AVX512: Combine AND + TESTM instrunctions

[igorb]

Summary

AVX512: Combine AND + TESTM instructions .

vpandq  %zmm0, %zmm1, %zmm2
vptestmq %zmm2, %zmm2, %k0

TO

vptestmq %zmm0, %zmm1, %k0

[delena]

LGTM

lib/Target/X86/X86ISelLowering.cpp

static SDValue combineLockSub(SDNode *N, SelectionDAG &DAG,
 
   RHS = DAG.getConstant(-1, DL, VT);
   MachineMemOperand *MMO = cast<MemSDNode>(N)->getMemOperand();
   return DAG.getMemIntrinsicNode(X86ISD::LADD, DL,
                                  DAG.getVTList(MVT::i32, MVT::Other),
                                  {Chain, LHS, RHS}, VT, MMO);
 }
 
+// TEST (AND a, b) ,(AND a, b) -> TEST a, b
+static SDValue PerformTESTM(SDNode *N, SelectionDAG &DAG) {
+  SDValue Op0 = N->getOperand(0);
+  SDValue Op1 = N->getOperand(1);
+
+  if (Op0 != Op1 || Op1->getOpcode() != ISD::AND)
+    return SDValue();
+
+  EVT VT = N->getValueType(0);
+  SDLoc DL(N);
+
+  return DAG.getNode(X86ISD::TESTM, DL, VT,
+                     Op0->getOperand(0), Op0->getOperand(1));
+}
+
 SDValue X86TargetLowering::PerformDAGCombine(SDNode *N,
                                              DAGCombinerInfo &DCI) const {
   SelectionDAG &DAG = DCI.DAG;
   switch (N->getOpcode()) {
   default: break;
   case ISD::EXTRACT_VECTOR_ELT: return combineExtractVectorElt(N, DAG, DCI);
   case ISD::VSELECT:
   case ISD::SELECT:
SDValue X86TargetLowering::PerformDAGCombine(SDNode *N,
   case X86ISD::MOVSD:
   case X86ISD::VPERMILPI:
   case X86ISD::VPERM2X128:
   case ISD::VECTOR_SHUFFLE: return combineShuffle(N, DAG, DCI,Subtarget);
   case ISD::FMA:            return combineFMA(N, DAG, Subtarget);
   case ISD::MGATHER:
   case ISD::MSCATTER:       return combineGatherScatter(N, DAG);
   case X86ISD::LSUB:        return combineLockSub(N, DAG, Subtarget);
+  case X86ISD::TESTM:       return PerformTESTM(N, DAG);
   }
 
   return SDValue();
 }
 
 /// Return true if the target has native support for the specified value type
 /// and it is 'desirable' to use the type for the given node type. e.g. On x86
 /// i16 is legal, but undesirable since i16 instruction encodings are longer and

lib/Target/X86/X86InstrAVX512.td

 defm VSCALEF : avx512_fp_scalef_all<0x2C, 0x2D, "vscalef", X86scalef>, T8PD;
 
 //===----------------------------------------------------------------------===//
 // AVX-512  VPTESTM instructions
 //===----------------------------------------------------------------------===//
 
 multiclass avx512_vptest<bits<8> opc, string OpcodeStr, SDNode OpNode,
                             X86VectorVTInfo _> {
+  let isCommutable = 1 in
   defm rr : AVX512_maskable_cmp<opc, MRMSrcReg, _, (outs _.KRC:$dst),
                    (ins _.RC:$src1, _.RC:$src2), OpcodeStr,
                       "$src2, $src1", "$src1, $src2",
                    (OpNode (_.VT _.RC:$src1), (_.VT _.RC:$src2))>,
                     EVEX_4V;
   let mayLoad = 1 in
   defm rm : AVX512_maskable_cmp<opc, MRMSrcMem, _, (outs _.KRC:$dst),
                    (ins _.RC:$src1, _.MemOp:$src2), OpcodeStr,

