111
|
1 ;; ARM Cortex-A15 NEON pipeline description
|
131
|
2 ;; Copyright (C) 2012-2018 Free Software Foundation, Inc.
|
111
|
3 ;;
|
|
4 ;; This file is part of GCC.
|
|
5 ;;
|
|
6 ;; GCC is free software; you can redistribute it and/or modify it
|
|
7 ;; under the terms of the GNU General Public License as published by
|
|
8 ;; the Free Software Foundation; either version 3, or (at your option)
|
|
9 ;; any later version.
|
|
10 ;;
|
|
11 ;; GCC is distributed in the hope that it will be useful, but
|
|
12 ;; WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
13 ;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
14 ;; General Public License for more details.
|
|
15 ;;
|
|
16 ;; You should have received a copy of the GNU General Public License
|
|
17 ;; along with GCC; see the file COPYING3. If not see
|
|
18 ;; <http://www.gnu.org/licenses/>.
|
|
19
|
|
20 (define_attr "cortex_a15_neon_type"
|
|
21 "neon_abd, neon_abd_q, neon_arith_acc, neon_arith_acc_q,
|
|
22 neon_arith_basic, neon_arith_complex,
|
|
23 neon_reduc_add_acc, neon_multiply, neon_multiply_q,
|
|
24 neon_multiply_long, neon_mla, neon_mla_q, neon_mla_long,
|
|
25 neon_sat_mla_long, neon_shift_acc, neon_shift_imm_basic,\
|
|
26 neon_shift_imm_complex,
|
|
27 neon_shift_reg_basic, neon_shift_reg_basic_q, neon_shift_reg_complex,
|
|
28 neon_shift_reg_complex_q, neon_fp_negabs, neon_fp_arith,
|
|
29 neon_fp_arith_q, neon_fp_cvt_int,
|
|
30 neon_fp_cvt_int_q, neon_fp_cvt16, neon_fp_minmax, neon_fp_mul,
|
|
31 neon_fp_mul_q, neon_fp_mla, neon_fp_mla_q, neon_fp_recpe_rsqrte,
|
|
32 neon_fp_recpe_rsqrte_q, neon_bitops, neon_bitops_q, neon_from_gp,
|
|
33 neon_from_gp_q, neon_move, neon_tbl3_tbl4, neon_zip_q, neon_to_gp,
|
|
34 neon_load_a, neon_load_b, neon_load_c, neon_load_d, neon_load_e,
|
|
35 neon_load_f, neon_store_a, neon_store_b, neon_store_c, neon_store_d,
|
|
36 neon_store_e, neon_store_f, neon_store_g, neon_store_h,
|
|
37 unknown"
|
|
38 (cond [
|
|
39 (eq_attr "type" "neon_abd, neon_abd_long")
|
|
40 (const_string "neon_abd")
|
|
41 (eq_attr "type" "neon_abd_q")
|
|
42 (const_string "neon_abd_q")
|
|
43 (eq_attr "type" "neon_arith_acc, neon_reduc_add_acc,\
|
|
44 neon_reduc_add_acc_q")
|
|
45 (const_string "neon_arith_acc")
|
|
46 (eq_attr "type" "neon_arith_acc_q")
|
|
47 (const_string "neon_arith_acc_q")
|
|
48 (eq_attr "type" "neon_add, neon_add_q, neon_add_long,\
|
|
49 neon_add_widen, neon_neg, neon_neg_q,\
|
|
50 neon_reduc_add, neon_reduc_add_q,\
|
|
51 neon_reduc_add_long, neon_sub, neon_sub_q,\
|
|
52 neon_sub_long, neon_sub_widen, neon_logic,\
|
|
53 neon_logic_q, neon_tst, neon_tst_q")
|
|
54 (const_string "neon_arith_basic")
|
|
55 (eq_attr "type" "neon_abs, neon_abs_q, neon_add_halve_narrow_q,\
|
|
56 neon_add_halve, neon_add_halve_q,\
|
|
57 neon_sub_halve, neon_sub_halve_q, neon_qabs,\
|
|
58 neon_qabs_q, neon_qadd, neon_qadd_q, neon_qneg,\
|
|
59 neon_qneg_q, neon_qsub, neon_qsub_q,\
|
|
60 neon_sub_halve_narrow_q,\
|
|
61 neon_compare, neon_compare_q,\
|
|
62 neon_compare_zero, neon_compare_zero_q,\
|
|
63 neon_minmax, neon_minmax_q, neon_reduc_minmax,\
|
|
64 neon_reduc_minmax_q")
|
|
65 (const_string "neon_arith_complex")
|
|
66
|
|
67 (eq_attr "type" "neon_mul_b, neon_mul_h, neon_mul_s,\
|
|
68 neon_mul_h_scalar, neon_mul_s_scalar,\
|
|
69 neon_sat_mul_b, neon_sat_mul_h,\
|
|
70 neon_sat_mul_s, neon_sat_mul_h_scalar,\
|
|
71 neon_sat_mul_s_scalar,\
|
|
72 neon_mul_b_long, neon_mul_h_long,\
|
|
73 neon_mul_s_long,\
|
|
74 neon_mul_h_scalar_long, neon_mul_s_scalar_long,\
|
|
75 neon_sat_mul_b_long, neon_sat_mul_h_long,\
|
|
76 neon_sat_mul_s_long, neon_sat_mul_h_scalar_long,\
|
|
77 neon_sat_mul_s_scalar_long")
|
|
