summaryrefslogtreecommitdiff
path: root/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
blob: 14477547affd383b52c6076161ab4b132b784cc0 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
//===-- AMDGPUISelLowering.cpp - AMDGPU Common DAG lowering functions -----===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This is the parent TargetLowering class for hardware code gen targets.
//
//===----------------------------------------------------------------------===//

#include "AMDGPUISelLowering.h"
#include "AMDILIntrinsicInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"

using namespace llvm;

AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) :
  TargetLowering(TM, new TargetLoweringObjectFileELF())
{

  // Initialize target lowering borrowed from AMDIL
  InitAMDILLowering();

  // We need to custom lower some of the intrinsics
  setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom);

  // Library functions.  These default to Expand, but we have instructions
  // for them.
  setOperationAction(ISD::FCEIL,  MVT::f32, Legal);
  setOperationAction(ISD::FEXP2,  MVT::f32, Legal);
  setOperationAction(ISD::FPOW,   MVT::f32, Legal);
  setOperationAction(ISD::FLOG2,  MVT::f32, Legal);
  setOperationAction(ISD::FABS,   MVT::f32, Legal);
  setOperationAction(ISD::FRINT,  MVT::f32, Legal);

  setOperationAction(ISD::UDIV, MVT::i32, Expand);
  setOperationAction(ISD::UDIVREM, MVT::i32, Custom);
  setOperationAction(ISD::UREM, MVT::i32, Expand);
}

//===---------------------------------------------------------------------===//
// TargetLowering Callbacks
//===---------------------------------------------------------------------===//

SDValue AMDGPUTargetLowering::LowerFormalArguments(
                                      SDValue Chain,
                                      CallingConv::ID CallConv,
                                      bool isVarArg,
                                      const SmallVectorImpl<ISD::InputArg> &Ins,
                                      DebugLoc DL, SelectionDAG &DAG,
                                      SmallVectorImpl<SDValue> &InVals) const
{
  for (unsigned i = 0, e = Ins.size(); i < e; ++i) {
    InVals.push_back(SDValue());
  }
  return Chain;
}

SDValue AMDGPUTargetLowering::LowerReturn(
                                     SDValue Chain,
                                     CallingConv::ID CallConv,
                                     bool isVarArg,
                                     const SmallVectorImpl<ISD::OutputArg> &Outs,
                                     const SmallVectorImpl<SDValue> &OutVals,
                                     DebugLoc DL, SelectionDAG &DAG) const
{
  return DAG.getNode(AMDGPUISD::RET_FLAG, DL, MVT::Other, Chain);
}

//===---------------------------------------------------------------------===//
// Target specific lowering
//===---------------------------------------------------------------------===//

SDValue AMDGPUTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG)
    const
{
  switch (Op.getOpcode()) {
  default:
    Op.getNode()->dump();
    assert(0 && "Custom lowering code for this"
        "instruction is not implemented yet!");
    break;
  // AMDIL DAG lowering
  case ISD::SDIV: return LowerSDIV(Op, DAG);
  case ISD::SREM: return LowerSREM(Op, DAG);
  case ISD::SIGN_EXTEND_INREG: return LowerSIGN_EXTEND_INREG(Op, DAG);
  case ISD::BRCOND: return LowerBRCOND(Op, DAG);
  // AMDGPU DAG lowering
  case ISD::INTRINSIC_WO_CHAIN: return LowerINTRINSIC_WO_CHAIN(Op, DAG);
  case ISD::UDIVREM: return LowerUDIVREM(Op, DAG);
  }
  return Op;
}

