111
|
1 ;; Marvell ARM Processor Pipeline Description
|
131
|
2 ;; Copyright (C) 2010-2018 Free Software Foundation, Inc.
|
111
|
3 ;; Contributed by Marvell.
|
|
4
|
|
5 ;; This file is part of GCC.
|
|
6
|
|
7 ;; GCC is free software; you can redistribute it and/or modify it
|
|
8 ;; under the terms of the GNU General Public License as published
|
|
9 ;; by the Free Software Foundation; either version 3, or (at your
|
|
10 ;; option) any later version.
|
|
11
|
|
12 ;; GCC is distributed in the hope that it will be useful, but WITHOUT
|
|
13 ;; ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
14 ;; or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
15 ;; License for more details.
|
|
16
|
|
17 ;; You should have received a copy of the GNU General Public License
|
|
18 ;; along with GCC; see the file COPYING3. If not see
|
|
19 ;; <http://www.gnu.org/licenses/>.
|
|
20
|
|
21 ;; Pipeline description for the Marvell PJ4, aka "Flareon".
|
|
22 (define_automaton "pj4")
|
|
23
|
|
24 ;; Issue resources
|
|
25 (define_cpu_unit "pj4_is1,pj4_is2" "pj4")
|
|
26 (define_reservation "pj4_is" "(pj4_is1|pj4_is2)")
|
|
27 (define_reservation "pj4_isb" "(pj4_is1+pj4_is2)")
|
|
28
|
|
29 ;; Functional units
|
|
30 (define_cpu_unit "pj4_alu1,pj4_alu2,pj4_mul,pj4_div" "pj4")
|
|
31
|
|
32 ;; Completion ports
|
|
33 (define_cpu_unit "pj4_w1,pj4_w2" "pj4")
|
|
34
|
|
35 ;; Complete/Retire control
|
|
36 (define_cpu_unit "pj4_c1,pj4_c2" "pj4")
|
|
37 (define_reservation "pj4_cp" "(pj4_c1|pj4_c2)")
|
|
38 (define_reservation "pj4_cpb" "(pj4_c1+pj4_c2)")
|
|
39
|
|
40 ;; Integer arithmetic instructions
|
|
41
|
|
42 (define_insn_reservation "pj4_alu_e1" 1
|
|
43 (and (eq_attr "tune" "marvell_pj4")
|
|
44 (eq_attr "type" "mov_imm,mov_reg,mvn_imm,mvn_reg")
|
|
45 (not (eq_attr "conds" "set")))
|
|
46 "pj4_is,(pj4_alu1,pj4_w1+pj4_cp)|(pj4_alu2,pj4_w2+pj4_cp)")
|
|
47
|
|
48 (define_insn_reservation "pj4_alu_e1_conds" 4
|
|
49 (and (eq_attr "tune" "marvell_pj4")
|
|
50 (eq_attr "type" "mov_imm,mov_reg,mvn_imm,mvn_reg")
|
|
51 (eq_attr "conds" "set"))
|
|
52 "pj4_is,(pj4_alu1,pj4_w1+pj4_cp)|(pj4_alu2,pj4_w2+pj4_cp)")
|
|
53
|
|
54 (define_insn_reservation "pj4_alu" 1
|
|
55 (and (eq_attr "tune" "marvell_pj4")
|
|
56 (eq_attr "type" "alu_imm,alus_imm,alu_sreg,alus_sreg,\
|
|
57 logic_imm,logics_imm,logic_reg,logics_reg,\
|
|
58 adc_imm,adcs_imm,adc_reg,adcs_reg,\
|
|
59 adr,bfm,rev,alu_dsp_reg,\
|
|
60 shift_imm,shift_reg")
|
|
61 (not (eq_attr "conds" "set")))
|
|
62 "pj4_is,(pj4_alu1,pj4_w1+pj4_cp)|(pj4_alu2,pj4_w2+pj4_cp)")
|
|
63
|
|
64 (define_insn_reservation "pj4_alu_conds" 4
|
|
65 (and (eq_attr "tune" "marvell_pj4")
|
|
66 (eq_attr "type" "alu_imm,alus_imm,alu_sreg,alus_sreg,\
