[ARM] Limit and+csinc fold to a single use.

With glue nodes between the CMP and the CSINC, generating multiple
uses of the CMP can fail when scheduling the DAG. This limits the fold
from 90f24bef47227d58f2ccd to a single use on the CSINC, to prevent the
CMP being needed by two nodes.

diff --git a/llvm/lib/Target/ARM/ARMInstrThumb2.td b/llvm/lib/Target/ARM/ARMInstrThumb2.td
index f06f9f1b99b57dff663b083a8c0039929235c22a..d8721cbebd2dc25566ff4c89594d1f35221e797a 100644 (file)
--- a/llvm/lib/Target/ARM/ARMInstrThumb2.td
+++ b/llvm/lib/Target/ARM/ARMInstrThumb2.td
@@ -5621,6 +5621,10 @@ def t2CSINC : CS<"csinc", 0b1001>;
 def t2CSINV : CS<"csinv", 0b1010>;
 def t2CSNEG : CS<"csneg", 0b1011>;
 
+def ARMcsinc_su : PatFrag<(ops node:$lhs, node:$rhs, node:$cond),
+                          (ARMcsinc node:$lhs, node:$rhs, node:$cond), [{
+  return N->hasOneUse();
+}]>;
 
 let Predicates = [HasV8_1MMainline] in {
   multiclass CSPats<SDNode Node, Instruction Insn> {
@@ -5660,7 +5664,7 @@ let Predicates = [HasV8_1MMainline] in {
 
   def : T2Pat<(ARMcmov (topbitsallzero32:$Rn), (i32 1), cmovpred:$imm),
               (t2CSINC $Rn, ZR, $imm)>;
-  def : T2Pat<(and (topbitsallzero32:$Rn), (ARMcsinc (i32 0), (i32 0), cmovpred:$imm)),
+  def : T2Pat<(and (topbitsallzero32:$Rn), (ARMcsinc_su (i32 0), (i32 0), cmovpred:$imm)),
               (t2CSEL ZR, $Rn, $imm)>;
 }
 

diff --git a/llvm/test/CodeGen/Thumb2/csel-andor-onebit.ll b/llvm/test/CodeGen/Thumb2/csel-andor-onebit.ll
index d9e18b5c0942e8c039bdbf3b9db0958ac5241a37..759a8169389cca74a2785cbb2d1a873883286373 100644 (file)
--- a/llvm/test/CodeGen/Thumb2/csel-andor-onebit.ll
+++ b/llvm/test/CodeGen/Thumb2/csel-andor-onebit.ll
@@ -170,3 +170,28 @@ define i64 @andi64_ne(i8 %x, i8 %y) {
   %a = and i64 %xa, %cz
   ret i64 %a
 }
+
+; Check for multiple uses on the csinc
+define i32 @t5(i32 %f.0, i32 %call) {
+; CHECK-LABEL: t5:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    cmp r1, #0
+; CHECK-NEXT:    cset r1, ne
+; CHECK-NEXT:    cmp r0, #13
+; CHECK-NEXT:    cset r0, eq
+; CHECK-NEXT:    and.w r2, r0, r1
+; CHECK-NEXT:    orrs r0, r1
+; CHECK-NEXT:    eor r0, r0, #1
+; CHECK-NEXT:    orrs r0, r2
+; CHECK-NEXT:    bx lr
+entry:
+  %tobool1.i = icmp ne i32 %call, 0
+  %cmp = icmp eq i32 %f.0, 13
+  %or.cond = select i1 %cmp, i1 %tobool1.i, i1 false
+  %or.cond7.not = select i1 %cmp, i1 true, i1 %tobool1.i
+  %or.cond7.not.not = xor i1 %or.cond7.not, true
+  %not.or.cond12 = select i1 %or.cond, i1 true, i1 %or.cond7.not.not
+  %g.0 = zext i1 %not.or.cond12 to i32
+  ret i32 %g.0
+}
+