-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathAVX512_Saturated_AddSub.cpp
352 lines (278 loc) · 12.7 KB
/
AVX512_Saturated_AddSub.cpp
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
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
#include "stdafx.h"
#include "AVX512_Saturated_AddSub.h"
extern CPU_Props cpu_props;
using namespace std;
/*
The saturated addition and substraction are supported only for 8- and 16-bit long data types in Intel SSE/AVX ISA:
- _mm512_adds_epi8/epu8
- _mm512_subs_epi8/epu8
- _mm512_adds_epi16/epu16
- _mm512_subs_epi16/epu16
This proposal is for the missing 32- and 64-bit versions:
- _mm512_adds_epi32/epu32
- _mm512_subs_epi32/epu32
- _mm512_adds_epi64/epu64
- _mm512_subs_epi64/epu64
*/
//Inspiration:
//https://twitter.com/Wunkolo/status/1579894616593076224
//credits:
//https://stackoverflow.com/questions/29498824/add-saturate-32-bit-signed-ints-intrinsics
//https://stackoverflow.com/questions/56526082/is-there-a-way-to-subtract-packed-unsigned-doublewords-saturated-on-x86-using
//https://godbolt.org/z/3o5KCm
//VPTERNLOGx imm8 Calculator:
//https://www.officedaytime.com/simd512e/simdimg/ternlogcalc.html
//c ? a : b
//imm8 = 0xE4
//
//a 1 1 1 1 0 0 0 0
//b 1 1 0 0 1 1 0 0
//c 1 0 1 0 1 0 1 0
//c ? a : b 1 1 1 0 0 1 0 0
// A signed addition overflow will happen if, and only if:
// - the signs of both inputs are the same, and
// - the sign of the sum (when added with wrap-around) is different from the input:
// sign(a) = sign(b), sign(a + b) != sign(a)
//eq(a, b) & neq(a, c)
//imm8 = 0x42
//a 1 1 1 1 0 0 0 0
//b 1 1 0 0 1 1 0 0
//a + b 1 0 1 0 1 0 1 0
//eq(a, b) & neq(a, a + b) 0 1 0 0 0 0 1 0
__m512i _mm512_adds_epi32(__m512i a, __m512i b) {
__m512i add = _mm512_add_epi32(a, b);
__m512i sign = _mm512_srai_epi32(add, 31);
__m512i of = _mm512_ternarylogic_epi32(a, b, add, 0x42);
__m512i ofmask = _mm512_srai_epi32(of, 31);
__m512i ofvalue = _mm512_xor_si512(_mm512_set1_epi32(LONG_MIN), sign);
__m512i val = _mm512_ternarylogic_epi32(ofvalue, add, ofmask, 0xe4);
return val;
}
__m512i _mm512_adds_epi64(__m512i a, __m512i b) {
__m512i add = _mm512_add_epi64(a, b);
__m512i sign = _mm512_srai_epi64(add, 63);
__m512i of = _mm512_ternarylogic_epi64(a, b, add, 0x42);
__m512i ofmask = _mm512_srai_epi64(of, 63);
__m512i ofvalue = _mm512_xor_si512(_mm512_set1_epi64(LLONG_MIN), sign);
__m512i val = _mm512_ternarylogic_epi64(ofvalue, add, ofmask, 0xe4);
return val;
}
// A signed substraction overflow will happen if, and only if:
// - if the signs of the two inputs are different, and
// - the sign of the difference (when substracted with wrap-around) is different from the minuend:
// sign(a) != sign(b), sign(a - b) != sign(a)
//imm8 = 0x18
//a 1 1 1 1 0 0 0 0
//b 1 1 0 0 1 1 0 0
//a - b 1 0 1 0 1 0 1 0
//neq(a, b) & neq(a - b, a) 0 0 0 1 1 0 0 0
__m512i _mm512_subs_epi32(__m512i a, __m512i b) {
__m512i sub = _mm512_sub_epi32(a, b);
__m512i sign = _mm512_srai_epi32(sub, 31);
__m512i of = _mm512_ternarylogic_epi32(a, b, sub, 0x18);
__m512i ofmask = _mm512_srai_epi32(of, 31);
__m512i ofvalue = _mm512_xor_si512(_mm512_set1_epi32(LONG_MIN), sign);
__m512i val = _mm512_ternarylogic_epi32(ofvalue, sub, ofmask, 0xe4);
return val;
