Implement proper handling for pcmpistri/pcmpestri intrinsics. Requires custom handling in DAGISelToDAG due to limitations in TableGen's implicit def handling. Fixes PR11305.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@161318 91177308-0d34-0410-b5e6-96231b3b80d8
2024-12-13 20:32:21 +00:00 · 2012-08-06 06:22:36 +00:00 · 2012-08-06 06:22:36 +00:00 · 4feb647283
commit 4feb647283
parent d230d20b73
6 changed files with 197 additions and 65 deletions
--- a/lib/Target/X86/X86ISelDAGToDAG.cpp
+++ b/lib/Target/X86/X86ISelDAGToDAG.cpp
@ -2600,6 +2600,85 @@ SDNode *X86DAGToDAGISel::Select(SDNode *Node) {

    return Result;
  }
+
+  // FIXME: Custom handling because TableGen doesn't support multiple implicit
+  // defs in an instruction pattern
+  case X86ISD::PCMPESTRI: {
+    SDValue N0 = Node->getOperand(0);
+    SDValue N1 = Node->getOperand(1);
+    SDValue N2 = Node->getOperand(2);
+    SDValue N3 = Node->getOperand(3);
+    SDValue N4 = Node->getOperand(4);
+
+    // Make sure last argument is a constant
+    ConstantSDNode *Cst = dyn_cast<ConstantSDNode>(N4);
+    if (!Cst)
+      break;
+
+    uint64_t Imm = Cst->getZExtValue();
+
+    SDValue InFlag = CurDAG->getCopyToReg(CurDAG->getEntryNode(), dl,
+                                          X86::EAX, N1, SDValue()).getValue(1);
+    InFlag = CurDAG->getCopyToReg(CurDAG->getEntryNode(), dl, X86::EDX,
+                                  N3, InFlag).getValue(1);
+
+    SDValue Ops[] = { N0, N2, getI8Imm(Imm), InFlag };
+    unsigned Opc = Subtarget->hasAVX() ? X86::VPCMPESTRIrr :
+                                         X86::PCMPESTRIrr;
+    InFlag = SDValue(CurDAG->getMachineNode(Opc, dl, MVT::Glue, Ops,
+                                            array_lengthof(Ops)), 0);
+
+    if (!SDValue(Node, 0).use_empty()) {
+      SDValue Result = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), dl,
+                                              X86::ECX, NVT, InFlag);
+      InFlag = Result.getValue(2);
+      ReplaceUses(SDValue(Node, 0), Result);
+    }
+    if (!SDValue(Node, 1).use_empty()) {
+      SDValue Result = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), dl,
+                                              X86::EFLAGS, NVT, InFlag);
+      InFlag = Result.getValue(2);
+      ReplaceUses(SDValue(Node, 1), Result);
+    }
+
+    return NULL;
+  }
+
+  // FIXME: Custom handling because TableGen doesn't support multiple implicit
+  // defs in an instruction pattern
+  case X86ISD::PCMPISTRI: {
+    SDValue N0 = Node->getOperand(0);
+    SDValue N1 = Node->getOperand(1);
+    SDValue N2 = Node->getOperand(2);
+
+    // Make sure last argument is a constant
+    ConstantSDNode *Cst = dyn_cast<ConstantSDNode>(N2);
+    if (!Cst)
+      break;
+
+    uint64_t Imm = Cst->getZExtValue();
+
+    SDValue Ops[] = { N0, N1, getI8Imm(Imm) };
+    unsigned Opc = Subtarget->hasAVX() ? X86::VPCMPISTRIrr :
+                                         X86::PCMPISTRIrr;
+    SDValue InFlag = SDValue(CurDAG->getMachineNode(Opc, dl, MVT::Glue, Ops,
+                                                    array_lengthof(Ops)), 0);
+
+    if (!SDValue(Node, 0).use_empty()) {
+      SDValue Result = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), dl,
+                                              X86::ECX, NVT, InFlag);
+      InFlag = Result.getValue(2);
+      ReplaceUses(SDValue(Node, 0), Result);
+    }
+    if (!SDValue(Node, 1).use_empty()) {
+      SDValue Result = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), dl,
+                                              X86::EFLAGS, NVT, InFlag);
+      InFlag = Result.getValue(2);
+      ReplaceUses(SDValue(Node, 1), Result);
+    }
+
+    return NULL;
+  }
  }

