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
352
353
354
355
356
357
358
359
360
361
362
363
364
|
//===------------ AMDILInstrInfo.td - AMDIL Target ------*-tablegen-*------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//==-----------------------------------------------------------------------===//
//
// This file describes the AMDIL instructions in TableGen format.
//
//===----------------------------------------------------------------------===//
// AMDIL Instruction Predicate Definitions
// Predicate that is set to true if the hardware supports double precision
// divide
def HasHWDDiv : Predicate<"Subtarget.device()"
"->getGeneration() > AMDILDeviceInfo::HD4XXX && "
"Subtarget.device()->usesHardware(AMDILDeviceInfo::DoubleOps)">;
// Predicate that is set to true if the hardware supports double, but not double
// precision divide in hardware
def HasSWDDiv : Predicate<"Subtarget.device()"
"->getGeneration() == AMDILDeviceInfo::HD4XXX &&"
"Subtarget.device()->usesHardware(AMDILDeviceInfo::DoubleOps)">;
// Predicate that is set to true if the hardware support 24bit signed
// math ops. Otherwise a software expansion to 32bit math ops is used instead.
def HasHWSign24Bit : Predicate<"Subtarget.device()"
"->getGeneration() > AMDILDeviceInfo::HD5XXX">;
// Predicate that is set to true if 64bit operations are supported or not
def HasHW64Bit : Predicate<"Subtarget.device()"
"->usesHardware(AMDILDeviceInfo::LongOps)">;
def HasSW64Bit : Predicate<"Subtarget.device()"
"->usesSoftware(AMDILDeviceInfo::LongOps)">;
// Predicate that is set to true if the timer register is supported
def HasTmrRegister : Predicate<"Subtarget.device()"
"->isSupported(AMDILDeviceInfo::TmrReg)">;
// Predicate that is true if we are at least evergreen series
def HasDeviceIDInst : Predicate<"Subtarget.device()"
"->getGeneration() >= AMDILDeviceInfo::HD5XXX">;
// Predicate that is true if we have region address space.
def hasRegionAS : Predicate<"Subtarget.device()"
"->usesHardware(AMDILDeviceInfo::RegionMem)">;
// Predicate that is false if we don't have region address space.
def noRegionAS : Predicate<"!Subtarget.device()"
"->isSupported(AMDILDeviceInfo::RegionMem)">;
// Predicate that is set to true if 64bit Mul is supported in the IL or not
def HasHW64Mul : Predicate<"Subtarget.calVersion()"
">= CAL_VERSION_SC_139"
"&& Subtarget.device()"
"->getGeneration() >="
"AMDILDeviceInfo::HD5XXX">;
def HasSW64Mul : Predicate<"Subtarget.calVersion()"
"< CAL_VERSION_SC_139">;
// Predicate that is set to true if 64bit Div/Mod is supported in the IL or not
def HasHW64DivMod : Predicate<"Subtarget.device()"
"->usesHardware(AMDILDeviceInfo::HW64BitDivMod)">;
def HasSW64DivMod : Predicate<"Subtarget.device()"
"->usesSoftware(AMDILDeviceInfo::HW64BitDivMod)">;
// Predicate that is set to true if 64bit pointer are used.
