yujianfeng
3df6845803
Fix fusion condition of reshape and transpose
5 years ago
mindspore-ci-bot
32405f9ab3
!2817 add internal output
Merge pull request !2817 from kisnwang/optimize-sub-graph-memcpy
5 years ago
kswang
e9067b4a10
add internal output
5 years ago
chujinjin
c880774942
change pynative batchnorm same as graph mode
5 years ago
He Wei
43e0967024
Decouple ir::Tensor class from python
5 years ago
mindspore-ci-bot
8b5166e569
!2393 fix bug of hccl kernel info and change cast's kernel info
Merge pull request !2393 from lianliguang/fix-bug-of-merge-cast-to-op-and-gene-hccl-kernel-info
5 years ago
gong chen
a6dfa281ea
Init GraphKernel.
- It provides a unified style to express graph and kernel for user.
- It provides a unified IR to represent graph and kernel for developer.
- It breaks the boundary between graph and kernel.
- It provides more opportunities to do compile optimization.
5 years ago
mindspore-ci-bot
a3e7b30457
!2384 Add split fission pass
Merge pull request !2384 from YuJianfeng/split
5 years ago
mindspore-ci-bot
24be3f82ad
!2336 Fix bug of not insert memcpy when hccl_op has more than one input
Merge pull request !2336 from huanghui/insert-memcpy-async-pass
5 years ago
WilliamLian
5f9d2759ee
fix bug of hccl kernel info and change cast's kernel info
5 years ago
yujianfeng
7ad877a948
Add Split fission pass
5 years ago
WilliamLian
ee79023e86
clean pclint warning
5 years ago
huanghui
c66fe00049
fix bug of not insert memcpy when hccl_op has more than one input
5 years ago
mindspore-ci-bot
230963d08e
!2332 Fix ConfusionSoftmaxGrad fusion pass work if the ReduceSum's attr of keep_dims set with False
Merge pull request !2332 from huanghui/fix-ConfusionSoftmaxGrad-fusion-pass
5 years ago
mindspore-ci-bot
2ad211df97
!2293 fix bug of transdata
Merge pull request !2293 from lianliguang/add-broadcast-and-reduce-kernel-select-rules
5 years ago
huanghui
c7bda5363d
fix ConfusionSoftmaxGrad fusion pass work if the ReduceSum's attr of keep_dims set False
5 years ago
huanghui
8463731bcc
make those AdamXX and LambXX fusion pass not work for unexpect data type
5 years ago
WilliamLian
e3a26c2229
fix transdata insert cast
5 years ago
WilliamLian
d3ea1e4298
fix code review
5 years ago
mindspore-ci-bot
294449697a
!2179 Add sort by index for each group of AllReduce
Merge pull request !2179 from YuJianfeng/all_reduce_index
5 years ago
yujianfeng
f15cb6b7c9
Add sort by index for each group of AllReduce
5 years ago
mindspore-ci-bot
04a23d138f
!2164 add more needtrans format for transdata
Merge pull request !2164 from lianliguang/master
5 years ago
WilliamLian
5d25bf7ca2
add more transform format insert transdata
5 years ago
huanghui
1243022f01
fix the back outputs of BNTrainingUpdateV3 may cover the front whitch is empty
5 years ago
lvliang
075da9a4b1
pynative-insert-transdata-for-hook-mode
5 years ago
mindspore-ci-bot
87371be65d
!2127 Fix exception when FusedBatchNorm's first input 's shape dims not equal 4
Merge pull request !2127 from huanghui/fix-fusebatchnorm-split
5 years ago
huanghui
ffe074b661
fix exception when FusedBatchNorm's first input 's shape dims not equal 4
5 years ago
mindspore-ci-bot
971f10d222
!1790 remove transdata only connected with control depend
Merge pull request !1790 from lianliguang/remove-the-useless-transdata-connected-with-the-control-depend
5 years ago
WilliamLian
b86016a26f
remove the useless transdata and cast of control depend node
5 years ago
huanghui
4acb61d59d
code review fix for buffer fusion
5 years ago
etone-chan
8f4e58472e
Change optimization pass order
5 years ago
jjfeing
caab25e09b
tbe select broadcast reduce dynamic
5 years ago
mindspore-ci-bot
3b809f2b0b
!1996 Fix single BatchNorm fission && SoftmaxGradExt fusion pass
Merge pull request !1996 from huanghui/single-batchnorm-fission-pass
5 years ago
huanghui
88eec2b894
fix single-batchnorm-fission && softmax-grad-ext-fusion pass
5 years ago
mindspore-ci-bot
dae35e0a71
!2048 fix bug of multioutput byway fusion pass
Merge pull request !2048 from Etone.Chan/June
5 years ago
etone-chan
23ba6291cc
fix bug of multioutput byway fusion pass
5 years ago
huanghui
1044310783
add op names mapping for one hwopt pass
5 years ago
huanghui
118496b3ec
enhance insert memcpy
5 years ago
liubuyu
0fd57bd15e
fix remove reshape pair pass
5 years ago
mindspore-ci-bot
bfc18f3adc
!1873 synchronize latest ascend software 04 Jun 2020
Merge pull request !1873 from yanghaoran/master
5 years ago
yanghaoran
8da4c1a763
synchronize latest ascend software 04 Jun 2020
5 years ago
mindspore-ci-bot
0624263496
!1940 Fix when batchnorm has just 4 outputs, it will be not fission
Merge pull request !1940 from huanghui/single-batchnorm-fission-pass
5 years ago
huanghui
36d1aadf1c
fix when Batchnorm's output is 0,1,2,4, fission doesn't work
5 years ago
WilliamLian
9808e47663
change checkAicpu to CheckAICPU & add charge Scalar function to charge the input or output is scalar
5 years ago
mindspore-ci-bot
686ced85ad
!1785 add dropout special kernel selected rules
Merge pull request !1785 from lianliguang/add-dropout-kernel-special-kernel-select-rules
5 years ago
mindspore-ci-bot
1b6816a851
!1880 add new buffer fusion pass
Merge pull request !1880 from Etone.Chan/June
5 years ago
huanghui
b4c0ed4b36
add signle batchnorm fission pass
5 years ago
WilliamLian
9c005b182c
convert dropoutdomask nodes's kernel info's first format which connected with same dropoutgenmask
5 years ago
etone-chan
1525fae8b9
add new buffer fusion pass
5 years ago
mindspore-ci-bot
10ebd81b10
!1845 convert parameter & valuenode's device type
Merge pull request !1845 from lianliguang/master
5 years ago