  SDNode *ResNode = SelectCode(Node);
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@ -9845,6 +9845,83 @@ X86TargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) const
                       DAG.getConstant(NewIntNo, MVT::i32),
                       Op.getOperand(1), ShAmt);
  }
+  case Intrinsic::x86_sse42_pcmpistria128:
+  case Intrinsic::x86_sse42_pcmpestria128:
+  case Intrinsic::x86_sse42_pcmpistric128:
+  case Intrinsic::x86_sse42_pcmpestric128:
+  case Intrinsic::x86_sse42_pcmpistrio128:
+  case Intrinsic::x86_sse42_pcmpestrio128:
+  case Intrinsic::x86_sse42_pcmpistris128:
+  case Intrinsic::x86_sse42_pcmpestris128:
+  case Intrinsic::x86_sse42_pcmpistriz128:
+  case Intrinsic::x86_sse42_pcmpestriz128: {
+    unsigned Opcode;
+    unsigned X86CC;
+    switch (IntNo) {
+    default: llvm_unreachable("Impossible intrinsic");  // Can't reach here.
+    case Intrinsic::x86_sse42_pcmpistria128:
+      Opcode = X86ISD::PCMPISTRI;
+      X86CC = X86::COND_A;
+      break;
+    case Intrinsic::x86_sse42_pcmpestria128:
+      Opcode = X86ISD::PCMPESTRI;
+      X86CC = X86::COND_A;
+      break;
+    case Intrinsic::x86_sse42_pcmpistric128:
+      Opcode = X86ISD::PCMPISTRI;
+      X86CC = X86::COND_B;
+      break;
+    case Intrinsic::x86_sse42_pcmpestric128:
+      Opcode = X86ISD::PCMPESTRI;
+      X86CC = X86::COND_B;
+      break;
+    case Intrinsic::x86_sse42_pcmpistrio128:
+      Opcode = X86ISD::PCMPISTRI;
+      X86CC = X86::COND_O;
+      break;
+    case Intrinsic::x86_sse42_pcmpestrio128:
+      Opcode = X86ISD::PCMPESTRI;
+      X86CC = X86::COND_O;
+      break;
+    case Intrinsic::x86_sse42_pcmpistris128:
+      Opcode = X86ISD::PCMPISTRI;
+      X86CC = X86::COND_S;
+      break;
+    case Intrinsic::x86_sse42_pcmpestris128:
+      Opcode = X86ISD::PCMPESTRI;
+      X86CC = X86::COND_S;
+      break;
+    case Intrinsic::x86_sse42_pcmpistriz128:
+      Opcode = X86ISD::PCMPISTRI;
+      X86CC = X86::COND_E;
+      break;
+    case Intrinsic::x86_sse42_pcmpestriz128:
+      Opcode = X86ISD::PCMPESTRI;
+      X86CC = X86::COND_E;
+      break;
+    }
+    SmallVector<SDValue, 5> NewOps;
+    NewOps.append(Op->op_begin()+1, Op->op_end());
+    SDVTList VTs = DAG.getVTList(Op.getValueType(), MVT::i32);
+    SDValue PCMP = DAG.getNode(Opcode, dl, VTs, NewOps.data(), NewOps.size());
+    SDValue SetCC = DAG.getNode(X86ISD::SETCC, dl, MVT::i8,
+                                DAG.getConstant(X86CC, MVT::i8),
+                                SDValue(PCMP.getNode(), 1));
+    return DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::i32, SetCC);
+  }
+  case Intrinsic::x86_sse42_pcmpistri128:
+  case Intrinsic::x86_sse42_pcmpestri128: {
+    unsigned Opcode;
+    if (IntNo == Intrinsic::x86_sse42_pcmpistri128)
+      Opcode = X86ISD::PCMPISTRI;
+    else
+      Opcode = X86ISD::PCMPESTRI;
+
+    SmallVector<SDValue, 5> NewOps;
+    NewOps.append(Op->op_begin()+1, Op->op_end());
+    SDVTList VTs = DAG.getVTList(Op.getValueType(), MVT::i32);
+    return DAG.getNode(Opcode, dl, VTs, NewOps.data(), NewOps.size());
+  }
  }
 }

