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
|
;; Scheduling description for Alpha EV4.
;; Copyright (C) 2002 Free Software Foundation, Inc.
;;
;; This file is part of GNU CC.
;;
;; GNU CC 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 2, or (at your option)
;; any later version.
;;
;; GNU CC 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 GNU CC; see the file COPYING. If not, write to
;; the Free Software Foundation, 59 Temple Place - Suite 330,
;; Boston, MA 02111-1307, USA.
; On EV4 there are two classes of resources to consider: resources needed
; to issue, and resources needed to execute. IBUS[01] are in the first
; category. ABOX, BBOX, EBOX, FBOX, IMUL & FDIV make up the second.
; (There are a few other register-like resources, but ...)
(define_automaton "ev4_0,ev4_1,ev4_2")
(define_cpu_unit "ev4_ib0,ev4_ib1,ev4_abox,ev4_bbox" "ev4_0")
(define_cpu_unit "ev4_ebox,ev4_imul" "ev4_1")
(define_cpu_unit "ev4_fbox,ev4_fdiv" "ev4_2")
(define_reservation "ev4_ib01" "ev4_ib0|ev4_ib1")
; Assume type "multi" single issues.
(define_insn_reservation "ev4_multi" 1
(and (eq_attr "cpu" "ev4")
(eq_attr "type" "multi"))
"ev4_ib0+ev4_ib1")
; Loads from L0 completes in three cycles. adjust_cost still factors
; in user-specified memory latency, so return 1 here.
(define_insn_reservation "ev4_ld" 1
(and (eq_attr "cpu" "ev4")
(eq_attr "type" "ild,fld,ldsym"))
"ev4_ib01+ev4_abox")
; Stores can issue before the data (but not address) is ready.
(define_insn_reservation "ev4_ist" 1
(and (eq_attr "cpu" "ev4")
(eq_attr "type" "ist"))
"ev4_ib1+ev4_abox")
(define_insn_reservation "ev4_fst" 1
(and (eq_attr "cpu" "ev4")
(eq_attr "type" "fst"))
"ev4_ib0+ev4_abox")
; Branches have no delay cost, but do tie up the unit for two cycles.
(define_insn_reservation "ev4_ibr" 2
(and (eq_attr "cpu" "ev4")
(eq_attr "type" "ibr,jsr"))
"ev4_ib1+ev4_bbox,ev4_bbox")
(define_insn_reservation "ev4_fbr" 2
(and (eq_attr "cpu" "ev4")
(eq_attr "type" "fbr"))
"ev4_ib0+ev4_bbox,ev4_bbox")
; Arithmetic insns are normally have their results available after
; two cycles. There are a number of exceptions.
(define_insn_reservation "ev4_iaddlog" 2
(and (eq_attr "cpu" "ev4")
(eq_attr "type" "iadd,ilog"))
"ev4_ib0+ev4_ebox")
(define_bypass 1
"ev4_iaddlog"
"ev4_ibr,ev4_iaddlog,ev4_shiftcm,ev4_icmp,ev4_imulsi,ev4_imuldi")
(define_insn_reservation "ev4_shiftcm" 2
(and (eq_attr "cpu" "ev4")
(eq_attr "type" "shift,icmov"))
"ev4_ib0+ev4_ebox")
(define_insn_reservation "ev4_icmp" 2
(and (eq_attr "cpu" "ev4")
(eq_attr "type" "icmp"))
"ev4_ib0+ev4_ebox")
(define_bypass 1 "ev4_icmp" "ev4_ibr")
(define_bypass 0
"ev4_iaddlog,ev4_shiftcm,ev4_icmp"
"ev4_ist"
"store_data_bypass_p")
; Multiplies use a non-piplined imul unit. Also, "no [ebox] insn can
; be issued exactly three cycles before an integer multiply completes".
(define_insn_reservation "ev4_imulsi" 21
(and (eq_attr "cpu" "ev4")
(and (eq_attr "type" "imul")
(eq_attr "opsize" "si")))
"ev4_ib0+ev4_imul,ev4_imul*18,ev4_ebox")
(define_bypass 20 "ev4_imulsi" "ev4_ist" "store_data_bypass_p")
(define_insn_reservation "ev4_imuldi" 23
(and (eq_attr "cpu" "ev4")
(and (eq_attr "type" "imul")
(eq_attr "opsize" "!si")))
"ev4_ib0+ev4_imul,ev4_imul*20,ev4_ebox")
(define_bypass 22 "ev4_imuldi" "ev4_ist" "store_data_bypass_p")
; Most FP insns have a 6 cycle latency, but with a 4 cycle bypass back in.
(define_insn_reservation "ev4_fpop" 6
(and (eq_attr "cpu" "ev4")
(eq_attr "type" "fadd,fmul,fcpys,fcmov"))
"ev4_ib1+ev4_fbox")
(define_bypass 4 "ev4_fpop" "ev4_fpop")
; The floating point divider is not pipelined. Also, "no FPOP insn can be
; issued exactly five or exactly six cycles before an fdiv insn completes".
(define_insn_reservation "ev4_fdivsf" 34
(and (eq_attr "cpu" "ev4")
(and (eq_attr "type" "fdiv")
(eq_attr "opsize" "si")))
"ev4_ib1+ev4_fdiv,ev4_fdiv*28,ev4_fdiv+ev4_fbox,ev4_fbox")
(define_insn_reservation "ev4_fdivdf" 63
(and (eq_attr "cpu" "ev4")
(and (eq_attr "type" "fdiv")
(eq_attr "opsize" "di")))
"ev4_ib1+ev4_fdiv,ev4_fdiv*57,ev4_fdiv+ev4_fbox,ev4_fbox")
; Traps don't consume or produce data.
(define_insn_reservation "ev4_misc" 1
(and (eq_attr "cpu" "ev4")
(eq_attr "type" "misc"))
"ev4_ib1")
|
