1 | /* Block Addition, Subtraction and Shifts with Carry |
---|
2 | Copyright (C) 2010, Robert D. Cameron |
---|
3 | Licensed to the public under the Open Software License 3.0. |
---|
4 | Licensed to International Characters Inc. |
---|
5 | under the Academic Free License version 3.0. |
---|
6 | |
---|
7 | This file defines addition, subtract and shift operations on |
---|
8 | 128-bit blocks. Different versions of the operations are |
---|
9 | selectable with the CARRY_STRATEGY preprocessor constant. |
---|
10 | |
---|
11 | Each implementation defines the following "abstract data type" |
---|
12 | for block operations with carry. |
---|
13 | |
---|
14 | Typename: CarryType |
---|
15 | Constant: Carry0 represents a value of 0 for the carry bit. |
---|
16 | Predicate: test_carry(x) returns nonzero if a carry bit is 1, 0 otherwise. |
---|
17 | Function: carry_or(carry1, carry2) forms the logical or of two carries. |
---|
18 | Function: adc128(x, y, carry, sum) computes (carry, sum) = x + y + carry, |
---|
19 | Function: advance_with_carry(cursor, carry, rslt) |
---|
20 | computes (carry, rslt) = cursor + cursor + carry |
---|
21 | Function: sbb128(x, y, borrow, diff) |
---|
22 | computes (borrow, diff) = y - x - borrow |
---|
23 | |
---|
24 | */ |
---|
25 | #ifndef BLOCK_CARRY_H |
---|
26 | #define BLOCK_CARRY_H |
---|
27 | |
---|
28 | |
---|
29 | |
---|
30 | |
---|
31 | |
---|
32 | /*------------------------------------------------------------*/ |
---|
33 | #include "sse_simd.h" |
---|
34 | |
---|
35 | #define SIMD_CARRY_STRATEGY 1 |
---|
36 | #define ADC64_STRATEGY 2 |
---|
37 | #define ADC64_SAHF_STRATEGY 3 |
---|
38 | |
---|
39 | #ifdef ADC64 |
---|
40 | #ifdef SAHFLAHF |
---|
41 | #define CARRY_STRATEGY ADC64_SAHF_STRATEGY |
---|
42 | #else |
---|
43 | #define CARRY_STRATEGY ADC64_STRATEGY |
---|
44 | #endif |
---|
45 | #else |
---|
46 | #define CARRY_STRATEGY SIMD_CARRY_STRATEGY |
---|
47 | #endif |
---|
48 | |
---|
49 | #if (CARRY_STRATEGY == ADC64_STRATEGY) |
---|
50 | typedef uint64_t CarryType; |
---|
51 | |
---|
52 | #define Carry0 0 |
---|
53 | |
---|
54 | #define test_carry(x) ((x) > 0) |
---|
55 | |
---|
56 | #define carry_or(carry1, carry2) (carry1 | carry2) |
---|
57 | |
---|
58 | #define clc() \ |
---|
59 | __asm__ __volatile__ ("clc\n\t":::) |
---|
60 | |
---|
61 | #define adc(x,y,carry,sum) \ |
---|
62 | __asm__ __volatile__ ("add %[carryflag], %[e]\n\t" \ |
---|
63 | "adc %[e], %[z]\n\t" \ |
---|
64 | "mov $0, %1\n\t" \ |
---|
65 | "adc %[carryflag], %1\n\t" \ |
---|
66 | : [z] "=r" (sum), [carryflag] "=a" (carry) \ |
---|
67 | : "[z]" (x), [e] "r" (y), "[carryflag]" (carry) \ |
---|
68 | : "cc") |
---|
69 | |
---|
70 | #define double_int64_adc(x1, x2, y1, y2, rslt1, rslt2, carry) \ |
---|
71 | __asm__ __volatile__ ("add %[carryin], %[z1]\n\t" \ |
---|
72 | "adc %[e1], %[z1]\n\t" \ |
---|
73 | "adc %[e2], %[z2]\n\t" \ |
---|
74 | "mov $0, %[carryout]\n\t" \ |
---|
75 | "adc $0, %[carryout]\n\t" \ |
---|
76 | : [z1] "=a" (rslt1), [z2] "=r" (rslt2), [carryout] "=r" (carry) \ |
---|
77 | : "0" (y1), "1" (y2), [carryin] "2" (carry), \ |
---|
78 | [e1] "b" (x1), [e2] "r" (x2)\ |
---|
79 | : "cc") |
---|
80 | |
---|
81 | #define adc128(first, second, carry, sum) \ |
---|
82 | do\ |
---|
83 | {\ |
---|
84 | union {__m128i bitblock;\ |
---|
85 | uint64_t int64[2];} rslt;\ |
---|
86 | \ |
---|
87 | union {__m128i bitblock;\ |
---|
88 | uint64_t int64[2];} x;\ |
---|
89 | \ |
---|
90 | union {__m128i bitblock;\ |
