[AArch64][GlobalISel] Select logical_imm32 and logical_imm64 patterns

[paquette]

Summary

Add a GlobalISel equivalent for the logical_imm32_XFORM and logical_imm64_XFORM SDNodeXForms in AArch64InstrFormats.td.

• Add select-logical-imm.mir, which contains tests for each imported pattern.

• Update select-pr32733.mir and select-scalar-shift-imm.mir, since they now

select instructions of this form.

[aemerson]

LGTM.

llvm/lib/Target/AArch64/AArch64InstrFormats.td

def logical_imm32_XFORM : SDNodeXForm<imm, [{
   uint64_t enc = AArch64_AM::encodeLogicalImmediate(N->getZExtValue(), 32);
   return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i32);
 }]>;
 def logical_imm64_XFORM : SDNodeXForm<imm, [{
   uint64_t enc = AArch64_AM::encodeLogicalImmediate(N->getZExtValue(), 64);
   return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i32);
 }]>;
 
+def gi_logical_imm32_XFORM : GICustomOperandRenderer<"renderLogicalImm32">,
+  GISDNodeXFormEquiv<logical_imm32_XFORM>;
+def gi_logical_imm64_XFORM : GICustomOperandRenderer<"renderLogicalImm64">,
+  GISDNodeXFormEquiv<logical_imm64_XFORM>;
+
 let DiagnosticType = "LogicalSecondSource" in {
   def LogicalImm32Operand : AsmOperandClass {
     let Name = "LogicalImm32";
     let PredicateMethod = "isLogicalImm<int32_t>";
     let RenderMethod = "addLogicalImmOperands<int32_t>";
   }
   def LogicalImm64Operand : AsmOperandClass {
     let Name = "LogicalImm64";

llvm/lib/Target/AArch64/AArch64InstructionSelector.cpp

private:
   ComplexRendererFns
   selectAddrModeShiftedExtendXReg(MachineOperand &Root,
                                   unsigned SizeInBytes) const;
   ComplexRendererFns selectAddrModeRegisterOffset(MachineOperand &Root) const;
   ComplexRendererFns selectAddrModeXRO(MachineOperand &Root,
                                        unsigned SizeInBytes) const;
 
   void renderTruncImm(MachineInstrBuilder &MIB, const MachineInstr &MI) const;
+  void renderLogicalImm32(MachineInstrBuilder &MIB, const MachineInstr &I) const;
+  void renderLogicalImm64(MachineInstrBuilder &MIB, const MachineInstr &I) const;
 
   // Materialize a GlobalValue or BlockAddress using a movz+movk sequence.
   void materializeLargeCMVal(MachineInstr &I, const Value *V,
                              unsigned OpFlags) const;
 
   // Optimization methods.
   bool tryOptVectorShuffle(MachineInstr &I) const;
   bool tryOptVectorDup(MachineInstr &MI) const;
void AArch64InstructionSelector::renderTruncImm(MachineInstrBuilder &MIB,
                                                 const MachineInstr &MI) const {
   const MachineRegisterInfo &MRI = MI.getParent()->getParent()->getRegInfo();
   assert(MI.getOpcode() == TargetOpcode::G_CONSTANT && "Expected G_CONSTANT");
   Optional<int64_t> CstVal = getConstantVRegVal(MI.getOperand(0).getReg(), MRI);
   assert(CstVal && "Expected constant value");
   MIB.addImm(CstVal.getValue());
 }
 
+void AArch64InstructionSelector::renderLogicalImm32(
+    MachineInstrBuilder &MIB, const MachineInstr &I) const {
+  assert(I.getOpcode() == TargetOpcode::G_CONSTANT && "Expected G_CONSTANT");
+  uint64_t CstVal = I.getOperand(1).getCImm()->getZExtValue();
+  uint64_t Enc = AArch64_AM::encodeLogicalImmediate(CstVal, 32);
+  MIB.addImm(Enc);
+}
+
+void AArch64InstructionSelector::renderLogicalImm64(
+    MachineInstrBuilder &MIB, const MachineInstr &I) const {
+  assert(I.getOpcode() == TargetOpcode::G_CONSTANT && "Expected G_CONSTANT");
+  uint64_t CstVal = I.getOperand(1).getCImm()->getZExtValue();
+  uint64_t Enc = AArch64_AM::encodeLogicalImmediate(CstVal, 64);
+  MIB.addImm(Enc);
+}
+
 namespace llvm {
 InstructionSelector *
 createAArch64InstructionSelector(const AArch64TargetMachine &TM,
                                  AArch64Subtarget &Subtarget,
                                  AArch64RegisterBankInfo &RBI) {
   return new AArch64InstructionSelector(TM, Subtarget, RBI);
 }
 }

