oneAPI Deep Neural Network Library (oneDNN)
Performance library for Deep Learning
2.1.3
dnnl.hpp
Go to the documentation of this file.
1/*******************************************************************************
2* Copyright 2016-2021 Intel Corporation
3*
4* Licensed under the Apache License, Version 2.0 (the "License");
5* you may not use this file except in compliance with the License.
6* You may obtain a copy of the License at
7*
8* http://www.apache.org/licenses/LICENSE-2.0
9*
10* Unless required by applicable law or agreed to in writing, software
11* distributed under the License is distributed on an "AS IS" BASIS,
12* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13* See the License for the specific language governing permissions and
14* limitations under the License.
15*******************************************************************************/
16
19
20#ifndef ONEAPI_DNNL_DNNL_HPP
21#define ONEAPI_DNNL_DNNL_HPP
22
23#include "oneapi/dnnl/dnnl_config.h"
24
26#include <algorithm>
27#include <cstdlib>
28#include <iterator>
29#include <memory>
30#include <string>
31#include <vector>
32#include <unordered_map>
33
34#include "oneapi/dnnl/dnnl.h"
35
37
38// __cpp_exceptions is referred from
39// https://gcc.gnu.org/onlinedocs/libstdc++/manual/using_exceptions.html
40// gcc < 5 does not define __cpp_exceptions but __EXCEPTIONS,
41// Microsoft C++ Compiler does not provide an option to disable exceptions
42#ifndef DNNL_ENABLE_EXCEPTIONS
43#if __cpp_exceptions || __EXCEPTIONS \
44 || (defined(_MSC_VER) && !defined(__clang__))
45#define DNNL_ENABLE_EXCEPTIONS 1
46#else
47#define DNNL_ENABLE_EXCEPTIONS 0
48#endif
49#endif
50
51#if defined(__GNUC__) || defined(__clang__)
52#define DNNL_TRAP() __builtin_trap()
53#elif defined(__INTEL_COMPILER) || defined(_MSC_VER)
54#define DNNL_TRAP() __debugbreak()
55#else
56#error "unknown compiler"
57#endif
58
59#if DNNL_ENABLE_EXCEPTIONS
60#define DNNL_THROW_ERROR(status, msg) throw error(status, msg)
61#else
62#include <cstdio>
63#define DNNL_THROW_ERROR(status, msg) \
64 do { \
65 fputs(msg, stderr); \
66 DNNL_TRAP(); \
67 } while (0)
68#endif
69
72
74namespace dnnl {
75
79
84struct error : public std::exception {
86 const char *message;
87
92 error(dnnl_status_t status, const char *message)
93 : status(status), message(message) {}
94
96 const char *what() const noexcept override { return message; }
97
103 static void wrap_c_api(dnnl_status_t status, const char *message) {
104 if (status != dnnl_success) DNNL_THROW_ERROR(status, message);
105 }
106};
107
109template <typename T>
110void validate_container_size(const T &v, const char *error_message,
111 int min_size = 1, int max_size = -1) {
112 const int size = (int)v.size();
113 if (size < min_size || (max_size >= 0 && size > max_size))
114 DNNL_THROW_ERROR(dnnl_invalid_arguments, error_message);
115}
117
119template <typename T>
121
135template <typename T, typename traits = handle_traits<T>>
136struct handle {
137private:
138 static dnnl_status_t dummy_destructor(T) { return dnnl_success; }
139 std::shared_ptr<typename std::remove_pointer<T>::type> data_ {0};
140
141protected:
142 bool operator==(const T other) const { return other == data_.get(); }
143 bool operator!=(const T other) const { return !(*this == other); }
144
145public:
153 handle() = default;
154
156 handle(const handle<T, traits> &) = default;
163
169 explicit handle(T t, bool weak = false) { reset(t, weak); }
170
176 void reset(T t, bool weak = false) {
177 data_.reset(t, weak ? &dummy_destructor : traits::destructor);
178 }
179
185 T get(bool allow_empty = false) const {
186 T result = data_.get();
187 if (allow_empty == false && result == nullptr)
188 DNNL_THROW_ERROR(
189 dnnl_invalid_arguments, "object is not initialized");
190 return result;
191 }
192
197 explicit operator T() const { return get(true); }
198
202 explicit operator bool() const { return get(true) != nullptr; }
203
210 bool operator==(const handle<T, traits> &other) const {
211 return other.data_.get() == data_.get();
212 }
213
220 bool operator!=(const handle &other) const { return !(*this == other); }
221};
222
224template <>
225struct handle_traits<dnnl_memory_t> {
226 static dnnl_status_t destructor(dnnl_memory_t p) {
227 return dnnl_memory_destroy(p);
228 }
229};
230
231template <>
232struct handle_traits<dnnl_primitive_desc_t> {
233 static dnnl_status_t destructor(dnnl_primitive_desc_t p) {
235 }
236};
237
238template <>
239struct handle_traits<dnnl_primitive_t> {
240 static dnnl_status_t destructor(dnnl_primitive_t p) {
241 return dnnl_primitive_destroy(p);
242 }
243};
244
245template <>
246struct handle_traits<dnnl_primitive_desc_iterator_t> {
247 static dnnl_status_t destructor(dnnl_primitive_desc_iterator_t p) {
249 }
250};
252
254
255struct stream;
256struct memory;
257struct primitive_desc;
258
263
267
269struct primitive : public handle<dnnl_primitive_t> {
271 enum class kind {
277 shuffle = dnnl_shuffle,
281 sum = dnnl_sum,
283 convolution = dnnl_convolution,
285 deconvolution = dnnl_deconvolution,
287 eltwise = dnnl_eltwise,
289 softmax = dnnl_softmax,
291 pooling = dnnl_pooling,
293 lrn = dnnl_lrn,
295 batch_normalization = dnnl_batch_normalization,
297 layer_normalization = dnnl_layer_normalization,
299 inner_product = dnnl_inner_product,
301 rnn = dnnl_rnn,
305 logsoftmax = dnnl_logsoftmax,
309 resampling = dnnl_resampling,
311 pooling_v2 = dnnl_pooling_v2,
315 prelu = dnnl_prelu,
316 };
317
318 using handle::handle;
319
321 primitive() = default;
322
327
332
338
342 inline kind get_kind() const;
343
356 void execute(const stream &astream,
357 const std::unordered_map<int, memory> &args) const;
358};
359
365 return static_cast<dnnl_primitive_kind_t>(akind);
366}
367
371 "could not get a primitive descriptor from a primitive");
372 return pd;
373}
374
377 // TODO (Roma): the code below is only needed because get_primitive_desc
378 // returns a C type.
381 pd, dnnl_query_primitive_kind, 0, (void *)&kind),
382 "could not get a primitive kind from a primitive descriptor");
383 return static_cast<dnnl::primitive::kind>(kind);
384}
385
387
399
401enum class scratchpad_mode {
424};
425
431 return static_cast<dnnl_scratchpad_mode_t>(mode);
432}
433
435enum class prop_kind {
459};
460
466 return static_cast<dnnl_prop_kind_t>(akind);
467}
468
470enum class algorithm {
472 undef = dnnl_alg_kind_undef,
600};
601
606 return static_cast<dnnl_alg_kind_t>(aalgorithm);
607}
608
610
613
615enum class normalization_flags : unsigned {
621
630
637
643};
644
649 return static_cast<dnnl_normalization_flags_t>(flags);
650}
651
653
656
658enum class rnn_flags : unsigned {
661};
662
667 return static_cast<dnnl_rnn_flags_t>(flags);
668}
669
670#define DNNL_DEFINE_BITMASK_OPS(enum_name) \
671 inline enum_name operator|(enum_name lhs, enum_name rhs) { \
672 return static_cast<enum_name>( \
673 static_cast<unsigned>(lhs) | static_cast<unsigned>(rhs)); \
674 } \
675\
676 inline enum_name operator&(enum_name lhs, enum_name rhs) { \
677 return static_cast<enum_name>( \
678 static_cast<unsigned>(lhs) & static_cast<unsigned>(rhs)); \
679 } \
680\
681 inline enum_name operator^(enum_name lhs, enum_name rhs) { \
682 return static_cast<enum_name>( \
683 static_cast<unsigned>(lhs) ^ static_cast<unsigned>(rhs)); \
684 } \
685\
686 inline enum_name &operator|=(enum_name &lhs, enum_name rhs) { \
687 lhs = static_cast<enum_name>( \
688 static_cast<unsigned>(lhs) | static_cast<unsigned>(rhs)); \
689 return lhs; \
690 } \
691\
692 inline enum_name &operator&=(enum_name &lhs, enum_name rhs) { \
693 lhs = static_cast<enum_name>( \
694 static_cast<unsigned>(lhs) & static_cast<unsigned>(rhs)); \
695 return lhs; \
696 } \
697\
698 inline enum_name &operator^=(enum_name &lhs, enum_name rhs) { \
699 lhs = static_cast<enum_name>( \
700 static_cast<unsigned>(lhs) ^ static_cast<unsigned>(rhs)); \
701 return lhs; \
702 } \
703\
704 inline enum_name operator~(enum_name rhs) { \
705 return static_cast<enum_name>(~static_cast<unsigned>(rhs)); \
706 }
707
708DNNL_DEFINE_BITMASK_OPS(normalization_flags)
709DNNL_DEFINE_BITMASK_OPS(rnn_flags)
710
711
712enum class rnn_direction {
725};
726
731 return static_cast<dnnl_rnn_direction_t>(dir);
732}
733
735
738
745enum class query {
748
753
758
765
770
775
778
781
816
835};
836
841 return static_cast<dnnl_query_t>(aquery);
842}
843
845
847
858
860template <>
861struct handle_traits<dnnl_engine_t> {
862 static dnnl_status_t destructor(dnnl_engine_t p) {
863 return dnnl_engine_destroy(p);
864 }
865};
867
869struct engine : public handle<dnnl_engine_t> {
870 friend struct primitive;
871 friend struct reorder;
872
874 enum class kind {
876 any = dnnl_any_engine,
878 cpu = dnnl_cpu,
880 gpu = dnnl_gpu,
881 };
882
883 using handle::handle;
884
887 engine() = default;
888
893 static size_t get_count(kind akind) {
894 return dnnl_engine_get_count(convert_to_c(akind));
895 }
896
902 engine(kind akind, size_t index) {
905 dnnl_engine_create(&engine, convert_to_c(akind), index),
906 "could not create an engine");
907 reset(engine);
908 }
909
915 dnnl_engine_t c_engine;
919 "could not get an engine from a primitive_desc");
920 reset(c_engine, true);
921 }
922
925 kind get_kind() const {
928 "could not get kind of an engine");
929 return static_cast<engine::kind>(kind);
930 }
931
937 template <typename primitive_desc>
938 static engine query(const primitive_desc &pd) {
939 return query(pd, dnnl::query::engine);
940 }
941
942private:
943 static dnnl_engine_kind_t convert_to_c(kind akind) {
944 return static_cast<dnnl_engine_kind_t>(akind);
945 }
946
947 template <typename primitive_desc>
948 static engine query(const primitive_desc &pd, dnnl::query what) {
949 dnnl_engine_t c_engine;
951 dnnl::convert_to_c(what), 0, &c_engine),
952 "could not get an engine from a primitive_desc");
953 return engine(c_engine, true);
954 }
955};
956
962 return static_cast<dnnl_engine_kind_t>(akind);
963}
964
966
974
976template <>
977struct handle_traits<dnnl_stream_t> {
978 static dnnl_status_t destructor(dnnl_stream_t p) {
979 return dnnl_stream_destroy(p);
980 }
981};
983
985struct stream : public handle<dnnl_stream_t> {
986 using handle::handle;
987
989 enum class flags : unsigned {
991 in_order = dnnl_stream_in_order,
993 out_of_order = dnnl_stream_out_of_order,
995 default_flags = dnnl_stream_default_flags,
996 };
997
1000 stream() = default;
1001
1007 stream(const engine &aengine, flags aflags = flags::default_flags) {
1010 static_cast<dnnl_stream_flags_t>(aflags)),
1011 "could not create a stream");
1012 reset(stream);
1013 }
1014
1017 dnnl_engine_t c_engine;
1019 "could not get an engine from a stream object");
1020 return engine(c_engine, true);
1021 }
1022
1027 dnnl_stream_wait(get()), "could not wait on a stream");
1028 return *this;
1029 }
1030};
1031
1032DNNL_DEFINE_BITMASK_OPS(stream::flags)
1033
1034
1035
1036
1101
1108struct memory : public handle<dnnl_memory_t> {
1109 using handle::handle;
1110
1115 typedef std::vector<dim> dims;
1116
1123 template <typename T>
1124 static void validate_dims(const std::vector<T> &v, int min_size = 0) {
1125 validate_container_size(
1126 v, "dimensions are invalid", min_size, DNNL_MAX_NDIMS);
1127 }
1128
1130 enum class data_type {
1134 f16 = dnnl_f16,
1137 bf16 = dnnl_bf16,
1139 f32 = dnnl_f32,
1141 s32 = dnnl_s32,
1143 s8 = dnnl_s8,
1145 u8 = dnnl_u8,
1146 };
1147
1149 enum class format_kind {
1158 blocked = dnnl_blocked,
1160 wino = dnnl_format_kind_wino,
1163 };
1164
1205 enum class format_tag {
1210 any = dnnl_format_tag_any,
1211
1213 a = dnnl_a,
1214
1216 ab = dnnl_ab,
1218 ba = dnnl_ba,
1219
1221 abc = dnnl_abc,
1223 acb = dnnl_acb,
1225 bac = dnnl_bac,
1227 bca = dnnl_bca,
1229 cba = dnnl_cba,
1230
1232 abcd = dnnl_abcd,
1234 abdc = dnnl_abdc,
1236 acdb = dnnl_acdb,
1238 bacd = dnnl_bacd,
1240 bcda = dnnl_bcda,
1242 cdba = dnnl_cdba,
1244 dcab = dnnl_dcab,
1245
1247 abcde = dnnl_abcde,
1249 abdec = dnnl_abdec,
1251 acbde = dnnl_acbde,
1253 acdeb = dnnl_acdeb,
1255 bacde = dnnl_bacde,
1257 bcdea = dnnl_bcdea,
1259 cdeba = dnnl_cdeba,
1261 decab = dnnl_decab,
1263 abced = dnnl_abced,
1264
1266 abcdef = dnnl_abcdef,
1268 abdfce = dnnl_abdfce,
1270 acbdef = dnnl_acbdef,
1272 abdefc = dnnl_abdefc,
1274 defcab = dnnl_defcab,
1276 abcdfe = dnnl_abcdfe,
1277
1279 abcdefg = dnnl_abcdefg,
1281 abcdegf = dnnl_abcdegf,
1282
1284 abcdefgh = dnnl_abcdefgh,
1286 abcdefhg = dnnl_abcdefhg,
1287
1289 abcdefghi = dnnl_abcdefghi,
1291 abcdefgih = dnnl_abcdefgih,
1292
1294 abcdefghij = dnnl_abcdefghij,
1296 abcdefghji = dnnl_abcdefghji,
1297
1299 abcdefghijk = dnnl_abcdefghijk,
1301 abcdefghikj = dnnl_abcdefghikj,
1302
1304 abcdefghijkl = dnnl_abcdefghijkl,
1306 abcdefghijlk = dnnl_abcdefghijlk,
1307
1309 x = a,
1311 nc = ab,
1313 cn = ba,
1315 tn = ab,
1317 nt = ba,
1319 ncw = abc,
1321 nwc = acb,
1323 nchw = abcd,
1325 nhwc = acdb,
1327 chwn = bcda,
1329 ncdhw = abcde,
1331 ndhwc = acdeb,
1332
1334 oi = ab,
1336 io = ba,
1338 oiw = abc,
1340 owi = acb,
1342 wio = cba,
1344 iwo = bca,
1346 oihw = abcd,
1348 hwio = cdba,
1350 ohwi = acdb,
1352 ihwo = bcda,
1354 iohw = bacd,
1356 oidhw = abcde,
1358 dhwio = cdeba,
1360 odhwi = acdeb,
1362 iodhw = bacde,
1364 idhwo = bcdea,
1365
1367 goiw = abcd,
1369 gowi = abdc,
1371 wigo = dcab,
1373 gohwi = abdec,
1375 goihw = abcde,
1377 hwigo = decab,
1379 giohw = acbde,
1381 goidhw = abcdef,
1383 giodhw = acbdef,
1385 godhwi = abdefc,
1387 dhwigo = defcab,
1388
1391 tnc = abc,
1394 ntc = bac,
1397 ldnc = abcd,
1405 ldigo = abcde,
1413 ldgoi = abdec,
1417 ldio = abcd,
1421 ldoi = abdc,
1429 ldgo = abcd,
1430
1431 // Opaque blocked formats
1432
1433 AB16b16a = dnnl_AB16b16a,
1434 AB16b32a = dnnl_AB16b32a,
1435 AB16b64a = dnnl_AB16b64a,
1436 AB8b16a2b = dnnl_AB8b16a2b,
1437 AB8b32a2b = dnnl_AB8b32a2b,
1438 AB8b64a2b = dnnl_AB8b64a2b,
1439 AB4b16a4b = dnnl_AB4b16a4b,
1440 AB4b32a4b = dnnl_AB4b32a4b,
1441 AB4b64a4b = dnnl_AB4b64a4b,
1442 AB16b16a4b = dnnl_AB16b16a4b,
1443 Abc16a = dnnl_Abc16a,
1444 ABc16a16b = dnnl_ABc16a16b,
1445 ABc4a4b = dnnl_ABc4a4b,
1446 aBc16b = dnnl_aBc16b,
1447 aBc32b = dnnl_aBc32b,
1448 ABc16b16a = dnnl_ABc16b16a,
1449 ABc16b32a = dnnl_ABc16b32a,
1450 ABc16b64a = dnnl_ABc16b64a,
1451 Abc4a = dnnl_Abc4a,
1452 aBc4b = dnnl_aBc4b,
1453 ABc4b16a4b = dnnl_ABc4b16a4b,
1454 ABc4b32a4b = dnnl_ABc4b32a4b,
1455 ABc4b64a4b = dnnl_ABc4b64a4b,
1456 ABc2b8a4b = dnnl_ABc2b8a4b,
1457 ABc16a16b2a = dnnl_ABc16a16b2a,
1458 ABc16b16a4b = dnnl_ABc16b16a4b,
1459 ABc16b16a2b = dnnl_ABc16b16a2b,
1460 ABc4b4a = dnnl_ABc4b4a,
1461 ABc8a16b2a = dnnl_ABc8a16b2a,
1462 ABc8a8b = dnnl_ABc8a8b,
1463 ABc8a4b = dnnl_ABc8a4b,
1464 aBc8b = dnnl_aBc8b,
1465 ABc8b16a2b = dnnl_ABc8b16a2b,
1466 ABc8b32a2b = dnnl_ABc8b32a2b,
1467 ABc8b64a2b = dnnl_ABc8b64a2b,
1468 ABc8b8a = dnnl_ABc8b8a,
1469 Abcd8a = dnnl_Abcd8a,
1470 Abcd16a = dnnl_Abcd16a,
1471 Abcd32a = dnnl_Abcd32a,
1472 ABcd16a16b = dnnl_ABcd16a16b,
1473 aBcd16b = dnnl_aBcd16b,
1474 aBcd32b = dnnl_aBcd32b,
1475 ABcd16b16a = dnnl_ABcd16b16a,
1476 ABcd16b32a = dnnl_ABcd16b32a,
1477 ABcd16b64a = dnnl_ABcd16b64a,
1478 aBCd16b16c = dnnl_aBCd16b16c,
1479 aBCd16c16b = dnnl_aBCd16c16b,
1480 Abcd4a = dnnl_Abcd4a,
1481 aBcd4b = dnnl_aBcd4b,
1482 ABcd4b16a4b = dnnl_ABcd4b16a4b,
1483 ABcd4b32a4b = dnnl_ABcd4b32a4b,
1484 ABcd4b64a4b = dnnl_ABcd4b64a4b,
1485 ABcd2b8a4b = dnnl_ABcd2b8a4b,
1486 ABcd4b4a = dnnl_ABcd4b4a,
1487 ABcd4a4b = dnnl_ABcd4a4b,
1488 aBCd4c16b4c = dnnl_aBCd4c16b4c,
1489 aBCd2c8b4c = dnnl_aBCd2c8b4c,
1490 ABcd16a16b2a = dnnl_ABcd16a16b2a,
1491 ABcd16b16a4b = dnnl_ABcd16b16a4b,
1492 ABcd16b16a2b = dnnl_ABcd16b16a2b,
1493 aBCd16b16c2b = dnnl_aBCd16b16c2b,
1494 aBCd16c16b4c = dnnl_aBCd16c16b4c,
1495 aBCd16c16b2c = dnnl_aBCd16c16b2c,
1496 aBCd4c4b = dnnl_aBCd4c4b,
1497 aBCd4b4c = dnnl_aBCd4b4c,
1498 ABcd8a16b2a = dnnl_ABcd8a16b2a,
1499 ABcd8a8b = dnnl_ABcd8a8b,
1500 ABcd8a4b = dnnl_ABcd8a4b,
1502 aBcd8b = dnnl_aBcd8b,
1503 ABcd8b16a2b = dnnl_ABcd8b16a2b,
1504 ABcd8b32a2b = dnnl_ABcd8b32a2b,
1505 ABcd8b64a2b = dnnl_ABcd8b64a2b,
1506 aBCd8b16c2b = dnnl_aBCd8b16c2b,
1508 ABcd8b8a = dnnl_ABcd8b8a,
1509 aBCd8b8c = dnnl_aBCd8b8c,
1510 aBCd8b4c = dnnl_aBCd8b4c,
1511 aBCd8c16b2c = dnnl_aBCd8c16b2c,
1512 aBCd8c8b = dnnl_aBCd8c8b,
1513 Abcde16a = dnnl_Abcde16a,
1514 Abcde32a = dnnl_Abcde32a,
1515 ABcde16a16b = dnnl_ABcde16a16b,
1516 aBcde16b = dnnl_aBcde16b,
1517 aBcde32b = dnnl_aBcde32b,
1518 ABcde16b16a = dnnl_ABcde16b16a,
1519 ABcde16b32a = dnnl_ABcde16b32a,
1520 ABcde16b64a = dnnl_ABcde16b64a,
1521 aBCde16b16c = dnnl_aBCde16b16c,
1522 aBCde16c16b = dnnl_aBCde16c16b,
1523 aBCde2c8b4c = dnnl_aBCde2c8b4c,
1524 Abcde4a = dnnl_Abcde4a,
1525 aBcde4b = dnnl_aBcde4b,
1526 ABcde4b4a = dnnl_ABcde4b4a,
1527 ABcde4a4b = dnnl_ABcde4a4b,
1528 aBCde4b4c = dnnl_aBCde4b4c,
1529 aBCde4c16b4c = dnnl_aBCde4c16b4c,
1530 aBCde16b16c2b = dnnl_aBCde16b16c2b,
1531 aBCde16c16b4c = dnnl_aBCde16c16b4c,
1532 aBCde16c16b2c = dnnl_aBCde16c16b2c,
1533 aBCdef16c16b2c = dnnl_aBCdef16c16b2c,
1534 aBCde4c4b = dnnl_aBCde4c4b,
1535 Abcde8a = dnnl_Abcde8a,
1536 ABcde8a8b = dnnl_ABcde8a8b,
1537 ABcde8a4b = dnnl_ABcde8a4b,
1538 aBcde8b = dnnl_aBcde8b,
1539 ABcde8b16a2b = dnnl_ABcde8b16a2b,
1540 ABcde8b32a2b = dnnl_ABcde8b32a2b,
1541 ABcde8b64a2b = dnnl_ABcde8b64a2b,
1542 ABcde4b16a4b = dnnl_ABcde4b16a4b,
1543 ABcde4b32a4b = dnnl_ABcde4b32a4b,
1544 ABcde4b64a4b = dnnl_ABcde4b64a4b,
1545 ABcde16b16a4b = dnnl_ABcde16b16a4b,
1546 ABcde16b16a2b = dnnl_ABcde16b16a2b,
1547 ABcde2b8a4b = dnnl_ABcde2b8a4b,
1548 aBCde8b16c2b = dnnl_aBCde8b16c2b,
1549 ABcde8b8a = dnnl_ABcde8b8a,
1550 aBCde8b8c = dnnl_aBCde8b8c,
1551 aBCde8b4c = dnnl_aBCde8b4c,
1552 ABcd4a8b8a4b = dnnl_ABcd4a8b8a4b,
1553 ABcd2a8b8a2b = dnnl_ABcd2a8b8a2b,
1554 aBCde4b8c8b4c = dnnl_aBCde4b8c8b4c,
1555 aBCde2b8c8b2c = dnnl_aBCde2b8c8b2c,
1556 aBCde8c16b2c = dnnl_aBCde8c16b2c,
1557 aBCde8c8b = dnnl_aBCde8c8b,
1558 aBcdef16b = dnnl_aBcdef16b,
1559 aBCdef16b16c = dnnl_aBCdef16b16c,
1560 aBCdef16c16b = dnnl_aBCdef16c16b,
1561 aBcdef4b = dnnl_aBcdef4b,
1562 aBCdef2c8b4c = dnnl_aBCdef2c8b4c,
1563 aBCdef4c4b = dnnl_aBCdef4c4b,
1564 aBCdef4b4c = dnnl_aBCdef4b4c,
1565 aBCdef8b8c = dnnl_aBCdef8b8c,
1566 aBCdef8b4c = dnnl_aBCdef8b4c,
1567 aBCdef8c16b2c = dnnl_aBCdef8c16b2c,
1568 aBCdef4c16b4c = dnnl_aBCdef4c16b4c,
1569 aBCdef8c8b = dnnl_aBCdef8c8b,
1570 aBdc16b = dnnl_aBdc16b,
1571 aBdc4b = dnnl_aBdc4b,
1572 aBdc8b = dnnl_aBdc8b,
1573 aBdec16b = dnnl_aBdec16b,
1574 aBdec4b = dnnl_aBdec4b,
1575 aBdec8b = dnnl_aBdec8b,
1576 aBdefc16b = dnnl_aBdefc16b,
1577 aCBdef16c16b = dnnl_aCBdef16c16b,
1578 aCBdef16b16c = dnnl_aCBdef16b16c,
1579 aBdefc4b = dnnl_aBdefc4b,
1580 aBdefc8b = dnnl_aBdefc8b,
1581 Acb16a = dnnl_Acb16a,
1582 Acb4a = dnnl_Acb4a,
1583 Acb8a = dnnl_Acb8a,
1584 aCBd16b16c = dnnl_aCBd16b16c,
1585 aCBd16c16b = dnnl_aCBd16c16b,
1586 aCBde16b16c = dnnl_aCBde16b16c,
1587 aCBde16c16b = dnnl_aCBde16c16b,
1588 Acdb16a = dnnl_Acdb16a,
1589 Acdb4a = dnnl_Acdb4a,
1590 Acdb8a = dnnl_Acdb8a,
1591 Acdeb16a = dnnl_Acdeb16a,
1592 Acdeb4a = dnnl_Acdeb4a,
1593 Acdeb8a = dnnl_Acdeb8a,
1594 BAc16a16b = dnnl_BAc16a16b,
1595 BAc16b16a = dnnl_BAc16b16a,
1596 BAcd16a16b = dnnl_BAcd16a16b,
1597 BAcd16b16a = dnnl_BAcd16b16a,
1598 ABcd32a32b = dnnl_ABcd32a32b,
1599 BAcde16b16a = dnnl_BAcde16b16a,
1600 BAcde16a16b = dnnl_BAcde16a16b,
1601 aBdec32b = dnnl_aBdec32b,
1602 Abcdef16a = dnnl_Abcdef16a,
1603 Abcdef32a = dnnl_Abcdef32a,
1604 Acdb32a = dnnl_Acdb32a,
1605 aBCd2b4c2b = dnnl_aBCd2b4c2b,
1606 aBCde2b4c2b = dnnl_aBCde2b4c2b,
1607 aBCdef2b4c2b = dnnl_aBCdef2b4c2b,
1608 aBCd2c4b2c = dnnl_aBCd2c4b2c,
1609 aBCde2c4b2c = dnnl_aBCde2c4b2c,
1610 aBCdef2c4b2c = dnnl_aBCdef2c4b2c,
1611 aBCd4b8c2b = dnnl_aBCd4b8c2b,
1612 aBCde4b8c2b = dnnl_aBCde4b8c2b,
1613 aBCdef4b8c2b = dnnl_aBCdef4b8c2b,
1614 aBCd4c8b2c = dnnl_aBCd4c8b2c,
1615 aBCde4c8b2c = dnnl_aBCde4c8b2c,
1616 aBCdef4c8b2c = dnnl_aBCdef4c8b2c,
1617 AB32a32b8a4b = dnnl_AB32a32b8a4b,
1618 AB32a32b8a2b = dnnl_AB32a32b8a2b,
1619 AB8a4b = dnnl_AB8a4b,
1620 AB8a2b = dnnl_AB8a2b,
1621 abDc32d = dnnl_abDc32d,
1622 abDC32d4c = dnnl_abDC32d4c,
1623 abdEc32e = dnnl_abdEc32e,
1624 abdEC32e2c = dnnl_abdEC32e2c,
1625 abdEC32e4c = dnnl_abdEC32e4c,
1626 aBCdef16c16b4c = dnnl_aBCdef16c16b4c,
1627 aBdC16b4c = dnnl_aBdC16b4c,
1628 aBdeC16b4c = dnnl_aBdeC16b4c,
1629 AcB16a4b = dnnl_AcB16a4b,
1630 AcdB16a2b = dnnl_AcdB16a2b,
1631 aBdefC16b4c = dnnl_aBdefC16b4c,
1632 AcdeB16a4b = dnnl_AcdeB16a4b,
1633
1634 Acb32a = dnnl_Acb32a,
1635 AcB32a2b = dnnl_AcB32a2b,
1636 AcB32a4b = dnnl_AcB32a4b,
1637 Acb48a = dnnl_Acb48a,
1638 AcB48a2b = dnnl_AcB48a2b,
1639 AcB48a4b = dnnl_AcB48a4b,
1640 Acb64a = dnnl_Acb64a,
1641 AcB64a2b = dnnl_AcB64a2b,
1642 AcB64a4b = dnnl_AcB64a4b,
1643 cBa2b = dnnl_cBa2b,
1644 cBa4b = dnnl_cBa4b,
1645 aBdc32b = dnnl_aBdc32b,
1646 aBdC32b2c = dnnl_aBdC32b2c,
1647 aBdC32b4c = dnnl_aBdC32b4c,
1648 aBdc48b = dnnl_aBdc48b,
1649 aBdC48b2c = dnnl_aBdC48b2c,
1650 aBdC48b4c = dnnl_aBdC48b4c,
1651 aBdc64b = dnnl_aBdc64b,
1652 aBdC64b2c = dnnl_aBdC64b2c,
1653 aBdC64b4c = dnnl_aBdC64b4c,
1654 adcb = dnnl_adcb,
1655 adCb2c = dnnl_adCb2c,
1656 adCb4c = dnnl_adCb4c,
1657 AcdB32a2b = dnnl_AcdB32a2b,
1658 AcdB32a4b = dnnl_AcdB32a4b,
1659 Acdb48a = dnnl_Acdb48a,
1660 AcdB48a2b = dnnl_AcdB48a2b,
1661 AcdB48a4b = dnnl_AcdB48a4b,
1662 Acdb64a = dnnl_Acdb64a,
1663 AcdB64a2b = dnnl_AcdB64a2b,
1664 AcdB64a4b = dnnl_AcdB64a4b,
1665 cdBa2b = dnnl_cdBa2b,
1666 cdBa4b = dnnl_cdBa4b,
1667 aBdeC32b2c = dnnl_aBdeC32b2c,
1668 aBdeC32b4c = dnnl_aBdeC32b4c,
1669 aBdec48b = dnnl_aBdec48b,
1670 aBdeC48b2c = dnnl_aBdeC48b2c,
1671 aBdeC48b4c = dnnl_aBdeC48b4c,
1672 aBdec64b = dnnl_aBdec64b,
1673 aBdeC64b2c = dnnl_aBdeC64b2c,
1674 aBdeC64b4c = dnnl_aBdeC64b4c,
1675 adecb = dnnl_adecb,
1676 adeCb2c = dnnl_adeCb2c,
1677 adeCb4c = dnnl_adeCb4c,
1678 Acdeb32a = dnnl_Acdeb32a,
1679 AcdeB32a2b = dnnl_AcdeB32a2b,
1680 AcdeB32a4b = dnnl_AcdeB32a4b,
1681 Acdeb48a = dnnl_Acdeb48a,
1682 AcdeB48a2b = dnnl_AcdeB48a2b,
1683 AcdeB48a4b = dnnl_AcdeB48a4b,
1684 Acdeb64a = dnnl_Acdeb64a,
1685 AcdeB64a2b = dnnl_AcdeB64a2b,
1686 AcdeB64a4b = dnnl_AcdeB64a4b,
1687 cdeBa2b = dnnl_cdeBa2b,
1688 cdeBa4b = dnnl_cdeBa4b,
1689 aBdefc32b = dnnl_aBdefc32b,
1690 aBdefC32b2c = dnnl_aBdefC32b2c,
1691 aBdefC32b4c = dnnl_aBdefC32b4c,
1692 aBdefc48b = dnnl_aBdefc48b,
1693 aBdefC48b2c = dnnl_aBdefC48b2c,
1694 aBdefC48b4c = dnnl_aBdefC48b4c,
1695 aBdefc64b = dnnl_aBdefc64b,
1696 aBdefC64b2c = dnnl_aBdefC64b2c,
1697 aBdefC64b4c = dnnl_aBdefC64b4c,
1698 adefcb = dnnl_adefcb,
1699 adefCb2c = dnnl_adefCb2c,
1700 adefCb4c = dnnl_adefCb4c,
1701
1702 format_tag_last = dnnl_format_tag_last,
1703
1704 nCdhw16c = dnnl_nCdhw16c,
1705 nCdhw4c = dnnl_nCdhw4c,
1706 nCdhw8c = dnnl_nCdhw8c,
1707 nChw16c = dnnl_nChw16c,
1708 nChw4c = dnnl_nChw4c,
1709 nChw8c = dnnl_nChw8c,
1710 nCw16c = dnnl_nCw16c,
1711 nCw4c = dnnl_nCw4c,
1712 nCw8c = dnnl_nCw8c,
1713 NCw16n16c = dnnl_NCw16n16c,
1714 NChw16n16c = dnnl_NChw16n16c,
1715 NCdhw16n16c = dnnl_NCdhw16n16c,
1716 NCdhw32n32c = dnnl_NCdhw32n32c,
1717 NChw32n32c = dnnl_NChw32n32c,
1718 IOhw16i16o = dnnl_IOhw16i16o,
1719 OI16i16o = dnnl_OI16i16o,
1720 OI16i32o = dnnl_OI16i32o,
1721 OI16i64o = dnnl_OI16i64o,
1722 OI8i16o2i = dnnl_OI8i16o2i,
1723 OI8i32o2i = dnnl_OI8i32o2i,
1724 OI8i64o2i = dnnl_OI8i64o2i,
1725 OI4i16o4i = dnnl_OI4i16o4i,
1726 OI4i32o4i = dnnl_OI4i32o4i,
1727 OI4i64o4i = dnnl_OI4i64o4i,
1728 Ohwi32o = dnnl_Ohwi32o,
1729 IOdhw16i16o = dnnl_IOdhw16i16o,
1730 gIOhw16i16o = dnnl_gIOhw16i16o,
1731 gOhwi32o = dnnl_gOhwi32o,
1732 Goidhw16g = dnnl_Goidhw16g,
1733 IOw16o16i = dnnl_IOw16o16i,
1734 OIw16i16o = dnnl_OIw16i16o,
1735 OIw16i32o = dnnl_OIw16i32o,
1736 OIw16i64o = dnnl_OIw16i64o,
1737 IOw16i16o = dnnl_IOw16i16o,
1738 gIOw16i16o = dnnl_gIOw16i16o,
1739 OIw16o16i = dnnl_OIw16o16i,
1740 Oiw16o = dnnl_Oiw16o,
1741 OIw4i16o4i = dnnl_OIw4i16o4i,
1742 OIw4i32o4i = dnnl_OIw4i32o4i,
1743 OIw4i64o4i = dnnl_OIw4i64o4i,
1744 OIw2i8o4i = dnnl_OIw2i8o4i,
1745 OIw4i4o = dnnl_OIw4i4o,
1746 OIw4o4i = dnnl_OIw4o4i,
1747 Oiw4o = dnnl_Oiw4o,
1748 OIw8i16o2i = dnnl_OIw8i16o2i,
1749 OIw8i32o2i = dnnl_OIw8i32o2i,
1750 OIw8i64o2i = dnnl_OIw8i64o2i,
1751 OIw8i8o = dnnl_OIw8i8o,
1752 OIw8o16i2o = dnnl_OIw8o16i2o,
1753 OIw8o8i = dnnl_OIw8o8i,
1754 OIw8o4i = dnnl_OIw8o4i,
1755 OIw16i16o4i = dnnl_OIw16i16o4i,
1756 OIw16i16o2i = dnnl_OIw16i16o2i,
1757 OIw16o16i2o = dnnl_OIw16o16i2o,
1758 Owi16o = dnnl_Owi16o,
1759 OwI16o2i = dnnl_OwI16o2i,
1760 Owi4o = dnnl_Owi4o,
1761 Owi8o = dnnl_Owi8o,
1762 IOhw16o16i = dnnl_IOhw16o16i,
1763 Ohwi16o = dnnl_Ohwi16o,
1764 OhwI16o2i = dnnl_OhwI16o2i,
1765 Ohwi4o = dnnl_Ohwi4o,
1766 Ohwi8o = dnnl_Ohwi8o,
1767 OIhw16i16o = dnnl_OIhw16i16o,
1768 OIhw16i32o = dnnl_OIhw16i32o,
1769 OIhw16i64o = dnnl_OIhw16i64o,
1770 OIhw16o16i = dnnl_OIhw16o16i,
1771 Oihw16o = dnnl_Oihw16o,
1772 OIhw4i16o4i = dnnl_OIhw4i16o4i,
1773 OIhw4i32o4i = dnnl_OIhw4i32o4i,
1774 OIhw4i64o4i = dnnl_OIhw4i64o4i,
1775 OIhw4i4o = dnnl_OIhw4i4o,
1776 OIhw4o4i = dnnl_OIhw4o4i,
1777 Oihw4o = dnnl_Oihw4o,
1778 OIhw8i16o2i = dnnl_OIhw8i16o2i,
1779 OIhw8i32o2i = dnnl_OIhw8i32o2i,
1780 OIhw8i64o2i = dnnl_OIhw8i64o2i,
1781 OIhw8i8o = dnnl_OIhw8i8o,
1782 OIhw8o16i2o = dnnl_OIhw8o16i2o,
1783 OIhw8o8i = dnnl_OIhw8o8i,
1784 OIhw8o4i = dnnl_OIhw8o4i,
1785 OIhw2i8o4i = dnnl_OIhw2i8o4i,
1786 IOdhw16o16i = dnnl_IOdhw16o16i,
1787 Odhwi16o = dnnl_Odhwi16o,
1788 OdhwI16o2i = dnnl_OdhwI16o2i,
1789 Odhwi4o = dnnl_Odhwi4o,
1790 Odhwi8o = dnnl_Odhwi8o,
1791 OIdhw16i16o = dnnl_OIdhw16i16o,
1792 OIdhw16i32o = dnnl_OIdhw16i32o,
1793 OIdhw16i64o = dnnl_OIdhw16i64o,
1794 OIdhw16o16i = dnnl_OIdhw16o16i,
1795 Oidhw16o = dnnl_Oidhw16o,
1796 OIdhw4i4o = dnnl_OIdhw4i4o,
1797 OIdhw4o4i = dnnl_OIdhw4o4i,
1798 Oidhw4o = dnnl_Oidhw4o,
1799 OIdhw8i16o2i = dnnl_OIdhw8i16o2i,
1800 OIdhw8i32o2i = dnnl_OIdhw8i32o2i,
1801 OIdhw8i64o2i = dnnl_OIdhw8i64o2i,
1802 OIdhw4i16o4i = dnnl_OIdhw4i16o4i,
1803 OIdhw16i16o4i = dnnl_OIdhw16i16o4i,
1804 OIdhw4i32o4i = dnnl_OIdhw4i32o4i,
1805 OIdhw4i64o4i = dnnl_OIdhw4i64o4i,
1806 OIdhw2i8o4i = dnnl_OIdhw2i8o4i,
1807 OIdhw8i8o = dnnl_OIdhw8i8o,
1808 OIdhw8o8i = dnnl_OIdhw8o8i,
1809 OIdhw8o4i = dnnl_OIdhw8o4i,
1810 gIOw16o16i = dnnl_gIOw16o16i,
1811 gOIw16i16o = dnnl_gOIw16i16o,
1812 gOIw16o16i = dnnl_gOIw16o16i,
1813 gOiw16o = dnnl_gOiw16o,
1814 gOIw4i16o4i = dnnl_gOIw4i16o4i,
1815 gOIw2i8o4i = dnnl_gOIw2i8o4i,
1816 gOIw4i4o = dnnl_gOIw4i4o,
1817 gOIw4o4i = dnnl_gOIw4o4i,
1818 gOiw4o = dnnl_gOiw4o,
1819 gOIw8i16o2i = dnnl_gOIw8i16o2i,
1820 gOIw8i8o = dnnl_gOIw8i8o,
1821 gOIw8o16i2o = dnnl_gOIw8o16i2o,
1822 gOIw8o8i = dnnl_gOIw8o8i,
1823 gOIw8o4i = dnnl_gOIw8o4i,
1824 gOIw16i16o4i = dnnl_gOIw16i16o4i,
1825 gOIw16i16o2i = dnnl_gOIw16i16o2i,
1826 gOIw16o16i2o = dnnl_gOIw16o16i2o,
1827 gOwi16o = dnnl_gOwi16o,
1828 gOwI16o2i = dnnl_gOwI16o2i,
1829 gOwi4o = dnnl_gOwi4o,
1830 gOwi8o = dnnl_gOwi8o,
1831 Goiw8g = dnnl_Goiw8g,
1832 Goiw16g = dnnl_Goiw16g,
1833 gIOhw16o16i = dnnl_gIOhw16o16i,
1834 gOhwi16o = dnnl_gOhwi16o,
1835 gOhwI16o2i = dnnl_gOhwI16o2i,
1836 gOhwi4o = dnnl_gOhwi4o,
1837 gOhwi8o = dnnl_gOhwi8o,
1838 Goihw16g = dnnl_Goihw16g,
1839 gOIhw16i16o = dnnl_gOIhw16i16o,
1840 gOIhw16o16i = dnnl_gOIhw16o16i,
1841 gOihw16o = dnnl_gOihw16o,
1842 gOIhw4i16o4i = dnnl_gOIhw4i16o4i,
1843 gOIhw2i8o4i = dnnl_gOIhw2i8o4i,
1844 gOIhw4i4o = dnnl_gOIhw4i4o,
1845 gOIhw4o4i = dnnl_gOIhw4o4i,
1846 gOihw4o = dnnl_gOihw4o,
1847 Goihw8g = dnnl_Goihw8g,
1848 gOIhw8i16o2i = dnnl_gOIhw8i16o2i,
1849 gOIhw8i8o = dnnl_gOIhw8i8o,
1850 gOIhw8o16i2o = dnnl_gOIhw8o16i2o,
1851 OIw4o8i8o4i = dnnl_OIw4o8i8o4i,
1852 OIdhw4o8i8o4i = dnnl_OIdhw4o8i8o4i,
1853 OIhw4o8i8o4i = dnnl_OIhw4o8i8o4i,
1854 OIhw2o8i8o2i = dnnl_OIhw2o8i8o2i,
1855 gOIw4o8i8o4i = dnnl_gOIw4o8i8o4i,
1856 gOIdhw4o8i8o4i = dnnl_gOIdhw4o8i8o4i,
1857 gOIhw4o8i8o4i = dnnl_gOIhw4o8i8o4i,
1858 gOIhw2o8i8o2i = dnnl_gOIhw2o8i8o2i,
1859 OIhw16i16o4i = dnnl_OIhw16i16o4i,
1860 OIhw16i16o2i = dnnl_OIhw16i16o2i,
1861 OIhw16o16i2o = dnnl_OIhw16o16i2o,
1862 OIdhw16i16o2i = dnnl_OIdhw16i16o2i,
1863 gOIhw16i16o4i = dnnl_gOIhw16i16o4i,
1864 gOIhw16i16o2i = dnnl_gOIhw16i16o2i,
1865 gOIhw16o16i2o = dnnl_gOIhw16o16i2o,
1866 gOIhw8o8i = dnnl_gOIhw8o8i,
1867 gOIhw8o4i = dnnl_gOIhw8o4i,
1868 gIOdhw16i16o = dnnl_gIOdhw16i16o,
1869 gIOdhw16o16i = dnnl_gIOdhw16o16i,
1870 gOdhwi16o = dnnl_gOdhwi16o,
1871 gOdhwI16o2i = dnnl_gOdhwI16o2i,
1872 gOdhwi4o = dnnl_gOdhwi4o,
1873 gOdhwi8o = dnnl_gOdhwi8o,
1874 gOIdhw16i16o = dnnl_gOIdhw16i16o,
1875 gOIdhw16o16i = dnnl_gOIdhw16o16i,
1876 gOidhw16o = dnnl_gOidhw16o,
1877 gOIdhw4i4o = dnnl_gOIdhw4i4o,
1878 gOIdhw4o4i = dnnl_gOIdhw4o4i,
1879 gOidhw4o = dnnl_gOidhw4o,
1880 gOIdhw8i16o2i = dnnl_gOIdhw8i16o2i,
1881 gOIdhw4i16o4i = dnnl_gOIdhw4i16o4i,
1882 gOIdhw16i16o4i = dnnl_gOIdhw16i16o4i,
1883 gOIdhw16i16o2i = dnnl_gOIdhw16i16o2i,
1884 gOIdhw2i8o4i = dnnl_gOIdhw2i8o4i,
1885 gOIdhw8i8o = dnnl_gOIdhw8i8o,
1886 gOIdhw8o8i = dnnl_gOIdhw8o8i,
1887 gOIdhw8o4i = dnnl_gOIdhw8o4i,
1888 gOIw2i4o2i = dnnl_gOIw2i4o2i,
1889 gOIhw2i4o2i = dnnl_gOIhw2i4o2i,
1890 gOIdhw2i4o2i = dnnl_gOIdhw2i4o2i,
1891 gOIw2o4i2o = dnnl_gOIw2o4i2o,
1892 gOIhw2o4i2o = dnnl_gOIhw2o4i2o,
1893 gOIdhw2o4i2o = dnnl_gOIdhw2o4i2o,
1894 gOIw4i8o2i = dnnl_gOIw4i8o2i,
1895 gOIhw4i8o2i = dnnl_gOIhw4i8o2i,
1896 gOIdhw4i8o2i = dnnl_gOIdhw4i8o2i,
1897 gOIw4o8i2o = dnnl_gOIw4o8i2o,
1898 gOIhw4o8i2o = dnnl_gOIhw4o8i2o,
1899 gOIdhw4o8i2o = dnnl_gOIdhw4o8i2o,
1900 ldOi32o = abDc32d,
1901 ldOI32o4i = abDC32d4c,
1902 ldgOi32o = abdEc32e,
1903 ldgOI32o2i = abdEC32e2c,
1904 ldgOI32o4i = abdEC32e4c,
1905 OwI16o4i = dnnl_OwI16o4i,
1906 OhwI16o4i = dnnl_OhwI16o4i,
1907 gOwI16o4i = dnnl_gOwI16o4i,
1908 gOhwI16o4i = dnnl_gOhwI16o4i,
1909 OdhwI16o4i = dnnl_OdhwI16o4i,
1910 gOdhwI16o4i = dnnl_gOdhwI16o4i,
1911
1912 Owi32o = dnnl_Owi32o,
1913 OwI32o2i = dnnl_OwI32o2i,
1914 OwI32o4i = dnnl_OwI32o4i,
1915 Owi48o = dnnl_Owi48o,
1916 OwI48o2i = dnnl_OwI48o2i,
1917 OwI48o4i = dnnl_OwI48o4i,
1918 Owi64o = dnnl_Owi64o,
1919 OwI64o2i = dnnl_OwI64o2i,
1920 OwI64o4i = dnnl_OwI64o4i,
1921 wIo2i = dnnl_wIo2i,
1922 wIo4i = dnnl_wIo4i,
1923 gOwi32o = dnnl_gOwi32o,
1924 gOwI32o2i = dnnl_gOwI32o2i,
1925 gOwI32o4i = dnnl_gOwI32o4i,
1926 gOwi48o = dnnl_gOwi48o,
1927 gOwI48o2i = dnnl_gOwI48o2i,
1928 gOwI48o4i = dnnl_gOwI48o4i,
1929 gOwi64o = dnnl_gOwi64o,
1930 gOwI64o2i = dnnl_gOwI64o2i,
1931 gOwI64o4i = dnnl_gOwI64o4i,
1932 gwio = dnnl_gwio,
1933 gwIo2i = dnnl_gwIo2i,
1934 gwIo4i = dnnl_gwIo4i,
1935 OhwI32o = dnnl_OhwI32o,
1936 OhwI32o2i = dnnl_OhwI32o2i,
1937 OhwI32o4i = dnnl_OhwI32o4i,
1938 Ohwi48o = dnnl_Ohwi48o,
1939 OhwI48o2i = dnnl_OhwI48o2i,
1940 OhwI48o4i = dnnl_OhwI48o4i,
1941 Ohwi64o = dnnl_Ohwi64o,
1942 OhwI64o2i = dnnl_OhwI64o2i,
1943 OhwI64o4i = dnnl_OhwI64o4i,
1944 hwIo2i = dnnl_hwIo2i,
1945 hwIo4i = dnnl_hwIo4i,
1946 gOhwI32o = dnnl_gOhwI32o,
1947 gOhwI32o2i = dnnl_gOhwI32o2i,
1948 gOhwI32o4i = dnnl_gOhwI32o4i,
1949 gOhwi48o = dnnl_gOhwi48o,
1950 gOhwI48o2i = dnnl_gOhwI48o2i,
1951 gOhwI48o4i = dnnl_gOhwI48o4i,
1952 gOhwi64o = dnnl_gOhwi64o,
1953 gOhwI64o2i = dnnl_gOhwI64o2i,
1954 gOhwI64o4i = dnnl_gOhwI64o4i,
1955 ghwio = dnnl_ghwio,
1956 ghwIo2i = dnnl_ghwIo2i,
1957 ghwIo4i = dnnl_ghwIo4i,
1958 Odhwi32o = dnnl_Odhwi32o,
1959 OdhwI32o2i = dnnl_OdhwI32o2i,
1960 OdhwI32o4i = dnnl_OdhwI32o4i,
1961 Odhwi48o = dnnl_Odhwi48o,
1962 OdhwI48o2i = dnnl_OdhwI48o2i,
1963 OdhwI48o4i = dnnl_OdhwI48o4i,
1964 Odhwi64o = dnnl_Odhwi64o,
1965 OdhwI64o2i = dnnl_OdhwI64o2i,
1966 OdhwI64o4i = dnnl_OdhwI64o4i,
1967 dhwIo2i = dnnl_dhwIo2i,
1968 dhwIo4i = dnnl_dhwIo4i,
1969 gOdhwi32o = dnnl_gOdhwi32o,
1970 gOdhwI32o2i = dnnl_gOdhwI32o2i,
1971 gOdhwI32o4i = dnnl_gOdhwI32o4i,
1972 gOdhwi48o = dnnl_gOdhwi48o,
1973 gOdhwI48o2i = dnnl_gOdhwI48o2i,
1974 gOdhwI48o4i = dnnl_gOdhwI48o4i,
1975 gOdhwi64o = dnnl_gOdhwi64o,
1976 gOdhwI64o2i = dnnl_gOdhwI64o2i,
1977 gOdhwI64o4i = dnnl_gOdhwI64o4i,
1978 gdhwio = dnnl_gdhwio,
1979 gdhwIo2i = dnnl_gdhwIo2i,
1980 gdhwIo4i = dnnl_gdhwIo4i,
1981 };
1982
1984 struct desc {
1985 friend struct memory;
1988
1991 desc() : data() {}
1992
2008 desc(const dims &adims, data_type adata_type, format_tag aformat_tag,
2009 bool allow_empty = false)
2010 : data() {
2011 validate_dims(adims);
2013 (int)adims.size(), adims.data(), convert_to_c(adata_type),
2014 convert_to_c(aformat_tag));
2015 if (!allow_empty)
2017 "could not construct a memory descriptor using a "
2018 "format tag");
2019 }
2020
2036 desc(const dims &adims, data_type adata_type, const dims &strides,
2037 bool allow_empty = false)
2038 : data() {
2039 validate_dims(adims);
2040 if (!strides.empty()) validate_dims(strides, (int)adims.size());
2042 (int)adims.size(), adims.data(), convert_to_c(adata_type),
2043 strides.empty() ? nullptr : &strides[0]);
2044 if (!allow_empty)
2046 "could not construct a memory descriptor using "
2047 "strides");
2048 }
2049
2053 desc(const dnnl_memory_desc_t &data) : data(data) {}
2054
2057 //
2066 desc submemory_desc(const dims &adims, const dims &offsets,
2067 bool allow_empty = false) const {
2068 validate_dims(adims, data.ndims);
2069 validate_dims(offsets, data.ndims);
2072 &sub_md, &data, adims.data(), offsets.data());
2073 if (!allow_empty)
2074 error::wrap_c_api(status, "could not construct a sub-memory");
2075 return desc(sub_md);
2076 }
2077
2122 desc reshape(const dims &adims, bool allow_empty = false) const {
2123 if (data.ndims) validate_dims(adims, 1);
2126 &out_md, &data, (int)adims.size(), adims.data());
2127 if (!allow_empty)
2129 status, "could not reshape a memory descriptor");
2130 return desc(out_md);
2131 }
2132
2170 desc permute_axes(const std::vector<int> &permutation,
2171 bool allow_empty = false) const {
2172 validate_dims(permutation, data.ndims);
2175 &out_md, &data, permutation.data());
2176 if (!allow_empty)
2178 "could not permute axes of a memory descriptor");
2179 return desc(out_md);
2180 }
2181
2187 return memory::dims(data.dims, data.dims + data.ndims);
2188 }
2189
