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
267
268
269
270
271
272
273
274
|
%%% -*- erlang-indent-level: 2 -*-
%%%
%%% %CopyrightBegin%
%%%
%%% Copyright Ericsson AB 2003-2016. All Rights Reserved.
%%%
%%% Licensed under the Apache License, Version 2.0 (the "License");
%%% you may not use this file except in compliance with the License.
%%% You may obtain a copy of the License at
%%%
%%% http://www.apache.org/licenses/LICENSE-2.0
%%%
%%% Unless required by applicable law or agreed to in writing, software
%%% distributed under the License is distributed on an "AS IS" BASIS,
%%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
%%% See the License for the specific language governing permissions and
%%% limitations under the License.
%%%
%%% %CopyrightEnd%
%%%
%%%----------------------------------------------------------------------
%%% File : hipe_x86_postpass.erl
%%% Author : Christoffer Vikström <[email protected]>
%%% Purpose : Contain postpass optimisations for x86-assembler code.
%%% Created : 5 Aug 2003 by Christoffer Vikström <[email protected]>
%%%----------------------------------------------------------------------
-ifndef(HIPE_X86_POSTPASS).
-define(HIPE_X86_POSTPASS, hipe_x86_postpass).
-endif.
-module(?HIPE_X86_POSTPASS).
-export([postpass/2]).
-include("../x86/hipe_x86.hrl").
%%>----------------------------------------------------------------------<
% Procedure : postpass/2
% Purpose : Function that performs a nr of postpass optimizations on
% the hipe x86-assembler code before it is encoded and loaded.
%%>----------------------------------------------------------------------<
postpass(#defun{code=Code0}=Defun, Options) ->
Code1 = pseudo_insn_expansion(Code0),
Code2 = case proplists:get_bool(peephole, Options) of
true -> peephole_optimization(Code1);
false -> Code1
end,
Code3 = trivial_goto_elimination(Code2),
Defun#defun{code=Code3}.
%%>----------------------------------------------------------------------<
% Procedure : peep/1
% Purpose : Function that does peephole optimizations. It works by
% moving a window over the code and looking at a sequence of
% a few instructions. Replaces long sequences of instructions
% with shorter ones and removes unnecesary ones.
% Arguments : Insns - List of pseudo x86-assembler records.
% Res - Returned list of pseudo x86-assembler records.
% Kept reversed, until it is returned.
% Return : An optimized list of pseudo x86-assembler records with
% (hopefully) fewer or faster instructions.
%%>----------------------------------------------------------------------<
peephole_optimization(Insns) ->
peep(Insns, [], []).
%% MoveSelf related peep-opts
%% ------------------------------
peep([#fmove{src=Src, dst=Src} | Insns], Res,Lst) ->
peep(Insns, Res, [moveSelf1|Lst]);
peep([I=#fmove{src=Src, dst=Dst},
#fmove{src=Dst, dst=Src} | Insns], Res,Lst) ->
peep(Insns, [I|Res], [moveSelf2|Lst]);
peep([#movsx{src=Src, dst=Src} | Insns], Res,Lst) ->
peep(Insns, Res, [moveSelf3|Lst]);
peep([I=#movsx{src=Src, dst=Dst},
#movsx{src=Dst, dst=Src} | Insns], Res,Lst) ->
peep(Insns, [I|Res], [moveSelf4|Lst]);
peep([#movzx{src=Src, dst=Src} | Insns], Res,Lst) ->
peep(Insns, Res, [moveSelf5|Lst]);
peep([I=#movzx{src=Src, dst=Dst},
#movzx{src=Dst, dst=Src} | Insns], Res,Lst) ->
peep(Insns, [I|Res], [moveSelf6|Lst]);