78 (const_string "neon_multiply")
|
|
79 (eq_attr "type" "neon_mul_b_q, neon_mul_h_q, neon_mul_s_q,\
|
|
80 neon_mul_h_scalar_q, neon_mul_s_scalar_q,\
|
|
81 neon_sat_mul_b_q, neon_sat_mul_h_q,\
|
|
82 neon_sat_mul_s_q, neon_sat_mul_h_scalar_q,\
|
|
83 neon_sat_mul_s_scalar_q")
|
|
84 (const_string "neon_multiply_q")
|
|
85 (eq_attr "type" "neon_mla_b, neon_mla_h, neon_mla_s,\
|
|
86 neon_mla_h_scalar, neon_mla_s_scalar,\
|
|
87 neon_mla_b_long, neon_mla_h_long,\
|
|
88 neon_mla_s_long,\
|
|
89 neon_mla_h_scalar_long, neon_mla_s_scalar_long")
|
|
90 (const_string "neon_mla")
|
|
91 (eq_attr "type" "neon_mla_b_q, neon_mla_h_q, neon_mla_s_q,\
|
|
92 neon_mla_h_scalar_q, neon_mla_s_scalar_q")
|
|
93 (const_string "neon_mla_q")
|
|
94 (eq_attr "type" "neon_sat_mla_b_long, neon_sat_mla_h_long,\
|
|
95 neon_sat_mla_s_long, neon_sat_mla_h_scalar_long,\
|
|
96 neon_sat_mla_s_scalar_long")
|
|
97 (const_string "neon_sat_mla_long")
|
|
98
|
|
99 (eq_attr "type" "neon_shift_acc, neon_shift_acc_q")
|
|
100 (const_string "neon_shift_acc")
|
|
101 (eq_attr "type" "neon_shift_imm, neon_shift_imm_q,\
|
|
102 neon_shift_imm_narrow_q, neon_shift_imm_long")
|
|
103 (const_string "neon_shift_imm_basic")
|
|
104 (eq_attr "type" "neon_sat_shift_imm, neon_sat_shift_imm_q,\
|
|
105 neon_sat_shift_imm_narrow_q")
|
|
106 (const_string "neon_shift_imm_complex")
|
|
107 (eq_attr "type" "neon_shift_reg")
|
|
108 (const_string "neon_shift_reg_basic")
|
|
109 (eq_attr "type" "neon_shift_reg_q")
|
|
110 (const_string "neon_shift_reg_basic_q")
|
|
111 (eq_attr "type" "neon_sat_shift_reg")
|
|
112 (const_string "neon_shift_reg_complex")
|
|
113 (eq_attr "type" "neon_sat_shift_reg_q")
|
|
114 (const_string "neon_shift_reg_complex_q")
|
|
115
|
|
116 (eq_attr "type" "neon_fp_neg_s, neon_fp_neg_s_q,\
|
|
117 neon_fp_abs_s, neon_fp_abs_s_q")
|
|
118 (const_string "neon_fp_negabs")
|
|
119 (eq_attr "type" "neon_fp_addsub_s, neon_fp_abd_s,\
|
|
120 neon_fp_reduc_add_s, neon_fp_compare_s,\
|
|
121 neon_fp_minmax_s, neon_fp_minmax_s_q,\
|
|
122 neon_fp_reduc_minmax_s, neon_fp_reduc_minmax_s_q")
|
|
123 (const_string "neon_fp_arith")
|
|
124 (eq_attr "type" "neon_fp_addsub_s_q, neon_fp_abd_s_q,\
|
|
125 neon_fp_reduc_add_s_q, neon_fp_compare_s_q")
|
|
126 (const_string "neon_fp_arith_q")
|
|
127 (eq_attr "type" "neon_fp_to_int_s, neon_int_to_fp_s")
|
|
128 (const_string "neon_fp_cvt_int")
|
|
129 (eq_attr "type" "neon_fp_to_int_s_q, neon_int_to_fp_s_q")
|
|
130 (const_string "neon_fp_cvt_int_q")
|
|
131 (eq_attr "type" "neon_fp_cvt_narrow_s_q, neon_fp_cvt_widen_h")
|
|
132 (const_string "neon_fp_cvt16")
|
|
133 (eq_attr "type" "neon_fp_mul_s, neon_fp_mul_s_scalar")
|
|
134 (const_string "neon_fp_mul")
|
|
135 (eq_attr "type" "neon_fp_mul_s_q, neon_fp_mul_s_scalar_q")
|
|
136 (const_string "neon_fp_mul_q")
|
|
137 (eq_attr "type" "neon_fp_mla_s, neon_fp_mla_s_scalar")
|
|
138 (const_string "neon_fp_mla")
|
|
139 (eq_attr "type" "neon_fp_mla_s_q, neon_fp_mla_s_scalar_q")
|
|
140 (const_string "neon_fp_mla_q")
|
|
141 (eq_attr "type" "neon_fp_recpe_s, neon_fp_rsqrte_s")
|
|
142 (const_string "neon_fp_recpe_rsqrte")
|
|
143 (eq_attr "type" "neon_fp_recpe_s_q, neon_fp_rsqrte_s_q")
|
|
144 (const_string "neon_fp_recpe_rsqrte_q")
|
|
145
|
|
146 (eq_attr "type" "neon_bsl, neon_cls, neon_cnt,\
|
|
147 neon_rev, neon_permute,\
|
|
148 neon_tbl1, neon_tbl2, neon_zip,\
|
|
149 neon_dup, neon_dup_q, neon_ext, neon_ext_q,\
|
|
150 neon_move, neon_move_q, neon_move_narrow_q")
|
|
151 (const_string "neon_bitops")
|
|
152 (eq_attr "type" "neon_bsl_q, neon_cls_q, neon_cnt_q,\
|
|
153 neon_rev_q, neon_permute_q")
|
|
154 (const_string "neon_bitops_q")
|
|
155 (eq_attr "type" "neon_from_gp")
