[X86][XOP] Merge rotation opcodes with AVX512 equivalents. NFCI.

[RKSimon]

Summary

The XOP rotations act as ROTL with +ve values and ROTR with -ve values, which means that we can treat them all as ROTL with unsigned modulo. We already check that we're only trying to lower as ROTL for XOP rotations.

[craig.topper]

lib/Target/X86/X86InstrFragmentsSIMD.td

215–216

Can we just replace all uses of this with X86vrotli?

[RKSimon]

lib/Target/X86/X86InstrFragmentsSIMD.td

215–216

We should - I hit a load of tblgen failures when I tried this in an earlier version of the patch and sort of brushed the problem under the carpet, I'll try again.

[RKSimon]

Removed X86vprot + X86vproti fragments

[craig.topper]

LGTM

lib/Target/X86/X86ISelLowering.h

enum NodeType : unsigned {
       SUBV_BROADCAST,
 
       // Extract vector element.
       VEXTRACT,
 
       /// SSE4A Extraction and Insertion.
       EXTRQI, INSERTQI,
 
-      // XOP variable/immediate rotations.
-      VPROT, VPROTI,
       // XOP arithmetic/logical shifts.
       VPSHA, VPSHL,
       // XOP signed/unsigned integer comparisons.
       VPCOM, VPCOMU,
       // XOP packed permute bytes.
       VPPERM,
       // XOP two source permutation.
       VPERMIL2,

lib/Target/X86/X86ISelLowering.cpp

if (Subtarget.hasAVX512()) {
     return Op;
   }
 
   assert(VT.isVector() && "Custom lowering only for vector rotates!");
   assert(Subtarget.hasXOP() && "XOP support required for vector rotates!");
   assert((Opcode == ISD::ROTL) && "Only ROTL supported");
 
   // XOP has 128-bit vector variable + immediate rotates.
-  // +ve/-ve Amt = rotate left/right.
+  // +ve/-ve Amt = rotate left/right - just need to handle ISD::ROTL.
 
   // Split 256-bit integers.
   if (VT.is256BitVector())
     return Lower256IntArith(Op, DAG);
 
   assert(VT.is128BitVector() && "Only rotate 128-bit vectors!");
 
   // Attempt to rotate by immediate.
   if (auto *BVAmt = dyn_cast<BuildVectorSDNode>(Amt)) {
     if (auto *RotateConst = BVAmt->getConstantSplatNode()) {
       uint64_t RotateAmt = RotateConst->getAPIntValue().getZExtValue();
       assert(RotateAmt < EltSizeInBits && "Rotation out of range");
-      return DAG.getNode(X86ISD::VPROTI, DL, VT, R,
+      return DAG.getNode(X86ISD::VROTLI, DL, VT, R,
                          DAG.getConstant(RotateAmt, DL, MVT::i8));
     }
   }
 
   // Use general rotate by variable (per-element).
-  return DAG.getNode(X86ISD::VPROT, DL, VT, R, Amt);
+  return Op;
 }
 
 static SDValue LowerXALUO(SDValue Op, SelectionDAG &DAG) {
   // Lower the "add/sub/mul with overflow" instruction into a regular ins plus
   // a "setcc" instruction that checks the overflow flag. The "brcond" lowering
   // looks for this combo and may remove the "setcc" instruction if the "setcc"
   // has only one use.
   SDNode *N = Op.getNode();
const char *X86TargetLowering::getTargetNodeName(unsigned Opcode) const {
   case X86ISD::MEMBARRIER:         return "X86ISD::MEMBARRIER";
   case X86ISD::MFENCE:             return "X86ISD::MFENCE";
   case X86ISD::SEG_ALLOCA:         return "X86ISD::SEG_ALLOCA";
   case X86ISD::SAHF:               return "X86ISD::SAHF";
   case X86ISD::RDRAND:             return "X86ISD::RDRAND";
   case X86ISD::RDSEED:             return "X86ISD::RDSEED";
   case X86ISD::VPMADDUBSW:         return "X86ISD::VPMADDUBSW";
   case X86ISD::VPMADDWD:           return "X86ISD::VPMADDWD";
-  case X86ISD::VPROT:              return "X86ISD::VPROT";
-  case X86ISD::VPROTI:             return "X86ISD::VPROTI";
   case X86ISD::VPSHA:              return "X86ISD::VPSHA";
   case X86ISD::VPSHL:              return "X86ISD::VPSHL";
   case X86ISD::VPCOM:              return "X86ISD::VPCOM";
   case X86ISD::VPCOMU:             return "X86ISD::VPCOMU";
   case X86ISD::VPERMIL2:           return "X86ISD::VPERMIL2";
   case X86ISD::FMSUB:              return "X86ISD::FMSUB";
   case X86ISD::FNMADD:             return "X86ISD::FNMADD";
   case X86ISD::FNMSUB:             return "X86ISD::FNMSUB";

lib/Target/X86/X86InstrFragmentsSIMD.td

def X86kshiftl : SDNode<"X86ISD::KSHIFTL",
                                              SDTCisVT<2, i8>]>>;
 def X86kshiftr : SDNode<"X86ISD::KSHIFTR",
                         SDTypeProfile<1, 2, [SDTCVecEltisVT<0, i1>,
                                              SDTCisSameAs<0, 1>,
                                              SDTCisVT<2, i8>]>>;
 
 def X86vrotli  : SDNode<"X86ISD::VROTLI", X86vshiftimm>;
 def X86vrotri  : SDNode<"X86ISD::VROTRI", X86vshiftimm>;
 
