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
|
//===- HexagonIntrinsicsV4.td - V4 Instruction intrinsics --*- tablegen -*-===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
// This is populated based on the following specs:
// Hexagon V4 Architecture Extensions
// Application-Level Specification
// 80-V9418-12 Rev. A
// June 15, 2010
// Vector reduce multiply word by signed half (32x16)
//Rdd=vrmpyweh(Rss,Rtt)[:<<1]
def : T_PP_pat <M4_vrmpyeh_s0, int_hexagon_M4_vrmpyeh_s0>;
def : T_PP_pat <M4_vrmpyeh_s1, int_hexagon_M4_vrmpyeh_s1>;
//Rdd=vrmpywoh(Rss,Rtt)[:<<1]
def : T_PP_pat <M4_vrmpyoh_s0, int_hexagon_M4_vrmpyoh_s0>;
def : T_PP_pat <M4_vrmpyoh_s1, int_hexagon_M4_vrmpyoh_s1>;
//Rdd+=vrmpyweh(Rss,Rtt)[:<<1]
def : T_PPP_pat <M4_vrmpyeh_acc_s0, int_hexagon_M4_vrmpyeh_acc_s0>;
def : T_PPP_pat <M4_vrmpyeh_acc_s1, int_hexagon_M4_vrmpyeh_acc_s1>;
//Rdd=vrmpywoh(Rss,Rtt)[:<<1]
def : T_PPP_pat <M4_vrmpyoh_acc_s0, int_hexagon_M4_vrmpyoh_acc_s0>;
def : T_PPP_pat <M4_vrmpyoh_acc_s1, int_hexagon_M4_vrmpyoh_acc_s1>;
// Vector multiply halfwords, signed by unsigned
// Rdd=vmpyhsu(Rs,Rt)[:<<1]:sat
def : T_RR_pat <M2_vmpy2su_s0, int_hexagon_M2_vmpy2su_s0>;
def : T_RR_pat <M2_vmpy2su_s1, int_hexagon_M2_vmpy2su_s1>;
// Rxx+=vmpyhsu(Rs,Rt)[:<<1]:sat
def : T_PRR_pat <M2_vmac2su_s0, int_hexagon_M2_vmac2su_s0>;
def : T_PRR_pat <M2_vmac2su_s1, int_hexagon_M2_vmac2su_s1>;
// Vector polynomial multiply halfwords
// Rdd=vpmpyh(Rs,Rt)
def : T_RR_pat <M4_vpmpyh, int_hexagon_M4_vpmpyh>;
// Rxx[^]=vpmpyh(Rs,Rt)
def : T_PRR_pat <M4_vpmpyh_acc, int_hexagon_M4_vpmpyh_acc>;
// Polynomial multiply words
// Rdd=pmpyw(Rs,Rt)
def : T_RR_pat <M4_pmpyw, int_hexagon_M4_pmpyw>;
// Rxx^=pmpyw(Rs,Rt)
def : T_PRR_pat <M4_pmpyw_acc, int_hexagon_M4_pmpyw_acc>;
//Rxx^=asr(Rss,Rt)
def : T_PPR_pat <S2_asr_r_p_xor, int_hexagon_S2_asr_r_p_xor>;
//Rxx^=asl(Rss,Rt)
def : T_PPR_pat <S2_asl_r_p_xor, int_hexagon_S2_asl_r_p_xor>;
//Rxx^=lsr(Rss,Rt)
def : T_PPR_pat <S2_lsr_r_p_xor, int_hexagon_S2_lsr_r_p_xor>;
//Rxx^=lsl(Rss,Rt)
def : T_PPR_pat <S2_lsl_r_p_xor, int_hexagon_S2_lsl_r_p_xor>;
// Multiply and use upper result
def : MType_R32_pat <int_hexagon_M2_mpysu_up, M2_mpysu_up>;
def : MType_R32_pat <int_hexagon_M2_mpy_up_s1, M2_mpy_up_s1>;
def : MType_R32_pat <int_hexagon_M2_hmmpyh_s1, M2_hmmpyh_s1>;
def : MType_R32_pat <int_hexagon_M2_hmmpyl_s1, M2_hmmpyl_s1>;
def : MType_R32_pat <int_hexagon_M2_mpy_up_s1_sat, M2_mpy_up_s1_sat>;
// Vector reduce add unsigned halfwords
