[X86][GlobalISel] Add instruction selection for G_SELECT (llvm#70753)

Author: e-kud

llvm/lib/Target/X86/GISel/X86InstructionSelector.cpp

@@ -20,6 +20,7 @@
 #include "X86Subtarget.h"
 #include "X86TargetMachine.h"
 #include "llvm/CodeGen/GlobalISel/GIMatchTableExecutorImpl.h"
+#include "llvm/CodeGen/GlobalISel/GenericMachineInstrs.h"
 #include "llvm/CodeGen/GlobalISel/InstructionSelector.h"
 #include "llvm/CodeGen/GlobalISel/Utils.h"
 #include "llvm/CodeGen/LowLevelType.h"
@@ -116,6 +117,8 @@ class X86InstructionSelector : public InstructionSelector {
   bool selectImplicitDefOrPHI(MachineInstr &I, MachineRegisterInfo &MRI) const;
   bool selectMulDivRem(MachineInstr &I, MachineRegisterInfo &MRI,
                        MachineFunction &MF) const;
+  bool selectSelect(MachineInstr &I, MachineRegisterInfo &MRI,
+                    MachineFunction &MF) const;
   bool selectIntrinsicWSideEffects(MachineInstr &I, MachineRegisterInfo &MRI,
                                    MachineFunction &MF) const;
 
@@ -429,6 +432,8 @@ bool X86InstructionSelector::select(MachineInstr &I) {
   case TargetOpcode::G_SREM:
   case TargetOpcode::G_UREM:
     return selectMulDivRem(I, MRI, MF);
+  case TargetOpcode::G_SELECT:
+    return selectSelect(I, MRI, MF);
   case TargetOpcode::G_INTRINSIC_W_SIDE_EFFECTS:
     return selectIntrinsicWSideEffects(I, MRI, MF);
   }
@@ -1789,6 +1794,49 @@ bool X86InstructionSelector::selectMulDivRem(MachineInstr &I,
   return true;
 }
 
+bool X86InstructionSelector::selectSelect(MachineInstr &I,
+                                          MachineRegisterInfo &MRI,
+                                          MachineFunction &MF) const {
+  GSelect &Sel = cast<GSelect>(I);
+  unsigned DstReg = Sel.getReg(0);
+  BuildMI(*Sel.getParent(), Sel, Sel.getDebugLoc(), TII.get(X86::TEST32rr))
+      .addReg(Sel.getCondReg())
+      .addReg(Sel.getCondReg());
+
+  unsigned OpCmp;
+  LLT Ty = MRI.getType(DstReg);
+  switch (Ty.getSizeInBits()) {
+  default:
+    return false;
+  case 8:
+    OpCmp = X86::CMOV_GR8;
+    break;
+  case 16:
+    OpCmp = STI.canUseCMOV() ? X86::CMOV16rr : X86::CMOV_GR16;
+    break;
+  case 32:
+    OpCmp = STI.canUseCMOV() ? X86::CMOV32rr : X86::CMOV_GR32;
+    break;
+  case 64:
+    assert(STI.is64Bit() && STI.canUseCMOV());
+    OpCmp = X86::CMOV64rr;
+    break;
+  }
+  BuildMI(*Sel.getParent(), Sel, Sel.getDebugLoc(), TII.get(OpCmp), DstReg)
+      .addReg(Sel.getTrueReg())
+      .addReg(Sel.getFalseReg())
+      .addImm(X86::COND_E);
+
+  const TargetRegisterClass *DstRC = getRegClass(Ty, DstReg, MRI);
+  if (!RBI.constrainGenericRegister(DstReg, *DstRC, MRI)) {
+    LLVM_DEBUG(dbgs() << "Failed to constrain CMOV\n");
+    return false;
+  }
+
+  Sel.eraseFromParent();
+  return true;
+}
+
 bool X86InstructionSelector::selectIntrinsicWSideEffects(
     MachineInstr &I, MachineRegisterInfo &MRI, MachineFunction &MF) const {
 

llvm/lib/Target/X86/GISel/X86LegalizerInfo.cpp

@@ -29,6 +29,7 @@ X86LegalizerInfo::X86LegalizerInfo(const X86Subtarget &STI,
     : Subtarget(STI) {
 
   bool Is64Bit = Subtarget.is64Bit();
+  bool HasCMOV = Subtarget.canUseCMOV();
   bool HasSSE1 = Subtarget.hasSSE1();
   bool HasSSE2 = Subtarget.hasSSE2();
   bool HasSSE41 = Subtarget.hasSSE41();
@@ -521,11 +522,10 @@ X86LegalizerInfo::X86LegalizerInfo(const X86Subtarget &STI,
 