-def X86vprot   : SDNode<"X86ISD::VPROT", X86vshiftvariable>;
-def X86vproti  : SDNode<"X86ISD::VPROTI", X86vshiftimm>;
-
 def X86vpshl   : SDNode<"X86ISD::VPSHL", X86vshiftvariable>;
 def X86vpsha   : SDNode<"X86ISD::VPSHA", X86vshiftvariable>;
 
 def X86vpcom   : SDNode<"X86ISD::VPCOM",
                         SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
                                              SDTCisSameAs<0,2>,
                                              SDTCisVT<3, i8>, SDTCisInt<0>]>>;
 def X86vpcomu  : SDNode<"X86ISD::VPCOMU",
                         SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,

lib/Target/X86/X86InstrXOP.td

multiclass xop3op<bits<8> opc, string OpcodeStr, SDNode OpNode,
   def rr_REV : IXOP<opc, MRMSrcReg, (outs VR128:$dst),
                (ins VR128:$src1, VR128:$src2),
                !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
                []>,
                XOP_4V, VEX_W, Sched<[WriteVarVecShift]>, FoldGenData<NAME#rr>;
 }
 
 let ExeDomain = SSEPackedInt in {
-  defm VPROTB : xop3op<0x90, "vprotb", X86vprot, v16i8>;
-  defm VPROTD : xop3op<0x92, "vprotd", X86vprot, v4i32>;
-  defm VPROTQ : xop3op<0x93, "vprotq", X86vprot, v2i64>;
-  defm VPROTW : xop3op<0x91, "vprotw", X86vprot, v8i16>;
+  defm VPROTB : xop3op<0x90, "vprotb", rotl, v16i8>;
+  defm VPROTD : xop3op<0x92, "vprotd", rotl, v4i32>;
+  defm VPROTQ : xop3op<0x93, "vprotq", rotl, v2i64>;
+  defm VPROTW : xop3op<0x91, "vprotw", rotl, v8i16>;
   defm VPSHAB : xop3op<0x98, "vpshab", X86vpsha, v16i8>;
   defm VPSHAD : xop3op<0x9A, "vpshad", X86vpsha, v4i32>;
   defm VPSHAQ : xop3op<0x9B, "vpshaq", X86vpsha, v2i64>;
   defm VPSHAW : xop3op<0x99, "vpshaw", X86vpsha, v8i16>;
   defm VPSHLB : xop3op<0x94, "vpshlb", X86vpshl, v16i8>;
   defm VPSHLD : xop3op<0x96, "vpshld", X86vpshl, v4i32>;
   defm VPSHLQ : xop3op<0x97, "vpshlq", X86vpshl, v2i64>;
   defm VPSHLW : xop3op<0x95, "vpshlw", X86vpshl, v8i16>;
multiclass xop3opimm<bits<8> opc, string OpcodeStr, SDNode OpNode,
   def mi : IXOPi8<opc, MRMSrcMem, (outs VR128:$dst),
            (ins i128mem:$src1, u8imm:$src2),
            !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
            [(set VR128:$dst,
               (vt128 (OpNode (vt128 (bitconvert (loadv2i64 addr:$src1))), imm:$src2)))]>, XOP;
 }
 
 let ExeDomain = SSEPackedInt in {
-  defm VPROTB : xop3opimm<0xC0, "vprotb", X86vproti, v16i8>;
-  defm VPROTD : xop3opimm<0xC2, "vprotd", X86vproti, v4i32>;
-  defm VPROTQ : xop3opimm<0xC3, "vprotq", X86vproti, v2i64>;
-  defm VPROTW : xop3opimm<0xC1, "vprotw", X86vproti, v8i16>;
+  defm VPROTB : xop3opimm<0xC0, "vprotb", X86vrotli, v16i8>;
+  defm VPROTD : xop3opimm<0xC2, "vprotd", X86vrotli, v4i32>;
+  defm VPROTQ : xop3opimm<0xC3, "vprotq", X86vrotli, v2i64>;
+  defm VPROTW : xop3opimm<0xC1, "vprotw", X86vrotli, v8i16>;
 }
 
