1 /* Bra86.c -- Branch converter for X86 code (BCJ)
2 2023-04-02 : Igor Pavlov : Public domain */
3
4 #include "Precomp.h"
5
6 #include "Bra.h"
7 #include "CpuArch.h"
8
9
10 #if defined(MY_CPU_SIZEOF_POINTER) \
11 && ( MY_CPU_SIZEOF_POINTER == 4 \
12 || MY_CPU_SIZEOF_POINTER == 8)
13 #define BR_CONV_USE_OPT_PC_PTR
14 #endif
15
16 #ifdef BR_CONV_USE_OPT_PC_PTR
17 #define BR_PC_INIT pc -= (UInt32)(SizeT)p; // (MY_uintptr_t)
18 #define BR_PC_GET (pc + (UInt32)(SizeT)p)
19 #else
20 #define BR_PC_INIT pc += (UInt32)size;
21 #define BR_PC_GET (pc - (UInt32)(SizeT)(lim - p))
22 // #define BR_PC_INIT
23 // #define BR_PC_GET (pc + (UInt32)(SizeT)(p - data))
24 #endif
25
26 #define BR_CONVERT_VAL(v, c) if (encoding) v += c; else v -= c;
27 // #define BR_CONVERT_VAL(v, c) if (!encoding) c = (UInt32)0 - c; v += c;
28
29 #define Z7_BRANCH_CONV_ST(name) z7_BranchConvSt_ ## name
30
31 #define BR86_NEED_CONV_FOR_MS_BYTE(b) ((((b) + 1) & 0xfe) == 0)
32
33 #ifdef MY_CPU_LE_UNALIGN
34 #define BR86_PREPARE_BCJ_SCAN const UInt32 v = GetUi32(p) ^ 0xe8e8e8e8;
35 #define BR86_IS_BCJ_BYTE(n) ((v & ((UInt32)0xfe << (n) * 8)) == 0)
36 #else
37 #define BR86_PREPARE_BCJ_SCAN
38 // bad for MSVC X86 (partial write to byte reg):
39 #define BR86_IS_BCJ_BYTE(n) ((p[n - 4] & 0xfe) == 0xe8)
40 // bad for old MSVC (partial write to byte reg):
41 // #define BR86_IS_BCJ_BYTE(n) (((*p ^ 0xe8) & 0xfe) == 0)
42 #endif
43
44 static
45 Z7_FORCE_INLINE
46 Z7_ATTRIB_NO_VECTOR
Z7_BRANCH_CONV_ST(X86)47 Byte *Z7_BRANCH_CONV_ST(X86)(Byte *p, SizeT size, UInt32 pc, UInt32 *state, int encoding)
48 {
49 if (size < 5)
50 return p;
51 {
52 // Byte *p = data;
53 const Byte *lim = p + size - 4;
54 unsigned mask = (unsigned)*state; // & 7;
55 #ifdef BR_CONV_USE_OPT_PC_PTR
56 /* if BR_CONV_USE_OPT_PC_PTR is defined: we need to adjust (pc) for (+4),
57 because call/jump offset is relative to the next instruction.
58 if BR_CONV_USE_OPT_PC_PTR is not defined : we don't need to adjust (pc) for (+4),
59 because BR_PC_GET uses (pc - (lim - p)), and lim was adjusted for (-4) before.