|
|
156 (const_string "neon_from_gp")
|
|
157 (eq_attr "type" "neon_from_gp_q")
|
|
158 (const_string "neon_from_gp_q")
|
|
159 (eq_attr "type" "neon_tbl3, neon_tbl4")
|
|
160 (const_string "neon_tbl3_tbl4")
|
|
161 (eq_attr "type" "neon_zip_q")
|
|
162 (const_string "neon_zip_q")
|
|
163 (eq_attr "type" "neon_to_gp, neon_to_gp_q")
|
|
164 (const_string "neon_to_gp")
|
|
165
|
|
166 (eq_attr "type" "f_loads, f_loadd,\
|
|
167 neon_load1_1reg, neon_load1_1reg_q,\
|
|
168 neon_load1_2reg, neon_load1_2reg_q")
|
|
169 (const_string "neon_load_a")
|
|
170 (eq_attr "type" "neon_load1_3reg, neon_load1_3reg_q,\
|
|
171 neon_load1_4reg, neon_load1_4reg_q")
|
|
172 (const_string "neon_load_b")
|
|
173 (eq_attr "type" "neon_load1_one_lane, neon_load1_one_lane_q,\
|
|
174 neon_load1_all_lanes, neon_load1_all_lanes_q,\
|
|
175 neon_load2_2reg, neon_load2_2reg_q,\
|
|
176 neon_load2_all_lanes, neon_load2_all_lanes_q")
|
|
177 (const_string "neon_load_c")
|
|
178 (eq_attr "type" "neon_load2_4reg, neon_load2_4reg_q,\
|
|
179 neon_load3_3reg, neon_load3_3reg_q,\
|
|
180 neon_load3_one_lane, neon_load3_one_lane_q,\
|
|
181 neon_load4_4reg, neon_load4_4reg_q")
|
|
182 (const_string "neon_load_d")
|
|
183 (eq_attr "type" "neon_load2_one_lane, neon_load2_one_lane_q,\
|
|
184 neon_load3_all_lanes, neon_load3_all_lanes_q,\
|
|
185 neon_load4_all_lanes, neon_load4_all_lanes_q")
|
|
186 (const_string "neon_load_e")
|
|
187 (eq_attr "type" "neon_load4_one_lane, neon_load4_one_lane_q")
|
|
188 (const_string "neon_load_f")
|
|
189
|
|
190 (eq_attr "type" "f_stores, f_stored,\
|
|
191 neon_store1_1reg, neon_store1_1reg_q")
|
|
192 (const_string "neon_store_a")
|
|
193 (eq_attr "type" "neon_store1_2reg, neon_store1_2reg_q")
|
|
194 (const_string "neon_store_b")
|
|
195 (eq_attr "type" "neon_store1_3reg, neon_store1_3reg_q")
|
|
196 (const_string "neon_store_c")
|
|
197 (eq_attr "type" "neon_store1_4reg, neon_store1_4reg_q")
|
|
198 (const_string "neon_store_d")
|
|
199 (eq_attr "type" "neon_store1_one_lane, neon_store1_one_lane_q,\
|
|
200 neon_store2_one_lane, neon_store2_one_lane_q")
|
|
201 (const_string "neon_store_e")
|
|
202 (eq_attr "type" "neon_store2_2reg, neon_store2_2reg_q,\
|
|
203 neon_store3_one_lane, neon_store3_one_lane_q,\
|
|
204 neon_store4_one_lane, neon_store4_one_lane_q")
|
|
205 (const_string "neon_store_f")
|
|
206 (eq_attr "type" "neon_store2_4reg, neon_store2_4reg_q,\
|
|
207 neon_store4_4reg, neon_store4_4reg_q")
|
|
208 (const_string "neon_store_g")
|
|
209 (eq_attr "type" "neon_store3_3reg, neon_store3_3reg_q")
|
|
210 (const_string "neon_store_h")]
|
|
211 (const_string "unknown")))
|
|
212
|
|
213 (define_automaton "cortex_a15_neon")
|
|
214
|
|
215 ;; Dispatch unit.
|
|
216 (define_cpu_unit "ca15_cx_ij, ca15_cx_ik" "cortex_a15_neon")
|
|
217
|
|
218 ;; Accumulate.
|
|
219 (define_cpu_unit "ca15_cx_acc" "cortex_a15_neon")
|
|
220
|
|
221 ;; The 32x32 integer multiply-accumulate pipeline.
|
|
222 (define_cpu_unit "ca15_cx_imac1" "cortex_a15_neon")
|
|
223 (define_reservation "ca15_cx_imac" "(ca15_cx_ij+ca15_cx_imac1)")
|
|
224
|
|
225
|
|
226 ;; The 64-bit ALU pipeline.
|
|
227 (define_cpu_unit "ca15_cx_ialu1, ca15_cx_ialu2" "cortex_a15_neon")
|
|
228
|
|
229 ;; IALU with accumulate.
|
|
230 (define_reservation "ca15_cx_ialu_with_acc" "ca15_cx_ik+ca15_cx_ialu2+ca15_cx_acc")
|
|
231
|
|
232 (define_reservation "ca15_cx_ialu"
|
|
233 "((ca15_cx_ij+ca15_cx_ialu1)|(ca15_cx_ik+ca15_cx_ialu2))")
|
|
234
|
|
235 ;; Integer shift pipeline.
|
|
236 (define_cpu_unit "ca15_cx_ishf" "cortex_a15_neon")
|
|
237 (define_reservation "ca15_cx_ishf_with_acc" "ca15_cx_ik+ca15_cx_ishf+ca15_cx_acc")
|
|
238
|
|
239 ;; SIMD multiply pipeline.