}
__m512i _mm512_subs_epi64(__m512i a, __m512i b) {
__m512i sub = _mm512_sub_epi64(a, b);
__m512i sign = _mm512_srai_epi64(sub, 63);
__m512i of = _mm512_ternarylogic_epi64(a, b, sub, 0x18);
__m512i ofmask = _mm512_srai_epi64(of, 63);
__m512i ofvalue = _mm512_xor_si512(_mm512_set1_epi64(LLONG_MIN), sign);
__m512i val = _mm512_ternarylogic_epi64(ofvalue, sub, ofmask, 0xe4);
return val;
}
//~a
//imm8 = 0x0F
//
//a 1 1 1 1 0 0 0 0
//b 1 1 0 0 1 1 0 0
//c 1 0 1 0 1 0 1 0
//~a 0 0 0 0 1 1 1 1
__m512i _mm512_adds_epu32(__m512i a, __m512i b) {
return _mm512_add_epi32(_mm512_min_epu32(a, _mm512_ternarylogic_epi32(b, b, b, 0x0f)), b);
}
__m512i _mm512_adds_epu64(__m512i a, __m512i b) {
return _mm512_add_epi64(_mm512_min_epu64(a, _mm512_ternarylogic_epi64(b, b, b, 0x0f)), b);
}
__m512i _mm512_subs_epu32(__m512i a, __m512i b) {
return _mm512_sub_epi32(_mm512_max_epu32(a, b), b);
}
__m512i _mm512_subs_epu64(__m512i a, __m512i b) {
return _mm512_sub_epi64(_mm512_max_epu64(a, b), b);
}
//the Zen4 versions consider the peculiarities of TERNLOG handling on AMD CPUs:
//https://uops.info/html-lat/ZEN4/VPTERNLOGD_ZMM_ZMM_ZMM_I8-Measurements.html
//Instruction: VPTERNLOGD ZMM0, ZMM1, ZMM2, 2
//Code:
// vpternlogd zmm0,zmm1,zmm2,0x2
// vandpd zmm0,zmm0,zmm0
// vandpd zmm0,zmm0,zmm0
// vandpd zmm0,zmm0,zmm0
// vandpd zmm0,zmm0,zmm0
// vandpd zmm0,zmm0,zmm0
//Results:
//MPERF: 5.17
//APERF: 6.0 -> 1 clk
//
//Instruction: VPTERNLOGD ZMM1, ZMM0, ZMM2, 2
//Code:
// vpternlogd zmm1,zmm0,zmm2,0x2
// vandpd zmm0,zmm1,zmm1
// vandpd zmm0,zmm0,zmm0
// vandpd zmm0,zmm0,zmm0
// vandpd zmm0,zmm0,zmm0
// vandpd zmm0,zmm0,zmm0
//Results:
//MPERF: 6.11
//APERF: 7.06 -> 2 clks
//
//Instruction: VPTERNLOGD ZMM1, ZMM2, ZMM0, 2
//Code:
// vpternlogd zmm1,zmm2,zmm0,0x2
// vandpd zmm0,zmm1,zmm1
// vandpd zmm0,zmm0,zmm0
// vandpd zmm0,zmm0,zmm0
// vandpd zmm0,zmm0,zmm0
// vandpd zmm0,zmm0,zmm0
//Results:
//MPERF: 6.19
//APERF: 7.0 -> 2 clks
//a ? b : c
//imm8 = 0xCA
//a 1 1 1 1 0 0 0 0
//b 1 1 0 0 1 1 0 0
//c 1 0 1 0 1 0 1 0
//a ? b : c 1 1 0 0 1 0 1 0
// A signed addition overflow will happen if, and only if:
// - if the signs of both inputs are the same, and
// - the sign of the sum (when added with wrap-around) is different from the input:
// sign(a) = sign(b), sign(a + b) != sign(a)
//imm8 = 0x18
//a + b 1 1 1 1 0 0 0 0
//a 1 1 0 0 1 1 0 0
//b 1 0 1 0 1 0 1 0
//eq(a, b) & neq(a+b, a) 0 0 0 1 1 0 0 0
__m512i _mm512_adds_Zen4_epi32(__m512i a, __m512i b) {
__m512i add = _mm512_add_epi32(a, b);
__m512i sign = _mm512_srai_epi32(add, 31);
__m512i of = _mm512_ternarylogic_epi32(add, a, b, 0x18);
__m512i ofmask = _mm512_srai_epi32(of, 31);
__m512i ofvalue = _mm512_xor_si512(sign, _mm512_set1_epi32(LONG_MIN));
__m512i val = _mm512_ternarylogic_epi32(ofmask, ofvalue, add, 0xca);
return val;
}
__m512i _mm512_adds_Zen4_epi64(__m512i a, __m512i b) {
__m512i add = _mm512_add_epi64(a, b);
__m512i sign = _mm512_srai_epi64(add, 63);
__m512i of = _mm512_ternarylogic_epi64(add, a, b, 0x18);
__m512i ofmask = _mm512_srai_epi64(of, 63);
__m512i ofvalue = _mm512_xor_si512(sign, _mm512_set1_epi64(LLONG_MIN));
__m512i val = _mm512_ternarylogic_epi64(ofmask, ofvalue, add, 0xca);
return val;