def : Pat <(int_hexagon_M2_vraddh DoubleRegs:$src1, DoubleRegs:$src2),
(M2_vraddh DoubleRegs:$src1, DoubleRegs:$src2)>;
def : T_P_pat <S2_brevp, int_hexagon_S2_brevp>;
def: T_P_pat <S2_ct0p, int_hexagon_S2_ct0p>;
def: T_P_pat <S2_ct1p, int_hexagon_S2_ct1p>;
def: T_RR_pat<C4_nbitsset, int_hexagon_C4_nbitsset>;
def: T_RR_pat<C4_nbitsclr, int_hexagon_C4_nbitsclr>;
def: T_RI_pat<C4_nbitsclri, int_hexagon_C4_nbitsclri>;
class vcmpImm_pat <InstHexagon MI, Intrinsic IntID, PatLeaf immPred> :
Pat <(IntID (i64 DoubleRegs:$src1), immPred:$src2),
(MI (i64 DoubleRegs:$src1), immPred:$src2)>;
def : vcmpImm_pat <A4_vcmpbeqi, int_hexagon_A4_vcmpbeqi, u8ImmPred>;
def : vcmpImm_pat <A4_vcmpbgti, int_hexagon_A4_vcmpbgti, s8ImmPred>;
def : vcmpImm_pat <A4_vcmpbgtui, int_hexagon_A4_vcmpbgtui, u7ImmPred>;
def : vcmpImm_pat <A4_vcmpheqi, int_hexagon_A4_vcmpheqi, s8ImmPred>;
def : vcmpImm_pat <A4_vcmphgti, int_hexagon_A4_vcmphgti, s8ImmPred>;
def : vcmpImm_pat <A4_vcmphgtui, int_hexagon_A4_vcmphgtui, u7ImmPred>;
def : vcmpImm_pat <A4_vcmpweqi, int_hexagon_A4_vcmpweqi, s8ImmPred>;
def : vcmpImm_pat <A4_vcmpwgti, int_hexagon_A4_vcmpwgti, s8ImmPred>;
def : vcmpImm_pat <A4_vcmpwgtui, int_hexagon_A4_vcmpwgtui, u7ImmPred>;
def : T_PP_pat<A4_vcmpbeq_any, int_hexagon_A4_vcmpbeq_any>;
def : T_RR_pat<A4_cmpbeq, int_hexagon_A4_cmpbeq>;
def : T_RR_pat<A4_cmpbgt, int_hexagon_A4_cmpbgt>;
def : T_RR_pat<A4_cmpbgtu, int_hexagon_A4_cmpbgtu>;
def : T_RR_pat<A4_cmpheq, int_hexagon_A4_cmpheq>;
def : T_RR_pat<A4_cmphgt, int_hexagon_A4_cmphgt>;
def : T_RR_pat<A4_cmphgtu, int_hexagon_A4_cmphgtu>;
def : T_RI_pat<A4_cmpbeqi, int_hexagon_A4_cmpbeqi>;
def : T_RI_pat<A4_cmpbgti, int_hexagon_A4_cmpbgti>;
def : T_RI_pat<A4_cmpbgtui, int_hexagon_A4_cmpbgtui>;
def : T_RI_pat<A4_cmpheqi, int_hexagon_A4_cmpheqi>;
def : T_RI_pat<A4_cmphgti, int_hexagon_A4_cmphgti>;
def : T_RI_pat<A4_cmphgtui, int_hexagon_A4_cmphgtui>;
def : T_RP_pat <A4_boundscheck, int_hexagon_A4_boundscheck>;
def : T_PR_pat<A4_tlbmatch, int_hexagon_A4_tlbmatch>;
def : Pat <(int_hexagon_M4_mpyrr_addr IntRegs:$src1, IntRegs:$src2,
IntRegs:$src3),
(M4_mpyrr_addr IntRegs:$src1, IntRegs:$src2, IntRegs:$src3)>;
def : T_IRR_pat <M4_mpyrr_addi, int_hexagon_M4_mpyrr_addi>;
def : T_IRI_pat <M4_mpyri_addi, int_hexagon_M4_mpyri_addi>;
def : T_RIR_pat <M4_mpyri_addr_u2, int_hexagon_M4_mpyri_addr_u2>;
def : T_RRI_pat <M4_mpyri_addr, int_hexagon_M4_mpyri_addr>;
// Multiply 32x32 and use upper result
