xref: /netbsd-src/external/gpl3/gcc.old/dist/gcc/config/arm/cortex-a57.md (revision 8feb0f0b7eaff0608f8350bbfa3098827b4bb91b)
1;; ARM Cortex-A57 pipeline description
2;; Copyright (C) 2014-2020 Free Software Foundation, Inc.
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_automaton "cortex_a57")
21
22(define_attr "cortex_a57_neon_type"
23  "neon_abd, neon_abd_q, neon_arith_acc, neon_arith_acc_q,
24   neon_arith_basic, neon_arith_complex,
25   neon_reduc_add_acc, neon_multiply, neon_multiply_q,
26   neon_multiply_long, neon_mla, neon_mla_q, neon_mla_long,
27   neon_sat_mla_long, neon_shift_acc, neon_shift_imm_basic,
28   neon_shift_imm_complex,
29   neon_shift_reg_basic, neon_shift_reg_basic_q, neon_shift_reg_complex,
30   neon_shift_reg_complex_q, neon_fp_negabs, neon_fp_arith,
31   neon_fp_arith_q, neon_fp_reductions_q, neon_fp_cvt_int,
32   neon_fp_cvt_int_q, neon_fp_cvt16, neon_fp_minmax, neon_fp_mul,
33   neon_fp_mul_q, neon_fp_mla, neon_fp_mla_q, neon_fp_recpe_rsqrte,
34   neon_fp_recpe_rsqrte_q, neon_fp_recps_rsqrts, neon_fp_recps_rsqrts_q,
35   neon_bitops, neon_bitops_q, neon_from_gp,
36   neon_from_gp_q, neon_move, neon_tbl3_tbl4, neon_zip_q, neon_to_gp,
37   neon_load_a, neon_load_b, neon_load_c, neon_load_d, neon_load_e,
38   neon_load_f, neon_store_a, neon_store_b, neon_store_complex,
39   unknown"
40  (cond [
41	  (eq_attr "type" "neon_abd, neon_abd_long")
42	    (const_string "neon_abd")
43	  (eq_attr "type" "neon_abd_q")
44	    (const_string "neon_abd_q")
45	  (eq_attr "type" "neon_arith_acc, neon_reduc_add_acc,\
46			   neon_reduc_add_acc_q")
47	    (const_string "neon_arith_acc")
48	  (eq_attr "type" "neon_arith_acc_q")
49	    (const_string "neon_arith_acc_q")
50	  (eq_attr "type" "neon_add, neon_add_q, neon_add_long,\
51			   neon_add_widen, neon_neg, neon_neg_q,\
52			   neon_reduc_add, neon_reduc_add_q,\
53			   neon_reduc_add_long, neon_sub, neon_sub_q,\
54			   neon_sub_long, neon_sub_widen, neon_logic,\
55			   neon_logic_q, neon_tst, neon_tst_q")
56	    (const_string "neon_arith_basic")
57	  (eq_attr "type" "neon_abs, neon_abs_q, neon_add_halve_narrow_q,\
58			   neon_add_halve, neon_add_halve_q,\
59			   neon_sub_halve, neon_sub_halve_q, neon_qabs,\
60			   neon_qabs_q, neon_qadd, neon_qadd_q, neon_qneg,\
61			   neon_qneg_q, neon_qsub, neon_qsub_q,\
62			   neon_sub_halve_narrow_q,\
63			   neon_compare, neon_compare_q,\
64			   neon_compare_zero, neon_compare_zero_q,\
65			   neon_minmax, neon_minmax_q, neon_reduc_minmax,\
66			   neon_reduc_minmax_q")
67	    (const_string "neon_arith_complex")
68
69	  (eq_attr "type" "neon_mul_b, neon_mul_h, neon_mul_s,\
70			   neon_mul_h_scalar, neon_mul_s_scalar,\
71			   neon_sat_mul_b, neon_sat_mul_h,\
72			   neon_sat_mul_s, neon_sat_mul_h_scalar,\
73			   neon_sat_mul_s_scalar,\
74			   neon_mul_b_long, neon_mul_h_long,\
75			   neon_mul_s_long, neon_mul_d_long,\
76			   neon_mul_h_scalar_long, neon_mul_s_scalar_long,\
77			   neon_sat_mul_b_long, neon_sat_mul_h_long,\
78			   neon_sat_mul_s_long, neon_sat_mul_h_scalar_long,\
79			   neon_sat_mul_s_scalar_long, crypto_pmull")
80	    (const_string "neon_multiply")
81	  (eq_attr "type" "neon_mul_b_q, neon_mul_h_q, neon_mul_s_q,\
82			   neon_mul_h_scalar_q, neon_mul_s_scalar_q,\
83			   neon_sat_mul_b_q, neon_sat_mul_h_q,\
84			   neon_sat_mul_s_q, neon_sat_mul_h_scalar_q,\
85			   neon_sat_mul_s_scalar_q")
86	    (const_string "neon_multiply_q")
87	  (eq_attr "type" "neon_mla_b, neon_mla_h, neon_mla_s,\
88			   neon_mla_h_scalar, neon_mla_s_scalar,\
89			   neon_mla_b_long, neon_mla_h_long,\
90			   neon_mla_s_long,\
91			   neon_mla_h_scalar_long, neon_mla_s_scalar_long")
