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
|
%%
%% %CopyrightBegin%
%%
%% Copyright Ericsson AB 2001-2009. All Rights Reserved.
%%
%% The contents of this file are subject to the Erlang Public License,
%% Version 1.1, (the "License"); you may not use this file except in
%% compliance with the License. You should have received a copy of the
%% Erlang Public License along with this software. If not, it can be
%% retrieved online at http://www.erlang.org/.
%%
%% Software distributed under the License is distributed on an "AS IS"
%% basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See
%% the License for the specific language governing rights and limitations
%% under the License.
%%
%% %CopyrightEnd%
%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%
%% ULTRASPARC MACHINE MODEL
%%
%% This module is used by the scheduler.
%% The following interface is used:
%% ...
%%
%% NOTES:
%% - the machine model is simple (on the verge of simplistic)
%% * all FUs are pipelined => model only one cycle at a time
%% * instruction latencies are mostly 1
%% * floating point is left for later (I _think_ it works, but ...)
%% - conservative: instructions that require multiple resources are
%% modelled as 'single'; instead, they could reserve IEU+BR or whatever
%% - possibly inefficient: I think machine state model could be turned into
%% a bitvector.
-module(hipe_ultra_mod2).
-export([init_resources/1,
init_instr_resources/2,
resources_available/4,
advance_cycle/1
]).
-export([raw_latency/2,
war_latency/2,
waw_latency/2,
%% m_raw_latency/2,
%% m_war_latency/2,
%% m_waw_latency/2,
m_raw_latency/0,
m_war_latency/0,
m_waw_latency/0,
br_to_unsafe_latency/2,
unsafe_to_br_latency/2,
br_br_latency/2
]).
-include("../sparc/hipe_sparc.hrl").
-define(debug(Str,Args),ok).
%-define(debug(Str,Args),io:format(Str,Args)).
-define(debug_ultra(Str,Args),ok).
%-define(debug_ultra(Str,Args),io:format(Str,Args)).
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%
%% Straightforward and somewhat simplistic model for UltraSparc:
%% - only one cycle at a time is modelled
%% - resources are simplified:
%% * ieu0, ieu1, ieu, mem, br, single
%% * per-cycle state = done | { I0, I1, NumI, X, Mem, Br }
%% * unoptimized representation (could be bit vector)
init_resources(_Size) ->
?debug_ultra('init res ~p~n',[_Size]),
empty_state().
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
init_instr_resources(N,Nodes) ->
ultra_instr_rsrcs(Nodes,hipe_vectors:new(N, '')).
ultra_instr_rsrcs([],I_res) -> I_res;
ultra_instr_rsrcs([N|Ns],I_res) ->
ultra_instr_rsrcs(Ns,ultra_instr_type(N,I_res)).
ultra_instr_type({N,I},I_res) ->
hipe_vectors:set(I_res,N-1,instr_type(I)).
instr_type(I) ->
case I of
#move{} ->
ieu;
#multimove{} -> %% TODO: expand multimoves before scheduling
ieu;
#alu{} ->
case hipe_sparc:alu_operator(I) of
'>>' -> ieu0;
'<<' -> ieu0;
_ -> ieu
end;
#alu_cc{} ->
ieu1;
#sethi{} ->
ieu;
#load{} ->
mem;
#store{} ->
mem;
#b{} ->
br;
#br{} ->
br;
#goto{} ->
br;
#jmp_link{} -> % imprecise; should be mem+br?
single;
#jmp{} -> % imprecise
br;
#call_link{} -> % imprecise; should be mem+br?
single;
#cmov_cc{} -> % imprecise
single;
#cmov_r{} -> % imprecise
single;
#load_atom{} -> % should be resolved to sethi/or
single;
#load_address{} -> % should be resolved to sethi/or
single;
#load_word_index{} -> % should be resolved to sethi/or
single;
%% uncommon types:
#label{} ->
none;
#nop{} ->
none;
#comment{} ->
none;
_ ->
exit({ultrasparc_instr_type,{cant_schedule,I}})
end.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
resources_available(_Cycle, I, Rsrc, I_res) ->
res_avail(instruction_resource(I_res, I), Rsrc).
instruction_resource(I_res, I) ->
hipe_vectors:get(I_res, I-1).
%% The following function checks resource availability.
%% * all function units are assumed to be fully pipelined, so only
%% one cycle at a time is modelled.
%% * for IEU0 and IEU1, these must precede all generic IEU instructions
%% (handled by X bit)
%% * at most 2 integer instructions can issue in a cycle
%% * mem is straightforward
%% * br closes the cycle (= returns done).
%% * single requires an entirely empty state and closes the cycle
res_avail(ieu0, { free, I1, NumI, free, Mem, Br })
when is_integer(NumI), NumI < 2 ->
{ yes, { occ, I1, NumI+1, free, Mem, Br }};
res_avail(ieu1, { _I0, free, NumI, free, Mem, Br })
when is_integer(NumI), NumI < 2 ->
{ yes, { free, occ, NumI+1, free, Mem, Br }};
res_avail(ieu, { I0, I1, NumI, _X, Mem, Br })
when is_integer(NumI), NumI < 2 ->
{ yes, { I0, I1, NumI+1, occ, Mem, Br }};
res_avail(mem, { I0, I1, NumI, X, free, Br }) ->
{ yes, { I0, I1, NumI, X, occ, Br }};
res_avail(br, { _I0, _I1, _NumI, _X, _Mem, free }) ->
{ yes, done };
res_avail(single, { free, free, 0, free, free, free }) ->
{ yes, done };
res_avail(_, _) ->
no.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
advance_cycle(_Rsrc) ->
empty_state().
empty_state() -> { free, free, 0, free, free, free }.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%% Latencies are taken from UltraSparc hardware manual
%%
%% *** UNFINISHED ***
%% more precisely, they are taken from my memory of the US-manual
%% at the moment.
%%
%% Note: all ld/st are assumed to hit in the L1 cache (D-cache),
%% which is sort of imprecise.
raw_latency(alu, store) -> 0;
raw_latency(load, _) -> 2; % only if load is L1 hit
raw_latency(alu_cc, b) -> 0;
raw_latency(_I0, _I1) ->
1.
war_latency(_I0, _I1) ->
0.
waw_latency(_I0, _I1) ->
1.
%% *** UNFINISHED ***
%% At present, all load/stores are assumed to hit in the L1 cache,
%% which isn't really satisfying.
%% m_raw_latency(_St, _Ld) ->
%% 1.
%%
%% m_war_latency(_Ld, _St) ->
%% 1.
%%
%% m_waw_latency(_St1, _St2) ->
%% 1.
%% Use these for 'default latencies' = do not permit reordering.
m_raw_latency() ->
1.
m_war_latency() ->
1.
m_waw_latency() ->
1.
br_to_unsafe_latency(_BrTy, _UTy) ->
0.
unsafe_to_br_latency(_UTy, _BrTy) ->
0.
br_br_latency(_BrTy1, _BrTy2) ->
0.
|
