECK-NEXT: %zext:_(s32) = G_ZEXT %cmp(s1)
+    ; CHECK-NEXT: $w0 = COPY %zext(s32)
+    ; CHECK-NEXT: RET_ReallyLR implicit $w0
+    %x:_(s32) = COPY $w0
+    %y:_(s32) = COPY $w1
+    %cmp:_(s1) = samesign G_ICMP intpred(eq), %y:_(s32), %y:_
+    %zext:_(s32) = G_ZEXT %cmp:_(s1)
+    $w0 = COPY %zext
+    RET_ReallyLR implicit $w0
+...
+---
+name:            icmp_differentsign
+body:             |
+  bb.0:
+    liveins: $w0, $w1
+    ; CHECK-LABEL: name: icmp_differentsign
+    ; CHECK: liveins: $w0, $w1
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: %x:_(s32) = COPY $w0
+    ; CHECK-NEXT: %y:_(s32) = COPY $w1
+    ; CHECK-NEXT: %cmp:_(s1) = G_ICMP intpred(eq), %y(s32), %y
+    ; CHECK-NEXT: %zext:_(s32) = G_ZEXT %cmp(s1)
+    ; CHECK-NEXT: $w0 = COPY %zext(s32)
+    ; CHECK-NEXT: RET_ReallyLR implicit $w0
+    %x:_(s32) = COPY $w0
+    %y:_(s32) = COPY $w1
+    %cmp:_(s1) = G_ICMP intpred(eq), %y:_(s32), %y:_
+    %zext:_(s32) = G_ZEXT %cmp:_(s1)
+    $w0 = COPY %zext
+    RET_ReallyLR implicit $w0
+---

llvm/test/CodeGen/AArch64/GlobalISel/irtranslater-samesign.ll

@@ -0,0 +1,69 @@
+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 4
+; RUN: llc -global-isel -mtriple=aarch64-linux-gnu -O0 -stop-after=irtranslator < %s | FileCheck %s
+
+
+define <2 x i1> @call_icmp_samesign_vector(<2 x i32> %a, <2 x i32> %b) {
+  ; CHECK-LABEL: name: call_icmp_samesign_vector
+  ; CHECK: bb.1.entry:
+  ; CHECK-NEXT:   liveins: $d0, $d1
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $d0
+  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $d1
+  ; CHECK-NEXT:   %2:_(<2 x s1>) = samesign G_ICMP intpred(ult), [[COPY]](<2 x s32>), [[COPY1]]
+  ; CHECK-NEXT:   [[ANYEXT:%[0-9]+]]:_(<2 x s32>) = G_ANYEXT %2(<2 x s1>)
+  ; CHECK-NEXT:   $d0 = COPY [[ANYEXT]](<2 x s32>)
+  ; CHECK-NEXT:   RET_ReallyLR implicit $d0
+entry:
+  %result = icmp samesign ult <2 x i32> %a, %b
+  ret <2 x i1> %result
+}
+
+define <2 x i1> @call_icmp_vector(<2 x i32> %a, <2 x i32> %b) {
+  ; CHECK-LABEL: name: call_icmp_vector
+  ; CHECK: bb.1.entry:
+  ; CHECK-NEXT:   liveins: $d0, $d1
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $d0
+  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $d1
+  ; CHECK-NEXT:   [[ICMP:%[0-9]+]]:_(<2 x s1>) = G_ICMP intpred(ult), [[COPY]](<2 x s32>), [[COPY1]]
+  ; CHECK-NEXT:   [[ANYEXT:%[0-9]+]]:_(<2 x s32>) = G_ANYEXT [[ICMP]](<2 x s1>)
+  ; CHECK-NEXT:   $d0 = COPY [[ANYEXT]](<2 x s32>)
+  ; CHECK-NEXT:   RET_ReallyLR implicit $d0
+entry:
+  %result = icmp ult <2 x i32> %a, %b
+  ret <2 x i1> %result
+}
+
+define i1 @call_icmp(i32 %a) {
+  ; CHECK-LABEL: name: call_icmp
+  ; CHECK: bb.1.entry:
+  ; CHECK-NEXT:   liveins: $w0
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:_(s32) = COPY $w0
+  ; CHECK-NEXT:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
+  ; CHECK-NEXT:   [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ult), [[COPY]](s32), [[C]]
+  ; CHECK-NEXT:   [[ZEXT:%[0-9]+]]:_(s8) = G_ZEXT [[ICMP]](s1)
+  ; CHECK-NEXT:   [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[ZEXT]](s8)
+  ; CHECK-NEXT:   $w0 = COPY [[ANYEXT]](s32)
+  ; CHECK-NEXT:   RET_ReallyLR implicit $w0
+entry:
+  %result = icmp ult i32 %a, 3
+  ret i1 %result
+}
+
+define i1 @call_icmp_samesign(i32 %a) {
+  ; CHECK-LABEL: name: call_icmp_samesign
+  ; CHECK: bb.1.entry:
+  ; CHECK-NEXT:   liveins: $w0
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:_(s32) = COPY $w0
+  ; CHECK-NEXT:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
+  ; CHECK-NEXT:   %2:_(s1) = samesign G_ICMP intpred(ult), [[COPY]](s32), [[C]]
+  ; CHECK-NEXT:   [[ZEXT:%[0-9]+]]:_(s8) = G_ZEXT %2(s1)
+  ; CHECK-NEXT:   [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[ZEXT]](s8)
+  ; CHECK-NEXT:   $w0 = COPY [[ANYEXT]](s32)
+  ; CHECK-NEXT:   RET_ReallyLR implicit $w0
+entry:
+  %result = icmp samesign ult i32 %a, 3
+  ret i1 %result
+}