92	    (const_string "neon_mla")
93	  (eq_attr "type" "neon_mla_b_q, neon_mla_h_q, neon_mla_s_q,\
94			   neon_mla_h_scalar_q, neon_mla_s_scalar_q")
95	    (const_string "neon_mla_q")
96	  (eq_attr "type" "neon_sat_mla_b_long, neon_sat_mla_h_long,\
97			   neon_sat_mla_s_long, neon_sat_mla_h_scalar_long,\
98			   neon_sat_mla_s_scalar_long")
99	    (const_string "neon_sat_mla_long")
100
101	  (eq_attr "type" "neon_shift_acc, neon_shift_acc_q")
102	    (const_string "neon_shift_acc")
103	  (eq_attr "type" "neon_shift_imm, neon_shift_imm_q,\
104			   neon_shift_imm_narrow_q, neon_shift_imm_long")
105	    (const_string "neon_shift_imm_basic")
106	  (eq_attr "type" "neon_sat_shift_imm, neon_sat_shift_imm_q,\
107			   neon_sat_shift_imm_narrow_q")
108	    (const_string "neon_shift_imm_complex")
109	  (eq_attr "type" "neon_shift_reg")
110	    (const_string "neon_shift_reg_basic")
111	  (eq_attr "type" "neon_shift_reg_q")
112	    (const_string "neon_shift_reg_basic_q")
113	  (eq_attr "type" "neon_sat_shift_reg")
114	    (const_string "neon_shift_reg_complex")
115	  (eq_attr "type" "neon_sat_shift_reg_q")
116	    (const_string "neon_shift_reg_complex_q")
117
118	  (eq_attr "type" "neon_fp_neg_s, neon_fp_neg_s_q,\
119			   neon_fp_abs_s, neon_fp_abs_s_q,\
120			   neon_fp_neg_d, neon_fp_neg_d_q,\
121			   neon_fp_abs_d, neon_fp_abs_d_q")
122	    (const_string "neon_fp_negabs")
123	  (eq_attr "type" "neon_fp_addsub_s, neon_fp_abd_s,\
124			   neon_fp_reduc_add_s, neon_fp_compare_s,\
125			   neon_fp_minmax_s, neon_fp_round_s,\
126			   neon_fp_addsub_d, neon_fp_abd_d,\
127			   neon_fp_reduc_add_d, neon_fp_compare_d,\
128			   neon_fp_minmax_d, neon_fp_round_d,\
129			   neon_fp_reduc_minmax_s, neon_fp_reduc_minmax_d")
130	    (const_string "neon_fp_arith")
131	  (eq_attr "type" "neon_fp_addsub_s_q, neon_fp_abd_s_q,\
132			   neon_fp_reduc_add_s_q, neon_fp_compare_s_q,\
133			   neon_fp_minmax_s_q, neon_fp_round_s_q,\
134			   neon_fp_addsub_d_q, neon_fp_abd_d_q,\
135			   neon_fp_reduc_add_d_q, neon_fp_compare_d_q,\
136			   neon_fp_minmax_d_q, neon_fp_round_d_q")
137	    (const_string "neon_fp_arith_q")
138	  (eq_attr "type" "neon_fp_reduc_minmax_s_q,\
139			   neon_fp_reduc_minmax_d_q,\
140			   neon_fp_reduc_add_s_q, neon_fp_reduc_add_d_q")
141	    (const_string "neon_fp_reductions_q")
142	  (eq_attr "type" "neon_fp_to_int_s, neon_int_to_fp_s,\
143			   neon_fp_to_int_d, neon_int_to_fp_d")
144	    (const_string "neon_fp_cvt_int")
145	  (eq_attr "type" "neon_fp_to_int_s_q, neon_int_to_fp_s_q,\
146			   neon_fp_to_int_d_q, neon_int_to_fp_d_q")
147	    (const_string "neon_fp_cvt_int_q")
148	  (eq_attr "type" "neon_fp_cvt_narrow_s_q, neon_fp_cvt_widen_h")
149	    (const_string "neon_fp_cvt16")
150	  (eq_attr "type" "neon_fp_mul_s, neon_fp_mul_s_scalar,\
151			   neon_fp_mul_d")
152	    (const_string "neon_fp_mul")
153	  (eq_attr "type" "neon_fp_mul_s_q, neon_fp_mul_s_scalar_q,\
154			   neon_fp_mul_d_q, neon_fp_mul_d_scalar_q")
155	    (const_string "neon_fp_mul_q")
156	  (eq_attr "type" "neon_fp_mla_s, neon_fp_mla_s_scalar,\
157			   neon_fp_mla_d")
158	    (const_string "neon_fp_mla")
159	  (eq_attr "type" "neon_fp_mla_s_q, neon_fp_mla_s_scalar_q,
160			   neon_fp_mla_d_q, neon_fp_mla_d_scalar_q")
161	    (const_string "neon_fp_mla_q")
162	  (eq_attr "type" "neon_fp_recpe_s, neon_fp_rsqrte_s,\
163			   neon_fp_recpx_s,\
164			   neon_fp_recpe_d, neon_fp_rsqrte_d,\
165			   neon_fp_recpx_d")
166	    (const_string "neon_fp_recpe_rsqrte")
167	  (eq_attr "type" "neon_fp_recpe_s_q, neon_fp_rsqrte_s_q,\
168			   neon_fp_recpx_s_q,\
169			   neon_fp_recpe_d_q, neon_fp_rsqrte_d_q,\
170			   neon_fp_recpx_d_q")