2193 return static_cast<memory::data_type>(data.data_type);
2194 }
2195
2200 size_t get_size() const { return dnnl_memory_desc_get_size(&data); }
2201
2205 bool is_zero() const { return data.ndims == 0; }
2206
2211 bool operator==(const desc &other) const {
2212 return dnnl_memory_desc_equal(&data, &other.data) != 0;
2213 }
2214
2219 bool operator!=(const desc &other) const { return !operator==(other); }
2220
2224 explicit operator bool() const { return data.ndims != 0; }
2225 };
2226
2231 memory() = default;
2232
2252 memory(const desc &md, const engine &aengine, void *handle) {
2253 dnnl_memory_t result;
2255 dnnl_memory_create(&result, &md.data, aengine.get(), handle),
2256 "could not create a memory object");
2257 reset(result);
2258 }
2259
2266 memory(const desc &md, const engine &aengine)
2267 : memory(md, aengine, DNNL_MEMORY_ALLOCATE) {}
2268
2270 desc get_desc() const {
2271 const dnnl_memory_desc_t *cdesc;
2273 "could not get a memory descriptor from a memory object");
2274 return desc(*cdesc);
2275 }
2276
2279 dnnl_engine_t c_engine;
2280 error::wrap_c_api(dnnl_memory_get_engine(get(), &c_engine),
2281 "could not get an engine from a memory object");
2282 return engine(c_engine, true);
2283 }
2284
2289 void *get_data_handle() const {
2290 void *handle;
2292 "could not get a native handle from a memory object");
2293 return handle;
2294 }
2295
2324 void set_data_handle(void *handle, const stream &astream) const {
2326 get(), handle, astream.get(true)),
2327 "could not set native handle of a memory object");
2328 }
2329
2340 void set_data_handle(void *handle) const {
2342 dnnl_memory_set_data_handle_v2(get(), handle, nullptr),
2343 "could not set native handle of a memory object");
2344 }
2345
2367 template <typename T = void>
2368 T *map_data() const {
2369 void *mapped_ptr;
2370 error::wrap_c_api(dnnl_memory_map_data(get(), &mapped_ptr),
2371 "could not map memory object data");
2372 return static_cast<T *>(mapped_ptr);
2373 }
2374
2385 void unmap_data(void *mapped_ptr) const {
2386 error::wrap_c_api(dnnl_memory_unmap_data(get(), mapped_ptr),
2387 "could not unmap memory object data");
2388 }
2389
2390 static dnnl_data_type_t convert_to_c(data_type adata_type) {
2391 return static_cast<dnnl_data_type_t>(adata_type);
2392 }
2393 static dnnl_format_tag_t convert_to_c(format_tag format) {
2394 return static_cast<dnnl_format_tag_t>(format);
2395 }
2396};
2397
2398inline bool operator==(dnnl_data_type_t a, memory::data_type b) {
2399 return a == memory::convert_to_c(b);
2400}
2401inline bool operator!=(dnnl_data_type_t a, memory::data_type b) {
2402 return !(a == b);
2403}
2404inline bool operator==(memory::data_type a, dnnl_data_type_t b) {
2405 return b == a;
2406}
2407inline bool operator!=(memory::data_type a, dnnl_data_type_t b) {
2408 return !(a == b);
2409}
2410
2411inline bool operator==(dnnl_format_tag_t a, memory::format_tag b) {
2412 return a == memory::convert_to_c(b);
2413}
2414inline bool operator!=(dnnl_format_tag_t a, memory::format_tag b) {
2415 return !(a == b);
2416}
2417inline bool operator==(memory::format_tag a, dnnl_format_tag_t b) {
2418 return b == a;
2419}
2420inline bool operator!=(memory::format_tag a, dnnl_format_tag_t b) {
2421 return !(a == b);
2422}
2423
2425
2433
2435template <>
2436struct handle_traits<dnnl_post_ops_t> {
2437 static dnnl_status_t destructor(dnnl_post_ops_t p) {
2438 return dnnl_post_ops_destroy(p);
2439 }
2440};
2442
2450struct post_ops : public handle<dnnl_post_ops_t> {
2452
2455 dnnl_post_ops_t result;
2457 dnnl_post_ops_create(&result), "could not create post-ops");
2458 reset(result);
2459 }
2460
2462 int len() const { return dnnl_post_ops_len(get()); }
2463
2467 primitive::kind kind(int index) const {
2469 "post-ops index is out of range");
2470 return static_cast<primitive::kind>(
2471 dnnl_post_ops_get_kind(get(), index));
2472 }
2473
2502 void append_sum(float scale = 1.f,
2504 if (data_type == memory::data_type::undef)
2506 "could not append a sum post-op");
2507 else
2509 memory::convert_to_c(data_type)),
2510 "could not append a sum post-op");
2511 }
2512
2517 void get_params_sum(int index, float &scale) const {
2519 "could not get parameters of a sum post-op");
2520 }
2521
2528 int index, float &scale, memory::data_type &data_type) const {
2529 dnnl_data_type_t c_data_type;
2531 get(), index, &scale, &c_data_type),
2532 "could not get parameters of a sum post-op");
2533 data_type = static_cast<memory::data_type>(c_data_type);
2534 }
2535
2550 float scale, algorithm aalgorithm, float alpha, float beta) {
2552 convert_to_c(aalgorithm), alpha, beta),
2553 "could not append an elementwise post-op");
2554 }
2555
2563 void get_params_eltwise(int index, float &scale, algorithm &aalgorithm,
2564 float &alpha, float &beta) const {
2565 dnnl_alg_kind_t c_alg;
2567 get(), index, &scale, &c_alg, &alpha, &beta),
2568 "could not get parameters of an elementwise post-op");
2569 aalgorithm = static_cast<dnnl::algorithm>(c_alg);
2570 }
2571
2600 void append_dw_k3s1p1(memory::data_type weights_data_type,
2601 memory::data_type bias_data_type, memory::data_type dst_data_type,
2602 int mask, const std::vector<float> &scales) {
2603
2605 memory::convert_to_c(weights_data_type),
2606 memory::convert_to_c(bias_data_type),
2607 memory::convert_to_c(dst_data_type),
2608 scales.size(), mask, &scales[0]),
2609 "could not append depthwise post-op");
2610 }
2611
2626 void get_params_dw_k3s1p1(int index, memory::data_type &weights_data_type,
2627 memory::data_type &bias_data_type, memory::data_type &dst_data_type,
2628 int &mask, std::vector<float> &scales) const {
2629
2630 dnnl_data_type_t c_weights_data_type;
2631 dnnl_data_type_t c_bias_data_type;
2632 dnnl_data_type_t c_dst_data_type;
2633 dnnl_dim_t count;
2634 int c_mask;
2635 const float *c_scales;
2637 &c_weights_data_type, &c_bias_data_type,
2638 &c_dst_data_type, &count, &c_mask, &c_scales),
2639 "could not get parameters of depthwise post-op");
2640
2641 weights_data_type = static_cast<memory::data_type>(c_weights_data_type);
2642 bias_data_type = static_cast<memory::data_type>(c_bias_data_type);
2643 dst_data_type = static_cast<memory::data_type>(c_dst_data_type);
2644 scales.resize(count);
2645
2646 mask = c_mask;
2647 for (dnnl_dim_t c = 0; c < count; ++c)
2648 scales[c] = c_scales[c];
2649 return;
2650 }
2651
2685 void append_dw_k3s2p1(memory::data_type weights_data_type,
2686 memory::data_type bias_data_type, memory::data_type dst_data_type,
2687 int mask, const std::vector<float> &scales) {
2688
2690 memory::convert_to_c(weights_data_type),
2691 memory::convert_to_c(bias_data_type),
2692 memory::convert_to_c(dst_data_type),
2693 scales.size(), mask, &scales[0]),
2694 "could not append depthwise post-op");
2695 }
2696
2711 void get_params_dw_k3s2p1(int index, memory::data_type &weights_data_type,
2712 memory::data_type &bias_data_type, memory::data_type &dst_data_type,
2713 int &mask, std::vector<float> &scales) const {
2714
2715 dnnl_data_type_t c_weights_data_type;
2716 dnnl_data_type_t c_bias_data_type;
2717 dnnl_data_type_t c_dst_data_type;
2718 dnnl_dim_t count;
2719 int c_mask;
2720 const float *c_scales;
2722 &c_weights_data_type, &c_bias_data_type,
2723 &c_dst_data_type, &count, &c_mask, &c_scales),
2724 "could not get parameters of depthwise post-op");
2725
2726 weights_data_type = static_cast<memory::data_type>(c_weights_data_type);
2727 bias_data_type = static_cast<memory::data_type>(c_bias_data_type);
2728 dst_data_type = static_cast<memory::data_type>(c_dst_data_type);
2729 scales.resize(count);
2730
2731 mask = c_mask;
2732 for (dnnl_dim_t c = 0; c < count; ++c)
2733 scales[c] = c_scales[c];
2734 return;
2735 }
2736
2751 void append_binary(algorithm aalgorithm, const memory::desc &src1_desc) {
2753 convert_to_c(aalgorithm), &src1_desc.data),
2754 "could not append a binary post-op");
2755 }
2756
2763 int index, algorithm &aalgorithm, memory::desc &src1_desc) const {
2764 dnnl_alg_kind_t c_alg;
2765 const dnnl_memory_desc_t *data;
2767 dnnl_post_ops_get_params_binary(get(), index, &c_alg, &data),
2768 "could not get parameters of a binary post-op");
2769 aalgorithm = static_cast<dnnl::algorithm>(c_alg);
2770 src1_desc.data = *data;
2771 }
2772};
2773
2775template <>
2776struct handle_traits<dnnl_primitive_attr_t> {
2777 static dnnl_status_t destructor(dnnl_primitive_attr_t p) {
2779 }
2780};
2782
2786struct primitive_attr : public handle<dnnl_primitive_attr_t> {
2788
2791 dnnl_primitive_attr_t result;
2793 "could not create primitive attribute");
2794 reset(result);
2795 }
2796
2803 : handle<dnnl_primitive_attr_t>(attr) {}
2804
2810 "could not get scratchpad mode primitive attribute");
2811 return scratchpad_mode(result);
2812 }
2813
2819 get(), dnnl::convert_to_c(mode)),
2820 "could not set scratchpad mode primitive attribute");
2821 }
2822
2832 void get_output_scales(int &mask, std::vector<float> &scales) const {
2833 dnnl_dim_t count;
2834 int c_mask;
2835 const float *c_scales;
2837 get(), &count, &c_mask, &c_scales),
2838 "could not get output scales primitive attribute");
2839 scales.resize(count);
2840
2841 mask = c_mask;
2842 for (dnnl_dim_t c = 0; c < count; ++c)
2843 scales[c] = c_scales[c];
2844 }
2845
2888 void set_output_scales(int mask, const std::vector<float> &scales) {
2891 get(), (dnnl_dim_t)scales.size(), mask, scales.data()),
2892 "could not set output scales primitive attribute");
2893 }
2894
2906 void get_scales(int arg, int &mask, std::vector<float> &scales) const {
2907 dnnl_dim_t count;
2908 int c_mask;
2909 const float *c_scales;
2911 get(), arg, &count, &c_mask, &c_scales),
2912 "could not get scales primitive attributes");
2913 scales.resize(count);
2914
2915 mask = c_mask;
2916 for (dnnl_dim_t c = 0; c < count; ++c)
2917 scales[c] = c_scales[c];
2918 }
2919
2936 void set_scales(int arg, int mask, const std::vector<float> &scales) {
2939 (dnnl_dim_t)scales.size(), mask, scales.data()),
2940 "could not set scales primitive attribute");
2941 }
2942
2954 int arg, int &mask, std::vector<int32_t> &zero_points) const {
2955 dnnl_dim_t count;
2956 int c_mask;
2957 const int32_t *c_zero_points;
2959 get(), arg, &count, &c_mask, &c_zero_points),
2960 "could not get zero points primitive attribute");
2961 zero_points.resize(count);
2962
2963 mask = c_mask;
2964 for (dnnl_dim_t c = 0; c < count; ++c)
2965 zero_points[c] = c_zero_points[c];
2966 }
2967
2989 int arg, int mask, const std::vector<int32_t> &zero_points) {
2991 (dnnl_dim_t)zero_points.size(), mask,
2992 zero_points.data()),
2993 "could not set zero points primitive attribute");
2994 }
2995
2999 const post_ops get_post_ops() const {
3000 post_ops result;
3001 const_dnnl_post_ops_t c_result;
3003 "could not get post-ops primitive attribute");
3004 result.reset(const_cast<dnnl_post_ops_t>(c_result), true);
3005 return result;
3006 }
3007
3016 void set_post_ops(const post_ops ops) {
3018 "could not set post-ops primitive attribute");
3019 }
3020
3054 void set_rnn_data_qparams(float scale, float shift) {
3057 "could not set RNN data quantization parameters primitive "
3058 "attribute");
3059 }
3060
3070 void get_rnn_data_qparams(float &scale, float &shift) {
3071 float c_scale, c_shift;
3073 get(), &c_scale, &c_shift),
3074 "could not set RNN data quantization parameters primitive "
3075 "attribute");
3076 scale = c_scale;
3077 shift = c_shift;
3078 }
3079
3106 void set_rnn_weights_qparams(int mask, const std::vector<float> &scales) {
3108 (int)scales.size(), mask, scales.data()),
3109 "could not set RNN weights quantization parameters primitive "
3110 "attribute");
3111 }
3112
3132 void get_rnn_weights_qparams(int &mask, std::vector<float> &scales) {
3133 dnnl_dim_t count;
3134 int c_mask;
3135 const float *c_scales;
3137 get(), &count, &c_mask, &c_scales),
3138 "could not get primitive RNN weights quantization "
3139 "parameters attributes");
3140 scales.resize(count);
3141
3142 mask = c_mask;
3143 for (dnnl_dim_t c = 0; c < count; c++)
3144 scales[c] = c_scales[c];
3145 }
3146
3148 // The low-precision configuration of the RNN primitives expect input
3149 // weights to use the signed 8-bit integer data type. The scaling factors
3150 // are used to quantize floating-point data to signed integer and must be
3174 int mask, const std::vector<float> &scales) {
3177 get(), (int)scales.size(), mask, scales.data()),
3178 "could not set primitive RNN weights projection quantization "
3179 "parameters attributes");
3180 }
3181
3202 int &mask, std::vector<float> &scales) {
3203 dnnl_dim_t count;
3204 int c_mask;
3205 const float *c_scales;
3208 get(), &count, &c_mask, &c_scales),
3209 "could not get primitive RNN weights projection quantization "
3210 "parameters attributes");
3211 scales.resize(count);
3212
3213 mask = c_mask;
3214 for (dnnl_dim_t c = 0; c < count; c++)
3215 scales[c] = c_scales[c];
3216 }
3217};
3218
3220
3223
3225struct primitive_desc_base : public handle<dnnl_primitive_desc_t> {
3227
3230
3233 engine get_engine() const { return engine::query(*this); }
3234
3237 const char *impl_info_str() const {
3238 const char *res;
3240 get(), dnnl_query_impl_info_str, 0, &res),
3241 "could not retrieve implementation info string from a "
3242 "primitive descriptor");
3243 return res;
3244 }
3245
3250 memory::dim res;
3252 get(), dnnl::convert_to_c(what), 0, &res);
3253 return status == dnnl_success ? res : 0;
3254 }
3255
3270 memory::desc query_md(query what, int idx = 0) const {
3271 std::vector<query> valid_q {query::src_md, query::diff_src_md,
3275 if (!std::any_of(valid_q.cbegin(), valid_q.cend(),
3276 [=](query q) { return what == q; }))
3277 DNNL_THROW_ERROR(dnnl_invalid_arguments,
3278 "memory descriptor query is invalid");
3279
3281 get(), dnnl::convert_to_c(what), idx);
3282 return cdesc ? memory::desc(*cdesc) : memory::desc();
3283 }
3284
3290 memory::desc src_desc(int idx) const {
3291 return query_md(query::src_md, idx);
3292 }
3293
3299 memory::desc dst_desc(int idx) const {
3300 return query_md(query::dst_md, idx);
3301 }
3302
3309 return query_md(query::weights_md, idx);
3310 }
3311
3318 return query_md(query::diff_src_md, idx);
3319 }
3320
3327 return query_md(query::diff_dst_md, idx);
3328 }
3329
3336 return query_md(query::diff_weights_md, idx);
3337 }
3338
3339 // Separate versions without the index argument for documentation
3340 // purposes.
3341
3346 memory::desc src_desc() const { return src_desc(0); }
3347
3352 memory::desc dst_desc() const { return dst_desc(0); }
3353
3359
3365
3371
3377
3383 return query_md(query::workspace_md, 0);
3384 }
3385
3392 return query_md(query::scratchpad_md, 0);
3393 }
3394
3398 dnnl_engine_t c_engine;
3401 0, &c_engine),
3402 "could not retrieve scratchpad engine from a primitive "
3403 "descriptor");
3404 return engine(c_engine, true);
3405 }
3406
3410 const_dnnl_primitive_attr_t const_c_attr;
3412 "could not get attributes from a primitive descriptor");
3413 dnnl_primitive_attr_t c_attr;
3414 error::wrap_c_api(dnnl_primitive_attr_clone(&c_attr, const_c_attr),
3415 "could not clone primitive attributes");
3416 return primitive_attr(c_attr);
3417 }
3418
3424 dnnl_query_primitive_kind, 0, (void *)&kind),
3425 "could not get primitive kind from a primitive descriptor");
3426 return static_cast<dnnl::primitive::kind>(kind);
3427 }
3428
3429protected:
3434 dnnl_primitive_desc_t new_pd;
3436 "could not clone a primitive descriptor");
3437 reset(new_pd);
3438 }
3439
3455 : primitive_desc_base(pd, prim_kind, dnnl::prop_kind::undef) {}
3456
3469 dnnl::primitive::kind prim_kind, dnnl::prop_kind aprop_kind)
3470 : primitive_desc_base(pd, prim_kind, aprop_kind, aprop_kind) {}
3471
3486 dnnl::primitive::kind prim_kind, dnnl::prop_kind prop_kind1,
3487 dnnl::prop_kind prop_kind2) {
3488 // It is OK to pass an empty primitive descriptor
3489 if (pd == nullptr) return;
3490
3491 dnnl_status_t rc;
3492
3493 dnnl_primitive_kind_t c_prim_kind = convert_to_c(prim_kind);
3494 dnnl_prop_kind_t c_prop_kind1 = convert_to_c(prop_kind1);
3495 dnnl_prop_kind_t c_prop_kind2 = convert_to_c(prop_kind2);
3496
3497 // Check that primitive kind matches
3498 dnnl_primitive_kind_t pd_kind;
3500 pd, dnnl_query_primitive_kind, 0, (void *)&pd_kind);
3502 rc, "could not get primitive kind from a primitive descriptor");
3503 if (pd_kind != c_prim_kind)
3504 DNNL_THROW_ERROR(dnnl_invalid_arguments,
3505 "primitive descriptor operation kind mismatch");
3506
3507 // Check that propagation kind matches
3508 dnnl_prop_kind_t pd_prop_kind;
3510 pd, dnnl_query_prop_kind, 0, (void *)&pd_prop_kind);
3511
3512 // Something went wrong
3513 if (rc != dnnl_success && rc != dnnl_unimplemented)
3514 DNNL_THROW_ERROR(dnnl_invalid_arguments,
3515 "could not get propagation kind from the primitive "
3516 "descriptor");
3517
3518 // Everything is fine
3519 if ((rc == dnnl_unimplemented && c_prop_kind1 == dnnl_prop_kind_undef)
3520 || (rc == dnnl_success
3521 && (pd_prop_kind == c_prop_kind1
3522 || pd_prop_kind == c_prop_kind2))) {
3523 reset_with_clone(pd);
3524 return;
3525 }
3526
3527 // We could get the propagation kind but there is a mismatch
3528 DNNL_THROW_ERROR(dnnl_invalid_arguments,
3529 "primitive descriptor propagation kind mismatch");
3530 }
3531
3532 using base = primitive_desc_base;
3533};
3534
3536
3545
3547struct reorder : public primitive {
3551
3553 primitive_desc() = default;
3554
3572 primitive_desc(const engine &src_engine, const memory::desc &src_md,
3573 const engine &dst_engine, const memory::desc &dst_md,
3574 const primitive_attr &attr = primitive_attr(),
3575 bool allow_empty = false) {
3576 dnnl_primitive_desc_t result;
3578 &src_md.data, src_engine.get(), &dst_md.data,
3579 dst_engine.get(), attr.get());
3580 if (!allow_empty)
3582 "could not create a primitive descriptor for a reorder "
3583 "primitive");
3585 }
3586
3598 primitive_desc(const memory &src, const memory &dst,
3599 const primitive_attr &attr = primitive_attr(),
3600 bool allow_empty = false) {
3601 dnnl_primitive_desc_t result;
3602 auto src_md = src.get_desc();
3603 auto dst_md = dst.get_desc();
3605 &src_md.data, src.get_engine().get(), &dst_md.data,
3606 dst.get_engine().get(), attr.get());
3607 if (!allow_empty)
3609 "could not create a primitive descriptor for a reorder "
3610 "primitive");
3612 }
3613
3620
3625 }
3626
3631 }
3632
3634 memory::desc src_desc() const { return base::src_desc(0); }
3635
3637 memory::desc dst_desc() const { return base::dst_desc(0); }
3638 };
3639
3641 reorder() = default;
3642
3645 reorder(const primitive_desc &pd) : primitive(pd.get()) {}
3646
3654 reorder(const memory &src, const memory &dst,
3655 const primitive_attr &attr = primitive_attr())
3656 : primitive(primitive_desc(src, dst, attr).get()) {}
3657
3658 using primitive::execute;
3659
3666 void execute(const stream &astream, memory &src, memory &dst) const {
3667 primitive::execute(astream, {{DNNL_ARG_FROM, src}, {DNNL_ARG_TO, dst}});
3668 }
3669};
3670
3672
3680
3682inline std::vector<dnnl_memory_desc_t> convert_to_c(
3683 const std::vector<memory::desc> &mems) {
3684 std::vector<dnnl_memory_desc_t> c_mems;
3685 c_mems.reserve(mems.size());
3686 for (const auto &s : mems)
3687 c_mems.push_back(s.data);
3688 return c_mems;
3689}
3691
3693struct concat : public primitive {
3697
3699 primitive_desc() = default;
3700
3711 primitive_desc(const memory::desc &dst, int concat_dimension,
3712 const std::vector<memory::desc> &srcs, const engine &aengine,
3713 const primitive_attr &attr = primitive_attr()) {
3714 auto c_srcs = convert_to_c(srcs);
3715
3716 dnnl_primitive_desc_t result;
3719 (int)c_srcs.size(), concat_dimension, c_srcs.data(),
3720 attr.get(), aengine.get()),
3721 "could not create a primitive descriptor for a concat "
3722 "primitive");
3723 reset(result);
3724 }
3725
3738 primitive_desc(int concat_dimension,
3739 const std::vector<memory::desc> &srcs, const engine &aengine,
3740 const primitive_attr &attr = primitive_attr()) {
3741 auto c_api_srcs = convert_to_c(srcs);
3742
3743 dnnl_primitive_desc_t result;
3745 dnnl_concat_primitive_desc_create(&result, nullptr,
3746 (int)c_api_srcs.size(), concat_dimension,
3747 c_api_srcs.data(), attr.get(), aengine.get()),
3748 "could not create a primitive descriptor for a concat "
3749 "primitive");
3750 reset(result);
3751 }
3752
3759
3761 memory::desc src_desc(int idx = 0) const { return base::src_desc(idx); }
3762
3764 memory::desc dst_desc() const { return base::dst_desc(0); }
3765 };
3766
3768 concat() = default;
3769
3772 concat(const primitive_desc &pd) : primitive(pd.get()) {}
3773};
3774
3776
3784
3786struct sum : public primitive {
3790
3792 primitive_desc() = default;
3793
3803 const std::vector<float> &scales,
3804 const std::vector<memory::desc> &srcs, const engine &aengine,
3805 const primitive_attr &attr = primitive_attr()) {
3806 validate_container_size(scales,
3807 "counts of scales and sources are not equal",
3808 (int)srcs.size(), (int)srcs.size());
3809
3810 auto c_api_srcs = convert_to_c(srcs);
3811
3812 dnnl_primitive_desc_t result;
3815 (int)c_api_srcs.size(), scales.data(),
3816 c_api_srcs.data(), attr.get(), aengine.get()),
3817 "could not create a primitive descriptor for a sum "
3818 "primitive");
3819 reset(result);
3820 }
3821
3832 primitive_desc(const std::vector<float> &scales,
3833 const std::vector<memory::desc> &srcs, const engine &aengine,
3834 const primitive_attr &attr = primitive_attr()) {
3835 validate_container_size(scales,
3836 "counts of scales and sources are not equal",
3837 (int)srcs.size(), (int)srcs.size());
3838
3839 auto c_api_srcs = convert_to_c(srcs);
3840 dnnl_primitive_desc_t result;
3842 dnnl_sum_primitive_desc_create(&result, nullptr,
3843 (int)c_api_srcs.size(), scales.data(),
3844 c_api_srcs.data(), attr.get(), aengine.get()),
3845 "could not create a primitive descriptor for a sum "
3846 "primitive");
3847 reset(result);
3848 }
3849
3856
3858 memory::desc src_desc(int idx = 0) const { return base::src_desc(idx); }
3859
3861 memory::desc dst_desc() const { return base::dst_desc(0); }
3862 };
3863
3865 sum() = default;
3866
3869 sum(const primitive_desc &pd) : primitive(pd.get()) {}
3870};
3871
3873
3876
3881
3882 primitive_desc() = default;
3883
3907 const engine &aengine, const_dnnl_primitive_desc_t hint_fwd_pd,
3908 bool allow_empty = false)
3909 : allow_empty_(allow_empty) {
3910 dnnl_primitive_desc_iterator_t iterator = nullptr;
3912 desc, attr ? attr->get() : nullptr, aengine.get(), hint_fwd_pd);
3913 if (!allow_empty)
3915 status, "could not create a primitive descriptor iterator");
3916 pd_iterator.reset(iterator);
3917 fetch_impl();
3918 }
3919
3924 bool next_impl() {
3926 = dnnl_primitive_desc_iterator_next(pd_iterator.get());
3927 if (status == dnnl_iterator_ends) return false;
3929 status, "could not advance a primitive descriptor iterator");
3930 fetch_impl();
3931 return true;
3932 }
3933
3934private:
3935 bool allow_empty_ = false;
3937 void fetch_impl() {
3939 pd_iterator.get(allow_empty_));
3940 error::wrap_c_api(pd != nullptr || allow_empty_ ? dnnl_success
3942 "could not fetch a primitive descriptor from a primitive "
3943 "descriptor iterator");
3944 reset(pd);
3945 }
3946};
3947
3949
3959
3963 struct desc {
3965
3996 desc(prop_kind aprop_kind, algorithm aalgorithm,
3997 const memory::desc &src_desc, const memory::desc &weights_desc,
3998 const memory::desc &bias_desc, const memory::desc &dst_desc,
3999 const memory::dims &strides, const memory::dims &padding_l,
4000 const memory::dims &padding_r) {
4001 memory::validate_dims(strides, src_desc.data.ndims - 2);
4002 memory::validate_dims(padding_l, src_desc.data.ndims - 2);
4003 memory::validate_dims(padding_r, src_desc.data.ndims - 2);
4006 dnnl::convert_to_c(aprop_kind),
4007 convert_to_c(aalgorithm), &src_desc.data,
4008 &weights_desc.data, &bias_desc.data, &dst_desc.data,
4009 &strides[0], &padding_l[0], &padding_r[0]),
4010 "could not create a descriptor for a convolution forward "
4011 "propagation primitive");
4012 }
4013
4042 desc(prop_kind aprop_kind, algorithm aalgorithm,
4043 const memory::desc &src_desc, const memory::desc &weights_desc,
4044 const memory::desc &dst_desc, const memory::dims &strides,
4045 const memory::dims &padding_l, const memory::dims &padding_r) {
4046 memory::validate_dims(strides, src_desc.data.ndims - 2);
4047 memory::validate_dims(padding_l, src_desc.data.ndims - 2);
4048 memory::validate_dims(padding_r, src_desc.data.ndims - 2);
4051 dnnl::convert_to_c(aprop_kind),
4052 convert_to_c(aalgorithm), &src_desc.data,
4053 &weights_desc.data, nullptr, &dst_desc.data,
4054 &strides[0], &padding_l[0], &padding_r[0]),
4055 "could not create a descriptor for a convolution forward "
4056 "propagation primitive");
4057 }
4058
4091 desc(prop_kind aprop_kind, algorithm aalgorithm,
4092 const memory::desc &src_desc, const memory::desc &weights_desc,
4093 const memory::desc &bias_desc, const memory::desc &dst_desc,
4094 const memory::dims &strides, const memory::dims &dilates,
4095 const memory::dims &padding_l, const memory::dims &padding_r) {
4096 memory::validate_dims(strides, src_desc.data.ndims - 2);
4097 memory::validate_dims(dilates, src_desc.data.ndims - 2);
4098 memory::validate_dims(padding_l, src_desc.data.ndims - 2);
4099 memory::validate_dims(padding_r, src_desc.data.ndims - 2);
4101 dnnl::convert_to_c(aprop_kind),
4102 convert_to_c(aalgorithm), &src_desc.data,
4103 &weights_desc.data, &bias_desc.data,
4104 &dst_desc.data, &strides[0], &dilates[0],
4105 &padding_l[0], &padding_r[0]),
4106 "could not create a descriptor for a dilated convolution "
4107 "forward propagation primitive");
4108 }
4109
4140 desc(prop_kind aprop_kind, algorithm aalgorithm,
4141 const memory::desc &src_desc, const memory::desc &weights_desc,
4142 const memory::desc &dst_desc, const memory::dims &strides,
4143 const memory::dims &dilates, const memory::dims &padding_l,
4144 const memory::dims &padding_r) {
4145 memory::validate_dims(strides, src_desc.data.ndims - 2);
4146 memory::validate_dims(dilates, src_desc.data.ndims - 2);
4147 memory::validate_dims(padding_l, src_desc.data.ndims - 2);
4148 memory::validate_dims(padding_r, src_desc.data.ndims - 2);
4150 dnnl::convert_to_c(aprop_kind),
4151 convert_to_c(aalgorithm), &src_desc.data,
4152 &weights_desc.data, nullptr,
4153 &dst_desc.data, &strides[0], &dilates[0],
4154 &padding_l[0], &padding_r[0]),
4155 "could not create a descriptor for a dilated convolution "
4156 "forward propagation primitive");
4157 }
4158 };
4159
4163 primitive_desc() = default;
4164
4175 primitive_desc(const desc &adesc, const engine &aengine,
4176 bool allow_empty = false)
4178 &adesc.data, nullptr, aengine, nullptr, allow_empty) {}
4179
4191 primitive_desc(const desc &adesc, const primitive_attr &attr,
4192 const engine &aengine, bool allow_empty = false)
4194 &adesc.data, &attr, aengine, nullptr, allow_empty) {}
4195
4203 : dnnl::primitive_desc(pd, dnnl::primitive::kind::convolution,
4206
4208 memory::desc src_desc() const { return base::src_desc(0); }
4209
4212
4214 memory::desc dst_desc() const { return base::dst_desc(0); }
4215
4221 };
4222
4225
4230};
4231
4234
4236 struct desc {
4238
4264 desc(algorithm aalgorithm, const memory::desc &diff_src_desc,
4265 const memory::desc &weights_desc,
4266 const memory::desc &diff_dst_desc, const memory::dims &strides,
4267 const memory::dims &padding_l, const memory::dims &padding_r) {
4268 memory::validate_dims(strides, diff_src_desc.data.ndims - 2);
4269 memory::validate_dims(padding_l, diff_src_desc.data.ndims - 2);
4270 memory::validate_dims(padding_r, diff_src_desc.data.ndims - 2);
4273 convert_to_c(aalgorithm), &diff_src_desc.data,
4274 &weights_desc.data, &diff_dst_desc.data,
4275 &strides[0], &padding_l[0], &padding_r[0]),
4276 "could not create a descriptor for a convolution backward "
4277 "propagation primitive");
4278 }
4279
4307 desc(algorithm aalgorithm, const memory::desc &diff_src_desc,
4308 const memory::desc &weights_desc,
4309 const memory::desc &diff_dst_desc, const memory::dims &strides,
4310 const memory::dims &dilates, const memory::dims &padding_l,
4311 const memory::dims &padding_r) {
4312 memory::validate_dims(strides, diff_src_desc.data.ndims - 2);
4313 memory::validate_dims(dilates, diff_src_desc.data.ndims - 2);
4314 memory::validate_dims(padding_l, diff_src_desc.data.ndims - 2);
4315 memory::validate_dims(padding_r, diff_src_desc.data.ndims - 2);
4318 convert_to_c(aalgorithm), &diff_src_desc.data,
4319 &weights_desc.data, &diff_dst_desc.data,
4320 &strides[0], &dilates[0], &padding_l[0],
4321 &padding_r[0]),
4322 "could not create a descriptor for a dilated convolution "
4323 "backward propagation primitive");
4324 }
4325 };
4326
4330 primitive_desc() = default;
4331
4345 primitive_desc(const desc &adesc, const engine &aengine,
4346 const convolution_forward::primitive_desc &hint_fwd_pd,
4347 bool allow_empty = false)
4348 : dnnl::primitive_desc(&adesc.data, nullptr, aengine,
4349 hint_fwd_pd.get(), allow_empty) {}
4350
4365 primitive_desc(const desc &adesc, const primitive_attr &attr,
4366 const engine &aengine,
4367 const convolution_forward::primitive_desc &hint_fwd_pd,
4368 bool allow_empty = false)
4369 : dnnl::primitive_desc(&adesc.data, &attr, aengine,
4370 hint_fwd_pd.get(), allow_empty) {}
4371
4379 : dnnl::primitive_desc(pd, dnnl::primitive::kind::convolution,
4381
4384
4387
4390 };
4391
4394
4399};
4400
4404 struct desc {
4406
4434 desc(algorithm aalgorithm, const memory::desc &src_desc,
4435 const memory::desc &diff_weights_desc,
4436 const memory::desc &diff_bias_desc,
4437 const memory::desc &diff_dst_desc, const memory::dims &strides,
4438 const memory::dims &padding_l, const memory::dims &padding_r) {
4439 memory::validate_dims(strides, src_desc.data.ndims - 2);
4440 memory::validate_dims(padding_l, src_desc.data.ndims - 2);
4441 memory::validate_dims(padding_r, src_desc.data.ndims - 2);
4444 convert_to_c(aalgorithm), &src_desc.data,
4445 &diff_weights_desc.data, &diff_bias_desc.data,
4446 &diff_dst_desc.data, &strides[0], &padding_l[0],
4447 &padding_r[0]),
4448 "could not create a descriptor for a convolution weights "
4449 "update primitive");
4450 }
4451
4477 desc(algorithm aalgorithm, const memory::desc &src_desc,
4478 const memory::desc &diff_weights_desc,
4479 const memory::desc &diff_dst_desc, const memory::dims &strides,
4480 const memory::dims &padding_l, const memory::dims &padding_r) {
4481 memory::validate_dims(strides, src_desc.data.ndims - 2);
4482 memory::validate_dims(padding_l, src_desc.data.ndims - 2);
4483 memory::validate_dims(padding_r, src_desc.data.ndims - 2);
4485 convert_to_c(aalgorithm), &src_desc.data,
4486 &diff_weights_desc.data, nullptr,
4487 &diff_dst_desc.data, &strides[0],
4488 &padding_l[0], &padding_r[0]),
4489 "could not create a descriptor for a convolution weights "
4490 "update primitive");
4491 }
4492
4522 desc(algorithm aalgorithm, const memory::desc &src_desc,
4523 const memory::desc &diff_weights_desc,
4524 const memory::desc &diff_bias_desc,
4525 const memory::desc &diff_dst_desc, const memory::dims &strides,
4526 const memory::dims &dilates, const memory::dims &padding_l,
4527 const memory::dims &padding_r) {
4528 memory::validate_dims(strides, src_desc.data.ndims - 2);
4529 memory::validate_dims(dilates, src_desc.data.ndims - 2);
4530 memory::validate_dims(padding_l, src_desc.data.ndims - 2);
4531 memory::validate_dims(padding_r, src_desc.data.ndims - 2);
4534 convert_to_c(aalgorithm), &src_desc.data,
4535 &diff_weights_desc.data, &diff_bias_desc.data,
4536 &diff_dst_desc.data, &strides[0], &dilates[0],
4537 &padding_l[0], &padding_r[0]),
4538 "could not create a descriptor for a dilated convolution "
4539 "weights gradient primitive");
4540 }
4541
4569 desc(algorithm aalgorithm, const memory::desc &src_desc,
4570 const memory::desc &diff_weights_desc,
4571 const memory::desc &diff_dst_desc, const memory::dims &strides,
4572 const memory::dims &dilates, const memory::dims &padding_l,
4573 const memory::dims &padding_r) {
4574 memory::validate_dims(strides, src_desc.data.ndims - 2);
4575 memory::validate_dims(dilates, src_desc.data.ndims - 2);
4576 memory::validate_dims(padding_l, src_desc.data.ndims - 2);
4577 memory::validate_dims(padding_r, src_desc.data.ndims - 2);
4580 convert_to_c(aalgorithm), &src_desc.data,
4581 &diff_weights_desc.data, nullptr,
4582 &diff_dst_desc.data, &strides[0], &dilates[0],
4583 &padding_l[0], &padding_r[0]),
4584 "could not create a descriptor for a dilated convolution "
4585 "weights gradient primitive");
4586 }
4587 };
4588
4592 primitive_desc() = default;
4593
4606 primitive_desc(const desc &adesc, const engine &aengine,
4607 const convolution_forward::primitive_desc &hint_fwd_pd,
4608 bool allow_empty = false)
4609 : dnnl::primitive_desc(&adesc.data, nullptr, aengine,
4610 hint_fwd_pd.get(), allow_empty) {}
4611
4625 primitive_desc(const desc &adesc, const primitive_attr &attr,
4626 const engine &aengine,
4627 const convolution_forward::primitive_desc &hint_fwd_pd,
4628 bool allow_empty = false)
4629 : dnnl::primitive_desc(&adesc.data, &attr, aengine,
4630 hint_fwd_pd.get(), allow_empty) {}
4631
4639 : dnnl::primitive_desc(pd, dnnl::primitive::kind::convolution,
4641
4643 memory::desc src_desc() const { return base::src_desc(0); }
4644
4647 return base::diff_weights_desc(0);
4648 }
4649
4652
4658 return base::diff_weights_desc(1);
4659 }
4660 };
4661
4664
4669};
4670
4672//
4680
4684 struct desc {
4686
4716 desc(prop_kind aprop_kind, algorithm aalgorithm,
4717 const memory::desc &src_desc, const memory::desc &weights_desc,
4718 const memory::desc &bias_desc, const memory::desc &dst_desc,
4719 const memory::dims &strides, const memory::dims &padding_l,
4720 const memory::dims &padding_r) {
