Attila Afra
f53274c9fe
merged 2.2.4
2021-06-28 18:51:36 +03:00
Attila Afra
59e3577cc3
merged 2.2.3
2021-06-28 18:32:17 +03:00
Pirogov, Vadim
c849328945
build: bumped version to v2.2.4
2021-06-14 16:36:55 -07:00
Denis Samoilov
eda1add956
common, gpu: include thread and limit headers to fix GCC 11 build issues
2021-06-14 14:01:34 -07:00
Andrzejewski, Kamil
cc77eef809
benchdnn: reorder: add new configuration to compensation tests
2021-06-09 17:03:08 +07:00
Andrzejewski, Kamil
5d3d1e1874
cpu: reorder: enable unsupported compensation reorder when src_tag=bxa
2021-06-09 17:03:08 +07:00
Andrzejewski, Kamil
4f05b76bb7
cpu: reorder: swap duplicate configuration to the corrected one
2021-06-09 17:03:08 +07:00
Barbieri, Kealan
3d202c2054
src: gpu: jit: ngen: rename gen12 to xe
2021-06-01 17:37:10 -07:00
Primak, Tatyana
7336ca9f05
build: bumped version to v2.2.3
2021-05-28 11:26:07 -07:00
Andrey Kharitonchik
f993b25dbe
cpu: matmul: enable bias + sum support in f32 & bf16 gemm impl
...
In bf16->f32 and f32->f32 gemm-based matmul with bias and sum
post-ops call gemm with accumulation buffer and compute sum in
pos-ops kernel to avoid falling back to reference implementation.
2021-05-25 13:15:53 -07:00
Andrey Kharitonchik
36dba73d0f
cpu: matmul: fix potential memory leak in gemm-based impl
...
avoid return before free() allocated memory for not
parallel_over_batch case with dynamic sizes in int8 and bf16
2021-05-25 13:15:53 -07:00
Xuxin, Zeng
4c0819c432
cpu: x64: reorder: enable optimized path for double blocking
2021-05-24 22:27:32 -07:00
Bogusevschi, Silviu
f3999b71d8
build: apply fp-model=precise for icx and dpcpp
2021-05-17 09:42:25 -07:00
Alok Bakshi
87fd4893ea
cpu: x64: gemm_bf16_ip: no advance in dst/acc pointers in jit prologue
2021-05-14 11:27:47 -07:00
Alok Bakshi
36d25d4308
cpu: matmul: disable gemm matmul with zero stride md
2021-05-14 11:27:47 -07:00
Xuxin, Zeng
fd2c6421f1
cpu: x64: pool: update reorder params in pool when transpose block fmt/ncsp
2021-05-14 11:27:31 -07:00
Xuxin, Zeng
10f8187577
cpu: x64: reorder: add inconsistent padding support in jit reorder
2021-05-14 11:27:31 -07:00
Xuxin, Zeng
23b2ec0d6f
pu: x64: reorder: update the code with uni instruction
2021-05-14 11:27:31 -07:00
Robert Dobrowolski
d51616bc7e
cpu: reorder: correct zero padding for multiblocked kernels
2021-05-12 17:59:19 -07:00
Soto Flores, Manuel
68f05d00ae
src: cpu: x64: reorders: add zero-padding for channel tails
2021-05-12 17:59:19 -07:00
Soto Flores, Manuel
f0e4af9616
src: cpu: x64: avx2: w_conv: fix zero-point buffer offset calculation
...
This fixes an incorrect buffer offset calculation when ngroups
dimension is padded. The error occurred in depthwise convolution
with 1D shapes.
2021-05-12 17:59:19 -07:00
Krzysztof Kozlowski
f3c3daf8a6
gpu: ocl: prelu: fix weight offset calculation
2021-05-11 13:48:37 +02:00
Soto Flores, Manuel
8a784c60fa
src: cpu: x64: dw_conv: fix zero-point buffer offset calculation
...
This fixes an incorrect buffer offset calculation when ngroups
dimension is padded. The error occurred in depthwise convolution
with 1D shapes.