171	    (const_string "neon_fp_recpe_rsqrte_q")
172	  (eq_attr "type" "neon_fp_recps_s, neon_fp_rsqrts_s,\
173			   neon_fp_recps_d, neon_fp_rsqrts_d")
174	    (const_string "neon_fp_recps_rsqrts")
175	  (eq_attr "type" "neon_fp_recps_s_q, neon_fp_rsqrts_s_q,\
176			   neon_fp_recps_d_q, neon_fp_rsqrts_d_q")
177	    (const_string "neon_fp_recps_rsqrts_q")
178	  (eq_attr "type" "neon_bsl, neon_cls, neon_cnt,\
179			   neon_rev, neon_permute, neon_rbit,\
180			   neon_tbl1, neon_tbl2, neon_zip,\
181			   neon_dup, neon_dup_q, neon_ext, neon_ext_q,\
182			   neon_move, neon_move_q, neon_move_narrow_q")
183	    (const_string "neon_bitops")
184	  (eq_attr "type" "neon_bsl_q, neon_cls_q, neon_cnt_q,\
185			   neon_rev_q, neon_permute_q, neon_rbit_q")
186	    (const_string "neon_bitops_q")
187	  (eq_attr "type" "neon_from_gp,f_mcr,f_mcrr")
188	    (const_string "neon_from_gp")
189	  (eq_attr "type" "neon_from_gp_q")
190	    (const_string "neon_from_gp_q")
191	  (eq_attr "type" "neon_tbl3, neon_tbl4")
192	    (const_string "neon_tbl3_tbl4")
193	  (eq_attr "type" "neon_zip_q")
194	    (const_string "neon_zip_q")
195	  (eq_attr "type" "neon_to_gp, neon_to_gp_q,f_mrc,f_mrrc")
196	    (const_string "neon_to_gp")
197
198	  (eq_attr "type" "f_loads, f_loadd,\
199			   neon_load1_1reg, neon_load1_1reg_q,\
200			   neon_load1_2reg, neon_load1_2reg_q")
201	    (const_string "neon_load_a")
202	  (eq_attr "type" "neon_load1_3reg, neon_load1_3reg_q,\
203			   neon_load1_4reg, neon_load1_4reg_q")
204	    (const_string "neon_load_b")
205	  (eq_attr "type" "neon_ldp, neon_ldp_q,\
206			   neon_load1_one_lane, neon_load1_one_lane_q,\
207			   neon_load1_all_lanes, neon_load1_all_lanes_q,\
208			   neon_load2_2reg, neon_load2_2reg_q,\
209			   neon_load2_all_lanes, neon_load2_all_lanes_q")
210	    (const_string "neon_load_c")
211	  (eq_attr "type" "neon_load2_4reg, neon_load2_4reg_q,\
212			   neon_load3_3reg, neon_load3_3reg_q,\
213			   neon_load3_one_lane, neon_load3_one_lane_q,\
214			   neon_load4_4reg, neon_load4_4reg_q")
215	    (const_string "neon_load_d")
216	  (eq_attr "type" "neon_load2_one_lane, neon_load2_one_lane_q,\
217			   neon_load3_all_lanes, neon_load3_all_lanes_q,\
218			   neon_load4_all_lanes, neon_load4_all_lanes_q")
219	    (const_string "neon_load_e")
220	  (eq_attr "type" "neon_load4_one_lane, neon_load4_one_lane_q")
221	    (const_string "neon_load_f")
222
223	  (eq_attr "type" "f_stores, f_stored,\
224			   neon_store1_1reg")
225	    (const_string "neon_store_a")
226	  (eq_attr "type" "neon_store1_2reg, neon_store1_1reg_q")
227	    (const_string "neon_store_b")
228	  (eq_attr "type" "neon_stp, neon_stp_q,\
229			   neon_store1_3reg, neon_store1_3reg_q,\
230			   neon_store3_3reg, neon_store3_3reg_q,\
231			   neon_store2_4reg, neon_store2_4reg_q,\
232			   neon_store4_4reg, neon_store4_4reg_q,\
233			   neon_store2_2reg, neon_store2_2reg_q,\
234			   neon_store3_one_lane, neon_store3_one_lane_q,\
235			   neon_store4_one_lane, neon_store4_one_lane_q,\
236			   neon_store1_4reg, neon_store1_4reg_q,\
237			   neon_store1_one_lane, neon_store1_one_lane_q,\
238			   neon_store2_one_lane, neon_store2_one_lane_q")
239	    (const_string "neon_store_complex")
240;; If it doesn't match any of the above that we want to treat specially but is
241;; still a NEON type, treat it as a basic NEON type.  This is better than
242;; dropping it on the floor and making no assumptions about it whatsoever.
243	  (eq_attr "is_neon_type" "yes")
244	    (const_string "neon_arith_basic")]
245	  (const_string "unknown")))
246
247;; The Cortex-A57 core is modelled as a triple issue pipeline that has
248;; the following functional units.