|
|
240 (define_cpu_unit "ca15_cx_fmul1, ca15_cx_fmul2, ca15_cx_fmul3, ca15_cx_fmul4"
|
|
241 "cortex_a15_neon")
|
|
242
|
|
243 (define_reservation "ca15_cx_fmul"
|
|
244 "(ca15_cx_ij+(ca15_cx_fmul1|ca15_cx_fmul2))|\
|
|
245 (ca15_cx_ik+(ca15_cx_fmul3|ca15_cx_fmul4))")
|
|
246
|
|
247 (define_reservation "ca15_cx_fmul_2"
|
|
248 "(ca15_cx_ij+(ca15_cx_fmul1|ca15_cx_fmul2))+\
|
|
249 (ca15_cx_ik+(ca15_cx_fmul3|ca15_cx_fmul4))")
|
|
250
|
|
251 ;; SIMD ALU pipeline.
|
|
252 (define_cpu_unit "ca15_cx_falu1, ca15_cx_falu2, ca15_cx_falu3, ca15_cx_falu4"
|
|
253 "cortex_a15_neon")
|
|
254
|
|
255 (define_reservation "ca15_cx_falu"
|
|
256 "(ca15_cx_ij+(ca15_cx_falu1|ca15_cx_falu2))|\
|
|
257 (ca15_cx_ik+(ca15_cx_falu3|ca15_cx_falu4))")
|
|
258
|
|
259 (define_reservation "ca15_cx_falu_2"
|
|
260 "(ca15_cx_ij+(ca15_cx_falu1|ca15_cx_falu2))+\
|
|
261 (ca15_cx_ik+(ca15_cx_falu3|ca15_cx_falu4))")
|
|
262
|
|
263 ;; SIMD multiply-accumulate pipeline.
|
|
264 ;; This can be used if fmul and falu are not reserved.
|
|
265 (define_reservation "ca15_cx_fmac"
|
|
266 "((ca15_cx_ij+ca15_cx_fmul1),nothing*2,ca15_cx_falu1)|\
|
|
267 ((ca15_cx_ij+ca15_cx_fmul2),nothing*2,ca15_cx_falu2)|\
|
|
268 ((ca15_cx_ik+ca15_cx_fmul3),nothing*2,ca15_cx_falu3)|\
|
|
269 ((ca15_cx_ik+ca15_cx_fmul4),nothing*2,ca15_cx_falu4)")
|
|
270
|
|
271 (define_reservation "ca15_cx_fmac_2"
|
|
272 "(((ca15_cx_ij+ca15_cx_fmul1),nothing*2,ca15_cx_falu1)|\
|
|
273 ((ca15_cx_ij+ca15_cx_fmul2),nothing*2,ca15_cx_falu2))+\
|
|
274 (((ca15_cx_ik+ca15_cx_fmul3),nothing*2,ca15_cx_falu3)|\
|
|
275 ((ca15_cx_ik+ca15_cx_fmul4),nothing*2,ca15_cx_falu4))")
|
|
276
|
|
277
|
|
278 ;; Vector FP multiply pipeline
|
|
279 (define_cpu_unit "ca15_cx_vfp_i" "cortex_a15_neon")
|
|
280
|
|
281 (define_reservation "ca15_cx_vfp" "ca15_cx_ik+ca15_cx_vfp_i")
|
|
282
|
|
283 ;; Load permute pipeline
|
|
284 (define_reservation "ca15_cx_perm" "ca15_cx_ij|ca15_cx_ik")
|
|
285 (define_reservation "ca15_cx_perm_2" "ca15_cx_ij+ca15_cx_ik")
|
|
286
|
|
287 ;; Integer Arithmetic Instructions.
|
|
288
|
|
289 (define_insn_reservation "cortex_a15_neon_abd" 5
|
|
290 (and (eq_attr "tune" "cortexa15")
|
|
291 (eq_attr "cortex_a15_neon_type" "neon_abd"))
|
|
292 "ca15_issue1,ca15_cx_ialu")
|
|
293
|
|
294 (define_insn_reservation "cortex_a15_neon_abd_q" 5
|
|
295 (and (eq_attr "tune" "cortexa15")
|
|
296 (eq_attr "cortex_a15_neon_type" "neon_abd_q"))
|
|
297 "ca15_issue2,ca15_cx_ialu*2")
|
|
298
|
|
299 (define_insn_reservation "cortex_a15_neon_aba" 7
|
|
300 (and (eq_attr "tune" "cortexa15")
|
|
301 (eq_attr "cortex_a15_neon_type" "neon_arith_acc"))
|
|
302 "ca15_issue1,ca15_cx_ialu_with_acc")
|
|
303
|
|
304 (define_insn_reservation "cortex_a15_neon_aba_q" 8
|
|
305 (and (eq_attr "tune" "cortexa15")
|
|
306 (eq_attr "cortex_a15_neon_type" "neon_arith_acc_q"))
|
|
307 "ca15_issue2,ca15_cx_ialu_with_acc*2")
|
|
308
|
|
309 (define_insn_reservation "cortex_a15_neon_arith_basic" 4
|
|
310 (and (eq_attr "tune" "cortexa15")
|
|
311 (eq_attr "cortex_a15_neon_type" "neon_arith_basic"))
|
|
312 "ca15_issue1,ca15_cx_ialu")
|
|
313
|
|
314 (define_insn_reservation "cortex_a15_neon_arith_complex" 5
|
|
315 (and (eq_attr "tune" "cortexa15")
|
|
316 (eq_attr "cortex_a15_neon_type" "neon_arith_complex"))
|
|
317 "ca15_issue1,ca15_cx_ialu")
|
|
318
|
|
319 ;; Integer Multiply Instructions.