llvm/test/CodeGen/AArch64/GlobalISel/select-logical-imm.mir

+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=aarch64-unknown-unknown -run-pass=instruction-select -verify-machineinstrs %s -o - | FileCheck %s
+
+---
+name:            logical_imm_64_and
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    liveins: $x0
+    ; CHECK-LABEL: name: logical_imm_64_and
+    ; CHECK: liveins: $x0
+    ; CHECK: [[COPY:%[0-9]+]]:gpr64 = COPY $x0
+    ; CHECK: [[ANDXri:%[0-9]+]]:gpr64sp = ANDXri [[COPY]], 4096
+    ; CHECK: $x0 = COPY [[ANDXri]]
+    ; CHECK: RET_ReallyLR implicit $x0
+    %0:gpr(s64) = COPY $x0
+    %1:gpr(s64) = G_CONSTANT i64 1
+    %2:gpr(s64) = G_AND %0, %1:gpr(s64)
+    $x0 = COPY %2:gpr(s64)
+    RET_ReallyLR implicit $x0
+...
+---
+name:            logical_imm_64_or
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    liveins: $x0
+    ; CHECK-LABEL: name: logical_imm_64_or
+    ; CHECK: liveins: $x0
+    ; CHECK: [[COPY:%[0-9]+]]:gpr64 = COPY $x0
+    ; CHECK: [[ORRXri:%[0-9]+]]:gpr64sp = ORRXri [[COPY]], 4096
+    ; CHECK: $x0 = COPY [[ORRXri]]
+    ; CHECK: RET_ReallyLR implicit $x0
+    %0:gpr(s64) = COPY $x0
+    %1:gpr(s64) = G_CONSTANT i64 1
+    %2:gpr(s64) = G_OR %0, %1:gpr(s64)
+    $x0 = COPY %2:gpr(s64)
+    RET_ReallyLR implicit $x0
+...
+---
+name:            logical_imm_64_xor
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    liveins: $x0
+    ; CHECK-LABEL: name: logical_imm_64_xor
+    ; CHECK: liveins: $x0
+    ; CHECK: [[COPY:%[0-9]+]]:gpr64 = COPY $x0
+    ; CHECK: [[EORXri:%[0-9]+]]:gpr64sp = EORXri [[COPY]], 4096
+    ; CHECK: $x0 = COPY [[EORXri]]
+    ; CHECK: RET_ReallyLR implicit $x0
+    %0:gpr(s64) = COPY $x0
+    %1:gpr(s64) = G_CONSTANT i64 1
+    %2:gpr(s64) = G_XOR %0, %1:gpr(s64)
+    $x0 = COPY %2:gpr(s64)
+    RET_ReallyLR implicit $x0
+...
+---
+name:            logical_imm_32_and
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    liveins: $w0
+    ; CHECK-LABEL: name: logical_imm_32_and
+    ; CHECK: liveins: $w0
+    ; CHECK: [[COPY:%[0-9]+]]:gpr32 = COPY $w0
+    ; CHECK: [[ANDWri:%[0-9]+]]:gpr32sp = ANDWri [[COPY]], 0
+    ; CHECK: $w0 = COPY [[ANDWri]]
+    ; CHECK: RET_ReallyLR implicit $w0
+    %0:gpr(s32) = COPY $w0
+    %1:gpr(s32) = G_CONSTANT i32 1
+    %2:gpr(s32) = G_AND %0, %1:gpr(s32)
+    $w0 = COPY %2:gpr(s32)
+    RET_ReallyLR implicit $w0
+...
+---
+name:            logical_imm_32_or
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    liveins: $w0
+    ; CHECK-LABEL: name: logical_imm_32_or
+    ; CHECK: liveins: $w0
+    ; CHECK: [[COPY:%[0-9]+]]:gpr32 = COPY $w0
+    ; CHECK: [[ORRWri:%[0-9]+]]:gpr32sp = ORRWri [[COPY]], 0
+    ; CHECK: $w0 = COPY [[ORRWri]]
+    ; CHECK: RET_ReallyLR implicit $w0
+    %0:gpr(s32) = COPY $w0
+    %1:gpr(s32) = G_CONSTANT i32 1
+    %2:gpr(s32) = G_OR %0, %1:gpr(s32)
+    $w0 = COPY %2:gpr(s32)
+    RET_ReallyLR implicit $w0
+...
+---
+name:            logical_imm_32_xor
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    liveins: $w0
+    ; CHECK-LABEL: name: logical_imm_32_xor
+    ; CHECK: liveins: $w0
+    ; CHECK: [[COPY:%[0-9]+]]:gpr32 = COPY $w0
+    ; CHECK: [[EORWri:%[0-9]+]]:gpr32sp = EORWri [[COPY]], 0
+    ; CHECK: $w0 = COPY [[EORWri]]
+    ; CHECK: RET_ReallyLR implicit $w0
+    %0:gpr(s32) = COPY $w0
+    %1:gpr(s32) = G_CONSTANT i32 1
+    %2:gpr(s32) = G_XOR %0, %1:gpr(s32)
+    $w0 = COPY %2:gpr(s32)
+    RET_ReallyLR implicit $w0
+...