249;; 1.  Two pipelines for integer operations: SX1, SX2
250
251(define_cpu_unit "ca57_sx1_issue" "cortex_a57")
252(define_reservation "ca57_sx1" "ca57_sx1_issue")
253
254(define_cpu_unit "ca57_sx2_issue" "cortex_a57")
255(define_reservation "ca57_sx2" "ca57_sx2_issue")
256
257;; 2.  One pipeline for complex integer operations: MX
258
259(define_cpu_unit "ca57_mx_issue"
260		 "cortex_a57")
261(define_reservation "ca57_mx" "ca57_mx_issue")
262(define_reservation "ca57_mx_block" "ca57_mx_issue")
263
264;; 3.  Two asymmetric pipelines for Neon and FP operations: CX1, CX2
265(define_automaton "cortex_a57_cx")
266
267(define_cpu_unit "ca57_cx1_issue"
268		 "cortex_a57_cx")
269(define_cpu_unit "ca57_cx2_issue"
270		 "cortex_a57_cx")
271
272(define_reservation "ca57_cx1" "ca57_cx1_issue")
273
274(define_reservation "ca57_cx2" "ca57_cx2_issue")
275(define_reservation "ca57_cx2_block" "ca57_cx2_issue*2")
276
277;; 4.  One pipeline for branch operations: BX
278
279(define_cpu_unit "ca57_bx_issue" "cortex_a57")
280(define_reservation "ca57_bx" "ca57_bx_issue")
281
282;; 5.  Two pipelines for load and store operations: LS1, LS2.  The most
283;;     valuable thing we can do is force a structural hazard to split
284;;     up loads/stores.
285
286(define_cpu_unit "ca57_ls_issue" "cortex_a57")
287(define_cpu_unit "ca57_ldr, ca57_str" "cortex_a57")
288(define_reservation "ca57_load_model" "ca57_ls_issue,ca57_ldr*2")
289(define_reservation "ca57_store_model" "ca57_ls_issue,ca57_str")
290
291;; Block all issue queues.
292
293(define_reservation "ca57_block" "ca57_cx1_issue + ca57_cx2_issue
294				  + ca57_mx_issue + ca57_sx1_issue
295				  + ca57_sx2_issue + ca57_ls_issue")
296
297;; Simple Execution Unit:
298;;
299;; Simple ALU without shift
300(define_insn_reservation "cortex_a57_alu" 2
301  (and (eq_attr "tune" "cortexa57")
302       (eq_attr "type" "alu_imm,alus_imm,logic_imm,logics_imm,\
303			alu_sreg,alus_sreg,logic_reg,logics_reg,\
304			adc_imm,adcs_imm,adc_reg,adcs_reg,\
305			adr,bfx,extend,clz,rbit,rev,alu_dsp_reg,\
306			rotate_imm,shift_imm,shift_reg,\
307			mov_imm,mov_reg,\
308			mvn_imm,mvn_reg,\
309			mrs,multiple"))
310  "ca57_sx1|ca57_sx2")
311
312;; ALU ops with immediate shift
313(define_insn_reservation "cortex_a57_alu_shift" 3
314  (and (eq_attr "tune" "cortexa57")
315       (eq_attr "type" "bfm,\
316			alu_shift_imm,alus_shift_imm,\
317			crc,logic_shift_imm,logics_shift_imm,\
318			mov_shift,mvn_shift"))
319  "ca57_mx")
320
321;; Multi-Cycle Execution Unit:
322;;
323;; ALU ops with register controlled shift
324(define_insn_reservation "cortex_a57_alu_shift_reg" 3
325  (and (eq_attr "tune" "cortexa57")
326       (eq_attr "type" "alu_shift_reg,alus_shift_reg,\
327			logic_shift_reg,logics_shift_reg,\
328			mov_shift_reg,mvn_shift_reg"))
329   "ca57_mx")
330
331;; All multiplies
332;; TODO: AArch32 and AArch64 have different behavior
333(define_insn_reservation "cortex_a57_mult32" 3
334  (and (eq_attr "tune" "cortexa57")
335       (ior (eq_attr "mul32" "yes")
336	    (eq_attr "widen_mul64" "yes")))
337  "ca57_mx")
338
339;; Integer divide
340(define_insn_reservation "cortex_a57_div" 10
341  (and (eq_attr "tune" "cortexa57")
342       (eq_attr "type" "udiv,sdiv"))
343  "ca57_mx_issue,ca57_mx_block*3")
344
345;; Block all issue pipes for a cycle
346(define_insn_reservation "cortex_a57_block" 1
347  (and (eq_attr "tune" "cortexa57")
348       (eq_attr "type" "block"))
349  "ca57_block")
350
351;; Branch execution Unit
352;;
353;; Branches take one issue slot.
354;; No latency as there is no result
355(define_insn_reservation "cortex_a57_branch" 0
356  (and (eq_attr "tune" "cortexa57")
357       (eq_attr "type" "branch"))
358  "ca57_bx")
359
360;; Load-store execution Unit
361;;
362;; Loads of up to two words.
363(define_insn_reservation "cortex_a57_load1" 5
364  (and (eq_attr "tune" "cortexa57")
365       (eq_attr "type" "load_byte,load_4,load_8"))
366  "ca57_load_model")
367
368;; Loads of three or four words.
369(define_insn_reservation "cortex_a57_load3" 5
370  (and (eq_attr "tune" "cortexa57")
371       (eq_attr "type" "load_12,load_16"))
372  "ca57_ls_issue*2,ca57_load_model")
373
374;; Stores of up to two words.
375(define_insn_reservation "cortex_a57_store1" 0
376  (and (eq_attr "tune" "cortexa57")
377       (eq_attr "type" "store_4,store_8"))
378  "ca57_store_model")
379
380;; Stores of three or four words.