SDValue AMDGPUTargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
    SelectionDAG &DAG) const
{
  unsigned IntrinsicID = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
  DebugLoc DL = Op.getDebugLoc();
  EVT VT = Op.getValueType();

  switch (IntrinsicID) {
    default: return Op;
    case AMDGPUIntrinsic::AMDIL_abs:
      return LowerIntrinsicIABS(Op, DAG);
    case AMDGPUIntrinsic::AMDIL_exp:
      return DAG.getNode(ISD::FEXP2, DL, VT, Op.getOperand(1));
    case AMDGPUIntrinsic::AMDGPU_lrp:
      return LowerIntrinsicLRP(Op, DAG);
    case AMDGPUIntrinsic::AMDIL_fraction:
      return DAG.getNode(AMDGPUISD::FRACT, DL, VT, Op.getOperand(1));
    case AMDGPUIntrinsic::AMDIL_mad:
      return DAG.getNode(AMDGPUISD::MAD, DL, VT, Op.getOperand(1),
                              Op.getOperand(2), Op.getOperand(3));
    case AMDGPUIntrinsic::AMDIL_max:
      return DAG.getNode(AMDGPUISD::FMAX, DL, VT, Op.getOperand(1),
                                                  Op.getOperand(2));
    case AMDGPUIntrinsic::AMDGPU_imax:
      return DAG.getNode(AMDGPUISD::SMAX, DL, VT, Op.getOperand(1),
                                                  Op.getOperand(2));
    case AMDGPUIntrinsic::AMDGPU_umax:
      return DAG.getNode(AMDGPUISD::UMAX, DL, VT, Op.getOperand(1),
                                                  Op.getOperand(2));
    case AMDGPUIntrinsic::AMDIL_min:
      return DAG.getNode(AMDGPUISD::FMIN, DL, VT, Op.getOperand(1),
                                                  Op.getOperand(2));
    case AMDGPUIntrinsic::AMDGPU_imin:
      return DAG.getNode(AMDGPUISD::SMIN, DL, VT, Op.getOperand(1),
                                                  Op.getOperand(2));
    case AMDGPUIntrinsic::AMDGPU_umin:
      return DAG.getNode(AMDGPUISD::UMIN, DL, VT, Op.getOperand(1),
                                                  Op.getOperand(2));
    case AMDGPUIntrinsic::AMDIL_round_nearest:
      return DAG.getNode(ISD::FRINT, DL, VT, Op.getOperand(1));
    case AMDGPUIntrinsic::AMDIL_round_posinf:
      return DAG.getNode(ISD::FCEIL, DL, VT, Op.getOperand(1));
  }
}

///IABS(a) = SMAX(sub(0, a), a)
SDValue AMDGPUTargetLowering::LowerIntrinsicIABS(SDValue Op,
    SelectionDAG &DAG) const
{

  DebugLoc DL = Op.getDebugLoc();
  EVT VT = Op.getValueType();
  SDValue Neg = DAG.getNode(ISD::SUB, DL, VT, DAG.getConstant(0, VT),
                                              Op.getOperand(1));

  return DAG.getNode(AMDGPUISD::SMAX, DL, VT, Neg, Op.getOperand(1));
}

/// Linear Interpolation
/// LRP(a, b, c) = muladd(a,  b, (1 - a) * c)
SDValue AMDGPUTargetLowering::LowerIntrinsicLRP(SDValue Op,
    SelectionDAG &DAG) const
{
  DebugLoc DL = Op.getDebugLoc();
  EVT VT = Op.getValueType();
  SDValue OneSubA = DAG.getNode(ISD::FSUB, DL, VT,
                                DAG.getConstantFP(1.0f, MVT::f32),
                                Op.getOperand(1));
  SDValue OneSubAC = DAG.getNode(ISD::FMUL, DL, VT, OneSubA,
                                                    Op.getOperand(3));
  return DAG.getNode(AMDGPUISD::MAD, DL, VT, Op.getOperand(1),
                                               Op.getOperand(2),
                                               OneSubAC);
}



SDValue AMDGPUTargetLowering::LowerUDIVREM(SDValue Op,
    SelectionDAG &DAG) const
{
  DebugLoc DL = Op.getDebugLoc();
  EVT VT = Op.getValueType();

  SDValue Num = Op.getOperand(0);
  SDValue Den = Op.getOperand(1);

  SmallVector<SDValue, 8> Results;