def Has64BitPtr : Predicate<"Subtarget.is64bit()">;
def Has32BitPtr : Predicate<"!Subtarget.is64bit()">;
//===--------------------------------------------------------------------===//
// Custom Operands
//===--------------------------------------------------------------------===//
def brtarget : Operand<OtherVT>;
//===--------------------------------------------------------------------===//
// Custom Selection DAG Type Profiles
//===--------------------------------------------------------------------===//
//===----------------------------------------------------------------------===//
// Generic Profile Types
//===----------------------------------------------------------------------===//
def SDTIL_GenBinaryOp : SDTypeProfile<1, 2, [
SDTCisSameAs<0, 1>, SDTCisSameAs<1, 2>
]>;
def SDTIL_GenTernaryOp : SDTypeProfile<1, 3, [
SDTCisSameAs<0, 1>, SDTCisSameAs<1, 2>, SDTCisSameAs<2, 3>
]>;
def SDTIL_GenVecBuild : SDTypeProfile<1, 1, [
SDTCisEltOfVec<1, 0>
]>;
//===----------------------------------------------------------------------===//
// Flow Control Profile Types
//===----------------------------------------------------------------------===//
// Branch instruction where second and third are basic blocks
def SDTIL_BRCond : SDTypeProfile<0, 2, [
SDTCisVT<0, OtherVT>
]>;
// Comparison instruction
def SDTIL_Cmp : SDTypeProfile<1, 3, [
SDTCisSameAs<0, 2>, SDTCisSameAs<2,3>, SDTCisVT<1, i32>
]>;
//===--------------------------------------------------------------------===//
// Custom Selection DAG Nodes
//===--------------------------------------------------------------------===//
//===----------------------------------------------------------------------===//
// Flow Control DAG Nodes
//===----------------------------------------------------------------------===//
def IL_brcond : SDNode<"AMDILISD::BRANCH_COND", SDTIL_BRCond, [SDNPHasChain]>;
//===----------------------------------------------------------------------===//
// Comparison DAG Nodes
//===----------------------------------------------------------------------===//
def IL_cmp : SDNode<"AMDILISD::CMP", SDTIL_Cmp>;
//===----------------------------------------------------------------------===//
// Call/Return DAG Nodes
//===----------------------------------------------------------------------===//
def IL_retflag : SDNode<"AMDILISD::RET_FLAG", SDTNone,
[SDNPHasChain, SDNPOptInGlue]>;
//===--------------------------------------------------------------------===//
// Instructions
//===--------------------------------------------------------------------===//
// Floating point math functions
def IL_cmov_logical : SDNode<"AMDILISD::CMOVLOG", SDTIL_GenTernaryOp>;
def IL_div_inf : SDNode<"AMDILISD::DIV_INF", SDTIL_GenBinaryOp>;
def IL_mad : SDNode<"AMDILISD::MAD", SDTIL_GenTernaryOp>;
//===----------------------------------------------------------------------===//
// Integer functions
//===----------------------------------------------------------------------===//
def IL_umul : SDNode<"AMDILISD::UMUL" , SDTIntBinOp,
[SDNPCommutative, SDNPAssociative]>;
//===----------------------------------------------------------------------===//
// Vector functions
//===----------------------------------------------------------------------===//
def IL_vbuild : SDNode<"AMDILISD::VBUILD", SDTIL_GenVecBuild,
[]>;
//===--------------------------------------------------------------------===//
// Custom Pattern DAG Nodes
//===--------------------------------------------------------------------===//
def global_store : PatFrag<(ops node:$val, node:$ptr),
(store node:$val, node:$ptr), [{
return isGlobalStore(dyn_cast<StoreSDNode>(N));
}]>;
//===----------------------------------------------------------------------===//
// Load pattern fragments
//===----------------------------------------------------------------------===//
// Global address space loads