381(define_insn_reservation "cortex_a57_store3" 0
382  (and (eq_attr "tune" "cortexa57")
383       (eq_attr "type" "store_12,store_16"))
384  "ca57_ls_issue*2,ca57_store_model")
385
386;; Advanced SIMD Unit - Integer Arithmetic Instructions.
387
388(define_insn_reservation  "cortex_a57_neon_abd" 5
389  (and (eq_attr "tune" "cortexa57")
390       (eq_attr "cortex_a57_neon_type" "neon_abd"))
391  "ca57_cx1|ca57_cx2")
392
393(define_insn_reservation  "cortex_a57_neon_abd_q" 5
394  (and (eq_attr "tune" "cortexa57")
395       (eq_attr "cortex_a57_neon_type" "neon_abd_q"))
396  "ca57_cx1+ca57_cx2")
397
398(define_insn_reservation  "cortex_a57_neon_aba" 7
399  (and (eq_attr "tune" "cortexa57")
400       (eq_attr "cortex_a57_neon_type" "neon_arith_acc"))
401  "ca57_cx2")
402
403(define_insn_reservation  "cortex_a57_neon_aba_q" 8
404  (and (eq_attr "tune" "cortexa57")
405       (eq_attr "cortex_a57_neon_type" "neon_arith_acc_q"))
406  "ca57_cx2+(ca57_cx2_issue,ca57_cx2)")
407
408(define_insn_reservation  "cortex_a57_neon_arith_basic" 4
409  (and (eq_attr "tune" "cortexa57")
410       (eq_attr "cortex_a57_neon_type" "neon_arith_basic"))
411  "ca57_cx1|ca57_cx2")
412
413(define_insn_reservation  "cortex_a57_neon_arith_complex" 5
414  (and (eq_attr "tune" "cortexa57")
415       (eq_attr "cortex_a57_neon_type" "neon_arith_complex"))
416  "ca57_cx1|ca57_cx2")
417
418;; Integer Multiply Instructions.
419
420(define_insn_reservation "cortex_a57_neon_multiply" 6
421  (and (eq_attr "tune" "cortexa57")
422       (eq_attr "cortex_a57_neon_type" "neon_multiply"))
423  "ca57_cx1")
424
425(define_insn_reservation "cortex_a57_neon_multiply_q" 7
426  (and (eq_attr "tune" "cortexa57")
427       (eq_attr "cortex_a57_neon_type" "neon_multiply_q"))
428  "ca57_cx1+(ca57_cx1_issue,ca57_cx1)")
429
430(define_insn_reservation "cortex_a57_neon_mla" 6
431  (and (eq_attr "tune" "cortexa57")
432       (eq_attr "cortex_a57_neon_type" "neon_mla"))
433  "ca57_cx1")
434
435(define_insn_reservation "cortex_a57_neon_mla_q" 7
436  (and (eq_attr "tune" "cortexa57")
437       (eq_attr "cortex_a57_neon_type" "neon_mla_q"))
438  "ca57_cx1+(ca57_cx1_issue,ca57_cx1)")
439
440(define_insn_reservation "cortex_a57_neon_sat_mla_long" 6
441  (and (eq_attr "tune" "cortexa57")
442       (eq_attr "cortex_a57_neon_type" "neon_sat_mla_long"))
443  "ca57_cx1")
444
445;; Integer Shift Instructions.
446
447(define_insn_reservation
448  "cortex_a57_neon_shift_acc" 7
449  (and (eq_attr "tune" "cortexa57")
450       (eq_attr "cortex_a57_neon_type" "neon_shift_acc"))
451  "ca57_cx2")
452
453(define_insn_reservation
454  "cortex_a57_neon_shift_imm_basic" 4
455  (and (eq_attr "tune" "cortexa57")
456       (eq_attr "cortex_a57_neon_type" "neon_shift_imm_basic"))
457  "ca57_cx2")
458
459(define_insn_reservation
460  "cortex_a57_neon_shift_imm_complex" 5
461  (and (eq_attr "tune" "cortexa57")
462       (eq_attr "cortex_a57_neon_type" "neon_shift_imm_complex"))
463  "ca57_cx2")
464
465(define_insn_reservation
466  "cortex_a57_neon_shift_reg_basic" 4
467  (and (eq_attr "tune" "cortexa57")
468       (eq_attr "cortex_a57_neon_type" "neon_shift_reg_basic"))
469  "ca57_cx2")
470
471(define_insn_reservation
472  "cortex_a57_neon_shift_reg_basic_q" 5
473  (and (eq_attr "tune" "cortexa57")
474       (eq_attr "cortex_a57_neon_type" "neon_shift_reg_basic_q"))
475  "ca57_cx2+(ca57_cx2_issue,ca57_cx2)")
476
477(define_insn_reservation
478  "cortex_a57_neon_shift_reg_complex" 5
479  (and (eq_attr "tune" "cortexa57")
480       (eq_attr "cortex_a57_neon_type" "neon_shift_reg_complex"))
481  "ca57_cx2")
482
483(define_insn_reservation
484  "cortex_a57_neon_shift_reg_complex_q" 6
485  (and (eq_attr "tune" "cortexa57")
486       (eq_attr "cortex_a57_neon_type" "neon_shift_reg_complex_q"))
487  "ca57_cx2+(ca57_cx2_issue,ca57_cx2)")
488
489;; Floating Point Instructions.