  // RCP =  URECIP(Den) = 2^32 / Den + e
  // e is rounding error.
  SDValue RCP = DAG.getNode(AMDGPUISD::URECIP, DL, VT, Den);

  // RCP_LO = umulo(RCP, Den) */
  SDValue RCP_LO = DAG.getNode(ISD::UMULO, DL, VT, RCP, Den);

  // RCP_HI = mulhu (RCP, Den) */
  SDValue RCP_HI = DAG.getNode(ISD::MULHU, DL, VT, RCP, Den);

  // NEG_RCP_LO = -RCP_LO
  SDValue NEG_RCP_LO = DAG.getNode(ISD::SUB, DL, VT, DAG.getConstant(0, VT),
                                                     RCP_LO);

  // ABS_RCP_LO = (RCP_HI == 0 ? NEG_RCP_LO : RCP_LO)
  SDValue ABS_RCP_LO = DAG.getSelectCC(DL, RCP_HI, DAG.getConstant(0, VT),
                                           NEG_RCP_LO, RCP_LO,
                                           ISD::SETEQ);
  // Calculate the rounding error from the URECIP instruction
  // E = mulhu(ABS_RCP_LO, RCP)
  SDValue E = DAG.getNode(ISD::MULHU, DL, VT, ABS_RCP_LO, RCP);

  // RCP_A_E = RCP + E
  SDValue RCP_A_E = DAG.getNode(ISD::ADD, DL, VT, RCP, E);

  // RCP_S_E = RCP - E
  SDValue RCP_S_E = DAG.getNode(ISD::SUB, DL, VT, RCP, E);

  // Tmp0 = (RCP_HI == 0 ? RCP_A_E : RCP_SUB_E)
  SDValue Tmp0 = DAG.getSelectCC(DL, RCP_HI, DAG.getConstant(0, VT),
                                     RCP_A_E, RCP_S_E,
                                     ISD::SETEQ);
  // Quotient = mulhu(Tmp0, Num)
  SDValue Quotient = DAG.getNode(ISD::MULHU, DL, VT, Tmp0, Num);

  // Num_S_Remainder = Quotient * Den
  SDValue Num_S_Remainder = DAG.getNode(ISD::UMULO, DL, VT, Quotient, Den);

  // Remainder = Num - Num_S_Remainder
  SDValue Remainder = DAG.getNode(ISD::SUB, DL, VT, Num, Num_S_Remainder);

  // Remainder_GE_Den = (Remainder >= Den ? -1 : 0)
  SDValue Remainder_GE_Den = DAG.getSelectCC(DL, Remainder, Den,
                                                 DAG.getConstant(-1, VT),
                                                 DAG.getConstant(0, VT),
                                                 ISD::SETGE);
  // Remainder_GE_Zero = (Remainder >= 0 ? -1 : 0)
  SDValue Remainder_GE_Zero = DAG.getSelectCC(DL, Remainder,
                                                  DAG.getConstant(0, VT),
                                                  DAG.getConstant(-1, VT),
                                                  DAG.getConstant(0, VT),
                                                  ISD::SETGE);
  // Tmp1 = Remainder_GE_Den & Remainder_GE_Zero
  SDValue Tmp1 = DAG.getNode(ISD::AND, DL, VT, Remainder_GE_Den,
                                               Remainder_GE_Zero);

  // Calculate Division result:

  // Quotient_A_One = Quotient + 1
  SDValue Quotient_A_One = DAG.getNode(ISD::ADD, DL, VT, Quotient,
                                                         DAG.getConstant(1, VT));

  // Quotient_S_One = Quotient - 1
  SDValue Quotient_S_One = DAG.getNode(ISD::SUB, DL, VT, Quotient,
                                                         DAG.getConstant(1, VT));

  // Div = (Tmp1 == 0 ? Quotient : Quotient_A_One)
  SDValue Div = DAG.getSelectCC(DL, Tmp1, DAG.getConstant(0, VT),
                                     Quotient, Quotient_A_One, ISD::SETEQ);

