Skip to content

Commit 47b2b9b

Browse files
authored
cpu: aarch64: Expand ARM SVE support for 1x1 convolution (#2075)
1 parent b1b2758 commit 47b2b9b

15 files changed

+479
-198
lines changed

include/oneapi/dnnl/dnnl.hpp

+13
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,6 @@
11
/*******************************************************************************
22
* Copyright 2016-2024 Intel Corporation
3+
* Copyright 2024 FUJITSU LIMITED
34
*
45
* Licensed under the Apache License, Version 2.0 (the "License");
56
* you may not use this file except in compliance with the License.
@@ -1412,6 +1413,7 @@ struct memory : public handle<dnnl_memory_t> {
14121413
aBdeC8b4c = dnnl_aBdeC8b4c,
14131414
aBdefc16b = dnnl_aBdefc16b,
14141415
aCBdef16c16b = dnnl_aCBdef16c16b,
1416+
aCBdef8b8c = dnnl_aCBdef8b8c,
14151417
aCBdef16b16c = dnnl_aCBdef16b16c,
14161418
aBdefc4b = dnnl_aBdefc4b,
14171419
aBdefc8b = dnnl_aBdefc8b,
@@ -1422,8 +1424,10 @@ struct memory : public handle<dnnl_memory_t> {
14221424
Acb8a = dnnl_Acb8a,
14231425
AcB8a2b = dnnl_AcB8a2b,
14241426
AcB8a4b = dnnl_AcB8a4b,
1427+
aCBd8b8c = dnnl_aCBd8b8c,
14251428
aCBd16b16c = dnnl_aCBd16b16c,
14261429
aCBd16c16b = dnnl_aCBd16c16b,
1430+
aCBde8b8c = dnnl_aCBde8b8c,
14271431
aCBde16b16c = dnnl_aCBde16b16c,
14281432
aCBde16c16b = dnnl_aCBde16c16b,
14291433
Acdb16a = dnnl_Acdb16a,
@@ -1436,12 +1440,15 @@ struct memory : public handle<dnnl_memory_t> {
14361440
Acdeb8a = dnnl_Acdeb8a,
14371441
AcdeB8a2b = dnnl_AcdeB8a2b,
14381442
AcdeB8a4b = dnnl_AcdeB8a4b,
1443+
BAc8a8b = dnnl_BAc8a8b,
14391444
BAc16a16b = dnnl_BAc16a16b,
14401445
BAc16b16a = dnnl_BAc16b16a,
1446+
BAcd8a8b = dnnl_BAcd8a8b,
14411447
BAcd16a16b = dnnl_BAcd16a16b,
14421448
BAcd16b16a = dnnl_BAcd16b16a,
14431449
ABcd32a32b = dnnl_ABcd32a32b,
14441450
BAcde16b16a = dnnl_BAcde16b16a,
1451+
BAcde8a8b = dnnl_BAcde8a8b,
14451452
BAcde16a16b = dnnl_BAcde16a16b,
14461453
aBdec32b = dnnl_aBdec32b,
14471454
Abcdef16a = dnnl_Abcdef16a,
@@ -1692,6 +1699,7 @@ struct memory : public handle<dnnl_memory_t> {
16921699
gIOhw16i16o = dnnl_gIOhw16i16o,
16931700
gOhwi32o = dnnl_gOhwi32o,
16941701
Goidhw16g = dnnl_Goidhw16g,
1702+
IOw8o8i = dnnl_IOw8o8i,
16951703
IOw16o16i = dnnl_IOw16o16i,
16961704
OIw16i16o = dnnl_OIw16i16o,
16971705
OwI16i16o = dnnl_OwI16i16o,
@@ -1748,6 +1756,7 @@ struct memory : public handle<dnnl_memory_t> {
17481756
Owi8o = dnnl_Owi8o,
17491757
OwI8o2i = dnnl_OwI8o2i,
17501758
OwI8o4i = dnnl_OwI8o4i,
1759+
IOhw8o8i = dnnl_IOhw8o8i,
17511760
IOhw16o16i = dnnl_IOhw16o16i,
17521761
Ohwi16o = dnnl_Ohwi16o,
17531762
OhwI16o2i = dnnl_OhwI16o2i,
@@ -1793,6 +1802,7 @@ struct memory : public handle<dnnl_memory_t> {
17931802
OIhw8o8i = dnnl_OIhw8o8i,
17941803
OIhw8o4i = dnnl_OIhw8o4i,
17951804
OIhw2i8o4i = dnnl_OIhw2i8o4i,
1805+
IOdhw8o8i = dnnl_IOdhw8o8i,
17961806
IOdhw16o16i = dnnl_IOdhw16o16i,
17971807
Odhwi16o = dnnl_Odhwi16o,
17981808
OdhwI16o2i = dnnl_OdhwI16o2i,
@@ -1846,6 +1856,7 @@ struct memory : public handle<dnnl_memory_t> {
18461856
OdhwI8i8o = dnnl_OdhwI8i8o,
18471857
OIdhw8o8i = dnnl_OIdhw8o8i,
18481858
OIdhw8o4i = dnnl_OIdhw8o4i,
1859+
gIOw8o8i = dnnl_gIOw8o8i,
18491860
gIOw16o16i = dnnl_gIOw16o16i,
18501861
gOIw16i16o = dnnl_gOIw16i16o,
18511862
gOIw16o16i = dnnl_gOIw16o16i,
@@ -1874,6 +1885,7 @@ struct memory : public handle<dnnl_memory_t> {
18741885
gOwI8o4i = dnnl_gOwI8o4i,
18751886
Goiw8g = dnnl_Goiw8g,
18761887
Goiw16g = dnnl_Goiw16g,
1888+
gIOhw8o8i = dnnl_gIOhw8o8i,
18771889
gIOhw16o16i = dnnl_gIOhw16o16i,
18781890
gOhwi16o = dnnl_gOhwi16o,
18791891
gOhwI16o2i = dnnl_gOhwI16o2i,
@@ -1920,6 +1932,7 @@ struct memory : public handle<dnnl_memory_t> {
19201932
gOIhw8o8i = dnnl_gOIhw8o8i,
19211933
gOIhw8o4i = dnnl_gOIhw8o4i,
19221934
gIOdhw16i16o = dnnl_gIOdhw16i16o,
1935+
gIOdhw8o8i = dnnl_gIOdhw8o8i,
19231936
gIOdhw16o16i = dnnl_gIOdhw16o16i,
19241937
gOdhwi16o = dnnl_gOdhwi16o,
19251938
gOdhwI16o2i = dnnl_gOdhwI16o2i,