lib/Target/X86/X86InstrFragmentsSIMD.td

 def X86avg     : SDNode<"X86ISD::AVG" , SDTIntBinOp>;
 def X86ptest   : SDNode<"X86ISD::PTEST", SDTX86CmpPTest>;
 def X86testp   : SDNode<"X86ISD::TESTP", SDTX86CmpPTest>;
 def X86kortest : SDNode<"X86ISD::KORTEST", SDTX86CmpPTest>;
 def X86ktest   : SDNode<"X86ISD::KTEST", SDTX86CmpPTest>;
 def X86testm   : SDNode<"X86ISD::TESTM", SDTypeProfile<1, 2, [SDTCisVec<0>,
                                           SDTCisVec<1>, SDTCisSameAs<2, 1>,
                                           SDTCVecEltisVT<0, i1>,
-                                          SDTCisSameNumEltsAs<0, 1>]>>;
+                                          SDTCisSameNumEltsAs<0, 1>]>, [SDNPCommutative]>;
 def X86testnm  : SDNode<"X86ISD::TESTNM", SDTypeProfile<1, 2, [SDTCisVec<0>,
                                           SDTCisVec<1>, SDTCisSameAs<2, 1>,
                                           SDTCVecEltisVT<0, i1>,
                                           SDTCisSameNumEltsAs<0, 1>]>>;
 def X86select  : SDNode<"X86ISD::SELECT"     , SDTSelect>;
 
 def X86pmuludq : SDNode<"X86ISD::PMULUDQ",
                         SDTypeProfile<1, 2, [SDTCVecEltisVT<0, i64>,

test/CodeGen/X86/combine-testm-and.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+;RUN: llc -mtriple=x86_64-apple-darwin -mcpu=skx < %s | FileCheck %s
+
+define i32 @combineTESTM_AND_1(<8 x i64> %a, <8 x i64> %b) {
+; CHECK-LABEL: combineTESTM_AND_1:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vptestmq %zmm0, %zmm1, %k0
+; CHECK-NEXT:    kmovb %k0, %eax
+; CHECK-NEXT:    movzbl %al, %eax
+; CHECK-NEXT:    retq
+  %and.i = and <8 x i64> %b, %a
+  %test.i = tail call i8 @llvm.x86.avx512.ptestm.q.512(<8 x i64> %and.i, <8 x i64> %and.i, i8 -1)
+  %conv = zext i8 %test.i to i32
+  ret i32 %conv
+}
+
+define i32 @combineTESTM_AND_2(<8 x i64> %a, <8 x i64> %b , i8 %mask) {
+; CHECK-LABEL: combineTESTM_AND_2:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovb %edi, %k1
+; CHECK-NEXT:    vptestmq %zmm0, %zmm1, %k0 {%k1}
+; CHECK-NEXT:    kmovb %k0, %eax
+; CHECK-NEXT:    movzbl %al, %eax
+; CHECK-NEXT:    retq
+  %and.i = and <8 x i64> %b, %a
+  %test.i = tail call i8 @llvm.x86.avx512.ptestm.q.512(<8 x i64> %and.i, <8 x i64> %and.i, i8 %mask)
+  %conv = zext i8 %test.i to i32
+  ret i32 %conv
+}
+
+define i32 @combineTESTM_AND_mask_3(<8 x i64> %a, <8 x i64>* %bptr , i8 %mask) {
+; CHECK-LABEL: combineTESTM_AND_mask_3:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovb %esi, %k1
+; CHECK-NEXT:    vptestmq (%rdi), %zmm0, %k0 {%k1}
+; CHECK-NEXT:    kmovb %k0, %eax
+; CHECK-NEXT:    movzbl %al, %eax
+; CHECK-NEXT:    retq
+  %b = load <8 x i64>, <8 x i64>* %bptr
+  %and.i = and <8 x i64> %a, %b
+  %test.i = tail call i8 @llvm.x86.avx512.ptestm.q.512(<8 x i64> %and.i, <8 x i64> %and.i, i8 %mask)
+  %conv = zext i8 %test.i to i32
+  ret i32 %conv
+}
+
+define i32 @combineTESTM_AND_mask_4(<8 x i64> %a, <8 x i64>* %bptr , i8 %mask) {
+; CHECK-LABEL: combineTESTM_AND_mask_4:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovb %esi, %k1
+; CHECK-NEXT:    vptestmq (%rdi), %zmm0, %k0 {%k1}
+; CHECK-NEXT:    kmovb %k0, %eax
+; CHECK-NEXT:    movzbl %al, %eax
+; CHECK-NEXT:    retq
+  %b = load <8 x i64>, <8 x i64>* %bptr
+  %and.i = and <8 x i64> %b, %a
+  %test.i = tail call i8 @llvm.x86.avx512.ptestm.q.512(<8 x i64> %and.i, <8 x i64> %and.i, i8 %mask)
+  %conv = zext i8 %test.i to i32
+  ret i32 %conv
+}
+
+declare i8 @llvm.x86.avx512.ptestm.q.512(<8 x i64>, <8 x i64>, i8)