  // Div = (Remainder_GE_Zero == 0 ? Quotient_S_One : Div)
  Div = DAG.getSelectCC(DL, Remainder_GE_Zero, DAG.getConstant(0, VT),
                            Quotient_S_One, Div, ISD::SETEQ);

  // Calculate Rem result:

  // Remainder_S_Den = Remainder - Den
  SDValue Remainder_S_Den = DAG.getNode(ISD::SUB, DL, VT, Remainder, Den);

  // Remainder_A_Den = Remainder + Den
  SDValue Remainder_A_Den = DAG.getNode(ISD::ADD, DL, VT, Remainder, Den);

  // Rem = (Tmp1 == 0 ? Remainder : Remainder_S_Den)
  SDValue Rem = DAG.getSelectCC(DL, Tmp1, DAG.getConstant(0, VT),
                                    Remainder, Remainder_S_Den, ISD::SETEQ);

  // Rem = (Remainder_GE_Zero == 0 ? Remainder_A_Den : Rem)
  Rem = DAG.getSelectCC(DL, Remainder_GE_Zero, DAG.getConstant(0, VT),
                            Remainder_A_Den, Rem, ISD::SETEQ);
  SDValue Ops[2];
  Ops[0] = Div;
  Ops[1] = Rem;
  return DAG.getMergeValues(Ops, 2, DL);
}

//===----------------------------------------------------------------------===//
// Helper functions
//===----------------------------------------------------------------------===//

bool AMDGPUTargetLowering::isHWTrueValue(SDValue Op) const
{
  if (ConstantFPSDNode * CFP = dyn_cast<ConstantFPSDNode>(Op)) {
    return CFP->isExactlyValue(1.0);
  }
  if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op)) {
    return C->isAllOnesValue();
  }
  return false;
}

bool AMDGPUTargetLowering::isHWFalseValue(SDValue Op) const
{
  if (ConstantFPSDNode * CFP = dyn_cast<ConstantFPSDNode>(Op)) {
    return CFP->getValueAPF().isZero();
  }
  if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op)) {
    return C->isNullValue();
  }
  return false;
}

SDValue AMDGPUTargetLowering::CreateLiveInRegister(SelectionDAG &DAG,
                                                  const TargetRegisterClass *RC,
                                                   unsigned Reg, EVT VT) const {
  MachineFunction &MF = DAG.getMachineFunction();
  MachineRegisterInfo &MRI = MF.getRegInfo();
  unsigned VirtualRegister;
  if (!MRI.isLiveIn(Reg)) {
    VirtualRegister = MRI.createVirtualRegister(RC);
    MRI.addLiveIn(Reg, VirtualRegister);
  } else {
    VirtualRegister = MRI.getLiveInVirtReg(Reg);
  }
  return DAG.getRegister(VirtualRegister, VT);
}

#define NODE_NAME_CASE(node) case AMDGPUISD::node: return #node;

const char* AMDGPUTargetLowering::getTargetNodeName(unsigned Opcode) const
{
  switch (Opcode) {
  default: return 0;
  // AMDIL DAG nodes
  NODE_NAME_CASE(MAD);
  NODE_NAME_CASE(CALL);
  NODE_NAME_CASE(UMUL);
  NODE_NAME_CASE(DIV_INF);
  NODE_NAME_CASE(RET_FLAG);
  NODE_NAME_CASE(BRANCH_COND);

  // AMDGPU DAG nodes
  NODE_NAME_CASE(FRACT)
  NODE_NAME_CASE(FMAX)
  NODE_NAME_CASE(SMAX)
  NODE_NAME_CASE(UMAX)
  NODE_NAME_CASE(FMIN)
  NODE_NAME_CASE(SMIN)
  NODE_NAME_CASE(UMIN)
  NODE_NAME_CASE(URECIP)
  NODE_NAME_CASE(INTERP)
  NODE_NAME_CASE(INTERP_P0)
  }
}