include/oneapi/dnnl/dnnl_types.h

+13
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,6 @@
11
/*******************************************************************************
22
* Copyright 2016-2024 Intel Corporation
3+
* Copyright 2024 FUJITSU LIMITED
34
*
45
* Licensed under the Apache License, Version 2.0 (the "License");
56
* you may not use this file except in compliance with the License.
@@ -1038,6 +1039,12 @@ typedef enum {
10381039
dnnl_cabd,
10391040
dnnl_dabc,
10401041
dnnl_Ab32a,
1042+
dnnl_aCBd8b8c,
1043+
dnnl_aCBde8b8c,
1044+
dnnl_BAc8a8b,
1045+
dnnl_BAcd8a8b,
1046+
dnnl_BAcde8a8b,
1047+
dnnl_aCBdef8b8c,
10411048

10421049
/// Just a sentinel, not real memory format tag. Must be changed after new
10431050
/// format tag is added.
@@ -1258,6 +1265,7 @@ typedef enum {
12581265
dnnl_OI8i8o = dnnl_AB8b8a,
12591266

12601267
// weights, 3D
1268+
dnnl_IOw8o8i = dnnl_BAc8a8b,
12611269
dnnl_IOw16o16i = dnnl_BAc16a16b,
12621270
dnnl_IOw16i16o = dnnl_BAc16b16a,
12631271
dnnl_OIw16i16o = dnnl_ABc16b16a,
@@ -1328,6 +1336,7 @@ typedef enum {
13281336