--- a/lib/Target/X86/X86ISelLowering.h
+++ b/lib/Target/X86/X86ISelLowering.h
@ -333,6 +333,10 @@ namespace llvm {
      // RDRAND - Get a random integer and indicate whether it is valid in CF.
      RDRAND,

+      // PCMP*STRI
+      PCMPISTRI,
+      PCMPESTRI,
+
      // ATOMADD64_DAG, ATOMSUB64_DAG, ATOMOR64_DAG, ATOMAND64_DAG,
      // ATOMXOR64_DAG, ATOMNAND64_DAG, ATOMSWAP64_DAG -
      // Atomic 64-bit binary operations.
--- a/lib/Target/X86/X86InstrFragmentsSIMD.td
+++ b/lib/Target/X86/X86InstrFragmentsSIMD.td
@ -173,6 +173,17 @@ def X86Fnmsub    : SDNode<"X86ISD::FNMSUB",    SDTFma>;
 def X86Fmaddsub  : SDNode<"X86ISD::FMSUBADD",  SDTFma>;
 def X86Fmsubadd  : SDNode<"X86ISD::FMADDSUB",  SDTFma>;

+def SDT_PCMPISTRI : SDTypeProfile<2, 3, [SDTCisVT<0, i32>, SDTCisVT<1, i32>,
+                                         SDTCisVT<2, v16i8>, SDTCisVT<3, v16i8>,
+                                         SDTCisVT<4, i8>]>;
+def SDT_PCMPESTRI : SDTypeProfile<2, 5, [SDTCisVT<0, i32>, SDTCisVT<1, i32>,
+                                         SDTCisVT<2, v16i8>, SDTCisVT<3, i32>,
+                                         SDTCisVT<4, v16i8>, SDTCisVT<5, i32>,
+                                         SDTCisVT<6, i8>]>;
+
+def X86pcmpistri : SDNode<"X86ISD::PCMPISTRI", SDT_PCMPISTRI>;
+def X86pcmpestri : SDNode<"X86ISD::PCMPESTRI", SDT_PCMPESTRI>;
+
 //===----------------------------------------------------------------------===//
 // SSE Complex Patterns
 //===----------------------------------------------------------------------===//
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@ -6841,81 +6841,42 @@ let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
 }

 // Packed Compare Implicit Length Strings, Return Index
-let Defs = [ECX, EFLAGS] in {
-  multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
+let Defs = [ECX, EFLAGS], neverHasSideEffects = 1 in {
+  multiclass SS42AI_pcmpistri<string asm> {
    def rr : SS42AI<0x63, MRMSrcReg, (outs),
      (ins VR128:$src1, VR128:$src2, i8imm:$src3),
      !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
-      [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
-       (implicit EFLAGS)]>, OpSize;
+      []>, OpSize;
+    let mayLoad = 1 in
    def rm : SS42AI<0x63, MRMSrcMem, (outs),
      (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
      !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
-      [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
-       (implicit EFLAGS)]>, OpSize;
+      []>, OpSize;
  }
 }