|
|
320
|
|
321 (define_insn_reservation "cortex_a15_neon_multiply" 6
|
|
322 (and (eq_attr "tune" "cortexa15")
|
|
323 (eq_attr "cortex_a15_neon_type" "neon_multiply"))
|
|
324 "ca15_issue1,ca15_cx_imac")
|
|
325
|
|
326 (define_insn_reservation "cortex_a15_neon_multiply_q" 7
|
|
327 (and (eq_attr "tune" "cortexa15")
|
|
328 (eq_attr "cortex_a15_neon_type" "neon_multiply_q"))
|
|
329 "ca15_issue2,ca15_cx_imac*2")
|
|
330
|
|
331 (define_insn_reservation "cortex_a15_neon_mla" 6
|
|
332 (and (eq_attr "tune" "cortexa15")
|
|
333 (eq_attr "cortex_a15_neon_type" "neon_mla"))
|
|
334 "ca15_issue1,ca15_cx_imac")
|
|
335
|
|
336 (define_insn_reservation "cortex_a15_neon_mla_q" 7
|
|
337 (and (eq_attr "tune" "cortexa15")
|
|
338 (eq_attr "cortex_a15_neon_type" "neon_mla_q"))
|
|
339 "ca15_issue1,ca15_cx_imac*2")
|
|
340
|
|
341 (define_insn_reservation "cortex_a15_neon_sat_mla_long" 6
|
|
342 (and (eq_attr "tune" "cortexa15")
|
|
343 (eq_attr "cortex_a15_neon_type" "neon_sat_mla_long"))
|
|
344 "ca15_issue1,ca15_cx_imac")
|
|
345
|
|
346 ;; Integer Shift Instructions.
|
|
347
|
|
348 (define_insn_reservation
|
|
349 "cortex_a15_neon_shift_acc" 7
|
|
350 (and (eq_attr "tune" "cortexa15")
|
|
351 (eq_attr "cortex_a15_neon_type" "neon_shift_acc"))
|
|
352 "ca15_issue1,ca15_cx_ishf_with_acc")
|
|
353
|
|
354 (define_insn_reservation
|
|
355 "cortex_a15_neon_shift_imm_basic" 4
|
|
356 (and (eq_attr "tune" "cortexa15")
|
|
357 (eq_attr "cortex_a15_neon_type" "neon_shift_imm_basic"))
|
|
358 "ca15_issue1,ca15_cx_ik+ca15_cx_ishf")
|
|
359
|
|
360 (define_insn_reservation
|
|
361 "cortex_a15_neon_shift_imm_complex" 5
|
|
362 (and (eq_attr "tune" "cortexa15")
|
|
363 (eq_attr "cortex_a15_neon_type" "neon_shift_imm_complex"))
|
|
364 "ca15_issue1,ca15_cx_ik+ca15_cx_ishf")
|
|
365
|
|
366 (define_insn_reservation
|
|
367 "cortex_a15_neon_shift_reg_basic" 4
|
|
368 (and (eq_attr "tune" "cortexa15")
|
|
369 (eq_attr "cortex_a15_neon_type" "neon_shift_reg_basic"))
|
|
370 "ca15_issue1,ca15_cx_ik+ca15_cx_ishf")
|
|
371
|
|
372 (define_insn_reservation
|
|
373 "cortex_a15_neon_shift_reg_basic_q" 5
|
|
374 (and (eq_attr "tune" "cortexa15")
|
|
375 (eq_attr "cortex_a15_neon_type" "neon_shift_reg_basic_q"))
|
|
376 "ca15_issue2,(ca15_cx_ik+ca15_cx_ishf*2)")
|
|
377
|
|
378 (define_insn_reservation
|
|
379 "cortex_a15_neon_shift_reg_complex" 5
|
|
380 (and (eq_attr "tune" "cortexa15")
|
|
381 (eq_attr "cortex_a15_neon_type" "neon_shift_reg_complex"))
|
|
382 "ca15_issue2,ca15_cx_ik+ca15_cx_ishf")
|
|
383
|
|
384 (define_insn_reservation
|
|
385 "cortex_a15_neon_shift_reg_complex_q" 6
|
|
386 (and (eq_attr "tune" "cortexa15")
|
|
387 (eq_attr "cortex_a15_neon_type" "neon_shift_reg_complex_q"))
|
|
388 "ca15_issue2,(ca15_cx_ik+ca15_cx_ishf)*2")
|
|
389
|
|
390 ;; Floating Point Instructions.