def global_load : PatFrag<(ops node:$ptr), (load node:$ptr), [{
return isGlobalLoad(dyn_cast<LoadSDNode>(N));
}]>;
// Constant address space loads
def constant_load : PatFrag<(ops node:$ptr), (load node:$ptr), [{
return isConstantLoad(dyn_cast<LoadSDNode>(N), -1);
}]>;
//===----------------------------------------------------------------------===//
// Complex addressing mode patterns
//===----------------------------------------------------------------------===//
def ADDR : ComplexPattern<i32, 2, "SelectADDR", [], []>;
def ADDRF : ComplexPattern<i32, 2, "SelectADDR", [frameindex], []>;
def ADDR64 : ComplexPattern<i64, 2, "SelectADDR64", [], []>;
def ADDR64F : ComplexPattern<i64, 2, "SelectADDR64", [frameindex], []>;
//===----------------------------------------------------------------------===//
// Conditional Instruction Pattern Leafs
//===----------------------------------------------------------------------===//
class IL_CC_Op<int N> : PatLeaf<(i32 N)>;
def IL_CC_D_EQ : IL_CC_Op<0>;
def IL_CC_D_GE : IL_CC_Op<1>;
def IL_CC_D_LT : IL_CC_Op<2>;
def IL_CC_D_NE : IL_CC_Op<3>;
def IL_CC_F_EQ : IL_CC_Op<4>;
def IL_CC_F_GE : IL_CC_Op<5>;
def IL_CC_F_LT : IL_CC_Op<6>;
def IL_CC_F_NE : IL_CC_Op<7>;
def IL_CC_I_EQ : IL_CC_Op<8>;
def IL_CC_I_GE : IL_CC_Op<9>;
def IL_CC_I_LT : IL_CC_Op<10>;
def IL_CC_I_NE : IL_CC_Op<11>;
def IL_CC_U_GE : IL_CC_Op<12>;
def IL_CC_U_LT : IL_CC_Op<13>;
// Pseudo IL comparison instructions that aren't natively supported
def IL_CC_F_GT : IL_CC_Op<14>;
def IL_CC_U_GT : IL_CC_Op<15>;
def IL_CC_I_GT : IL_CC_Op<16>;
def IL_CC_D_GT : IL_CC_Op<17>;
def IL_CC_F_LE : IL_CC_Op<18>;
def IL_CC_U_LE : IL_CC_Op<19>;
def IL_CC_I_LE : IL_CC_Op<20>;
def IL_CC_D_LE : IL_CC_Op<21>;
def IL_CC_F_UNE : IL_CC_Op<22>;
def IL_CC_F_UEQ : IL_CC_Op<23>;
def IL_CC_F_ULT : IL_CC_Op<24>;
def IL_CC_F_UGT : IL_CC_Op<25>;
def IL_CC_F_ULE : IL_CC_Op<26>;
def IL_CC_F_UGE : IL_CC_Op<27>;
def IL_CC_F_ONE : IL_CC_Op<28>;
def IL_CC_F_OEQ : IL_CC_Op<29>;
def IL_CC_F_OLT : IL_CC_Op<30>;
def IL_CC_F_OGT : IL_CC_Op<31>;
def IL_CC_F_OLE : IL_CC_Op<32>;
def IL_CC_F_OGE : IL_CC_Op<33>;
def IL_CC_D_UNE : IL_CC_Op<34>;
def IL_CC_D_UEQ : IL_CC_Op<35>;
def IL_CC_D_ULT : IL_CC_Op<36>;
def IL_CC_D_UGT : IL_CC_Op<37>;
def IL_CC_D_ULE : IL_CC_Op<38>;
def IL_CC_D_UGE : IL_CC_Op<39>;
def IL_CC_D_ONE : IL_CC_Op<30>;
def IL_CC_D_OEQ : IL_CC_Op<41>;
def IL_CC_D_OLT : IL_CC_Op<42>;
def IL_CC_D_OGT : IL_CC_Op<43>;
def IL_CC_D_OLE : IL_CC_Op<44>;
def IL_CC_D_OGE : IL_CC_Op<45>;
def IL_CC_U_EQ : IL_CC_Op<46>;
def IL_CC_U_NE : IL_CC_Op<47>;
def IL_CC_F_O : IL_CC_Op<48>;
def IL_CC_D_O : IL_CC_Op<49>;
def IL_CC_F_UO : IL_CC_Op<50>;
def IL_CC_D_UO : IL_CC_Op<51>;
def IL_CC_L_LE : IL_CC_Op<52>;
def IL_CC_L_GE : IL_CC_Op<53>;
def IL_CC_L_EQ : IL_CC_Op<54>;
def IL_CC_L_NE : IL_CC_Op<55>;
def IL_CC_L_LT : IL_CC_Op<56>;
def IL_CC_L_GT : IL_CC_Op<57>;
def IL_CC_UL_LE : IL_CC_Op<58>;
def IL_CC_UL_GE : IL_CC_Op<59>;
def IL_CC_UL_EQ : IL_CC_Op<60>;
def IL_CC_UL_NE : IL_CC_Op<61>;
def IL_CC_UL_LT : IL_CC_Op<62>;
def IL_CC_UL_GT : IL_CC_Op<63>;
//===----------------------------------------------------------------------===//
// Instruction format classes
//===----------------------------------------------------------------------===//
class ILFormat<dag outs, dag ins, string asmstr, list<dag> pattern>
: Instruction {
let Namespace = "AMDGPU";
dag OutOperandList = outs;
dag InOperandList = ins;
let Pattern = pattern;
let AsmString = !strconcat(asmstr, "\n");
let isPseudo = 1;
let Itinerary = NullALU;
bit hasIEEEFlag = 0;
bit hasZeroOpFlag = 0;
}
//===--------------------------------------------------------------------===//
// Multiclass Instruction formats
//===--------------------------------------------------------------------===//
// Multiclass that handles branch instructions
multiclass BranchConditional<SDNode Op> {
def _i32 : ILFormat<(outs),
(ins brtarget:$target, GPRI32:$src0),