-let Predicates = [HasAVX] in {
-defm VPCMPISTRI  : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
-                                    VEX;
-defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
-                                    VEX;
-defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
-                                    VEX;
-defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
-                                    VEX;
-defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
-                                    VEX;
-defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
-                                    VEX;
-}
-
-defm PCMPISTRI  : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
-defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
-defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
-defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
-defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
-defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
+let Predicates = [HasAVX] in
+defm VPCMPISTRI  : SS42AI_pcmpistri<"vpcmpistri">, VEX;
+defm PCMPISTRI   : SS42AI_pcmpistri<"pcmpistri">;

 // Packed Compare Explicit Length Strings, Return Index
-let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
-  multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
+let Defs = [ECX, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
+  multiclass SS42AI_pcmpestri<string asm> {
    def rr : SS42AI<0x61, MRMSrcReg, (outs),
      (ins VR128:$src1, VR128:$src3, i8imm:$src5),
      !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
-      [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
-       (implicit EFLAGS)]>, OpSize;
+      []>, OpSize;
+    let mayLoad = 1 in
    def rm : SS42AI<0x61, MRMSrcMem, (outs),
      (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
      !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
-       [(set ECX,
-             (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
-        (implicit EFLAGS)]>, OpSize;
+      []>, OpSize;
  }
 }

-let Predicates = [HasAVX] in {
-defm VPCMPESTRI  : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
-                                    VEX;
-defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
-                                    VEX;
-defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
-                                    VEX;
-defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
-                                    VEX;
-defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
-                                    VEX;
-defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
-                                    VEX;
-}
-
-defm PCMPESTRI  : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
-defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
-defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
-defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
-defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
-defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
+let Predicates = [HasAVX] in
+defm VPCMPESTRI  : SS42AI_pcmpestri<"vpcmpestri">, VEX;
+defm PCMPESTRI   : SS42AI_pcmpestri<"pcmpestri">;

 //===----------------------------------------------------------------------===//
 // SSE4.2 - CRC Instructions
--- a/test/CodeGen/X86/avx-intrinsics-x86.ll
+++ b/test/CodeGen/X86/avx-intrinsics-x86.ll
@ -1154,7 +1154,7 @@ define i32 @test_x86_sse42_pcmpestria128(<16 x i8> %a0, <16 x i8> %a2) {
  ; CHECK: movl
  ; CHECK: movl
  ; CHECK: vpcmpestri
-  ; CHECK: movl
+  ; CHECK: seta
  %res = call i32 @llvm.x86.sse42.pcmpestria128(<16 x i8> %a0, i32 7, <16 x i8> %a2, i32 7, i8 7) ; <i32> [#uses=1]
  ret i32 %res
 }
@ -1165,7 +1165,7 @@ define i32 @test_x86_sse42_pcmpestric128(<16 x i8> %a0, <16 x i8> %a2) {
  ; CHECK: movl
  ; CHECK: movl
  ; CHECK: vpcmpestri
-  ; CHECK: movl
+  ; CHECK: sbbl
  %res = call i32 @llvm.x86.sse42.pcmpestric128(<16 x i8> %a0, i32 7, <16 x i8> %a2, i32 7, i8 7) ; <i32> [#uses=1]
  ret i32 %res
 }
@ -1176,7 +1176,7 @@ define i32 @test_x86_sse42_pcmpestrio128(<16 x i8> %a0, <16 x i8> %a2) {
  ; CHECK: movl
  ; CHECK: movl
  ; CHECK: vpcmpestri
-  ; CHECK: movl
+  ; CHECK: seto
  %res = call i32 @llvm.x86.sse42.pcmpestrio128(<16 x i8> %a0, i32 7, <16 x i8> %a2, i32 7, i8 7) ; <i32> [#uses=1]
  ret i32 %res
 }
@ -1187,7 +1187,7 @@ define i32 @test_x86_sse42_pcmpestris128(<16 x i8> %a0, <16 x i8> %a2) {
  ; CHECK: movl
  ; CHECK: movl
  ; CHECK: vpcmpestri
-  ; CHECK: movl
+  ; CHECK: sets
  %res = call i32 @llvm.x86.sse42.pcmpestris128(<16 x i8> %a0, i32 7, <16 x i8> %a2, i32 7, i8 7) ; <i32> [#uses=1]
  ret i32 %res
 }
@ -1198,7 +1198,7 @@ define i32 @test_x86_sse42_pcmpestriz128(<16 x i8> %a0, <16 x i8> %a2) {
  ; CHECK: movl
  ; CHECK: movl
  ; CHECK: vpcmpestri
-  ; CHECK: movl
+  ; CHECK: sete
  %res = call i32 @llvm.x86.sse42.pcmpestriz128(<16 x i8> %a0, i32 7, <16 x i8> %a2, i32 7, i8 7) ; <i32> [#uses=1]
  ret i32 %res
 }
@ -1227,7 +1227,7 @@ declare i32 @llvm.x86.sse42.pcmpistri128(<16 x i8>, <16 x i8>, i8) nounwind read

