Commit | Line | Data |
---|---|---|
1a0c09c4 MW |
1 | /// -*- mode: asm; asm-comment-char: ?/ -*- |
2 | /// | |
3 | /// Fancy SIMD implementation of Salsa20 | |
4 | /// | |
5 | /// (c) 2015 Straylight/Edgeware | |
6 | /// | |
7 | ||
8 | ///----- Licensing notice --------------------------------------------------- | |
9 | /// | |
10 | /// This file is part of Catacomb. | |
11 | /// | |
12 | /// Catacomb is free software; you can redistribute it and/or modify | |
13 | /// it under the terms of the GNU Library General Public License as | |
14 | /// published by the Free Software Foundation; either version 2 of the | |
15 | /// License, or (at your option) any later version. | |
16 | /// | |
17 | /// Catacomb is distributed in the hope that it will be useful, | |
18 | /// but WITHOUT ANY WARRANTY; without even the implied warranty of | |
19 | /// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
20 | /// GNU Library General Public License for more details. | |
21 | /// | |
22 | /// You should have received a copy of the GNU Library General Public | |
23 | /// License along with Catacomb; if not, write to the Free | |
24 | /// Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, | |
25 | /// MA 02111-1307, USA. | |
26 | ||
27 | ///-------------------------------------------------------------------------- | |
28 | /// General definitions. | |
29 | ||
f8e509a9 MW |
30 | // Some useful variables. |
31 | .L$_subsec = 0 | |
32 | ||
33 | // Literal pools done the hard way. | |
34 | #define _LIT .text .L$_subsec + 1 | |
35 | #define _ENDLIT .text .L$_subsec | |
36 | ||
1a0c09c4 MW |
37 | // Announcing an external function. |
38 | #define FUNC(name) \ | |
39 | .globl F(name); \ | |
40 | TYPE_FUNC(name); \ | |
41 | .macro ENDFUNC; _ENDFUNC(name); .endm; \ | |
42 | FUNC_PREHOOK(name); \ | |
43 | F(name): \ | |
44 | FUNC_POSTHOOK(name) | |
45 | ||
46 | // Marking the end of a function. | |
47 | #define _ENDFUNC(name) \ | |
48 | .purgem ENDFUNC; \ | |
49 | SIZE_OBJ(name); \ | |
f8e509a9 MW |
50 | ENDFUNC_HOOK(name); \ |
51 | .L$_subsec = .L$_subsec + 2; \ | |
52 | .text .L$_subsec | |
1a0c09c4 MW |
53 | |
54 | ///-------------------------------------------------------------------------- | |
55 | /// ELF-specific hacking. | |
56 | ||
57 | #if __ELF__ | |
58 | ||
59 | #if __PIC__ || __PIE__ | |
60 | # define WANT_PIC 1 | |
61 | #endif | |
62 | ||
63 | #define TYPE_FUNC(name) .type name, STT_FUNC | |
64 | ||
65 | #define SIZE_OBJ(name) .size name, . - name | |
66 | ||
67 | #endif | |
68 | ||
69 | ///-------------------------------------------------------------------------- | |
0f23f75f MW |
70 | /// Windows-specific hacking. |
71 | ||
72 | #if ABI_WIN | |
1a0c09c4 MW |
73 | |
74 | #if CPUFAM_X86 | |
0f23f75f MW |
75 | # define F(name) _##name |
76 | #endif | |
77 | ||
78 | #endif | |
79 | ||
80 | ///-------------------------------------------------------------------------- | |
81 | /// x86- and amd64-specific hacking. | |
82 | /// | |