llvm/test/CodeGen/AArch64/GlobalISel/select-pr32733.mir

frameInfo:
   hasOpaqueSPAdjustment: false
   hasVAStart:      false
   hasMustTailInVarArgFunc: false
 body:             |
   bb.1.entry:
     liveins: $w0
     ; CHECK-LABEL: name: main
     ; CHECK: liveins: $w0
-    ; CHECK: [[MOVi32imm:%[0-9]+]]:gpr32 = MOVi32imm 1
     ; CHECK: [[COPY:%[0-9]+]]:gpr32 = COPY $w0
-    ; CHECK: [[EONWrr:%[0-9]+]]:gpr32 = EONWrr [[COPY]], [[MOVi32imm]]
-    ; CHECK: $w0 = COPY [[EONWrr]]
+    ; CHECK: [[ORNWrr:%[0-9]+]]:gpr32 = ORNWrr $wzr, [[COPY]]
+    ; CHECK: [[EORWri:%[0-9]+]]:gpr32sp = EORWri [[ORNWrr]], 0
+    ; CHECK: $w0 = COPY [[EORWri]]
     %0(s32) = G_CONSTANT i32 -1
     %3(s32) = G_CONSTANT i32 1
     %1(s32) = COPY $w0
     %2(s32) = G_XOR %1, %0
     %4(s32) = G_XOR %2, %3
     $w0 = COPY %4(s32)
 ...

llvm/test/CodeGen/AArch64/GlobalISel/select-scalar-shift-imm.mir

 regBankSelected: true
 body:             |
   bb.1:
     liveins: $w0
 
     ; CHECK-LABEL: name: lshr_32_notimm64
     ; CHECK: [[COPY:%[0-9]+]]:gpr32 = COPY $w0
     ; CHECK: [[MOVi64imm:%[0-9]+]]:gpr64 = MOVi64imm 8
-    ; CHECK: [[ANDXrr:%[0-9]+]]:gpr64 = ANDXrr [[MOVi64imm]], [[MOVi64imm]]
-    ; CHECK: [[COPY1:%[0-9]+]]:gpr32 = COPY [[ANDXrr]].sub_32
+    ; CHECK: [[ANDXri:%[0-9]+]]:gpr64sp = ANDXri [[MOVi64imm]], 8000
+    ; CHECK: [[COPY1:%[0-9]+]]:gpr32 = COPY [[ANDXri]].sub_32
     ; CHECK: [[LSRVWr:%[0-9]+]]:gpr32 = LSRVWr [[COPY]], [[COPY1]]
     ; CHECK: $w0 = COPY [[LSRVWr]]
     ; CHECK: RET_ReallyLR implicit $w0
     %0:gpr(s32) = COPY $w0
     %3:gpr(s64) = G_CONSTANT i64 8
     %4:gpr(s64) = G_AND %3, %3
     %2:gpr(s32) = G_LSHR %0, %4(s64)
     $w0 = COPY %2(s32)
     RET_ReallyLR implicit $w0
 
 ...
 ---
 name:            ashr_32_notimm64
 legalized:       true
 regBankSelected: true
 body:             |
   bb.1:
     liveins: $w0
 
     ; CHECK-LABEL: name: ashr_32_notimm64
     ; CHECK: [[COPY:%[0-9]+]]:gpr32 = COPY $w0
     ; CHECK: [[MOVi64imm:%[0-9]+]]:gpr64 = MOVi64imm 8
-    ; CHECK: [[ANDXrr:%[0-9]+]]:gpr64 = ANDXrr [[MOVi64imm]], [[MOVi64imm]]
-    ; CHECK: [[COPY1:%[0-9]+]]:gpr32 = COPY [[ANDXrr]].sub_32
+    ; CHECK: [[ANDXri:%[0-9]+]]:gpr64sp = ANDXri [[MOVi64imm]], 8000
+    ; CHECK: [[COPY1:%[0-9]+]]:gpr32 = COPY [[ANDXri]].sub_32
     ; CHECK: [[ASRVWr:%[0-9]+]]:gpr32 = ASRVWr [[COPY]], [[COPY1]]
     ; CHECK: $w0 = COPY [[ASRVWr]]
     ; CHECK: RET_ReallyLR implicit $w0
     %0:gpr(s32) = COPY $w0
     %3:gpr(s64) = G_CONSTANT i64 8
     %4:gpr(s64) = G_AND %3, %3
     %2:gpr(s32) = G_ASHR %0, %4(s64)
     $w0 = COPY %2(s32)
     RET_ReallyLR implicit $w0
 
 ...