 define i32 @test_x86_sse42_pcmpistria128(<16 x i8> %a0, <16 x i8> %a1) {
  ; CHECK: vpcmpistri
-  ; CHECK: movl
+  ; CHECK: seta
  %res = call i32 @llvm.x86.sse42.pcmpistria128(<16 x i8> %a0, <16 x i8> %a1, i8 7) ; <i32> [#uses=1]
  ret i32 %res
 }
@ -1236,7 +1236,7 @@ declare i32 @llvm.x86.sse42.pcmpistria128(<16 x i8>, <16 x i8>, i8) nounwind rea

 define i32 @test_x86_sse42_pcmpistric128(<16 x i8> %a0, <16 x i8> %a1) {
  ; CHECK: vpcmpistri
-  ; CHECK: movl
+  ; CHECK: sbbl
  %res = call i32 @llvm.x86.sse42.pcmpistric128(<16 x i8> %a0, <16 x i8> %a1, i8 7) ; <i32> [#uses=1]
  ret i32 %res
 }
@ -1245,7 +1245,7 @@ declare i32 @llvm.x86.sse42.pcmpistric128(<16 x i8>, <16 x i8>, i8) nounwind rea

 define i32 @test_x86_sse42_pcmpistrio128(<16 x i8> %a0, <16 x i8> %a1) {
  ; CHECK: vpcmpistri
-  ; CHECK: movl
+  ; CHECK: seto
  %res = call i32 @llvm.x86.sse42.pcmpistrio128(<16 x i8> %a0, <16 x i8> %a1, i8 7) ; <i32> [#uses=1]
  ret i32 %res
 }
@ -1254,7 +1254,7 @@ declare i32 @llvm.x86.sse42.pcmpistrio128(<16 x i8>, <16 x i8>, i8) nounwind rea

 define i32 @test_x86_sse42_pcmpistris128(<16 x i8> %a0, <16 x i8> %a1) {
  ; CHECK: vpcmpistri
-  ; CHECK: movl
+  ; CHECK: sets
  %res = call i32 @llvm.x86.sse42.pcmpistris128(<16 x i8> %a0, <16 x i8> %a1, i8 7) ; <i32> [#uses=1]
  ret i32 %res
 }
@ -1263,7 +1263,7 @@ declare i32 @llvm.x86.sse42.pcmpistris128(<16 x i8>, <16 x i8>, i8) nounwind rea

 define i32 @test_x86_sse42_pcmpistriz128(<16 x i8> %a0, <16 x i8> %a1) {
  ; CHECK: vpcmpistri
-  ; CHECK: movl
+  ; CHECK: sete
  %res = call i32 @llvm.x86.sse42.pcmpistriz128(<16 x i8> %a0, <16 x i8> %a1, i8 7) ; <i32> [#uses=1]
  ret i32 %res
 }