490
491(define_insn_reservation
492  "cortex_a57_neon_fp_negabs" 4
493  (and (eq_attr "tune" "cortexa57")
494       (eq_attr "cortex_a57_neon_type" "neon_fp_negabs"))
495  "(ca57_cx1|ca57_cx2)")
496
497(define_insn_reservation
498  "cortex_a57_neon_fp_arith" 6
499  (and (eq_attr "tune" "cortexa57")
500       (eq_attr "cortex_a57_neon_type" "neon_fp_arith"))
501  "(ca57_cx1|ca57_cx2)")
502
503(define_insn_reservation
504  "cortex_a57_neon_fp_arith_q" 6
505  (and (eq_attr "tune" "cortexa57")
506       (eq_attr "cortex_a57_neon_type" "neon_fp_arith_q"))
507  "(ca57_cx1+ca57_cx2)")
508
509(define_insn_reservation
510  "cortex_a57_neon_fp_reductions_q" 10
511  (and (eq_attr "tune" "cortexa57")
512       (eq_attr "cortex_a57_neon_type" "neon_fp_reductions_q"))
513  "(ca57_cx1+ca57_cx2),(ca57_cx1|ca57_cx2)")
514
515(define_insn_reservation
516  "cortex_a57_neon_fp_cvt_int" 6
517  (and (eq_attr "tune" "cortexa57")
518       (eq_attr "cortex_a57_neon_type" "neon_fp_cvt_int"))
519  "(ca57_cx1|ca57_cx2)")
520
521(define_insn_reservation
522  "cortex_a57_neon_fp_cvt_int_q" 6
523  (and (eq_attr "tune" "cortexa57")
524       (eq_attr "cortex_a57_neon_type" "neon_fp_cvt_int_q"))
525  "(ca57_cx1+ca57_cx2)")
526
527(define_insn_reservation
528  "cortex_a57_neon_fp_cvt16" 10
529  (and (eq_attr "tune" "cortexa57")
530       (eq_attr "cortex_a57_neon_type" "neon_fp_cvt16"))
531  "(ca57_cx1_issue+ca57_cx2_issue),(ca57_cx1|ca57_cx2)")
532
533(define_insn_reservation
534  "cortex_a57_neon_fp_mul" 5
535  (and (eq_attr "tune" "cortexa57")
536       (eq_attr "cortex_a57_neon_type" "neon_fp_mul"))
537  "(ca57_cx1|ca57_cx2)")
538
539(define_insn_reservation
540  "cortex_a57_neon_fp_mul_q" 5
541  (and (eq_attr "tune" "cortexa57")
542       (eq_attr "cortex_a57_neon_type" "neon_fp_mul_q"))
543  "(ca57_cx1+ca57_cx2)")
544
545(define_insn_reservation
546  "cortex_a57_neon_fp_mla" 9
547  (and (eq_attr "tune" "cortexa57")
548       (eq_attr "cortex_a57_neon_type" "neon_fp_mla"))
549  "(ca57_cx1,ca57_cx1)|(ca57_cx2,ca57_cx2)")
550
551(define_insn_reservation
552  "cortex_a57_neon_fp_mla_q" 9
553  (and (eq_attr "tune" "cortexa57")
554       (eq_attr "cortex_a57_neon_type" "neon_fp_mla_q"))
555  "(ca57_cx1+ca57_cx2),(ca57_cx1,ca57_cx2)")
556
557(define_insn_reservation
558  "cortex_a57_neon_fp_recpe_rsqrte" 6
559  (and (eq_attr "tune" "cortexa57")
560       (eq_attr "cortex_a57_neon_type" "neon_fp_recpe_rsqrte"))
561  "(ca57_cx1|ca57_cx2)")
562
563(define_insn_reservation
564  "cortex_a57_neon_fp_recpe_rsqrte_q" 6
565  (and (eq_attr "tune" "cortexa57")
566       (eq_attr "cortex_a57_neon_type" "neon_fp_recpe_rsqrte_q"))
567  "(ca57_cx1+ca57_cx2)")
568
569(define_insn_reservation
570  "cortex_a57_neon_fp_recps_rsqrts" 10
571  (and (eq_attr "tune" "cortexa57")
572       (eq_attr "cortex_a57_neon_type" "neon_fp_recps_rsqrts"))
573  "(ca57_cx1|ca57_cx2)")
574
575(define_insn_reservation
576  "cortex_a57_neon_fp_recps_rsqrts_q" 10
577  (and (eq_attr "tune" "cortexa57")
578       (eq_attr "cortex_a57_neon_type" "neon_fp_recps_rsqrts_q"))
579  "(ca57_cx1+ca57_cx2)")
580
581;; Miscellaneous Instructions.