13291337
// weights, 4D
13301338
dnnl_IOhw16i16o = dnnl_BAcd16b16a,
1339+
dnnl_IOhw8o8i = dnnl_BAcd8a8b,
13311340
dnnl_IOhw16o16i = dnnl_BAcd16a16b,
13321341
dnnl_Ohwi16o = dnnl_Acdb16a,
13331342
dnnl_OhwI16o2i = dnnl_AcdB16a2b,
@@ -1460,6 +1469,7 @@ typedef enum {
14601469
dnnl_OIdhw8o4i = dnnl_ABcde8a4b,
14611470
dnnl_IOdhw16i16o = dnnl_BAcde16b16a,
14621471
dnnl_OIdhw4o8i8o4i = dnnl_ABcde4a8b8a4b,
1472+
dnnl_IOdhw8o8i = dnnl_BAcde8a8b,
14631473
dnnl_IOdhw16o16i = dnnl_BAcde16a16b,
14641474
dnnl_OIdhw16o16i2o = dnnl_ABcde16a16b2a,
14651475
dnnl_OIdhw8i32o = dnnl_ABcde8b32a,
@@ -1473,6 +1483,7 @@ typedef enum {
14731483
dnnl_Goiw16g = dnnl_Abcd16a,
14741484
dnnl_Goiw8g = dnnl_Abcd8a,
14751485
dnnl_Goiw4g = dnnl_Abcd4a,
1486+
dnnl_gIOw8o8i = dnnl_aCBd8b8c,
14761487
dnnl_gIOw16o16i = dnnl_aCBd16b16c,
14771488
dnnl_gIOw16i16o = dnnl_aCBd16c16b,
14781489
dnnl_gOIw16i16o = dnnl_aBCd16c16b,
@@ -1518,6 +1529,7 @@ typedef enum {
15181529

15191530
// weights w/ groups, 4D
15201531
dnnl_gIOhw16i16o = dnnl_aCBde16c16b,
1532+
dnnl_gIOhw8o8i = dnnl_aCBde8b8c,
15211533
dnnl_gIOhw16o16i = dnnl_aCBde16b16c,
15221534
dnnl_gOhwi16o = dnnl_aBdec16b,
15231535
dnnl_gOhwI16o2i = dnnl_aBdeC16b2c,
@@ -1585,6 +1597,7 @@ typedef enum {
15851597

15861598
// weights w/ groups, 6D
15871599
dnnl_gIOdhw16i16o = dnnl_aCBdef16c16b,
1600+
dnnl_gIOdhw8o8i = dnnl_aCBdef8b8c,
15881601
dnnl_gIOdhw16o16i = dnnl_aCBdef16b16c,
15891602
dnnl_gOdhwi16o = dnnl_aBdefc16b,
15901603
dnnl_gOdhwI16o2i = dnnl_aBdefC16b2c,