|
|
391
|
|
392 (define_insn_reservation
|
|
393 "cortex_a15_neon_fp_negabs" 4
|
|
394 (and (eq_attr "tune" "cortexa15")
|
|
395 (eq_attr "cortex_a15_neon_type" "neon_fp_negabs"))
|
|
396 "ca15_issue1,ca15_cx_falu")
|
|
397
|
|
398 (define_insn_reservation
|
|
399 "cortex_a15_neon_fp_arith" 6
|
|
400 (and (eq_attr "tune" "cortexa15")
|
|
401 (eq_attr "cortex_a15_neon_type" "neon_fp_arith"))
|
|
402 "ca15_issue1,ca15_cx_falu")
|
|
403
|
|
404 (define_insn_reservation
|
|
405 "cortex_a15_neon_fp_arith_q" 6
|
|
406 (and (eq_attr "tune" "cortexa15")
|
|
407 (eq_attr "cortex_a15_neon_type" "neon_fp_arith_q"))
|
|
408 "ca15_issue2,ca15_cx_falu_2")
|
|
409
|
|
410 (define_insn_reservation
|
|
411 "cortex_a15_neon_fp_cvt_int" 6
|
|
412 (and (eq_attr "tune" "cortexa15")
|
|
413 (eq_attr "cortex_a15_neon_type" "neon_fp_cvt_int"))
|
|
414 "ca15_issue1,ca15_cx_falu+ca15_cx_ishf")
|
|
415
|
|
416 (define_insn_reservation
|
|
417 "cortex_a15_neon_fp_cvt_int_q" 6
|
|
418 (and (eq_attr "tune" "cortexa15")
|
|
419 (eq_attr "cortex_a15_neon_type" "neon_fp_cvt_int_q"))
|
|
420 "ca15_issue2,(ca15_cx_falu+ca15_cx_ishf)*2")
|
|
421
|
|
422 (define_insn_reservation
|
|
423 "cortex_a15_neon_fp_cvt16" 10
|
|
424 (and (eq_attr "tune" "cortexa15")
|
|
425 (eq_attr "cortex_a15_neon_type" "neon_fp_cvt16"))
|
|
426 "ca15_issue3,(ca15_cx_falu+ca15_cx_ishf)*2+ca15_cx_falu")
|
|
427
|
|
428 (define_insn_reservation
|
|
429 "cortex_a15_neon_fp_mul" 5
|
|
430 (and (eq_attr "tune" "cortexa15")
|
|
431 (eq_attr "cortex_a15_neon_type" "neon_fp_mul"))
|
|
432 "ca15_issue1,ca15_cx_fmul")
|
|
433
|
|
434 (define_insn_reservation
|
|
435 "cortex_a15_neon_fp_mul_q" 5
|
|
436 (and (eq_attr "tune" "cortexa15")
|
|
437 (eq_attr "cortex_a15_neon_type" "neon_fp_mul_q"))
|
|
438 "ca15_issue2,ca15_cx_fmul_2")
|
|
439
|
|
440 (define_insn_reservation
|
|
441 "cortex_a15_neon_fp_mla" 9
|
|
442 (and (eq_attr "tune" "cortexa15")
|
|
443 (eq_attr "cortex_a15_neon_type" "neon_fp_mla"))
|
|
444 "ca15_issue1,ca15_cx_fmul")
|
|
445
|
|
446 (define_insn_reservation
|
|
447 "cortex_a15_neon_fp_mla_q" 9
|
|
448 (and (eq_attr "tune" "cortexa15")
|
|
449 (eq_attr "cortex_a15_neon_type" "neon_fp_mla_q"))
|
|
450 "ca15_issue2,ca15_cx_fmul_2")
|
|
451
|
|
452 (define_insn_reservation
|
|
453 "cortex_a15_neon_fp_recps_rsqrte" 9
|
|
454 (and (eq_attr "tune" "cortexa15")
|
|
455 (eq_attr "cortex_a15_neon_type" "neon_fp_recpe_rsqrte"))
|
|
456 "ca15_issue1,ca15_cx_fmac")
|
|
457
|
|
458 (define_insn_reservation
|
|
459 "cortex_a15_neon_fp_recps_rsqrte_q" 9
|
|
460 (and (eq_attr "tune" "cortexa15")
|
|
461 (eq_attr "cortex_a15_neon_type" "neon_fp_recpe_rsqrte_q"))
|
|
462 "ca15_issue2,ca15_cx_fmac_2")
|
|
463
|
|
464 ;; Miscelaaneous Instructions.
|
|
465
|
|
466 (define_insn_reservation
|
|
467 "cortex_a15_neon_bitops" 4
|
|
468 (and (eq_attr "tune" "cortexa15")
|
|
469 (eq_attr "cortex_a15_neon_type" "neon_bitops"))
|
|
470 "ca15_issue1,ca15_cx_perm")
|
|
471
|
|
472 (define_insn_reservation
|
|
473 "cortex_a15_neon_bitops_q" 4
|
|
474 (and (eq_attr "tune" "cortexa15")
|
|
475 (eq_attr "cortex_a15_neon_type" "neon_bitops_q"))
|
|
476 "ca15_issue2,ca15_cx_perm_2")
|
|
477
|
|
478 (define_insn_reservation
|
|
479 "cortex_a15_neon_from_gp" 9
|
|
480 (and (eq_attr "tune" "cortexa15")
|
|
481 (eq_attr "cortex_a15_neon_type" "neon_from_gp"))
|
|
482 "ca15_issue2,ca15_ls1+ca15_ls2+ca15_cx_perm")
|
|
483
|
|
484 (define_insn_reservation
|
|
485 "cortex_a15_neon_from_gp_q" 9
|
|
486 (and (eq_attr "tune" "cortexa15")
|
|
487 (eq_attr "cortex_a15_neon_type" "neon_from_gp_q"))
|
|
488 "ca15_issue2,ca15_ls1+ca15_ls2+ca15_cx_perm_2")
|
|
489
|
|
490 (define_insn_reservation
|
|
491 "cortex_a15_neon_tbl3_tbl4" 7
|
|
492 (and (eq_attr "tune" "cortexa15")
|
|
493 (eq_attr "cortex_a15_neon_type" "neon_tbl3_tbl4"))
|
|
494 "ca15_issue2,ca15_cx_perm_2")
|
|
495
|
|
496 (define_insn_reservation
|
|
497 "cortex_a15_neon_zip_q" 7
|
|
498 (and (eq_attr "tune" "cortexa15")
|
|
499 (eq_attr "cortex_a15_neon_type" "neon_zip_q"))
|
|
500 "ca15_issue3,ca15_cx_perm*3")
|
|
501
|
|
502 (define_insn_reservation
|
|
503 "cortex_a15_neon_to_gp" 7
|
|
504 (and (eq_attr "tune" "cortexa15")
|
|
505 (eq_attr "cortex_a15_neon_type" "neon_to_gp"))
|
|
506 "ca15_issue2,ca15_ls1+ca15_ls2")
|
|
507
|
|
508 ;; Load Instructions.