peep([#cmovcc{src=Src, dst=Src} | Insns], Res,Lst) ->
peep(Insns, Res, [moveSelf7|Lst]);
peep([I=#cmovcc{src=Src, dst=Dst},
#cmovcc{src=Dst, dst=Src}|Insns], Res,Lst) ->
peep(Insns, [I|Res], [moveSelf8|Lst]);
peep([#move{src=#x86_temp{reg=X},
dst=#x86_temp{reg=X}} | Insns], Res,Lst) ->
peep(Insns, Res, [moveSelf9|Lst]);
peep([I=#move{src=#x86_temp{reg=Src}, dst=#x86_temp{reg=Dst}},
#move{src=#x86_temp{reg=Dst}, dst=#x86_temp{reg=Src}} | Insns], Res,Lst) ->
peep(Insns, [I|Res], [moveSelf0|Lst]);
%% ElimBinALMDouble
%% ----------------
peep([Move=#move{src=Src, dst=Dst}, Alu=#alu{src=Src, dst=Dst}|Insns], Res, Lst)
when not is_record(Dst, x86_mem) ->
peep([Alu#alu{src=Dst}|Insns], [Move|Res], [elimBinALMDouble|Lst]);
%% ElimFBinDouble
%% --------------
peep([Move=#fmove{src=Src, dst=Dst},
BinOp=#fp_binop{src=Src, dst=Dst}|Insns], Res, Lst) ->
peep([BinOp#fp_binop{src=Dst}|Insns], [Move|Res], [elimFBinDouble|Lst]);
%% CommuteBinALMD
%% --------------
peep([#move{src=Src1, dst=Dst},
#alu{aluop=Op,src=Src2,dst=Dst}|Insns], Res, Lst)
when (Src1 =:= #x86_imm{}) and (Src2 =/= #x86_imm{}) and
((Op =:= 'add') or (Op =:= 'and') or (Op =:= 'or') or (Op =:= 'xor')) ->
peep(Insns, [#alu{aluop=Op,src=Src1,dst=Dst},
#move{src=Src2, dst=Dst}|Res],
[commuteBinALMD|Lst]);
%% ElimCmp0
%% --------
peep([#cmp{src=#x86_imm{value=0}, dst=Dst=#x86_temp{}}|Insns],Res,Lst) ->
%% TEST leaves the adjust flag undefined, whereas CMP sets it properly (in
%% this case to 0). However, since HiPE does not use any instructions that
%% read the adjust flag, we can do this transform safely.
peep(Insns, [#test{src=Dst, dst=Dst} | Res], [elimCmp0_1|Lst]);
peep([#cmp{src=Src=#x86_temp{}, dst=#x86_imm{value=0}},
J=#jcc{cc=Cond}|Insns],Res,Lst)
when Cond =:= 'e'; Cond =:= 'ne' -> % We're commuting the comparison
peep(Insns, [J, #test{src=Src, dst=Src} | Res], [elimCmp0_2|Lst]);
%% ElimCmpTest
%% -----------
peep([I|Insns],Res,Lst) when (I =:= #cmp{}) or (I =:= #test{}) ->
case check(Insns) of
#jcc{} ->
peep(Insns, [I|Res], Lst);
#jmp_fun{} ->
peep(Insns, [I|Res], Lst);
#jmp_label{} ->
peep(Insns, [I|Res], Lst);
#jmp_switch{} ->
peep(Insns, [I|Res], Lst);
#cmovcc{} ->
peep(Insns, [I|Res], Lst);
#ret{} ->
peep(Insns, [I|Res], Lst);
_ ->
peep(Insns, Res, [elimCmpTest|Lst])
end;
%% ElimPushPop
%% -----------
peep([#push{src=Opr}, #pop{dst=Opr} | Insns], Res, Lst) ->
peep(Insns, Res, [elimPushPop|Lst]);
% %% ElimIFF
% %% -------
peep([#jcc{label=Lab}, I=#label{label=Lab}|Insns], Res, Lst) ->
peep(Insns, [I, #jmp_label{label=Lab}|Res], [elimIFF|Lst]);
%% ElimSet0
%% --------
peep([#move{src=#x86_imm{value=0},dst=Dst}|Insns],Res,Lst)
when (Dst==#x86_temp{}) ->
peep(Insns, [#alu{aluop='xor', src=Dst, dst=Dst}|Res], [elimSet0|Lst]);
%% ElimMDPow2
%% ----------
peep([B = #alu{aluop=Op,src=#x86_imm{value=Val},dst=Dst}|Insns], Res, Lst) ->
{IsLog2, Size, Sign} = log2(Val),
case ((Op =:= imul) or (Op =:= idiv)) and IsLog2 of
true ->
Sh = case Sign of positive -> 'bsl'; negative -> 'bsr' end,
peep(Insns,
[#shift{shiftop=Sh, src=#x86_imm{value=Size}, dst=Dst}|Res],
[elimMDPow2|Lst]);
false ->
peep(Insns, [B|Res], Lst)
end;
%% SubToDec
%% This rule turns "subl $1,Dst; jl Lab" into "decl Dst; jl Lab", which
%% changes reduction counter tests to use decl instead of subl.