582
583(define_insn_reservation
584  "cortex_a57_neon_bitops" 4
585  (and (eq_attr "tune" "cortexa57")
586       (eq_attr "cortex_a57_neon_type" "neon_bitops"))
587  "(ca57_cx1|ca57_cx2)")
588
589(define_insn_reservation
590  "cortex_a57_neon_bitops_q" 4
591  (and (eq_attr "tune" "cortexa57")
592       (eq_attr "cortex_a57_neon_type" "neon_bitops_q"))
593  "(ca57_cx1+ca57_cx2)")
594
595(define_insn_reservation
596  "cortex_a57_neon_from_gp" 9
597  (and (eq_attr "tune" "cortexa57")
598       (eq_attr "cortex_a57_neon_type" "neon_from_gp"))
599  "(ca57_ls_issue+ca57_cx1_issue,ca57_cx1)
600	       |(ca57_ls_issue+ca57_cx2_issue,ca57_cx2)")
601
602(define_insn_reservation
603  "cortex_a57_neon_from_gp_q" 9
604  (and (eq_attr "tune" "cortexa57")
605       (eq_attr "cortex_a57_neon_type" "neon_from_gp_q"))
606  "(ca57_ls_issue+ca57_cx1_issue,ca57_cx1)
607	       +(ca57_ls_issue+ca57_cx2_issue,ca57_cx2)")
608
609(define_insn_reservation
610  "cortex_a57_neon_tbl3_tbl4" 7
611  (and (eq_attr "tune" "cortexa57")
612       (eq_attr "cortex_a57_neon_type" "neon_tbl3_tbl4"))
613  "(ca57_cx1_issue,ca57_cx1)
614	       +(ca57_cx2_issue,ca57_cx2)")
615
616(define_insn_reservation
617  "cortex_a57_neon_zip_q" 7
618  (and (eq_attr "tune" "cortexa57")
619       (eq_attr "cortex_a57_neon_type" "neon_zip_q"))
620  "(ca57_cx1_issue,ca57_cx1)
621	       +(ca57_cx2_issue,ca57_cx2)")
622
623(define_insn_reservation
624  "cortex_a57_neon_to_gp" 7
625  (and (eq_attr "tune" "cortexa57")
626       (eq_attr "cortex_a57_neon_type" "neon_to_gp"))
627  "((ca57_ls_issue+ca57_sx1_issue),ca57_sx1)
628   |((ca57_ls_issue+ca57_sx2_issue),ca57_sx2)")
629
630;; Load Instructions.
631
632(define_insn_reservation
633  "cortex_a57_neon_load_a" 6
634  (and (eq_attr "tune" "cortexa57")
635       (eq_attr "cortex_a57_neon_type" "neon_load_a"))
636  "ca57_load_model")
637
638(define_insn_reservation
639  "cortex_a57_neon_load_b" 7
640  (and (eq_attr "tune" "cortexa57")
641       (eq_attr "cortex_a57_neon_type" "neon_load_b"))
642  "ca57_ls_issue,ca57_ls_issue+ca57_ldr,ca57_ldr*2")
643
644(define_insn_reservation
645  "cortex_a57_neon_load_c" 9
646  (and (eq_attr "tune" "cortexa57")
647       (eq_attr "cortex_a57_neon_type" "neon_load_c"))
648  "ca57_load_model+(ca57_cx1|ca57_cx2)")
649
650(define_insn_reservation
651  "cortex_a57_neon_load_d" 11
652  (and (eq_attr "tune" "cortexa57")
653       (eq_attr "cortex_a57_neon_type" "neon_load_d"))
654  "ca57_cx1_issue+ca57_cx2_issue,
655   ca57_ls_issue+ca57_ls_issue,ca57_ldr*2")
656
657(define_insn_reservation
658  "cortex_a57_neon_load_e" 9
659  (and (eq_attr "tune" "cortexa57")
660       (eq_attr "cortex_a57_neon_type" "neon_load_e"))
661  "ca57_load_model+(ca57_cx1|ca57_cx2)")
662
663(define_insn_reservation
664  "cortex_a57_neon_load_f" 11
665  (and (eq_attr "tune" "cortexa57")
666       (eq_attr "cortex_a57_neon_type" "neon_load_f"))
667  "ca57_cx1_issue+ca57_cx2_issue,
668   ca57_ls_issue+ca57_ls_issue,ca57_ldr*2")
669
670;; Store Instructions.
671
672(define_insn_reservation
673  "cortex_a57_neon_store_a" 0
674  (and (eq_attr "tune" "cortexa57")
675       (eq_attr "cortex_a57_neon_type" "neon_store_a"))
676  "ca57_store_model")
677
678(define_insn_reservation
679  "cortex_a57_neon_store_b" 0
680  (and (eq_attr "tune" "cortexa57")
681       (eq_attr "cortex_a57_neon_type" "neon_store_b"))
682  "ca57_store_model")
683
684;; These block issue for a number of cycles proportional to the number
685;; of 64-bit chunks they will store, we don't attempt to model that
686;; precisely, treat them as blocking execution for two cycles when
687;; issued.
688(define_insn_reservation
689  "cortex_a57_neon_store_complex" 0
690  (and (eq_attr "tune" "cortexa57")
691       (eq_attr "cortex_a57_neon_type" "neon_store_complex"))
692  "ca57_block*2")
693
694;; Floating-Point Operations.