def : T_RRR_pat <M4_mac_up_s1_sat, int_hexagon_M4_mac_up_s1_sat>;
def : T_RRR_pat <M4_nac_up_s1_sat, int_hexagon_M4_nac_up_s1_sat>;
// Complex multiply 32x16
def : T_PR_pat <M4_cmpyi_wh, int_hexagon_M4_cmpyi_wh>;
def : T_PR_pat <M4_cmpyr_wh, int_hexagon_M4_cmpyr_wh>;
def : T_PR_pat <M4_cmpyi_whc, int_hexagon_M4_cmpyi_whc>;
def : T_PR_pat <M4_cmpyr_whc, int_hexagon_M4_cmpyr_whc>;
def : T_PP_pat<A4_andnp, int_hexagon_A4_andnp>;
def : T_PP_pat<A4_ornp, int_hexagon_A4_ornp>;
// Complex add/sub halfwords/words
def : T_PP_pat <S4_vxaddsubw, int_hexagon_S4_vxaddsubw>;
def : T_PP_pat <S4_vxsubaddw, int_hexagon_S4_vxsubaddw>;
def : T_PP_pat <S4_vxaddsubh, int_hexagon_S4_vxaddsubh>;
def : T_PP_pat <S4_vxsubaddh, int_hexagon_S4_vxsubaddh>;
def : T_PP_pat <S4_vxaddsubhr, int_hexagon_S4_vxaddsubhr>;
def : T_PP_pat <S4_vxsubaddhr, int_hexagon_S4_vxsubaddhr>;
// Extract bitfield
def : T_PP_pat <S4_extractp_rp, int_hexagon_S4_extractp_rp>;
def : T_RP_pat <S4_extract_rp, int_hexagon_S4_extract_rp>;
def : T_PII_pat <S4_extractp, int_hexagon_S4_extractp>;
def : T_RII_pat <S4_extract, int_hexagon_S4_extract>;
// Vector conditional negate
// Rdd=vcnegh(Rss,Rt)
def : T_PR_pat <S2_vcnegh, int_hexagon_S2_vcnegh>;
// Shift an immediate left by register amount
def : T_IR_pat<S4_lsli, int_hexagon_S4_lsli>;
// Vector reduce maximum halfwords
def : T_PPR_pat <A4_vrmaxh, int_hexagon_A4_vrmaxh>;
def : T_PPR_pat <A4_vrmaxuh, int_hexagon_A4_vrmaxuh>;
// Vector reduce maximum words
def : T_PPR_pat <A4_vrmaxw, int_hexagon_A4_vrmaxw>;
def : T_PPR_pat <A4_vrmaxuw, int_hexagon_A4_vrmaxuw>;
// Vector reduce minimum halfwords
def : T_PPR_pat <A4_vrminh, int_hexagon_A4_vrminh>;
def : T_PPR_pat <A4_vrminuh, int_hexagon_A4_vrminuh>;
// Vector reduce minimum words
def : T_PPR_pat <A4_vrminw, int_hexagon_A4_vrminw>;
def : T_PPR_pat <A4_vrminuw, int_hexagon_A4_vrminuw>;
// Rotate and reduce bytes
def : Pat <(int_hexagon_S4_vrcrotate DoubleRegs:$src1, IntRegs:$src2,
u2ImmPred:$src3),
(S4_vrcrotate DoubleRegs:$src1, IntRegs:$src2, u2ImmPred:$src3)>;
// Rotate and reduce bytes with accumulation
// Rxx+=vrcrotate(Rss,Rt,#u2)
def : Pat <(int_hexagon_S4_vrcrotate_acc DoubleRegs:$src1, DoubleRegs:$src2,
IntRegs:$src3, u2ImmPred:$src4),
(S4_vrcrotate_acc DoubleRegs:$src1, DoubleRegs:$src2,
IntRegs:$src3, u2ImmPred:$src4)>;
// Vector conditional negate
def : T_PPR_pat<S2_vrcnegh, int_hexagon_S2_vrcnegh>;
// Logical xor with xor accumulation
def : T_PPP_pat<M4_xor_xacc, int_hexagon_M4_xor_xacc>;