83 | /// It's (slightly) easier to deal with both of these in one go. | |
84 | ||
85 | #if CPUFAM_X86 || CPUFAM_AMD64 | |
1a0c09c4 MW |
86 | |
87 | // Set the function hooks. | |
88 | #define FUNC_PREHOOK(_) .balign 16 | |
89 | ||
90 | // Don't use the wretched AT&T syntax. It's festooned with pointless | |
91 | // punctuation, and all of the data movement is backwards. Ugh! | |
92 | .intel_syntax noprefix | |
93 | ||
94 | // Call external subroutine at ADDR, possibly via PLT. | |
95 | .macro callext addr | |
96 | #if WANT_PIC | |
97 | call \addr@PLT | |
98 | #else | |
99 | call \addr | |
100 | #endif | |
101 | .endm | |
102 | ||
103 | // Do I need to arrange a spare GOT register? | |
104 | #if WANT_PIC && CPUFAM_X86 | |
105 | # define NEED_GOT 1 | |
106 | #endif | |
107 | #define GOTREG ebx // Not needed in AMD64 so don't care. | |
108 | ||
109 | // Maybe load GOT address into GOT. | |
110 | .macro ldgot got=GOTREG | |
0f23f75f | 111 | #if WANT_PIC && CPUFAM_X86 |
1a0c09c4 MW |
112 | call _where_am_i.\got |
113 | add \got, offset _GLOBAL_OFFSET_TABLE_ | |
114 | #endif | |
115 | .endm | |
116 | ||
117 | // Maybe build a helper subroutine for `ldgot GOT'. | |
118 | .macro gotaux got=GOTREG | |
0f23f75f | 119 | #if WANT_PIC && CPUFAM_X86 |
1a0c09c4 MW |
120 | .align 16 |
121 | _where_am_i.\got : | |
122 | mov \got, [esp] | |
123 | ret | |
124 | #endif | |
125 | .endm | |
126 | ||
127 | // Load address of external symbol ADDR into REG, maybe using GOT. | |
128 | .macro leaext reg, addr, got=GOTREG | |
129 | #if WANT_PIC | |
0f23f75f | 130 | # if CPUFAM_X86 |
1a0c09c4 | 131 | mov \reg, [\got + \addr@GOT] |
0f23f75f MW |
132 | # endif |
133 | # if CPUFAM_AMD64 | |
134 | mov \reg, \addr@GOTPCREL[rip] | |
135 | # endif | |
1a0c09c4 | 136 | #else |
0f23f75f | 137 | # if CPUFAM_X86 |
1a0c09c4 | 138 | mov \reg, offset \addr |
0f23f75f MW |
139 | # endif |
140 | # if CPUFAM_AMD64 | |
141 | lea \reg, \addr[rip] | |
142 | # endif | |
1a0c09c4 MW |
143 | #endif |
144 | .endm | |
145 | ||
146 | // Address expression (possibly using a base register, and a displacement) | |
147 | // referring to ADDR, which is within our module, maybe using GOT. | |
148 | #define INTADDR(...) INTADDR__0(__VA_ARGS__, GOTREG, dummy) | |
149 | #define INTADDR__0(addr, got, ...) INTADDR__1(addr, got) | |
0f23f75f MW |
150 | #if CPUFAM_AMD64 |
151 | # define INTADDR__1(addr, got) addr + rip | |
152 | #elif WANT_PIC | |
1a0c09c4 MW |
153 | # define INTADDR__1(addr, got) got + addr@GOTOFF |
154 | #else | |
155 | # define INTADDR__1(addr, got) addr | |
156 | #endif | |
157 | ||
158 | #endif | |
159 | ||
61bd904b MW |
160 | ///-------------------------------------------------------------------------- |
161 | /// ARM-specific hacking. | |
162 | ||
163 | #if CPUFAM_ARM | |
164 | ||
9f6eb05d MW |
165 | // ARM/Thumb mode things. Use ARM by default. |
166 | #define ARM .arm; .L$_pcoff = 8 | |
167 | #define THUMB .thumb; .L$_pcoff = 4 | |
168 | ARM | |
169 | ||
61bd904b MW |