4721 memory::validate_dims(strides, src_desc.data.ndims - 2);
4722 memory::validate_dims(padding_l, src_desc.data.ndims - 2);
4723 memory::validate_dims(padding_r, src_desc.data.ndims - 2);
4726 dnnl::convert_to_c(aprop_kind),
4727 convert_to_c(aalgorithm), &src_desc.data,
4728 &weights_desc.data, &bias_desc.data, &dst_desc.data,
4729 &strides[0], &padding_l[0], &padding_r[0]),
4730 "could not create a descriptor for a deconvolution forward "
4731 "propagation primitive");
4732 }
4733
4761 desc(prop_kind aprop_kind, algorithm aalgorithm,
4762 const memory::desc &src_desc, const memory::desc &weights_desc,
4763 const memory::desc &dst_desc, const memory::dims &strides,
4764 const memory::dims &padding_l, const memory::dims &padding_r) {
4765 memory::validate_dims(strides, src_desc.data.ndims - 2);
4766 memory::validate_dims(padding_l, src_desc.data.ndims - 2);
4767 memory::validate_dims(padding_r, src_desc.data.ndims - 2);
4770 dnnl::convert_to_c(aprop_kind),
4771 convert_to_c(aalgorithm), &src_desc.data,
4772 &weights_desc.data, nullptr, &dst_desc.data,
4773 &strides[0], &padding_l[0], &padding_r[0]),
4774 "could not create a descriptor for a deconvolution forward "
4775 "propagation primitive");
4776 }
4777
4809 desc(prop_kind aprop_kind, algorithm aalgorithm,
4810 const memory::desc &src_desc, const memory::desc &weights_desc,
4811 const memory::desc &bias_desc, const memory::desc &dst_desc,
4812 const memory::dims &strides, const memory::dims &dilates,
4813 const memory::dims &padding_l, const memory::dims &padding_r) {
4814 memory::validate_dims(strides, src_desc.data.ndims - 2);
4815 memory::validate_dims(dilates, src_desc.data.ndims - 2);
4816 memory::validate_dims(padding_l, src_desc.data.ndims - 2);
4817 memory::validate_dims(padding_r, src_desc.data.ndims - 2);
4819 &data, dnnl::convert_to_c(aprop_kind),
4820 convert_to_c(aalgorithm), &src_desc.data,
4821 &weights_desc.data, &bias_desc.data,
4822 &dst_desc.data, &strides[0], &dilates[0],
4823 &padding_l[0], &padding_r[0]),
4824 "could not create a descriptor for a dilated deconvolution "
4825 "forward propagation primitive");
4826 }
4827
4857 desc(prop_kind aprop_kind, algorithm aalgorithm,
4858 const memory::desc &src_desc, const memory::desc &weights_desc,
4859 const memory::desc &dst_desc, const memory::dims &strides,
4860 const memory::dims &dilates, const memory::dims &padding_l,
4861 const memory::dims &padding_r) {
4862 memory::validate_dims(strides, src_desc.data.ndims - 2);
4863 memory::validate_dims(dilates, src_desc.data.ndims - 2);
4864 memory::validate_dims(padding_l, src_desc.data.ndims - 2);
4865 memory::validate_dims(padding_r, src_desc.data.ndims - 2);
4867 &data, dnnl::convert_to_c(aprop_kind),
4868 convert_to_c(aalgorithm), &src_desc.data,
4869 &weights_desc.data, nullptr,
4870 &dst_desc.data, &strides[0], &dilates[0],
4871 &padding_l[0], &padding_r[0]),
4872 "could not create a descriptor for a dilated deconvolution "
4873 "forward propagation primitive");
4874 }
4875 };
4876
4880 primitive_desc() = default;
4881
4892 primitive_desc(const desc &adesc, const engine &aengine,
4893 bool allow_empty = false)
4895 &adesc.data, nullptr, aengine, nullptr, allow_empty) {}
4896
4908 primitive_desc(const desc &adesc, const primitive_attr &attr,
4909 const engine &aengine, bool allow_empty = false)
4911 &adesc.data, &attr, aengine, nullptr, allow_empty) {}
4912
4920 : dnnl::primitive_desc(pd, dnnl::primitive::kind::deconvolution,
4923
4925 memory::desc src_desc() const { return base::src_desc(0); }
4926
4929
4931 memory::desc dst_desc() const { return base::dst_desc(0); }
4932
4935 };
4936
4939
4944};
4945
4949 struct desc {
4951
4976 desc(algorithm aalgorithm, const memory::desc &diff_src_desc,
4977 const memory::desc &weights_desc,
4978 const memory::desc &diff_dst_desc, const memory::dims &strides,
4979 const memory::dims &padding_l, const memory::dims &padding_r) {
4980 memory::validate_dims(strides, diff_src_desc.data.ndims - 2);
4981 memory::validate_dims(padding_l, diff_src_desc.data.ndims - 2);
4982 memory::validate_dims(padding_r, diff_src_desc.data.ndims - 2);
4985 convert_to_c(aalgorithm), &diff_src_desc.data,
4986 &weights_desc.data, &diff_dst_desc.data,
4987 &strides[0], &padding_l[0], &padding_r[0]),
4988 "could not create a descriptor for a deconvolution "
4989 "backward propagation primitive");
4990 }
4991
5018 desc(algorithm aalgorithm, const memory::desc &diff_src_desc,
5019 const memory::desc &weights_desc,
5020 const memory::desc &diff_dst_desc, const memory::dims &strides,
5021 const memory::dims &dilates, const memory::dims &padding_l,
5022 const memory::dims &padding_r) {
5023 memory::validate_dims(strides, diff_src_desc.data.ndims - 2);
5024 memory::validate_dims(dilates, diff_src_desc.data.ndims - 2);
5025 memory::validate_dims(padding_l, diff_src_desc.data.ndims - 2);
5026 memory::validate_dims(padding_r, diff_src_desc.data.ndims - 2);
5029 convert_to_c(aalgorithm), &diff_src_desc.data,
5030 &weights_desc.data, &diff_dst_desc.data,
5031 &strides[0], &dilates[0], &padding_l[0],
5032 &padding_r[0]),
5033 "could not create a descriptor for a dilated deconvolution "
5034 "backward propagation primitive");
5035 }
5036 };
5037
5041 primitive_desc() = default;
5042
5056 primitive_desc(const desc &adesc, const engine &aengine,
5057 const deconvolution_forward::primitive_desc &hint_fwd_pd,
5058 bool allow_empty = false)
5059 : dnnl::primitive_desc(&adesc.data, nullptr, aengine,
5060 hint_fwd_pd.get(), allow_empty) {}
5061
5076 primitive_desc(const desc &adesc, const primitive_attr &attr,
5077 const engine &aengine,
5078 const deconvolution_forward::primitive_desc &hint_fwd_pd,
5079 bool allow_empty = false)
5080 : dnnl::primitive_desc(&adesc.data, &attr, aengine,
5081 hint_fwd_pd.get(), allow_empty) {}
5082
5090 : dnnl::primitive_desc(pd, dnnl::primitive::kind::deconvolution,
5092
5095
5098
5101 };
5102
5105
5110};
5111
5115 struct desc {
5117
5144 desc(algorithm aalgorithm, const memory::desc &src_desc,
5145 const memory::desc &diff_weights_desc,
5146 const memory::desc &diff_bias_desc,
5147 const memory::desc &diff_dst_desc, const memory::dims &strides,
5148 const memory::dims &padding_l, const memory::dims &padding_r) {
5149 memory::validate_dims(strides, src_desc.data.ndims - 2);
5150 memory::validate_dims(padding_l, src_desc.data.ndims - 2);
5151 memory::validate_dims(padding_r, src_desc.data.ndims - 2);
5154 convert_to_c(aalgorithm), &src_desc.data,
5155 &diff_weights_desc.data, &diff_bias_desc.data,
5156 &diff_dst_desc.data, &strides[0], &padding_l[0],
5157 &padding_r[0]),
5158 "could not create a descriptor for a deconvolution weights "
5159 "update primitive");
5160 }
5161
5186 desc(algorithm aalgorithm, const memory::desc &src_desc,
5187 const memory::desc &diff_weights_desc,
5188 const memory::desc &diff_dst_desc, const memory::dims &strides,
5189 const memory::dims &padding_l, const memory::dims &padding_r) {
5190 memory::validate_dims(strides, src_desc.data.ndims - 2);
5191 memory::validate_dims(padding_l, src_desc.data.ndims - 2);
5192 memory::validate_dims(padding_r, src_desc.data.ndims - 2);
5194 &data, convert_to_c(aalgorithm),
5195 &src_desc.data, &diff_weights_desc.data,
5196 nullptr, &diff_dst_desc.data, &strides[0],
5197 &padding_l[0], &padding_r[0]),
5198 "could not create a descriptor for a deconvolution weights "
5199 "update primitive");
5200 }
5201
5230 desc(algorithm aalgorithm, const memory::desc &src_desc,
5231 const memory::desc &diff_weights_desc,
5232 const memory::desc &diff_bias_desc,
5233 const memory::desc &diff_dst_desc, const memory::dims &strides,
5234 const memory::dims &dilates, const memory::dims &padding_l,
5235 const memory::dims &padding_r) {
5236 memory::validate_dims(strides, src_desc.data.ndims - 2);
5237 memory::validate_dims(dilates, src_desc.data.ndims - 2);
5238 memory::validate_dims(padding_l, src_desc.data.ndims - 2);
5239 memory::validate_dims(padding_r, src_desc.data.ndims - 2);
5242 convert_to_c(aalgorithm), &src_desc.data,
5243 &diff_weights_desc.data, &diff_bias_desc.data,
5244 &diff_dst_desc.data, &strides[0], &dilates[0],
5245 &padding_l[0], &padding_r[0]),
5246 "could not create a descriptor for a dilated deconvolution "
5247 "weights gradient primitive");
5248 }
5249
5276 desc(algorithm aalgorithm, const memory::desc &src_desc,
5277 const memory::desc &diff_weights_desc,
5278 const memory::desc &diff_dst_desc, const memory::dims &strides,
5279 const memory::dims &dilates, const memory::dims &padding_l,
5280 const memory::dims &padding_r) {
5281 memory::validate_dims(strides, src_desc.data.ndims - 2);
5282 memory::validate_dims(dilates, src_desc.data.ndims - 2);
5283 memory::validate_dims(padding_l, src_desc.data.ndims - 2);
5284 memory::validate_dims(padding_r, src_desc.data.ndims - 2);
5287 convert_to_c(aalgorithm), &src_desc.data,
5288 &diff_weights_desc.data, nullptr,
5289 &diff_dst_desc.data, &strides[0], &dilates[0],
5290 &padding_l[0], &padding_r[0]),
5291 "could not create a descriptor for a dilated deconvolution "
5292 "weights gradient primitive");
5293 }
5294 };
5295
5299 primitive_desc() = default;
5300
5314 primitive_desc(const desc &adesc, const engine &aengine,
5315 const deconvolution_forward::primitive_desc &hint_fwd_pd,
5316 bool allow_empty = false)
5317 : dnnl::primitive_desc(&adesc.data, nullptr, aengine,
5318 hint_fwd_pd.get(), allow_empty) {}
5319
5334 primitive_desc(const desc &adesc, const primitive_attr &attr,
5335 const engine &aengine,
5336 const deconvolution_forward::primitive_desc &hint_fwd_pd,
5337 bool allow_empty = false)
5338 : dnnl::primitive_desc(&adesc.data, &attr, aengine,
5339 hint_fwd_pd.get(), allow_empty) {}
5340
5348 : dnnl::primitive_desc(pd, dnnl::primitive::kind::deconvolution,
5350
5352 memory::desc src_desc() const { return base::src_desc(0); }
5353
5356 return base::diff_weights_desc(0);
5357 }
5358
5361
5364 return base::diff_weights_desc(1);
5365 }
5366 };
5367
5370
5375};
5376
5378
5387
5389struct lrn_forward : public primitive {
5391 struct desc {
5392 dnnl_lrn_desc_t data;
5393
5407 desc(prop_kind aprop_kind, algorithm aalgorithm,
5408 const memory::desc &data_desc, memory::dim local_size,
5409 float alpha, float beta, float k = 1.f) {
5411 dnnl::convert_to_c(aprop_kind),
5412 convert_to_c(aalgorithm), &data_desc.data,
5413 local_size, alpha, beta, k),
5414 "could not create a descriptor for a lrn forward "
5415 "propagation primitive");
5416 }
5417 };
5418
5422 primitive_desc() = default;
5423
5433 primitive_desc(const desc &adesc, const engine &aengine,
5434 bool allow_empty = false)
5436 &adesc.data, nullptr, aengine, nullptr, allow_empty) {}
5437
5448 primitive_desc(const desc &adesc, const primitive_attr &attr,
5449 const engine &aengine, bool allow_empty = false)
5451 &adesc.data, &attr, aengine, nullptr, allow_empty) {}
5452
5460 : dnnl::primitive_desc(pd, dnnl::primitive::kind::lrn,
5463
5465 memory::desc src_desc() const { return base::src_desc(0); }
5466
5468 memory::desc dst_desc() const { return base::dst_desc(0); }
5469
5472 };
5473
5475 lrn_forward() = default;
5476
5481};
5482
5484struct lrn_backward : public primitive {
5486 struct desc {
5487 dnnl_lrn_desc_t data;
5488
5501 desc(algorithm aalgorithm, const memory::desc &data_desc,
5502 const memory::desc &diff_data_desc, memory::dim local_size,
5503 float alpha, float beta, float k = 1.f) {
5505 dnnl_lrn_backward_desc_init(&data, convert_to_c(aalgorithm),
5506 &diff_data_desc.data, &data_desc.data, local_size,
5507 alpha, beta, k),
5508 "could not create a descriptor for a lrn backward "
5509 "propagation primitive");
5510 }
5511 };
5512
5516 primitive_desc() = default;
5517
5530 primitive_desc(const desc &adesc, const engine &aengine,
5531 const lrn_forward::primitive_desc &hint_fwd_pd,
5532 bool allow_empty = false)
5533 : dnnl::primitive_desc(&adesc.data, nullptr, aengine,
5534 hint_fwd_pd.get(), allow_empty) {}
5535
5549 primitive_desc(const desc &adesc, const primitive_attr &attr,
5550 const engine &aengine,
5551 const lrn_forward::primitive_desc &hint_fwd_pd,
5552 bool allow_empty = false)
5553 : dnnl::primitive_desc(&adesc.data, &attr, aengine,
5554 hint_fwd_pd.get(), allow_empty) {}
5555
5563 : dnnl::primitive_desc(pd, dnnl::primitive::kind::lrn,
5565
5568
5571
5574 };
5575
5577 lrn_backward() = default;
5578
5583};
5584
5586
5594
5598 struct desc {
5600
5625 desc(prop_kind aprop_kind, algorithm aalgorithm,
5626 const memory::desc &src_desc, const memory::desc &dst_desc,
5627 const memory::dims &strides, const memory::dims &kernel,
5628 const memory::dims &padding_l, const memory::dims &padding_r) {
5629 memory::validate_dims(strides, src_desc.data.ndims - 2);
5630 memory::validate_dims(kernel, src_desc.data.ndims - 2);
5631 memory::validate_dims(padding_l, src_desc.data.ndims - 2);
5632 memory::validate_dims(padding_r, src_desc.data.ndims - 2);
5634 dnnl::convert_to_c(aprop_kind),
5635 convert_to_c(aalgorithm), &src_desc.data,
5636 &dst_desc.data, &strides[0], &kernel[0],
5637 &padding_l[0], &padding_r[0]),
5638 "could not create a descriptor for a pooling forward "
5639 "propagation primitive");
5640 }
5641 };
5642
5646 primitive_desc() = default;
5647
5657 primitive_desc(const desc &adesc, const engine &aengine,
5658 bool allow_empty = false)
5660 &adesc.data, nullptr, aengine, nullptr, allow_empty) {}
5661
5672 primitive_desc(const desc &adesc, const primitive_attr &attr,
5673 const engine &aengine, bool allow_empty = false)
5675 &adesc.data, &attr, aengine, nullptr, allow_empty) {}
5676
5684 : dnnl::primitive_desc(pd, dnnl::primitive::kind::pooling,
5687
5689 memory::desc src_desc() const { return base::src_desc(0); }
5690
5692 memory::desc dst_desc() const { return base::dst_desc(0); }
5693
5696 };
5697
5699 pooling_forward() = default;
5700
5705};
5706
5710 struct desc {
5712
5734 desc(algorithm aalgorithm, const memory::desc &diff_src_desc,
5735 const memory::desc &diff_dst_desc, const memory::dims &strides,
5736 const memory::dims &kernel, const memory::dims &padding_l,
5737 const memory::dims &padding_r) {
5738 memory::validate_dims(strides, diff_src_desc.data.ndims - 2);
5739 memory::validate_dims(kernel, diff_src_desc.data.ndims - 2);
5740 memory::validate_dims(padding_l, diff_src_desc.data.ndims - 2);
5741 memory::validate_dims(padding_r, diff_src_desc.data.ndims - 2);
5744 convert_to_c(aalgorithm), &diff_src_desc.data,
5745 &diff_dst_desc.data, &strides[0], &kernel[0],
5746 &padding_l[0], &padding_r[0]),
5747 "could not create a descriptor for a pooling backward "
5748 "propagation primitive");
5749 }
5750 };
5751
5755 primitive_desc() = default;
5756
5769 primitive_desc(const desc &adesc, const engine &aengine,
5770 const pooling_forward::primitive_desc &hint_fwd_pd,
5771 bool allow_empty = false)
5772 : dnnl::primitive_desc(&adesc.data, nullptr, aengine,
5773 hint_fwd_pd.get(), allow_empty) {}
5774
5788 primitive_desc(const desc &adesc, const primitive_attr &attr,
5789 const engine &aengine,
5790 const pooling_forward::primitive_desc &hint_fwd_pd,
5791 bool allow_empty = false)
5792 : dnnl::primitive_desc(&adesc.data, &attr, aengine,
5793 hint_fwd_pd.get(), allow_empty) {}
5794
5802 : dnnl::primitive_desc(pd, dnnl::primitive::kind::pooling,
5804
5807
5810
5813 };
5814
5816 pooling_backward() = default;
5817
5822};
5823
5825
5846
5850 struct desc {
5852
5865 desc(prop_kind aprop_kind, algorithm aalgorithm,
5866 const memory::desc &data_desc, float alpha = 0,
5867 float beta = 0) {
5869 dnnl::convert_to_c(aprop_kind),
5870 dnnl::convert_to_c(aalgorithm),
5871 &data_desc.data, alpha, beta),
5872 "could not create a descriptor for an eltwise forward "
5873 "propagation primitive");
5874 }
5875 };
5876
5880 primitive_desc() = default;
5881
5892 primitive_desc(const desc &adesc, const engine &aengine,
5893 bool allow_empty = false)
5895 &adesc.data, nullptr, aengine, nullptr, allow_empty) {}
5896
5908 primitive_desc(const desc &adesc, const primitive_attr &attr,
5909 const engine &aengine, bool allow_empty = false)
5911 &adesc.data, &attr, aengine, nullptr, allow_empty) {}
5912
5920 : dnnl::primitive_desc(pd, dnnl::primitive::kind::eltwise,
5923
5925 memory::desc src_desc() const { return base::src_desc(0); }
5926
5928 memory::desc dst_desc() const { return base::dst_desc(0); }
5929 };
5930
5932 eltwise_forward() = default;
5933
5938};
5939
5943 struct desc {
5945
5957 desc(algorithm aalgorithm, const memory::desc &diff_data_desc,
5958 const memory::desc &data_desc, float alpha = 0,
5959 float beta = 0) {
5962 dnnl::convert_to_c(aalgorithm),
5963 &diff_data_desc.data, &data_desc.data, alpha, beta),
5964 "could not create a descriptor for an eltwise backward "
5965 "propagation primitive");
5966 }
5967 };
5968
5972 primitive_desc() = default;
5973
5987 primitive_desc(const desc &adesc, const engine &aengine,
5988 const eltwise_forward::primitive_desc &hint_fwd_pd,
5989 bool allow_empty = false)
5990 : dnnl::primitive_desc(&adesc.data, nullptr, aengine,
5991 hint_fwd_pd.get(), allow_empty) {}
5992
6007 primitive_desc(const desc &adesc, const primitive_attr &attr,
6008 const engine &aengine,
6009 const eltwise_forward::primitive_desc &hint_fwd_pd,
6010 bool allow_empty = false)
6011 : dnnl::primitive_desc(&adesc.data, &attr, aengine,
6012 hint_fwd_pd.get(), allow_empty) {}
6013
6021 : dnnl::primitive_desc(pd, dnnl::primitive::kind::eltwise,
6023
6025 memory::desc src_desc() const { return base::src_desc(0); }
6026
6029
6032 };
6033
6035 eltwise_backward() = default;
6036
6041};
6042
6044
6052
6056 struct desc {
6058
6060 desc() = default;
6061
6070 desc(prop_kind aprop_kind, const memory::desc &data_desc,
6071 int softmax_axis) {
6073 dnnl::convert_to_c(aprop_kind),
6074 &data_desc.data, softmax_axis),
6075 "could not create a descriptor for a softmax forward "
6076 "propagation primitive");
6077 }
6078 };
6079
6083 primitive_desc() = default;
6084
6095 primitive_desc(const desc &adesc, const engine &aengine,
6096 bool allow_empty = false)
6098 &adesc.data, nullptr, aengine, nullptr, allow_empty) {}
6099
6111 primitive_desc(const desc &adesc, const primitive_attr &attr,
6112 const engine &aengine, bool allow_empty = false)
6114 &adesc.data, &attr, aengine, nullptr, allow_empty) {}
6115
6123 : dnnl::primitive_desc(pd, dnnl::primitive::kind::softmax,
6126
6128 memory::desc src_desc() const { return base::src_desc(0); }
6129
6131 memory::desc dst_desc() const { return base::dst_desc(0); }
6132 };
6133
6135 softmax_forward() = default;
6136
6141};
6142
6146 struct desc {
6148
6150 desc() = default;
6151
6159 desc(const memory::desc &diff_data_desc, const memory::desc &data_desc,
6160 int softmax_axis) {
6162 dnnl_softmax_backward_desc_init(&data, &diff_data_desc.data,
6163 &data_desc.data, softmax_axis),
6164 "could not create a descriptor for a softmax backward "
6165 "propagation primitive");
6166 }
6167 };
6168
6172 primitive_desc() = default;
6173
6187 primitive_desc(const desc &adesc, const engine &aengine,
6188 const softmax_forward::primitive_desc &hint_fwd_pd,
6189 bool allow_empty = false)
6190 : dnnl::primitive_desc(&adesc.data, nullptr, aengine,
6191 hint_fwd_pd.get(), allow_empty) {}
6192
6207 primitive_desc(const desc &adesc, const primitive_attr &attr,
6208 const engine &aengine,
6209 const softmax_forward::primitive_desc &hint_fwd_pd,
6210 bool allow_empty = false)
6211 : dnnl::primitive_desc(&adesc.data, &attr, aengine,
6212 hint_fwd_pd.get(), allow_empty) {}
6213
6221 : dnnl::primitive_desc(pd, dnnl::primitive::kind::softmax,
6223
6225 memory::desc dst_desc() const { return base::dst_desc(0); }
6226
6229
6232 };
6233
6235 softmax_backward() = default;
6236
6241};
6242
6244
6252
6256 struct desc {
6258
6260 desc() = default;
6261
6270 desc(prop_kind aprop_kind, const memory::desc &data_desc,
6271 int logsoftmax_axis) {
6273 dnnl::convert_to_c(aprop_kind),
6274 &data_desc.data, logsoftmax_axis),
6275 "could not create a descriptor for a logsoftmax forward "
6276 "propagation primitive");
6277 }
6278 };
6279
6283 primitive_desc() = default;
6284
6295 primitive_desc(const desc &adesc, const engine &aengine,
6296 bool allow_empty = false)
6298 &adesc.data, nullptr, aengine, nullptr, allow_empty) {}
6299
6311 primitive_desc(const desc &adesc, const primitive_attr &attr,
6312 const engine &aengine, bool allow_empty = false)
6314 &adesc.data, &attr, aengine, nullptr, allow_empty) {}
6315
6323 : dnnl::primitive_desc(pd,
6324 // Logsoftmax and softmax share the implementation and
6325 // currently report the same primitive kind. Hence this
6326 // must be softmax and not logsoftmax.
6327 dnnl::primitive::kind::softmax,
6330
6332 memory::desc src_desc() const { return base::src_desc(0); }
6333
6335 memory::desc dst_desc() const { return base::dst_desc(0); }
6336 };
6337
6340
6345};
6346
6350 struct desc {
6352
6354 desc() = default;
6355
6363 desc(const memory::desc &diff_data_desc, const memory::desc &data_desc,
6364 int logsoftmax_axis) {
6366 &diff_data_desc.data, &data_desc.data,
6367 logsoftmax_axis),
6368 "could not create a descriptor for a logsoftmax backward "
6369 "propagation primitive");
6370 }
6371 };
6372
6376 primitive_desc() = default;
6377
6391 primitive_desc(const desc &adesc, const engine &aengine,
6392 const logsoftmax_forward::primitive_desc &hint_fwd_pd,
6393 bool allow_empty = false)
6394 : dnnl::primitive_desc(&adesc.data, nullptr, aengine,
6395 hint_fwd_pd.get(), allow_empty) {}
6396
6411 primitive_desc(const desc &adesc, const primitive_attr &attr,
6412 const engine &aengine,
6413 const logsoftmax_forward::primitive_desc &hint_fwd_pd,
6414 bool allow_empty = false)
6415 : dnnl::primitive_desc(&adesc.data, &attr, aengine,
6416 hint_fwd_pd.get(), allow_empty) {}
6417
6425 : dnnl::primitive_desc(pd,
6426 // Logsoftmax and softmax share the implementation and
6427 // currently report the same primitive kind. Hence this
6428 // must be softmax and not logsoftmax.
6429 dnnl::primitive::kind::softmax,
6431
6433 memory::desc dst_desc() const { return base::dst_desc(0); }
6434
6437
6440 };
6441
6444
6449};
6450
6452
6472
6476 struct desc {
6478
6493 desc(prop_kind aprop_kind, const memory::desc &data_desc, float epsilon,
6494 normalization_flags flags) {
6497 dnnl::convert_to_c(aprop_kind), &data_desc.data,
6498 epsilon, convert_to_c(flags)),
6499 "could not create a descriptor for a batch normalization "
6500 "forward propagation primitive");
6501 }
6502 };
6503
6508 primitive_desc() = default;
6509
6520 primitive_desc(const desc &adesc, const engine &aengine,
6521 bool allow_empty = false)
6523 &adesc.data, nullptr, aengine, nullptr, allow_empty) {}
6524
6536 primitive_desc(const desc &adesc, const primitive_attr &attr,
6537 const engine &aengine, bool allow_empty = false)
6539 &adesc.data, &attr, aengine, nullptr, allow_empty) {}
6540
6548 : dnnl::primitive_desc(pd,
6549 dnnl::primitive::kind::batch_normalization,
6552
6554 memory::desc src_desc() const { return base::src_desc(0); }
6555
6557 memory::desc dst_desc() const { return base::dst_desc(0); }
6558
6561
6564
6567 memory::desc mean_desc() const { return stat_desc(mean); }
6568
6571 memory::desc variance_desc() const { return stat_desc(var); }
6572
6573 private:
6574 enum {
6575 mean = 1,
6576 var = 2,
6577 };
6578 memory::desc stat_desc(int kind) const {
6583 &p),
6584 "could not retrieve a descriptor from a primitive "
6585 "descriptor for batch normalization forward propagation "
6586 "primitive");
6587 return query_md(p->flags & dnnl_use_global_stats ? query::src_md
6588 : query::dst_md,
6589 kind);
6590 }
6591 };
6592
6595
6600};
6601
6605 struct desc {
6607
6620 desc(prop_kind aprop_kind, const memory::desc &diff_data_desc,
6621 const memory::desc &data_desc, float epsilon,
6622 normalization_flags flags) {
6624 dnnl::convert_to_c(aprop_kind),
6625 &diff_data_desc.data, &data_desc.data,
6626 epsilon, convert_to_c(flags)),
6627 "could not create a descriptor for a batch normalization "
6628 "backward propagation primitive");
6629 }
6630 };
6631
6636 primitive_desc() = default;
6637
6651 primitive_desc(const desc &adesc, const engine &aengine,
6653 bool allow_empty = false)
6654 : dnnl::primitive_desc(&adesc.data, nullptr, aengine,
6655 hint_fwd_pd.get(), allow_empty) {}
6656
6671 primitive_desc(const desc &adesc, const primitive_attr &attr,
6672 const engine &aengine,
6674 bool allow_empty = false)
6675 : dnnl::primitive_desc(&adesc.data, &attr, aengine,
6676 hint_fwd_pd.get(), allow_empty) {}
6677
6685 : dnnl::primitive_desc(pd,
6686 dnnl::primitive::kind::batch_normalization,
6688 }
6689
6691 memory::desc src_desc() const { return base::src_desc(0); }
6692
6695
6697 memory::desc dst_desc() const { return base::dst_desc(0); }
6698
6701
6704
6707 return base::diff_weights_desc(0);
6708 }
6709
6712
6715 return query_md(query::src_md, 2);
6716 }
6717
6720 };
6721
6724
6729};
6730
6732
6754
6758 struct desc {
6760
6772 desc(prop_kind aprop_kind, const memory::desc &data_desc,
6773 const memory::desc &stat_desc, float epsilon,
6774 normalization_flags flags) {
6777 dnnl::convert_to_c(aprop_kind), &data_desc.data,
6778 &stat_desc.data, epsilon, convert_to_c(flags)),
6779 "could not create a descriptor for a layer normalization "
6780 "forward propagation primitive");
6781 }
6782
6793 desc(prop_kind aprop_kind, const memory::desc &data_desc, float epsilon,
6794 normalization_flags flags) {
6797 dnnl::convert_to_c(aprop_kind), &data_desc.data,
6798 nullptr, epsilon, convert_to_c(flags)),
6799 "could not create a descriptor for a layer normalization "
6800 "forward propagation primitive");
6801 }
6802 };
6803
6808 primitive_desc() = default;
6809
6820 primitive_desc(const desc &adesc, const engine &aengine,
6821 bool allow_empty = false)
6823 &adesc.data, nullptr, aengine, nullptr, allow_empty) {}
6824
6836 primitive_desc(const desc &adesc, const primitive_attr &attr,
6837 const engine &aengine, bool allow_empty = false)
6839 &adesc.data, &attr, aengine, nullptr, allow_empty) {}
6840
6848 : dnnl::primitive_desc(pd,
6849 dnnl::primitive::kind::layer_normalization,
6852
6854 memory::desc src_desc() const { return base::src_desc(0); }
6855
6857 memory::desc dst_desc() const { return base::dst_desc(0); }
6858
6861
6864
6866 memory::desc mean_desc() const { return stat_desc(mean); }
6867
6869 memory::desc variance_desc() const { return stat_desc(var); }
6870
6871 private:
6872 enum {
6873 mean = 1,
6874 var = 2,
6875 };
6876 memory::desc stat_desc(int kind) const {
6881 &p),
6882 "could not retrieve a descriptor from a primitive "
6883 "descriptor for layer normalization forward propagation "
6884 "primitive");
6885 return query_md(p->flags & dnnl_use_global_stats ? query::src_md
6886 : query::dst_md,
6887 kind);
6888 }
6889 };
6890
6893
6898};
6899
6903 struct desc {
6905
6919 desc(prop_kind aprop_kind, const memory::desc &diff_data_desc,
6920 const memory::desc &data_desc, const memory::desc &stat_desc,
6921 float epsilon, normalization_flags flags) {
6924 dnnl::convert_to_c(aprop_kind),
6925 &diff_data_desc.data, &data_desc.data,
6926 &stat_desc.data, epsilon, convert_to_c(flags)),
6927 "could not create a descriptor for a batch normalization "
6928 "backward propagation primitive");
6929 }
6930
6943 desc(prop_kind aprop_kind, const memory::desc &diff_data_desc,
6944 const memory::desc &data_desc, float epsilon,
6945 normalization_flags flags) {
6947 dnnl::convert_to_c(aprop_kind),
6948 &diff_data_desc.data, &data_desc.data,
6949 nullptr, epsilon, convert_to_c(flags)),
6950 "could not create a descriptor for a batch normalization "
6951 "backward propagation primitive");
6952 }
6953 };
6954
6959 primitive_desc() = default;
6960
6974 primitive_desc(const desc &adesc, const engine &aengine,
6976 bool allow_empty = false)
6977 : dnnl::primitive_desc(&adesc.data, nullptr, aengine,
6978 hint_fwd_pd.get(), allow_empty) {}
6979
6994 primitive_desc(const desc &adesc, const primitive_attr &attr,
6995 const engine &aengine,
6997 bool allow_empty = false)
6998 : dnnl::primitive_desc(&adesc.data, &attr, aengine,
6999 hint_fwd_pd.get(), allow_empty) {}
7000
7008 : dnnl::primitive_desc(pd,
7009 dnnl::primitive::kind::layer_normalization,
7011 }
7012
7014 memory::desc src_desc() const { return base::src_desc(0); }
7015
7018
7020 memory::desc dst_desc() const { return base::dst_desc(0); }
7021
7024
7027
7030 return base::diff_weights_desc(0);
7031 }
7032
7035
7038 return query_md(query::src_md, 2);
7039 }
7040
7043 };
7044
7047
7052};
7053
7055
7063
7067 struct desc {
7069
7084 desc(prop_kind aprop_kind, const memory::desc &src_desc,
7085 const memory::desc &weights_desc, const memory::desc &bias_desc,
7086 const memory::desc &dst_desc) {
7088 dnnl::convert_to_c(aprop_kind),
7089 &src_desc.data, &weights_desc.data,
7090 &bias_desc.data, &dst_desc.data),
7091 "could not create a descriptor for an inner product "
7092 "forward propagation primitive");
7093 }
7094
7108 desc(prop_kind aprop_kind, const memory::desc &src_desc,
7109 const memory::desc &weights_desc,
7110 const memory::desc &dst_desc) {
7113 dnnl::convert_to_c(aprop_kind), &src_desc.data,
7114 &weights_desc.data, nullptr, &dst_desc.data),
7115 "could not create a descriptor for an inner product "
7116 "forward propagation primitive");
7117 }
7118 };
7119
7123 primitive_desc() = default;
7124
7135 primitive_desc(const desc &adesc, const engine &aengine,
7136 bool allow_empty = false)
7138 &adesc.data, nullptr, aengine, nullptr, allow_empty) {}
7139
7151 primitive_desc(const desc &adesc, const primitive_attr &attr,
7152 const engine &aengine, bool allow_empty = false)
7154 &adesc.data, &attr, aengine, nullptr, allow_empty) {}
7155
7163 : dnnl::primitive_desc(pd, dnnl::primitive::kind::inner_product,
7166
7168 memory::desc src_desc() const { return base::src_desc(0); }
7169
7172
7174 memory::desc dst_desc() const { return base::dst_desc(0); }
7175
7178 };
7179
7182
7187};
7188
7192 struct desc {
7194
7205 desc(const memory::desc &diff_src_desc,
7206 const memory::desc &weights_desc,
7207 const memory::desc &diff_dst_desc) {
7209 &diff_src_desc.data, &weights_desc.data,
7210 &diff_dst_desc.data),
7211 "could not create a descriptor for an inner product "
7212 "backward propagation primitive");
7213 }
7214 };
7215
7220 primitive_desc() = default;
7221
7235 primitive_desc(const desc &adesc, const engine &aengine,
7236 const inner_product_forward::primitive_desc &hint_fwd_pd,
7237 bool allow_empty = false)
7238 : dnnl::primitive_desc(&adesc.data, nullptr, aengine,
7239 hint_fwd_pd.get(), allow_empty) {}
7240
7255 primitive_desc(const desc &adesc, const primitive_attr &attr,
7256 const engine &aengine,
7257 const inner_product_forward::primitive_desc &hint_fwd_pd,
7258 bool allow_empty = false)
7259 : dnnl::primitive_desc(&adesc.data, &attr, aengine,
7260 hint_fwd_pd.get(), allow_empty) {}
7261
7269 : dnnl::primitive_desc(pd, dnnl::primitive::kind::inner_product,
7271
7274
7277
7280 };
7281
7284
7289};
7290
7294 struct desc {
7296
7308 desc(const memory::desc &src_desc,
7309 const memory::desc &diff_weights_desc,
7310 const memory::desc &diff_bias_desc,
7311 const memory::desc &diff_dst_desc) {
7314 &src_desc.data, &diff_weights_desc.data,
7315 &diff_bias_desc.data, &diff_dst_desc.data),
7316 "could not create a descriptor for an inner product "
7317 "weights gradient primitive");
7318 }
7319
7330 desc(const memory::desc &src_desc,
7331 const memory::desc &diff_weights_desc,
7332 const memory::desc &diff_dst_desc) {
7335 &src_desc.data, &diff_weights_desc.data, nullptr,
7336 &diff_dst_desc.data),
7337 "could not create a descriptor for an inner product "
7338 "weights gradient primitive");
7339 }
7340 };
7341
7345 primitive_desc() = default;
7346
7360 primitive_desc(const desc &adesc, const engine &aengine,
7361 const inner_product_forward::primitive_desc &hint_fwd_pd,
7362 bool allow_empty = false)
7363 : dnnl::primitive_desc(&adesc.data, nullptr, aengine,
7364 hint_fwd_pd.get(), allow_empty) {}
7365
7380 primitive_desc(const desc &adesc, const primitive_attr &attr,
7381 const engine &aengine,
7382 const inner_product_forward::primitive_desc &hint_fwd_pd,
7383 bool allow_empty = false)
7384 : dnnl::primitive_desc(&adesc.data, &attr, aengine,
7385 hint_fwd_pd.get(), allow_empty) {}
7386
7394 : dnnl::primitive_desc(pd, dnnl::primitive::kind::inner_product,
7396
7398 memory::desc src_desc() const { return base::src_desc(0); }
7399
7402 return base::diff_weights_desc(0);
7403 }
7404
7407
7410 return base::diff_weights_desc(1);
7411 }
7412 };
7413
7416
7421};
7422
7424
7432
7435 using primitive_desc::primitive_desc;
7436
7439
7448 dnnl::prop_kind aprop_kind, dnnl::algorithm cell_kind)
7449 : rnn_primitive_desc_base(pd, aprop_kind, aprop_kind, cell_kind) {}
7450
7455 }
7456
7463 }
7464
7469 }
7470
7475 }
7476
7481 }
7482
7487 }
7488
7493 }
7494
7501 }
7502
7507 }
7508
7515 }
7516
7521 }
7522
7527 }
7528
7535 }
7536
7541 }
7542
7547 }
7548
7553 }
7554
7558 return base::query_md(
7560 }
7561
7565 return base::query_md(
7567 }
7568
7575 }
7576
7581 }
7582
7589 }
7590
7595 }
7596
7597protected:
7598 using rnn_base = rnn_primitive_desc_base;
7599
7600 // (Deliberately not using doxygen comments)
7601 //
7602 // Constructs an RNN primitive descriptor base from a C API primitive
7603 // descriptor while checking that it actually describes the expected
7604 // primitive by comparing propagation and primitive kinds. Caller can
7605 // pass two options propagation kinds. This is typically used to check
7606 // that propagation kind is inference or training forward propagation.
7607 //
7608 // @param pd C API primitive descriptor.
7609 // @param prop_kind1 Expected propagation kind.
7610 // @param prop_kind2 Expected propagation kind.
7611 // @param cell_kind Expected cell kind.