|
|
509
|
|
510 (define_insn_reservation
|
|
511 "cortex_a15_neon_load_a" 6
|
|
512 (and (eq_attr "tune" "cortexa15")
|
|
513 (eq_attr "cortex_a15_neon_type" "neon_load_a"))
|
|
514 "ca15_issue1,ca15_ls,ca15_ldr")
|
|
515
|
|
516 (define_insn_reservation
|
|
517 "cortex_a15_neon_load_b" 7
|
|
518 (and (eq_attr "tune" "cortexa15")
|
|
519 (eq_attr "cortex_a15_neon_type" "neon_load_b"))
|
|
520 "ca15_issue2,ca15_ls1+ca15_ls2,ca15_ldr,ca15_ldr")
|
|
521
|
|
522 (define_insn_reservation
|
|
523 "cortex_a15_neon_load_c" 9
|
|
524 (and (eq_attr "tune" "cortexa15")
|
|
525 (eq_attr "cortex_a15_neon_type" "neon_load_c"))
|
|
526 "ca15_issue2,ca15_ls1+ca15_ls2,ca15_ldr,ca15_ldr")
|
|
527
|
|
528 (define_insn_reservation
|
|
529 "cortex_a15_neon_load_d" 11
|
|
530 (and (eq_attr "tune" "cortexa15")
|
|
531 (eq_attr "cortex_a15_neon_type" "neon_load_d"))
|
|
532 "ca15_issue1,ca15_issue3+ca15_ls1+ca15_ls2,ca15_ldr*2")
|
|
533
|
|
534 (define_insn_reservation
|
|
535 "cortex_a15_neon_load_e" 9
|
|
536 (and (eq_attr "tune" "cortexa15")
|
|
537 (eq_attr "cortex_a15_neon_type" "neon_load_e"))
|
|
538 "ca15_issue3+ca15_ls1+ca15_ls2,ca15_ldr*2")
|
|
539
|
|
540 (define_insn_reservation
|
|
541 "cortex_a15_neon_load_f" 11
|
|
542 (and (eq_attr "tune" "cortexa15")
|
|
543 (eq_attr "cortex_a15_neon_type" "neon_load_f"))
|
|
544 "ca15_issue3,ca15_issue3+ca15_ls1+ca15_ls2,ca15_ldr*2")
|
|
545
|
|
546 ;; Store Instructions.
|
|
547
|
|
548 (define_insn_reservation
|
|
549 "cortex_a15_neon_store_a" 0
|
|
550 (and (eq_attr "tune" "cortexa15")
|
|
551 (eq_attr "cortex_a15_neon_type" "neon_store_a"))
|
|
552 "ca15_issue1,ca15_ls1+ca15_ls2,ca15_str")
|
|
553
|
|
554 (define_insn_reservation
|
|
555 "cortex_a15_neon_store_b" 0
|
|
556 (and (eq_attr "tune" "cortexa15")
|
|
557 (eq_attr "cortex_a15_neon_type" "neon_store_b"))
|
|
558 "ca15_issue2,ca15_ls1+ca15_ls2,ca15_str*2")
|
|
559
|
|
560 (define_insn_reservation
|
|
561 "cortex_a15_neon_store_c" 0
|
|
562 (and (eq_attr "tune" "cortexa15")
|
|
563 (eq_attr "cortex_a15_neon_type" "neon_store_c"))
|
|
564 "ca15_issue3,ca15_ls1+ca15_ls2,ca15_str*3")
|
|
565
|
|
566 (define_insn_reservation
|
|
567 "cortex_a15_neon_store_d" 0
|
|
568 (and (eq_attr "tune" "cortexa15")
|
|
569 (eq_attr "cortex_a15_neon_type" "neon_store_d"))
|
|
570 "ca15_issue3,ca15_issue1,ca15_ls1+ca15_ls2,ca15_str*4")
|
|
571
|
|
572 (define_insn_reservation
|
|
573 "cortex_a15_neon_store_e" 0
|
|
574 (and (eq_attr "tune" "cortexa15")
|
|
575 (eq_attr "cortex_a15_neon_type" "neon_store_e"))
|
|
576 "ca15_issue2,ca15_ls1+ca15_ls2,ca15_str+ca15_cx_perm")
|
|
577
|
|
578 (define_insn_reservation
|
|
579 "cortex_a15_neon_store_f" 0
|
|
580 (and (eq_attr "tune" "cortexa15")
|
|
581 (eq_attr "cortex_a15_neon_type" "neon_store_f"))
|
|
582 "ca15_issue3,ca15_ls1+ca15_ls2,ca15_str*2+ca15_cx_perm")
|
|
583
|
|
584 (define_insn_reservation
|
|
585 "cortex_a15_neon_store_g" 0
|
|
586 (and (eq_attr "tune" "cortexa15")
|
|
587 (eq_attr "cortex_a15_neon_type" "neon_store_g"))
|
|
588 "ca15_issue3,ca15_issue3+ca15_cx_perm+ca15_ls1+ca15_ls2,ca15_str*2")
|
|
589
|
|
590 (define_insn_reservation
|
|
591 "cortex_a15_neon_store_h" 0
|
|
592 (and (eq_attr "tune" "cortexa15")
|
|
593 (eq_attr "cortex_a15_neon_type" "neon_store_h"))
|
|
594 "ca15_issue3,ca15_issue2+ca15_cx_perm+ca15_ls1+ca15_ls2,ca15_str*2")
|
|
595
|
|
596 ;; VFP Operations.
