Megvii Engine Team
39d77fb55a
feat(arm): add arm rnn_cell/lstm_cell/lstm optimized kernel
GitOrigin-RevId: b9bb7352bc
3 years ago
Megvii Engine Team
f509b1be9b
fix(build): split elemwise_multi_type cpp
GitOrigin-RevId: 13267e9db6
3 years ago
Megvii Engine Team
3251f50114
fix(mgb/cuda-stub): add libcuda-wrap_11.4.h to fit the CUDA11.4 toolchain
GitOrigin-RevId: efa38f00d1
3 years ago
Megvii Engine Team
ee0b95e935
feat(dnn/elemwise/arm_common): support part of arm ternary elemwise multithread
BCAST111C_VEC_BCAST111C and BCAST101_VEC_BCAST101
GitOrigin-RevId: 0e26553c90
3 years ago
Megvii Engine Team
cbbca5fb10
feat(mge): add softmax op use cudnn api
GitOrigin-RevId: 7734ebf8c4
3 years ago
Megvii Engine Team
20b42a8c3b
fix(dnn): add naive lstm kernel
GitOrigin-RevId: f08ef810cf
3 years ago
Megvii Engine Team
2faa6ea5a9
Merge pull request #213 from kxz18:rnn
GitOrigin-RevId: 9e9215c115
3 years ago
Megvii Engine Team
82be0aaced
test(dnn): fix compute capability requirement for NCHWX test
GitOrigin-RevId: d2f8022be1
3 years ago
Megvii Engine Team
3b41840b68
fix(mgb): change caffepooling log level
GitOrigin-RevId: 290d243ef5
3 years ago
Megvii Engine Team
1999307015
feat(mgb/opr): add dropout kernel
GitOrigin-RevId: d248bd2005
3 years ago
Megvii Engine Team
32717b0ca4
fix(build): split some cpp, which consume two many mem when build
make build possible at 8G ddr env, when -j8
GitOrigin-RevId: d0c442b41d
3 years ago
Megvii Engine Team
a93741815b
feat(mgb/opr): add layernorm forward and backward kernel
GitOrigin-RevId: 0cd484e753
3 years ago
Megvii Engine Team
a404cd7d06
fix(mgb/src): add tensorRT version check
GitOrigin-RevId: 7abfd30cab
3 years ago
Megvii Engine Team
c53cad2049
feat(cmake): format all cmake file
GitOrigin-RevId: 0a4ecab99b
3 years ago
Megvii Engine Team
a5803058b4
fix(dnn/x86): opt algo order
GitOrigin-RevId: 6dd14f9a96
3 years ago
Megvii Engine Team
93310c0e4b
fix(mgb/gopt): fix cpu global layout transform fastrun error
GitOrigin-RevId: ea254297e5
3 years ago
Megvii Engine Team
c90e0b54be
perf(arm): optimize arm uint16 relayout with n=4
GitOrigin-RevId: 5779c6b9c1
3 years ago
Megvii Engine Team
f6d9909460
feat(dnn): add elemwise multi type support i16xf32 and u8xf32
GitOrigin-RevId: 2fe469bb4e
3 years ago
Megvii Engine Team
d9a46ea47b
fix(dnn): correct behaviour of floor div for int tensor
GitOrigin-RevId: 1444f69cce
3 years ago
Megvii Engine Team
0ad5eeaedd
feat(mgb/gopt): global layout transform support opencl
GitOrigin-RevId: 132605c7d9
3 years ago
kxz@thumt102-1
8f48da7ffe
feat(mgb/opr): add cell level rnn/lstm and sequence level rnn/lstm
3 years ago
Megvii Engine Team
2881934cb8
feat(dnn/check_non_finite): addmul scale to check_non_finite opr
GitOrigin-RevId: c35a219e52
3 years ago
Megvii Engine Team
6bb5409976
feat(dnn/src): add images2neibs kernel of opencl and related test
GitOrigin-RevId: 82242b7437
3 years ago
Megvii Engine Team
6ce4a34403
feat(dnn): add fallback postprocess
GitOrigin-RevId: 4201a0f158
3 years ago
Megvii Engine Team
c96dbd29b8
fix(dnn/arm_common): support more monotonous case in arm typecvt for performance
GitOrigin-RevId: 9e28a64d93
3 years ago
Megvii Engine Team
ead611e11d
perf(dnn): slightly improve arm neon transcendental function performance
GitOrigin-RevId: 210d88f81e
3 years ago
Megvii Engine Team
0d16952470
fix(mgb/cuda): fix conv error when the input tensor is too large
GitOrigin-RevId: 1b1d693795
3 years ago
Megvii Engine Team
02d5f46d90
fix(mgb/x86): fix convbias crash on X86
GitOrigin-RevId: cc7283c6a2
3 years ago
Megvii Engine Team