src/common/c_types_map.hpp

+13
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,6 @@
11
/*******************************************************************************
22
* Copyright 2016-2024 Intel Corporation
3+
* Copyright 2024 FUJITSU LIMITED
34
*
45
* Licensed under the Apache License, Version 2.0 (the "License");
56
* you may not use this file except in compliance with the License.
@@ -620,6 +621,7 @@ const format_tag_t aBdefc16b = dnnl_aBdefc16b;
620621
const format_tag_t aBdefC16b2c = dnnl_aBdefC16b2c;
621622
const format_tag_t aBdefC16b4c = dnnl_aBdefC16b4c;
622623
const format_tag_t aCBdef16c16b = dnnl_aCBdef16c16b;
624+
const format_tag_t aCBdef8b8c = dnnl_aCBdef8b8c;
623625
const format_tag_t aCBdef16b16c = dnnl_aCBdef16b16c;
624626
const format_tag_t aBdefc4b = dnnl_aBdefc4b;
625627
const format_tag_t aBdefc8b = dnnl_aBdefc8b;
@@ -634,8 +636,10 @@ const format_tag_t Acb4a = dnnl_Acb4a;
634636
const format_tag_t Acb8a = dnnl_Acb8a;
635637
const format_tag_t AcB8a2b = dnnl_AcB8a2b;
636638
const format_tag_t AcB8a4b = dnnl_AcB8a4b;
639+
const format_tag_t aCBd8b8c = dnnl_aCBd8b8c;
637640
const format_tag_t aCBd16b16c = dnnl_aCBd16b16c;
638641
const format_tag_t aCBd16c16b = dnnl_aCBd16c16b;
642+
const format_tag_t aCBde8b8c = dnnl_aCBde8b8c;
639643
const format_tag_t aCBde16b16c = dnnl_aCBde16b16c;
640644
const format_tag_t aCBde16c16b = dnnl_aCBde16c16b;
641645
const format_tag_t Acdb16a = dnnl_Acdb16a;
@@ -654,7 +658,9 @@ const format_tag_t AcdeB8a2b = dnnl_AcdeB8a2b;
654658
const format_tag_t AcdeB8a4b = dnnl_AcdeB8a4b;
655659
const format_tag_t Acedb16a = dnnl_Acedb16a;
656660
const format_tag_t Adcb16a = dnnl_Adcb16a;
661+
const format_tag_t BAc8a8b = dnnl_BAc8a8b;
657662
const format_tag_t BAc16a16b = dnnl_BAc16a16b;
663+
const format_tag_t BAcd8a8b = dnnl_BAcd8a8b;
658664
const format_tag_t BAcd16a16b = dnnl_BAcd16a16b;
659665
const format_tag_t ABc32a16b = dnnl_ABc32a16b;
660666
const format_tag_t ABcd32a16b = dnnl_ABcd32a16b;
@@ -663,6 +669,7 @@ const format_tag_t ABc40a16b = dnnl_ABc40a16b;
663669
const format_tag_t ABcd40a16b = dnnl_ABcd40a16b;
664670
const format_tag_t ABcde40a16b = dnnl_ABcde40a16b;
665671
const format_tag_t ABc32a32b = dnnl_ABc32a32b;
672+
const format_tag_t BAcde8a8b = dnnl_BAcde8a8b;
666673
const format_tag_t BAcde16a16b = dnnl_BAcde16a16b;
667674
const format_tag_t ABcd32a32b = dnnl_ABcd32a32b;
668675
const format_tag_t ABcde32a32b = dnnl_ABcde32a32b;
@@ -1169,6 +1176,7 @@ const format_tag_t Ohwi32o = dnnl_Ohwi32o;
11691176
const format_tag_t gIOhw16i16o = dnnl_gIOhw16i16o;
11701177
const format_tag_t gOhwi32o = dnnl_gOhwi32o;
11711178
const format_tag_t Goidhw16g = dnnl_Goidhw16g;
1179+
const format_tag_t IOw8o8i = dnnl_IOw8o8i;
11721180
const format_tag_t IOw16o16i = dnnl_IOw16o16i;
11731181
const format_tag_t IOw16i16o = dnnl_IOw16i16o;
11741182
const format_tag_t gIOw16i16o = dnnl_gIOw16i16o;
@@ -1224,7 +1232,9 @@ const format_tag_t Owi4o = dnnl_Owi4o;
12241232
const format_tag_t Owi8o = dnnl_Owi8o;
12251233
const format_tag_t OwI8o2i = dnnl_OwI8o2i;
12261234
const format_tag_t OwI8o4i = dnnl_OwI8o4i;
1235+
const format_tag_t IOdhw8o8i = dnnl_IOdhw8o8i;
12271236
const format_tag_t IOdhw16o16i = dnnl_IOdhw16o16i;
1237+
const format_tag_t IOhw8o8i = dnnl_IOhw8o8i;
12281238
const format_tag_t IOhw16o16i = dnnl_IOhw16o16i;
12291239
const format_tag_t Ohwi16o = dnnl_Ohwi16o;
12301240
const format_tag_t OhwI16o2i = dnnl_OhwI16o2i;
@@ -1332,6 +1342,7 @@ const format_tag_t OIdhw8i8o = dnnl_OIdhw8i8o;
13321342
const format_tag_t OdhwI8i8o = dnnl_OdhwI8i8o;
13331343
const format_tag_t OIdhw8o8i = dnnl_OIdhw8o8i;
13341344
const format_tag_t OIdhw8o4i = dnnl_OIdhw8o4i;
1345+
const format_tag_t gIOw8o8i = dnnl_gIOw8o8i;
13351346
const format_tag_t gIOw16o16i = dnnl_gIOw16o16i;
13361347
const format_tag_t Goiw16g = dnnl_Goiw16g;
13371348
const format_tag_t Goiw8g = dnnl_Goiw8g;
@@ -1360,7 +1371,9 @@ const format_tag_t gOwi4o = dnnl_gOwi4o;
13601371
const format_tag_t gOwi8o = dnnl_gOwi8o;
13611372
const format_tag_t gOwI8o2i = dnnl_gOwI8o2i;
13621373
const format_tag_t gOwI8o4i = dnnl_gOwI8o4i;
1374+
const format_tag_t gIOdhw8o8i = dnnl_gIOdhw8o8i;
13631375
const format_tag_t gIOdhw16o16i = dnnl_gIOdhw16o16i;
1376+
const format_tag_t gIOhw8o8i = dnnl_gIOhw8o8i;
13641377
const format_tag_t gIOhw16o16i = dnnl_gIOhw16o16i;
13651378
const format_tag_t gOhwi16o = dnnl_gOhwi16o;
13661379
const format_tag_t gOhwI16o2i = dnnl_gOhwI16o2i;

