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
|
;; AMD K6/K6-2 Scheduling
;; Copyright (C) 2002-2013 Free Software Foundation, Inc.
;;
;; This file is part of GCC.
;;
;; GCC is free software; you can redistribute it and/or modify
;; it under the terms of the GNU General Public License as published by
;; the Free Software Foundation; either version 3, or (at your option)
;; any later version.
;;
;; GCC is distributed in the hope that it will be useful,
;; but WITHOUT ANY WARRANTY; without even the implied warranty of
;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
;; GNU General Public License for more details.
;;
;; You should have received a copy of the GNU General Public License
;; along with GCC; see the file COPYING3. If not see
;; <http://www.gnu.org/licenses/>.
;;
;; The K6 architecture is quite similar to PPro. Important difference is
;; that there are only two decoders and they seems to be much slower than
;; any of the execution units. So we have to pay much more attention to
;; proper scheduling for the decoders.
;; FIXME: We don't do that right now. A good start would be to sort the
;; instructions based on length.
;;
;; This description is based on data from the following documents:
;;
;; "AMD-K6 Processor Data Sheet (Preliminary information)"
;; Advanced Micro Devices, Inc., 1998.
;;
;; "AMD-K6 Processor Code Optimization Application Note"
;; Advanced Micro Devices, Inc., 2000.
;;
;; CPU execution units of the K6:
;;
;; store describes the Store unit. This unit is not modelled
;; completely and it is only used to model lea operation.
;; Otherwise it lies outside of any critical path.
;; load describes the Load unit
;; alux describes the Integer X unit
;; mm describes the Multimedia unit, which shares a pipe
;; with the Integer X unit. This unit is used for MMX,
;; which is not implemented for K6.
;; aluy describes the Integer Y unit
;; fpu describes the FPU unit
;; branch describes the Branch unit
;;
;; The fp unit is not pipelined, and it can only do one operation per two
;; cycles, including fxcg.
;;
;; Generally this is a very poor description, but at least no worse than
;; the old description, and a lot easier to extend to something more
;; reasonable if anyone still cares enough about this architecture in 2004.
;;
;; ??? fxch isn't handled; not an issue until sched3 after reg-stack is real.
(define_automaton "k6_decoder,k6_load_unit,k6_store_unit,k6_integer_units,k6_fpu_unit,k6_branch_unit")
;; The K6 instruction decoding begins before the on-chip instruction cache is
;; filled. Depending on the length of the instruction, two simple instructions
;; can be decoded in two parallel short decoders, or one complex instruction can
;; be decoded in either the long or the vector decoder. For all practical
;; purposes, the long and vector decoder can be modelled as one decoder.
(define_cpu_unit "k6_decode_short0" "k6_decoder")
(define_cpu_unit "k6_decode_short1" "k6_decoder")
(define_cpu_unit "k6_decode_long" "k6_decoder")
(exclusion_set "k6_decode_long" "k6_decode_short0,k6_decode_short1")
(define_reservation "k6_decode_short" "k6_decode_short0|k6_decode_short1")
(define_reservation "k6_decode_vector" "k6_decode_long")
(define_cpu_unit "k6_store" "k6_store_unit")
(define_cpu_unit "k6_load" "k6_load_unit")
(define_cpu_unit "k6_alux,k6_aluy" "k6_integer_units")
(define_cpu_unit "k6_fpu" "k6_fpu_unit")
(define_cpu_unit "k6_branch" "k6_branch_unit")
;; Shift instructions and certain arithmetic are issued only on Integer X.
(define_insn_reservation "k6_alux_only" 1
(and (eq_attr "cpu" "k6")
(and (eq_attr "type" "ishift,ishift1,rotate,rotate1,alu1,negnot")
(eq_attr "memory" "none")))
"k6_decode_short,k6_alux")
(define_insn_reservation "k6_alux_only_load" 3
(and (eq_attr "cpu" "k6")
(and (eq_attr "type" "ishift,ishift1,rotate,rotate1,alu1,negnot")
(eq_attr "memory" "load")))
"k6_decode_short,k6_load,k6_alux")
(define_insn_reservation "k6_alux_only_store" 3
(and (eq_attr "cpu" "k6")
(and (eq_attr "type" "ishift,ishift1,rotate,rotate1,alu1,negnot")
(eq_attr "memory" "store,both,unknown")))
"k6_decode_long,k6_load,k6_alux,k6_store")
;; Integer divide and multiply can only be issued on Integer X, too.
(define_insn_reservation "k6_alu_imul" 2
(and (eq_attr "cpu" "k6")
(eq_attr "type" "imul"))
"k6_decode_vector,k6_alux*3")
(define_insn_reservation "k6_alu_imul_load" 4
(and (eq_attr "cpu" "k6")
(and (eq_attr "type" "imul")
(eq_attr "memory" "load")))
"k6_decode_vector,k6_load,k6_alux*3")
(define_insn_reservation "k6_alu_imul_store" 4
(and (eq_attr "cpu" "k6")
(and (eq_attr "type" "imul")
(eq_attr "memory" "store,both,unknown")))
"k6_decode_vector,k6_load,k6_alux*3,k6_store")
;; ??? Guessed latencies based on the old pipeline description.
(define_insn_reservation "k6_alu_idiv" 17
(and (eq_attr "cpu" "k6")
(and (eq_attr "type" "idiv")
(eq_attr "memory" "none")))
"k6_decode_vector,k6_alux*17")
(define_insn_reservation "k6_alu_idiv_mem" 19
(and (eq_attr "cpu" "k6")
(and (eq_attr "type" "idiv")
(eq_attr "memory" "!none")))
"k6_decode_vector,k6_load,k6_alux*17")
;; Basic word and doubleword ALU ops can be issued on both Integer units.