2021-05-05 13:11:39 -10:00
Primak, Tatyana
518b5e0369
build: bumped version to v2.2.2
2021-04-28 08:30:54 -10:00
Roy Oursler
7d78eb240d
gpu: prevent segfaults from unchecked kernel creation failure
2021-04-27 17:08:06 -07:00
Bhavani Subramanian
714b1fd7f9
cpu: x64: brgemm_ip_fwd: fallback to gemm-based ip when oc = 1 for f32
2021-04-27 12:07:33 -10:00
Andrey Kalinin
3555d4a76e
src: cpu: x64: brgemm convolutions: grouped convolutions with small ic should go to direct JIT implementation
2021-04-23 14:34:22 -07:00
Denis Samoilov
1fcb867e37
build: examples: let SYCL to choose C++ standard
2021-04-21 18:43:13 -07:00
Kentaro Kawakami
25c4ceaca3
aarch64: use getauxval for Linux to get CPU features
2021-04-21 13:33:13 -07:00
Sadia, Haleema
9b03091429
fix: gpu: ocl: rnn: fixing buffer access in initializing workspace
2021-04-16 16:32:03 -07:00
Sadia, Haleema
17a95f2c2a
gpu: ocl: rnn: change copyrights to 2021
2021-04-16 16:32:03 -07:00
Wierschem, Keola
6bb1bce3ed
benchdnn: add regression test for 1x1 conv with mixed strides
2021-04-14 17:05:04 -07:00
Wierschem, Keola
d0b3e3fe0b
cpu: x64: brgemm 1x1 conv: fix a bug for mixed strides (eg sh1sw2)
2021-04-14 17:05:04 -07:00
Wierschem, Keola
46f4ee7f84
cpu: x64: brgemm 1x1 conv: refactor check for os blocking
...
Consolidate the check for os blocking into a single location.
2021-04-14 17:05:04 -07:00
Bogusevschi, Silviu
3e40e5e92e
build: tests: add fp-model precise to dpcpp
2021-04-13 13:42:13 -07:00
jreniecki
bcf823c485
gpu: ocl: eltwise: gen9: fix out-of-bounds writes ( #410 )
2021-04-12 11:15:28 +02:00
Pirogov, Vadim
f58682cd8b
build: bumped version to v2.2.1
2021-04-09 17:16:09 -07:00
Denis Samoilov
8b7715faf4
common: fix a size for memcmp
2021-04-09 15:06:10 -07:00
Wierschem, Keola
0326b096ef
fixup: common: make comparison operator for attr nan friendly
2021-04-08 16:49:05 -07:00
Denis Samoilov
315fe46a94
gtests: add a test for comparison operators
2021-04-08 09:02:50 -07:00
Denis Samoilov
e6d05ecf20
common: make comparison operator for op_desc nan friendly
2021-04-08 09:02:50 -07:00
Denis Samoilov
dbca1e9370
common: make comparison operator for attr nan friendly
2021-04-08 09:02:50 -07:00
Dmitrii Zarukin
4c3a114384
src, tests: fix engine creation failure for sub-devices + test
2021-04-07 14:10:41 -07:00
Dmitrii Zarukin
70d70a8d06
common: verbose: fast fix for really long lines
2021-04-07 14:10:41 -07:00
Shustrov, Nikita
a5a0b49561
tests: benchdnn: add regression tests for amx bwd/wu
2021-04-06 18:22:46 -07:00
Shustrov, Nikita
a3a73a3707
cpu: x64: amx: fix of choosing wrong harness in bwd/wu convolution
2021-04-06 18:22:46 -07:00
Piotr Chmiel
9ac85d8508
cpu: x64: binary post-ops: fix reference kernel execution instead
2021-04-01 09:31:53 -07:00
Denis Samoilov
2f39155b25
build: skip msvc version detection for CMake 3.19.0 and 3.19.1
2021-03-31 16:31:43 -07:00
Denis Samoilov
1b026f5e30
build: let SYCL to choose C++ standard
2021-03-31 16:31:43 -07:00
Kentaro Kawakami
4a129541fd
aarch64: add ASIMD isa check
2021-03-30 17:23:06 -07:00
Attila Afra