}
// A signed substraction overflow will happen if, and only if:
// - the signs of the two inputs are different, and
// - the sign of the difference (when substracted with wrap-around) is different from the minuend:
// sign(a) != sign(b), sign(a - b) != sign(a)
//imm8 = 0x24
//a - b 1 1 1 1 0 0 0 0
//a 1 1 0 0 1 1 0 0
//b 1 0 1 0 1 0 1 0
//neq(a, b) & neq(a - b, a) 0 0 1 0 0 1 0 0
__m512i _mm512_subs_Zen4_epi32(__m512i a, __m512i b) {
__m512i sub = _mm512_sub_epi32(a, b);
__m512i sign = _mm512_srai_epi32(sub, 31);
__m512i of = _mm512_ternarylogic_epi32(sub, a, b, 0x24);
__m512i ofmask = _mm512_srai_epi32(of, 31);
__m512i ofvalue = _mm512_xor_si512(sign, _mm512_set1_epi32(LONG_MIN));
__m512i val = _mm512_ternarylogic_epi32(ofmask, ofvalue, sub, 0xca);
return val;
}
__m512i _mm512_subs_Zen4_epi64(__m512i a, __m512i b) {
__m512i sub = _mm512_sub_epi64(a, b);
__m512i sign = _mm512_srai_epi64(sub, 63);
__m512i of = _mm512_ternarylogic_epi64(sub, a, b, 0x24);
__m512i ofmask = _mm512_srai_epi64(of, 63);
__m512i ofvalue = _mm512_xor_si512(sign, _mm512_set1_epi64(LLONG_MIN));
__m512i val = _mm512_ternarylogic_epi64(ofmask, ofvalue, sub, 0xca);
return val;
}
//~a
//imm8 = 0x0F
//
//a 1 1 1 1 0 0 0 0
//b 1 1 0 0 1 1 0 0
//c 1 0 1 0 1 0 1 0
//~a 0 0 0 0 1 1 1 1
__m512i _mm512_adds_Zen4_epu32(__m512i a, __m512i b) {
__m512i u = _mm512_undefined_epi32();
return _mm512_add_epi32(_mm512_min_epu32(a, _mm512_ternarylogic_epi32(b, u, u, 0x0f)), b);
}
__m512i _mm512_adds_Zen4_epu64(__m512i a, __m512i b) {
__m512i u = _mm512_undefined_epi32(); // there isn't _mm512_undefined_epi64();
return _mm512_add_epi64(_mm512_min_epu64(a, _mm512_ternarylogic_epi64(b, u, u, 0x0f)), b);
}
__m512i _mm512_subs_Zen4_epu32(__m512i a, __m512i b) {
return _mm512_sub_epi32(_mm512_max_epu32(a, b), b);
}
__m512i _mm512_subs_Zen4_epu64(__m512i a, __m512i b) {
return _mm512_sub_epi64(_mm512_max_epu64(a, b), b);
}
void AVX512_Saturated_AddSub_Demo(void) {
uint16_t x16 = __rdtsc() & 0xffff;
uint32_t x32 = x16 * 0x10001;
uint64_t x64 = x16 * 0x1000100010001;
__m256i testcases_a16 = _mm256_set_epi16(0, 0, 0, 0, x16, x16, x16, x16, SHRT_MIN, SHRT_MIN, SHRT_MIN, SHRT_MIN, SHRT_MAX, SHRT_MAX, SHRT_MAX, SHRT_MAX);
__m256i testcases_b16 = _mm256_set_epi16(0, x16, SHRT_MIN, SHRT_MAX, 0, x16, SHRT_MIN, SHRT_MAX, 0, x16, SHRT_MIN, SHRT_MAX, 0, x16, SHRT_MIN, SHRT_MAX);
printRes16("Testcases_a16 :", testcases_a16);
printRes16("Testcases_b16 :", testcases_b16);
printRes16("Saturated signed add epi16:", _mm256_adds_epi16(testcases_a16, testcases_b16));
printRes16("Saturated signed sub epi16:", _mm256_subs_epi16(testcases_a16, testcases_b16));
printRes16("Saturated unsigned add epu16:", _mm256_adds_epu16(testcases_a16, testcases_b16));
printRes16("Saturated unsigned sub epu16:", _mm256_subs_epu16(testcases_a16, testcases_b16));
cout << endl;
__m512i testcases_a32 = _mm512_set_epi32(0, 0, 0, 0, x32, x32, x32, x32, LONG_MIN, LONG_MIN, LONG_MIN, LONG_MIN, LONG_MAX, LONG_MAX, LONG_MAX, LONG_MAX);
__m512i testcases_b32 = _mm512_set_epi32(0, x32, LONG_MIN, LONG_MAX, 0, x32, LONG_MIN, LONG_MAX, 0, x32, LONG_MIN, LONG_MAX, 0, x32, LONG_MIN, LONG_MAX);
printRes32("Testcases_a32 :", testcases_a32);
