[SimplifyDemandedVec] Strengthen handling all undef lanes (particular…

…ly GEPs)

A change of two parts:
1) A generic enhancement for all callers of SDVE to exploit the fact that if all lanes are undef, the result is undef.
2) A GEP specific piece to strengthen/fix the vector index undef element handling, and call into the generic infrastructure when visiting the GEP.

The result is that we replace a vector gep with at least one undef in each lane with a undef.  We can also do the same for vector intrinsics.  Once the masked.load patch (D57372) has landed, I'll update to include call tests as well.

Differential Revision: https://reviews.llvm.org/D57468

llvm-svn: 356293

llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp

@@ -1908,8 +1908,7 @@ Instruction *InstCombiner::visitCallInst(CallInst &CI) {
     if (Changed) return II;
   }
 
-  // For vector result intrinsics, use the generic demanded vector support to
-  // simplify any operands before moving on to the per-intrinsic rules.    
+  // For vector result intrinsics, use the generic demanded vector support.
   if (II->getType()->isVectorTy()) {
     auto VWidth = II->getType()->getVectorNumElements();
     APInt UndefElts(VWidth, 0);

llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp

@@ -1175,9 +1175,18 @@ Value *InstCombiner::SimplifyDemandedVectorElts(Value *V, APInt DemandedElts,
     // wouldn't have a vector result to get here. Note that we intentionally
     // merge the undef bits here since gepping with either an undef base or
     // index results in undef. 
-    for (unsigned i = 0; i < I->getNumOperands(); i++)
-      if (I->getOperand(i)->getType()->isVectorTy())
-        simplifyAndSetOp(I, i, DemandedElts, UndefElts);
+    for (unsigned i = 0; i < I->getNumOperands(); i++) {
+      if (isa<UndefValue>(I->getOperand(i))) {
+        // If the entire vector is undefined, just return this info.
+        UndefElts = EltMask;
+        return nullptr;
+      }
+      if (I->getOperand(i)->getType()->isVectorTy()) {
+        APInt UndefEltsOp(VWidth, 0);
+        simplifyAndSetOp(I, i, DemandedElts, UndefEltsOp);
+        UndefElts |= UndefEltsOp;
+      }
+    }
 
     break;
   }
@@ -1663,5 +1672,10 @@ Value *InstCombiner::SimplifyDemandedVectorElts(Value *V, APInt DemandedElts,
     UndefElts &= UndefElts2;
   }
 
+  // If we've proven all of the lanes undef, return an undef value.
+  // TODO: Intersect w/demanded lanes
+  if (UndefElts.isAllOnesValue())
+    return UndefValue::get(I->getType());;
+
   return MadeChange ? I : nullptr;
 }

llvm/lib/Transforms/InstCombine/InstructionCombining.cpp

@@ -1557,6 +1557,19 @@ Instruction *InstCombiner::visitGetElementPtrInst(GetElementPtrInst &GEP) {
   if (Value *V = SimplifyGEPInst(GEPEltType, Ops, SQ.getWithInstruction(&GEP)))
     return replaceInstUsesWith(GEP, V);
 
+  // For vector geps, use the generic demanded vector support.
+  if (GEP.getType()->isVectorTy()) {
+    auto VWidth = GEP.getType()->getVectorNumElements();
+    APInt UndefElts(VWidth, 0);
+    APInt AllOnesEltMask(APInt::getAllOnesValue(VWidth));
+    if (Value *V = SimplifyDemandedVectorElts(&GEP, AllOnesEltMask,
+                                              UndefElts)) {
+      if (V != &GEP)
+        return replaceInstUsesWith(GEP, V);
+      return &GEP;
+    }
+  }
+
   Value *PtrOp = GEP.getOperand(0);
 
   // Eliminate unneeded casts for indices, and replace indices which displace

llvm/test/Transforms/InstCombine/vec_demanded_elts.ll

@@ -620,10 +620,7 @@ define i32* @gep_splat_both(i32* %base, i64 %idx) {
 
 define <2 x i32*> @gep_all_lanes_undef(i32* %base, i64 %idx) {;
 ; CHECK-LABEL: @gep_all_lanes_undef(
-; CHECK-NEXT:    [[BASEVEC:%.*]] = insertelement <2 x i32*> undef, i32* [[BASE:%.*]], i32 0
-; CHECK-NEXT:    [[IDXVEC:%.*]] = insertelement <2 x i64> undef, i64 [[IDX:%.*]], i32 1
-; CHECK-NEXT:    [[GEP:%.*]] = getelementptr i32, <2 x i32*> [[BASEVEC]], <2 x i64> [[IDXVEC]]
-; CHECK-NEXT:    ret <2 x i32*> [[GEP]]
+; CHECK-NEXT:    ret <2 x i32*> undef
 ;
   %basevec = insertelement <2 x i32*> undef, i32* %base, i32 0
   %idxvec = insertelement <2 x i64> undef, i64 %idx, i32 1
@@ -641,4 +638,3 @@ define i32* @gep_demanded_lane_undef(i32* %base, i64 %idx) {
   %ee = extractelement <2 x i32*> %gep, i32 1
   ret i32* %ee
 }
-