%% However, on Athlon64 this leads to a small but measurable decrease
%% in performance. The use of dec is also not recommended on P4, so
%% this transformation is disabled.
%% peep([#alu{aluop='sub',src=#x86_imm{value=1},dst=Dst},J=#jcc{cc='l'}|Insns], Res, Lst) ->
%% peep(Insns, [J, #dec{dst=Dst} | Res], [subToDec|Lst]);
%% Standard list recursion clause
%% ------------------------------
peep([I | Insns], Res, Lst) ->
peep(Insns, [I|Res], Lst);
peep([], Res, _Lst) ->
lists:reverse(Res).
%% Simple goto elimination
%% -----------------------
trivial_goto_elimination(Insns) -> goto_elim(Insns, []).
goto_elim([#jmp_label{label=Label}, I = #label{label=Label}|Insns], Res) ->
goto_elim([I|Insns], Res);
goto_elim([I | Insns], Res) ->
goto_elim(Insns, [I|Res]);
goto_elim([], Res) ->
lists:reverse(Res).
%%>----------------------------------------------------------------------<
%% Procedure : expand/1
%% Purpose : Expands pseudo instructions.
%% Arguments : Insns - An x86-instruction list.
%% Return : An expanded instruction list.
%% Notes :
%%>----------------------------------------------------------------------<
pseudo_insn_expansion(Insns) -> expand(Insns, []).
expand([I|Tail], Res) ->
case I of
#pseudo_jcc{cc=Cc,true_label=TrueLab,false_label=FalseLab} ->
expand(Tail, [hipe_x86:mk_jmp_label(FalseLab),
hipe_x86:mk_jcc(Cc, TrueLab) | Res]);
#pseudo_tailcall_prepare{} ->
expand(Tail, Res);
#pseudo_call{'fun'=Fun,sdesc=SDesc,contlab=ContLab,linkage=Linkage} ->
expand(Tail, [hipe_x86:mk_jmp_label(ContLab),
hipe_x86:mk_call(Fun, SDesc, Linkage) | Res]);
_ ->
expand(Tail, [I|Res])
end;
expand([], Res) -> lists:reverse(Res).
%% Log2 function
%% -------------
%% Used by ElimMDPow2 clause of peep(..)
log2(Nr) -> log2(Nr, 0).
log2(0, _) -> {false, 0, positive};
log2(Nr, I) ->
case (Nr band 1) =:= 1 of
true ->
case Nr of
1 ->
{true, I, positive};
-1 ->
{true, I, negative};
_ ->
{false, 0, positive}
end;
false ->
log2((Nr bsr 1), I+1)
end.
%% Skips through all comments and move instructions and returns the next one
%% -------------------------------------------------------------------------
%% Used by ElimCmpTest above.
check([I|Ins]) ->
case I of
#comment{} ->
check(Ins);
#move{} ->
check(Ins);
#fmove{} ->
check(Ins);
#movsx{} ->
check(Ins);
#movzx{} ->
check(Ins);
OtherI ->
OtherI
end.
|
