[DAG] foldAddToAvg - optimize nested m_Reassociatable matchers (#171681)

The use of nested m_Reassociatable matchers by #169644 can result in
high compile times as the inner m_Reassociatable call is being repeated
a lot while the outer call is trying to match. Place the inner
m_ReassociatableAnd at the beginning of the pattern so it is not
repeatedly matched in recursion.

Author: RKSimon

llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

@@ -3171,19 +3171,19 @@ SDValue DAGCombiner::foldAddToAvg(SDNode *N, const SDLoc &DL) {
                 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_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()))))) {
+                       m_ReassociatableAnd(m_Value(A), m_Value(B), m_One()),
+                       m_Srl(m_Deferred(A), m_One()),
+                       m_Srl(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_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()))))) {
+                       m_ReassociatableAnd(m_Value(A), m_Value(B), m_One()),
+                       m_Sra(m_Deferred(A), m_One()),
+                       m_Sra(m_Deferred(B), m_One()))))) {
     return DAG.getNode(ISD::AVGFLOORS, DL, VT, A, B);
   }
 

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

@@ -317,11 +317,11 @@ define i64 @test_lsb_i64(i64 %a0, i64 %a1) nounwind {
 ;
 ; X64-LABEL: test_lsb_i64:
 ; X64:       # %bb.0:
-; X64-NEXT:    movq %rsi, %rax
-; X64-NEXT:    andq %rdi, %rax
-; X64-NEXT:    xorq %rdi, %rsi
-; X64-NEXT:    sarq %rsi
-; X64-NEXT:    addq %rsi, %rax
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    andq %rsi, %rax
+; X64-NEXT:    xorq %rsi, %rdi
+; X64-NEXT:    sarq %rdi
+; X64-NEXT:    addq %rdi, %rax
 ; X64-NEXT:    retq
   %s0 = ashr i64 %a0, 1
   %s1 = ashr i64 %a1, 1

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

@@ -293,11 +293,11 @@ define i64 @test_lsb_i64(i64 %a0, i64 %a1) nounwind {
 ;
 ; X64-LABEL: test_lsb_i64:
 ; X64:       # %bb.0:
-; X64-NEXT:    movq %rsi, %rax
-; X64-NEXT:    andq %rdi, %rax
-; X64-NEXT:    xorq %rdi, %rsi
-; X64-NEXT:    shrq %rsi
-; X64-NEXT:    addq %rsi, %rax
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    andq %rsi, %rax
+; X64-NEXT:    xorq %rsi, %rdi
+; X64-NEXT:    shrq %rdi
+; X64-NEXT:    addq %rdi, %rax
 ; X64-NEXT:    retq
   %s0 = lshr i64 %a0, 1
   %s1 = lshr i64 %a1, 1