 // Instruction where second source can be memory, but third must be register
 multiclass xop4opm2<bits<8> opc, string OpcodeStr, Intrinsic Int> {
   let isCommutable = 1 in
   def rr : IXOPi8Reg<opc, MRMSrcReg, (outs VR128:$dst),
            (ins VR128:$src1, VR128:$src2, VR128:$src3),
            !strconcat(OpcodeStr,

lib/Target/X86/X86IntrinsicsInfo.h

   X86_INTRINSIC_DATA(xop_vpcomuq,       INTR_TYPE_3OP, X86ISD::VPCOMU, 0),
   X86_INTRINSIC_DATA(xop_vpcomuw,       INTR_TYPE_3OP, X86ISD::VPCOMU, 0),
   X86_INTRINSIC_DATA(xop_vpcomw,        INTR_TYPE_3OP, X86ISD::VPCOM, 0),
   X86_INTRINSIC_DATA(xop_vpermil2pd,     INTR_TYPE_4OP, X86ISD::VPERMIL2, 0),
   X86_INTRINSIC_DATA(xop_vpermil2pd_256, INTR_TYPE_4OP, X86ISD::VPERMIL2, 0),
   X86_INTRINSIC_DATA(xop_vpermil2ps,     INTR_TYPE_4OP, X86ISD::VPERMIL2, 0),
   X86_INTRINSIC_DATA(xop_vpermil2ps_256, INTR_TYPE_4OP, X86ISD::VPERMIL2, 0),
   X86_INTRINSIC_DATA(xop_vpperm,        INTR_TYPE_3OP, X86ISD::VPPERM, 0),
-  X86_INTRINSIC_DATA(xop_vprotb,        INTR_TYPE_2OP, X86ISD::VPROT, 0),
-  X86_INTRINSIC_DATA(xop_vprotbi,       INTR_TYPE_2OP, X86ISD::VPROTI, 0),
-  X86_INTRINSIC_DATA(xop_vprotd,        INTR_TYPE_2OP, X86ISD::VPROT, 0),
-  X86_INTRINSIC_DATA(xop_vprotdi,       INTR_TYPE_2OP, X86ISD::VPROTI, 0),
-  X86_INTRINSIC_DATA(xop_vprotq,        INTR_TYPE_2OP, X86ISD::VPROT, 0),
-  X86_INTRINSIC_DATA(xop_vprotqi,       INTR_TYPE_2OP, X86ISD::VPROTI, 0),
-  X86_INTRINSIC_DATA(xop_vprotw,        INTR_TYPE_2OP, X86ISD::VPROT, 0),
-  X86_INTRINSIC_DATA(xop_vprotwi,       INTR_TYPE_2OP, X86ISD::VPROTI, 0),
+  X86_INTRINSIC_DATA(xop_vprotb,        INTR_TYPE_2OP, ISD::ROTL, 0),
+  X86_INTRINSIC_DATA(xop_vprotbi,       INTR_TYPE_2OP, X86ISD::VROTLI, 0),
+  X86_INTRINSIC_DATA(xop_vprotd,        INTR_TYPE_2OP, ISD::ROTL, 0),
+  X86_INTRINSIC_DATA(xop_vprotdi,       INTR_TYPE_2OP, X86ISD::VROTLI, 0),
+  X86_INTRINSIC_DATA(xop_vprotq,        INTR_TYPE_2OP, ISD::ROTL, 0),
+  X86_INTRINSIC_DATA(xop_vprotqi,       INTR_TYPE_2OP, X86ISD::VROTLI, 0),
+  X86_INTRINSIC_DATA(xop_vprotw,        INTR_TYPE_2OP, ISD::ROTL, 0),
+  X86_INTRINSIC_DATA(xop_vprotwi,       INTR_TYPE_2OP, X86ISD::VROTLI, 0),
   X86_INTRINSIC_DATA(xop_vpshab,        INTR_TYPE_2OP, X86ISD::VPSHA, 0),
   X86_INTRINSIC_DATA(xop_vpshad,        INTR_TYPE_2OP, X86ISD::VPSHA, 0),
   X86_INTRINSIC_DATA(xop_vpshaq,        INTR_TYPE_2OP, X86ISD::VPSHA, 0),
   X86_INTRINSIC_DATA(xop_vpshaw,        INTR_TYPE_2OP, X86ISD::VPSHA, 0),
   X86_INTRINSIC_DATA(xop_vpshlb,        INTR_TYPE_2OP, X86ISD::VPSHL, 0),
   X86_INTRINSIC_DATA(xop_vpshld,        INTR_TYPE_2OP, X86ISD::VPSHL, 0),
   X86_INTRINSIC_DATA(xop_vpshlq,        INTR_TYPE_2OP, X86ISD::VPSHL, 0),
   X86_INTRINSIC_DATA(xop_vpshlw,        INTR_TYPE_2OP, X86ISD::VPSHL, 0)