|
|
597
|
|
598 (define_insn_reservation "cortex_a15_vfp_const" 4
|
|
599 (and (eq_attr "tune" "cortexa15")
|
|
600 (eq_attr "type" "fconsts,fconstd"))
|
|
601 "ca15_issue1,ca15_cx_perm")
|
|
602
|
|
603 (define_insn_reservation "cortex_a15_vfp_adds_subs" 6
|
|
604 (and (eq_attr "tune" "cortexa15")
|
|
605 (eq_attr "type" "fadds"))
|
|
606 "ca15_issue1,ca15_cx_vfp")
|
|
607
|
|
608 (define_insn_reservation "cortex_a15_vfp_addd_subd" 10
|
|
609 (and (eq_attr "tune" "cortexa15")
|
|
610 (eq_attr "type" "faddd"))
|
|
611 "ca15_issue2,ca15_cx_vfp*2")
|
|
612
|
|
613 (define_insn_reservation "cortex_a15_vfp_muls" 7
|
|
614 (and (eq_attr "tune" "cortexa15")
|
|
615 (eq_attr "type" "fmuls"))
|
|
616 "ca15_issue1,ca15_cx_vfp")
|
|
617
|
|
618 (define_insn_reservation "cortex_a15_vfp_muld" 12
|
|
619 (and (eq_attr "tune" "cortexa15")
|
|
620 (eq_attr "type" "fmuld"))
|
|
621 "ca15_issue2,ca15_cx_vfp*2")
|
|
622
|
|
623 (define_insn_reservation "cortex_a15_vfp_macs" 6
|
|
624 (and (eq_attr "tune" "cortexa15")
|
|
625 (eq_attr "type" "fmacs,ffmas"))
|
|
626 "ca15_issue1,ca15_cx_vfp")
|
|
627
|
|
628 (define_insn_reservation "cortex_a15_vfp_macd" 11
|
|
629 (and (eq_attr "tune" "cortexa15")
|
|
630 (eq_attr "type" "fmacd,ffmad"))
|
|
631 "ca15_issue2,ca15_cx_vfp*2")
|
|
632
|
|
633 (define_insn_reservation "cortex_a15_vfp_cvt" 6
|
|
634 (and (eq_attr "tune" "cortexa15")
|
|
635 (eq_attr "type" "f_cvt,f_cvtf2i,f_cvti2f"))
|
|
636 "ca15_issue1,ca15_cx_vfp")
|
|
637
|
|
638 (define_insn_reservation "cortex_a15_vfp_cmpd" 8
|
|
639 (and (eq_attr "tune" "cortexa15")
|
|
640 (eq_attr "type" "fcmpd"))
|
|
641 "ca15_issue2,ca15_cx_perm,ca15_cx_vfp")
|
|
642
|
|
643 (define_insn_reservation "cortex_a15_vfp_cmps" 8
|
|
644 (and (eq_attr "tune" "cortexa15")
|
|
645 (eq_attr "type" "fcmps"))
|
|
646 "ca15_issue2,ca15_cx_perm,ca15_cx_vfp")
|
|
647
|
|
648 (define_insn_reservation "cortex_a15_vfp_arithd" 7
|
|
649 (and (eq_attr "tune" "cortexa15")
|
|
650 (eq_attr "type" "ffarithd"))
|
|
651 "ca15_issue2,ca15_cx_perm*2")
|
|
652
|
|
653 (define_insn_reservation "cortex_a15_vfp_cpys" 4
|
|
654 (and (eq_attr "tune" "cortexa15")
|
|
655 (eq_attr "type" "fmov"))
|
|
656 "ca15_issue1,ca15_cx_perm")
|
|
657
|
|
658 (define_insn_reservation "cortex_a15_gp_to_vfp" 5
|
|
659 (and (eq_attr "tune" "cortexa15")
|
|
660 (eq_attr "type" "f_mcr, f_mcrr"))
|
|
661 "ca15_issue1,ca15_ls")
|
|
662
|
|
663 (define_insn_reservation "cortex_a15_mov_vfp_to_gp" 5
|
|
664 (and (eq_attr "tune" "cortexa15")
|
|
665 (eq_attr "type" "f_mrc, f_mrrc"))
|
|
666 "ca15_issue1,ca15_ls")
|
|
667
|
|
668 ;; Moves from floating point registers to general purpose registers
|
|
669 ;; induce additional latency.
|
|
670 (define_bypass 10 "cortex_a15_vfp*, cortex_a15_neon*, cortex_a15_gp_to_vfp" "cortex_a15_mov_vfp_to_gp")
|
|
671
|
|
672
|
|
673 (define_insn_reservation "cortex_a15_vfp_ariths" 7
|
|
674 (and (eq_attr "tune" "cortexa15")
|
|
675 (eq_attr "type" "ffariths"))
|
|
676 "ca15_issue1,ca15_cx_perm")
|
|
677
|
|
678 (define_insn_reservation "cortex_a15_vfp_divs" 10
|
|
679 (and (eq_attr "tune" "cortexa15")
|
|
680 (eq_attr "type" "fdivs, fsqrts"))
|
|
681 "ca15_issue1,ca15_cx_ik")
|
|
682
|
|
683 (define_insn_reservation "cortex_a15_vfp_divd" 18
|
|
684 (and (eq_attr "tune" "cortexa15")
|
|
685 (eq_attr "type" "fdivd, fsqrtd"))
|
|
686 "ca15_issue1,ca15_cx_ik")
|
|
687
|