[TargetLowering] SimplifyDemandedBits ANY_EXTEND/ANY_EXTEND_VECTOR_INREG multi-use handling

RKSimon · RKSimon · commit 651fa669a237 · 2020-01-21T14:07:19.000Z
Call SimplifyMultipleUseDemandedBits to peek through extended source args with multiple uses
diff --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -1779,6 +1779,11 @@ bool TargetLowering::SimplifyDemandedBits(
     assert(!Known.hasConflict() && "Bits known to be one AND zero?");
     assert(Known.getBitWidth() == InBits && "Src width has changed?");
     Known = Known.zext(BitWidth, false /* => any extend */);
+
+    // Attempt to avoid multi-use ops if we don't need anything from them.
+    if (SDValue NewSrc = SimplifyMultipleUseDemandedBits(
+            Src, InDemandedBits, InDemandedElts, TLO.DAG, Depth + 1))
+      return TLO.CombineTo(Op, TLO.DAG.getNode(Op.getOpcode(), dl, VT, NewSrc));
     break;
   }
   case ISD::TRUNCATE: {
diff --git a/llvm/test/CodeGen/X86/vector-sext.ll b/llvm/test/CodeGen/X86/vector-sext.ll
@@ -1748,11 +1748,11 @@ define <4 x i64> @load_sext_4i1_to_4i64(<4 x i1> *%ptr) {
 ; SSE41-NEXT:    pinsrd $1, %ecx, %xmm1
 ; SSE41-NEXT:    movl %eax, %ecx
 ; SSE41-NEXT:    shrl $2, %ecx
+; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm0 = xmm1[0],zero,xmm1[1],zero
 ; SSE41-NEXT:    pinsrd $2, %ecx, %xmm1
 ; SSE41-NEXT:    shrl $3, %eax
 ; SSE41-NEXT:    pinsrd $3, %eax, %xmm1
 ; SSE41-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm0 = xmm1[0],zero,xmm1[1],zero
 ; SSE41-NEXT:    psllq $63, %xmm0
 ; SSE41-NEXT:    psrad $31, %xmm0
 ; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
@@ -1851,11 +1851,11 @@ define <4 x i64> @load_sext_4i1_to_4i64(<4 x i1> *%ptr) {
 ; X32-SSE41-NEXT:    pinsrd $1, %ecx, %xmm1
 ; X32-SSE41-NEXT:    movl %eax, %ecx
 ; X32-SSE41-NEXT:    shrl $2, %ecx
+; X32-SSE41-NEXT:    pmovzxdq {{.*#+}} xmm0 = xmm1[0],zero,xmm1[1],zero
 ; X32-SSE41-NEXT:    pinsrd $2, %ecx, %xmm1
 ; X32-SSE41-NEXT:    shrl $3, %eax
 ; X32-SSE41-NEXT:    pinsrd $3, %eax, %xmm1
 ; X32-SSE41-NEXT:    pand {{\.LCPI.*}}, %xmm1
-; X32-SSE41-NEXT:    pmovzxdq {{.*#+}} xmm0 = xmm1[0],zero,xmm1[1],zero
 ; X32-SSE41-NEXT:    psllq $63, %xmm0
 ; X32-SSE41-NEXT:    psrad $31, %xmm0
 ; X32-SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