|
|
67 logic_imm,logics_imm,logic_reg,logics_reg,\
|
|
68 adc_imm,adcs_imm,adc_reg,adcs_reg,\
|
|
69 adr,bfm,rev,alu_dsp_reg,\
|
|
70 shift_imm,shift_reg")
|
|
71 (eq_attr "conds" "set"))
|
|
72 "pj4_is,(pj4_alu1,pj4_w1+pj4_cp)|(pj4_alu2,pj4_w2+pj4_cp)")
|
|
73
|
|
74 (define_insn_reservation "pj4_shift" 1
|
|
75 (and (eq_attr "tune" "marvell_pj4")
|
|
76 (eq_attr "type" "alu_shift_imm,logic_shift_imm,\
|
|
77 alus_shift_imm,logics_shift_imm,\
|
|
78 alu_shift_reg,logic_shift_reg,\
|
|
79 alus_shift_reg,logics_shift_reg,\
|
|
80 extend,\
|
|
81 mov_shift,mvn_shift,mov_shift_reg,mvn_shift_reg")
|
|
82 (not (eq_attr "conds" "set"))
|
|
83 (eq_attr "shift" "1")) "pj4_is,(pj4_alu1,pj4_w1+pj4_cp)|(pj4_alu2,pj4_w2+pj4_cp)")
|
|
84
|
|
85 (define_insn_reservation "pj4_shift_conds" 4
|
|
86 (and (eq_attr "tune" "marvell_pj4")
|
|
87 (eq_attr "type" "alu_shift_imm,logic_shift_imm,\
|
|
88 alus_shift_imm,logics_shift_imm,\
|
|
89 alu_shift_reg,logic_shift_reg,\
|
|
90 alus_shift_reg,logics_shift_reg,\
|
|
91 extend,\
|
|
92 mov_shift,mvn_shift,mov_shift_reg,mvn_shift_reg")
|
|
93 (eq_attr "conds" "set")
|
|
94 (eq_attr "shift" "1")) "pj4_is,(pj4_alu1,pj4_w1+pj4_cp)|(pj4_alu2,pj4_w2+pj4_cp)")
|
|
95
|
|
96 (define_insn_reservation "pj4_alu_shift" 1
|
|
97 (and (eq_attr "tune" "marvell_pj4")
|
|
98 (not (eq_attr "conds" "set"))
|
|
99 (eq_attr "type" "alu_shift_imm,logic_shift_imm,\
|
|
100 alus_shift_imm,logics_shift_imm,\
|
|
101 alu_shift_reg,logic_shift_reg,\
|
|
102 alus_shift_reg,logics_shift_reg,\
|
|
103 extend,\
|
|
104 mov_shift,mvn_shift,mov_shift_reg,mvn_shift_reg"))
|
|
105 "pj4_is,(pj4_alu1,nothing,pj4_w1+pj4_cp)|(pj4_alu2,nothing,pj4_w2+pj4_cp)")
|
|
106
|
|
107 (define_insn_reservation "pj4_alu_shift_conds" 4
|
|
108 (and (eq_attr "tune" "marvell_pj4")
|
|
109 (eq_attr "conds" "set")
|
|
110 (eq_attr "type" "alu_shift_imm,logic_shift_imm,alus_shift_imm,logics_shift_imm,\
|
|
111 alu_shift_reg,logic_shift_reg,alus_shift_reg,logics_shift_reg,\
|
|
112 extend,\
|
|
113 mov_shift,mvn_shift,mov_shift_reg,mvn_shift_reg"))
|
|
114 "pj4_is,(pj4_alu1,nothing,pj4_w1+pj4_cp)|(pj4_alu2,nothing,pj4_w2+pj4_cp)")
|
|
115
|
|
116 (define_bypass 2 "pj4_alu_shift,pj4_shift"
|
|
117 "pj4_ir_mul,pj4_ir_div,pj4_core_to_vfp")
|
|
118
|
|
119 (define_insn_reservation "pj4_ir_mul" 3
|
|
120 (and (eq_attr "tune" "marvell_pj4")
|
|
121 (ior (eq_attr "mul32" "yes")
|
|
122 (eq_attr "mul64" "yes")))
|
|
123 "pj4_is,pj4_mul,nothing*2,pj4_cp")
|
|
124
|
|
125 (define_insn_reservation "pj4_ir_div" 20
|
|
126 (and (eq_attr "tune" "marvell_pj4")
|
|
127 (eq_attr "type" "udiv,sdiv")) "pj4_is,pj4_div*19,pj4_cp")
|
|
128
|
|
129 ;; Branches and calls.
