111
|
1 ;; Scheduling description for the Xilinx PowerPC 405 APU Floating Point Unit.
|
131
|
2 ;; Copyright (C) 2008-2018 Free Software Foundation, Inc.
|
111
|
3 ;; Contributed by Michael Eager (eager@eagercon.com).
|
|
4 ;;
|
|
5 ;; This file is part of GCC.
|
|
6 ;;
|
|
7 ;; GCC is free software; you can redistribute it and/or modify
|
|
8 ;; it under the terms of the GNU General Public License as published by
|
|
9 ;; the Free Software Foundation; either version 3, or (at your option)
|
|
10 ;; any later version.
|
|
11 ;;
|
|
12 ;; GCC is distributed in the hope that it will be useful,
|
|
13 ;; but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
14 ;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
15 ;; GNU General Public 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 ;;----------------------------------------------------
|
|
22 ;; Xilinx APU FPU Pipeline Description
|
|
23 ;;
|
|
24 ;; - attr 'type' and 'fp_type' should definitely
|
|
25 ;; be cleaned up at some point in the future.
|
|
26 ;; ddiv,sdiv,dmul,smul etc are quite confusing.
|
|
27 ;; Should use consistent fp* attrs. 'fp_type'
|
|
28 ;; should also go away, leaving us only with 'fp'
|
|
29 ;;
|
|
30 ;;----------------------------------------------------
|
|
31
|
|
32 ;; -------------------------------------------------------------------------
|
|
33 ;; Latencies
|
|
34 ;; Latest latency figures (all in FCB cycles). PowerPC to FPU frequency ratio
|
|
35 ;; assumed to be 1/2. (most common deployment)
|
|
36 ;; Add 2 PPC cycles for (register file access + wb) and 2 PPC cycles
|
|
37 ;; for issue (from PPC)
|
|
38 ;; SP DP
|
|
39 ;; Loads: 4 6
|
|
40 ;; Stores: 1 2 (from availability of data)
|
|
41 ;; Move/Abs/Neg: 1 1
|
|
42 ;; Add/Subtract: 5 7
|
|
43 ;; Multiply: 4 11
|
|
44 ;; Multiply-add: 10 19
|
|
45 ;; Convert (any): 4 6
|
|
46 ;; Divide/Sqrt: 27 56
|
|
47 ;; Compares: 1 2
|
|
48 ;;
|
|
49 ;; bypasses needed for forwarding capability of the FPU.
|
|
50 ;; Add this at some future time.
|
|
51 ;; -------------------------------------------------------------------------
|
|
52 (define_automaton "Xfpu")
|
|
53 (define_cpu_unit "Xfpu_issue,Xfpu_addsub,Xfpu_mul,Xfpu_div,Xfpu_sqrt" "Xfpu")
|
|
54
|
|
55
|
|
56 (define_insn_reservation "fp-default" 2
|
|
57 (and (and
|
|
58 (eq_attr "type" "fp,fpsimple")
|
|
59 (eq_attr "fp_type" "fp_default"))
|
|
60 (eq_attr "cpu" "ppc405"))
|
|
61 "Xfpu_issue*2")
|
|
62
|
|
63 (define_insn_reservation "fp-compare" 6
|
|
64 (and (eq_attr "type" "fpcompare") ;; Inconsistent naming
|
|
65 (eq_attr "cpu" "ppc405"))
|
|
66 "Xfpu_issue*2,Xfpu_addsub")
|
|
67
|
|
68 (define_insn_reservation "fp-addsub-s" 14
|
|
69 (and (and
|
|
70 (eq_attr "type" "fp,fpsimple")
|
|
71 (eq_attr "fp_type" "fp_addsub_s"))
|
|
72 (eq_attr "cpu" "ppc405"))
|
|
73 "Xfpu_issue*2,Xfpu_addsub")
|
|
74
|
|
75 (define_insn_reservation "fp-addsub-d" 18
|
|
76 (and (and
|
|
77 (eq_attr "type" "fp,fpsimple")
|
|
78 (eq_attr "fp_type" "fp_addsub_d"))
|
|
79 (eq_attr "cpu" "ppc405"))
|
|
80 "Xfpu_issue*2,Xfpu_addsub")
|
|
81
|
|
82 (define_insn_reservation "fp-mul-s" 12
|
|
83 (and (and
|
|
84 (eq_attr "type" "fp")
|
|
85 (eq_attr "fp_type" "fp_mul_s"))
|
|
86 (eq_attr "cpu" "ppc405"))
|
|
87 "Xfpu_issue*2,Xfpu_mul")
|
|
88
|
|
89 (define_insn_reservation "fp-mul-d" 16 ;; Actually 28. Long latencies are killing the automaton formation. Need to figure out why.