7613 dnnl::prop_kind prop_kind1, dnnl::prop_kind prop_kind2,
7614 dnnl::algorithm cell_kind) {
7616 dnnl_status_t rc;
7617 rc = dnnl_primitive_desc_query(pd, dnnl_query_rnn_d, 0, &rnn_d);
7619 "could not retrieve a descriptor from a primitive descriptor "
7620 "for an RNN primitive");
7621
7622 dnnl_prop_kind_t c_prop_kind1 = convert_to_c(prop_kind1);
7623 dnnl_prop_kind_t c_prop_kind2 = convert_to_c(prop_kind2);
7624 dnnl_alg_kind_t c_cell_kind = convert_to_c(cell_kind);
7625
7626 bool ok = rnn_d->primitive_kind == dnnl_rnn
7627 && (rnn_d->prop_kind == c_prop_kind1
7628 || rnn_d->prop_kind == c_prop_kind2)
7629 && rnn_d->cell_kind == c_cell_kind;
7630
7631 if (!ok)
7632 DNNL_THROW_ERROR(dnnl_invalid_arguments,
7633 "mismatch between expected and provided descriptors for an "
7634 "RNN primitive");
7635
7636 reset_with_clone(pd);
7637 }
7638};
7639
7643 struct desc {
7644 dnnl_rnn_desc_t data;
7645
7686 desc(prop_kind aprop_kind, algorithm activation,
7687 rnn_direction direction, const memory::desc &src_layer_desc,
7688 const memory::desc &src_iter_desc,
7689 const memory::desc &weights_layer_desc,
7690 const memory::desc &weights_iter_desc,
7691 const memory::desc &bias_desc,
7692 const memory::desc &dst_layer_desc,
7693 const memory::desc &dst_iter_desc,
7694 rnn_flags flags = rnn_flags::undef, float alpha = 0.0f,
7695 float beta = 0.0f) {
7698 dnnl::convert_to_c(aprop_kind),
7699 dnnl::convert_to_c(activation),
7700 dnnl::convert_to_c(direction), &src_layer_desc.data,
7701 &src_iter_desc.data, &weights_layer_desc.data,
7702 &weights_iter_desc.data, &bias_desc.data,
7703 &dst_layer_desc.data, &dst_iter_desc.data,
7704 dnnl::convert_to_c(flags), alpha, beta),
7705 "could not create a descriptor for a vanilla RNN forward "
7706 "propagation primitive");
7707 }
7708 };
7709
7713 primitive_desc() = default;
7714
7725 primitive_desc(const desc &adesc, const engine &aengine,
7726 bool allow_empty = false)
7728 &adesc.data, nullptr, aengine, nullptr, allow_empty) {}
7729
7741 primitive_desc(const desc &adesc, const primitive_attr &attr,
7742 const engine &aengine, bool allow_empty = false)
7744 &adesc.data, &attr, aengine, nullptr, allow_empty) {}
7745
7756
7759 return rnn_base::src_layer_desc();
7760 }
7761
7764
7768 }
7769
7773 }
7774
7777
7780 return rnn_base::dst_layer_desc();
7781 }
7782
7785
7788 return rnn_base::workspace_desc();
7789 }
7790 };
7791
7794
7799};
7800
7804 struct desc {
7805 dnnl_rnn_desc_t data;
7806
7859 desc(prop_kind aprop_kind, algorithm activation,
7860 rnn_direction direction, const memory::desc &src_layer_desc,
7861 const memory::desc &src_iter_desc,
7862 const memory::desc &weights_layer_desc,
7863 const memory::desc &weights_iter_desc,
7864 const memory::desc &bias_desc,
7865 const memory::desc &dst_layer_desc,
7866 const memory::desc &dst_iter_desc,
7867 const memory::desc &diff_src_layer_desc,
7868 const memory::desc &diff_src_iter_desc,
7869 const memory::desc &diff_weights_layer_desc,
7870 const memory::desc &diff_weights_iter_desc,
7871 const memory::desc &diff_bias_desc,
7872 const memory::desc &diff_dst_layer_desc,
7873 const memory::desc &diff_dst_iter_desc,
7874 rnn_flags flags = rnn_flags::undef, float alpha = 0.0f,
7875 float beta = 0.0f) {
7878 dnnl::convert_to_c(aprop_kind),
7879 dnnl::convert_to_c(activation),
7880 dnnl::convert_to_c(direction), &src_layer_desc.data,
7881 &src_iter_desc.data, &weights_layer_desc.data,
7882 &weights_iter_desc.data, &bias_desc.data,
7883 &dst_layer_desc.data, &dst_iter_desc.data,
7884 &diff_src_layer_desc.data, &diff_src_iter_desc.data,
7885 &diff_weights_layer_desc.data,
7886 &diff_weights_iter_desc.data, &diff_bias_desc.data,
7887 &diff_dst_layer_desc.data, &diff_dst_iter_desc.data,
7888 dnnl::convert_to_c(flags), alpha, beta),
7889 "could not create a descriptor for a vanilla RNN backward "
7890 "propagation primitive");
7891 }
7892 };
7893
7897 primitive_desc() = default;
7898
7912 primitive_desc(const desc &adesc, const engine &aengine,
7913 const vanilla_rnn_forward::primitive_desc &hint_fwd_pd,
7914 bool allow_empty = false)
7915 : rnn_primitive_desc_base(&adesc.data, nullptr, aengine,
7916 hint_fwd_pd.get(), allow_empty) {}
7917
7932 primitive_desc(const desc &adesc, const primitive_attr &attr,
7933 const engine &aengine,
7934 const vanilla_rnn_forward::primitive_desc &hint_fwd_pd,
7935 bool allow_empty = false)
7936 : rnn_primitive_desc_base(&adesc.data, &attr, aengine,
7937 hint_fwd_pd.get(), allow_empty) {}
7938
7948
7951 return rnn_base::src_layer_desc();
7952 }
7953
7956
7960 }
7961
7965 }
7966
7969
7972 return rnn_base::dst_layer_desc();
7973 }
7974
7977
7980 return rnn_base::workspace_desc();
7981 }
7982
7986 }
7987
7991 }
7992
7996 }
7997
8001 }
8002
8005 return rnn_base::diff_bias_desc();
8006 }
8007
8011 }
8012
8016 }
8017 };
8018
8021
8026};
8027
8029struct lstm_forward : public primitive {
8031 struct desc {
8032 dnnl_rnn_desc_t data;
8033
8082 desc(prop_kind aprop_kind, rnn_direction direction,
8083 const memory::desc &src_layer_desc,
8084 const memory::desc &src_iter_desc,
8085 const memory::desc &src_iter_c_desc,
8086 const memory::desc &weights_layer_desc,
8087 const memory::desc &weights_iter_desc,
8088 const memory::desc &weights_peephole_desc,
8089 const memory::desc &weights_projection_desc,
8090 const memory::desc &bias_desc,
8091 const memory::desc &dst_layer_desc,
8092 const memory::desc &dst_iter_desc,
8093 const memory::desc &dst_iter_c_desc,
8094 rnn_flags flags = rnn_flags::undef) {
8097 dnnl::convert_to_c(aprop_kind),
8098 dnnl::convert_to_c(direction), &src_layer_desc.data,
8099 &src_iter_desc.data, &src_iter_c_desc.data,
8100 &weights_layer_desc.data, &weights_iter_desc.data,
8101 &weights_peephole_desc.data,
8102 &weights_projection_desc.data, &bias_desc.data,
8103 &dst_layer_desc.data, &dst_iter_desc.data,
8104 &dst_iter_c_desc.data, dnnl::convert_to_c(flags)),
8105 "could not create a descriptor for an LSTM forward "
8106 "propagation primitive");
8107 }
8108
8150 desc(prop_kind aprop_kind, rnn_direction direction,
8151 const memory::desc &src_layer_desc,
8152 const memory::desc &src_iter_desc,
8153 const memory::desc &src_iter_c_desc,
8154 const memory::desc &weights_layer_desc,
8155 const memory::desc &weights_iter_desc,
8156 const memory::desc &weights_peephole_desc,
8157 const memory::desc &bias_desc,
8158 const memory::desc &dst_layer_desc,
8159 const memory::desc &dst_iter_desc,
8160 const memory::desc &dst_iter_c_desc,
8161 rnn_flags flags = rnn_flags::undef) {
8164 dnnl::convert_to_c(aprop_kind),
8165 dnnl::convert_to_c(direction), &src_layer_desc.data,
8166 &src_iter_desc.data, &src_iter_c_desc.data,
8167 &weights_layer_desc.data, &weights_iter_desc.data,
8168 &weights_peephole_desc.data, &bias_desc.data,
8169 &dst_layer_desc.data, &dst_iter_desc.data,
8170 &dst_iter_c_desc.data, dnnl::convert_to_c(flags)),
8171 "could not create a descriptor for an LSTM forward "
8172 "propagation primitive");
8173 }
8174
8211 desc(prop_kind aprop_kind, rnn_direction direction,
8212 const memory::desc &src_layer_desc,
8213 const memory::desc &src_iter_desc,
8214 const memory::desc &src_iter_c_desc,
8215 const memory::desc &weights_layer_desc,
8216 const memory::desc &weights_iter_desc,
8217 const memory::desc &bias_desc,
8218 const memory::desc &dst_layer_desc,
8219 const memory::desc &dst_iter_desc,
8220 const memory::desc &dst_iter_c_desc,
8221 rnn_flags flags = rnn_flags::undef) {
8224 dnnl::convert_to_c(aprop_kind),
8225 dnnl::convert_to_c(direction), &src_layer_desc.data,
8226 &src_iter_desc.data, &src_iter_c_desc.data,
8227 &weights_layer_desc.data, &weights_iter_desc.data,
8228 &bias_desc.data, &dst_layer_desc.data,
8229 &dst_iter_desc.data, &dst_iter_c_desc.data,
8230 dnnl::convert_to_c(flags)),
8231 "could not create a descriptor for an LSTM forward "
8232 "propagation primitive");
8233 }
8234 };
8235
8239 primitive_desc() = default;
8240
8250 primitive_desc(const desc &adesc, const engine &aengine,
8251 bool allow_empty = false)
8253 &adesc.data, nullptr, aengine, nullptr, allow_empty) {}
8254
8265 primitive_desc(const desc &adesc, const primitive_attr &attr,
8266 const engine &aengine, bool allow_empty = false)
8268 &adesc.data, &attr, aengine, nullptr, allow_empty) {}
8269
8280
8283 return rnn_base::src_layer_desc();
8284 }
8285
8288
8292 }
8293
8297 }
8298
8302 }
8303
8307 }
8308
8312 }
8313
8316
8319 return rnn_base::dst_layer_desc();
8320 }
8321
8324
8328 }
8329
8332 return rnn_base::workspace_desc();
8333 }
8334 };
8335
8337 lstm_forward() = default;
8338
8343};
8344
8346struct lstm_backward : public primitive {
8348 struct desc {
8349 dnnl_rnn_desc_t data;
8350
8426 desc(prop_kind aprop_kind, rnn_direction direction,
8427 const memory::desc &src_layer_desc,
8428 const memory::desc &src_iter_desc,
8429 const memory::desc &src_iter_c_desc,
8430 const memory::desc &weights_layer_desc,
8431 const memory::desc &weights_iter_desc,
8432 const memory::desc &weights_peephole_desc,
8433 const memory::desc &weights_projection_desc,
8434 const memory::desc &bias_desc,
8435 const memory::desc &dst_layer_desc,
8436 const memory::desc &dst_iter_desc,
8437 const memory::desc &dst_iter_c_desc,
8438 const memory::desc &diff_src_layer_desc,
8439 const memory::desc &diff_src_iter_desc,
8440 const memory::desc &diff_src_iter_c_desc,
8441 const memory::desc &diff_weights_layer_desc,
8442 const memory::desc &diff_weights_iter_desc,
8443 const memory::desc &diff_weights_peephole_desc,
8444 const memory::desc &diff_weights_projection_desc,
8445 const memory::desc &diff_bias_desc,
8446 const memory::desc &diff_dst_layer_desc,
8447 const memory::desc &diff_dst_iter_desc,
8448 const memory::desc &diff_dst_iter_c_desc,
8449 rnn_flags flags = rnn_flags::undef) {
8452 dnnl::convert_to_c(aprop_kind),
8453 dnnl::convert_to_c(direction), &src_layer_desc.data,
8454 &src_iter_desc.data, &src_iter_c_desc.data,
8455 &weights_layer_desc.data, &weights_iter_desc.data,
8456 &weights_peephole_desc.data,
8457 &weights_projection_desc.data, &bias_desc.data,
8458 &dst_layer_desc.data, &dst_iter_desc.data,
8459 &dst_iter_c_desc.data, &diff_src_layer_desc.data,
8460 &diff_src_iter_desc.data,
8461 &diff_src_iter_c_desc.data,
8462 &diff_weights_layer_desc.data,
8463 &diff_weights_iter_desc.data,
8464 &diff_weights_peephole_desc.data,
8465 &diff_weights_projection_desc.data,
8466 &diff_bias_desc.data, &diff_dst_layer_desc.data,
8467 &diff_dst_iter_desc.data,
8468 &diff_dst_iter_c_desc.data,
8469 dnnl::convert_to_c(flags)),
8470 "could not create a descriptor for an LSTM backward "
8471 "propagation primitive");
8472 }
8473
8538 desc(prop_kind aprop_kind, rnn_direction direction,
8539 const memory::desc &src_layer_desc,
8540 const memory::desc &src_iter_desc,
8541 const memory::desc &src_iter_c_desc,
8542 const memory::desc &weights_layer_desc,
8543 const memory::desc &weights_iter_desc,
8544 const memory::desc &weights_peephole_desc,
8545 const memory::desc &bias_desc,
8546 const memory::desc &dst_layer_desc,
8547 const memory::desc &dst_iter_desc,
8548 const memory::desc &dst_iter_c_desc,
8549 const memory::desc &diff_src_layer_desc,
8550 const memory::desc &diff_src_iter_desc,
8551 const memory::desc &diff_src_iter_c_desc,
8552 const memory::desc &diff_weights_layer_desc,
8553 const memory::desc &diff_weights_iter_desc,
8554 const memory::desc &diff_weights_peephole_desc,
8555 const memory::desc &diff_bias_desc,
8556 const memory::desc &diff_dst_layer_desc,
8557 const memory::desc &diff_dst_iter_desc,
8558 const memory::desc &diff_dst_iter_c_desc,
8559 rnn_flags flags = rnn_flags::undef) {
8562 dnnl::convert_to_c(aprop_kind),
8563 dnnl::convert_to_c(direction), &src_layer_desc.data,
8564 &src_iter_desc.data, &src_iter_c_desc.data,
8565 &weights_layer_desc.data, &weights_iter_desc.data,
8566 &weights_peephole_desc.data, &bias_desc.data,
8567 &dst_layer_desc.data, &dst_iter_desc.data,
8568 &dst_iter_c_desc.data, &diff_src_layer_desc.data,
8569 &diff_src_iter_desc.data,
8570 &diff_src_iter_c_desc.data,
8571 &diff_weights_layer_desc.data,
8572 &diff_weights_iter_desc.data,
8573 &diff_weights_peephole_desc.data,
8574 &diff_bias_desc.data, &diff_dst_layer_desc.data,
8575 &diff_dst_iter_desc.data,
8576 &diff_dst_iter_c_desc.data,
8577 dnnl::convert_to_c(flags)),
8578 "could not create a descriptor for an LSTM backward "
8579 "propagation primitive");
8580 }
8581
8637 desc(prop_kind aprop_kind, rnn_direction direction,
8638 const memory::desc &src_layer_desc,
8639 const memory::desc &src_iter_desc,
8640 const memory::desc &src_iter_c_desc,
8641 const memory::desc &weights_layer_desc,
8642 const memory::desc &weights_iter_desc,
8643 const memory::desc &bias_desc,
8644 const memory::desc &dst_layer_desc,
8645 const memory::desc &dst_iter_desc,
8646 const memory::desc &dst_iter_c_desc,
8647 const memory::desc &diff_src_layer_desc,
8648 const memory::desc &diff_src_iter_desc,
8649 const memory::desc &diff_src_iter_c_desc,
8650 const memory::desc &diff_weights_layer_desc,
8651 const memory::desc &diff_weights_iter_desc,
8652 const memory::desc &diff_bias_desc,
8653 const memory::desc &diff_dst_layer_desc,
8654 const memory::desc &diff_dst_iter_desc,
8655 const memory::desc &diff_dst_iter_c_desc,
8656 rnn_flags flags = rnn_flags::undef) {
8659 dnnl::convert_to_c(aprop_kind),
8660 dnnl::convert_to_c(direction), &src_layer_desc.data,
8661 &src_iter_desc.data, &src_iter_c_desc.data,
8662 &weights_layer_desc.data, &weights_iter_desc.data,
8663 &bias_desc.data, &dst_layer_desc.data,
8664 &dst_iter_desc.data, &dst_iter_c_desc.data,
8665 &diff_src_layer_desc.data, &diff_src_iter_desc.data,
8666 &diff_src_iter_c_desc.data,
8667 &diff_weights_layer_desc.data,
8668 &diff_weights_iter_desc.data, &diff_bias_desc.data,
8669 &diff_dst_layer_desc.data, &diff_dst_iter_desc.data,
8670 &diff_dst_iter_c_desc.data,
8671 dnnl::convert_to_c(flags)),
8672 "could not create a descriptor for an LSTM backward "
8673 "propagation primitive");
8674 }
8675 };
8676
8680 primitive_desc() = default;
8681
8694 primitive_desc(const desc &adesc, const engine &aengine,
8695 const lstm_forward::primitive_desc &hint_fwd_pd,
8696 bool allow_empty = false)
8697 : rnn_primitive_desc_base(&adesc.data, nullptr, aengine,
8698 hint_fwd_pd.get(), allow_empty) {}
8699
8713 primitive_desc(const desc &adesc, const primitive_attr &attr,
8714 const engine &aengine,
8715 const lstm_forward::primitive_desc &hint_fwd_pd,
8716 bool allow_empty = false)
8717 : rnn_primitive_desc_base(&adesc.data, &attr, aengine,
8718 hint_fwd_pd.get(), allow_empty) {}
8719
8729
8732 return rnn_base::src_layer_desc();
8733 }
8734
8737
8741 }
8742
8746 }
8747
8751 }
8752
8756 }
8757
8761 }
8762
8765
8768 return rnn_base::dst_layer_desc();
8769 }
8770
8773
8777 }
8778
8781 return rnn_base::workspace_desc();
8782 }
8783
8787 }
8788
8792 }
8793
8797 }
8798
8802 }
8803
8807 }
8808
8812 }
8813
8817 }
8818
8821 return rnn_base::diff_bias_desc();
8822 }
8823
8827 }
8828
8832 }
8833
8837 }
8838 };
8839
8841 lstm_backward() = default;
8842
8847};
8848
8850struct gru_forward : public primitive {
8852 struct desc {
8853 dnnl_rnn_desc_t data;
8854
8887 desc(prop_kind aprop_kind, rnn_direction direction,
8888 const memory::desc &src_layer_desc,
8889 const memory::desc &src_iter_desc,
8890 const memory::desc &weights_layer_desc,
8891 const memory::desc &weights_iter_desc,
8892 const memory::desc &bias_desc,
8893 const memory::desc &dst_layer_desc,
8894 const memory::desc &dst_iter_desc,
8895 rnn_flags flags = rnn_flags::undef) {
8898 dnnl::convert_to_c(aprop_kind),
8899 dnnl::convert_to_c(direction), &src_layer_desc.data,
8900 &src_iter_desc.data, &weights_layer_desc.data,
8901 &weights_iter_desc.data, &bias_desc.data,
8902 &dst_layer_desc.data, &dst_iter_desc.data,
8903 dnnl::convert_to_c(flags)),
8904 "could not create a descriptor for a GRU forward "
8905 "propagation primitive");
8906 }
8907 };
8908
8912 primitive_desc() = default;
8913
8923 primitive_desc(const desc &adesc, const engine &aengine,
8924 bool allow_empty = false)
8926 &adesc.data, nullptr, aengine, nullptr, allow_empty) {}
8927
8938 primitive_desc(const desc &adesc, const primitive_attr &attr,
8939 const engine &aengine, bool allow_empty = false)
8941 &adesc.data, &attr, aengine, nullptr, allow_empty) {}
8942
8953
8956 return rnn_base::src_layer_desc();
8957 }
8958
8961
8965 }
8966
8970 }
8971
8974
8977 return rnn_base::dst_layer_desc();
8978 }
8979
8982
8985 return rnn_base::workspace_desc();
8986 }
8987 };
8988
8990 gru_forward() = default;
8991
8996};
8997
8999struct gru_backward : public primitive {
9001 struct desc {
9002 dnnl_rnn_desc_t data;
9003
9048 desc(prop_kind aprop_kind, rnn_direction direction,
9049 const memory::desc &src_layer_desc,
9050 const memory::desc &src_iter_desc,
9051 const memory::desc &weights_layer_desc,
9052 const memory::desc &weights_iter_desc,
9053 const memory::desc &bias_desc,
9054 const memory::desc &dst_layer_desc,
9055 const memory::desc &dst_iter_desc,
9056 const memory::desc &diff_src_layer_desc,
9057 const memory::desc &diff_src_iter_desc,
9058 const memory::desc &diff_weights_layer_desc,
9059 const memory::desc &diff_weights_iter_desc,
9060 const memory::desc &diff_bias_desc,
9061 const memory::desc &diff_dst_layer_desc,
9062 const memory::desc &diff_dst_iter_desc,
9063 rnn_flags flags = rnn_flags::undef) {
9066 dnnl::convert_to_c(aprop_kind),
9067 dnnl::convert_to_c(direction), &src_layer_desc.data,
9068 &src_iter_desc.data, &weights_layer_desc.data,
9069 &weights_iter_desc.data, &bias_desc.data,
9070 &dst_layer_desc.data, &dst_iter_desc.data,
9071 &diff_src_layer_desc.data, &diff_src_iter_desc.data,
9072 &diff_weights_layer_desc.data,
9073 &diff_weights_iter_desc.data, &diff_bias_desc.data,
9074 &diff_dst_layer_desc.data, &diff_dst_iter_desc.data,
9075 dnnl::convert_to_c(flags)),
9076 "could not create a descriptor for a GRU backward "
9077 "propagation primitive");
9078 }
9079 };
9080
9084 primitive_desc() = default;
9085
9098 primitive_desc(const desc &adesc, const engine &aengine,
9099 const gru_forward::primitive_desc &hint_fwd_pd,
9100 bool allow_empty = false)
9101 : rnn_primitive_desc_base(&adesc.data, nullptr, aengine,
9102 hint_fwd_pd.get(), allow_empty) {}
9103
9117 primitive_desc(const desc &adesc, const primitive_attr &attr,
9118 const engine &aengine,
9119 const gru_forward::primitive_desc &hint_fwd_pd,
9120 bool allow_empty = false)
9121 : rnn_primitive_desc_base(&adesc.data, &attr, aengine,
9122 hint_fwd_pd.get(), allow_empty) {}
9123
9133
9136 return rnn_base::src_layer_desc();
9137 }
9138
9141
9145 }
9146
9150 }
9151
9154
9157 return rnn_base::dst_layer_desc();
9158 }
9159
9162
9165 return rnn_base::workspace_desc();
9166 }
9167
9171 }
9172
9176 }
9177
9181 }
9182
9186 }
9187
9190 return rnn_base::diff_bias_desc();
9191 }
9192
9196 }
9197
9201 }
9202 };
9203
9205 gru_backward() = default;
9206
9211};
9212
9216 struct desc {
9217 dnnl_rnn_desc_t data;
9218
9252 desc(prop_kind aprop_kind, rnn_direction direction,
9253 const memory::desc &src_layer_desc,
9254 const memory::desc &src_iter_desc,
9255 const memory::desc &weights_layer_desc,
9256 const memory::desc &weights_iter_desc,
9257 const memory::desc &bias_desc,
9258 const memory::desc &dst_layer_desc,
9259 const memory::desc &dst_iter_desc,
9260 rnn_flags flags = rnn_flags::undef) {
9263 dnnl::convert_to_c(aprop_kind),
9264 dnnl::convert_to_c(direction), &src_layer_desc.data,
9265 &src_iter_desc.data, &weights_layer_desc.data,
9266 &weights_iter_desc.data, &bias_desc.data,
9267 &dst_layer_desc.data, &dst_iter_desc.data,
9268 dnnl::convert_to_c(flags)),
9269 "could not create a descriptor for an LBR GRU forward "
9270 "propagation primitive");
9271 }
9272 };
9273
9277 primitive_desc() = default;
9278
9289 primitive_desc(const desc &adesc, const engine &aengine,
9290 bool allow_empty = false)
9292 &adesc.data, nullptr, aengine, nullptr, allow_empty) {}
9293
9305 primitive_desc(const desc &adesc, const primitive_attr &attr,
9306 const engine &aengine, bool allow_empty = false)
9308 &adesc.data, &attr, aengine, nullptr, allow_empty) {}
9309
9319 dnnl::algorithm::lbr_gru) {}
9320
9323 return rnn_base::src_layer_desc();
9324 }
9325
9328
9332 }
9333
9337 }
9338
9341
9344 return rnn_base::dst_layer_desc();
9345 }
9346
9349
9352 return rnn_base::workspace_desc();
9353 }
9354 };
9355
9357 lbr_gru_forward() = default;
9358
9363};
9364
9368 struct desc {
9369 dnnl_rnn_desc_t data;
9370
9416 desc(prop_kind aprop_kind, rnn_direction direction,
9417 const memory::desc &src_layer_desc,
9418 const memory::desc &src_iter_desc,
9419 const memory::desc &weights_layer_desc,
9420 const memory::desc &weights_iter_desc,
9421 const memory::desc &bias_desc,
9422 const memory::desc &dst_layer_desc,
9423 const memory::desc &dst_iter_desc,
9424 const memory::desc &diff_src_layer_desc,
9425 const memory::desc &diff_src_iter_desc,
9426 const memory::desc &diff_weights_layer_desc,
9427 const memory::desc &diff_weights_iter_desc,
9428 const memory::desc &diff_bias_desc,
9429 const memory::desc &diff_dst_layer_desc,
9430 const memory::desc &diff_dst_iter_desc,
9431 rnn_flags flags = rnn_flags::undef) {
9434 dnnl::convert_to_c(aprop_kind),
9435 dnnl::convert_to_c(direction), &src_layer_desc.data,
9436 &src_iter_desc.data, &weights_layer_desc.data,
9437 &weights_iter_desc.data, &bias_desc.data,
9438 &dst_layer_desc.data, &dst_iter_desc.data,
9439 &diff_src_layer_desc.data, &diff_src_iter_desc.data,
9440 &diff_weights_layer_desc.data,
9441 &diff_weights_iter_desc.data, &diff_bias_desc.data,
9442 &diff_dst_layer_desc.data, &diff_dst_iter_desc.data,
9443 dnnl::convert_to_c(flags)),
9444 "could not create a descriptor for an LBR GRU backward "
9445 "propagation primitive");
9446 }
9447 };
9448
9452 primitive_desc() = default;
9453
9467 primitive_desc(const desc &adesc, const engine &aengine,
9468 const lbr_gru_forward::primitive_desc &hint_fwd_pd,
9469 bool allow_empty = false)
9470 : rnn_primitive_desc_base(&adesc.data, nullptr, aengine,
9471 hint_fwd_pd.get(), allow_empty) {}
9472
9487 primitive_desc(const desc &adesc, const primitive_attr &attr,
9488 const engine &aengine,
9489 const lbr_gru_forward::primitive_desc &hint_fwd_pd,
9490 bool allow_empty = false)
9491 : rnn_primitive_desc_base(&adesc.data, &attr, aengine,
9492 hint_fwd_pd.get(), allow_empty) {}
9493
9503
9506 return rnn_base::src_layer_desc();
9507 }
9508
9511
9515 }
9516
9520 }
9521
9524
9527 return rnn_base::dst_layer_desc();
9528 }
9529
9532
9535 return rnn_base::workspace_desc();
9536 }
9537
9541 }
9542
9546 }
9547
9551 }
9552
9556 }
9557
9560 return rnn_base::diff_bias_desc();
9561 }
9562
9566 }
9567
9571 }
9572 };
9573
9575 lbr_gru_backward() = default;
9576
9581};
9582
9584
9592
9596 struct desc {
9598
9608 desc(prop_kind aprop_kind, const memory::desc &data_desc, int axis,
9609 int group_size) {
9611 dnnl::convert_to_c(aprop_kind),
9612 &data_desc.data, axis, group_size),
9613 "could not create a descriptor for a shuffle forward "
9614 "propagation primitive");
9615 }
9616 };
9617
9621 primitive_desc() = default;
9622
9634 primitive_desc(const desc &adesc, const engine &aengine,
9635 const primitive_attr &attr = primitive_attr(),
9636 bool allow_empty = false)
9638 &adesc.data, &attr, aengine, nullptr, allow_empty) {}
9639
9647 : dnnl::primitive_desc(pd, dnnl::primitive::kind::shuffle,
9650
9652 memory::desc src_desc() const { return base::src_desc(0); }
9653
9655 memory::desc dst_desc() const { return base::dst_desc(0); }
9656 };
9657
9659 shuffle_forward() = default;
9660
9665};
9666
9671 struct desc {
9673
9681 desc(const memory::desc &diff_data_desc, int axis, int group_size) {
9683 &diff_data_desc.data, axis, group_size),
9684 "could not create a descriptor for a shuffle backward "
9685 "propagation primitive");
9686 }
9687 };
9688
9692 primitive_desc() = default;
9693
9708 primitive_desc(const desc &adesc, const engine &aengine,
9709 const shuffle_forward::primitive_desc &hint_fwd_pd,
9710 const primitive_attr &attr = primitive_attr(),
9711 bool allow_empty = false)
9712 : dnnl::primitive_desc(&adesc.data, &attr, aengine,
9713 hint_fwd_pd.get(), allow_empty) {}
9714
9722 : dnnl::primitive_desc(pd, dnnl::primitive::kind::shuffle,
9724
9727
9730 };
9731
9733 shuffle_backward() = default;
9734
9739};
9740
9742
9750
9752struct binary : public primitive {
9754 struct desc {
9757
9759 desc() = default;
9760
9768 desc(algorithm aalgorithm, const memory::desc &src0,
9769 const memory::desc &src1, const memory::desc &dst) {
9772 &src0.data, &src1.data, &dst.data),
9773 "could not create a descriptor for a binary operation "
9774 "primitive");
9775 }
9776 };
9777
9781 primitive_desc() = default;
9782
9792 primitive_desc(const desc &adesc, const engine &aengine,
9793 bool allow_empty = false)
9795 &adesc.data, nullptr, aengine, nullptr, allow_empty) {}
9796
9807 primitive_desc(const desc &adesc, const primitive_attr &attr,
9808 const engine &aengine, bool allow_empty = false)
9810 &adesc.data, &attr, aengine, nullptr, allow_empty) {}
9811
9818
9820 memory::desc src_desc(int idx = 0) const { return base::src_desc(idx); }
9821
9824
9827
9829 memory::desc dst_desc() const { return base::dst_desc(0); }
9830 };
9831
9833 binary() = default;
9834
9838 binary(const primitive_desc &pd) : primitive(pd) {}
9839};
9840
9842
9852
9854struct matmul : public primitive {
9856 struct desc {
9857 dnnl_matmul_desc_t data;
9858
9864 desc(const memory::desc &src_desc, const memory::desc &weights_desc,
9865 const memory::desc &dst_desc) {
9867 dnnl_matmul_desc_init(&data, &src_desc.data,
9868 &weights_desc.data, nullptr, &dst_desc.data),
9869 "could not create a descriptor for a matmul primitive");
9870 }
9871
9878 desc(const memory::desc &src_desc, const memory::desc &weights_desc,
9879 const memory::desc &bias_desc, const memory::desc &dst_desc) {
9881 &weights_desc.data, &bias_desc.data,
9882 &dst_desc.data),
9883 "could not create a descriptor for a matmul primitive");
9884 }
9885 };
9886
9890 primitive_desc() = default;
9891
9900 primitive_desc(const desc &adesc, const engine &aengine,
9901 bool allow_empty = false)
9903 &adesc.data, nullptr, aengine, nullptr, allow_empty) {}
9904
9914 primitive_desc(const desc &adesc, const primitive_attr &attr,
9915 const engine &aengine, bool allow_empty = false)
9917 &adesc.data, &attr, aengine, nullptr, allow_empty) {}
9918
9925
9928
9931 return query_md(query::weights_md, 0);
9932 }
9933
9936 return query_md(query::weights_md, 1);
9937 }
9938
9941 };
9942
9944 matmul() = default;
9945
9948 matmul(const primitive_desc &pd) : primitive(pd) {}
9949};
9950
9952
9962
9966 struct desc {
9968
9984 desc(prop_kind aprop_kind, algorithm aalgorithm,
9985 const memory::desc &src_desc, const memory::desc &dst_desc) {
9987 dnnl::convert_to_c(aprop_kind),
9988 convert_to_c(aalgorithm), nullptr,
9989 &src_desc.data, &dst_desc.data),
9990 "could not create a resampling forward descriptor");
9991 }
9992
10004 desc(prop_kind aprop_kind, algorithm aalgorithm,
10005 const std::vector<float> &factors,
10006 const memory::desc &src_desc) {
10007 memory::validate_dims(factors, src_desc.data.ndims - 2);
10009 dnnl::convert_to_c(aprop_kind),
10010 convert_to_c(aalgorithm), &factors[0],
10011 &src_desc.data, nullptr),
10012 "could not create a resampling forward descriptor");
10013 }
10014
10031 desc(prop_kind aprop_kind, algorithm aalgorithm,
10032 const std::vector<float> &factors, const memory::desc &src_desc,
10033 const memory::desc &dst_desc) {
10034 if (!factors.empty())
10035 memory::validate_dims(factors, src_desc.data.ndims - 2);
10037 dnnl::convert_to_c(aprop_kind),
10038 convert_to_c(aalgorithm), factors.data(),
10039 &src_desc.data, &dst_desc.data),
10040 "could not create a resampling forward descriptor");
10041 }
10042 };
10043
10047 primitive_desc() = default;
10048
10059 primitive_desc(const desc &adesc, const engine &aengine,
10060 bool allow_empty = false)
10062 &adesc.data, nullptr, aengine, nullptr, allow_empty) {}
10063
10075 primitive_desc(const desc &adesc, const primitive_attr &attr,
10076 const engine &aengine, bool allow_empty = false)
10078 &adesc.data, &attr, aengine, nullptr, allow_empty) {}
10079
10087 : dnnl::primitive_desc(pd, dnnl::primitive::kind::resampling,
10090
10092 memory::desc src_desc() const { return base::src_desc(0); }
10093
10095 memory::desc dst_desc() const { return base::dst_desc(0); }
10096 };
10097
10100
10105};
10106
10110 struct desc {
10112
10121 desc(algorithm aalgorithm, const memory::desc &diff_src_desc,
10122 const memory::desc &diff_dst_desc) {
10124 convert_to_c(aalgorithm), nullptr,
10125 &diff_src_desc.data, &diff_dst_desc.data),
10126 "could not create a resampling backward data descriptor");
10127 }
10128
10138 desc(algorithm aalgorithm, const std::vector<float> &factors,
10139 const memory::desc &diff_src_desc,
10140 const memory::desc &diff_dst_desc) {
10141 if (!factors.empty())
10142 memory::validate_dims(factors, diff_src_desc.data.ndims - 2);
10144 convert_to_c(aalgorithm), factors.data(),
10145 &diff_src_desc.data, &diff_dst_desc.data),
10146 "could not create a resampling backward data descriptor");
10147 }
10148 };
10149
10153 primitive_desc() = default;
10154
10168 primitive_desc(const desc &adesc, const engine &aengine,
10169 const resampling_forward::primitive_desc &hint_fwd_pd,
10170 bool allow_empty = false)
10171 : dnnl::primitive_desc(&adesc.data, nullptr, aengine,
10172 hint_fwd_pd.get(), allow_empty) {}
10173
10188 primitive_desc(const desc &adesc, const primitive_attr &attr,
10189 const engine &aengine,
10190 const resampling_forward::primitive_desc &hint_fwd_pd,
10191 bool allow_empty = false)
10192 : dnnl::primitive_desc(&adesc.data, &attr, aengine,
10193 hint_fwd_pd.get(), allow_empty) {}
10194
10202 : dnnl::primitive_desc(pd, dnnl::primitive::kind::resampling,
10204
10207
10210 };
10211
10214
10219};
10220
10222
10230
10234 struct desc {
10236
10263 desc(prop_kind aprop_kind, algorithm aalgorithm,
10264 const memory::desc &src_desc, const memory::desc &dst_desc,
10265 const memory::dims &strides, const memory::dims &kernel,
10266 const memory::dims &dilation, const memory::dims &padding_l,
10267 const memory::dims &padding_r) {
10268 memory::validate_dims(strides, src_desc.data.ndims - 2);
10269 memory::validate_dims(kernel, src_desc.data.ndims - 2);
10270 memory::validate_dims(padding_l, src_desc.data.ndims - 2);
10271 memory::validate_dims(padding_r, src_desc.data.ndims - 2);
10272 memory::validate_dims(dilation, src_desc.data.ndims - 2);
10275 dnnl::convert_to_c(aprop_kind),
10276 convert_to_c(aalgorithm), &src_desc.data,
10277 &dst_desc.data, &strides[0], &kernel[0],
10278 &dilation[0], &padding_l[0], &padding_r[0]),
10279 "could not create a descriptor for a pooling forward "
10280 "propagation primitive");
10281 }
10282 };
10283
10287 primitive_desc() = default;
10288
10299 primitive_desc(const desc &adesc, const engine &aengine,
10300 bool allow_empty = false)
10302 &adesc.data, nullptr, aengine, nullptr, allow_empty) {}
10303
10315 primitive_desc(const desc &adesc, const primitive_attr &attr,
10316 const engine &aengine, bool allow_empty = false)
10318 &adesc.data, &attr, aengine, nullptr, allow_empty) {}
10319
10328 : dnnl::primitive_desc(pd, dnnl::primitive::kind::pooling_v2,
10331
10333 memory::desc src_desc() const { return base::src_desc(0); }
10334
10336 memory::desc dst_desc() const { return base::dst_desc(0); }
10337
10340 };
10341
10344
10350};
10351
10355 struct desc {
10357
10381 desc(algorithm aalgorithm, const memory::desc &diff_src_desc,
10382 const memory::desc &diff_dst_desc, const memory::dims &strides,
10383 const memory::dims &kernel, const memory::dims &dilation,
10384 const memory::dims &padding_l, const memory::dims &padding_r) {
10385 memory::validate_dims(strides, diff_src_desc.data.ndims - 2);
10386 memory::validate_dims(kernel, diff_src_desc.data.ndims - 2);
10387 memory::validate_dims(padding_l, diff_src_desc.data.ndims - 2);
10388 memory::validate_dims(padding_r, diff_src_desc.data.ndims - 2);
10389 memory::validate_dims(dilation, diff_src_desc.data.ndims - 2);
10392 convert_to_c(aalgorithm), &diff_src_desc.data,
10393 &diff_dst_desc.data, &strides[0], &kernel[0],
10394 &dilation[0], &padding_l[0], &padding_r[0]),
10395 "could not create a descriptor for a pooling backward "
10396 "propagation primitive");
10397 }
10398 };
10399
10404 primitive_desc() = default;
10405
10419 primitive_desc(const desc &adesc, const engine &aengine,
10420 const pooling_v2_forward::primitive_desc &hint_fwd_pd,
10421 bool allow_empty = false)
10422 : dnnl::primitive_desc(&adesc.data, nullptr, aengine,
10423 hint_fwd_pd.get(), allow_empty) {}
10424
10439 primitive_desc(const desc &adesc, const primitive_attr &attr,
10440 const engine &aengine,
10441 const pooling_v2_forward::primitive_desc &hint_fwd_pd,
10442 bool allow_empty = false)
10443 : dnnl::primitive_desc(&adesc.data, &attr, aengine,
10444 hint_fwd_pd.get(), allow_empty) {}
10445
10454 : dnnl::primitive_desc(pd, dnnl::primitive::kind::pooling_v2,
10456
10459
10462
10465 };
10466
10469
10475};
10476
10478
10487
10489struct prelu_forward : public primitive {
10491 struct desc {
10492 dnnl_prelu_desc_t data;
10493
10502 desc(prop_kind aprop_kind, const memory::desc &data_desc,
10503 const memory::desc &weight_desc) {
10505 dnnl::convert_to_c(aprop_kind),
10506 &data_desc.data, &weight_desc.data),
10507 "could not create a descriptor for a prelu forward "
10508 "propagation primitive");
10509 }
10510 };
10511
10515 primitive_desc() = default;
10516
10527 primitive_desc(const desc &adesc, const engine &aengine,
10528 bool allow_empty = false)
10530 &adesc.data, nullptr, aengine, nullptr, allow_empty) {}
10531
10543 primitive_desc(const desc &adesc, const primitive_attr &attr,
10544 const engine &aengine, bool allow_empty = false)
10546 &adesc.data, &attr, aengine, nullptr, allow_empty) {}
10547
10555 : dnnl::primitive_desc(pd, dnnl::primitive::kind::prelu,
10558
10560 memory::desc src_desc() const { return base::src_desc(0); }
10561
10563 memory::desc dst_desc() const { return base::dst_desc(0); }
10564 };
10565
10567 prelu_forward() = default;
10568
10573};
10574
10576struct prelu_backward : public primitive {
10578 struct desc {
10579 dnnl_prelu_desc_t data;
10580
10589 desc(const memory::desc &data_desc, const memory::desc &weight_desc,
10590 const memory::desc &diff_data_desc,
10591 const memory::desc &diff_weights_desc) {
10593 dnnl_prelu_backward_desc_init(&data, &data_desc.data,
10594 &weight_desc.data, &diff_data_desc.data,
10595 &diff_weights_desc.data),
10596 "could not create a descriptor for a prelu backward "
10597 "propagation primitive");
10598 }
10599 };
10600
10604 primitive_desc() = default;
10605
10619 primitive_desc(const desc &adesc, const engine &aengine,
10620 const prelu_forward::primitive_desc &hint_fwd_pd,
10621 bool allow_empty = false)
10622 : dnnl::primitive_desc(&adesc.data, nullptr, aengine,
10623 hint_fwd_pd.get(), allow_empty) {}
10624
10639 primitive_desc(const desc &adesc, const primitive_attr &attr,
10640 const engine &aengine,
10641 const prelu_forward::primitive_desc &hint_fwd_pd,
10642 bool allow_empty = false)
10643 : dnnl::primitive_desc(&adesc.data, &attr, aengine,
10644 hint_fwd_pd.get(), allow_empty) {}
10645
10653 : dnnl::primitive_desc(pd, dnnl::primitive::kind::prelu,
10655
10657 memory::desc src_desc() const { return base::src_desc(0); }
10658
10661
10664 };
10665
10667 prelu_backward() = default;
10668
10673};
10674
10676
10685
10687struct reduction : public primitive {
10689 struct desc {
10691
10693 desc() = default;
10694
10712 desc(algorithm aalgorithm, const memory::desc &src_desc,
10713 const memory::desc &dst_desc, float p, float eps) {
10715 dnnl_reduction_desc_init(&data, convert_to_c(aalgorithm),
10716 &src_desc.data, &dst_desc.data, p, eps),
10717 "could not create a reduction descriptor");
10718 }
10719 };
10720
10724 primitive_desc() = default;
10725
10734 primitive_desc(const desc &adesc, const engine &aengine,
10735 bool allow_empty = false)
10737 &adesc.data, nullptr, aengine, nullptr, allow_empty) {}
10738
10748 primitive_desc(const desc &adesc, const primitive_attr &attr,
10749 const engine &aengine, bool allow_empty = false)
10751 &adesc.data, &attr, aengine, nullptr, allow_empty) {}
10752
10759
10761 memory::desc src_desc() const { return base::src_desc(0); }
10762
10764 memory::desc dst_desc() const { return base::dst_desc(0); }
10765 };
10766
10768 reduction() = default;
10769
10773};
10774
10776
10778
10784
10787
10789enum class status {
10804};
10805
10807inline status set_verbose(int level) {
10808 return static_cast<status>(dnnl_set_verbose(level));
10809}
10810
10812inline const version_t *version() {
10813 return dnnl_version();
10814}
10815
10817inline status set_jit_dump(int enable) {
10818 return static_cast<status>(dnnl_set_jit_dump(enable));
10819}
10820
10822inline status set_jit_profiling_flags(unsigned flags) {
10823 return static_cast<status>(dnnl_set_jit_profiling_flags(flags));
10824}
10825
10827inline status set_jit_profiling_jitdumpdir(const std::string &dir) {
10828 return static_cast<status>(dnnl_set_jit_profiling_jitdumpdir(dir.c_str()));
10829}
10830
10832enum class cpu_isa {
10855};
10856
10859 return static_cast<status>(
10860 dnnl_set_max_cpu_isa(static_cast<dnnl_cpu_isa_t>(isa)));
10861}
10862
10865 return static_cast<cpu_isa>(dnnl_get_effective_cpu_isa());
10866}
10867
10869enum class cpu_isa_hints {
10874};
10875
10878 return static_cast<status>(dnnl_set_cpu_isa_hints(
10879 static_cast<dnnl_cpu_isa_hints_t>(isa_hints)));
10880}
10881
10884 return static_cast<cpu_isa_hints>(dnnl_get_cpu_isa_hints());
10885}
10886
10888
10894
10898 int result = 0;
10900 "could not get primitive cache capacity");
10901 return result;
10902}
10903
10905inline void set_primitive_cache_capacity(int capacity) {
10907 "could not set primitive cache capacity");
10908}
10909
10911
10918
10920inline status sgemm(char transa, char transb, dnnl_dim_t M, dnnl_dim_t N,
10921 dnnl_dim_t K, float alpha, const float *A, dnnl_dim_t lda,
10922 const float *B, dnnl_dim_t ldb, float beta, float *C, dnnl_dim_t ldc) {
10923 return static_cast<status>(dnnl_sgemm(
10924 transa, transb, M, N, K, alpha, A, lda, B, ldb, beta, C, ldc));
10925}
10926
10928inline status gemm_u8s8s32(char transa, char transb, char offsetc, dnnl_dim_t M,
10929 dnnl_dim_t N, dnnl_dim_t K, float alpha, const uint8_t *A,
10930 dnnl_dim_t lda, uint8_t ao, const int8_t *B, dnnl_dim_t ldb, int8_t bo,
10931 float beta, int32_t *C, dnnl_dim_t ldc, const int32_t *co) {
10932 return static_cast<status>(dnnl_gemm_u8s8s32(transa, transb, offsetc, M, N,
10933 K, alpha, A, lda, ao, B, ldb, bo, beta, C, ldc, co));
10934}
10935
10937inline status gemm_s8s8s32(char transa, char transb, char offsetc, dnnl_dim_t M,
10938 dnnl_dim_t N, dnnl_dim_t K, float alpha, const int8_t *A,
10939 dnnl_dim_t lda, int8_t ao, const int8_t *B, dnnl_dim_t ldb, int8_t bo,
10940 float beta, int32_t *C, dnnl_dim_t ldc, const int32_t *co) {
10941 return static_cast<status>(dnnl_gemm_s8s8s32(transa, transb, offsetc, M, N,
10942 K, alpha, A, lda, ao, B, ldb, bo, beta, C, ldc, co));
10943}
10944
10946
10947// implementation section
10948
10951 dnnl_primitive_t result;
10953 "could not create a primitive");
10954 reset(result);
10955}
10956
10957inline primitive::primitive(const primitive_desc &pd) : primitive(pd.get()) {}
10958
10959inline void primitive::execute(const stream &astream,
10960 const std::unordered_map<int, memory> &args) const {
10961 std::vector<dnnl_exec_arg_t> c_args;
10962 c_args.reserve(args.size());
10963 for (const auto &a : args)
10964 c_args.push_back({a.first, a.second.get(true)});
10965
10966 error::wrap_c_api(dnnl_primitive_execute(get(), astream.get(),
10967 (int)c_args.size(), c_args.data()),
10968 "could not execute a primitive");
10969}
10970
10972
10973#undef DNNL_DEFINE_BITMASK_OPS
10974
10975} // namespace dnnl
10976
10978
10981namespace oneapi {
10982// Note: without this guard, doxygen warns of potentially recursive namespace
10983#ifndef DOXYGEN_SHOULD_SKIP_THIS
10985namespace dnnl = ::dnnl;
10986#endif
10987} // namespace oneapi
10988
10990
10991#endif /* ONEAPI_DNNL_DNNL_HPP */
algorithm
Kinds of algorithms.
Definition: dnnl.hpp:470
dnnl_status_t DNNL_API dnnl_primitive_attr_set_rnn_data_qparams(dnnl_primitive_attr_t attr, const float scale, const float shift)
Set quantization scale and shift parameters for RNN data tensors.
dnnl_status_t DNNL_API dnnl_post_ops_get_params_sum_v2(const_dnnl_post_ops_t post_ops, int index, float *scale, dnnl_data_type_t *data_type)
Returns the parameters of an accumulation (sum) post-op with a data type parameter.
dnnl_status_t DNNL_API dnnl_post_ops_get_params_dw_k3s2p1(const_dnnl_post_ops_t post_ops, int index, dnnl_data_type_t *weights_data_type, dnnl_data_type_t *bias_data_type, dnnl_data_type_t *dst_data_type, dnnl_dim_t *count, int *mask, const float **scales)
Returns the parameters of an depthwise post-op with stride 2.
dnnl_status_t DNNL_API dnnl_primitive_attr_set_scratchpad_mode(dnnl_primitive_attr_t attr, dnnl_scratchpad_mode_t mode)
Sets primitive attributes scratchpad mode.
dnnl_status_t DNNL_API dnnl_primitive_attr_get_post_ops(const_dnnl_primitive_attr_t attr, const_dnnl_post_ops_t *post_ops)
Returns primitive attributes post-ops.
dnnl_status_t DNNL_API dnnl_primitive_attr_get_rnn_weights_qparams(const_dnnl_primitive_attr_t attr, dnnl_dim_t *count, int *mask, const float **scales)
Returns the quantization scaling factors for RNN weights tensors.
dnnl_status_t DNNL_API dnnl_post_ops_append_dw_k3s1p1(dnnl_post_ops_t post_ops, dnnl_data_type_t weights_data_type, dnnl_data_type_t bias_data_type, dnnl_data_type_t dst_data_type, dnnl_dim_t count, int mask, const float *scales)
Appends a depthwise post-op convolution with stride 1.
dnnl_status_t DNNL_API dnnl_post_ops_destroy(dnnl_post_ops_t post_ops)
Destroys post-ops.
dnnl_status_t DNNL_API dnnl_primitive_attr_set_zero_points(dnnl_primitive_attr_t attr, int arg, dnnl_dim_t count, int mask, const int32_t *zero_points)
Sets primitive attributes zero points for primitive operations for a given memory argument.
dnnl_status_t DNNL_API dnnl_primitive_attr_set_post_ops(dnnl_primitive_attr_t attr, const_dnnl_post_ops_t post_ops)
Sets primitive attributes post-ops.
dnnl_status_t DNNL_API dnnl_post_ops_append_sum(dnnl_post_ops_t post_ops, float scale)
Appends an accumulation (sum) to post-ops.
dnnl_status_t DNNL_API dnnl_primitive_attr_set_rnn_weights_qparams(dnnl_primitive_attr_t attr, dnnl_dim_t count, int mask, const float *scales)
Sets quantization scaling factors for RNN weights tensors.
dnnl_status_t DNNL_API dnnl_primitive_attr_destroy(dnnl_primitive_attr_t attr)
Destroys primitive attributes.
dnnl_status_t DNNL_API dnnl_post_ops_append_sum_v2(dnnl_post_ops_t post_ops, float scale, dnnl_data_type_t data_type)
Appends an accumulation v2 (sum) to post-ops.
int DNNL_API dnnl_post_ops_len(const_dnnl_post_ops_t post_ops)
Returns the length of post-ops.
dnnl_status_t DNNL_API dnnl_post_ops_append_binary(dnnl_post_ops_t post_ops, dnnl_alg_kind_t alg_kind, const dnnl_memory_desc_t *src1_desc)
Appends a binary post-op.
dnnl_status_t DNNL_API dnnl_post_ops_append_dw_k3s2p1(dnnl_post_ops_t post_ops, dnnl_data_type_t weights_data_type, dnnl_data_type_t bias_data_type, dnnl_data_type_t dst_data_type, dnnl_dim_t count, int mask, const float *scales)
Appends a depthwise post-op convolution with stride 2.
dnnl_status_t DNNL_API dnnl_primitive_attr_get_rnn_weights_projection_qparams(const_dnnl_primitive_attr_t attr, dnnl_dim_t *count, int *mask, const float **scales)
Returns the quantization scaling factors for RNN projection weights tensors.
dnnl_status_t DNNL_API dnnl_post_ops_get_params_eltwise(const_dnnl_post_ops_t post_ops, int index, float *scale, dnnl_alg_kind_t *alg_kind, float *alpha, float *beta)
Returns the parameters of an elementwise post-op.
dnnl_status_t DNNL_API dnnl_post_ops_create(dnnl_post_ops_t *post_ops)
Creates empty post-ops sequence.
dnnl_status_t DNNL_API dnnl_primitive_attr_set_scales(dnnl_primitive_attr_t attr, int arg, dnnl_dim_t count, int mask, const float *scales)
Sets primitive attributes scaling factors for primitive operations for a given memory argument.
dnnl_status_t DNNL_API dnnl_primitive_attr_get_scratchpad_mode(const_dnnl_primitive_attr_t attr, dnnl_scratchpad_mode_t *mode)
Returns the primitive attributes scratchpad mode.
dnnl_status_t DNNL_API dnnl_primitive_attr_clone(dnnl_primitive_attr_t *attr, const_dnnl_primitive_attr_t existing_attr)
Clones primitive attributes.
dnnl_status_t DNNL_API dnnl_post_ops_get_params_dw_k3s1p1(const_dnnl_post_ops_t post_ops, int index, dnnl_data_type_t *weights_data_type, dnnl_data_type_t *bias_data_type, dnnl_data_type_t *dst_data_type, dnnl_dim_t *count, int *mask, const float **scales)
Returns the parameters of an depthwise post-op with stride 1.
dnnl_primitive_kind_t DNNL_API dnnl_post_ops_get_kind(const_dnnl_post_ops_t post_ops, int index)
Returns the kind of a post-op entry.
scratchpad_mode
Scratchpad mode.
Definition: dnnl.hpp:401
dnnl_status_t DNNL_API dnnl_primitive_attr_set_rnn_weights_projection_qparams(dnnl_primitive_attr_t attr, dnnl_dim_t count, int mask, const float *scales)
Sets quantization scaling factors for RNN projection weights tensors.
prop_kind
Propagation kind.
Definition: dnnl.hpp:435
dnnl_scratchpad_mode_t
Scratchpad mode.
Definition: dnnl_types.h:2201
dnnl_status_t DNNL_API dnnl_post_ops_append_eltwise(dnnl_post_ops_t post_ops, float scale, dnnl_alg_kind_t alg_kind, float alpha, float beta)
Appends an elementwise post-op.
dnnl_status_t DNNL_API dnnl_primitive_attr_get_zero_points(const_dnnl_primitive_attr_t attr, int arg, dnnl_dim_t *count, int *mask, const int32_t **zero_points)
Returns count, correspondence zero point mask, and a pointer to a constant int32_t array of zero_poin...
dnnl_status_t DNNL_API dnnl_post_ops_get_params_sum(const_dnnl_post_ops_t post_ops, int index, float *scale)
Returns the parameters of an accumulation (sum) post-op.
dnnl_status_t DNNL_API dnnl_post_ops_get_params_binary(const_dnnl_post_ops_t post_ops, int index, dnnl_alg_kind_t *alg_kind, const dnnl_memory_desc_t **src1_desc)
Returns the parameters of a binary post-op.
dnnl_status_t DNNL_API dnnl_primitive_attr_get_rnn_data_qparams(const_dnnl_primitive_attr_t attr, float *scale, float *shift)
Returns the quantization scale and shift parameters for RNN data tensors.
dnnl_status_t DNNL_API dnnl_primitive_attr_set_output_scales(dnnl_primitive_attr_t attr, dnnl_dim_t count, int mask, const float *scales)
Sets output scaling factors correspondence mask and values.
dnnl_status_t DNNL_API dnnl_primitive_attr_get_scales(dnnl_primitive_attr_t attr, int arg, dnnl_dim_t *count, int *mask, const float **scales)
Returns primitive attributes scaling factors correspondence mask and values for a given memory argume...
dnnl_status_t DNNL_API dnnl_primitive_attr_create(dnnl_primitive_attr_t *attr)
Creates an empty (default) primitive attributes with all the parameters set to their default values.
dnnl_status_t DNNL_API dnnl_primitive_attr_get_output_scales(const_dnnl_primitive_attr_t attr, dnnl_dim_t *count, int *mask, const float **scales)
Returns primitive attributes output scaling factors correspondence mask and values.
@ resampling_linear
Linear (Bilinear, Trilinear) resampling method.
@ binary_mul
Binary mul.
@ resampling_nearest
Nearest Neighbor resampling method.
@ eltwise_elu_use_dst_for_bwd
Elementwise: exponential linear unit (ELU) (dst for backward)
@ eltwise_tanh_use_dst_for_bwd
Elementwise: hyperbolic tangent non-linearity (tanh) (dst for backward)
@ reduction_norm_lp_power_p_sum
Reduction using norm_lp_power_p_sum operation.
@ eltwise_linear
Elementwise: linear.
@ eltwise_clip_v2
Eltwise: clip version 2.
@ eltwise_soft_relu
Elementwise: soft_relu.
@ vanilla_gru
GRU cell.
@ eltwise_logistic
Elementwise: logistic.
@ binary_div
Binary div.
@ eltwise_clip
Elementwise: clip.
@ eltwise_abs
Elementwise: abs.
@ eltwise_pow
Elementwise: pow.
@ eltwise_tanh
Elementwise: hyperbolic tangent non-linearity (tanh)
@ eltwise_logistic_use_dst_for_bwd
Elementwise: logistic (dst for backward)
@ eltwise_bounded_relu
Elementwise: bounded_relu.
@ reduction_norm_lp_power_p_max
Reduction using norm_lp_power_p_max operation.
@ reduction_max
Reduction using max operation.
@ eltwise_clip_v2_use_dst_for_bwd
Elementwise: clip version 2 (dst for backward)
@ eltwise_square
Elementwise: square.
@ binary_max
Binary max.
@ convolution_direct
Direct convolution.