|
|
130
|
|
131 (define_insn_reservation "pj4_branches" 0
|
|
132 (and (eq_attr "tune" "marvell_pj4") (eq_attr "type" "branch")) "pj4_is")
|
|
133
|
|
134 (define_insn_reservation "pj4_calls" 32
|
|
135 (and (eq_attr "tune" "marvell_pj4") (eq_attr "type" "call")) "pj4_is")
|
|
136
|
|
137 ;; Load/store instructions
|
|
138
|
|
139 (define_insn_reservation "pj4_ldr" 3
|
|
140 (and (eq_attr "tune" "marvell_pj4")
|
|
141 (eq_attr "type" "load_byte,load_4"))
|
|
142 "pj4_is,pj4_alu1,nothing*2,pj4_cp")
|
|
143
|
|
144 (define_insn_reservation "pj4_ldrd" 3
|
|
145 (and (eq_attr "tune" "marvell_pj4")
|
|
146 (eq_attr "type" "load_8"))
|
|
147 "pj4_is,pj4_alu1,nothing*2,pj4_cpb")
|
|
148
|
|
149 (define_insn_reservation "pj4_str" 1
|
|
150 (and (eq_attr "tune" "marvell_pj4")
|
|
151 (eq_attr "type" "store_4"))
|
|
152 "pj4_is,pj4_alu1,nothing*2,pj4_cp")
|
|
153
|
|
154 (define_insn_reservation "pj4_strd" 1
|
|
155 (and (eq_attr "tune" "marvell_pj4")
|
|
156 (eq_attr "type" "store_8"))
|
|
157 "pj4_is,pj4_alu1,nothing*2,pj4_cpb")
|
|
158
|
|
159 (define_insn_reservation "pj4_ldm" 4
|
|
160 (and (eq_attr "tune" "marvell_pj4")
|
|
161 (eq_attr "type" "load_12,load_16")) "pj4_isb,pj4_isb+pj4_alu1,pj4_alu1,nothing,pj4_cp,pj4_cp")
|
|
162
|
|
163 (define_insn_reservation "pj4_stm" 2
|
|
164 (and (eq_attr "tune" "marvell_pj4")
|
|
165 (eq_attr "type" "store_12,store_16")) "pj4_isb,pj4_isb+pj4_alu1,pj4_alu1,nothing,pj4_cp,pj4_cp")
|
|
166
|
|
167 ;; Loads forward at WR-stage to ALU pipes
|
|
168 (define_bypass 2 "pj4_ldr,pj4_ldrd" "pj4_alu")
|
|
169 (define_bypass 2 "pj4_ldr,pj4_ldrd" "pj4_alu_shift" "arm_no_early_alu_shift_dep")
|
|
170
|
|
171 (define_bypass 4 "pj4_ldr,pj4_ldrd" "pj4_ir_mul,pj4_ir_div,pj4_core_to_vfp")
|
|
172 (define_bypass 5 "pj4_ldm" "pj4_ir_mul,pj4_ir_div,pj4_core_to_vfp")
|
|
173
|
|
174 ;; Loads to stores can back-to-back forward
|
|
175 (define_bypass 1 "pj4_ldr,pj4_ldrd" "pj4_str,pj4_strd" "arm_no_early_store_addr_dep")
|
|
176
|
|
177 ;; PJ4 VFP floating point unit
|
|
178 (define_automaton "pj4_vfp")
|
|
179
|
|
180 (define_cpu_unit "vissue" "pj4_vfp")
|
|
181 (define_cpu_unit "vadd" "pj4_vfp")
|
|
182 (define_cpu_unit "vmul" "pj4_vfp")
|
|
183 (define_cpu_unit "vdiv" "pj4_vfp")
|
|