170 | // Set the function hooks. |
171 | #define FUNC_PREHOOK(_) .balign 4 | |
172 | #define ENDFUNC_HOOK(name) .ltorg | |
173 | ||
174 | // Call external subroutine at ADDR, possibly via PLT. | |
175 | .macro callext addr, cond= | |
176 | #if WANT_PIC | |
177 | bl\cond \addr(PLT) | |
178 | #else | |
179 | bl\cond \addr | |
180 | #endif | |
181 | .endm | |
182 | ||
183 | // Do I need to arrange a spare GOT register? | |
184 | #if WANT_PIC | |
185 | # define NEED_GOT 1 | |
186 | #endif | |
187 | #define GOTREG r9 | |
188 | ||
189 | // Maybe load GOT address into GOT. | |
2d03a881 | 190 | .macro ldgot cond=, got=GOTREG |
61bd904b | 191 | #if WANT_PIC |
adca2a18 MW |
192 | ldr\cond \got, .L$_ldgot$\@ |
193 | .L$_ldgot_pc$\@: | |
2d03a881 | 194 | add\cond \got, pc, \got |
adca2a18 MW |
195 | _LIT |
196 | .balign 4 | |
197 | .L$_ldgot$\@: | |
9f6eb05d | 198 | .word _GLOBAL_OFFSET_TABLE_ - .L$_ldgot_pc$\@ - .L$_pcoff |
adca2a18 | 199 | _ENDLIT |
61bd904b MW |
200 | #endif |
201 | .endm | |
202 | ||
203 | // Load address of external symbol ADDR into REG, maybe using GOT. | |
204 | .macro leaext reg, addr, cond=, got=GOTREG | |
205 | #if WANT_PIC | |
adca2a18 | 206 | ldr\cond \reg, .L$_leaext$\@ |
2d03a881 | 207 | ldr\cond \reg, [\got, \reg] |
adca2a18 MW |
208 | _LIT |
209 | .balign 4 | |
210 | .L$_leaext$\@: | |
211 | .word \addr(GOT) | |
212 | _ENDLIT | |
61bd904b | 213 | #else |
2d03a881 | 214 | ldr\cond \reg, =\addr |
61bd904b MW |
215 | #endif |
216 | .endm | |
217 | ||
0c53ac58 MW |
218 | // Load address of external symbol ADDR into REG directly. |
219 | .macro leaextq reg, addr, cond= | |
220 | #if WANT_PIC | |
221 | ldr\cond \reg, .L$_leaextq$\@ | |
222 | .L$_leaextq_pc$\@: | |
9f6eb05d | 223 | .if .L$_pcoff == 8 |
0c53ac58 | 224 | ldr\cond \reg, [pc, \reg] |
9f6eb05d MW |
225 | .else |
226 | add\cond \reg, pc | |
227 | ldr\cond \reg, [\reg] | |
228 | .endif | |
0c53ac58 MW |
229 | _LIT |
230 | .balign 4 | |
231 | .L$_leaextq$\@: | |
9f6eb05d | 232 | .word \addr(GOT_PREL) + (. - .L$_leaextq_pc$\@ - .L$_pcoff) |
0c53ac58 MW |
233 | _ENDLIT |
234 | #else | |
235 | ldr\cond \reg, =\addr | |
236 | #endif | |
237 | .endm | |
238 | ||
61bd904b MW |
239 | #endif |
240 | ||
1a0c09c4 MW |
241 | ///-------------------------------------------------------------------------- |
242 | /// Final stuff. | |
243 | ||
244 | // Default values for the various hooks. | |
245 | #ifndef FUNC_PREHOOK | |
246 | # define FUNC_PREHOOK(name) | |
247 | #endif | |
248 | #ifndef FUNC_POSTHOOK | |
249 | # define FUNC_POSTHOOK(name) | |
250 | #endif | |
251 | #ifndef ENDFUNC_HOOK | |
252 | # define ENDFUNC_HOOK(name) | |
253 | #endif | |
254 | ||
255 | #ifndef F | |
256 | # define F(name) name | |
257 | #endif | |
258 | ||
259 | #ifndef TYPE_FUNC | |
260 | # define TYPE_FUNC(name) | |
261 | #endif | |
262 | ||
263 | #ifndef SIZE_OBJ | |
264 | # define SIZE_OBJ(name) | |
265 | #endif | |
266 | ||
267 | ///----- That's all, folks -------------------------------------------------- |