src/common/dnnl_debug_autogenerated.cpp

+15-2
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,6 @@
11
/*******************************************************************************
22
* Copyright 2018-2024 Intel Corporation
3+
* Copyright 2024 FUJITSU LIMITED
34
*
45
* Licensed under the Apache License, Version 2.0 (the "License");
56
* you may not use this file except in compliance with the License.
@@ -96,8 +97,8 @@ const char *dnnl_engine_kind2str(dnnl_engine_kind_t v) {
9697
const char *dnnl_sparse_encoding2str(dnnl_sparse_encoding_t v) {
9798
if (v == dnnl_sparse_encoding_undef) return "undef";
9899
if (v == dnnl_csr) return "csr";
99-
if (v == dnnl_coo) return "coo";
100100
if (v == dnnl_packed) return "packed";
101+
if (v == dnnl_coo) return "coo";
101102
assert(!"unknown sparse_encoding");
102103
return "unknown sparse_encoding";
103104
}
@@ -927,7 +928,6 @@ const char *dnnl_fmt_tag2str(dnnl_format_tag_t v) {
927928
if (v == dnnl_AcdeB4b8a4b) return "AcdeB4b8a4b";
928929
if (v == dnnl_Ab4a) return "Ab4a";
929930
if (v == dnnl_Ab8a) return "Ab8a";
930-
if (v == dnnl_Ab32a) return "Ab32a";
931931
if (v == dnnl_BA4b4a) return "BA4b4a";
932932
if (v == dnnl_BA8b4a) return "BA8b4a";
933933
if (v == dnnl_BA2a24b) return "BA2a24b";
@@ -943,6 +943,13 @@ const char *dnnl_fmt_tag2str(dnnl_format_tag_t v) {
943943
if (v == dnnl_bcad) return "bcad";
944944
if (v == dnnl_cabd) return "cabd";
945945
if (v == dnnl_dabc) return "dabc";
946+
if (v == dnnl_Ab32a) return "Ab32a";
947+
if (v == dnnl_aCBd8b8c) return "aCBd8b8c";
948+
if (v == dnnl_aCBde8b8c) return "aCBde8b8c";
949+
if (v == dnnl_BAc8a8b) return "BAc8a8b";
950+
if (v == dnnl_BAcd8a8b) return "BAcd8a8b";
951+
if (v == dnnl_BAcde8a8b) return "BAcde8a8b";
952+
if (v == dnnl_aCBdef8b8c) return "aCBdef8b8c";
946953
if (v == dnnl_format_tag_last) return "format_tag_last";
947954
if (v == dnnl_x) return "x";
948955
if (v == dnnl_nc) return "nc";
@@ -1048,6 +1055,7 @@ const char *dnnl_fmt_tag2str(dnnl_format_tag_t v) {
10481055
if (v == dnnl_OI8i24o) return "OI8i24o";
10491056
if (v == dnnl_OI8i16o) return "OI8i16o";
10501057
if (v == dnnl_OI8i8o) return "OI8i8o";
1058+
if (v == dnnl_IOw8o8i) return "IOw8o8i";
10511059
if (v == dnnl_IOw16o16i) return "IOw16o16i";
10521060
if (v == dnnl_IOw16i16o) return "IOw16i16o";
10531061