   // todo: vectors and address spaces
   getActionDefinitionsBuilder(G_SELECT)
-       .legalFor({{s8, s32}, {s16, s32}, {s32, s32}, {s64, s32},
-                  {p0, s32}})
-    .widenScalarToNextPow2(0, /*Min=*/8)
-    .clampScalar(0, s8, sMaxScalar)
-    .clampScalar(1, s32, s32);
+      .legalFor({{s8, s32}, {s16, s32}, {s32, s32}, {s64, s32}, {p0, s32}})
+      .widenScalarToNextPow2(0, /*Min=*/8)
+      .clampScalar(0, HasCMOV ? s16 : s8, sMaxScalar)
+      .clampScalar(1, s32, s32);
 
   // memory intrinsics
   getActionDefinitionsBuilder({G_MEMCPY, G_MEMMOVE, G_MEMSET}).libcall();

llvm/test/CodeGen/X86/GlobalISel/legalize-select.mir

@@ -41,6 +41,7 @@ body:             |
     ; X86-NEXT: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[SELECT]](s32), [[SELECT1]](s32)
     ; X86-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY [[MV]](s64)
     ; X86-NEXT: RET 0, implicit [[COPY]](s64)
+    ;
     ; X64-LABEL: name: test_select64
     ; X64: [[DEF:%[0-9]+]]:_(s64) = IMPLICIT_DEF
     ; X64-NEXT: [[DEF1:%[0-9]+]]:_(s64) = IMPLICIT_DEF
@@ -101,14 +102,26 @@ body:             |
 name:            test_select8
 body:             |
   bb.1:
-    ; CHECK-LABEL: name: test_select8
-    ; CHECK: [[DEF:%[0-9]+]]:_(s8) = IMPLICIT_DEF
-    ; CHECK-NEXT: [[DEF1:%[0-9]+]]:_(s8) = IMPLICIT_DEF
-    ; CHECK-NEXT: [[DEF2:%[0-9]+]]:_(s1) = IMPLICIT_DEF
-    ; CHECK-NEXT: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[DEF2]](s1)
-    ; CHECK-NEXT: [[SELECT:%[0-9]+]]:_(s8) = G_SELECT [[ZEXT]](s32), [[DEF1]], [[DEF]]
-    ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s8) = COPY [[SELECT]](s8)
-    ; CHECK-NEXT: RET 0, implicit [[COPY]](s8)
+    ; X86-LABEL: name: test_select8
+    ; X86: [[DEF:%[0-9]+]]:_(s8) = IMPLICIT_DEF
+    ; X86-NEXT: [[DEF1:%[0-9]+]]:_(s8) = IMPLICIT_DEF
+    ; X86-NEXT: [[DEF2:%[0-9]+]]:_(s1) = IMPLICIT_DEF
+    ; X86-NEXT: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[DEF2]](s1)
+    ; X86-NEXT: [[SELECT:%[0-9]+]]:_(s8) = G_SELECT [[ZEXT]](s32), [[DEF1]], [[DEF]]
+    ; X86-NEXT: [[COPY:%[0-9]+]]:_(s8) = COPY [[SELECT]](s8)
+    ; X86-NEXT: RET 0, implicit [[COPY]](s8)
+    ;
+    ; X64-LABEL: name: test_select8
+    ; X64: [[DEF:%[0-9]+]]:_(s8) = IMPLICIT_DEF
+    ; X64-NEXT: [[DEF1:%[0-9]+]]:_(s8) = IMPLICIT_DEF
+    ; X64-NEXT: [[DEF2:%[0-9]+]]:_(s1) = IMPLICIT_DEF
+    ; X64-NEXT: [[ANYEXT:%[0-9]+]]:_(s16) = G_ANYEXT [[DEF1]](s8)
+    ; X64-NEXT: [[ANYEXT1:%[0-9]+]]:_(s16) = G_ANYEXT [[DEF]](s8)
+    ; X64-NEXT: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[DEF2]](s1)
+    ; X64-NEXT: [[SELECT:%[0-9]+]]:_(s16) = G_SELECT [[ZEXT]](s32), [[ANYEXT]], [[ANYEXT1]]
+    ; X64-NEXT: [[TRUNC:%[0-9]+]]:_(s8) = G_TRUNC [[SELECT]](s16)
+    ; X64-NEXT: [[COPY:%[0-9]+]]:_(s8) = COPY [[TRUNC]](s8)
+    ; X64-NEXT: RET 0, implicit [[COPY]](s8)
     %0:_(s8) = IMPLICIT_DEF
     %1:_(s8) = IMPLICIT_DEF
     %2:_(s1) = IMPLICIT_DEF