accb2d8d47
fix(mgb/serialize): fix flatbuffer compatibility issues
GitOrigin-RevId: e4771d6bc4
3 years ago
Megvii Engine Team
5e07e1e0f9
fix(dnn/falback): let cpu be able to execute int4 model
GitOrigin-RevId: 1a6b78f3b6
3 years ago
Megvii Engine Team
2696e4efaa
feat(dnn): add float16 for remap backward
GitOrigin-RevId: 0263030051
3 years ago
Megvii Engine Team
1f0cc891b0
feat(dnn): enable eye to support bool
GitOrigin-RevId: 76d874d5b7
3 years ago
Megvii Engine Team
11d75fecb5
feat(dnn/check_non_finite): add batch check_non_finite
GitOrigin-RevId: e108133282
3 years ago
Megvii Engine Team
2318ea3f15
fix(dnn): fix naive average pooling overflow bug for int8 type
GitOrigin-RevId: b60a7b6cf8
3 years ago
Megvii Engine Team
2d54ad185b
feat(lite): add global layout transform interface for load and run
GitOrigin-RevId: 65c2430ec2
3 years ago
Megvii Engine Team
ba2f0c2e48
fix(dnn/cuda): fix cudnn_conv algo of conv_bias opr for fp16 add z cases
GitOrigin-RevId: b29b009de0
3 years ago
Megvii Engine Team
30976c239f
fix(mgb/gopt): fix global layout transform
add a special opr_format modify function for concat operators to modify concat axis when input's layout has been changed
GitOrigin-RevId: 4094208057
3 years ago
Megvii Engine Team
ca7cec7a5d
fix(mgb/gopt): minor fixes for global layout transform
Merge Target::ARM and Target::X86 into Target::CPU to make global layout transform easier to use
GitOrigin-RevId: cc9363fa38
3 years ago
Megvii Engine Team
fe93013a6e
feat(mgb/gopt): global layout transform support nchw_nchwxx hybrid mode
GitOrigin-RevId: 6d5b55d7fc
3 years ago
Megvii Engine Team
3d45d35241
feat(mgb/gopt): profiler support checking algo availability
GitOrigin-RevId: 39cad612cc
3 years ago
Megvii Engine Team
b59e8ccf24
fix(mgb): fix cambricon bangc copybara
GitOrigin-RevId: 22b57364e8
3 years ago
Megvii Engine Team
3116e128c5
fix(ci/integration_test): fix benchmark torch version
GitOrigin-RevId: bd964ed505
3 years ago
Megvii Engine Team
c85631aa77
feat(dnn): use ref ptr interface for all backends
GitOrigin-RevId: f65feae5cc
3 years ago
Megvii Engine Team
d90cb7763c
feat(src/core): record support change ptr basic
GitOrigin-RevId: b12f1c4a66
3 years ago
Megvii Engine Team
89186edc5d
fix(dnn): correct reduce/argmxx/fakequant calculation with nan
GitOrigin-RevId: 7e78bdae91
3 years ago
Megvii Engine Team
68cdabd288
feat(opr): indexing_multi_axis_vec support nd index
GitOrigin-RevId: 07b1248bdc
3 years ago
Megvii Engine Team
a1cba6cc27
fix(dnn): fix convbias crash on X86
GitOrigin-RevId: 3b33c83d03
3 years ago
Megvii Engine Team
9b4cd92ba3
fix(mgb/dnn): fix cudnnConvBiasActivation crash on nchw32 int8 with oc > 256
GitOrigin-RevId: 20c0b90575
3 years ago
Megvii Engine Team
23c1fda7e6
perf(arm_common): optimize sigmoid
GitOrigin-RevId: 7cb248a15b
3 years ago
Megvii Engine Team
25ec2530ba
feat(whl/api/lar): enable megengine dll on Windows
1: reduce python whl package size
2: unify api link logic on all OS
3: add option: MGE_WINDOWS_BUILD_WITH_STATIC_CRT
--- default OFF
--- if build CRT(vc runtime) with STATIC with megengine.dll
some CRT api will crash, for example, flush, so if you
build with static megengine, and do not want to install CRT
you can set MGE_WINDOWS_BUILD_WITH_STATIC_CRT TRUE
--- how to install CRT:
https://docs.microsoft.com/en-us/cpp/windows/latest-supported-vc-redist?view=msvc-160
install VC_redist.x64.exe
4: rename megengine_export to megengine_shared(only export needed symbols ),
caused by runtime symbols conflict with pytorch
GitOrigin-RevId: 93d8d80f29
3 years ago