@ eltwise_exp
Elementwise: exponent.
@ reduction_norm_lp_max
Reduction using norm_lp_max operation.
@ eltwise_elu
Elementwise: exponential linear unit (ELU)
@ convolution_winograd
Winograd convolution.
@ vanilla_lstm
LSTM cell.
@ deconvolution_direct
Direct deconvolution.
@ pooling_avg
Average pooling exclude padding, alias for dnnl::algorithm::pooling_avg_include_padding.
@ lbr_gru
GRU cell with linear before reset.
@ pooling_avg_exclude_padding
Average pooling exclude padding.
@ eltwise_gelu
Elementwise: gelu alias for dnnl::algorithm::eltwise_gelu_tanh.
@ eltwise_sqrt
Elementwise: square root.
@ pooling_max
Max pooling.
@ reduction_min
Reduction using min operation.
@ eltwise_gelu_erf
Elementwise: erf-based gelu.
@ eltwise_swish
Elementwise: swish ( )
@ binary_sub
Binary sub.
@ lrn_within_channel
LRN within a single channel.
@ reduction_mul
Reduction using mul operation.
@ vanilla_rnn
RNN cell.
@ binary_add
Binary add.
@ lrn_across_channels
Local response normalization (LRN) across multiple channels.
@ eltwise_relu
Elementwise: rectified linear unit (ReLU)
@ eltwise_gelu_tanh
Elementwise: tanh-based gelu.
@ eltwise_relu_use_dst_for_bwd
Elementwise: rectified linar unit (ReLU) (dst for backward)
@ eltwise_logsigmoid
Elementwise: logsigmoid.
@ convolution_auto
Convolution algorithm that is chosen to be either direct or Winograd automatically.
@ binary_min
Binary min.
@ eltwise_exp_use_dst_for_bwd
Elementwise: exponent (dst for backward)
@ eltwise_round
Elementwise: round.
@ eltwise_sqrt_use_dst_for_bwd
Elementwise: square root (dst for backward)
@ pooling_avg_include_padding
Average pooling include padding.
@ reduction_norm_lp_sum
Reduction using norm_lp_sum operation.
@ reduction_mean
Reduction using mean operation.
@ deconvolution_winograd
Winograd deconvolution.
@ eltwise_log
Elementwise: natural logarithm.
@ reduction_sum
Reduction using sum operation.
@ library
The library manages the scratchpad allocation according to the policy specified by the DNNL_ENABLE_CO...
@ user
The user manages the scratchpad allocation by querying and providing the scratchpad memory to primiti...
@ backward
Backward propagation (with respect to all parameters).
@ backward_weights
Backward weights propagation.
@ forward_training
Forward data propagation (training mode).
@ forward_inference
Forward data propagation (inference mode).
@ forward_scoring
Forward data propagation, alias for dnnl::prop_kind::forward_inference.
@ forward
Forward data propagation, alias for dnnl::prop_kind::forward_training.
@ backward_data
Backward data propagation.
@ backward_bias
Backward bias propagation.
@ undef
Undefined propagation kind.
@ dnnl_scratchpad_mode_user
The user manages the scratchpad allocation by querying and providing the scratchpad memory to primiti...
Definition: dnnl_types.h:2223
@ dnnl_scratchpad_mode_library
The library manages the scratchpad allocation according to the policy specified by the DNNL_ENABLE_CO...
Definition: dnnl_types.h:2218
dnnl_status_t DNNL_API dnnl_batch_normalization_backward_desc_init(dnnl_batch_normalization_desc_t *bnrm_desc, dnnl_prop_kind_t prop_kind, const dnnl_memory_desc_t *diff_data_desc, const dnnl_memory_desc_t *data_desc, float epsilon, unsigned flags)
Initializes a descriptor for a batch normalization backward propagation primitive.
dnnl_status_t DNNL_API dnnl_batch_normalization_forward_desc_init(dnnl_batch_normalization_desc_t *bnrm_desc, dnnl_prop_kind_t prop_kind, const dnnl_memory_desc_t *data_desc, float epsilon, unsigned flags)
Initializes a descriptor for a batch normalization forward propagation primitive.
dnnl_status_t DNNL_API dnnl_binary_desc_init(dnnl_binary_desc_t *binary_desc, dnnl_alg_kind_t alg_kind, const dnnl_memory_desc_t *src0_desc, const dnnl_memory_desc_t *src1_desc, const dnnl_memory_desc_t *dst_desc)
Initializes a descriptor for a binary primitive.
dnnl_status_t DNNL_API dnnl_gemm_s8s8s32(char transa, char transb, char offsetc, dnnl_dim_t M, dnnl_dim_t N, dnnl_dim_t K, float alpha, const int8_t *A, dnnl_dim_t lda, int8_t ao, const int8_t *B, dnnl_dim_t ldb, int8_t bo, float beta, int32_t *C, dnnl_dim_t ldc, const int32_t *co)
Performs integer matrix-matrix multiply on 8-bit signed matrix A, 8-bit signed matrix B,...
status gemm_u8s8s32(char transa, char transb, char offsetc, dnnl_dim_t M, dnnl_dim_t N, dnnl_dim_t K, float alpha, const uint8_t *A, dnnl_dim_t lda, uint8_t ao, const int8_t *B, dnnl_dim_t ldb, int8_t bo, float beta, int32_t *C, dnnl_dim_t ldc, const int32_t *co)
Performs integer matrix-matrix multiply on 8-bit unsigned matrix A, 8-bit signed matrix B,...
Definition: dnnl.hpp:10928
status gemm_s8s8s32(char transa, char transb, char offsetc, dnnl_dim_t M, dnnl_dim_t N, dnnl_dim_t K, float alpha, const int8_t *A, dnnl_dim_t lda, int8_t ao, const int8_t *B, dnnl_dim_t ldb, int8_t bo, float beta, int32_t *C, dnnl_dim_t ldc, const int32_t *co)
Performs integer matrix-matrix multiply on 8-bit signed matrix A, 8-bit signed matrix B,...
Definition: dnnl.hpp:10937
dnnl_status_t DNNL_API dnnl_sgemm(char transa, char transb, dnnl_dim_t M, dnnl_dim_t N, dnnl_dim_t K, float alpha, const float *A, dnnl_dim_t lda, const float *B, dnnl_dim_t ldb, float beta, float *C, dnnl_dim_t ldc)
Performs single-precision matrix-matrix multiply.
status sgemm(char transa, char transb, dnnl_dim_t M, dnnl_dim_t N, dnnl_dim_t K, float alpha, const float *A, dnnl_dim_t lda, const float *B, dnnl_dim_t ldb, float beta, float *C, dnnl_dim_t ldc)
Performs single-precision matrix-matrix multiply.
Definition: dnnl.hpp:10920
dnnl_status_t DNNL_API dnnl_gemm_u8s8s32(char transa, char transb, char offsetc, dnnl_dim_t M, dnnl_dim_t N, dnnl_dim_t K, float alpha, const uint8_t *A, dnnl_dim_t lda, uint8_t ao, const int8_t *B, dnnl_dim_t ldb, int8_t bo, float beta, int32_t *C, dnnl_dim_t ldc, const int32_t *co)
Performs integer matrix-matrix multiply on 8-bit unsigned matrix A, 8-bit signed matrix B,...
dnnl_status_t DNNL_API dnnl_concat_primitive_desc_create(dnnl_primitive_desc_t *concat_primitive_desc, const dnnl_memory_desc_t *dst_desc, int n, int concat_dimension, const dnnl_memory_desc_t *src_descs, const_dnnl_primitive_attr_t attr, dnnl_engine_t engine)
Creates a primitive descriptor for an out-of-place concatenation primitive.
dnnl_status_t DNNL_API dnnl_convolution_forward_desc_init(dnnl_convolution_desc_t *conv_desc, dnnl_prop_kind_t prop_kind, dnnl_alg_kind_t alg_kind, const dnnl_memory_desc_t *src_desc, const dnnl_memory_desc_t *weights_desc, const dnnl_memory_desc_t *bias_desc, const dnnl_memory_desc_t *dst_desc, const dnnl_dims_t strides, const dnnl_dims_t padding_l, const dnnl_dims_t padding_r)
Initializes a descriptor for a convolution forward propagation primitive.
dnnl_status_t DNNL_API dnnl_dilated_convolution_backward_weights_desc_init(dnnl_convolution_desc_t *conv_desc, dnnl_alg_kind_t alg_kind, const dnnl_memory_desc_t *src_desc, const dnnl_memory_desc_t *diff_weights_desc, const dnnl_memory_desc_t *diff_bias_desc, const dnnl_memory_desc_t *diff_dst_desc, const dnnl_dims_t strides, const dnnl_dims_t dilates, const dnnl_dims_t padding_l, const dnnl_dims_t padding_r)
Initializes a descriptor for a dilated convolution weights gradient primitive.
dnnl_status_t DNNL_API dnnl_dilated_convolution_forward_desc_init(dnnl_convolution_desc_t *conv_desc, dnnl_prop_kind_t prop_kind, dnnl_alg_kind_t alg_kind, const dnnl_memory_desc_t *src_desc, const dnnl_memory_desc_t *weights_desc, const dnnl_memory_desc_t *bias_desc, const dnnl_memory_desc_t *dst_desc, const dnnl_dims_t strides, const dnnl_dims_t dilates, const dnnl_dims_t padding_l, const dnnl_dims_t padding_r)
Initializes a descriptor for a dilated convolution forward propagation primitive.
dnnl_status_t DNNL_API dnnl_convolution_backward_weights_desc_init(dnnl_convolution_desc_t *conv_desc, dnnl_alg_kind_t alg_kind, const dnnl_memory_desc_t *src_desc, const dnnl_memory_desc_t *diff_weights_desc, const dnnl_memory_desc_t *diff_bias_desc, const dnnl_memory_desc_t *diff_dst_desc, const dnnl_dims_t strides, const dnnl_dims_t padding_l, const dnnl_dims_t padding_r)
Initializes a descriptor for a convolution weights gradient primitive.
dnnl_status_t DNNL_API dnnl_dilated_convolution_backward_data_desc_init(dnnl_convolution_desc_t *conv_desc, dnnl_alg_kind_t alg_kind, const dnnl_memory_desc_t *diff_src_desc, const dnnl_memory_desc_t *weights_desc, const dnnl_memory_desc_t *diff_dst_desc, const dnnl_dims_t strides, const dnnl_dims_t dilates, const dnnl_dims_t padding_l, const dnnl_dims_t padding_r)
Initializes a descriptor for a dilated convolution backward propagation primitive.
dnnl_status_t DNNL_API dnnl_convolution_backward_data_desc_init(dnnl_convolution_desc_t *conv_desc, dnnl_alg_kind_t alg_kind, const dnnl_memory_desc_t *diff_src_desc, const dnnl_memory_desc_t *weights_desc, const dnnl_memory_desc_t *diff_dst_desc, const dnnl_dims_t strides, const dnnl_dims_t padding_l, const dnnl_dims_t padding_r)
Initializes a descriptor for a convolution backward propagation primitive.
dnnl_status_t DNNL_API dnnl_dilated_deconvolution_backward_data_desc_init(dnnl_deconvolution_desc_t *deconv_desc, dnnl_alg_kind_t alg_kind, const dnnl_memory_desc_t *diff_src_desc, const dnnl_memory_desc_t *weights_desc, const dnnl_memory_desc_t *diff_dst_desc, const dnnl_dims_t strides, const dnnl_dims_t dilates, const dnnl_dims_t padding_l, const dnnl_dims_t padding_r)
Initializes a descriptor for a dilated deconvolution backward propagation primitive.
dnnl_status_t DNNL_API dnnl_deconvolution_forward_desc_init(dnnl_deconvolution_desc_t *deconv_desc, dnnl_prop_kind_t prop_kind, dnnl_alg_kind_t alg_kind, const dnnl_memory_desc_t *src_desc, const dnnl_memory_desc_t *weights_desc, const dnnl_memory_desc_t *bias_desc, const dnnl_memory_desc_t *dst_desc, const dnnl_dims_t strides, const dnnl_dims_t padding_l, const dnnl_dims_t padding_r)
Initializes a descriptor for a deconvolution forward propagation primitive.
dnnl_status_t DNNL_API dnnl_deconvolution_backward_weights_desc_init(dnnl_deconvolution_desc_t *deconv_desc, dnnl_alg_kind_t alg_kind, const dnnl_memory_desc_t *src_desc, const dnnl_memory_desc_t *diff_weights_desc, const dnnl_memory_desc_t *diff_bias_desc, const dnnl_memory_desc_t *diff_dst_desc, const dnnl_dims_t strides, const dnnl_dims_t padding_l, const dnnl_dims_t padding_r)
Initializes a descriptor for a deconvolution weights gradient primitive.
dnnl_status_t DNNL_API dnnl_deconvolution_backward_data_desc_init(dnnl_deconvolution_desc_t *deconv_desc, dnnl_alg_kind_t alg_kind, const dnnl_memory_desc_t *diff_src_desc, const dnnl_memory_desc_t *weights_desc, const dnnl_memory_desc_t *diff_dst_desc, const dnnl_dims_t strides, const dnnl_dims_t padding_l, const dnnl_dims_t padding_r)
Initializes a descriptor for a deconvolution backward propagation primitive.
dnnl_status_t DNNL_API dnnl_dilated_deconvolution_forward_desc_init(dnnl_deconvolution_desc_t *deconv_desc, dnnl_prop_kind_t prop_kind, dnnl_alg_kind_t alg_kind, const dnnl_memory_desc_t *src_desc, const dnnl_memory_desc_t *weights_desc, const dnnl_memory_desc_t *bias_desc, const dnnl_memory_desc_t *dst_desc, const dnnl_dims_t strides, const dnnl_dims_t dilates, const dnnl_dims_t padding_l, const dnnl_dims_t padding_r)
Initializes a descriptor for a dilated deconvolution forward propagation primitive.
dnnl_status_t DNNL_API dnnl_dilated_deconvolution_backward_weights_desc_init(dnnl_deconvolution_desc_t *deconv_desc, dnnl_alg_kind_t alg_kind, const dnnl_memory_desc_t *src_desc, const dnnl_memory_desc_t *diff_weights_desc, const dnnl_memory_desc_t *diff_bias_desc, const dnnl_memory_desc_t *diff_dst_desc, const dnnl_dims_t strides, const dnnl_dims_t dilates, const dnnl_dims_t padding_l, const dnnl_dims_t padding_r)
Initializes a descriptor for a dilated deconvolution weights gradient primitive.
dnnl_status_t DNNL_API dnnl_eltwise_forward_desc_init(dnnl_eltwise_desc_t *eltwise_desc, dnnl_prop_kind_t prop_kind, dnnl_alg_kind_t alg_kind, const dnnl_memory_desc_t *data_desc, float alpha, float beta)
Initializes a descriptor for eltwise forward propagation primitive.
dnnl_status_t DNNL_API dnnl_eltwise_backward_desc_init(dnnl_eltwise_desc_t *eltwise_desc, dnnl_alg_kind_t alg_kind, const dnnl_memory_desc_t *diff_data_desc, const dnnl_memory_desc_t *data_desc, float alpha, float beta)
Initializes a descriptor for eltwise backward propagation primitive.
dnnl_engine_kind_t
Kinds of engines.
Definition: dnnl_types.h:2147
dnnl_status_t DNNL_API dnnl_engine_get_kind(dnnl_engine_t engine, dnnl_engine_kind_t *kind)
Returns the kind of an engine.
dnnl_status_t DNNL_API dnnl_engine_destroy(dnnl_engine_t engine)
Destroys an engine.
dnnl_status_t DNNL_API dnnl_engine_create(dnnl_engine_t *engine, dnnl_engine_kind_t kind, size_t index)
Creates an engine.
size_t DNNL_API dnnl_engine_get_count(dnnl_engine_kind_t kind)
Returns the number of engines of a particular kind.
dnnl_engine_kind_t convert_to_c(engine::kind akind)
Converts engine kind enum value from C++ API to C API type.
Definition: dnnl.hpp:961
@ dnnl_gpu
GPU engine.
Definition: dnnl_types.h:2153
@ dnnl_cpu
CPU engine.
Definition: dnnl_types.h:2151
@ dnnl_any_engine
An unspecified engine.
Definition: dnnl_types.h:2149
dnnl_status_t DNNL_API dnnl_inner_product_forward_desc_init(dnnl_inner_product_desc_t *ip_desc, dnnl_prop_kind_t prop_kind, const dnnl_memory_desc_t *src_desc, const dnnl_memory_desc_t *weights_desc, const dnnl_memory_desc_t *bias_desc, const dnnl_memory_desc_t *dst_desc)
Initializes descriptor for inner product forward propagation.
dnnl_status_t DNNL_API dnnl_inner_product_backward_weights_desc_init(dnnl_inner_product_desc_t *ip_desc, const dnnl_memory_desc_t *src_desc, const dnnl_memory_desc_t *diff_weights_desc, const dnnl_memory_desc_t *diff_bias_desc, const dnnl_memory_desc_t *diff_dst_desc)
Initializes descriptor for inner product weights gradient primitive.
dnnl_status_t DNNL_API dnnl_inner_product_backward_data_desc_init(dnnl_inner_product_desc_t *ip_desc, const dnnl_memory_desc_t *diff_src_desc, const dnnl_memory_desc_t *weights_desc, const dnnl_memory_desc_t *diff_dst_desc)
Initializes descriptor for inner product backward propagation.
dnnl_status_t DNNL_API dnnl_layer_normalization_backward_desc_init(dnnl_layer_normalization_desc_t *lnrm_desc, dnnl_prop_kind_t prop_kind, const dnnl_memory_desc_t *diff_data_desc, const dnnl_memory_desc_t *data_desc, const dnnl_memory_desc_t *stat_desc, float epsilon, unsigned flags)
Initializes a descriptor for a layer normalization backward propagation primitive.
dnnl_status_t DNNL_API dnnl_layer_normalization_forward_desc_init(dnnl_layer_normalization_desc_t *lnrm_desc, dnnl_prop_kind_t prop_kind, const dnnl_memory_desc_t *data_desc, const dnnl_memory_desc_t *stat_desc, float epsilon, unsigned flags)
Initializes a descriptor for layer normalization forward propagation primitive.
dnnl_status_t DNNL_API dnnl_logsoftmax_forward_desc_init(dnnl_logsoftmax_desc_t *logsoftmax_desc, dnnl_prop_kind_t prop_kind, const dnnl_memory_desc_t *data_desc, int logsoftmax_axis)
Initializes a descriptor for logsoftmax forward propagation primitive.
dnnl_status_t DNNL_API dnnl_logsoftmax_backward_desc_init(dnnl_logsoftmax_desc_t *logsoftmax_desc, const dnnl_memory_desc_t *diff_data_desc, const dnnl_memory_desc_t *data_desc, int logsoftmax_axis)
Initializes a descriptor for logsoftmax backward propagation primitive.
dnnl_status_t DNNL_API dnnl_lrn_backward_desc_init(dnnl_lrn_desc_t *lrn_desc, dnnl_alg_kind_t alg_kind, const dnnl_memory_desc_t *diff_data_desc, const dnnl_memory_desc_t *data_desc, dnnl_dim_t local_size, float alpha, float beta, float k)
Initializes a descriptor for LRN backward propagation primitive.
dnnl_status_t DNNL_API dnnl_lrn_forward_desc_init(dnnl_lrn_desc_t *lrn_desc, dnnl_prop_kind_t prop_kind, dnnl_alg_kind_t alg_kind, const dnnl_memory_desc_t *data_desc, dnnl_dim_t local_size, float alpha, float beta, float k)
Initializes a descriptor for LRN forward propagation primitive.
dnnl_status_t DNNL_API dnnl_matmul_desc_init(dnnl_matmul_desc_t *matmul_desc, const dnnl_memory_desc_t *src_desc, const dnnl_memory_desc_t *weights_desc, const dnnl_memory_desc_t *bias_desc, const dnnl_memory_desc_t *dst_desc)
Initializes a matrix multiplication descriptor.
dnnl_data_type_t
Data type specification.
Definition: dnnl_types.h:62
dnnl_status_t DNNL_API dnnl_memory_desc_init_submemory(dnnl_memory_desc_t *memory_desc, const dnnl_memory_desc_t *parent_memory_desc, const dnnl_dims_t dims, const dnnl_dims_t offsets)
Initializes a memory descriptor for a region inside an area described by an existing memory descripto...
dnnl_format_tag_t
Memory format tag specification.
Definition: dnnl_types.h:164
dnnl_status_t DNNL_API dnnl_memory_desc_permute_axes(dnnl_memory_desc_t *out_memory_desc, const dnnl_memory_desc_t *in_memory_desc, const int *permutation)
Initializes a memory descriptor by permuting axes in an existing one.
dnnl_status_t DNNL_API dnnl_memory_unmap_data(const_dnnl_memory_t memory, void *mapped_ptr)
Unmaps a memory object and writes back any changes made to the previously mapped memory buffer.
dnnl_status_t DNNL_API dnnl_memory_create(dnnl_memory_t *memory, const dnnl_memory_desc_t *memory_desc, dnnl_engine_t engine, void *handle)
Creates a memory object.
dnnl_status_t DNNL_API dnnl_memory_get_engine(const_dnnl_memory_t memory, dnnl_engine_t *engine)
Returns the engine of a memory object.
dnnl_status_t DNNL_API dnnl_memory_desc_reshape(dnnl_memory_desc_t *out_memory_desc, const dnnl_memory_desc_t *in_memory_desc, int ndims, const dnnl_dims_t dims)
Initializes a memory descriptor by reshaping an existing one.
dnnl_status_t DNNL_API dnnl_memory_get_memory_desc(const_dnnl_memory_t memory, const dnnl_memory_desc_t **memory_desc)
Returns the memory descriptor for a memory object.
dnnl_status_t DNNL_API dnnl_memory_get_data_handle(const_dnnl_memory_t memory, void **handle)
Returns memory object's data handle.
dnnl_status_t DNNL_API dnnl_memory_set_data_handle_v2(dnnl_memory_t memory, void *handle, dnnl_stream_t stream)
Sets the underlying memory buffer.
dnnl_status_t DNNL_API dnnl_memory_desc_init_by_strides(dnnl_memory_desc_t *memory_desc, int ndims, const dnnl_dims_t dims, dnnl_data_type_t data_type, const dnnl_dims_t strides)
Initializes a memory descriptor using dimensions and strides.
int64_t dnnl_dim_t
A type to describe tensor dimension.
Definition: dnnl_types.h:1333
dnnl_status_t DNNL_API dnnl_memory_destroy(dnnl_memory_t memory)
Destroys a memory object.
int DNNL_API dnnl_memory_desc_equal(const dnnl_memory_desc_t *lhs, const dnnl_memory_desc_t *rhs)
Compares two memory descriptors.
#define DNNL_MAX_NDIMS
Maximum number of dimensions a tensor can have.
Definition: dnnl_types.h:1301
dnnl_status_t DNNL_API dnnl_memory_map_data(const_dnnl_memory_t memory, void **mapped_ptr)
Maps a memory object and returns a host-side pointer to a memory buffer with a copy of its contents.
size_t DNNL_API dnnl_memory_desc_get_size(const dnnl_memory_desc_t *memory_desc)
Returns the size of a memory descriptor.
#define DNNL_MEMORY_ALLOCATE
Special pointer value that indicates that the library needs to allocate an underlying buffer for a me...
Definition: dnnl_types.h:1510
dnnl_status_t DNNL_API dnnl_memory_desc_init_by_tag(dnnl_memory_desc_t *memory_desc, int ndims, const dnnl_dims_t dims, dnnl_data_type_t data_type, dnnl_format_tag_t tag)
Initializes a memory descriptor using dimensions and memory format tag.
@ dnnl_f16
16-bit/half-precision floating point.
Definition: dnnl_types.h:66
@ dnnl_bf16
non-standard 16-bit (bfloat16 w/ 7 bit mantissa) floating point.
Definition: dnnl_types.h:68
@ dnnl_f32
32-bit/single-precision floating point.
Definition: dnnl_types.h:70
@ dnnl_data_type_undef
Undefined data type, used for empty memory descriptors.
Definition: dnnl_types.h:64
@ dnnl_s8
8-bit signed integer.
Definition: dnnl_types.h:74
@ dnnl_s32
32-bit signed integer.
Definition: dnnl_types.h:72
@ dnnl_u8
8-bit unsigned integer.
Definition: dnnl_types.h:76
@ dnnl_abcdefhg
permuted 8D tensor
Definition: dnnl_types.h:216
@ dnnl_aBCdef2b4c2b
6D tensor blocked by 3rd dimension with block size 4
Definition: dnnl_types.h:362
@ dnnl_abcdefghi
plain 9D tensor
Definition: dnnl_types.h:186
@ dnnl_acdeb
permuted 5D tensor
Definition: dnnl_types.h:199
@ dnnl_abcdefgh
plain 8D tensor
Definition: dnnl_types.h:185
@ dnnl_abcdefghikj
permuted 11D tensor
Definition: dnnl_types.h:219
@ dnnl_ab
plain 2D tensor
Definition: dnnl_types.h:178
@ dnnl_ABcd8b8a
4D tensor blocked by 1st and 2nd dimension with block size 8
Definition: dnnl_types.h:288
@ dnnl_cdba
permuted 4D tensor
Definition: dnnl_types.h:208
@ dnnl_abcdefghijkl
plain 12D tensor
Definition: dnnl_types.h:189
@ dnnl_aBcdef4b
6D tensor blocked by 2nd dimension with block size 4
Definition: dnnl_types.h:364
@ dnnl_abcdegf
permuted 7D tensor
Definition: dnnl_types.h:215
@ dnnl_abcdfe
permuted 6D tensor
Definition: dnnl_types.h:214
@ dnnl_aBcd4b
4D tensor blocked by 2nd dimension with block size 4
Definition: dnnl_types.h:263
@ dnnl_nCdhw16c
5D CNN activations tensor blocked by channels with block size 16, an alias to dnnl_aBcde16b
Definition: dnnl_types.h:682
@ dnnl_abcde
plain 5D tensor
Definition: dnnl_types.h:182
@ dnnl_decab
permuted 5D tensor
Definition: dnnl_types.h:211
@ dnnl_bca
permuted 3D tensor
Definition: dnnl_types.h:204
@ dnnl_aBcde4b
5D tensor blocked by 2nd dimension with block size 4
Definition: dnnl_types.h:315
@ dnnl_aBc16b
3D tensor blocked by 2nd dimension with block size 16
Definition: dnnl_types.h:229
@ dnnl_aBcdef16b
6D tensor blocked by 2nd dimension with block size 16
Definition: dnnl_types.h:354
@ dnnl_aBCde2b4c2b
5D tensor blocked by 3rd dimension with block size 4
Definition: dnnl_types.h:352
@ dnnl_aBc4b
3D tensor blocked by 2nd dimension with block size 4
Definition: dnnl_types.h:235
@ dnnl_abcdefghijk
plain 11D tensor
Definition: dnnl_types.h:188
@ dnnl_bacde
permuted 5D tensor
Definition: dnnl_types.h:203
@ dnnl_aBcd16b
4D tensor blocked by 2nd dimension with block size 16
Definition: dnnl_types.h:255
@ dnnl_cba
permuted 3D tensor
Definition: dnnl_types.h:207
@ dnnl_ba
permuted 2D tensor
Definition: dnnl_types.h:200
@ dnnl_ABcde2b8a4b
5D tensor blocked by 1st dimension with block size 8
Definition: dnnl_types.h:304
@ dnnl_abcd
plain 4D tensor
Definition: dnnl_types.h:180
@ dnnl_format_tag_undef
Undefined memory format tag.
Definition: dnnl_types.h:166
@ dnnl_nCdhw4c
5D CNN activations tensor blocked by channels with block size 4, an alias to dnnl_aBcde4b
Definition: dnnl_types.h:685
@ dnnl_defcab
permuted 6D tensor
Definition: dnnl_types.h:212
@ dnnl_abcdef
plain 6D tensor
Definition: dnnl_types.h:183
@ dnnl_nChw8c
4D CNN activations tensor blocked by channels with block size 8, an alias to dnnl_aBcd8b
Definition: dnnl_types.h:700
@ dnnl_a
plain 1D tensor
Definition: dnnl_types.h:177
@ dnnl_nChw4c
4D CNN activations tensor blocked by channels with block size 4, an alias to dnnl_aBcd4b
Definition: dnnl_types.h:697
@ dnnl_acbdef
permuted 6D tensor
Definition: dnnl_types.h:197
@ dnnl_acdb
permuted 4D tensor
Definition: dnnl_types.h:198
@ dnnl_aBcd8b
4D tensor blocked by 2nd dimension with block size 8
Definition: dnnl_types.h:282
@ dnnl_aBc8b
3D tensor blocked by 2nd dimension with block size 8
Definition: dnnl_types.h:245
@ dnnl_nCw4c
3D CNN activations tensor blocked by channels with block size 4, an alias to dnnl_aBc4b
Definition: dnnl_types.h:709
@ dnnl_abcdefg
plain 7D tensor
Definition: dnnl_types.h:184
@ dnnl_aBcde8b
5D tensor blocked by 2nd dimension with block size 8
Definition: dnnl_types.h:330
@ dnnl_nChw16c
4D CNN activations tensor blocked by channels with block size 16, an alias to dnnl_aBcd16b
Definition: dnnl_types.h:694
@ dnnl_abdfce
permuted 6D tensor
Definition: dnnl_types.h:424
@ dnnl_abdec
permuted 5D tensor
Definition: dnnl_types.h:194
@ dnnl_bacd
permuted 4D tensor
Definition: dnnl_types.h:202
@ dnnl_nCdhw8c
5D CNN activations tensor blocked by channels with block size 8, an alias to dnnl_aBcde8b
Definition: dnnl_types.h:688
@ dnnl_aBcde32b
5D tensor blocked by 2nd dimension with block size 32
Definition: dnnl_types.h:313
@ dnnl_abced
permuted 5D tensor
Definition: dnnl_types.h:213
@ dnnl_bcda
permuted 4D tensor
Definition: dnnl_types.h:205
@ dnnl_acbde
permuted 5D tensor
Definition: dnnl_types.h:196
@ dnnl_aBCd2b4c2b
4D tensor blocked by 3rd dimension with block size 4
Definition: dnnl_types.h:300
@ dnnl_abcdefgih
permuted 9D tensor
Definition: dnnl_types.h:217
@ dnnl_bcdea
permuted 5D tensor
Definition: dnnl_types.h:206
@ dnnl_abdefc
permuted 6D tensor
Definition: dnnl_types.h:425
@ dnnl_aBcde16b
5D tensor blocked by 2nd dimension with block size 16
Definition: dnnl_types.h:306
@ dnnl_nCw8c
3D CNN activations tensor blocked by channels with block size 8, an alias to dnnl_aBc8b
Definition: dnnl_types.h:712
@ dnnl_abdc
permuted 4D tensor
Definition: dnnl_types.h:193
@ dnnl_ABcde4b16a4b
5D tensor blocked by 1st dimension with block size 16
Definition: dnnl_types.h:302
@ dnnl_aBcd32b
4D tensor blocked by 2nd dimension with block size 32
Definition: dnnl_types.h:261
@ dnnl_abcdefghijlk
permuted 12D tensor
Definition: dnnl_types.h:220
@ dnnl_format_tag_last
Just a sentinel, not real memory format tag.
Definition: dnnl_types.h:543
@ dnnl_abc
plain 3D tensor
Definition: dnnl_types.h:179
@ dnnl_bac
permuted 3D tensor
Definition: dnnl_types.h:201
@ dnnl_dcab
permuted 4D tensor
Definition: dnnl_types.h:209
@ dnnl_cdeba
permuted 5D tensor
Definition: dnnl_types.h:210
@ dnnl_acb
permuted 3D tensor
Definition: dnnl_types.h:195
@ dnnl_aBc32b
3D tensor blocked by 2nd dimension with block size 32
Definition: dnnl_types.h:233
@ dnnl_abcdefghji
permuted 10D tensor
Definition: dnnl_types.h:218
@ dnnl_nCw16c
3D CNN activations tensor blocked by channels with block size 16, an alias to dnnl_aBc16b
Definition: dnnl_types.h:706
@ dnnl_aBCdef2c8b4c
6D tensor blocked by 2nd dimension with block size 8
Definition: dnnl_types.h:359
@ dnnl_abcdefghij
plain 10D tensor
Definition: dnnl_types.h:187
@ dnnl_format_tag_any
Undefined memory format tag.
Definition: dnnl_types.h:169
@ dnnl_blocked
A tensor in a generic format described by the stride and blocking values in each dimension.
Definition: dnnl_types.h:89
@ dnnl_format_kind_wino
Weights format used in 8bit Winograd convolution.
Definition: dnnl_types.h:91
@ dnnl_format_kind_any
Unspecified format kind.
Definition: dnnl_types.h:85
@ dnnl_format_kind_undef
Undefined memory format kind, used for empty memory descriptors.
Definition: dnnl_types.h:82
@ dnnl_format_kind_rnn_packed
Packed weights format used in RNN.
Definition: dnnl_types.h:93
dnnl_status_t DNNL_API dnnl_pooling_v2_backward_desc_init(dnnl_pooling_v2_desc_t *pool_desc, dnnl_alg_kind_t alg_kind, const dnnl_memory_desc_t *diff_src_desc, const dnnl_memory_desc_t *diff_dst_desc, const dnnl_dims_t strides, const dnnl_dims_t kernel, const dnnl_dims_t dilation, const dnnl_dims_t padding_l, const dnnl_dims_t padding_r)
Initializes a descriptor for pooling v2 (pooling with dilation support) backward propagation primitiv...
dnnl_status_t DNNL_API dnnl_pooling_v2_forward_desc_init(dnnl_pooling_v2_desc_t *pool_desc, dnnl_prop_kind_t prop_kind, dnnl_alg_kind_t alg_kind, const dnnl_memory_desc_t *src_desc, const dnnl_memory_desc_t *dst_desc, const dnnl_dims_t strides, const dnnl_dims_t kernel, const dnnl_dims_t dilation, const dnnl_dims_t padding_l, const dnnl_dims_t padding_r)
Initializes a descriptor for pooling v2 (pooling with dilation support) forward propagation primitive...
dnnl_status_t DNNL_API dnnl_pooling_forward_desc_init(dnnl_pooling_desc_t *pool_desc, dnnl_prop_kind_t prop_kind, dnnl_alg_kind_t alg_kind, const dnnl_memory_desc_t *src_desc, const dnnl_memory_desc_t *dst_desc, const dnnl_dims_t strides, const dnnl_dims_t kernel, const dnnl_dims_t padding_l, const dnnl_dims_t padding_r)
Initializes a descriptor for pooling forward propagation primitive.
dnnl_status_t DNNL_API dnnl_pooling_backward_desc_init(dnnl_pooling_desc_t *pool_desc, dnnl_alg_kind_t alg_kind, const dnnl_memory_desc_t *diff_src_desc, const dnnl_memory_desc_t *diff_dst_desc, const dnnl_dims_t strides, const dnnl_dims_t kernel, const dnnl_dims_t padding_l, const dnnl_dims_t padding_r)
Initializes a descriptor for pooling backward propagation primitive.
dnnl_status_t DNNL_API dnnl_prelu_forward_desc_init(dnnl_prelu_desc_t *prelu_desc, dnnl_prop_kind_t prop_kind, const dnnl_memory_desc_t *data_desc, const dnnl_memory_desc_t *weights_desc)
Initializes a descriptor for PReLU (leaky ReLU with trainable alpha parameter) forward propagation pr...
dnnl_status_t DNNL_API dnnl_prelu_backward_desc_init(dnnl_prelu_desc_t *prelu_desc, const dnnl_memory_desc_t *data_desc, const dnnl_memory_desc_t *weights_desc, const dnnl_memory_desc_t *diff_data_desc, const dnnl_memory_desc_t *diff_weights_desc)
Initializes a descriptor for PReLU (leaky ReLU with trainable alpha parameter) backward propagation p...
void set_primitive_cache_capacity(int capacity)
Sets a number of primitives that can be held in the primitive cache at a time.
Definition: dnnl.hpp:10905
dnnl_status_t DNNL_API dnnl_set_primitive_cache_capacity(int capacity)
Sets a number of primitives that can be held in the primitive cache at a time.
dnnl_status_t DNNL_API dnnl_get_primitive_cache_capacity(int *capacity)
Returns the number of primitives that can be held in the primitive cache at the same time.
int get_primitive_cache_capacity()
Returns the number of primitives that can be held in the primitive cache at the same time.
Definition: dnnl.hpp:10897
dnnl_status_t DNNL_API dnnl_primitive_desc_query(const_dnnl_primitive_desc_t primitive_desc, dnnl_query_t what, int index, void *result)
Queries a primitive descriptor for various pieces of information.
#define DNNL_ARG_DST_ITER
A special mnemonic for RNN input recurrent hidden state vector.
Definition: dnnl_types.h:2318
dnnl_status_t DNNL_API dnnl_primitive_desc_iterator_destroy(dnnl_primitive_desc_iterator_t iterator)
Destroys a primitive descriptor iterator.
#define DNNL_ARG_WEIGHTS_LAYER
A special mnemonic for RNN weights applied to the layer input.
Definition: dnnl_types.h:2336
#define DNNL_ARG_DIFF_BIAS
Gradient (diff) of the bias tensor argument.
Definition: dnnl_types.h:2443
#define DNNL_ARG_DIFF_SRC_ITER_C
A special mnemonic for gradient (diff) of RNN input recurrent cell state vector.
Definition: dnnl_types.h:2389
#define DNNL_ARG_DIFF_SRC_LAYER
A special mnemonic for gradient (diff) of RNN input vector.
Definition: dnnl_types.h:2377
#define DNNL_ARG_DIFF_WEIGHTS_PEEPHOLE
A special mnemonic for diff of RNN weights applied to the peephole weights.
Definition: dnnl_types.h:2434
#define DNNL_ARG_WEIGHTS_PROJECTION
A special mnemonic for RNN weights applied to the projection weights.
Definition: dnnl_types.h:2354
dnnl_normalization_flags_t
Flags for normalization primitives.
Definition: dnnl_types.h:1241
#define DNNL_ARG_DIFF_WEIGHTS_PROJECTION
A special mnemonic for diff of RNN weights applied to the projection weights.
Definition: dnnl_types.h:2440
dnnl_status_t DNNL_API dnnl_primitive_desc_get_attr(const_dnnl_primitive_desc_t primitive_desc, const_dnnl_primitive_attr_t *attr)
Returns a constant reference to the attributes of a primitive descriptor.
#define DNNL_ARG_DIFF_WEIGHTS_ITER
A special mnemonic for diff of RNN weights applied to the recurrent input.
Definition: dnnl_types.h:2428
#define DNNL_ARG_DIFF_SRC_ITER
A special mnemonic for gradient (diff) of RNN input recurrent hidden state vector.
Definition: dnnl_types.h:2383
#define DNNL_ARG_DIFF_DST_ITER_C
A special mnemonic for gradient (diff) of RNN input recurrent cell state vector.
Definition: dnnl_types.h:2410
dnnl_status_t DNNL_API dnnl_primitive_execute(const_dnnl_primitive_t primitive, dnnl_stream_t stream, int nargs, const dnnl_exec_arg_t *args)
Executes a primitive.
#define DNNL_ARG_WEIGHTS_ITER
A special mnemonic for RNN weights applied to the recurrent input.
Definition: dnnl_types.h:2342
dnnl_status_t DNNL_API dnnl_primitive_desc_iterator_next(dnnl_primitive_desc_iterator_t iterator)
Advances the primitive descriptor iterator to point to the next available implementation.
dnnl_status_t DNNL_API dnnl_primitive_desc_destroy(dnnl_primitive_desc_t primitive_desc)
Destroys a primitive descriptor.
const void * const_dnnl_op_desc_t
A pointer to any of the operation descriptors (constant variant).
Definition: dnnl_types.h:1522
const_dnnl_primitive_desc_t get_primitive_desc() const
Returns the C API primitive descriptor of the underlying C API primitive.
Definition: dnnl.hpp:368
dnnl_status_t DNNL_API dnnl_primitive_get_primitive_desc(const_dnnl_primitive_t primitive, const_dnnl_primitive_desc_t *primitive_desc)
Retrieves a constant reference to the primitive descriptor of a given primitive.
#define DNNL_ARG_DST_ITER_C
A special mnemonic for LSTM output recurrent cell state vector.
Definition: dnnl_types.h:2324
#define DNNL_ARG_SRC_ITER_C
A special mnemonic for RNN input recurrent cell state vector.
Definition: dnnl_types.h:2301
query
Primitive descriptor query specification.
Definition: dnnl.hpp:745
#define DNNL_ARG_FROM
A special mnemonic for reorder source argument.
Definition: dnnl_types.h:2289
dnnl_alg_kind_t
Kinds of algorithms.
Definition: dnnl_types.h:1107
dnnl_primitive_kind_t
Kinds of primitives.
Definition: dnnl_types.h:1053
dnnl_query_t
Primitive descriptor query specification.
Definition: dnnl_types.h:2514
const dnnl_memory_desc_t DNNL_API * dnnl_primitive_desc_query_md(const_dnnl_primitive_desc_t primitive_desc, dnnl_query_t what, int index)
Queries primitive descriptor for a memory descriptor.
dnnl_primitive_kind_t convert_to_c(primitive::kind akind)
Converts primitive kind enum value from C++ API to C API type.
Definition: dnnl.hpp:364
struct dnnl_primitive_desc * dnnl_primitive_desc_t
A primitive descriptor handle.
Definition: dnnl_types.h:2190
#define DNNL_ARG_WEIGHTS_PEEPHOLE
A special mnemonic for RNN weights applied to the peephole weights.
Definition: dnnl_types.h:2348
kind get_kind() const
Returns the kind of the primitive.
Definition: dnnl.hpp:375
#define DNNL_ARG_SRC_LAYER
A special mnemonic for RNN input vector.
Definition: dnnl_types.h:2286
dnnl_status_t DNNL_API dnnl_primitive_destroy(dnnl_primitive_t primitive)
Destroys a primitive.
#define DNNL_ARG_DIFF_WEIGHTS_LAYER
A special mnemonic for diff of RNN weights applied to the layer input.
Definition: dnnl_types.h:2422
dnnl_status_t DNNL_API dnnl_primitive_desc_iterator_create(dnnl_primitive_desc_iterator_t *iterator, const_dnnl_op_desc_t op_desc, const_dnnl_primitive_attr_t attr, dnnl_engine_t engine, const_dnnl_primitive_desc_t hint_forward_primitive_desc)
Creates a primitive descriptor iterator.
#define DNNL_ARG_DST_LAYER
A special mnemonic for RNN output vector. An alias for DNNL_ARG_DST_0.
Definition: dnnl_types.h:2312
dnnl_status_t DNNL_API dnnl_primitive_create(dnnl_primitive_t *primitive, const_dnnl_primitive_desc_t primitive_desc)
Creates a primitive.
#define DNNL_ARG_BIAS
Bias tensor argument.
Definition: dnnl_types.h:2357
normalization_flags
Flags for normalization primitives.
Definition: dnnl.hpp:615
#define DNNL_ARG_DIFF_DST_ITER
A special mnemonic for gradient (diff) of RNN input recurrent hidden state vector.
Definition: dnnl_types.h:2404
dnnl_prop_kind_t
Kinds of propagation.
Definition: dnnl_types.h:1026
dnnl_status_t DNNL_API dnnl_primitive_desc_clone(dnnl_primitive_desc_t *primitive_desc, const_dnnl_primitive_desc_t existing_primitive_desc)
Clones a primitive descriptor.
#define DNNL_ARG_SRC_ITER
A special mnemonic for RNN input recurrent hidden state vector.
Definition: dnnl_types.h:2295
dnnl_primitive_desc_t DNNL_API dnnl_primitive_desc_iterator_fetch(const_dnnl_primitive_desc_iterator_t iterator)
Fetches the current primitive descriptor from a primitive descriptor iterator.
#define DNNL_ARG_TO
A special mnemonic for reorder destination argument.
Definition: dnnl_types.h:2310
#define DNNL_ARG_DIFF_DST_LAYER
A special mnemonic for gradient (diff) of RNN output vector.
Definition: dnnl_types.h:2398
@ dnnl_fuse_norm_relu
Fuse with ReLU.
Definition: dnnl_types.h:1289
@ dnnl_normalization_flags_none
Use no normalization flags.
Definition: dnnl_types.h:1250
@ dnnl_use_scaleshift
Use scale and shift parameters.
Definition: dnnl_types.h:1276
@ dnnl_use_global_stats
Use global statistics.
Definition: dnnl_types.h:1263
@ batch_normalization_d
batch normalization descriptor
@ weights_md
weights memory descriptor desc
@ memory_consumption_s64
memory required for scratchpad (bytes)
@ shuffle_d
shuffle descriptor
@ deconvolution_d
deconvolution descriptor
@ impl_info_str
implementation name
@ diff_weights_md
weights gradient (diff) memory desc
@ workspace_md
workspace memory desc
@ reduction_d
reduction descriptor
@ eltwise_d
eltwise descriptor
@ matmul_d
matmul descriptor
@ rnn_d
rnn descriptor
@ softmax_d
softmax descriptor
@ num_of_outputs_s32
number of outputs expected
@ primitive_kind
primitive kind
@ dst_md
destination memory desc
@ scratchpad_engine
scratchpad engine
@ reorder_src_engine
reorder source engine
@ op_d
operation descriptor
@ layer_normalization_d
layer normalization descriptor
@ logsoftmax_d
logsoftmax descriptor
@ pooling_d
pooling descriptor
@ num_of_inputs_s32
number of inputs expected
@ diff_src_md
source gradient (diff) memory desc
@ src_md
source memory desc
@ scratchpad_md
scratchpad memory desc
@ reorder_dst_engine
reorder destination engine
@ engine
execution engine
@ convolution_d
convolution descriptor
@ time_estimate_f64
runtime estimation (seconds), unimplemented
@ binary_d
binary descriptor
@ diff_dst_md
destination gradient (diff) memory desc
@ exec_arg_md
memory desc of an execute argument
@ inner_product_d
inner product descriptor
@ lrn_d
lrn descriptor
@ resampling_d
resampling descriptor
@ dnnl_pooling_avg_exclude_padding
Average pooling exclude padding.
Definition: dnnl_types.h:1183
@ dnnl_eltwise_clip
Eltwise: clip.
Definition: dnnl_types.h:1153
@ dnnl_eltwise_tanh_use_dst_for_bwd
Eltwise: hyperbolic tangent non-linearity (tanh) (dst for backward)
Definition: dnnl_types.h:1167
@ dnnl_eltwise_logsigmoid
Eltwise: logsigmoid.
Definition: dnnl_types.h:1163
@ dnnl_pooling_avg
Average pooling (alias for dnnl_pooling_avg_exclude_padding)
Definition: dnnl_types.h:1185
@ dnnl_eltwise_gelu_tanh
Eltwise: gelu.
Definition: dnnl_types.h:1145
@ dnnl_resampling_linear
Linear Resampling Method.
Definition: dnnl_types.h:1219
@ dnnl_eltwise_sqrt
Eltwise: square root.
Definition: dnnl_types.h:1130
@ dnnl_binary_min
Binary min.
Definition: dnnl_types.h:1211
@ dnnl_reduction_norm_lp_sum
Reduction using lp norm.
Definition: dnnl_types.h:1233
@ dnnl_eltwise_abs
Eltwise: abs.
Definition: dnnl_types.h:1128
@ dnnl_reduction_norm_lp_power_p_max
Reduction using lp norm without final pth-root.
Definition: dnnl_types.h:1235
@ dnnl_reduction_min
Reduction using min.
Definition: dnnl_types.h:1223
@ dnnl_eltwise_sqrt_use_dst_for_bwd
Eltwise: square root (dst for backward)
Definition: dnnl_types.h:1171
@ dnnl_eltwise_exp
Eltwise: exponent.
Definition: dnnl_types.h:1140
@ dnnl_eltwise_square
Eltwise: square.