184 (define_cpu_unit "vfast" "pj4_vfp")
|
|
185
|
|
186 (define_insn_reservation "pj4_vfp_add" 5
|
|
187 (and (eq_attr "tune" "marvell_pj4")
|
|
188 (eq_attr "type" "fadds,faddd")) "pj4_is,nothing*2,vissue,vadd,nothing*3")
|
|
189
|
|
190 (define_insn_reservation "pj4_vfp_mul" 6
|
|
191 (and (eq_attr "tune" "marvell_pj4")
|
|
192 (eq_attr "type" "fmuls,fmuld")) "pj4_is,nothing*2,vissue,vmul,nothing*4")
|
|
193
|
|
194 (define_insn_reservation "pj4_vfp_divs" 20
|
|
195 (and (eq_attr "tune" "marvell_pj4")
|
|
196 (eq_attr "type" "fdivs, fsqrts")) "pj4_is,nothing*2,vissue,vdiv*18,nothing")
|
|
197
|
|
198 (define_insn_reservation "pj4_vfp_divd" 34
|
|
199 (and (eq_attr "tune" "marvell_pj4")
|
|
200 (eq_attr "type" "fdivd, fsqrtd")) "pj4_is,nothing*2,vissue,vdiv*32,nothing")
|
|
201
|
|
202 (define_insn_reservation "pj4_vfp_mac" 9
|
|
203 (and (eq_attr "tune" "marvell_pj4")
|
|
204 (eq_attr "type" "fmacs,fmacd"))
|
|
205 "pj4_is,nothing*2,vissue,vmul,nothing*3,vadd,nothing*3")
|
|
206
|
|
207 (define_bypass 5 "pj4_vfp_mac" "pj4_vfp_mac" "arm_no_early_mul_dep")
|
|
208
|
|
209 (define_insn_reservation "pj4_vfp_cpy" 4
|
|
210 (and (eq_attr "tune" "marvell_pj4")
|
|
211 (eq_attr "type" "fmov,ffariths,ffarithd,fconsts,fconstd,\
|
|
212 fcmps,fcmpd,f_cvt,f_cvtf2i,f_cvti2f"))
|
|
213 "pj4_is,nothing*2,vissue,vfast,nothing*2")
|
|
214
|
|
215 ;; Enlarge latency, and wish that more nondependent insns are
|
|
216 ;; scheduled immediately after VFP load.
|
|
217 (define_insn_reservation "pj4_vfp_load" 4
|
|
218 (and (eq_attr "tune" "marvell_pj4")
|
|
219 (eq_attr "type" "f_loads,f_loadd")) "pj4_isb,pj4_alu1,nothing,vissue,pj4_cp")
|
|
220
|
|
221 (define_insn_reservation "pj4_vfp_store" 1
|
|
222 (and (eq_attr "tune" "marvell_pj4")
|
|
223 (eq_attr "type" "f_stores,f_stored")) "pj4_isb,pj4_alu1,nothing,vissue,pj4_cp")
|
|
224
|
|
225 (define_insn_reservation "pj4_vfp_to_core" 7
|
|
226 (and (eq_attr "tune" "marvell_pj4")
|
|
227 (eq_attr "type" "f_mrc,f_mrrc,f_flag")) "pj4_isb,nothing,nothing,vissue,vfast,nothing*2")
|
|
228
|
|
229 (define_insn_reservation "pj4_core_to_vfp" 2
|
|
230 (and (eq_attr "tune" "marvell_pj4")
|
|
231 (eq_attr "type" "f_mcr,f_mcrr")) "pj4_isb,pj4_alu1,pj4_w1,vissue,pj4_cp")
|
|
232
|