2 files changed, 44 insertions, 7 deletions

diff --git a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
index 861c22d605a..684b99d8bae 100644
--- a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
@@ -2249,6 +2249,8 @@ LegalizerHelper::lower(MachineInstr &MI, unsigned TypeIdx, LLT Ty) {
     return lowerDynStackAlloc(MI);
   case G_EXTRACT:
     return lowerExtract(MI);
+  case G_INSERT:
+    return lowerInsert(MI);
   }
 }
 
@@ -4134,3 +4136,42 @@ LegalizerHelper::lowerExtract(MachineInstr &MI) {
 
   return UnableToLegalize;
 }
+
+LegalizerHelper::LegalizeResult LegalizerHelper::lowerInsert(MachineInstr &MI) {
+  Register Dst = MI.getOperand(0).getReg();
+  Register Src = MI.getOperand(1).getReg();
+  Register InsertSrc = MI.getOperand(2).getReg();
+  uint64_t Offset = MI.getOperand(3).getImm();
+
+  LLT DstTy = MRI.getType(Src);
+  LLT InsertTy = MRI.getType(InsertSrc);
+
+  if (InsertTy.isScalar() &&
+      (DstTy.isScalar() ||
+       (DstTy.isVector() && DstTy.getElementType() == InsertTy))) {
+    LLT IntDstTy = DstTy;
+    if (!DstTy.isScalar()) {
+      IntDstTy = LLT::scalar(DstTy.getSizeInBits());
+      Src = MIRBuilder.buildBitcast(IntDstTy, Src).getReg(0);
+    }
+
+    Register ExtInsSrc = MIRBuilder.buildZExt(IntDstTy, InsertSrc).getReg(0);
+    if (Offset != 0) {
+      auto ShiftAmt = MIRBuilder.buildConstant(IntDstTy, Offset);
+      ExtInsSrc = MIRBuilder.buildShl(IntDstTy, ExtInsSrc, ShiftAmt).getReg(0);
+    }
+
+    APInt MaskVal = ~APInt::getBitsSet(DstTy.getSizeInBits(), Offset,
+                                       InsertTy.getSizeInBits());
+
+    auto Mask = MIRBuilder.buildConstant(IntDstTy, MaskVal);
+    auto MaskedSrc = MIRBuilder.buildAnd(IntDstTy, Src, Mask);
+    auto Or = MIRBuilder.buildOr(IntDstTy, MaskedSrc, ExtInsSrc);
+
+    MIRBuilder.buildBitcast(Dst, Or);
+    MI.eraseFromParent();
+    return Legalized;
+  }
+
+  return UnableToLegalize;
+}
diff --git a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
index ce724e360ae..dfb8ed55d6b 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
@@ -912,13 +912,9 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST_,
     unsigned LitTyIdx = Op == G_EXTRACT ? 0 : 1;
 
     // FIXME: Doesn't handle extract of illegal sizes.
-    auto &Builder = getActionDefinitionsBuilder(Op);
-
-    // FIXME: Cleanup when G_INSERT lowering implemented.
-    if (Op == G_EXTRACT)
-      Builder.lowerIf(all(typeIs(LitTyIdx, S16), sizeIs(BigTyIdx, 32)));
-
-    Builder
+    getActionDefinitionsBuilder(Op)
+      .lowerIf(all(typeIs(LitTyIdx, S16), sizeIs(BigTyIdx, 32)))
+      // FIXME: Multiples of 16 should not be legal.
       .legalIf([=](const LegalityQuery &Query) {
           const LLT BigTy = Query.Types[BigTyIdx];
           const LLT LitTy = Query.Types[LitTyIdx];