if (v == dnnl_OIw16i16o) return "OIw16i16o";
@@ -1116,6 +1124,7 @@ const char *dnnl_fmt_tag2str(dnnl_format_tag_t v) {
11161124
if (v == dnnl_OwI8i16o) return "OwI8i16o";
11171125
if (v == dnnl_OwI8o4i) return "OwI8o4i";
11181126
if (v == dnnl_IOhw16i16o) return "IOhw16i16o";
1127+
if (v == dnnl_IOhw8o8i) return "IOhw8o8i";
11191128
if (v == dnnl_IOhw16o16i) return "IOhw16o16i";
11201129
if (v == dnnl_Ohwi16o) return "Ohwi16o";
11211130
if (v == dnnl_OhwI16o2i) return "OhwI16o2i";
@@ -1246,6 +1255,7 @@ const char *dnnl_fmt_tag2str(dnnl_format_tag_t v) {
12461255
if (v == dnnl_OIdhw8o4i) return "OIdhw8o4i";
12471256
if (v == dnnl_IOdhw16i16o) return "IOdhw16i16o";
12481257
if (v == dnnl_OIdhw4o8i8o4i) return "OIdhw4o8i8o4i";
1258+
if (v == dnnl_IOdhw8o8i) return "IOdhw8o8i";
12491259
if (v == dnnl_IOdhw16o16i) return "IOdhw16o16i";
12501260
if (v == dnnl_OIdhw16o16i2o) return "OIdhw16o16i2o";
12511261
if (v == dnnl_OIdhw8i32o) return "OIdhw8i32o";
@@ -1257,6 +1267,7 @@ const char *dnnl_fmt_tag2str(dnnl_format_tag_t v) {
12571267
if (v == dnnl_Goiw16g) return "Goiw16g";
12581268
if (v == dnnl_Goiw8g) return "Goiw8g";
12591269
if (v == dnnl_Goiw4g) return "Goiw4g";
1270+
if (v == dnnl_gIOw8o8i) return "gIOw8o8i";
12601271
if (v == dnnl_gIOw16o16i) return "gIOw16o16i";
12611272
if (v == dnnl_gIOw16i16o) return "gIOw16i16o";
12621273
if (v == dnnl_gOIw16i16o) return "gOIw16i16o";
@@ -1300,6 +1311,7 @@ const char *dnnl_fmt_tag2str(dnnl_format_tag_t v) {
13001311
if (v == dnnl_goIw4i) return "goIw4i";
13011312
if (v == dnnl_goIw32i) return "goIw32i";
13021313
if (v == dnnl_gIOhw16i16o) return "gIOhw16i16o";
1314+
if (v == dnnl_gIOhw8o8i) return "gIOhw8o8i";
13031315
if (v == dnnl_gIOhw16o16i) return "gIOhw16o16i";
13041316
if (v == dnnl_gOhwi16o) return "gOhwi16o";
13051317
if (v == dnnl_gOhwI16o2i) return "gOhwI16o2i";
@@ -1363,6 +1375,7 @@ const char *dnnl_fmt_tag2str(dnnl_format_tag_t v) {
13631375
if (v == dnnl_gOIhw4i8o2i) return "gOIhw4i8o2i";
13641376
if (v == dnnl_gOIhw4o8i2o) return "gOIhw4o8i2o";
13651377
if (v == dnnl_gIOdhw16i16o) return "gIOdhw16i16o";
1378+
if (v == dnnl_gIOdhw8o8i) return "gIOdhw8o8i";
13661379
if (v == dnnl_gIOdhw16o16i) return "gIOdhw16o16i";
13671380
if (v == dnnl_gOdhwi16o) return "gOdhwi16o";
13681381
if (v == dnnl_gOdhwI16o2i) return "gOdhwI16o2i";