Definition: dnnl_types.h:1126
@ dnnl_eltwise_gelu
Eltwise: tanh-based gelu (alias for dnnl_eltwise_gelu_tanh)
Definition: dnnl_types.h:1147
@ dnnl_convolution_winograd
Winograd convolution.
Definition: dnnl_types.h:1112
@ dnnl_eltwise_clip_v2_use_dst_for_bwd
Eltwise: clip version 2 (dst for backward)
Definition: dnnl_types.h:1177
@ dnnl_lrn_across_channels
Local response normalization (LRN) across multiple channels.
Definition: dnnl_types.h:1187
@ dnnl_binary_sub
Binary sub.
Definition: dnnl_types.h:1215
@ dnnl_deconvolution_direct
Direct deconvolution.
Definition: dnnl_types.h:1116
@ dnnl_eltwise_relu
Eltwise: ReLU.
Definition: dnnl_types.h:1120
@ dnnl_convolution_auto
Convolution algorithm(either direct or Winograd) is chosen just in time.
Definition: dnnl_types.h:1114
@ dnnl_eltwise_swish
Eltwise: swish.
Definition: dnnl_types.h:1149
@ dnnl_vanilla_rnn
RNN cell.
Definition: dnnl_types.h:1191
@ dnnl_eltwise_gelu_erf
Eltwise: erf-based gelu.
Definition: dnnl_types.h:1159
@ dnnl_vanilla_lstm
LSTM cell.
Definition: dnnl_types.h:1193
@ dnnl_eltwise_elu
Eltwise: exponential linear unit (elu)
Definition: dnnl_types.h:1124
@ dnnl_vanilla_gru
GRU cell.
Definition: dnnl_types.h:1195
@ dnnl_lbr_gru
GRU cell with linear before reset.
Definition: dnnl_types.h:1203
@ dnnl_eltwise_tanh
Eltwise: hyperbolic tangent non-linearity (tanh)
Definition: dnnl_types.h:1122
@ dnnl_convolution_direct
Direct convolution.
Definition: dnnl_types.h:1110
@ dnnl_eltwise_soft_relu
Eltwise: soft_relu.
Definition: dnnl_types.h:1136
@ dnnl_eltwise_log
Eltwise: natural logarithm.
Definition: dnnl_types.h:1151
@ dnnl_eltwise_clip_v2
Eltwise: clip version 2.
Definition: dnnl_types.h:1155
@ dnnl_lrn_within_channel
LRN within a single channel.
Definition: dnnl_types.h:1189
@ dnnl_eltwise_elu_use_dst_for_bwd
Eltwise: exponential linear unit (elu) (dst for backward)
Definition: dnnl_types.h:1169
@ dnnl_deconvolution_winograd
Winograd deconvolution.
Definition: dnnl_types.h:1118
@ dnnl_reduction_mul
Reduction using mul.
Definition: dnnl_types.h:1227
@ dnnl_eltwise_pow
Eltwise: pow.
Definition: dnnl_types.h:1157
@ dnnl_eltwise_relu_use_dst_for_bwd
Eltwise: ReLU (dst for backward)
Definition: dnnl_types.h:1165
@ dnnl_reduction_max
Reduction using max.
Definition: dnnl_types.h:1221
@ dnnl_eltwise_logistic
Eltwise: logistic.
Definition: dnnl_types.h:1138
@ dnnl_pooling_avg_include_padding
Average pooling include padding.
Definition: dnnl_types.h:1181
@ dnnl_reduction_mean
Reduction using mean.
Definition: dnnl_types.h:1229
@ dnnl_pooling_max
Max pooling.
Definition: dnnl_types.h:1179
@ dnnl_eltwise_logistic_use_dst_for_bwd
Eltwise: logistic (dst for backward)
Definition: dnnl_types.h:1173
@ dnnl_binary_add
Binary add.
Definition: dnnl_types.h:1205
@ dnnl_binary_div
Binary div.
Definition: dnnl_types.h:1213
@ dnnl_reduction_norm_lp_max
Reduction using lp norm.
Definition: dnnl_types.h:1231
@ dnnl_reduction_norm_lp_power_p_sum
Reduction using lp norm without final pth-root.
Definition: dnnl_types.h:1237
@ dnnl_eltwise_round
Eltwise: round.
Definition: dnnl_types.h:1161
@ dnnl_binary_mul
Binary mul.
Definition: dnnl_types.h:1207
@ dnnl_reduction_sum
Reduction using sum.
Definition: dnnl_types.h:1225
@ dnnl_eltwise_exp_use_dst_for_bwd
Eltwise: exp (dst for backward)
Definition: dnnl_types.h:1175
@ dnnl_eltwise_bounded_relu
Eltwise: bounded_relu.
Definition: dnnl_types.h:1134
@ dnnl_eltwise_linear
Eltwise: linear.
Definition: dnnl_types.h:1132
@ dnnl_resampling_nearest
Nearest Neighbor Resampling Method.
Definition: dnnl_types.h:1217
@ dnnl_binary_max
Binary max.
Definition: dnnl_types.h:1209
@ dnnl_binary
A binary primitive.
Definition: dnnl_types.h:1087
@ dnnl_concat
A (out-of-place) concat primitive.
Definition: dnnl_types.h:1061
@ dnnl_reorder
A reorder primitive.
Definition: dnnl_types.h:1057
@ dnnl_convolution
A convolution primitive.
Definition: dnnl_types.h:1065
@ dnnl_inner_product
An inner product primitive.
Definition: dnnl_types.h:1081
@ dnnl_resampling
A resampling primitive.
Definition: dnnl_types.h:1093
@ dnnl_batch_normalization
A batch normalization primitive.
Definition: dnnl_types.h:1077
@ dnnl_undefined_primitive
Undefined primitive.
Definition: dnnl_types.h:1055
@ dnnl_sum
A sum primitive.
Definition: dnnl_types.h:1063
@ dnnl_pooling_v2
A pooling version 2 primitive (pooling with dilation support).
Definition: dnnl_types.h:1095
@ dnnl_layer_normalization
A layer normalization primitive.
Definition: dnnl_types.h:1079
@ dnnl_prelu
A PReLU primitive.
Definition: dnnl_types.h:1099
@ dnnl_eltwise
An element-wise primitive.
Definition: dnnl_types.h:1069
@ dnnl_matmul
A matrix multiplication primitive.
Definition: dnnl_types.h:1091
@ dnnl_shuffle
A shuffle primitive.
Definition: dnnl_types.h:1059
@ dnnl_logsoftmax
A logsoftmax primitive.
Definition: dnnl_types.h:1089
@ dnnl_pooling
A pooling primitive.
Definition: dnnl_types.h:1073
@ dnnl_deconvolution
A deconvolution primitive.
Definition: dnnl_types.h:1067
@ dnnl_softmax
A softmax primitive.
Definition: dnnl_types.h:1071
@ dnnl_rnn
A rnn primitive.
Definition: dnnl_types.h:1083
@ dnnl_reduction
A reduction primitive.
Definition: dnnl_types.h:1097
@ dnnl_lrn
An LRN primitive.
Definition: dnnl_types.h:1075
@ dnnl_query_resampling_d
resampling descriptor
Definition: dnnl_types.h:2557
@ dnnl_query_num_of_outputs_s32
number of outputs expected
Definition: dnnl_types.h:2521
@ dnnl_query_convolution_d
convolution descriptor
Definition: dnnl_types.h:2542
@ dnnl_query_weights_md
weights memory descriptor desc
Definition: dnnl_types.h:2566
@ dnnl_query_src_md
source memory desc
Definition: dnnl_types.h:2564
@ dnnl_query_softmax_d
softmax descriptor
Definition: dnnl_types.h:2546
@ dnnl_query_binary_d
binary descriptor
Definition: dnnl_types.h:2554
@ dnnl_query_workspace_md
workspace memory desc
Definition: dnnl_types.h:2570
@ dnnl_query_matmul_d
matrix multiplication (matmul) descriptor
Definition: dnnl_types.h:2556
@ dnnl_query_num_of_inputs_s32
number of inputs expected
Definition: dnnl_types.h:2520
@ dnnl_query_op_d
op descriptor
Definition: dnnl_types.h:2541
@ dnnl_query_diff_src_md
source gradient memory desc
Definition: dnnl_types.h:2565
@ dnnl_query_scratchpad_md
scratchpad memory desc
Definition: dnnl_types.h:2571
@ dnnl_query_shuffle_d
shuffle descriptor
Definition: dnnl_types.h:2544
@ dnnl_query_memory_consumption_s64
memory consumption – extra
Definition: dnnl_types.h:2524
@ dnnl_query_inner_product_d
inner product descriptor
Definition: dnnl_types.h:2551
@ dnnl_query_deconvolution_d
deconvolution descriptor
Definition: dnnl_types.h:2543
@ dnnl_query_primitive_kind
primitive kind
Definition: dnnl_types.h:2518
@ dnnl_query_batch_normalization_d
batch normalization descriptor
Definition: dnnl_types.h:2549
@ dnnl_query_impl_info_str
for creating scratchpad memory
Definition: dnnl_types.h:2532
@ dnnl_query_time_estimate_f64
runtime estimation (seconds)
Definition: dnnl_types.h:2523
@ dnnl_query_eltwise_d
eltwise descriptor
Definition: dnnl_types.h:2545
@ dnnl_query_diff_weights_md
weights grad. memory desc
Definition: dnnl_types.h:2567
@ dnnl_query_reduction_d
reduction descriptor
Definition: dnnl_types.h:2559
@ dnnl_query_reorder_dst_engine
destination engine
Definition: dnnl_types.h:2535
@ dnnl_query_reorder_src_engine
source engine
Definition: dnnl_types.h:2534
@ dnnl_query_scratchpad_engine
(scratch) memory, additional to all inputs and outputs memory (bytes)
Definition: dnnl_types.h:2529
@ dnnl_query_undef
no query
Definition: dnnl_types.h:2515
@ dnnl_query_prop_kind
propagation kind
Definition: dnnl_types.h:2537
@ dnnl_query_pooling_d
pooling descriptor
Definition: dnnl_types.h:2547
@ dnnl_query_exec_arg_md
memory desc of an execute argument
Definition: dnnl_types.h:2572
@ dnnl_query_engine
execution engine
Definition: dnnl_types.h:2517
@ dnnl_query_rnn_d
rnn descriptor
Definition: dnnl_types.h:2552
@ dnnl_query_layer_normalization_d
layer normalization descriptor
Definition: dnnl_types.h:2550
@ dnnl_query_lrn_d
lrn descriptor
Definition: dnnl_types.h:2548
@ dnnl_query_dst_md
destination memory desc
Definition: dnnl_types.h:2568
@ dnnl_query_diff_dst_md
destination grad. memory desc
Definition: dnnl_types.h:2569
@ dnnl_query_logsoftmax_d
logsoftmax descriptor
Definition: dnnl_types.h:2555
@ use_scale_shift
Use scale and shift parameters.
@ none
Use no normalization flags.
@ fuse_norm_relu
Fuse normalization with ReLU.
@ use_global_stats
Use global statistics.
@ dnnl_backward_weights
Backward weights propagation.
Definition: dnnl_types.h:1046
@ dnnl_forward_inference
Forward data propagation (inference mode).
Definition: dnnl_types.h:1036
@ dnnl_backward
Backward propagation (with respect to all parameters).
Definition: dnnl_types.h:1042
@ dnnl_backward_data
Backward data propagation.
Definition: dnnl_types.h:1044
@ dnnl_prop_kind_undef
Undefined propagation type.
Definition: dnnl_types.h:1029
@ dnnl_forward
Forward data propagation (alias for dnnl_forward_training).
Definition: dnnl_types.h:1040
@ dnnl_forward_training
Forward data propagation (training mode).
Definition: dnnl_types.h:1032
@ dnnl_backward_bias
Backward bias propagation.
Definition: dnnl_types.h:1048
@ dnnl_forward_scoring
Forward data propagation (alias for dnnl_forward_inference).
Definition: dnnl_types.h:1038
dnnl_status_t DNNL_API dnnl_reduction_desc_init(dnnl_reduction_desc_t *desc, dnnl_alg_kind_t alg_kind, const dnnl_memory_desc_t *src_desc, const dnnl_memory_desc_t *dst_desc, float p, float eps)
Initializes a descriptor for a reduction primitive.
dnnl_status_t DNNL_API dnnl_reorder_primitive_desc_create(dnnl_primitive_desc_t *reorder_primitive_desc, const dnnl_memory_desc_t *src_desc, dnnl_engine_t src_engine, const dnnl_memory_desc_t *dst_desc, dnnl_engine_t dst_engine, const_dnnl_primitive_attr_t attr)
Creates a primitive descriptor for a reorder primitive.
dnnl_status_t DNNL_API dnnl_resampling_backward_desc_init(dnnl_resampling_desc_t *resampling_desc, dnnl_alg_kind_t alg_kind, const float *factors, const dnnl_memory_desc_t *diff_src_desc, const dnnl_memory_desc_t *diff_dst_desc)
Initializes a descriptor for resampling backward propagation primitive.
dnnl_status_t DNNL_API dnnl_resampling_forward_desc_init(dnnl_resampling_desc_t *resampling_desc, dnnl_prop_kind_t prop_kind, dnnl_alg_kind_t alg_kind, const float *factors, const dnnl_memory_desc_t *src_desc, const dnnl_memory_desc_t *dst_desc)
Initializes a descriptor for a resampling forward propagation primitive.
dnnl_status_t DNNL_API dnnl_lbr_gru_backward_desc_init(dnnl_rnn_desc_t *rnn_desc, dnnl_prop_kind_t prop_kind, dnnl_rnn_direction_t direction, const dnnl_memory_desc_t *src_layer_desc, const dnnl_memory_desc_t *src_iter_desc, const dnnl_memory_desc_t *weights_layer_desc, const dnnl_memory_desc_t *weights_iter_desc, const dnnl_memory_desc_t *bias_desc, const dnnl_memory_desc_t *dst_layer_desc, const dnnl_memory_desc_t *dst_iter_desc, const dnnl_memory_desc_t *diff_src_layer_desc, const dnnl_memory_desc_t *diff_src_iter_desc, const dnnl_memory_desc_t *diff_weights_layer_desc, const dnnl_memory_desc_t *diff_weights_iter_desc, const dnnl_memory_desc_t *diff_bias_desc, const dnnl_memory_desc_t *diff_dst_layer_desc, const dnnl_memory_desc_t *diff_dst_iter_desc, unsigned flags)
Initializes a descriptor for LBR GRU backward propagation primitive.
dnnl_status_t DNNL_API dnnl_gru_forward_desc_init(dnnl_rnn_desc_t *rnn_desc, dnnl_prop_kind_t prop_kind, dnnl_rnn_direction_t direction, const dnnl_memory_desc_t *src_layer_desc, const dnnl_memory_desc_t *src_iter_desc, const dnnl_memory_desc_t *weights_layer_desc, const dnnl_memory_desc_t *weights_iter_desc, const dnnl_memory_desc_t *bias_desc, const dnnl_memory_desc_t *dst_layer_desc, const dnnl_memory_desc_t *dst_iter_desc, unsigned flags)
Initializes a descriptor for GRU forward propagation primitive.
rnn_direction
A direction of RNN primitive execution.
Definition: dnnl.hpp:712
dnnl_rnn_flags_t
Flags for RNN cell.
Definition: dnnl_types.h:1931
dnnl_status_t DNNL_API dnnl_vanilla_rnn_forward_desc_init(dnnl_rnn_desc_t *rnn_desc, dnnl_prop_kind_t prop_kind, const dnnl_alg_kind_t activation, const dnnl_rnn_direction_t direction, const dnnl_memory_desc_t *src_layer_desc, const dnnl_memory_desc_t *src_iter_desc, const dnnl_memory_desc_t *weights_layer_desc, const dnnl_memory_desc_t *weights_iter_desc, const dnnl_memory_desc_t *bias_desc, const dnnl_memory_desc_t *dst_layer_desc, const dnnl_memory_desc_t *dst_iter_desc, unsigned flags, float alpha, float beta)
Initializes a descriptor for vanilla RNN forward propagation primitive.
dnnl_status_t DNNL_API dnnl_lstm_backward_desc_init(dnnl_rnn_desc_t *rnn_desc, dnnl_prop_kind_t prop_kind, dnnl_rnn_direction_t direction, const dnnl_memory_desc_t *src_layer_desc, const dnnl_memory_desc_t *src_iter_desc, const dnnl_memory_desc_t *src_iter_c_desc, const dnnl_memory_desc_t *weights_layer_desc, const dnnl_memory_desc_t *weights_iter_desc, const dnnl_memory_desc_t *bias_desc, const dnnl_memory_desc_t *dst_layer_desc, const dnnl_memory_desc_t *dst_iter_desc, const dnnl_memory_desc_t *dst_iter_c_desc, const dnnl_memory_desc_t *diff_src_layer_desc, const dnnl_memory_desc_t *diff_src_iter_desc, const dnnl_memory_desc_t *diff_src_iter_c_desc, const dnnl_memory_desc_t *diff_weights_layer_desc, const dnnl_memory_desc_t *diff_weights_iter_desc, const dnnl_memory_desc_t *diff_bias_desc, const dnnl_memory_desc_t *diff_dst_layer_desc, const dnnl_memory_desc_t *diff_dst_iter_desc, const dnnl_memory_desc_t *diff_dst_iter_c_desc, unsigned flags)
Initializes a descriptor for an LSTM backward propagation primitive.
dnnl_rnn_direction_t
A direction of RNN primitive execution.
Definition: dnnl_types.h:1937
dnnl_status_t DNNL_API dnnl_vanilla_rnn_backward_desc_init(dnnl_rnn_desc_t *rnn_desc, dnnl_prop_kind_t prop_kind, const dnnl_alg_kind_t activation, const dnnl_rnn_direction_t direction, const dnnl_memory_desc_t *src_layer_desc, const dnnl_memory_desc_t *src_iter_desc, const dnnl_memory_desc_t *weights_layer_desc, const dnnl_memory_desc_t *weights_iter_desc, const dnnl_memory_desc_t *bias_desc, const dnnl_memory_desc_t *dst_layer_desc, const dnnl_memory_desc_t *dst_iter_desc, const dnnl_memory_desc_t *diff_src_layer_desc, const dnnl_memory_desc_t *diff_src_iter_desc, const dnnl_memory_desc_t *diff_weights_layer_desc, const dnnl_memory_desc_t *diff_weights_iter_desc, const dnnl_memory_desc_t *diff_bias_desc, const dnnl_memory_desc_t *diff_dst_layer_desc, const dnnl_memory_desc_t *diff_dst_iter_desc, unsigned flags, float alpha, float beta)
Initializes a descriptor for vanilla RNN backward propagation primitive.
dnnl_status_t DNNL_API dnnl_lstm_backward_desc_init_v3(dnnl_rnn_desc_t *rnn_desc, dnnl_prop_kind_t prop_kind, dnnl_rnn_direction_t direction, const dnnl_memory_desc_t *src_layer_desc, const dnnl_memory_desc_t *src_iter_desc, const dnnl_memory_desc_t *src_iter_c_desc, const dnnl_memory_desc_t *weights_layer_desc, const dnnl_memory_desc_t *weights_iter_desc, const dnnl_memory_desc_t *weights_peephole_desc, const dnnl_memory_desc_t *weights_projection_desc, const dnnl_memory_desc_t *bias_desc, const dnnl_memory_desc_t *dst_layer_desc, const dnnl_memory_desc_t *dst_iter_desc, const dnnl_memory_desc_t *dst_iter_c_desc, const dnnl_memory_desc_t *diff_src_layer_desc, const dnnl_memory_desc_t *diff_src_iter_desc, const dnnl_memory_desc_t *diff_src_iter_c_desc, const dnnl_memory_desc_t *diff_weights_layer_desc, const dnnl_memory_desc_t *diff_weights_iter_desc, const dnnl_memory_desc_t *diff_weights_peephole_desc, const dnnl_memory_desc_t *diff_weights_projection_desc, const dnnl_memory_desc_t *diff_bias_desc, const dnnl_memory_desc_t *diff_dst_layer_desc, const dnnl_memory_desc_t *diff_dst_iter_desc, const dnnl_memory_desc_t *diff_dst_iter_c_desc, unsigned flags)
Initializes a descriptor for an LSTM (with or without peephole and with or with out recurrent project...
dnnl_status_t DNNL_API dnnl_lstm_backward_desc_init_v2(dnnl_rnn_desc_t *rnn_desc, dnnl_prop_kind_t prop_kind, dnnl_rnn_direction_t direction, const dnnl_memory_desc_t *src_layer_desc, const dnnl_memory_desc_t *src_iter_desc, const dnnl_memory_desc_t *src_iter_c_desc, const dnnl_memory_desc_t *weights_layer_desc, const dnnl_memory_desc_t *weights_iter_desc, const dnnl_memory_desc_t *weights_peephole_desc, const dnnl_memory_desc_t *bias_desc, const dnnl_memory_desc_t *dst_layer_desc, const dnnl_memory_desc_t *dst_iter_desc, const dnnl_memory_desc_t *dst_iter_c_desc, const dnnl_memory_desc_t *diff_src_layer_desc, const dnnl_memory_desc_t *diff_src_iter_desc, const dnnl_memory_desc_t *diff_src_iter_c_desc, const dnnl_memory_desc_t *diff_weights_layer_desc, const dnnl_memory_desc_t *diff_weights_iter_desc, const dnnl_memory_desc_t *diff_weights_peephole_desc, const dnnl_memory_desc_t *diff_bias_desc, const dnnl_memory_desc_t *diff_dst_layer_desc, const dnnl_memory_desc_t *diff_dst_iter_desc, const dnnl_memory_desc_t *diff_dst_iter_c_desc, unsigned flags)
Initializes a descriptor for an LSTM (with or without peephole) backward propagation primitive.
dnnl_status_t DNNL_API dnnl_lstm_forward_desc_init(dnnl_rnn_desc_t *rnn_desc, dnnl_prop_kind_t prop_kind, dnnl_rnn_direction_t direction, const dnnl_memory_desc_t *src_layer_desc, const dnnl_memory_desc_t *src_iter_desc, const dnnl_memory_desc_t *src_iter_c_desc, const dnnl_memory_desc_t *weights_layer_desc, const dnnl_memory_desc_t *weights_iter_desc, const dnnl_memory_desc_t *bias_desc, const dnnl_memory_desc_t *dst_layer_desc, const dnnl_memory_desc_t *dst_iter_desc, const dnnl_memory_desc_t *dst_iter_c_desc, unsigned flags)
Initializes a descriptor for LSTM forward propagation primitive.
dnnl_status_t DNNL_API dnnl_lbr_gru_forward_desc_init(dnnl_rnn_desc_t *rnn_desc, dnnl_prop_kind_t prop_kind, dnnl_rnn_direction_t direction, const dnnl_memory_desc_t *src_layer_desc, const dnnl_memory_desc_t *src_iter_desc, const dnnl_memory_desc_t *weights_layer_desc, const dnnl_memory_desc_t *weights_iter_desc, const dnnl_memory_desc_t *bias_desc, const dnnl_memory_desc_t *dst_layer_desc, const dnnl_memory_desc_t *dst_iter_desc, unsigned flags)
Initializes a descriptor for LBR GRU forward propagation primitive.
dnnl_status_t DNNL_API dnnl_lstm_forward_desc_init_v3(dnnl_rnn_desc_t *rnn_desc, dnnl_prop_kind_t prop_kind, dnnl_rnn_direction_t direction, const dnnl_memory_desc_t *src_layer_desc, const dnnl_memory_desc_t *src_iter_desc, const dnnl_memory_desc_t *src_iter_c_desc, const dnnl_memory_desc_t *weights_layer_desc, const dnnl_memory_desc_t *weights_iter_desc, const dnnl_memory_desc_t *weights_peephole_desc, const dnnl_memory_desc_t *weights_projection_desc, const dnnl_memory_desc_t *bias_desc, const dnnl_memory_desc_t *dst_layer_desc, const dnnl_memory_desc_t *dst_iter_desc, const dnnl_memory_desc_t *dst_iter_c_desc, unsigned flags)
Initializes a descriptor for an LSTM (with or without peephole and with or without recurrent projecti...
rnn_flags
RNN cell flags.
Definition: dnnl.hpp:658
dnnl_status_t DNNL_API dnnl_lstm_forward_desc_init_v2(dnnl_rnn_desc_t *rnn_desc, dnnl_prop_kind_t prop_kind, dnnl_rnn_direction_t direction, const dnnl_memory_desc_t *src_layer_desc, const dnnl_memory_desc_t *src_iter_desc, const dnnl_memory_desc_t *src_iter_c_desc, const dnnl_memory_desc_t *weights_layer_desc, const dnnl_memory_desc_t *weights_iter_desc, const dnnl_memory_desc_t *weights_peephole_desc, const dnnl_memory_desc_t *bias_desc, const dnnl_memory_desc_t *dst_layer_desc, const dnnl_memory_desc_t *dst_iter_desc, const dnnl_memory_desc_t *dst_iter_c_desc, unsigned flags)
Initializes a descriptor for an LSTM (with or without peephole) forward propagation primitive.
dnnl_status_t DNNL_API dnnl_gru_backward_desc_init(dnnl_rnn_desc_t *rnn_desc, dnnl_prop_kind_t prop_kind, dnnl_rnn_direction_t direction, const dnnl_memory_desc_t *src_layer_desc, const dnnl_memory_desc_t *src_iter_desc, const dnnl_memory_desc_t *weights_layer_desc, const dnnl_memory_desc_t *weights_iter_desc, const dnnl_memory_desc_t *bias_desc, const dnnl_memory_desc_t *dst_layer_desc, const dnnl_memory_desc_t *dst_iter_desc, const dnnl_memory_desc_t *diff_src_layer_desc, const dnnl_memory_desc_t *diff_src_iter_desc, const dnnl_memory_desc_t *diff_weights_layer_desc, const dnnl_memory_desc_t *diff_weights_iter_desc, const dnnl_memory_desc_t *diff_bias_desc, const dnnl_memory_desc_t *diff_dst_layer_desc, const dnnl_memory_desc_t *diff_dst_iter_desc, unsigned flags)
Initializes a descriptor for GRU backward propagation primitive.
@ unidirectional_left2right
Unidirectional execution of RNN primitive from left to right.
@ unidirectional_right2left
Unidirectional execution of RNN primitive from right to left.
@ bidirectional_concat
Bidirectional execution of RNN primitive with concatenation of the results.
@ unidirectional
Alias for dnnl::rnn_direction::unidirectional_left2right.
@ bidirectional_sum
Bidirectional execution of RNN primitive with summation of the results.
@ dnnl_rnn_flags_undef
Undefined RNN flags.
Definition: dnnl_types.h:1933
@ dnnl_unidirectional
Alias for dnnl_unidirectional_left2right.
Definition: dnnl_types.h:1949
@ dnnl_bidirectional_concat
Bidirectional execution of RNN primitive with concatenation of the results.
Definition: dnnl_types.h:1944
@ dnnl_bidirectional_sum
Bidirectional execution of RNN primitive with summation of the results.
Definition: dnnl_types.h:1947
@ dnnl_unidirectional_left2right
Unidirectional execution of RNN primitive from left to right.
Definition: dnnl_types.h:1939
@ dnnl_unidirectional_right2left
Unidirectional execution of RNN primitive from right to left.
Definition: dnnl_types.h:1941
@ undef
Undefined RNN flags.
dnnl_status_t DNNL_API dnnl_set_jit_dump(int enable)
Configures dumping of JIT-generated code.
status set_max_cpu_isa(cpu_isa isa)
Sets the maximal ISA the library can dispatch to on the CPU.
Definition: dnnl.hpp:10858
dnnl_status_t DNNL_API dnnl_set_verbose(int level)
Configures verbose output to stdout.
const version_t * version()
Returns library version information.
Definition: dnnl.hpp:10812
status set_jit_dump(int enable)
Configures dumping of JIT-generated code.
Definition: dnnl.hpp:10817
status set_cpu_isa_hints(cpu_isa_hints isa_hints)
Sets the hints flag for the CPU ISA.
Definition: dnnl.hpp:10877
dnnl_cpu_isa_t
CPU instruction set flags.
Definition: dnnl_types.h:2664
status set_verbose(int level)
Configures verbose output to stdout.
Definition: dnnl.hpp:10807
cpu_isa get_effective_cpu_isa()
Gets the maximal ISA the library can dispatch to on the CPU.
Definition: dnnl.hpp:10864
dnnl_status_t DNNL_API dnnl_set_max_cpu_isa(dnnl_cpu_isa_t isa)
Sets the maximal ISA the library can dispatch to on the CPU.
dnnl_status_t DNNL_API dnnl_set_jit_profiling_flags(unsigned flags)
Sets library profiling flags.
status set_jit_profiling_jitdumpdir(const std::string &dir)
Sets JIT dump output path.
Definition: dnnl.hpp:10827
status
Status values returned by the library functions.
Definition: dnnl.hpp:10789
const dnnl_version_t DNNL_API * dnnl_version(void)
Returns library version information.
cpu_isa_hints get_cpu_isa_hints()
Gets the ISA specific hints that library can follow.
Definition: dnnl.hpp:10883
status set_jit_profiling_flags(unsigned flags)
Sets library profiling flags.
Definition: dnnl.hpp:10822
cpu_isa
CPU instruction set flags.
Definition: dnnl.hpp:10832
dnnl_cpu_isa_t DNNL_API dnnl_get_effective_cpu_isa(void)
Gets the maximal ISA the library can dispatch to on the CPU.
dnnl_status_t DNNL_API dnnl_set_cpu_isa_hints(dnnl_cpu_isa_hints_t isa_hints)
Sets the hints flag for the CPU ISA.
dnnl_cpu_isa_hints_t DNNL_API dnnl_get_cpu_isa_hints(void)
Gets the ISA specific hints that library can follow.
dnnl_cpu_isa_hints_t
CPU ISA hints flags.
Definition: dnnl_types.h:2710
cpu_isa_hints
CPU ISA hints flags.
Definition: dnnl.hpp:10869
dnnl_status_t DNNL_API dnnl_set_jit_profiling_jitdumpdir(const char *dir)
Sets JIT dump output path.
@ dnnl_cpu_isa_avx512_mic
Intel Advanced Vector Extensions 512 (Intel AVX-512) subset for Intel Xeon Phi processors x200 Series...
Definition: dnnl_types.h:2679
@ dnnl_cpu_isa_avx
Intel Advanced Vector Extensions (Intel AVX)
Definition: dnnl_types.h:2672
@ dnnl_cpu_isa_avx512_core_amx
Intel AVX-512, Intel DL Boost and bfloat16 support and Intel AMX with 8-bit integer and bfloat16 supp...
Definition: dnnl_types.h:2702
@ dnnl_cpu_isa_avx512_core_vnni
Intel AVX-512 and Intel Deep Learning Boost (Intel DL Boost) support for Intel Xeon Scalable processo...
Definition: dnnl_types.h:2692
@ dnnl_cpu_isa_avx2
Intel Advanced Vector Extensions 2 (Intel AVX2)
Definition: dnnl_types.h:2675
@ dnnl_cpu_isa_all
Any ISA (excepting those listed as initial support)
Definition: dnnl_types.h:2666
@ dnnl_cpu_isa_avx512_core
Intel AVX-512 subset for Intel Xeon Scalable processor family and Intel Core processor family.
Definition: dnnl_types.h:2687
@ dnnl_cpu_isa_sse41
Intel Streaming SIMD Extensions 4.1 (Intel SSE4.1)
Definition: dnnl_types.h:2669
@ dnnl_cpu_isa_avx2_vnni
Intel AVX2 and Intel Deep Learning Boost (Intel DL Boost) support.
Definition: dnnl_types.h:2705
@ dnnl_cpu_isa_avx512_core_bf16
Intel AVX-512, Intel DL Boost and bfloat16 support for Intel Xeon Scalable processor family and Intel...
Definition: dnnl_types.h:2697
@ dnnl_cpu_isa_avx512_mic_4ops
Intel AVX-512 subset for Intel Xeon Phi processors 7235, 7285, 7295 Series.
Definition: dnnl_types.h:2683
@ not_required
Queried element is not required for given primitive.
@ invalid_arguments
The operation failed because of incorrect function arguments.
@ success
The operation was successful.
@ unimplemented
The operation failed because requested functionality is not implemented.
@ runtime_error
Primitive or engine failed on execution.
@ out_of_memory
The operation failed due to an out-of-memory condition.
@ iterator_ends
Primitive iterator passed over last primitive descriptor.
@ avx512_mic
Intel Advanced Vector Extensions 512 (Intel AVX-512) subset for Intel Xeon Phi processors x200 Series...
@ avx2
Intel Advanced Vector Extensions 2 (Intel AVX2)
@ avx2_vnni
Intel AVX2 and Intel Deep Learning Boost (Intel DL Boost) support.
@ avx
Intel Advanced Vector Extensions (Intel AVX)
@ all
Any ISA (excepting those listed as initial support)
@ avx512_core
Intel AVX-512 subset for Intel Xeon Scalable processor family and Intel Core processor family.
@ avx512_mic_4ops
Intel AVX-512 subset for Intel Xeon Phi processors 7235, 7285, 7295 Series.
@ sse41
Intel Streaming SIMD Extensions 4.1 (Intel SSE4.1)
@ avx512_core_vnni
Intel AVX-512 and Intel Deep Learning Boost (Intel DL Boost) support for Intel Xeon Scalable processo...
@ avx512_core_amx
Intel AVX-512, Intel DL Boost and bfloat16 support and Intel AMX with 8-bit integer and bfloat16 supp...
@ avx512_core_bf16
Intel AVX-512, Intel DL Boost and bfloat16 support for Intel Xeon Scalable processor family and Intel...
@ dnnl_cpu_isa_no_hints
No hints (use default features)
Definition: dnnl_types.h:2712
@ dnnl_cpu_isa_prefer_ymm
Prefer to exclusively use Ymm registers for computations.
Definition: dnnl_types.h:2715
@ no_hints
No hints (use default features)
@ prefer_ymm
Prefer to exclusively use Ymm registers for computations.
dnnl_status_t DNNL_API dnnl_shuffle_forward_desc_init(dnnl_shuffle_desc_t *shuffle_desc, dnnl_prop_kind_t prop_kind, const dnnl_memory_desc_t *data_desc, int axis, dnnl_dim_t group_size)
Initializes a descriptor for shuffle forward propagation primitive.
dnnl_status_t DNNL_API dnnl_shuffle_backward_desc_init(dnnl_shuffle_desc_t *shuffle_desc, const dnnl_memory_desc_t *diff_data_desc, int axis, dnnl_dim_t group_size)
Initializes a descriptor for shuffle backward propagation primitive.
dnnl_status_t DNNL_API dnnl_softmax_backward_desc_init(dnnl_softmax_desc_t *softmax_desc, const dnnl_memory_desc_t *diff_data_desc, const dnnl_memory_desc_t *data_desc, int softmax_axis)
Initializes a descriptor for softmax backward propagation primitive.
dnnl_status_t DNNL_API dnnl_softmax_forward_desc_init(dnnl_softmax_desc_t *softmax_desc, dnnl_prop_kind_t prop_kind, const dnnl_memory_desc_t *data_desc, int softmax_axis)
Initializes a descriptor for softmax forward propagation primitive.
dnnl_stream_flags_t
Stream flags.
Definition: dnnl_types.h:2586
dnnl_status_t DNNL_API dnnl_stream_wait(dnnl_stream_t stream)
Waits for all primitives in the execution stream to finish computations.
dnnl_status_t DNNL_API dnnl_stream_get_engine(const_dnnl_stream_t stream, dnnl_engine_t *engine)
Returns the engine of a stream object.
dnnl_status_t DNNL_API dnnl_stream_destroy(dnnl_stream_t stream)
Destroys an execution stream.
dnnl_status_t DNNL_API dnnl_stream_create(dnnl_stream_t *stream, dnnl_engine_t engine, unsigned flags)
Creates an execution stream.
@ dnnl_stream_out_of_order
Out-of-order execution.
Definition: dnnl_types.h:2590
@ dnnl_stream_default_flags
Default stream configuration.
Definition: dnnl_types.h:2592
dnnl_status_t DNNL_API dnnl_sum_primitive_desc_create(dnnl_primitive_desc_t *sum_primitive_desc, const dnnl_memory_desc_t *dst_desc, int n, const float *scales, const dnnl_memory_desc_t *src_descs, const_dnnl_primitive_attr_t attr, dnnl_engine_t engine)
Creates a primitive descriptor for an (out-of-place) sum primitive.
dnnl_status_t
Status values returned by the library functions.
Definition: dnnl_types.h:39
@ dnnl_iterator_ends
Primitive iterator passed over last primitive descriptor.
Definition: dnnl_types.h:49
@ dnnl_runtime_error
Primitive or engine failed on execution.
Definition: dnnl_types.h:51
@ dnnl_unimplemented
The operation failed because requested functionality is not implemented.
Definition: dnnl_types.h:47
@ dnnl_out_of_memory
The operation failed due to an out-of-memory condition.
Definition: dnnl_types.h:43
@ dnnl_success
The operation was successful.
Definition: dnnl_types.h:41
@ dnnl_invalid_arguments
The operation failed because of incorrect function arguments.
Definition: dnnl_types.h:45
@ dnnl_not_required
Queried element is not required for given primitive.
Definition: dnnl_types.h:53
oneDNN namespace
Definition: dnnl.hpp:74
oneAPI namespace
Definition: dnnl.hpp:10981
C API.
Descriptor for a batch normalization backward propagation primitive.
Definition: dnnl.hpp:6605
desc(prop_kind aprop_kind, const memory::desc &diff_data_desc, const memory::desc &data_desc, float epsilon, normalization_flags flags)
Constructs a batch normalization descriptor for backward propagation.
Definition: dnnl.hpp:6620
Primitive descriptor for a batch normalization backward propagation primitive.
Definition: dnnl.hpp:6634
primitive_desc(const desc &adesc, const engine &aengine, const batch_normalization_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a batch normalization backward propagation primitive.
Definition: dnnl.hpp:6651
memory::desc weights_desc() const
Returns a weights memory descriptor.
Definition: dnnl.hpp:6694
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a batch normalization backward propagation primitive from a C A...
Definition: dnnl.hpp:6684
primitive_desc()=default
Default constructor. Produces an empty object.
memory::desc workspace_desc() const
Returns the workspace memory descriptor.
Definition: dnnl.hpp:6719
memory::desc diff_src_desc() const
Returns a diff source memory descriptor.
Definition: dnnl.hpp:6700
memory::desc variance_desc() const
Returns memory descriptor for variance.
Definition: dnnl.hpp:6714
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, const batch_normalization_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a batch normalization backward propagation primitive.
Definition: dnnl.hpp:6671
memory::desc dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:6697
memory::desc src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:6691
memory::desc diff_dst_desc() const
Returns a diff destination memory descriptor.
Definition: dnnl.hpp:6703
memory::desc diff_weights_desc() const
Returns a diff weights memory descriptor.
Definition: dnnl.hpp:6706
memory::desc mean_desc() const
Returns memory descriptor for mean.
Definition: dnnl.hpp:6711
Batch normalization backward propagation primitive.
Definition: dnnl.hpp:6603
batch_normalization_backward()=default
Default constructor. Produces an empty object.
batch_normalization_backward(const primitive_desc &pd)
Constructs a batch normalization backward propagation primitive.
Definition: dnnl.hpp:6728
Descriptor for a batch normalization forward propagation primitive.
Definition: dnnl.hpp:6476
desc(prop_kind aprop_kind, const memory::desc &data_desc, float epsilon, normalization_flags flags)
Constructs a batch normalization descriptor for forward propagation.
Definition: dnnl.hpp:6493
Primitive descriptor for a batch normalization forward propagation primitive.
Definition: dnnl.hpp:6506
memory::desc src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:6554
primitive_desc()=default
Default constructor. Produces an empty object.
memory::desc weights_desc() const
Returns a weights memory descriptor.
Definition: dnnl.hpp:6560
memory::desc mean_desc() const
Returns memory descriptor for mean.
Definition: dnnl.hpp:6567
primitive_desc(const desc &adesc, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a batch normalization forward propagation primitive.
Definition: dnnl.hpp:6520
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a batch normalization forward propagation primitive.
Definition: dnnl.hpp:6536
memory::desc workspace_desc() const
Returns the workspace memory descriptor.
Definition: dnnl.hpp:6563
memory::desc variance_desc() const
Returns memory descriptor for variance.
Definition: dnnl.hpp:6571
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a batch normalization forward propagation primitive from a C AP...
Definition: dnnl.hpp:6547
memory::desc dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:6557
Batch normalization forward propagation primitive.
Definition: dnnl.hpp:6474
batch_normalization_forward()=default
Default constructor. Produces an empty object.
batch_normalization_forward(const primitive_desc &pd)
Constructs a batch normalization forward propagation primitive.
Definition: dnnl.hpp:6599
Descriptor for an elementwise binary operator primitive.
Definition: dnnl.hpp:9754
desc()=default
Default constructor. Produces an empty object.
dnnl_binary_desc_t data
Underlying C operation descriptor.
Definition: dnnl.hpp:9756
desc(algorithm aalgorithm, const memory::desc &src0, const memory::desc &src1, const memory::desc &dst)
Constructs a descriptor for an elementwise binary operator primitive.
Definition: dnnl.hpp:9768
Primitive descriptor for an elementwise binary operator primitive.
Definition: dnnl.hpp:9779
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for an elementwise binary operator primitive.
Definition: dnnl.hpp:9807
memory::desc src_desc(int idx=0) const
Returns a source memory descriptor.
Definition: dnnl.hpp:9820
memory::desc src0_desc() const
Returns the memory descriptor for source #0.
Definition: dnnl.hpp:9823
primitive_desc()=default
Default constructor. Produces an empty object.
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a binary primitive from a C API primitive descriptor that must ...
Definition: dnnl.hpp:9816
memory::desc dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:9829
memory::desc src1_desc() const
Returns the memory descriptor for source #1.
Definition: dnnl.hpp:9826
primitive_desc(const desc &adesc, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for an elementwise binary operator primitive.
Definition: dnnl.hpp:9792
Elementwise binary operator primitive.
Definition: dnnl.hpp:9752
binary()=default
Default constructor. Produces an empty object.
binary(const primitive_desc &pd)
Constructs an elementwise binary operation primitive.
Definition: dnnl.hpp:9838
Primitive descriptor for a concat primitive.
Definition: dnnl.hpp:3695
memory::desc dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:3764
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for concat primitive from a C API primitive descriptor which must h...
Definition: dnnl.hpp:3757
primitive_desc(const memory::desc &dst, int concat_dimension, const std::vector< memory::desc > &srcs, const engine &aengine, const primitive_attr &attr=primitive_attr())
Constructs a primitive descriptor for an out-of-place concatenation primitive.
Definition: dnnl.hpp:3711
primitive_desc()=default
Default constructor. Produces an empty object.
primitive_desc(int concat_dimension, const std::vector< memory::desc > &srcs, const engine &aengine, const primitive_attr &attr=primitive_attr())
Constructs a primitive descriptor for an out-of-place concatenation primitive.
Definition: dnnl.hpp:3738
memory::desc src_desc(int idx=0) const
Returns a source memory descriptor.
Definition: dnnl.hpp:3761
Tensor concatenation (concat) primitive.
Definition: dnnl.hpp:3693
concat()=default
Default constructor. Produces an empty object.
concat(const primitive_desc &pd)
Constructs a concatenation primitive.
Definition: dnnl.hpp:3772
Descriptor for a convolution backward propagation primitive.
Definition: dnnl.hpp:4236
desc(algorithm aalgorithm, const memory::desc &diff_src_desc, const memory::desc &weights_desc, const memory::desc &diff_dst_desc, const memory::dims &strides, const memory::dims &dilates, const memory::dims &padding_l, const memory::dims &padding_r)
Constructs a descriptor for dilated convolution backward propagation primitive.
Definition: dnnl.hpp:4307
desc(algorithm aalgorithm, const memory::desc &diff_src_desc, const memory::desc &weights_desc, const memory::desc &diff_dst_desc, const memory::dims &strides, const memory::dims &padding_l, const memory::dims &padding_r)
Constructs a descriptor for a convolution backward propagation primitive.
Definition: dnnl.hpp:4264
Primitive descriptor for a convolution backward propagation primitive.
Definition: dnnl.hpp:4328
memory::desc weights_desc() const
Returns a weights memory descriptor.
Definition: dnnl.hpp:4386
memory::desc diff_dst_desc() const
Returns a diff destination memory descriptor.
Definition: dnnl.hpp:4389
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a convolution backward propagation primitive from a C API primi...
Definition: dnnl.hpp:4378
primitive_desc()=default
Default constructor. Produces an empty object.
primitive_desc(const desc &adesc, const engine &aengine, const convolution_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a convolution backward propagation primitive.
Definition: dnnl.hpp:4345
memory::desc diff_src_desc() const
Returns a diff source memory descriptor.
Definition: dnnl.hpp:4383
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, const convolution_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a convolution backward propagation primitive.
Definition: dnnl.hpp:4365
Convolution backward propagation primitive.
Definition: dnnl.hpp:4233
convolution_backward_data()=default
Default constructor. Produces an empty object.
convolution_backward_data(const primitive_desc &pd)
Constructs a convolution backward propagation primitive.
Definition: dnnl.hpp:4398
Descriptor for a convolution weights gradient primitive.
Definition: dnnl.hpp:4404
desc(algorithm aalgorithm, const memory::desc &src_desc, const memory::desc &diff_weights_desc, const memory::desc &diff_dst_desc, const memory::dims &strides, const memory::dims &padding_l, const memory::dims &padding_r)
Constructs a descriptor for a convolution weights gradient primitive without bias.
Definition: dnnl.hpp:4477
desc(algorithm aalgorithm, const memory::desc &src_desc, const memory::desc &diff_weights_desc, const memory::desc &diff_bias_desc, const memory::desc &diff_dst_desc, const memory::dims &strides, const memory::dims &dilates, const memory::dims &padding_l, const memory::dims &padding_r)
Constructs a descriptor for a dilated convolution weights gradient primitive with bias.
Definition: dnnl.hpp:4522
desc(algorithm aalgorithm, const memory::desc &src_desc, const memory::desc &diff_weights_desc, const memory::desc &diff_dst_desc, const memory::dims &strides, const memory::dims &dilates, const memory::dims &padding_l, const memory::dims &padding_r)
Constructs a descriptor for a dilated convolution weights gradient primitive without bias.
Definition: dnnl.hpp:4569
desc(algorithm aalgorithm, const memory::desc &src_desc, const memory::desc &diff_weights_desc, const memory::desc &diff_bias_desc, const memory::desc &diff_dst_desc, const memory::dims &strides, const memory::dims &padding_l, const memory::dims &padding_r)
Constructs a descriptor for a convolution weights gradient primitive with bias.
Definition: dnnl.hpp:4434
Primitive descriptor for a convolution weights gradient primitive.
Definition: dnnl.hpp:4590
memory::desc diff_bias_desc() const
Returns the diff bias memory descriptor.
Definition: dnnl.hpp:4657
memory::desc diff_weights_desc() const
Returns a diff weights memory descriptor.
Definition: dnnl.hpp:4646
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, const convolution_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a convolution weights gradient primitive.
Definition: dnnl.hpp:4625
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a convolution weights gradient primitive from a C API primitive...
Definition: dnnl.hpp:4638
memory::desc src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:4643
primitive_desc()=default
Default constructor. Produces an empty object.
primitive_desc(const desc &adesc, const engine &aengine, const convolution_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a convolution weights gradient primitive.
Definition: dnnl.hpp:4606
memory::desc diff_dst_desc() const
Returns a diff destination memory descriptor.
Definition: dnnl.hpp:4651
Convolution weights gradient primitive.
Definition: dnnl.hpp:4402
convolution_backward_weights()=default
Default constructor. Produces an empty object.
convolution_backward_weights(const primitive_desc &pd)
Constructs a convolution weights gradient primitive.
Definition: dnnl.hpp:4668
Descriptor for a convolution forward propagation primitive.