60 */
61 pc += 4;
62 #endif
63 BR_PC_INIT
64 goto start;
65
66 for (;; mask |= 4)
67 {
68 // cont: mask |= 4;
69 start:
70 if (p >= lim)
71 goto fin;
72 {
73 BR86_PREPARE_BCJ_SCAN
74 p += 4;
75 if (BR86_IS_BCJ_BYTE(0)) { goto m0; } mask >>= 1;
76 if (BR86_IS_BCJ_BYTE(1)) { goto m1; } mask >>= 1;
77 if (BR86_IS_BCJ_BYTE(2)) { goto m2; } mask = 0;
78 if (BR86_IS_BCJ_BYTE(3)) { goto a3; }
79 }
80 goto main_loop;
81
82 m0: p--;
83 m1: p--;
84 m2: p--;
85 if (mask == 0)
86 goto a3;
87 if (p > lim)
88 goto fin_p;
89
90 // if (((0x17u >> mask) & 1) == 0)
91 if (mask > 4 || mask == 3)
92 {
93 mask >>= 1;
94 continue; // goto cont;
95 }
96 mask >>= 1;
97 if (BR86_NEED_CONV_FOR_MS_BYTE(p[mask]))
98 continue; // goto cont;
99 // if (!BR86_NEED_CONV_FOR_MS_BYTE(p[3])) continue; // goto cont;
100 {
101 UInt32 v = GetUi32(p);
102 UInt32 c;
103 v += (1 << 24); if (v & 0xfe000000) continue; // goto cont;
104 c = BR_PC_GET;
105 BR_CONVERT_VAL(v, c)
106 {
107 mask <<= 3;
108 if (BR86_NEED_CONV_FOR_MS_BYTE(v >> mask))
109 {
110 v ^= (((UInt32)0x100 << mask) - 1);
111 #ifdef MY_CPU_X86
112 // for X86 : we can recalculate (c) to reduce register pressure
113 c = BR_PC_GET;
114 #endif
115 BR_CONVERT_VAL(v, c)
116 }
117 mask = 0;
118 }
119 // v = (v & ((1 << 24) - 1)) - (v & (1 << 24));
120 v &= (1 << 25) - 1; v -= (1 << 24);
121 SetUi32(p, v)
122 p += 4;
123 goto main_loop;
124 }
125
126 main_loop:
127 if (p >= lim)
128 goto fin;
129 for (;;)
130 {
131 BR86_PREPARE_BCJ_SCAN
132 p += 4;
133 if (BR86_IS_BCJ_BYTE(0)) { goto a0; }
134 if (BR86_IS_BCJ_BYTE(1)) { goto a1; }
135 if (BR86_IS_BCJ_BYTE(2)) { goto a2; }
136 if (BR86_IS_BCJ_BYTE(3)) { goto a3; }
137 if (p >= lim)
138 goto fin;
139 }
140
141 a0: p--;
142 a1: p--;
143 a2: p--;
144 a3:
145 if (p > lim)
146 goto fin_p;
147 // if (!BR86_NEED_CONV_FOR_MS_BYTE(p[3])) continue; // goto cont;
148 {
149 UInt32 v = GetUi32(p);
150 UInt32 c;
151 v += (1 << 24); if (v & 0xfe000000) continue; // goto cont;
152 c = BR_PC_GET;
153 BR_CONVERT_VAL(v, c)
154 // v = (v & ((1 << 24) - 1)) - (v & (1 << 24));
155 v &= (1 << 25) - 1; v -= (1 << 24);
156 SetUi32(p, v)
157 p += 4;
158 goto main_loop;
159 }
160 }
161
162 fin_p:
163 p--;
164 fin:
165 // the following processing for tail is optional and can be commented
166 /*
167 lim += 4;
168 for (; p < lim; p++, mask >>= 1)
169 if ((*p & 0xfe) == 0xe8)
170 break;
171 */
172 *state = (UInt32)mask;
173 return p;
174 }
175 }
176
177
178 #define Z7_BRANCH_CONV_ST_FUNC_IMP(name, m, encoding) \
179 Z7_NO_INLINE \
180 Z7_ATTRIB_NO_VECTOR \
181 Byte *m(name)(Byte *data, SizeT size, UInt32 pc, UInt32 *state) \
182 { return Z7_BRANCH_CONV_ST(name)(data, size, pc, state, encoding); }
183
184 Z7_BRANCH_CONV_ST_FUNC_IMP(X86, Z7_BRANCH_CONV_ST_DEC, 0)
185 #ifndef Z7_EXTRACT_ONLY
186 Z7_BRANCH_CONV_ST_FUNC_IMP(X86, Z7_BRANCH_CONV_ST_ENC, 1)
187 #endif
188