---|
91 | uint64_t int64[2];} y;\ |
---|
92 | \ |
---|
93 | x.bitblock = first;\ |
---|
94 | y.bitblock = second;\ |
---|
95 | \ |
---|
96 | double_int64_adc(x.int64[0], x.int64[1], y.int64[0], y.int64[1], rslt.int64[0], rslt.int64[1], carry);\ |
---|
97 | sum = rslt.bitblock;\ |
---|
98 | }while(0) |
---|
99 | |
---|
100 | |
---|
101 | |
---|
102 | #define advance_with_carry(cursor, carry, rslt)\ |
---|
103 | adc128(cursor, cursor, carry, rslt) |
---|
104 | |
---|
105 | |
---|
106 | #define double_int64_sbb(x1, x2, y1, y2, rslt1, rslt2, carry) \ |
---|
107 | __asm__ ("sahf\n\t" \ |
---|
108 | "sbb %[e1], %[z1]\n\t" \ |
---|
109 | "sbb %[e2], %[z2]\n\t" \ |
---|
110 | "lahf\n\t" \ |
---|
111 | : [z1] "=r" (rslt1), [z2] "=r" (rslt2), [carryflag] "=a" (carry) \ |
---|
112 | : "[z1]" (x1), "[z2]" (x2), \ |
---|
113 | [e1] "r" (y1), [e2] "r" (y2), \ |
---|
114 | "[carryflag]" (carry) \ |
---|
115 | : "cc") |
---|
116 | |
---|
117 | #define sbb128(first, second, borrow, diff) \ |
---|
118 | do\ |
---|
119 | { union {__m128i bitblock;\ |
---|
120 | uint64_t int64[2];} rslt;\ |
---|
121 | \ |
---|
122 | union {__m128i bitblock;\ |
---|
123 | uint64_t int64[2];} x;\ |
---|
124 | \ |
---|
125 | union {__m128i bitblock;\ |
---|
126 | uint64_t int64[2];} y;\ |
---|
127 | \ |
---|
128 | x.bitblock = first;\ |
---|
129 | y.bitblock = second;\ |
---|
130 | \ |
---|
131 | double_int64_sbb(x.int64[0], x.int64[1], y.int64[0], y.int64[1], \ |
---|
132 | rslt.int64[0], rslt.int64[1], borrow);\ |
---|
133 | diff = rslt.bitblock;\ |
---|
134 | }while(0) |
---|
135 | |
---|
136 | #endif |
---|
137 | |
---|
138 | #if (CARRY_STRATEGY == ADC64_SAHF_STRATEGY) |
---|
139 | typedef uint64_t CarryType; |
---|
140 | |
---|
141 | #define Carry0 0 |
---|
142 | |
---|
143 | #define test_carry(x) (((x)&256) > 0) |
---|
144 | |
---|
145 | #define carry_or(carry1, carry2) (carry1 | carry2) |
---|
146 | |
---|
147 | #define double_int64_adc(x1, x2, y1, y2, rslt1, rslt2, carry) \ |
---|
148 | __asm__ ("sahf\n\t" \ |
---|
149 | "adc %[e1], %[z1]\n\t" \ |
---|
150 | "adc %[e2], %[z2]\n\t" \ |
---|
151 | "lahf\n\t" \ |
---|
152 | : [z1] "=r" (rslt1), [z2] "=r" (rslt2), [carryflag] "=a" (carry) \ |
---|
153 | : "[z1]" (x1), "[z2]" (x2), \ |
---|
154 | [e1] "r" (y1), [e2] "r" (y2), \ |
---|
155 | "[carryflag]" (carry) \ |
---|
156 | : "cc") |
---|
157 | |
---|
158 | #define adc128(first, second, carry, sum) \ |
---|
159 | do\ |
---|
160 | {\ |
---|
161 | union {__m128i bitblock;\ |
---|
162 | uint64_t int64[2];} rslt;\ |
---|
163 | \ |
---|
164 | union {__m128i bitblock;\ |
---|
165 | uint64_t int64[2];} x;\ |
---|
166 | \ |
---|
167 | union {__m128i bitblock;\ |
---|
168 | uint64_t int64[2];} y;\ |
---|
169 | \ |
---|
170 | x.bitblock = first;\ |
---|
171 | y.bitblock = second;\ |
---|
172 | \ |
---|
173 | double_int64_adc(x.int64[0], x.int64[1], y.int64[0], y.int64[1], rslt.int64[0], rslt.int64[1], carry);\ |
---|
174 | sum = rslt.bitblock;\ |
---|
175 | }while(0) |
---|
176 | |
---|
177 | |
---|
178 | |
---|
179 | #define double_int64_advance(x1, x2, rslt1, rslt2, carry) \ |
---|
180 | __asm__ ("sahf\n\t" \ |
---|
181 | "adc %[z1], %[z1]\n\t" \ |
---|
182 | "adc %[z2], %[z2]\n\t" \ |
---|
183 | "lahf\n\t" \ |
---|
184 | : [z1] "=r" (rslt1), [z2] "=r" (rslt2), [carryflag] "=a" (carry) \ |
---|
185 | : "[z1]" (x1), "[z2]" (x2), \ |
---|
186 | "[carryflag]" (carry) \ |
---|
187 | : "cc") |
---|
188 | |
---|
189 | |
---|