Definition: dnnl.hpp:3963
desc(prop_kind aprop_kind, algorithm aalgorithm, const memory::desc &src_desc, const memory::desc &weights_desc, const memory::desc &bias_desc, const memory::desc &dst_desc, const memory::dims &strides, const memory::dims &dilates, const memory::dims &padding_l, const memory::dims &padding_r)
Constructs a descriptor for a dilated convolution forward propagation primitive with bias.
Definition: dnnl.hpp:4091
desc(prop_kind aprop_kind, algorithm aalgorithm, const memory::desc &src_desc, const memory::desc &weights_desc, const memory::desc &dst_desc, const memory::dims &strides, const memory::dims &padding_l, const memory::dims &padding_r)
Constructs a descriptor for a convolution forward propagation primitive without bias.
Definition: dnnl.hpp:4042
desc(prop_kind aprop_kind, algorithm aalgorithm, const memory::desc &src_desc, const memory::desc &weights_desc, const memory::desc &dst_desc, const memory::dims &strides, const memory::dims &dilates, const memory::dims &padding_l, const memory::dims &padding_r)
Constructs a descriptor for a dilated convolution forward propagation primitive without bias.
Definition: dnnl.hpp:4140
desc(prop_kind aprop_kind, algorithm aalgorithm, const memory::desc &src_desc, const memory::desc &weights_desc, const memory::desc &bias_desc, const memory::desc &dst_desc, const memory::dims &strides, const memory::dims &padding_l, const memory::dims &padding_r)
Constructs a descriptor for a convolution forward propagation primitive with bias.
Definition: dnnl.hpp:3996
Primitive descriptor for a convolution forward propagation primitive.
Definition: dnnl.hpp:4161
primitive_desc()=default
Default constructor. Produces an empty object.
primitive_desc(const desc &adesc, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a convolution forward propagation primitive.
Definition: dnnl.hpp:4175
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a convolution forward propagation primitive.
Definition: dnnl.hpp:4191
memory::desc src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:4208
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a convolution forward propagation primitive from a C API primit...
Definition: dnnl.hpp:4202
memory::desc bias_desc() const
Returns the bias memory descriptor.
Definition: dnnl.hpp:4220
memory::desc weights_desc() const
Returns a weights memory descriptor.
Definition: dnnl.hpp:4211
memory::desc dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:4214
Convolution forward propagation primitive.
Definition: dnnl.hpp:3961
convolution_forward(const primitive_desc &pd)
Constructs a convolution forward propagation primitive.
Definition: dnnl.hpp:4229
convolution_forward()=default
Default constructor. Produces an empty object.
Descriptor for a deconvolution backward propagation primitive.
Definition: dnnl.hpp:4949
desc(algorithm aalgorithm, const memory::desc &diff_src_desc, const memory::desc &weights_desc, const memory::desc &diff_dst_desc, const memory::dims &strides, const memory::dims &dilates, const memory::dims &padding_l, const memory::dims &padding_r)
Constructs a descriptor for a dilated deconvolution backward propagation primitive.
Definition: dnnl.hpp:5018
desc(algorithm aalgorithm, const memory::desc &diff_src_desc, const memory::desc &weights_desc, const memory::desc &diff_dst_desc, const memory::dims &strides, const memory::dims &padding_l, const memory::dims &padding_r)
Constructs a descriptor for a deconvolution backward propagation primitive.
Definition: dnnl.hpp:4976
Primitive descriptor for a deconvolution backward propagation primitive.
Definition: dnnl.hpp:5039
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, const deconvolution_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a deconvolution backward propagation primitive.
Definition: dnnl.hpp:5076
memory::desc weights_desc() const
Returns a weights memory descriptor.
Definition: dnnl.hpp:5097
memory::desc diff_dst_desc() const
Returns a diff destination memory descriptor.
Definition: dnnl.hpp:5100
memory::desc diff_src_desc() const
Returns a diff source memory descriptor.
Definition: dnnl.hpp:5094
primitive_desc(const desc &adesc, const engine &aengine, const deconvolution_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a deconvolution backward propagation primitive.
Definition: dnnl.hpp:5056
primitive_desc()=default
Default constructor. Produces an empty object.
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a deconvolution backward propagation primitive from a C API pri...
Definition: dnnl.hpp:5089
Deconvolution backward propagation primitive.
Definition: dnnl.hpp:4947
deconvolution_backward_data()=default
Default constructor. Produces an empty object.
deconvolution_backward_data(const primitive_desc &pd)
Constructs a deconvolution backward propagation primitive.
Definition: dnnl.hpp:5109
Descriptor for a deconvolution weights gradient primitive.
Definition: dnnl.hpp:5115
desc(algorithm aalgorithm, const memory::desc &src_desc, const memory::desc &diff_weights_desc, const memory::desc &diff_dst_desc, const memory::dims &strides, const memory::dims &dilates, const memory::dims &padding_l, const memory::dims &padding_r)
Constructs a descriptor for a dilated deconvolution weights gradient primitive without bias.
Definition: dnnl.hpp:5276
desc(algorithm aalgorithm, const memory::desc &src_desc, const memory::desc &diff_weights_desc, const memory::desc &diff_bias_desc, const memory::desc &diff_dst_desc, const memory::dims &strides, const memory::dims &dilates, const memory::dims &padding_l, const memory::dims &padding_r)
Constructs a descriptor for a dilated deconvolution weights gradient primitive with bias.
Definition: dnnl.hpp:5230
desc(algorithm aalgorithm, const memory::desc &src_desc, const memory::desc &diff_weights_desc, const memory::desc &diff_dst_desc, const memory::dims &strides, const memory::dims &padding_l, const memory::dims &padding_r)
Constructs a descriptor for a deconvolution weights gradient primitive without bias.
Definition: dnnl.hpp:5186
desc(algorithm aalgorithm, const memory::desc &src_desc, const memory::desc &diff_weights_desc, const memory::desc &diff_bias_desc, const memory::desc &diff_dst_desc, const memory::dims &strides, const memory::dims &padding_l, const memory::dims &padding_r)
Constructs a descriptor for a deconvolution weights gradient primitive with bias.
Definition: dnnl.hpp:5144
Primitive descriptor for a deconvolution weights gradient primitive.
Definition: dnnl.hpp:5297
memory::desc src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:5352
memory::desc diff_dst_desc() const
Returns a diff destination memory descriptor.
Definition: dnnl.hpp:5360
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a deconvolution weights gradient primitive from a C API primiti...
Definition: dnnl.hpp:5347
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, const deconvolution_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a deconvolution weights update primitive.
Definition: dnnl.hpp:5334
primitive_desc(const desc &adesc, const engine &aengine, const deconvolution_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a deconvolution weights update primitive.
Definition: dnnl.hpp:5314
memory::desc diff_weights_desc() const
Returns a diff weights memory descriptor.
Definition: dnnl.hpp:5355
memory::desc diff_bias_desc() const
Returns the diff bias memory descriptor.
Definition: dnnl.hpp:5363
primitive_desc()=default
Default constructor. Produces an empty object.
Deconvolution weights gradient primitive.
Definition: dnnl.hpp:5113
deconvolution_backward_weights()=default
Default constructor. Produces an empty object.
deconvolution_backward_weights(const primitive_desc &pd)
Constructs a deconvolution weights gradient primitive.
Definition: dnnl.hpp:5374
Descriptor for a deconvolution forward propagation primitive.
Definition: dnnl.hpp:4684
desc(prop_kind aprop_kind, algorithm aalgorithm, const memory::desc &src_desc, const memory::desc &weights_desc, const memory::desc &bias_desc, const memory::desc &dst_desc, const memory::dims &strides, const memory::dims &dilates, const memory::dims &padding_l, const memory::dims &padding_r)
Constructs a descriptor for a dilated deconvolution forward propagation primitive with bias.
Definition: dnnl.hpp:4809
desc(prop_kind aprop_kind, algorithm aalgorithm, const memory::desc &src_desc, const memory::desc &weights_desc, const memory::desc &dst_desc, const memory::dims &strides, const memory::dims &padding_l, const memory::dims &padding_r)
Constructs a descriptor for a deconvolution forward propagation primitive without bias.
Definition: dnnl.hpp:4761
desc(prop_kind aprop_kind, algorithm aalgorithm, const memory::desc &src_desc, const memory::desc &weights_desc, const memory::desc &bias_desc, const memory::desc &dst_desc, const memory::dims &strides, const memory::dims &padding_l, const memory::dims &padding_r)
Constructs a descriptor for a deconvolution forward propagation primitive with bias.
Definition: dnnl.hpp:4716
desc(prop_kind aprop_kind, algorithm aalgorithm, const memory::desc &src_desc, const memory::desc &weights_desc, const memory::desc &dst_desc, const memory::dims &strides, const memory::dims &dilates, const memory::dims &padding_l, const memory::dims &padding_r)
Constructs a descriptor for a dilated deconvolution forward propagation primitive without bias.
Definition: dnnl.hpp:4857
Primitive descriptor for a deconvolution forward propagation primitive.
Definition: dnnl.hpp:4878
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a deconvolution forward propagation primitive from a C API prim...
Definition: dnnl.hpp:4919
memory::desc dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:4931
memory::desc src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:4925
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a deconvolution forward propagation primitive.
Definition: dnnl.hpp:4908
primitive_desc(const desc &adesc, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a deconvolution forward propagation primitive.
Definition: dnnl.hpp:4892
primitive_desc()=default
Default constructor. Produces an empty object.
memory::desc bias_desc() const
Returns the bias memory descriptor.
Definition: dnnl.hpp:4934
memory::desc weights_desc() const
Returns a weights memory descriptor.
Definition: dnnl.hpp:4928
Deconvolution forward propagation primitive.
Definition: dnnl.hpp:4682
deconvolution_forward(const primitive_desc &pd)
Constructs a deconvolution forward propagation primitive.
Definition: dnnl.hpp:4943
deconvolution_forward()=default
Default constructor. Produces an empty object.
Descriptor for an elementwise backward propagation primitive.
Definition: dnnl.hpp:5943
desc(algorithm aalgorithm, const memory::desc &diff_data_desc, const memory::desc &data_desc, float alpha=0, float beta=0)
Constructs a descriptor for an elementwise backward propagation primitive.
Definition: dnnl.hpp:5957
Primitive descriptor for eltwise backward propagation.
Definition: dnnl.hpp:5970
memory::desc diff_src_desc() const
Returns a diff source memory descriptor.
Definition: dnnl.hpp:6028
primitive_desc()=default
Default constructor. Produces an empty object.
primitive_desc(const desc &adesc, const engine &aengine, const eltwise_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for an elementwise backward propagation primitive.
Definition: dnnl.hpp:5987
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, const eltwise_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for an elementwise backward propagation primitive.
Definition: dnnl.hpp:6007
memory::desc src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:6025
memory::desc diff_dst_desc() const
Returns a diff destination memory descriptor.
Definition: dnnl.hpp:6031
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for an eltwise backward propagation primitive from a C API primitiv...
Definition: dnnl.hpp:6020
Elementwise unary operation backward propagation primitive.
Definition: dnnl.hpp:5941
eltwise_backward()=default
Default constructor. Produces an empty object.
eltwise_backward(const primitive_desc &pd)
Constructs an eltwise backward propagation primitive.
Definition: dnnl.hpp:6040
Descriptor for an elementwise forward propagation primitive.
Definition: dnnl.hpp:5850
desc(prop_kind aprop_kind, algorithm aalgorithm, const memory::desc &data_desc, float alpha=0, float beta=0)
Constructs a descriptor for an elementwise forward propagation primitive.
Definition: dnnl.hpp:5865
Primitive descriptor for an elementwise forward propagation primitive.
Definition: dnnl.hpp:5878
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for an elementwise forward propagation primitive.
Definition: dnnl.hpp:5908
primitive_desc()=default
Default constructor. Produces an empty object.
memory::desc dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:5928
memory::desc src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:5925
primitive_desc(const desc &adesc, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for an elementwise forward propagation primitive.
Definition: dnnl.hpp:5892
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for an eltwise forward propagation primitive from a C API primitive...
Definition: dnnl.hpp:5919
Elementwise unary operation forward propagation primitive.
Definition: dnnl.hpp:5848
eltwise_forward(const primitive_desc &pd)
Constructs an eltwise forward propagation primitive.
Definition: dnnl.hpp:5937
eltwise_forward()=default
Default constructor. Produces an empty object.
An execution engine.
Definition: dnnl.hpp:869
static engine query(const primitive_desc &pd)
Returns the engine of a primitive descriptor.
Definition: dnnl.hpp:938
kind
Kinds of engines.
Definition: dnnl.hpp:874
engine(kind akind, size_t index)
Constructs an engine.
Definition: dnnl.hpp:902
engine()=default
Constructs an empty engine.
static size_t get_count(kind akind)
Returns the number of engines of a certain kind.
Definition: dnnl.hpp:893
engine(const handle< dnnl_primitive_desc_t > &pd)
Constructs an engine based on a primitive from the primitive descriptor pd by querying its engine.
Definition: dnnl.hpp:914
kind get_kind() const
Returns the kind of the engine.
Definition: dnnl.hpp:925
oneDNN exception class.
Definition: dnnl.hpp:84
error(dnnl_status_t status, const char *message)
Constructs an instance of an exception class.
Definition: dnnl.hpp:92
static void wrap_c_api(dnnl_status_t status, const char *message)
A convenience function for wrapping calls to C API functions.
Definition: dnnl.hpp:103
const char * what() const noexcept override
Returns the explanatory string.
Definition: dnnl.hpp:96
Descriptor for a GRU backward propagation primitive.
Definition: dnnl.hpp:9001
desc(prop_kind aprop_kind, rnn_direction direction, const memory::desc &src_layer_desc, const memory::desc &src_iter_desc, const memory::desc &weights_layer_desc, const memory::desc &weights_iter_desc, const memory::desc &bias_desc, const memory::desc &dst_layer_desc, const memory::desc &dst_iter_desc, const memory::desc &diff_src_layer_desc, const memory::desc &diff_src_iter_desc, const memory::desc &diff_weights_layer_desc, const memory::desc &diff_weights_iter_desc, const memory::desc &diff_bias_desc, const memory::desc &diff_dst_layer_desc, const memory::desc &diff_dst_iter_desc, rnn_flags flags=rnn_flags::undef)
Constructs a descriptor for a GRU backward propagation primitive.
Definition: dnnl.hpp:9048
Primitive descriptor for a GRU backward propagation primitive.
Definition: dnnl.hpp:9082
primitive_desc(const desc &adesc, const engine &aengine, const gru_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a GRU backward propagation primitive.
Definition: dnnl.hpp:9098
memory::desc diff_weights_iter_desc() const
Returns diff weights iteration memory descriptor.
Definition: dnnl.hpp:9184
memory::desc dst_layer_desc() const
Returns destination layer memory descriptor.
Definition: dnnl.hpp:9156
memory::desc weights_layer_desc() const
Returns weights layer memory descriptor.
Definition: dnnl.hpp:9143
memory::desc src_iter_desc() const
Returns source iteration memory descriptor.
Definition: dnnl.hpp:9140
memory::desc diff_bias_desc() const
Returns diff bias memory descriptor.
Definition: dnnl.hpp:9189
memory::desc weights_iter_desc() const
Returns weights iteration memory descriptor.
Definition: dnnl.hpp:9148
memory::desc bias_desc() const
Returns bias memory descriptor.
Definition: dnnl.hpp:9153
primitive_desc()=default
Default constructor. Produces an empty object.
memory::desc diff_dst_iter_desc() const
Returns diff destination iteration memory descriptor.
Definition: dnnl.hpp:9199
memory::desc diff_dst_layer_desc() const
Returns diff destination layer memory descriptor.
Definition: dnnl.hpp:9194
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a GRU backward propagation primitive from a C API primitive des...
Definition: dnnl.hpp:9130
memory::desc src_layer_desc() const
Returns source layer memory descriptor.
Definition: dnnl.hpp:9135
memory::desc workspace_desc() const
Returns the workspace memory descriptor.
Definition: dnnl.hpp:9164
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, const gru_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a GRU backward propagation primitive.
Definition: dnnl.hpp:9117
memory::desc diff_src_layer_desc() const
Returns diff source layer memory descriptor.
Definition: dnnl.hpp:9169
memory::desc diff_src_iter_desc() const
Returns diff source iteration memory descriptor.
Definition: dnnl.hpp:9174
memory::desc diff_weights_layer_desc() const
Returns diff weights layer memory descriptor.
Definition: dnnl.hpp:9179
memory::desc dst_iter_desc() const
Returns destination iteration memory descriptor.
Definition: dnnl.hpp:9161
GRU backward propagation primitive.
Definition: dnnl.hpp:8999
gru_backward()=default
Default constructor. Produces an empty object.
gru_backward(const primitive_desc &pd)
Constructs a GRU backward propagation primitive.
Definition: dnnl.hpp:9210
Descriptor for a GRU forward propagation primitive.
Definition: dnnl.hpp:8852
desc(prop_kind aprop_kind, rnn_direction direction, const memory::desc &src_layer_desc, const memory::desc &src_iter_desc, const memory::desc &weights_layer_desc, const memory::desc &weights_iter_desc, const memory::desc &bias_desc, const memory::desc &dst_layer_desc, const memory::desc &dst_iter_desc, rnn_flags flags=rnn_flags::undef)
Constructs a descriptor for a GRU forward propagation primitive.
Definition: dnnl.hpp:8887
Primitive descriptor for a GRU forward propagation primitive.
Definition: dnnl.hpp:8910
primitive_desc(const desc &adesc, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a GRU forward propagation primitive.
Definition: dnnl.hpp:8923
memory::desc weights_iter_desc() const
Returns weights iteration memory descriptor.
Definition: dnnl.hpp:8968
memory::desc src_layer_desc() const
Returns source layer memory descriptor.
Definition: dnnl.hpp:8955
memory::desc dst_layer_desc() const
Returns destination layer memory descriptor.
Definition: dnnl.hpp:8976
memory::desc weights_layer_desc() const
Returns weights layer memory descriptor.
Definition: dnnl.hpp:8963
memory::desc bias_desc() const
Returns bias memory descriptor.
Definition: dnnl.hpp:8973
primitive_desc()=default
Default constructor. Produces an empty object.
memory::desc dst_iter_desc() const
Returns destination iteration memory descriptor.
Definition: dnnl.hpp:8981
memory::desc workspace_desc() const
Returns the workspace memory descriptor.
Definition: dnnl.hpp:8984
memory::desc src_iter_desc() const
Returns source iteration memory descriptor.
Definition: dnnl.hpp:8960
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a GRU forward propagation primitive from a C API primitive desc...
Definition: dnnl.hpp:8949
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a GRU forward propagation primitive.
Definition: dnnl.hpp:8938
GRU forward propagation primitive.
Definition: dnnl.hpp:8850
gru_forward(const primitive_desc &pd)
Constructs a GRU forward propagation primitive.
Definition: dnnl.hpp:8995
gru_forward()=default
Default constructor. Produces an empty object.
A class that provides the destructor for a oneDNN C API handle.
Definition: dnnl.hpp:120
oneDNN C API handle wrapper class.
Definition: dnnl.hpp:136
handle(const handle< T, traits > &)=default
Copy constructor.
bool operator==(const handle< T, traits > &other) const
Equality operator.
Definition: dnnl.hpp:210
handle< T, traits > & operator=(const handle< T, traits > &)=default
Assignment operator.
bool operator!=(const handle &other) const
Inequality operator.
Definition: dnnl.hpp:220
T get(bool allow_empty=false) const
Returns the underlying C API handle.
Definition: dnnl.hpp:185
handle()=default
Constructs an empty handle object.
void reset(T t, bool weak=false)
Resets the handle wrapper objects to wrap a new C API handle.
Definition: dnnl.hpp:176
handle(T t, bool weak=false)
Constructs a handle wrapper object from a C API handle.
Definition: dnnl.hpp:169
handle(handle< T, traits > &&)=default
Move constructor.
handle< T, traits > & operator=(handle< T, traits > &&)=default
Move assignment operator.
Descriptor for an inner product backward propagation primitive.
Definition: dnnl.hpp:7192
desc(const memory::desc &diff_src_desc, const memory::desc &weights_desc, const memory::desc &diff_dst_desc)
Constructs a descriptor for an inner product backward propagation primitive.
Definition: dnnl.hpp:7205
Primitive descriptor for an inner product backward propagation primitive.
Definition: dnnl.hpp:7218
memory::desc diff_dst_desc() const
Returns a diff destination memory descriptor.
Definition: dnnl.hpp:7279
primitive_desc(const desc &adesc, const engine &aengine, const inner_product_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for an inner product backward propagation primitive.
Definition: dnnl.hpp:7235
memory::desc weights_desc() const
Returns a weights memory descriptor.
Definition: dnnl.hpp:7276
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for an inner product backward propagation primitive from a C API pr...
Definition: dnnl.hpp:7268
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, const inner_product_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for an inner product backward propagation primitive.
Definition: dnnl.hpp:7255
memory::desc diff_src_desc() const
Returns a diff source memory descriptor.
Definition: dnnl.hpp:7273
primitive_desc()=default
Default constructor. Produces an empty object.
Inner product backward propagation primitive.
Definition: dnnl.hpp:7190
inner_product_backward_data(const primitive_desc &pd)
Constructs an inner product backward propagation primitive.
Definition: dnnl.hpp:7288
inner_product_backward_data()=default
Default constructor. Produces an empty object.
Descriptor for an inner product weights gradient primitive.
Definition: dnnl.hpp:7294
desc(const memory::desc &src_desc, const memory::desc &diff_weights_desc, const memory::desc &diff_bias_desc, const memory::desc &diff_dst_desc)
Constructs a descriptor for an inner product descriptor weights update primitive with bias.
Definition: dnnl.hpp:7308
desc(const memory::desc &src_desc, const memory::desc &diff_weights_desc, const memory::desc &diff_dst_desc)
Constructs a descriptor for an inner product descriptor weights update primitive without bias.
Definition: dnnl.hpp:7330
Primitive descriptor for an inner product weights gradient primitive.
Definition: dnnl.hpp:7343
memory::desc src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:7398
memory::desc diff_weights_desc() const
Returns a diff weights memory descriptor.
Definition: dnnl.hpp:7401
primitive_desc()=default
Default constructor. Produces an empty object.
memory::desc diff_dst_desc() const
Returns a diff destination memory descriptor.
Definition: dnnl.hpp:7406
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for an inner product weights update primitive from a C API primitiv...
Definition: dnnl.hpp:7393
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, const inner_product_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for an inner product weights update primitive.
Definition: dnnl.hpp:7380
memory::desc diff_bias_desc() const
Returns the diff bias memory descriptor.
Definition: dnnl.hpp:7409
primitive_desc(const desc &adesc, const engine &aengine, const inner_product_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for an inner product weights update primitive.
Definition: dnnl.hpp:7360
Inner product weights gradient primitive.
Definition: dnnl.hpp:7292
inner_product_backward_weights(const primitive_desc &pd)
Constructs an inner product weights gradient primitive.
Definition: dnnl.hpp:7420
inner_product_backward_weights()=default
Default constructor. Produces an empty object.
Descriptor for an inner product forward propagation primitive.
Definition: dnnl.hpp:7067
desc(prop_kind aprop_kind, const memory::desc &src_desc, const memory::desc &weights_desc, const memory::desc &dst_desc)
Constructs a descriptor for an inner product forward propagation primitive without bias.
Definition: dnnl.hpp:7108
desc(prop_kind aprop_kind, const memory::desc &src_desc, const memory::desc &weights_desc, const memory::desc &bias_desc, const memory::desc &dst_desc)
Constructs a descriptor for an inner product forward propagation primitive with bias.
Definition: dnnl.hpp:7084
Primitive descriptor for an inner product forward propagation primitive.
Definition: dnnl.hpp:7121
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for an inner product forward propagation primitive from a C API pri...
Definition: dnnl.hpp:7162
memory::desc dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:7174
primitive_desc(const desc &adesc, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for an inner product forward propagation primitive.
Definition: dnnl.hpp:7135
memory::desc weights_desc() const
Returns a weights memory descriptor.
Definition: dnnl.hpp:7171
primitive_desc()=default
Default constructor. Produces an empty object.
memory::desc bias_desc() const
Returns the bias memory descriptor.
Definition: dnnl.hpp:7177
memory::desc src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:7168
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for an inner product forward propagation primitive.
Definition: dnnl.hpp:7151
Inner product forward propagation primitive.
Definition: dnnl.hpp:7065
inner_product_forward(const primitive_desc &pd)
Constructs an inner product forward propagation primitive.
Definition: dnnl.hpp:7186
inner_product_forward()=default
Default constructor. Produces an empty object.
Descriptor for a layer normalization backward propagation primitive.
Definition: dnnl.hpp:6903
desc(prop_kind aprop_kind, const memory::desc &diff_data_desc, const memory::desc &data_desc, float epsilon, normalization_flags flags)
Constructs a descriptor for layer normalization backward propagation primitive.
Definition: dnnl.hpp:6943
desc(prop_kind aprop_kind, const memory::desc &diff_data_desc, const memory::desc &data_desc, const memory::desc &stat_desc, float epsilon, normalization_flags flags)
Constructs a descriptor for layer normalization backward propagation primitive.
Definition: dnnl.hpp:6919
Primitive descriptor for a layer normalization backward propagation primitive.
Definition: dnnl.hpp:6957
primitive_desc()=default
Default constructor. Produces an empty object.
memory::desc diff_src_desc() const
Returns a diff source memory descriptor.
Definition: dnnl.hpp:7023
memory::desc mean_desc() const
Returns memory descriptor for mean.
Definition: dnnl.hpp:7034
memory::desc workspace_desc() const
Returns the workspace memory descriptor.
Definition: dnnl.hpp:7042
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a layer normalization backward propagation primitive from a C A...
Definition: dnnl.hpp:7007
memory::desc diff_dst_desc() const
Returns a diff destination memory descriptor.
Definition: dnnl.hpp:7026
memory::desc dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:7020
memory::desc variance_desc() const
Returns memory descriptor for variance.
Definition: dnnl.hpp:7037
memory::desc weights_desc() const
Returns a weights memory descriptor.
Definition: dnnl.hpp:7017
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, const layer_normalization_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a layer normalization backward propagation primitive.
Definition: dnnl.hpp:6994
memory::desc diff_weights_desc() const
Returns a diff weights memory descriptor.
Definition: dnnl.hpp:7029
primitive_desc(const desc &adesc, const engine &aengine, const layer_normalization_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a layer normalization backward propagation primitive.
Definition: dnnl.hpp:6974
memory::desc src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:7014
Layer normalization backward propagation primitive.
Definition: dnnl.hpp:6901
layer_normalization_backward(const primitive_desc &pd)
Constructs a layer normalization backward propagation primitive.
Definition: dnnl.hpp:7051
layer_normalization_backward()=default
Default constructor. Produces an empty object.
Descriptor for a layer normalization forward propagation primitive.
Definition: dnnl.hpp:6758
desc(prop_kind aprop_kind, const memory::desc &data_desc, const memory::desc &stat_desc, float epsilon, normalization_flags flags)
Constructs a descriptor for layer normalization forward propagation primitive.
Definition: dnnl.hpp:6772
desc(prop_kind aprop_kind, const memory::desc &data_desc, float epsilon, normalization_flags flags)
Constructs a descriptor for layer normalization forward propagation primitive.
Definition: dnnl.hpp:6793
Primitive descriptor for a layer normalization forward propagation primitive.
Definition: dnnl.hpp:6806
primitive_desc(const desc &adesc, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a layer normalization forward propagation primitive.
Definition: dnnl.hpp:6820
memory::desc dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:6857
memory::desc src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:6854
memory::desc workspace_desc() const
Returns the workspace memory descriptor.
Definition: dnnl.hpp:6863
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a layer normalization forward propagation primitive from a C AP...
Definition: dnnl.hpp:6847
memory::desc variance_desc() const
Returns memory descriptor for variance.
Definition: dnnl.hpp:6869
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a layer normalization forward propagation primitive.
Definition: dnnl.hpp:6836
memory::desc weights_desc() const
Returns a weights memory descriptor.
Definition: dnnl.hpp:6860
primitive_desc()=default
Default constructor. Produces an empty object.
memory::desc mean_desc() const
Returns memory descriptor for mean.
Definition: dnnl.hpp:6866
Layer normalization forward propagation primitive.
Definition: dnnl.hpp:6756
layer_normalization_forward()=default
Default constructor. Produces an empty object.
layer_normalization_forward(const primitive_desc &pd)
Constructs a layer normalization forward propagation primitive.
Definition: dnnl.hpp:6897
Descriptor for a LBR GRU backward propagation primitive.
Definition: dnnl.hpp:9368
desc(prop_kind aprop_kind, rnn_direction direction, const memory::desc &src_layer_desc, const memory::desc &src_iter_desc, const memory::desc &weights_layer_desc, const memory::desc &weights_iter_desc, const memory::desc &bias_desc, const memory::desc &dst_layer_desc, const memory::desc &dst_iter_desc, const memory::desc &diff_src_layer_desc, const memory::desc &diff_src_iter_desc, const memory::desc &diff_weights_layer_desc, const memory::desc &diff_weights_iter_desc, const memory::desc &diff_bias_desc, const memory::desc &diff_dst_layer_desc, const memory::desc &diff_dst_iter_desc, rnn_flags flags=rnn_flags::undef)
Constructs a descriptor for LBR GRU backward propagation primitive.
Definition: dnnl.hpp:9416
Primitive descriptor for an LBR GRU backward propagation primitive.
Definition: dnnl.hpp:9450
memory::desc weights_layer_desc() const
Returns weights layer memory descriptor.
Definition: dnnl.hpp:9513
memory::desc diff_weights_layer_desc() const
Returns diff weights layer memory descriptor.
Definition: dnnl.hpp:9549
primitive_desc(const desc &adesc, const engine &aengine, const lbr_gru_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for an LBR GRU backward propagation primitive.
Definition: dnnl.hpp:9467
memory::desc diff_dst_iter_desc() const
Returns diff destination iteration memory descriptor.
Definition: dnnl.hpp:9569
memory::desc diff_bias_desc() const
Returns diff bias memory descriptor.
Definition: dnnl.hpp:9559
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, const lbr_gru_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for an LBR GRU backward propagation primitive.
Definition: dnnl.hpp:9487
primitive_desc()=default
Default constructor. Produces an empty object.
memory::desc dst_iter_desc() const
Returns destination iteration memory descriptor.
Definition: dnnl.hpp:9531
memory::desc weights_iter_desc() const
Returns weights iteration memory descriptor.
Definition: dnnl.hpp:9518
memory::desc src_iter_desc() const
Returns source iteration memory descriptor.
Definition: dnnl.hpp:9510
memory::desc diff_src_iter_desc() const
Returns diff source iteration memory descriptor.
Definition: dnnl.hpp:9544
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a LBR GRU backward propagation primitive from a C API primitive...
Definition: dnnl.hpp:9500
memory::desc workspace_desc() const
Returns the workspace memory descriptor.
Definition: dnnl.hpp:9534
memory::desc bias_desc() const
Returns bias memory descriptor.
Definition: dnnl.hpp:9523
memory::desc dst_layer_desc() const
Returns destination layer memory descriptor.
Definition: dnnl.hpp:9526
memory::desc src_layer_desc() const
Returns source layer memory descriptor.
Definition: dnnl.hpp:9505
memory::desc diff_weights_iter_desc() const
Returns diff weights iteration memory descriptor.
Definition: dnnl.hpp:9554
memory::desc diff_dst_layer_desc() const
Returns diff destination layer memory descriptor.
Definition: dnnl.hpp:9564
memory::desc diff_src_layer_desc() const
Returns diff source layer memory descriptor.
Definition: dnnl.hpp:9539
LBR GRU backward propagation primitive.
Definition: dnnl.hpp:9366
lbr_gru_backward(const primitive_desc &pd)
Constructs an LBR GRU backward propagation primitive.
Definition: dnnl.hpp:9580
lbr_gru_backward()=default
Default constructor. Produces an empty object.
Descriptor for an LBR GRU forward propagation primitive.
Definition: dnnl.hpp:9216
desc(prop_kind aprop_kind, rnn_direction direction, const memory::desc &src_layer_desc, const memory::desc &src_iter_desc, const memory::desc &weights_layer_desc, const memory::desc &weights_iter_desc, const memory::desc &bias_desc, const memory::desc &dst_layer_desc, const memory::desc &dst_iter_desc, rnn_flags flags=rnn_flags::undef)
Constructs a descriptor for LBR GRU forward propagation primitive.
Definition: dnnl.hpp:9252
Primitive descriptor for an LBR GRU forward propagation primitive.
Definition: dnnl.hpp:9275
memory::desc dst_iter_desc() const
Returns destination iteration memory descriptor.
Definition: dnnl.hpp:9348
memory::desc src_iter_desc() const
Returns source iteration memory descriptor.
Definition: dnnl.hpp:9327
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a LBR GRU forward propagation primitive.
Definition: dnnl.hpp:9305
memory::desc dst_layer_desc() const
Returns destination layer memory descriptor.
Definition: dnnl.hpp:9343
memory::desc workspace_desc() const
Returns the workspace memory descriptor.
Definition: dnnl.hpp:9351
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a LBR GRU forward propagation primitive from a C API primitive ...
Definition: dnnl.hpp:9316
primitive_desc(const desc &adesc, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a LBR GRU forward propagation primitive.
Definition: dnnl.hpp:9289
memory::desc bias_desc() const
Returns bias memory descriptor.
Definition: dnnl.hpp:9340
memory::desc src_layer_desc() const
Returns source layer memory descriptor.
Definition: dnnl.hpp:9322
primitive_desc()=default
Default constructor. Produces an empty object.
memory::desc weights_iter_desc() const
Returns weights iteration memory descriptor.
Definition: dnnl.hpp:9335
memory::desc weights_layer_desc() const
Returns weights layer memory descriptor.
Definition: dnnl.hpp:9330
LBR GRU forward propagation primitive.
Definition: dnnl.hpp:9214
lbr_gru_forward()=default
Default constructor. Produces an empty object.
lbr_gru_forward(const primitive_desc &pd)
Constructs an LBR GRU forward propagation primitive.
Definition: dnnl.hpp:9362
Descriptor for a logsoftmax backward propagation primitive.
Definition: dnnl.hpp:6350
desc()=default
Default constructor. Produces an empty object.
desc(const memory::desc &diff_data_desc, const memory::desc &data_desc, int logsoftmax_axis)
Constructs a descriptor for a logsoftmax backward propagation primitive.
Definition: dnnl.hpp:6363
Primitive descriptor for a logsoftmax backward propagation primitive.
Definition: dnnl.hpp:6374
memory::desc dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:6433
primitive_desc()=default
Default constructor. Produces an empty object.
memory::desc diff_dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:6439
memory::desc diff_src_desc() const
Returns a diff source memory descriptor.
Definition: dnnl.hpp:6436
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a logsoftmax backward propagation primitive from a C API primit...
Definition: dnnl.hpp:6424
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, const logsoftmax_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a logsoftmax backward propagation primitive.
Definition: dnnl.hpp:6411
primitive_desc(const desc &adesc, const engine &aengine, const logsoftmax_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a logsoftmax backward propagation primitive.
Definition: dnnl.hpp:6391
Logsoftmax backward propagation primitive.
Definition: dnnl.hpp:6348
logsoftmax_backward(const primitive_desc &pd)
Constructs a logsoftmax backward propagation primitive.
Definition: dnnl.hpp:6448
logsoftmax_backward()=default
Default constructor. Produces an empty object.
Descriptor for a logsoftmax forward propagation primitive.
Definition: dnnl.hpp:6256
desc(prop_kind aprop_kind, const memory::desc &data_desc, int logsoftmax_axis)
Constructs a descriptor for a logsoftmax forward propagation primitive.
Definition: dnnl.hpp:6270
desc()=default
Default constructor. Produces an empty object.
Primitive descriptor for a logsoftmax forward propagation primitive.
Definition: dnnl.hpp:6281
memory::desc dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:6335
memory::desc src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:6332
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a logsoftmax forward propagation primitive.
Definition: dnnl.hpp:6311
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a logsoftmax forward propagation primitive from a C API primiti...
Definition: dnnl.hpp:6322
primitive_desc(const desc &adesc, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a logsoftmax forward propagation primitive.
Definition: dnnl.hpp:6295
primitive_desc()=default
Default constructor. Produces an empty object.
Logsoftmax forward propagation primitive.
Definition: dnnl.hpp:6254
logsoftmax_forward()=default
Default constructor. Produces an empty object.
logsoftmax_forward(const primitive_desc &pd)
Constructs a logsoftmax forward propagation primitive.
Definition: dnnl.hpp:6344
Descriptor for an LRN backward propagation primitive.
Definition: dnnl.hpp:5486
desc(algorithm aalgorithm, const memory::desc &data_desc, const memory::desc &diff_data_desc, memory::dim local_size, float alpha, float beta, float k=1.f)
Constructs a descriptor for an LRN backward propagation primitive.
Definition: dnnl.hpp:5501
Primitive descriptor for an LRN backward propagation primitive.
Definition: dnnl.hpp:5514
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, const lrn_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for an LRN backward propagation primitive.
Definition: dnnl.hpp:5549
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for an LRN backward propagation primitive from a C API primitive de...
Definition: dnnl.hpp:5562
memory::desc diff_dst_desc() const
Returns a diff destination memory descriptor.
Definition: dnnl.hpp:5570
primitive_desc(const desc &adesc, const engine &aengine, const lrn_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for an LRN backward propagation primitive.
Definition: dnnl.hpp:5530
memory::desc workspace_desc() const
Returns the workspace memory descriptor.
Definition: dnnl.hpp:5573
memory::desc diff_src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:5567
primitive_desc()=default
Default constructor. Produces an empty object.
Local response normalization (LRN) backward propagation primitive.
Definition: dnnl.hpp:5484
lrn_backward(const primitive_desc &pd)
Constructs an LRN backward propagation primitive.
Definition: dnnl.hpp:5582
lrn_backward()=default
Default constructor. Produces an empty object.
Descriptor for an LRN forward propagation primitive.
Definition: dnnl.hpp:5391
desc(prop_kind aprop_kind, algorithm aalgorithm, const memory::desc &data_desc, memory::dim local_size, float alpha, float beta, float k=1.f)
Constructs a descriptor for a LRN forward propagation primitive.
Definition: dnnl.hpp:5407
Primitive descriptor for an LRN forward propagation primitive.
Definition: dnnl.hpp:5420
memory::desc src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:5465
memory::desc dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:5468
primitive_desc()=default
Default constructor. Produces an empty object.
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for an LRN forward propagation primitive.
Definition: dnnl.hpp:5448
memory::desc workspace_desc() const
Returns the workspace memory descriptor.
Definition: dnnl.hpp:5471
primitive_desc(const desc &adesc, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for an LRN forward propagation primitive.
Definition: dnnl.hpp:5433
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for an LRN forward propagation primitive from a C API primitive des...
Definition: dnnl.hpp:5459
Local response normalization (LRN) forward propagation primitive.
Definition: dnnl.hpp:5389
lrn_forward()=default
Default constructor. Produces an empty object.
lrn_forward(const primitive_desc &pd)
Constructs an LRN forward propagation primitive.
Definition: dnnl.hpp:5480
Descriptor for an LSTM backward propagation primitive.
Definition: dnnl.hpp:8348
desc(prop_kind aprop_kind, rnn_direction direction, const memory::desc &src_layer_desc, const memory::desc &src_iter_desc, const memory::desc &src_iter_c_desc, const memory::desc &weights_layer_desc, const memory::desc &weights_iter_desc, const memory::desc &weights_peephole_desc, const memory::desc &weights_projection_desc, const memory::desc &bias_desc, const memory::desc &dst_layer_desc, const memory::desc &dst_iter_desc, const memory::desc &dst_iter_c_desc, const memory::desc &diff_src_layer_desc, const memory::desc &diff_src_iter_desc, const memory::desc &diff_src_iter_c_desc, const memory::desc &diff_weights_layer_desc, const memory::desc &diff_weights_iter_desc, const memory::desc &diff_weights_peephole_desc, const memory::desc &diff_weights_projection_desc, const memory::desc &diff_bias_desc, const memory::desc &diff_dst_layer_desc, const memory::desc &diff_dst_iter_desc, const memory::desc &diff_dst_iter_c_desc, rnn_flags flags=rnn_flags::undef)
Constructs an LSTM (with or without peephole and with or without projection) descriptor for backward ...
Definition: dnnl.hpp:8426
desc(prop_kind aprop_kind, rnn_direction direction, const memory::desc &src_layer_desc, const memory::desc &src_iter_desc, const memory::desc &src_iter_c_desc, const memory::desc &weights_layer_desc, const memory::desc &weights_iter_desc, const memory::desc &bias_desc, const memory::desc &dst_layer_desc, const memory::desc &dst_iter_desc, const memory::desc &dst_iter_c_desc, const memory::desc &diff_src_layer_desc, const memory::desc &diff_src_iter_desc, const memory::desc &diff_src_iter_c_desc, const memory::desc &diff_weights_layer_desc, const memory::desc &diff_weights_iter_desc, const memory::desc &diff_bias_desc, const memory::desc &diff_dst_layer_desc, const memory::desc &diff_dst_iter_desc, const memory::desc &diff_dst_iter_c_desc, rnn_flags flags=rnn_flags::undef)
Constructs an LSTM descriptor for backward propagation using prop_kind, direction,...
Definition: dnnl.hpp:8637
desc(prop_kind aprop_kind, rnn_direction direction, const memory::desc &src_layer_desc, const memory::desc &src_iter_desc, const memory::desc &src_iter_c_desc, const memory::desc &weights_layer_desc, const memory::desc &weights_iter_desc, const memory::desc &weights_peephole_desc, const memory::desc &bias_desc, const memory::desc &dst_layer_desc, const memory::desc &dst_iter_desc, const memory::desc &dst_iter_c_desc, const memory::desc &diff_src_layer_desc, const memory::desc &diff_src_iter_desc, const memory::desc &diff_src_iter_c_desc, const memory::desc &diff_weights_layer_desc, const memory::desc &diff_weights_iter_desc, const memory::desc &diff_weights_peephole_desc, const memory::desc &diff_bias_desc, const memory::desc &diff_dst_layer_desc, const memory::desc &diff_dst_iter_desc, const memory::desc &diff_dst_iter_c_desc, rnn_flags flags=rnn_flags::undef)
Constructs an LSTM (with or without peephole) descriptor for backward propagation using prop_kind,...
Definition: dnnl.hpp:8538
Primitive descriptor for an LSTM backward propagation primitive.
Definition: dnnl.hpp:8678
memory::desc weights_iter_desc() const
Returns weights iteration memory descriptor.
Definition: dnnl.hpp:8749
memory::desc diff_dst_iter_desc() const
Returns diff destination iteration memory descriptor.
Definition: dnnl.hpp:8830
memory::desc diff_weights_projection_desc() const
Returns diff weights projection memory descriptor.
Definition: dnnl.hpp:8815
memory::desc weights_peephole_desc() const
Returns weights peephole memory descriptor.
Definition: dnnl.hpp:8754
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for an LSTM backward propagation primitive from a C API primitive d...
Definition: dnnl.hpp:8726
memory::desc diff_weights_peephole_desc() const
Returns diff weights peephole memory descriptor.
Definition: dnnl.hpp:8810
memory::desc dst_iter_c_desc() const
Returns source iteration memory descriptor.
Definition: dnnl.hpp:8775
memory::desc src_layer_desc() const
Returns source layer memory descriptor.
Definition: dnnl.hpp:8731
memory::desc dst_iter_desc() const
Returns destination iteration memory descriptor.
Definition: dnnl.hpp:8772
primitive_desc()=default
Default constructor. Produces an empty object.
primitive_desc(const desc &adesc, const engine &aengine, const lstm_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for an LSTM backward propagation primitive.
Definition: dnnl.hpp:8694
memory::desc diff_src_layer_desc() const
Returns diff source layer memory descriptor.
Definition: dnnl.hpp:8785
memory::desc src_iter_desc() const
Returns source iteration memory descriptor.
Definition: dnnl.hpp:8736
memory::desc diff_weights_iter_desc() const
Returns diff weights iteration memory descriptor.
Definition: dnnl.hpp:8805
memory::desc weights_projection_desc() const
Returns weights projection memory descriptor.
Definition: dnnl.hpp:8759
memory::desc diff_bias_desc() const
Returns diff bias memory descriptor.
Definition: dnnl.hpp:8820
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, const lstm_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for an LSTM backward propagation primitive.
Definition: dnnl.hpp:8713
memory::desc bias_desc() const
Returns bias memory descriptor.
Definition: dnnl.hpp:8764
memory::desc src_iter_c_desc() const
Returns source iteration memory descriptor.
Definition: dnnl.hpp:8739
memory::desc dst_layer_desc() const
Returns destination layer memory descriptor.
Definition: dnnl.hpp:8767
memory::desc diff_dst_iter_c_desc() const
Returns diff destination recurrent cell state memory descriptor.
Definition: dnnl.hpp:8835
memory::desc diff_src_iter_desc() const
Returns diff source iteration memory descriptor.
Definition: dnnl.hpp:8790
memory::desc diff_dst_layer_desc() const
Returns diff destination layer memory descriptor.
Definition: dnnl.hpp:8825
memory::desc weights_layer_desc() const
Returns weights layer memory descriptor.
Definition: dnnl.hpp:8744
memory::desc diff_weights_layer_desc() const
Returns diff weights layer memory descriptor.
Definition: dnnl.hpp:8800
memory::desc workspace_desc() const
Returns the workspace memory descriptor.
Definition: dnnl.hpp:8780
memory::desc diff_src_iter_c_desc() const
Returns diff source recurrent cell state memory descriptor.
Definition: dnnl.hpp:8795
LSTM backward propagation primitive.
Definition: dnnl.hpp:8346
lstm_backward()=default
Default constructor. Produces an empty object.
lstm_backward(const primitive_desc &pd)
Constructs an LSTM backward propagation primitive.
Definition: dnnl.hpp:8846
Descriptor for an LSTM forward propagation primitive.
Definition: dnnl.hpp:8031
desc(prop_kind aprop_kind, rnn_direction direction, const memory::desc &src_layer_desc, const memory::desc &src_iter_desc, const memory::desc &src_iter_c_desc, const memory::desc &weights_layer_desc, const memory::desc &weights_iter_desc, const memory::desc &bias_desc, const memory::desc &dst_layer_desc, const memory::desc &dst_iter_desc, const memory::desc &dst_iter_c_desc, rnn_flags flags=rnn_flags::undef)
Constructs a descriptor for an LSTM forward propagation primitive.
Definition: dnnl.hpp:8211
desc(prop_kind aprop_kind, rnn_direction direction, const memory::desc &src_layer_desc, const memory::desc &src_iter_desc, const memory::desc &src_iter_c_desc, const memory::desc &weights_layer_desc, const memory::desc &weights_iter_desc, const memory::desc &weights_peephole_desc, const memory::desc &weights_projection_desc, const memory::desc &bias_desc, const memory::desc &dst_layer_desc, const memory::desc &dst_iter_desc, const memory::desc &dst_iter_c_desc, rnn_flags flags=rnn_flags::undef)
Constructs a descriptor for an LSTM (with or without peephole and with or without projection) forward...
Definition: dnnl.hpp:8082
desc(prop_kind aprop_kind, rnn_direction direction, const memory::desc &src_layer_desc, const memory::desc &src_iter_desc, const memory::desc &src_iter_c_desc, const memory::desc &weights_layer_desc, const memory::desc &weights_iter_desc, const memory::desc &weights_peephole_desc, const memory::desc &bias_desc, const memory::desc &dst_layer_desc, const memory::desc &dst_iter_desc, const memory::desc &dst_iter_c_desc, rnn_flags flags=rnn_flags::undef)
Constructs a descriptor for an LSTM (with or without peephole) forward propagation primitive.
Definition: dnnl.hpp:8150
Primitive descriptor for an LSTM forward propagation primitive.