|
|
90 (and (and
|
|
91 (eq_attr "type" "fp")
|
|
92 (eq_attr "fp_type" "fp_mul_d"))
|
|
93 (eq_attr "cpu" "ppc405"))
|
|
94 "Xfpu_issue*2,Xfpu_mul")
|
|
95
|
|
96 (define_insn_reservation "fp-div-s" 24 ;; Actually 34
|
|
97 (and (eq_attr "type" "sdiv") ;; Inconsistent attr naming
|
|
98 (eq_attr "cpu" "ppc405"))
|
|
99 "Xfpu_issue*2,Xfpu_div*10") ;; Unpipelined
|
|
100
|
|
101 (define_insn_reservation "fp-div-d" 34 ;; Actually 116
|
|
102 (and (eq_attr "type" "ddiv")
|
|
103 (eq_attr "cpu" "ppc405")) ;; Inconsistent attr naming
|
|
104 "Xfpu_issue*2,Xfpu_div*10") ;; Unpipelined
|
|
105
|
|
106 (define_insn_reservation "fp-maddsub-s" 24
|
|
107 (and (and
|
|
108 (eq_attr "type" "fp")
|
|
109 (eq_attr "fp_type" "fp_maddsub_s"))
|
|
110 (eq_attr "cpu" "ppc405"))
|
|
111 "Xfpu_issue*2,Xfpu_mul,nothing*7,Xfpu_addsub")
|
|
112
|
|
113 (define_insn_reservation "fp-maddsub-d" 34 ;; Actually 42
|
|
114 (and (and
|
|
115 (eq_attr "type" "dmul") ;; Inconsistent attr naming
|
|
116 (eq_attr "fp_type" "fp_maddsub_d"))
|
|
117 (eq_attr "cpu" "ppc405"))
|
|
118 "Xfpu_issue*2,Xfpu_mul,nothing*7,Xfpu_addsub")
|
|
119
|
|
120 (define_insn_reservation "fp-load" 10 ;; FIXME. Is double/single precision the same ?
|
|
121 (and (eq_attr "type" "fpload")
|
|
122 (eq_attr "cpu" "ppc405"))
|
|
123 "Xfpu_issue*10")
|
|
124
|
|
125 (define_insn_reservation "fp-store" 4
|
|
126 (and (eq_attr "type" "fpstore")
|
|
127 (eq_attr "cpu" "ppc405"))
|
|
128 "Xfpu_issue*4")
|
|
129
|
|
130 (define_insn_reservation "fp-sqrt-s" 24 ;; Actually 56
|
|
131 (and (eq_attr "type" "ssqrt")
|
|
132 (eq_attr "cpu" "ppc405"))
|
|
133 "Xfpu_issue*2,Xfpu_sqrt*10") ;; Unpipelined
|
|
134
|
|
135
|
|
136 (define_insn_reservation "fp-sqrt-d" 34 ;; Actually 116
|
|
137 (and (eq_attr "type" "dsqrt")
|
|
138 (eq_attr "cpu" "ppc405"))
|
|
139 "Xfpu_issue*2,Xfpu_sqrt*10") ;; Unpipelined
|
|
140
|