printRes32("Testcases_b32 :", testcases_b32);
printRes32("Saturated signed add epi32:", _mm512_adds_epi32(testcases_a32, testcases_b32));
printRes32("Saturated signed add epi32:", _mm512_adds_Zen4_epi32(testcases_a32, testcases_b32));
printRes32("Saturated signed sub epi32:", _mm512_subs_epi32(testcases_a32, testcases_b32));
printRes32("Saturated signed sub epi32:", _mm512_subs_Zen4_epi32(testcases_a32, testcases_b32));
printRes32("Saturated unsigned add epu32:", _mm512_adds_epu32(testcases_a32, testcases_b32));
printRes32("Saturated unsigned add epu32:", _mm512_adds_Zen4_epu32(testcases_a32, testcases_b32));
printRes32("Saturated unsigned sub epu32:", _mm512_subs_epu32(testcases_a32, testcases_b32));
printRes32("Saturated unsigned sub epu32:", _mm512_subs_Zen4_epu32(testcases_a32, testcases_b32));
cout << endl;
__m512i testcases_a0_64 = _mm512_set_epi64( 0, 0, 0, 0, x64, x64, x64, x64);
__m512i testcases_a1_64 = _mm512_set_epi64(LLONG_MIN, LLONG_MIN, LLONG_MIN, LLONG_MIN, LLONG_MAX, LLONG_MAX, LLONG_MAX, LLONG_MAX);
__m512i testcases_b0_64 = _mm512_set_epi64( 0, x64, LLONG_MIN, LLONG_MAX, 0, x64, LLONG_MIN, LLONG_MAX);
__m512i testcases_b1_64 = _mm512_set_epi64( 0, x64, LLONG_MIN, LLONG_MAX, 0, x64, LLONG_MIN, LLONG_MAX);
printRes("Testcases_a0_64 :", testcases_a0_64);
printRes("Testcases_a1_64 :", testcases_a1_64);
printRes("Testcases_b0_64 :", testcases_b0_64);
printRes("Testcases_b1_64 :", testcases_b1_64);
cout << endl;
printRes("Saturated signed add epi64:", _mm512_adds_epi64(testcases_a0_64, testcases_b0_64));
printRes("Saturated signed add epi64:", _mm512_adds_Zen4_epi64(testcases_a0_64, testcases_b0_64));
printRes("Saturated signed add epi64:", _mm512_adds_epi64(testcases_a1_64, testcases_b1_64));
printRes("Saturated signed add epi64:", _mm512_adds_Zen4_epi64(testcases_a1_64, testcases_b1_64));
cout << endl;
printRes("Saturated signed sub epi64:", _mm512_subs_epi64(testcases_a0_64, testcases_b0_64));
printRes("Saturated signed sub epi64:", _mm512_subs_Zen4_epi64(testcases_a0_64, testcases_b0_64));
printRes("Saturated signed sub epi64:", _mm512_subs_epi64(testcases_a1_64, testcases_b1_64));
printRes("Saturated signed sub epi64:", _mm512_subs_Zen4_epi64(testcases_a1_64, testcases_b1_64));
cout << endl;
printRes("Saturated unsigned add epu64:", _mm512_adds_epu64(testcases_a0_64, testcases_b0_64));
printRes("Saturated unsigned add epu64:", _mm512_adds_Zen4_epu64(testcases_a0_64, testcases_b0_64));
printRes("Saturated unsigned add epu64:", _mm512_adds_epu64(testcases_a1_64, testcases_b1_64));
printRes("Saturated unsigned add epu64:", _mm512_adds_Zen4_epu64(testcases_a1_64, testcases_b1_64));
cout << endl;
printRes("Saturated unsigned sub epu64:", _mm512_subs_epu64(testcases_a0_64, testcases_b0_64));
printRes("Saturated unsigned sub epu64:", _mm512_subs_Zen4_epu64(testcases_a0_64, testcases_b0_64));
printRes("Saturated unsigned sub epu64:", _mm512_subs_epu64(testcases_a1_64, testcases_b1_64));
printRes("Saturated unsigned sub epu64:", _mm512_subs_Zen4_epu64(testcases_a1_64, testcases_b1_64));
}