Definition: dnnl.hpp:8237
memory::desc dst_iter_desc() const
Returns destination iteration memory descriptor.
Definition: dnnl.hpp:8323
memory::desc weights_peephole_desc() const
Returns weights peephole memory descriptor.
Definition: dnnl.hpp:8305
memory::desc weights_iter_desc() const
Returns weights iteration memory descriptor.
Definition: dnnl.hpp:8300
memory::desc dst_layer_desc() const
Returns destination layer memory descriptor.
Definition: dnnl.hpp:8318
primitive_desc()=default
Default constructor. Produces an empty object.
memory::desc workspace_desc() const
Returns the workspace memory descriptor.
Definition: dnnl.hpp:8331
primitive_desc(const desc &adesc, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for an LSTM forward propagation primitive.
Definition: dnnl.hpp:8250
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for an LSTM forward propagation primitive from a C API primitive de...
Definition: dnnl.hpp:8276
memory::desc dst_iter_c_desc() const
Returns source iteration memory descriptor.
Definition: dnnl.hpp:8326
memory::desc weights_layer_desc() const
Returns weights layer memory descriptor.
Definition: dnnl.hpp:8295
memory::desc weights_projection_desc() const
Returns weights projection memory descriptor.
Definition: dnnl.hpp:8310
memory::desc src_iter_c_desc() const
Returns source iteration memory descriptor.
Definition: dnnl.hpp:8290
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for an LSTM forward propagation primitive.
Definition: dnnl.hpp:8265
memory::desc src_iter_desc() const
Returns source iteration memory descriptor.
Definition: dnnl.hpp:8287
memory::desc bias_desc() const
Returns bias memory descriptor.
Definition: dnnl.hpp:8315
memory::desc src_layer_desc() const
Returns source layer memory descriptor.
Definition: dnnl.hpp:8282
LSTM forward propagation primitive.
Definition: dnnl.hpp:8029
lstm_forward(const primitive_desc &pd)
Constructs an LSTM forward propagation primitive.
Definition: dnnl.hpp:8342
lstm_forward()=default
Default constructor. Produces an empty object.
Descriptor for a matmul primitive.
Definition: dnnl.hpp:9856
desc(const memory::desc &src_desc, const memory::desc &weights_desc, const memory::desc &dst_desc)
Constructs a descriptor for a matmul primitive.
Definition: dnnl.hpp:9864
desc(const memory::desc &src_desc, const memory::desc &weights_desc, const memory::desc &bias_desc, const memory::desc &dst_desc)
Constructs a descriptor for a matmul primitive.
Definition: dnnl.hpp:9878
Primitive descriptor for a matmul primitive.
Definition: dnnl.hpp:9888
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a matmul primitive.
Definition: dnnl.hpp:9914
memory::desc weights_desc() const
Returns a weights memory descriptor.
Definition: dnnl.hpp:9930
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a matmul primitive from a C API primitive descriptor that must ...
Definition: dnnl.hpp:9923
memory::desc bias_desc() const
Returns the bias memory descriptor.
Definition: dnnl.hpp:9935
primitive_desc()=default
Default constructor. Produces an empty object.
memory::desc src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:9927
memory::desc dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:9940
primitive_desc(const desc &adesc, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a matmul primitive.
Definition: dnnl.hpp:9900
Matrix multiplication (matmul) primitive.
Definition: dnnl.hpp:9854
matmul(const primitive_desc &pd)
Constructs a matmul primitive.
Definition: dnnl.hpp:9948
matmul()=default
Default constructor. Produces an empty object.
A memory descriptor.
Definition: dnnl.hpp:1984
desc(const dims &adims, data_type adata_type, format_tag aformat_tag, bool allow_empty=false)
Constructs a memory descriptor.
Definition: dnnl.hpp:2008
desc()
Constructs a zero (empty) memory descriptor.
Definition: dnnl.hpp:1991
bool operator!=(const desc &other) const
An inequality operator.
Definition: dnnl.hpp:2219
desc permute_axes(const std::vector< int > &permutation, bool allow_empty=false) const
Constructs a memory descriptor by permuting axes in an existing one.
Definition: dnnl.hpp:2170
desc submemory_desc(const dims &adims, const dims &offsets, bool allow_empty=false) const
Constructs a memory descriptor for a region inside an area described by this memory descriptor.
Definition: dnnl.hpp:2066
bool operator==(const desc &other) const
An equality operator.
Definition: dnnl.hpp:2211
bool is_zero() const
Checks whether the memory descriptor is zero (empty).
Definition: dnnl.hpp:2205
memory::dims dims() const
Returns dimensions of the memory descriptor.
Definition: dnnl.hpp:2186
memory::data_type data_type() const
Returns the data type of the memory descriptor.
Definition: dnnl.hpp:2192
desc reshape(const dims &adims, bool allow_empty=false) const
Constructs a memory descriptor by reshaping an existing one.
Definition: dnnl.hpp:2122
desc(const dims &adims, data_type adata_type, const dims &strides, bool allow_empty=false)
Constructs a memory descriptor by strides.
Definition: dnnl.hpp:2036
size_t get_size() const
Returns size of the memory descriptor in bytes.
Definition: dnnl.hpp:2200
desc(const dnnl_memory_desc_t &data)
Constructs a memory descriptor from a C API data structure.
Definition: dnnl.hpp:2053
dnnl_memory_desc_t data
The underlying C API data structure.
Definition: dnnl.hpp:1987
Memory object.
Definition: dnnl.hpp:1108
void unmap_data(void *mapped_ptr) const
Unmaps a memory object and writes back any changes made to the previously mapped memory buffer.
Definition: dnnl.hpp:2385
static void validate_dims(const std::vector< T > &v, int min_size=0)
Helper function that validates that an std::vector of dimensions can be safely converted to the C API...
Definition: dnnl.hpp:1124
T * map_data() const
Maps a memory object and returns a host-side pointer to a memory buffer with a copy of its contents.
Definition: dnnl.hpp:2368
memory()=default
Default constructor.
dnnl_dim_t dim
Integer type for representing dimension sizes and indices.
Definition: dnnl.hpp:1112
memory(const desc &md, const engine &aengine, void *handle)
Constructs a memory object.
Definition: dnnl.hpp:2252
void set_data_handle(void *handle, const stream &astream) const
Sets the underlying memory buffer.
Definition: dnnl.hpp:2324
format_tag
Memory format tag specification.
Definition: dnnl.hpp:1205
data_type
Data type specification.
Definition: dnnl.hpp:1130
@ undef
Undefined data type (used for empty memory descriptors).
engine get_engine() const
Returns the associated engine.
Definition: dnnl.hpp:2278
format_kind
Memory format kind.
Definition: dnnl.hpp:1149
memory(const desc &md, const engine &aengine)
Constructs a memory object.
Definition: dnnl.hpp:2266
void set_data_handle(void *handle) const
Sets the underlying memory buffer.
Definition: dnnl.hpp:2340
desc get_desc() const
Returns the associated memory descriptor.
Definition: dnnl.hpp:2270
void * get_data_handle() const
Returns the underlying memory buffer.
Definition: dnnl.hpp:2289
std::vector< dim > dims
Vector of dimensions.
Definition: dnnl.hpp:1115
Descriptor for a pooling backward propagation primitive.
Definition: dnnl.hpp:5710
desc(algorithm aalgorithm, const memory::desc &diff_src_desc, const memory::desc &diff_dst_desc, const memory::dims &strides, const memory::dims &kernel, const memory::dims &padding_l, const memory::dims &padding_r)
Constructs a descriptor for pooling backward propagation primitive.
Definition: dnnl.hpp:5734
Primitive descriptor for a pooling backward propagation primitive.
Definition: dnnl.hpp:5753
primitive_desc(const desc &adesc, const engine &aengine, const pooling_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a pooling backward propagation primitive.
Definition: dnnl.hpp:5769
memory::desc diff_dst_desc() const
Returns a diff destination memory descriptor.
Definition: dnnl.hpp:5809
memory::desc workspace_desc() const
Returns the workspace memory descriptor.
Definition: dnnl.hpp:5812
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, const pooling_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a pooling backward propagation primitive.
Definition: dnnl.hpp:5788
primitive_desc()=default
Default constructor. Produces an empty object.
memory::desc diff_src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:5806
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a pooling backward propagation primitive from a C API primitive...
Definition: dnnl.hpp:5801
Pooling backward propagation primitive.
Definition: dnnl.hpp:5708
pooling_backward()=default
Default constructor. Produces an empty object.
pooling_backward(const primitive_desc &pd)
Constructs a pooling backward propagation primitive.
Definition: dnnl.hpp:5821
Descriptor for a pooling forward propagation primitive.
Definition: dnnl.hpp:5598
desc(prop_kind aprop_kind, algorithm aalgorithm, const memory::desc &src_desc, const memory::desc &dst_desc, const memory::dims &strides, const memory::dims &kernel, const memory::dims &padding_l, const memory::dims &padding_r)
Constructs a descriptor for pooling forward propagation primitive.
Definition: dnnl.hpp:5625
Primitive descriptor for a pooling forward propagation primitive.
Definition: dnnl.hpp:5644
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a pooling forward propagation primitive.
Definition: dnnl.hpp:5672
memory::desc dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:5692
primitive_desc()=default
Default constructor. Produces an empty object.
memory::desc src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:5689
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a pooling forward propagation primitive from a C API primitive ...
Definition: dnnl.hpp:5683
memory::desc workspace_desc() const
Returns the workspace memory descriptor.
Definition: dnnl.hpp:5695
primitive_desc(const desc &adesc, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a pooling forward propagation primitive.
Definition: dnnl.hpp:5657
Pooling forward propagation primitive.
Definition: dnnl.hpp:5596
pooling_forward(const primitive_desc &pd)
Constructs a pooling forward propagation primitive.
Definition: dnnl.hpp:5704
pooling_forward()=default
Default constructor. Produces an empty object.
Descriptor for a pooling backward propagation primitive.
Definition: dnnl.hpp:10355
desc(algorithm aalgorithm, const memory::desc &diff_src_desc, const memory::desc &diff_dst_desc, const memory::dims &strides, const memory::dims &kernel, const memory::dims &dilation, const memory::dims &padding_l, const memory::dims &padding_r)
Constructs a descriptor for pooling v2 (dilated pooling) backward propagation primitive.
Definition: dnnl.hpp:10381
Primitive descriptor for a pooling v2 (dilated pooling) backward propagation primitive.
Definition: dnnl.hpp:10402
memory::desc diff_dst_desc() const
Returns a diff destination memory descriptor.
Definition: dnnl.hpp:10461
memory::desc diff_src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:10458
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a pooling v2 (dilated pooling) backward propagation primitive f...
Definition: dnnl.hpp:10453
primitive_desc()=default
Default constructor. Produces an empty object.
primitive_desc(const desc &adesc, const engine &aengine, const pooling_v2_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a pooling v2 (dilated pooling) backward propagation primitive.
Definition: dnnl.hpp:10419
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, const pooling_v2_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a pooling v2 (dilated pooling) backward propagation primitive.
Definition: dnnl.hpp:10439
memory::desc workspace_desc() const
Returns the workspace memory descriptor.
Definition: dnnl.hpp:10464
Pooling v2 (dilated pooling) backward propagation primitive.
Definition: dnnl.hpp:10353
pooling_v2_backward(const primitive_desc &pd)
Constructs a pooling v2 (dilated pooling) backward propagation primitive.
Definition: dnnl.hpp:10474
pooling_v2_backward()=default
Default constructor. Produces an empty object.
Descriptor for a pooling forward propagation primitive.
Definition: dnnl.hpp:10234
desc(prop_kind aprop_kind, algorithm aalgorithm, const memory::desc &src_desc, const memory::desc &dst_desc, const memory::dims &strides, const memory::dims &kernel, const memory::dims &dilation, const memory::dims &padding_l, const memory::dims &padding_r)
Constructs a descriptor for pooling v2 (dilated pooling) forward propagation primitive.
Definition: dnnl.hpp:10263
Primitive descriptor for a pooling forward propagation primitive.
Definition: dnnl.hpp:10285
memory::desc workspace_desc() const
Returns the workspace memory descriptor.
Definition: dnnl.hpp:10339
memory::desc dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:10336
memory::desc src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:10333
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a pooling v2 (dilated pooling) forward propagation primitive.
Definition: dnnl.hpp:10315
primitive_desc()=default
Default constructor. Produces an empty object.
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a pooling v2 (dilated pooling) forward propagation primitive fr...
Definition: dnnl.hpp:10327
primitive_desc(const desc &adesc, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a pooling v2 (dilated pooling) forward propagation primitive.
Definition: dnnl.hpp:10299
Pooling v2 (dilated pooling) forward propagation primitive.
Definition: dnnl.hpp:10232
pooling_v2_forward()=default
Default constructor. Produces an empty object.
pooling_v2_forward(const primitive_desc &pd)
Constructs a pooling v2 (dilated pooling) forward propagation primitive.
Definition: dnnl.hpp:10349
Post-ops.
Definition: dnnl.hpp:2450
void get_params_dw_k3s1p1(int index, memory::data_type &weights_data_type, memory::data_type &bias_data_type, memory::data_type &dst_data_type, int &mask, std::vector< float > &scales) const
Returns the parameters of an depthwise post-op with stride 1.
Definition: dnnl.hpp:2626
void get_params_binary(int index, algorithm &aalgorithm, memory::desc &src1_desc) const
Returns the parameters of a binary post-op.
Definition: dnnl.hpp:2762
void get_params_sum(int index, float &scale, memory::data_type &data_type) const
Returns the parameters of an accumulation (sum) post-op.
Definition: dnnl.hpp:2527
void append_eltwise(float scale, algorithm aalgorithm, float alpha, float beta)
Appends an elementwise post-op.
Definition: dnnl.hpp:2549
void append_binary(algorithm aalgorithm, const memory::desc &src1_desc)
Appends a binary post-op.
Definition: dnnl.hpp:2751
void append_dw_k3s1p1(memory::data_type weights_data_type, memory::data_type bias_data_type, memory::data_type dst_data_type, int mask, const std::vector< float > &scales)
Appends a depthwise post-op convolution with stride 1.
Definition: dnnl.hpp:2600
primitive::kind kind(int index) const
Returns the primitive kind of post-op at entry with a certain index.
Definition: dnnl.hpp:2467
int len() const
Returns the number of post-ops entries.
Definition: dnnl.hpp:2462
void append_dw_k3s2p1(memory::data_type weights_data_type, memory::data_type bias_data_type, memory::data_type dst_data_type, int mask, const std::vector< float > &scales)
Appends a depthwise post-op convolution with stride 2.
Definition: dnnl.hpp:2685
post_ops()
Constructs an empty sequence of post-ops.
Definition: dnnl.hpp:2454
void get_params_dw_k3s2p1(int index, memory::data_type &weights_data_type, memory::data_type &bias_data_type, memory::data_type &dst_data_type, int &mask, std::vector< float > &scales) const
Returns the parameters of an depthwise post-op with stride 2.
Definition: dnnl.hpp:2711
void get_params_eltwise(int index, float &scale, algorithm &aalgorithm, float &alpha, float &beta) const
Returns parameters of an elementwise post-op.
Definition: dnnl.hpp:2563
void get_params_sum(int index, float &scale) const
Returns the parameters of an accumulation (sum) post-op.
Definition: dnnl.hpp:2517
void append_sum(float scale=1.f, memory::data_type data_type=memory::data_type::undef)
Appends an accumulation (sum) post-op.
Definition: dnnl.hpp:2502
Descriptor for a PReLU backward propagation primitive.
Definition: dnnl.hpp:10578
desc(const memory::desc &data_desc, const memory::desc &weight_desc, const memory::desc &diff_data_desc, const memory::desc &diff_weights_desc)
Constructs a descriptor for a PReLU backward propagation primitive.
Definition: dnnl.hpp:10589
Primitive descriptor for prelu backward propagation.
Definition: dnnl.hpp:10602
memory::desc src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:10657
memory::desc diff_src_desc() const
Returns a diff source memory descriptor.
Definition: dnnl.hpp:10660
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, const prelu_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a PReLU backward propagation primitive.
Definition: dnnl.hpp:10639
primitive_desc(const desc &adesc, const engine &aengine, const prelu_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a PReLU backward propagation primitive.
Definition: dnnl.hpp:10619
memory::desc diff_dst_desc() const
Returns a diff destination memory descriptor.
Definition: dnnl.hpp:10663
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a prelu backward propagation primitive from a C API primitive d...
Definition: dnnl.hpp:10652
primitive_desc()=default
Default constructor. Produces an empty object.
PReLU backward propagation primitive.
Definition: dnnl.hpp:10576
prelu_backward()=default
Default constructor. Produces an empty object.
prelu_backward(const primitive_desc &pd)
Constructs a prelu backward propagation primitive.
Definition: dnnl.hpp:10672
Descriptor for a PReLU forward propagation primitive.
Definition: dnnl.hpp:10491
desc(prop_kind aprop_kind, const memory::desc &data_desc, const memory::desc &weight_desc)
Constructs a descriptor for a PReLU forward propagation primitive.
Definition: dnnl.hpp:10502
Primitive descriptor for a PReLU forward propagation primitive.
Definition: dnnl.hpp:10513
primitive_desc()=default
Default constructor. Produces an empty object.
memory::desc dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:10563
memory::desc src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:10560
primitive_desc(const desc &adesc, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a PReLU forward propagation primitive.
Definition: dnnl.hpp:10527
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a PReLU forward propagation primitive.
Definition: dnnl.hpp:10543
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a prelu forward propagation primitive from a C API primitive de...
Definition: dnnl.hpp:10554
PReLU forward propagation primitive.
Definition: dnnl.hpp:10489
prelu_forward(const primitive_desc &pd)
Constructs a prelu forward propagation primitive.
Definition: dnnl.hpp:10572
prelu_forward()=default
Default constructor. Produces an empty object.
Primitive attributes.
Definition: dnnl.hpp:2786
void get_zero_points(int arg, int &mask, std::vector< int32_t > &zero_points) const
Returns zero points correspondence mask and values.
Definition: dnnl.hpp:2953
const post_ops get_post_ops() const
Returns post-ops previously set via set_post_ops().
Definition: dnnl.hpp:2999
void set_rnn_data_qparams(float scale, float shift)
Sets quantization scale and shift parameters for RNN data tensors.
Definition: dnnl.hpp:3054
void get_rnn_weights_qparams(int &mask, std::vector< float > &scales)
Returns the quantization scaling factors for RNN projection weights tensors.
Definition: dnnl.hpp:3132
void get_rnn_data_qparams(float &scale, float &shift)
Returns the quantization scale and shift parameters for RNN data tensors.
Definition: dnnl.hpp:3070
void set_output_scales(int mask, const std::vector< float > &scales)
Sets output scaling factors correspondence mask and values.
Definition: dnnl.hpp:2888
void get_rnn_weights_projection_qparams(int &mask, std::vector< float > &scales)
Returns the quantization scaling factors for RNN projection weights tensors.
Definition: dnnl.hpp:3201
void set_rnn_weights_qparams(int mask, const std::vector< float > &scales)
Sets quantization scaling factors for RNN weights tensors.
Definition: dnnl.hpp:3106
void set_rnn_weights_projection_qparams(int mask, const std::vector< float > &scales)
Sets quantization scaling factors for RNN projection weights tensors.
Definition: dnnl.hpp:3173
void set_scratchpad_mode(scratchpad_mode mode)
Sets scratchpad mode.
Definition: dnnl.hpp:2817
void set_scales(int arg, int mask, const std::vector< float > &scales)
Sets scaling factors for primitive operations for a given memory argument.
Definition: dnnl.hpp:2936
void get_scales(int arg, int &mask, std::vector< float > &scales) const
Returns scaling factors correspondence mask and values for a given memory argument.
Definition: dnnl.hpp:2906
void get_output_scales(int &mask, std::vector< float > &scales) const
Returns output scaling factors correspondence mask and values.
Definition: dnnl.hpp:2832
primitive_attr(dnnl_primitive_attr_t attr)
Creates primitive attributes from a C API dnnl_primitive_attr_t handle.
Definition: dnnl.hpp:2802
void set_post_ops(const post_ops ops)
Sets post-ops.
Definition: dnnl.hpp:3016
primitive_attr()
Constructs default (empty) primitive attributes.
Definition: dnnl.hpp:2790
void set_zero_points(int arg, int mask, const std::vector< int32_t > &zero_points)
Sets zero points for primitive operations for a given memory argument.
Definition: dnnl.hpp:2988
scratchpad_mode get_scratchpad_mode() const
Returns the scratchpad mode.
Definition: dnnl.hpp:2806
Base class for all primitive descriptors.
Definition: dnnl.hpp:3225
primitive_attr get_primitive_attr() const
Returns the primitive attributes.
Definition: dnnl.hpp:3409
memory::desc diff_weights_desc(int idx) const
Returns a diff weights memory descriptor.
Definition: dnnl.hpp:3335
primitive_desc_base()=default
Default constructor. Produces an empty object.
engine get_engine() const
Returns the engine of the primitive descriptor.
Definition: dnnl.hpp:3233
memory::desc query_md(query what, int idx=0) const
Returns a memory descriptor.
Definition: dnnl.hpp:3270
memory::desc dst_desc(int idx) const
Returns a destination memory descriptor.
Definition: dnnl.hpp:3299
memory::desc diff_dst_desc(int idx) const
Returns a diff destination memory descriptor.
Definition: dnnl.hpp:3326
memory::desc scratchpad_desc() const
Returns the scratchpad memory descriptor.
Definition: dnnl.hpp:3391
void reset_with_clone(const_dnnl_primitive_desc_t pd)
Resets the value of the handle to a clone of a C API primitive descriptor.
Definition: dnnl.hpp:3433
dnnl::primitive::kind get_kind() const
Returns the kind of the primitive descriptor.
Definition: dnnl.hpp:3421
memory::desc diff_dst_desc() const
Returns a diff destination memory descriptor.
Definition: dnnl.hpp:3370
memory::desc diff_src_desc(int idx) const
Returns a diff source memory descriptor.
Definition: dnnl.hpp:3317
memory::desc weights_desc() const
Returns a weights memory descriptor.
Definition: dnnl.hpp:3358
const char * impl_info_str() const
Returns implementation name.
Definition: dnnl.hpp:3237
primitive_desc_base(dnnl_primitive_desc_t pd, dnnl::primitive::kind prim_kind, dnnl::prop_kind prop_kind1, dnnl::prop_kind prop_kind2)
Constructs a primitive descriptor base object from a clone of a C API primitive descriptor after veri...
Definition: dnnl.hpp:3485
primitive_desc_base(dnnl_primitive_desc_t pd, dnnl::primitive::kind prim_kind)
Constructs a primitive descriptor base object from a clone of a C API primitive descriptor after veri...
Definition: dnnl.hpp:3453
memory::desc diff_src_desc() const
Returns a diff source memory descriptor.
Definition: dnnl.hpp:3364
memory::desc weights_desc(int idx) const
Returns a weights memory descriptor.
Definition: dnnl.hpp:3308
memory::dim query_s64(query what) const
Returns a memory::dim value (same as int64_t).
Definition: dnnl.hpp:3249
memory::desc workspace_desc() const
Returns the workspace memory descriptor.
Definition: dnnl.hpp:3382
engine scratchpad_engine() const
Returns the engine on which the scratchpad memory is located.
Definition: dnnl.hpp:3397
memory::desc dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:3352
memory::desc src_desc(int idx) const
Returns a source memory descriptor.
Definition: dnnl.hpp:3290
memory::desc src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:3346
primitive_desc_base(dnnl_primitive_desc_t pd, dnnl::primitive::kind prim_kind, dnnl::prop_kind aprop_kind)
Constructs a primitive descriptor base object from a clone of a C API primitive descriptor after veri...
Definition: dnnl.hpp:3468
memory::desc diff_weights_desc() const
Returns a diff weights memory descriptor.
Definition: dnnl.hpp:3376
A base class for descriptors of all primitives that have an operation descriptor and that support ite...
Definition: dnnl.hpp:3879
primitive_desc(const_dnnl_op_desc_t desc, const primitive_attr *attr, const engine &aengine, const_dnnl_primitive_desc_t hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor.
Definition: dnnl.hpp:3906
bool next_impl()
Advances the primitive iterator to the next implementation.
Definition: dnnl.hpp:3924
Base class for all computational primitives.
Definition: dnnl.hpp:269
void execute(const stream &astream, const std::unordered_map< int, memory > &args) const
Executes computations specified by the primitive in a specified stream.
primitive()=default
Default constructor. Constructs an empty object.
primitive(const primitive_desc &pd)
Constructs a primitive from a primitive descriptor.
kind
Kinds of primitives supported by the library.
Definition: dnnl.hpp:271
primitive(const_dnnl_primitive_desc_t c_pd)
Constructs a primitive from a C API primitive descriptor.
Descriptor for reduction.
Definition: dnnl.hpp:10689
desc()=default
Default constructor. Produces an empty object.
desc(algorithm aalgorithm, const memory::desc &src_desc, const memory::desc &dst_desc, float p, float eps)
Constructs a descriptor for a reduction primitive using algorithm specific parameters,...
Definition: dnnl.hpp:10712
Primitive descriptor for a reduction primitive.
Definition: dnnl.hpp:10722
primitive_desc()=default
Default constructor. Produces an empty object.
memory::desc src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:10761
memory::desc dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:10764
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a reduction primitive from a C API primitive descriptor that mu...
Definition: dnnl.hpp:10757
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a reduction primitive.
Definition: dnnl.hpp:10748
primitive_desc(const desc &adesc, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a reduction primitive.
Definition: dnnl.hpp:10734
Reduction.
Definition: dnnl.hpp:10687
reduction(const primitive_desc &pd)
Constructs a reduction primitive.
Definition: dnnl.hpp:10772
reduction()=default
Default constructor. Produces an empty object.
Primitive descriptor for a reorder primitive.
Definition: dnnl.hpp:3549
memory::desc src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:3634
primitive_desc(const engine &src_engine, const memory::desc &src_md, const engine &dst_engine, const memory::desc &dst_md, const primitive_attr &attr=primitive_attr(), bool allow_empty=false)
Constructs a primitive descriptor for reorder primitive.
Definition: dnnl.hpp:3572
primitive_desc(const memory &src, const memory &dst, const primitive_attr &attr=primitive_attr(), bool allow_empty=false)
Constructs a primitive descriptor for reorder primitive.
Definition: dnnl.hpp:3598
engine get_src_engine() const
Returns the engine on which the source memory is allocated.
Definition: dnnl.hpp:3623
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for reorder primitive from a C API primitive descriptor which must ...
Definition: dnnl.hpp:3618
engine get_dst_engine() const
Returns the engine on which the destination memory is allocated.
Definition: dnnl.hpp:3629
primitive_desc()=default
Default constructor. Produces an empty object.
memory::desc dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:3637
Reorder primitive.
Definition: dnnl.hpp:3547
reorder(const primitive_desc &pd)
Constructs a reorder primitive.
Definition: dnnl.hpp:3645
void execute(const stream &astream, memory &src, memory &dst) const
Executes the reorder primitive.
Definition: dnnl.hpp:3666
reorder()=default
Default constructor. Produces an empty object.
reorder(const memory &src, const memory &dst, const primitive_attr &attr=primitive_attr())
Constructs a reorder primitive that would reorder data between memory objects having the same memory ...
Definition: dnnl.hpp:3654
Descriptor for a resampling backward propagation primitive.
Definition: dnnl.hpp:10110
desc(algorithm aalgorithm, const memory::desc &diff_src_desc, const memory::desc &diff_dst_desc)
Constructs a descriptor for a resampling backward propagation primitive using source and destination ...
Definition: dnnl.hpp:10121
desc(algorithm aalgorithm, const std::vector< float > &factors, const memory::desc &diff_src_desc, const memory::desc &diff_dst_desc)
Constructs a descriptor for resampling backward propagation primitive.
Definition: dnnl.hpp:10138
Primitive descriptor for resampling backward propagation primitive.
Definition: dnnl.hpp:10151
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a resampling backward propagation primitive from a C API primit...
Definition: dnnl.hpp:10201
memory::desc diff_src_desc() const
Returns a diff source memory descriptor.
Definition: dnnl.hpp:10206
primitive_desc(const desc &adesc, const engine &aengine, const resampling_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a resampling backward propagation primitive.
Definition: dnnl.hpp:10168
primitive_desc()=default
Default constructor. Produces an empty object.
memory::desc diff_dst_desc() const
Returns a diff destination memory descriptor.
Definition: dnnl.hpp:10209
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, const resampling_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a resampling backward propagation primitive.
Definition: dnnl.hpp:10188
Resampling backward propagation primitive.
Definition: dnnl.hpp:10108
resampling_backward(const primitive_desc &pd)
Constructs a resampling backward propagation primitive.
Definition: dnnl.hpp:10218
resampling_backward()=default
Default constructor. Produces an empty object.
Descriptor for resampling forward propagation.
Definition: dnnl.hpp:9966
desc(prop_kind aprop_kind, algorithm aalgorithm, const memory::desc &src_desc, const memory::desc &dst_desc)
Constructs a descriptor for a resampling forward propagation primitive using source and destination m...
Definition: dnnl.hpp:9984
desc(prop_kind aprop_kind, algorithm aalgorithm, const std::vector< float > &factors, const memory::desc &src_desc)
Constructs a descriptor for a resampling forward propagation primitive using source memory descriptor...
Definition: dnnl.hpp:10004
desc(prop_kind aprop_kind, algorithm aalgorithm, const std::vector< float > &factors, const memory::desc &src_desc, const memory::desc &dst_desc)
Constructs a descriptor for a resampling forward propagation primitive.
Definition: dnnl.hpp:10031
Primitive descriptor for a resampling forward propagation primitive.
Definition: dnnl.hpp:10045
primitive_desc(const desc &adesc, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a resampling forward propagation primitive.
Definition: dnnl.hpp:10059
memory::desc dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:10095
memory::desc src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:10092
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a resampling forward propagation primitive from a C API primiti...
Definition: dnnl.hpp:10086
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a resampling forward propagation primitive.
Definition: dnnl.hpp:10075
primitive_desc()=default
Default constructor. Produces an empty object.
Resampling forward propagation.
Definition: dnnl.hpp:9964
resampling_forward()=default
Default constructor. Produces an empty object.
resampling_forward(const primitive_desc &pd)
Constructs a resampling forward propagation primitive.
Definition: dnnl.hpp:10104
Base class for primitive descriptors for RNN primitives.
Definition: dnnl.hpp:7434
memory::desc dst_iter_c_desc() const
Returns destination recurrent cell state memory descriptor.
Definition: dnnl.hpp:7519
memory::desc weights_peephole_desc() const
Returns weights peephole memory descriptor.
Definition: dnnl.hpp:7485
memory::desc diff_weights_layer_desc() const
Returns diff weights layer memory descriptor.
Definition: dnnl.hpp:7545
memory::desc weights_layer_desc() const
Returns weights layer memory descriptor.
Definition: dnnl.hpp:7473
memory::desc weights_iter_desc() const
Returns weights iteration memory descriptor.
Definition: dnnl.hpp:7479
memory::desc diff_src_iter_desc() const
Returns diff source iteration memory descriptor.
Definition: dnnl.hpp:7533
memory::desc diff_dst_iter_c_desc() const
Returns diff destination recurrent cell state memory descriptor.
Definition: dnnl.hpp:7593
memory::desc diff_weights_iter_desc() const
Returns diff weights iteration memory descriptor.
Definition: dnnl.hpp:7551
memory::desc diff_dst_iter_desc() const
Returns diff destination iteration memory descriptor.
Definition: dnnl.hpp:7587
rnn_primitive_desc_base()=default
Default constructor. Produces an empty object.
memory::desc diff_src_iter_c_desc() const
Returns diff source recurrent cell state memory descriptor.
Definition: dnnl.hpp:7539
rnn_primitive_desc_base(dnnl_primitive_desc_t pd, dnnl::prop_kind aprop_kind, dnnl::algorithm cell_kind)
Constructs an RNN primitive descriptor base from a C API primitive descriptor while checking that it ...
Definition: dnnl.hpp:7447
memory::desc diff_bias_desc() const
Returns diff bias memory descriptor.
Definition: dnnl.hpp:7573
memory::desc dst_layer_desc() const
Returns destination layer memory descriptor.
Definition: dnnl.hpp:7505
memory::desc diff_weights_projection_desc() const
Returns diff weights projection memory descriptor.
Definition: dnnl.hpp:7564
memory::desc src_iter_c_desc() const
Returns source recurrent cell state memory descriptor.
Definition: dnnl.hpp:7467
memory::desc src_iter_desc() const
Returns source iteration memory descriptor.
Definition: dnnl.hpp:7461
memory::desc bias_desc() const
Returns bias memory descriptor.
Definition: dnnl.hpp:7499
memory::desc weights_projection_desc() const
Returns weights projection memory descriptor.
Definition: dnnl.hpp:7491
memory::desc src_layer_desc() const
Returns source layer memory descriptor.
Definition: dnnl.hpp:7453
memory::desc diff_dst_layer_desc() const
Returns diff destination layer memory descriptor.
Definition: dnnl.hpp:7579
memory::desc dst_iter_desc() const
Returns destination iteration memory descriptor.
Definition: dnnl.hpp:7513
memory::desc diff_weights_peephole_desc() const
Returns diff weights peephole memory descriptor.
Definition: dnnl.hpp:7557
memory::desc diff_src_layer_desc() const
Returns diff source layer memory descriptor.
Definition: dnnl.hpp:7525
Descriptor for a shuffle primitive backward propagation primitive.
Definition: dnnl.hpp:9671
desc(const memory::desc &diff_data_desc, int axis, int group_size)
Constructs a descriptor for a shuffle backward propagation primitive.
Definition: dnnl.hpp:9681
Primitive descriptor for a shuffle backward propagation primitive.
Definition: dnnl.hpp:9690
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a shuffle backward propagation primitive from a C API primitive...
Definition: dnnl.hpp:9721
memory::desc diff_src_desc() const
Returns a diff source memory descriptor.
Definition: dnnl.hpp:9726
primitive_desc()=default
Default constructor. Produces an empty object.
primitive_desc(const desc &adesc, const engine &aengine, const shuffle_forward::primitive_desc &hint_fwd_pd, const primitive_attr &attr=primitive_attr(), bool allow_empty=false)
Constructs a primitive descriptor for a shuffle backward propagation primitive.
Definition: dnnl.hpp:9708
memory::desc diff_dst_desc() const
Returns a diff destination memory descriptor.
Definition: dnnl.hpp:9729
Shuffle backward propagation primitive.
Definition: dnnl.hpp:9668
shuffle_backward()=default
Default constructor. Produces an empty object.
shuffle_backward(const primitive_desc &pd)
Constructs a shuffle backward propagation primitive.
Definition: dnnl.hpp:9738
Descriptor for a shuffle forward propagation primitive.
Definition: dnnl.hpp:9596
desc(prop_kind aprop_kind, const memory::desc &data_desc, int axis, int group_size)
Constructs a descriptor for a shuffle forward propagation primitive.
Definition: dnnl.hpp:9608
Primitive descriptor for a shuffle forward propagation primitive.
Definition: dnnl.hpp:9619
memory::desc dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:9655
memory::desc src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:9652
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a shuffle forward propagation primitive from a C API primitive ...
Definition: dnnl.hpp:9646
primitive_desc(const desc &adesc, const engine &aengine, const primitive_attr &attr=primitive_attr(), bool allow_empty=false)
Constructs a primitive descriptor for a shuffle forward propagation primitive.
Definition: dnnl.hpp:9634
primitive_desc()=default
Default constructor. Produces an empty object.
Shuffle forward propagation primitive.
Definition: dnnl.hpp:9594
shuffle_forward()=default
Default constructor. Produces an empty object.
shuffle_forward(const primitive_desc &pd)
Constructs a shuffle forward propagation primitive.
Definition: dnnl.hpp:9664
Descriptor for a softmax backward propagation primitive.
Definition: dnnl.hpp:6146
desc(const memory::desc &diff_data_desc, const memory::desc &data_desc, int softmax_axis)
Constructs a descriptor for a softmax backward propagation primitive.
Definition: dnnl.hpp:6159
desc()=default
Default constructor. Produces an empty object.
Primitive descriptor for a softmax backward propagation primitive.
Definition: dnnl.hpp:6170
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a softmax backward propagation primitive from a C API primitive...
Definition: dnnl.hpp:6220
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, const softmax_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a softmax backward propagation primitive.
Definition: dnnl.hpp:6207
memory::desc diff_dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:6231
memory::desc diff_src_desc() const
Returns a diff source memory descriptor.
Definition: dnnl.hpp:6228
primitive_desc()=default
Default constructor. Produces an empty object.
primitive_desc(const desc &adesc, const engine &aengine, const softmax_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a softmax backward propagation primitive.
Definition: dnnl.hpp:6187
memory::desc dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:6225
Softmax backward propagation primitive.
Definition: dnnl.hpp:6144
softmax_backward()=default
Default constructor. Produces an empty object.
softmax_backward(const primitive_desc &pd)
Constructs a softmax backward propagation primitive.
Definition: dnnl.hpp:6240
Descriptor for a softmax forward propagation primitive.
Definition: dnnl.hpp:6056
desc(prop_kind aprop_kind, const memory::desc &data_desc, int softmax_axis)
Constructs a descriptor for a softmax forward propagation primitive.
Definition: dnnl.hpp:6070
desc()=default
Default constructor. Produces an empty object.
Primitive descriptor for a softmax forward propagation primitive.
Definition: dnnl.hpp:6081
memory::desc src_desc() const
Returns a source memory descriptor.
Definition: dnnl.hpp:6128
primitive_desc(const desc &adesc, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a softmax forward propagation primitive.
Definition: dnnl.hpp:6095
memory::desc dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:6131
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a softmax forward propagation primitive from a C API primitive ...
Definition: dnnl.hpp:6122
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a softmax forward propagation primitive.
Definition: dnnl.hpp:6111
primitive_desc()=default
Default constructor. Produces an empty object.
Softmax forward propagation primitive.
Definition: dnnl.hpp:6054
softmax_forward()=default
Default constructor. Produces an empty object.
softmax_forward(const primitive_desc &pd)
Constructs a softmax forward propagation primitive.
Definition: dnnl.hpp:6140
An execution stream.
Definition: dnnl.hpp:985
engine get_engine() const
Returns the associated engine.
Definition: dnnl.hpp:1016
stream & wait()
Waits for all primitives executing in the stream to finish.
Definition: dnnl.hpp:1025
stream(const engine &aengine, flags aflags=flags::default_flags)
Constructs a stream for the specified engine and with behavior controlled by the specified flags.
Definition: dnnl.hpp:1007
flags
Stream flags. Can be combined using the bitwise OR operator.
Definition: dnnl.hpp:989
@ default_flags
Default stream configuration.
stream()=default
Constructs an empty stream.
Primitive descriptor for a sum primitive.
Definition: dnnl.hpp:3788
memory::desc dst_desc() const
Returns a destination memory descriptor.
Definition: dnnl.hpp:3861
primitive_desc()=default
Default constructor. Produces an empty object.
memory::desc src_desc(int idx=0) const
Returns a source memory descriptor.
Definition: dnnl.hpp:3858
primitive_desc(const memory::desc &dst, const std::vector< float > &scales, const std::vector< memory::desc > &srcs, const engine &aengine, const primitive_attr &attr=primitive_attr())
Constructs a primitive descriptor for a sum primitive.
Definition: dnnl.hpp:3802
primitive_desc(const std::vector< float > &scales, const std::vector< memory::desc > &srcs, const engine &aengine, const primitive_attr &attr=primitive_attr())
Constructs a primitive descriptor for a sum primitive.
Definition: dnnl.hpp:3832
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for sum primitive from a C API primitive descriptor which must have...
Definition: dnnl.hpp:3854
Out-of-place summation (sum) primitive.
Definition: dnnl.hpp:3786
sum()=default
Default constructor. Produces an empty object.
sum(const primitive_desc &pd)
Constructs a sum primitive.
Definition: dnnl.hpp:3869
Descriptor for a vanilla RNN backward propagation primitive.
Definition: dnnl.hpp:7804
desc(prop_kind aprop_kind, algorithm activation, rnn_direction direction, const memory::desc &src_layer_desc, const memory::desc &src_iter_desc, const memory::desc &weights_layer_desc, const memory::desc &weights_iter_desc, const memory::desc &bias_desc, const memory::desc &dst_layer_desc, const memory::desc &dst_iter_desc, const memory::desc &diff_src_layer_desc, const memory::desc &diff_src_iter_desc, const memory::desc &diff_weights_layer_desc, const memory::desc &diff_weights_iter_desc, const memory::desc &diff_bias_desc, const memory::desc &diff_dst_layer_desc, const memory::desc &diff_dst_iter_desc, rnn_flags flags=rnn_flags::undef, float alpha=0.0f, float beta=0.0f)
Constructs a descriptor for a vanilla RNN backward propagation primitive.
Definition: dnnl.hpp:7859
Primitive descriptor for an RNN backward propagation primitive.
Definition: dnnl.hpp:7895
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, const vanilla_rnn_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a vanilla RNN backward propagation primitive.
Definition: dnnl.hpp:7932
memory::desc src_iter_desc() const
Returns source iteration memory descriptor.
Definition: dnnl.hpp:7955
memory::desc diff_dst_layer_desc() const
Returns diff destination layer memory descriptor.
Definition: dnnl.hpp:8009
memory::desc dst_layer_desc() const
Returns destination layer memory descriptor.
Definition: dnnl.hpp:7971
memory::desc diff_src_iter_desc() const
Returns diff source iteration memory descriptor.
Definition: dnnl.hpp:7989
memory::desc diff_weights_iter_desc() const
Returns diff weights iteration memory descriptor.
Definition: dnnl.hpp:7999
primitive_desc()=default
Default constructor. Produces an empty object.
primitive_desc(const desc &adesc, const engine &aengine, const vanilla_rnn_forward::primitive_desc &hint_fwd_pd, bool allow_empty=false)
Constructs a primitive descriptor for a vanilla RNN backward propagation primitive.
Definition: dnnl.hpp:7912
memory::desc diff_bias_desc() const
Returns diff bias memory descriptor.
Definition: dnnl.hpp:8004
memory::desc weights_iter_desc() const
Returns weights iteration memory descriptor.
Definition: dnnl.hpp:7963
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a vanilla RNN backward propagation primitive from a C API primi...
Definition: dnnl.hpp:7945
memory::desc weights_layer_desc() const
Returns weights layer memory descriptor.
Definition: dnnl.hpp:7958
memory::desc bias_desc() const
Returns bias memory descriptor.
Definition: dnnl.hpp:7968
memory::desc dst_iter_desc() const
Returns destination iteration memory descriptor.
Definition: dnnl.hpp:7976
memory::desc diff_dst_iter_desc() const
Returns diff destination iteration memory descriptor.
Definition: dnnl.hpp:8014
memory::desc diff_src_layer_desc() const
Returns diff source layer memory descriptor.
Definition: dnnl.hpp:7984
memory::desc src_layer_desc() const
Returns source layer memory descriptor.
Definition: dnnl.hpp:7950
memory::desc diff_weights_layer_desc() const
Returns diff weights layer memory descriptor.
Definition: dnnl.hpp:7994
memory::desc workspace_desc() const
Returns the workspace memory descriptor.
Definition: dnnl.hpp:7979
Vanilla RNN backward propagation primitive.
Definition: dnnl.hpp:7802
vanilla_rnn_backward(const primitive_desc &pd)
Constructs a vanilla RNN backward propagation primitive.
Definition: dnnl.hpp:8025
vanilla_rnn_backward()=default
Default constructor. Produces an empty object.
Descriptor for a vanilla RNN forward propagation primitive.
Definition: dnnl.hpp:7643
desc(prop_kind aprop_kind, algorithm activation, rnn_direction direction, const memory::desc &src_layer_desc, const memory::desc &src_iter_desc, const memory::desc &weights_layer_desc, const memory::desc &weights_iter_desc, const memory::desc &bias_desc, const memory::desc &dst_layer_desc, const memory::desc &dst_iter_desc, rnn_flags flags=rnn_flags::undef, float alpha=0.0f, float beta=0.0f)
Constructs a descriptor for a vanilla RNN forward propagation primitive.
Definition: dnnl.hpp:7686
Primitive descriptor for a vanilla RNN forward propagation primitive.
Definition: dnnl.hpp:7711
primitive_desc(const desc &adesc, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a vanilla RNN forward propagation primitive.
Definition: dnnl.hpp:7725
primitive_desc(dnnl_primitive_desc_t pd)
Constructs a primitive descriptor for a vanilla RNN forward propagation primitive from a C API primit...
Definition: dnnl.hpp:7752
memory::desc src_layer_desc() const
Returns source layer memory descriptor.
Definition: dnnl.hpp:7758
memory::desc src_iter_desc() const
Returns source iteration memory descriptor.
Definition: dnnl.hpp:7763
memory::desc weights_iter_desc() const
Returns weights iteration memory descriptor.
Definition: dnnl.hpp:7771
memory::desc weights_layer_desc() const
Returns weights layer memory descriptor.
Definition: dnnl.hpp:7766
memory::desc workspace_desc() const
Returns the workspace memory descriptor.
Definition: dnnl.hpp:7787
memory::desc dst_iter_desc() const
Returns destination iteration memory descriptor.
Definition: dnnl.hpp:7784
primitive_desc()=default
Default constructor. Produces an empty object.
primitive_desc(const desc &adesc, const primitive_attr &attr, const engine &aengine, bool allow_empty=false)
Constructs a primitive descriptor for a vanilla RNN forward propagation primitive.
Definition: dnnl.hpp:7741
memory::desc dst_layer_desc() const
Returns destination layer memory descriptor.
Definition: dnnl.hpp:7779
memory::desc bias_desc() const
Returns bias memory descriptor.
Definition: dnnl.hpp:7776
Vanilla RNN forward propagation primitive.
Definition: dnnl.hpp:7641
vanilla_rnn_forward()=default
Default constructor. Produces an empty object.
vanilla_rnn_forward(const primitive_desc &pd)
Constructs a vanilla RNN forward propagation primitive.
Definition: dnnl.hpp:7798
A descriptor of a Batch Normalization operation.
Definition: dnnl_types.h:1827
A descriptor of a binary operation.
Definition: dnnl_types.h:2035
A descriptor of a convolution operation.
Definition: dnnl_types.h:1534
A descriptor of a element-wise operation.
Definition: dnnl_types.h:1609
An opaque structure to describe an engine.
A descriptor of an inner product operation.
Definition: dnnl_types.h:1897
A descriptor of a Layer Normalization operation.
Definition: dnnl_types.h:1860
A descriptor of a Local Response Normalization (LRN) operation.
Definition: dnnl_types.h:1796
A descriptor of a matrix multiplication operation.
Definition: dnnl_types.h:2061
Memory descriptor.
Definition: dnnl_types.h:1445
dnnl_data_type_t data_type
Data type of the tensor elements.
Definition: dnnl_types.h:1465
dnnl_dims_t dims
Dimensions in the following order:
Definition: dnnl_types.h:1462
int ndims
Number of dimensions.
Definition: dnnl_types.h:1447
An opaque structure to describe a memory.
A descriptor of a pooling operation.
Definition: dnnl_types.h:1696
A descriptor of a pooling operation.
Definition: dnnl_types.h:1734
An opaque structure for a chain of post operations.
An opaque structure for primitive descriptor attributes.
An opaque structure to describe a primitive descriptor iterator.
An opaque structure to describe a primitive descriptor.
An opaque structure to describe a primitive.
A descriptor of reduction operation.
Definition: dnnl_types.h:2111
A descriptor of resampling operation.
Definition: dnnl_types.h:2083
A descriptor for an RNN operation.
Definition: dnnl_types.h:1953
A descriptor of a shuffle operation.
Definition: dnnl_types.h:1587
A descriptor of a Softmax operation.
Definition: dnnl_types.h:1666
An opaque structure to describe an execution stream.
Structure containing version information as per Semantic Versioning
Definition: dnnl_types.h:2634