"; i32 Pseudo branch instruction",
[(Op bb:$target, GPRI32:$src0)]>;
def _f32 : ILFormat<(outs),
(ins brtarget:$target, GPRF32:$src0),
"; f32 Pseudo branch instruction",
[(Op bb:$target, GPRF32:$src0)]>;
}
// Only scalar types should generate flow control
multiclass BranchInstr<string name> {
def _i32 : ILFormat<(outs), (ins GPRI32:$src),
!strconcat(name, " $src"), []>;
def _f32 : ILFormat<(outs), (ins GPRF32:$src),
!strconcat(name, " $src"), []>;
}
// Only scalar types should generate flow control
multiclass BranchInstr2<string name> {
def _i32 : ILFormat<(outs), (ins GPRI32:$src0, GPRI32:$src1),
!strconcat(name, " $src0, $src1"), []>;
def _f32 : ILFormat<(outs), (ins GPRF32:$src0, GPRF32:$src1),
!strconcat(name, " $src0, $src1"), []>;
}
//===--------------------------------------------------------------------===//
// Intrinsics support
//===--------------------------------------------------------------------===//
include "AMDILIntrinsics.td"
//===--------------------------------------------------------------------===//
// Instructions support
//===--------------------------------------------------------------------===//
//===---------------------------------------------------------------------===//
// Custom Inserter for Branches and returns, this eventually will be a
// seperate pass
//===---------------------------------------------------------------------===//
let isTerminator = 1 in {
def BRANCH : ILFormat<(outs), (ins brtarget:$target),
"; Pseudo unconditional branch instruction",
[(br bb:$target)]>;
defm BRANCH_COND : BranchConditional<IL_brcond>;
}
//===---------------------------------------------------------------------===//
// return instructions
//===---------------------------------------------------------------------===//
let isTerminator = 1, isReturn = 1, isBarrier = 1, hasCtrlDep = 1 in {
def RETURN : ILFormat<(outs), (ins variable_ops),
"RETURN", [(IL_retflag)]>;
}
//===---------------------------------------------------------------------===//
// Flow and Program control Instructions
//===---------------------------------------------------------------------===//
let isTerminator=1 in {
def SWITCH : ILFormat< (outs), (ins GPRI32:$src),
!strconcat("SWITCH", " $src"), []>;
def CASE : ILFormat< (outs), (ins GPRI32:$src),
!strconcat("CASE", " $src"), []>;
def BREAK : ILFormat< (outs), (ins),
"BREAK", []>;
def CONTINUE : ILFormat< (outs), (ins),
"CONTINUE", []>;
def DEFAULT : ILFormat< (outs), (ins),
"DEFAULT", []>;
def ELSE : ILFormat< (outs), (ins),
"ELSE", []>;
def ENDSWITCH : ILFormat< (outs), (ins),
"ENDSWITCH", []>;
def ENDMAIN : ILFormat< (outs), (ins),
"ENDMAIN", []>;
def END : ILFormat< (outs), (ins),
"END", []>;
def ENDFUNC : ILFormat< (outs), (ins),
"ENDFUNC", []>;
def ENDIF : ILFormat< (outs), (ins),
"ENDIF", []>;
def WHILELOOP : ILFormat< (outs), (ins),
"WHILE", []>;
def ENDLOOP : ILFormat< (outs), (ins),
"ENDLOOP", []>;
def FUNC : ILFormat< (outs), (ins),
"FUNC", []>;
def RETDYN : ILFormat< (outs), (ins),
"RET_DYN", []>;
// This opcode has custom swizzle pattern encoded in Swizzle Encoder
defm IF_LOGICALNZ : BranchInstr<"IF_LOGICALNZ">;
// This opcode has custom swizzle pattern encoded in Swizzle Encoder
defm IF_LOGICALZ : BranchInstr<"IF_LOGICALZ">;
// This opcode has custom swizzle pattern encoded in Swizzle Encoder
defm BREAK_LOGICALNZ : BranchInstr<"BREAK_LOGICALNZ">;
// This opcode has custom swizzle pattern encoded in Swizzle Encoder
defm BREAK_LOGICALZ : BranchInstr<"BREAK_LOGICALZ">;
// This opcode has custom swizzle pattern encoded in Swizzle Encoder
defm CONTINUE_LOGICALNZ : BranchInstr<"CONTINUE_LOGICALNZ">;
// This opcode has custom swizzle pattern encoded in Swizzle Encoder
defm CONTINUE_LOGICALZ : BranchInstr<"CONTINUE_LOGICALZ">;
defm IFC : BranchInstr2<"IFC">;
defm BREAKC : BranchInstr2<"BREAKC">;
defm CONTINUEC : BranchInstr2<"CONTINUEC">;
}
|