// ALU64 - Vector min/max byte
def : T_PP_pat <A2_vminb, int_hexagon_A2_vminb>;
def : T_PP_pat <A2_vmaxb, int_hexagon_A2_vmaxb>;
// Shift and add/sub/and/or
def : T_IRI_pat <S4_andi_asl_ri, int_hexagon_S4_andi_asl_ri>;
def : T_IRI_pat <S4_ori_asl_ri, int_hexagon_S4_ori_asl_ri>;
def : T_IRI_pat <S4_addi_asl_ri, int_hexagon_S4_addi_asl_ri>;
def : T_IRI_pat <S4_subi_asl_ri, int_hexagon_S4_subi_asl_ri>;
def : T_IRI_pat <S4_andi_lsr_ri, int_hexagon_S4_andi_lsr_ri>;
def : T_IRI_pat <S4_ori_lsr_ri, int_hexagon_S4_ori_lsr_ri>;
def : T_IRI_pat <S4_addi_lsr_ri, int_hexagon_S4_addi_lsr_ri>;
def : T_IRI_pat <S4_subi_lsr_ri, int_hexagon_S4_subi_lsr_ri>;
// Split bitfield
def : T_RI_pat <A4_bitspliti, int_hexagon_A4_bitspliti>;
def : T_RR_pat <A4_bitsplit, int_hexagon_A4_bitsplit>;
def: T_RR_pat<S4_parity, int_hexagon_S4_parity>;
def: T_RI_pat<S4_ntstbit_i, int_hexagon_S4_ntstbit_i>;
def: T_RR_pat<S4_ntstbit_r, int_hexagon_S4_ntstbit_r>;
def: T_RI_pat<S4_clbaddi, int_hexagon_S4_clbaddi>;
def: T_PI_pat<S4_clbpaddi, int_hexagon_S4_clbpaddi>;
def: T_P_pat <S4_clbpnorm, int_hexagon_S4_clbpnorm>;
/********************************************************************
* ALU32/ALU *
*********************************************************************/
// ALU32 / ALU / Logical Operations.
def: T_RR_pat<A4_andn, int_hexagon_A4_andn>;
def: T_RR_pat<A4_orn, int_hexagon_A4_orn>;
/********************************************************************
* ALU32/PERM *
*********************************************************************/
// Combine Words Into Doublewords.
def: T_RI_pat<A4_combineri, int_hexagon_A4_combineri, s8ExtPred>;
def: T_IR_pat<A4_combineir, int_hexagon_A4_combineir, s8ExtPred>;
/********************************************************************
* ALU32/PRED *
*********************************************************************/
// Compare
def : T_RI_pat<C4_cmpneqi, int_hexagon_C4_cmpneqi, s10ExtPred>;
def : T_RI_pat<C4_cmpltei, int_hexagon_C4_cmpltei, s10ExtPred>;
def : T_RI_pat<C4_cmplteui, int_hexagon_C4_cmplteui, u9ExtPred>;
def: T_RR_pat<A4_rcmpeq, int_hexagon_A4_rcmpeq>;
def: T_RR_pat<A4_rcmpneq, int_hexagon_A4_rcmpneq>;
def: T_RI_pat<A4_rcmpeqi, int_hexagon_A4_rcmpeqi>;
def: T_RI_pat<A4_rcmpneqi, int_hexagon_A4_rcmpneqi>;
/********************************************************************
* CR *
*********************************************************************/
// CR / Logical Operations On Predicates.