695
696(define_insn_reservation "cortex_a57_fp_const" 4
697  (and (eq_attr "tune" "cortexa57")
698       (eq_attr "type" "fconsts,fconstd"))
699  "(ca57_cx1|ca57_cx2)")
700
701(define_insn_reservation "cortex_a57_fp_add_sub" 6
702  (and (eq_attr "tune" "cortexa57")
703       (eq_attr "type" "fadds,faddd"))
704  "(ca57_cx1|ca57_cx2)")
705
706(define_insn_reservation "cortex_a57_fp_mul" 6
707  (and (eq_attr "tune" "cortexa57")
708       (eq_attr "type" "fmuls,fmuld"))
709  "(ca57_cx1|ca57_cx2)")
710
711(define_insn_reservation "cortex_a57_fp_mac" 10
712  (and (eq_attr "tune" "cortexa57")
713       (eq_attr "type" "fmacs,ffmas,fmacd,ffmad"))
714  "(ca57_cx1,nothing,nothing,ca57_cx1) \
715   |(ca57_cx2,nothing,nothing,ca57_cx2)")
716
717(define_insn_reservation "cortex_a57_fp_cvt" 6
718  (and (eq_attr "tune" "cortexa57")
719       (eq_attr "type" "f_cvt,f_cvtf2i,f_cvti2f"))
720  "(ca57_cx1|ca57_cx2)")
721
722(define_insn_reservation "cortex_a57_fp_cmp" 7
723  (and (eq_attr "tune" "cortexa57")
724       (eq_attr "type" "fcmps,fcmpd,fccmps,fccmpd"))
725  "ca57_cx2")
726
727(define_insn_reservation "cortex_a57_fp_arith" 4
728  (and (eq_attr "tune" "cortexa57")
729       (eq_attr "type" "ffariths,ffarithd"))
730  "(ca57_cx1|ca57_cx2)")
731
732(define_insn_reservation "cortex_a57_fp_cpys" 4
733  (and (eq_attr "tune" "cortexa57")
734       (eq_attr "type" "fmov,fcsel"))
735  "(ca57_cx1|ca57_cx2)")
736
737(define_insn_reservation "cortex_a57_fp_divs" 12
738  (and (eq_attr "tune" "cortexa57")
739       (eq_attr "type" "fdivs, fsqrts,\
740			neon_fp_div_s, neon_fp_sqrt_s"))
741  "ca57_cx2_block*5")
742
743(define_insn_reservation "cortex_a57_fp_divd" 16
744  (and (eq_attr "tune" "cortexa57")
745       (eq_attr "type" "fdivd, fsqrtd, neon_fp_div_d, neon_fp_sqrt_d"))
746  "ca57_cx2_block*3")
747
748(define_insn_reservation "cortex_a57_neon_fp_div_q" 20
749  (and (eq_attr "tune" "cortexa57")
750       (eq_attr "type" "fdivd, fsqrtd,\
751			 neon_fp_div_s_q, neon_fp_div_d_q,\
752			 neon_fp_sqrt_s_q, neon_fp_sqrt_d_q"))
753  "ca57_cx2_block*3")
754
755(define_insn_reservation "cortex_a57_crypto_simple" 3
756  (and (eq_attr "tune" "cortexa57")
757       (eq_attr "type" "crypto_aese,crypto_aesmc,crypto_sha1_fast,crypto_sha256_fast"))
758  "ca57_cx1")
759
760(define_insn_reservation "cortex_a57_crypto_complex" 6
761  (and (eq_attr "tune" "cortexa57")
762       (eq_attr "type" "crypto_sha1_slow,crypto_sha256_slow"))
763  "ca57_cx1*2")
764
765(define_insn_reservation "cortex_a57_crypto_xor" 6
766  (and (eq_attr "tune" "cortexa57")
767       (eq_attr "type" "crypto_sha1_xor"))
768  "(ca57_cx1*2)|(ca57_cx2*2)")
769
770;; We lie with calls.  They take up all issue slots, but are otherwise
771;; not harmful.
772(define_insn_reservation "cortex_a57_call" 1
773  (and (eq_attr "tune" "cortexa57")
774       (eq_attr "type" "call"))
775  "ca57_sx1_issue+ca57_sx2_issue+ca57_cx1_issue+ca57_cx2_issue\
776    +ca57_mx_issue+ca57_bx_issue+ca57_ls_issue"
777)
778
779;; Simple execution unit bypasses
780(define_bypass 1 "cortex_a57_alu"
781	         "cortex_a57_alu,cortex_a57_alu_shift,cortex_a57_alu_shift_reg")
782(define_bypass 2 "cortex_a57_alu_shift"
783	         "cortex_a57_alu,cortex_a57_alu_shift,cortex_a57_alu_shift_reg")
784(define_bypass 2 "cortex_a57_alu_shift_reg"
785	         "cortex_a57_alu,cortex_a57_alu_shift,cortex_a57_alu_shift_reg")
786(define_bypass 1 "cortex_a57_alu" "cortex_a57_load1,cortex_a57_load3")
787(define_bypass 2 "cortex_a57_alu_shift" "cortex_a57_load1,cortex_a57_load3")
788(define_bypass 2 "cortex_a57_alu_shift_reg"
789	         "cortex_a57_load1,cortex_a57_load3")
790
791;; An MLA or a MUL can feed a dependent MLA.
792(define_bypass 5 "cortex_a57_neon_*mla*,cortex_a57_neon_*mul*"
793		 "cortex_a57_neon_*mla*")
794
795(define_bypass 5 "cortex_a57_fp_mul,cortex_a57_fp_mac"
796		 "cortex_a57_fp_mac")
797
798;; We don't need to care about control hazards, either the branch is
799;; predicted in which case we pay no penalty, or the branch is
800;; mispredicted in which case instruction scheduling will be unlikely to
801;; help.
802(define_bypass 1 "cortex_a57_*"
803		 "cortex_a57_call,cortex_a57_branch")
804