src/common/memory_desc_wrapper.cpp

+7
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,6 @@
11
/*******************************************************************************
22
* Copyright 2016-2024 Intel Corporation
3+
* Copyright 2024 FUJITSU LIMITED
34
*
45
* Licensed under the Apache License, Version 2.0 (the "License");
56
* you may not use this file except in compliance with the License.
@@ -516,7 +517,9 @@ status_t memory_desc_wrapper::compute_blocking(
516517
C(Acb8a, {0, 2, 1}, {8}, {0});
517518
C(AcB8a2b, {0, 2, 1}, {8, 2}, {0, 1});
518519
C(AcB8a4b, {0, 2, 1}, {8, 4}, {0, 1});
520+
C(aCBd8b8c, {0, 2, 1, 3}, {8, 8}, {1, 2});
519521
C(aCBd16b16c, {0, 2, 1, 3}, {16, 16}, {1, 2});
522+
C(aCBde8b8c, {0, 2, 1, 3, 4}, {8, 8}, {1, 2});
520523
C(aCBde16b16c, {0, 2, 1, 3, 4}, {16, 16}, {1, 2});
521524
C(Acdb16a, {0, 2, 3, 1}, {16}, {0});
522525
C(AcdB16a2b, {0, 2, 3, 1}, {16, 2}, {0, 1});
@@ -532,7 +535,9 @@ status_t memory_desc_wrapper::compute_blocking(
532535
C(AcdeB8a4b, {0, 2, 3, 4, 1}, {8, 4}, {0, 1});
533536
C(Acedb16a, {0, 2, 4, 3, 1}, {16}, {0});
534537
C(Adcb16a, {0, 3, 2, 1}, {16}, {0});
538+
C(BAc8a8b, {1, 0, 2}, {8, 8}, {0, 1});
535539
C(BAc16a16b, {1, 0, 2}, {16, 16}, {0, 1});
540+
C(BAcd8a8b, {1, 0, 2, 3}, {8, 8}, {0, 1});
536541
C(BAcd16a16b, {1, 0, 2, 3}, {16, 16}, {0, 1});
537542
C(ABc32a16b, {0, 1, 2}, {32, 16}, {0, 1});
538543
C(ABcd32a16b, {0, 1, 2, 3}, {32, 16}, {0, 1});
@@ -585,13 +590,15 @@ status_t memory_desc_wrapper::compute_blocking(
585590
C(aBCde2b8c8b2c, {0, 1, 2, 3, 4}, {2, 8, 8, 2}, {1, 2, 1, 2});
586591
C(aBdec32b, {0, 1, 3, 4, 2}, {32}, {1});
587592
C(aCBdef16c16b, {0, 2, 1, 3, 4, 5}, {16, 16}, {2, 1});
593+
C(aCBdef8b8c, {0, 2, 1, 3, 4, 5}, {8, 8}, {1, 2});
588594
C(aCBdef16b16c, {0, 2, 1, 3, 4, 5}, {16, 16}, {1, 2});
589595
C(Abcdef16a, {0, 1, 2, 3, 4, 5}, {16}, {0});
590596
C(Abcdef32a, {0, 1, 2, 3, 4, 5}, {32}, {0});
591597
C(aCBd16c16b, {0, 2, 1, 3}, {16, 16}, {2, 1});
592598
C(aCBde16c16b, {0, 2, 1, 3, 4}, {16, 16}, {2, 1});
593599
C(Acdb32a, {0, 2, 3, 1}, {32}, {0});
594600
C(BAcd16b16a, {1, 0, 2, 3}, {16, 16}, {1, 0});
601+
C(BAcde8a8b, {1, 0, 2, 3, 4}, {8, 8}, {0, 1});
595602
C(BAcde16a16b, {1, 0, 2, 3, 4}, {16, 16}, {0, 1});
596603
C(BAc16b16a, {1, 0, 2}, {16, 16}, {1, 0});
597604
C(aBCd2b4c2b, {0, 1, 2, 3}, {2, 4, 2}, {1, 2, 1});

0 commit comments

Comments
 (0)