Use m_ReassociatableAdd now that llvm#170061 is fixed

Author: RKSimon

llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

@@ -3166,25 +3166,24 @@ SDValue DAGCombiner::foldAddToAvg(SDNode *N, const SDLoc &DL) {
   EVT VT = N0.getValueType();
   SDValue A, B;
 
-  // FIXME: m_ReassociatableAdd can't handle m_Value/m_Deferred mixing.
   if ((!LegalOperations || hasOperation(ISD::AVGFLOORU, VT)) &&
       (sd_match(N,
                 m_Add(m_And(m_Value(A), m_Value(B)),
                       m_Srl(m_Xor(m_Deferred(A), m_Deferred(B)), m_One()))) ||
-       sd_match(N, m_Add(m_Add(m_Srl(m_Value(A), m_One()),
-                               m_Srl(m_Value(B), m_One())),
-                         m_ReassociatableAnd(m_Deferred(A), m_Deferred(B),
-                                             m_One()))))) {
+       sd_match(N, m_ReassociatableAdd(
+                       m_Srl(m_Value(A), m_One()), m_Srl(m_Value(B), m_One()),
+                       m_ReassociatableAnd(m_Deferred(A), m_Deferred(B),
+                                           m_One()))))) {
     return DAG.getNode(ISD::AVGFLOORU, DL, VT, A, B);
   }
   if ((!LegalOperations || hasOperation(ISD::AVGFLOORS, VT)) &&
       (sd_match(N,
                 m_Add(m_And(m_Value(A), m_Value(B)),
                       m_Sra(m_Xor(m_Deferred(A), m_Deferred(B)), m_One()))) ||
-       sd_match(N, m_Add(m_Add(m_Sra(m_Value(A), m_One()),
-                               m_Sra(m_Value(B), m_One())),
-                         m_ReassociatableAnd(m_Deferred(A), m_Deferred(B),
-                                             m_One()))))) {
+       sd_match(N, m_ReassociatableAdd(
+                       m_Sra(m_Value(A), m_One()), m_Sra(m_Value(B), m_One()),
+                       m_ReassociatableAnd(m_Deferred(A), m_Deferred(B),
+                                           m_One()))))) {
     return DAG.getNode(ISD::AVGFLOORS, DL, VT, A, B);
   }
 

llvm/test/CodeGen/X86/avgfloors-scalar.ll

@@ -199,23 +199,19 @@ define i32 @test_lsb_i32(i32 %a0, i32 %a1) nounwind {
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movl %eax, %edx
-; X86-NEXT:    sarl %edx
-; X86-NEXT:    andl %ecx, %eax
-; X86-NEXT:    sarl %ecx
-; X86-NEXT:    addl %edx, %ecx
-; X86-NEXT:    andl $1, %eax
-; X86-NEXT:    addl %ecx, %eax
+; X86-NEXT:    andl %ecx, %edx
+; X86-NEXT:    xorl %ecx, %eax
+; X86-NEXT:    sarl %eax
+; X86-NEXT:    addl %edx, %eax
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_lsb_i32:
 ; X64:       # %bb.0:
-; X64-NEXT:    movl %edi, %eax
-; X64-NEXT:    sarl %eax
-; X64-NEXT:    andl %esi, %edi
-; X64-NEXT:    sarl %esi
-; X64-NEXT:    addl %esi, %eax
-; X64-NEXT:    andl $1, %edi
-; X64-NEXT:    addl %edi, %eax
+; X64-NEXT:    movslq %esi, %rcx
+; X64-NEXT:    movslq %edi, %rax
+; X64-NEXT:    addq %rcx, %rax
+; X64-NEXT:    shrq %rax
+; X64-NEXT:    # kill: def $eax killed $eax killed $rax
 ; X64-NEXT:    retq
   %s0 = ashr i32 %a0, 1
   %s1 = ashr i32 %a1, 1

llvm/test/CodeGen/X86/avgflooru-scalar.ll

@@ -199,23 +199,19 @@ define i32 @test_lsb_i32(i32 %a0, i32 %a1) nounwind {
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movl %eax, %edx
-; X86-NEXT:    shrl %edx
-; X86-NEXT:    andl %ecx, %eax
-; X86-NEXT:    shrl %ecx
-; X86-NEXT:    addl %edx, %ecx
-; X86-NEXT:    andl $1, %eax
-; X86-NEXT:    addl %ecx, %eax
+; X86-NEXT:    andl %ecx, %edx
+; X86-NEXT:    xorl %ecx, %eax
+; X86-NEXT:    shrl %eax
+; X86-NEXT:    addl %edx, %eax
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_lsb_i32:
 ; X64:       # %bb.0:
+; X64-NEXT:    movl %esi, %ecx
 ; X64-NEXT:    movl %edi, %eax
-; X64-NEXT:    shrl %eax
-; X64-NEXT:    andl %esi, %edi
-; X64-NEXT:    shrl %esi
-; X64-NEXT:    addl %esi, %eax
-; X64-NEXT:    andl $1, %edi
-; X64-NEXT:    addl %edi, %eax
+; X64-NEXT:    addq %rcx, %rax
+; X64-NEXT:    shrq %rax
+; X64-NEXT:    # kill: def $eax killed $eax killed $rax
 ; X64-NEXT:    retq
   %s0 = lshr i32 %a0, 1
   %s1 = lshr i32 %a1, 1