(define_insn_reservation "k6_alu" 1
(and (eq_attr "cpu" "k6")
(and (eq_attr "type" "alu,alu1,negnot,icmp,test,imovx,incdec,setcc")
(eq_attr "memory" "none")))
"k6_decode_short,k6_alux|k6_aluy")
(define_insn_reservation "k6_alu_load" 3
(and (eq_attr "cpu" "k6")
(and (eq_attr "type" "alu,alu1,negnot,icmp,test,imovx,incdec,setcc")
(eq_attr "memory" "load")))
"k6_decode_short,k6_load,k6_alux|k6_aluy")
(define_insn_reservation "k6_alu_store" 3
(and (eq_attr "cpu" "k6")
(and (eq_attr "type" "alu,alu1,negnot,icmp,test,imovx,incdec,setcc")
(eq_attr "memory" "store,both,unknown")))
"k6_decode_long,k6_load,k6_alux|k6_aluy,k6_store")
;; A "load immediate" operation does not require execution at all,
;; it is available immediately after decoding. Special-case this.
(define_insn_reservation "k6_alu_imov" 1
(and (eq_attr "cpu" "k6")
(and (eq_attr "type" "imov")
(and (eq_attr "memory" "none")
(match_operand 1 "nonimmediate_operand"))))
"k6_decode_short,k6_alux|k6_aluy")
(define_insn_reservation "k6_alu_imov_imm" 0
(and (eq_attr "cpu" "k6")
(and (eq_attr "type" "imov")
(and (eq_attr "memory" "none")
(match_operand 1 "immediate_operand"))))
"k6_decode_short")
(define_insn_reservation "k6_alu_imov_load" 2
(and (eq_attr "cpu" "k6")
(and (eq_attr "type" "imov")
(eq_attr "memory" "load")))
"k6_decode_short,k6_load")
(define_insn_reservation "k6_alu_imov_store" 1
(and (eq_attr "cpu" "k6")
(and (eq_attr "type" "imov")
(eq_attr "memory" "store")))
"k6_decode_short,k6_store")
(define_insn_reservation "k6_alu_imov_both" 2
(and (eq_attr "cpu" "k6")
(and (eq_attr "type" "imov")
(eq_attr "memory" "both,unknown")))
"k6_decode_long,k6_load,k6_alux|k6_aluy")
;; The branch unit.
(define_insn_reservation "k6_branch_call" 1
(and (eq_attr "cpu" "k6")
(eq_attr "type" "call,callv"))
"k6_decode_vector,k6_branch")
(define_insn_reservation "k6_branch_branch" 1
(and (eq_attr "cpu" "k6")
(eq_attr "type" "ibr"))
"k6_decode_short,k6_branch")
;; The load and units have two pipeline stages. The load latency is
;; two cycles.
(define_insn_reservation "k6_load_pop" 3
(and (eq_attr "cpu" "k6")
(ior (eq_attr "type" "pop")
(eq_attr "memory" "load,both")))
"k6_decode_short,k6_load")
(define_insn_reservation "k6_load_leave" 5
(and (eq_attr "cpu" "k6")
(eq_attr "type" "leave"))
"k6_decode_long,k6_load,(k6_alux|k6_aluy)*2")
;; ??? From the old pipeline description. Egad!
;; ??? Apparently we take care of this reservation in adjust_cost.
(define_insn_reservation "k6_load_str" 10
(and (eq_attr "cpu" "k6")
(and (eq_attr "type" "str")
(eq_attr "memory" "load,both")))
"k6_decode_vector,k6_load*10")
;; The store unit handles lea and push. It is otherwise unmodelled.
(define_insn_reservation "k6_store_lea" 2
(and (eq_attr "cpu" "k6")
(eq_attr "type" "lea"))
"k6_decode_short,k6_store,k6_alux|k6_aluy")
(define_insn_reservation "k6_store_push" 2
(and (eq_attr "cpu" "k6")
(ior (eq_attr "type" "push")
(eq_attr "memory" "store,both")))
"k6_decode_short,k6_store")
(define_insn_reservation "k6_store_str" 10
(and (eq_attr "cpu" "k6")
(eq_attr "type" "str"))
"k6_store*10")
;; Most FPU instructions have latency 2 and throughput 2.
(define_insn_reservation "k6_fpu" 2
(and (eq_attr "cpu" "k6")
(and (eq_attr "type" "fop,fmov,fcmp,fistp")
(eq_attr "memory" "none")))
"k6_decode_vector,k6_fpu*2")
(define_insn_reservation "k6_fpu_load" 6
(and (eq_attr "cpu" "k6")
(and (eq_attr "type" "fop,fmov,fcmp,fistp")
(eq_attr "memory" "load,both")))
"k6_decode_short,k6_load,k6_fpu*2")
(define_insn_reservation "k6_fpu_store" 6
(and (eq_attr "cpu" "k6")
(and (eq_attr "type" "fop,fmov,fcmp,fistp")
(eq_attr "memory" "store")))
"k6_decode_short,k6_store,k6_fpu*2")
(define_insn_reservation "k6_fpu_fmul" 2
(and (eq_attr "cpu" "k6")
(and (eq_attr "type" "fmul")
(eq_attr "memory" "none")))
"k6_decode_short,k6_fpu*2")
(define_insn_reservation "k6_fpu_fmul_load" 2
(and (eq_attr "cpu" "k6")
(and (eq_attr "type" "fmul")
(eq_attr "memory" "load,both")))
"k6_decode_short,k6_load,k6_fpu*2")
;; ??? Guessed latencies from the old pipeline description.
(define_insn_reservation "k6_fpu_expensive" 56
(and (eq_attr "cpu" "k6")
(eq_attr "type" "fdiv,fpspc"))
"k6_decode_short,k6_fpu*56")
|