class qi_CRInst_qiqiqi_pat<Intrinsic IntID, InstHexagon Inst> :
Pat<(i32 (IntID IntRegs:$Rs, IntRegs:$Rt, IntRegs:$Ru)),
(i32 (C2_tfrpr (Inst (C2_tfrrp IntRegs:$Rs),
(C2_tfrrp IntRegs:$Rt),
(C2_tfrrp IntRegs:$Ru))))>;
def: qi_CRInst_qiqiqi_pat<int_hexagon_C4_and_and, C4_and_and>;
def: qi_CRInst_qiqiqi_pat<int_hexagon_C4_and_andn, C4_and_andn>;
def: qi_CRInst_qiqiqi_pat<int_hexagon_C4_and_or, C4_and_or>;
def: qi_CRInst_qiqiqi_pat<int_hexagon_C4_and_orn, C4_and_orn>;
def: qi_CRInst_qiqiqi_pat<int_hexagon_C4_or_and, C4_or_and>;
def: qi_CRInst_qiqiqi_pat<int_hexagon_C4_or_andn, C4_or_andn>;
def: qi_CRInst_qiqiqi_pat<int_hexagon_C4_or_or, C4_or_or>;
def: qi_CRInst_qiqiqi_pat<int_hexagon_C4_or_orn, C4_or_orn>;
/********************************************************************
* XTYPE/ALU *
*********************************************************************/
// Add And Accumulate.
def : T_RRI_pat <S4_addaddi, int_hexagon_S4_addaddi>;
def : T_RIR_pat <S4_subaddi, int_hexagon_S4_subaddi>;
// XTYPE / ALU / Logical-logical Words.
def : T_RRR_pat <M4_or_xor, int_hexagon_M4_or_xor>;
def : T_RRR_pat <M4_and_xor, int_hexagon_M4_and_xor>;
def : T_RRR_pat <M4_or_and, int_hexagon_M4_or_and>;
def : T_RRR_pat <M4_and_and, int_hexagon_M4_and_and>;
def : T_RRR_pat <M4_xor_and, int_hexagon_M4_xor_and>;
def : T_RRR_pat <M4_or_or, int_hexagon_M4_or_or>;
def : T_RRR_pat <M4_and_or, int_hexagon_M4_and_or>;
def : T_RRR_pat <M4_xor_or, int_hexagon_M4_xor_or>;
def : T_RRR_pat <M4_or_andn, int_hexagon_M4_or_andn>;
def : T_RRR_pat <M4_and_andn, int_hexagon_M4_and_andn>;
def : T_RRR_pat <M4_xor_andn, int_hexagon_M4_xor_andn>;
def : T_RRI_pat <S4_or_andi, int_hexagon_S4_or_andi>;
def : T_RRI_pat <S4_or_andix, int_hexagon_S4_or_andix>;
def : T_RRI_pat <S4_or_ori, int_hexagon_S4_or_ori>;
// Modulo wrap.
def : T_RR_pat <A4_modwrapu, int_hexagon_A4_modwrapu>;
// Arithmetic/Convergent round
// Rd=[cround|round](Rs,Rt)[:sat]
// Rd=[cround|round](Rs,#u5)[:sat]
def : T_RI_pat <A4_cround_ri, int_hexagon_A4_cround_ri>;
def : T_RR_pat <A4_cround_rr, int_hexagon_A4_cround_rr>;
def : T_RI_pat <A4_round_ri, int_hexagon_A4_round_ri>;
def : T_RR_pat <A4_round_rr, int_hexagon_A4_round_rr>;
def : T_RI_pat <A4_round_ri_sat, int_hexagon_A4_round_ri_sat>;
def : T_RR_pat <A4_round_rr_sat, int_hexagon_A4_round_rr_sat>;
def : T_P_pat <A2_roundsat, int_hexagon_A2_roundsat>;
|