190 | #define advance_with_carry(cursor, carry, rslt)\ |
---|
191 | do\ |
---|
192 | {\ |
---|
193 | union {__m128i bitblock;\ |
---|
194 | uint64_t int64[2];} z;\ |
---|
195 | \ |
---|
196 | union {__m128i bitblock;\ |
---|
197 | uint64_t int64[2];} x;\ |
---|
198 | \ |
---|
199 | x.bitblock = cursor;\ |
---|
200 | \ |
---|
201 | double_int64_advance(x.int64[0], x.int64[1], z.int64[0], z.int64[1], carry);\ |
---|
202 | rslt = z.bitblock;\ |
---|
203 | }while(0) |
---|
204 | |
---|
205 | |
---|
206 | |
---|
207 | |
---|
208 | #define double_int64_sbb(x1, x2, y1, y2, rslt1, rslt2, carry) \ |
---|
209 | __asm__ ("sahf\n\t" \ |
---|
210 | "sbb %[e1], %[z1]\n\t" \ |
---|
211 | "sbb %[e2], %[z2]\n\t" \ |
---|
212 | "lahf\n\t" \ |
---|
213 | : [z1] "=r" (rslt1), [z2] "=r" (rslt2), [carryflag] "=a" (carry) \ |
---|
214 | : "[z1]" (x1), "[z2]" (x2), \ |
---|
215 | [e1] "r" (y1), [e2] "r" (y2), \ |
---|
216 | "[carryflag]" (carry) \ |
---|
217 | : "cc") |
---|
218 | |
---|
219 | #define sbb128(first, second, borrow, diff) \ |
---|
220 | do\ |
---|
221 | { union {__m128i bitblock;\ |
---|
222 | uint64_t int64[2];} rslt;\ |
---|
223 | \ |
---|
224 | union {__m128i bitblock;\ |
---|
225 | uint64_t int64[2];} x;\ |
---|
226 | \ |
---|
227 | union {__m128i bitblock;\ |
---|
228 | uint64_t int64[2];} y;\ |
---|
229 | \ |
---|
230 | x.bitblock = first;\ |
---|
231 | y.bitblock = second;\ |
---|
232 | \ |
---|
233 | double_int64_sbb(x.int64[0], x.int64[1], y.int64[0], y.int64[1], \ |
---|
234 | rslt.int64[0], rslt.int64[1], borrow);\ |
---|
235 | diff = rslt.bitblock;\ |
---|
236 | }while(0) |
---|
237 | |
---|
238 | #endif |
---|
239 | |
---|
240 | |
---|
241 | |
---|
242 | #if (CARRY_STRATEGY == SIMD_CARRY_STRATEGY) |
---|
243 | |
---|
244 | typedef SIMD_type CarryType; |
---|
245 | |
---|
246 | #define Carry0 simd_const_1(0) |
---|
247 | |
---|
248 | #define test_carry(x) bitblock_has_bit(x) |
---|
249 | |
---|
250 | #define carry_or(carry1, carry2) simd_or(carry1, carry2) |
---|
251 | |
---|
252 | #define adc128(x, y, carry, sum) \ |
---|
253 | do{ \ |
---|
254 | SIMD_type gen = simd_and(x, y); \ |
---|
255 | SIMD_type prop = simd_or(x, y); \ |
---|
256 | SIMD_type partial = simd_add_64(simd_add_64(x, y), carry); \ |
---|
257 | SIMD_type c1 = sisd_slli(simd_srli_64(simd_or(gen, simd_andc(prop, partial)), 63), 64); \ |
---|
258 | sum = simd_add_64(c1, partial); \ |
---|
259 | carry = sisd_srli(simd_or(gen, simd_andc(prop, sum)), 127); \ |
---|
260 | } while(0) |
---|
261 | |
---|
262 | |
---|
263 | #define sbb128(x, y, borrow, difference) \ |
---|
264 | do {\ |
---|
265 | SIMD_type gen = simd_andc(y, x); \ |
---|
266 | SIMD_type prop = simd_not(simd_xor(x, y)); \ |
---|
267 | SIMD_type partial = simd_sub_64(simd_sub_64(x, y), borrow); \ |
---|
268 | SIMD_type b1 = sisd_slli(simd_srli_64(simd_or(gen, simd_and(prop, partial)), 63), 64); \ |
---|
269 | difference = simd_sub_64(partial, b1); \ |
---|
270 | borrow = sisd_srli(simd_or(gen, simd_and(prop, difference)), 127); \ |
---|
271 | }while(0) |
---|
272 | |
---|
273 | |
---|
274 | #define advance_with_carry(cursor, carry, rslt)\ |
---|
275 | do{\ |
---|
276 | SIMD_type shift_out = simd_srli_64(cursor, 63);\ |
---|
277 | SIMD_type low_bits = simd_mergel_64(shift_out, carry);\ |
---|
278 | carry = sisd_srli(shift_out, 64);\ |
---|
279 | rslt = simd_or(simd_add_64(cursor, cursor), low_bits);\ |
---|
280 | }while(0) |
---|
281 | |
---|
282 | #endif |
---|
283 | #endif |
---|
284 | |
---|
285 | |
---|