He Wei
7d9a783993
[auto-monad] Support side-effects by auto-monad
The basic idea is: exploits data dependency to control the execution order
of side-effect operations, and keep the semantics of ANF unchanged.
The ControlDepend primitive is removed and there are two primitives added:
1. UpdateState:
```
a = Assign(para, value)
```
became:
```
a = Assign(para, value, u)
u = UpdateState(u, a)
```
2. Load:
```
x = Add(para, value)
```
became:
```
p = Load(para, u)
x = Add(p, value)
u = UpdateState(u, p)
```
5 years ago
buxue
81d26eaa34
unify parameter output type
5 years ago
buxue
6ccc4379b4
do not broaden scalar
5 years ago
buxue
6d395c1d3f
keep consistent in Graph mode and PyNative mode for 'isinstance'
5 years ago
mxm
82268dc703
support DynamicShape and StridedSlice primitive return range of shape value.
5 years ago
mxm
59f9ac9975
Primitive 'AddN' support dynamic shape
5 years ago
huanghui
bda74495f3
Add trace mangager around front opt
5 years ago
mindspore-ci-bot
57899791d3
!9063 Throw exception if using call stack for CNodes in the same func graph.
From: @zh_qh
Reviewed-by: @stsuteng,@chenfei52
Signed-off-by: @stsuteng
5 years ago
Zhang Qinghua
b43d71c130
Throw exception if using call stack for CNodes in the same func graph.
5 years ago
mindspore-ci-bot
617c175a98
!8854 Fix some redundant codes
From: @chenfei52
Reviewed-by:
Signed-off-by:
5 years ago
chenfei
f154abdff8
change some redundant code
normalize codes of session
5 years ago
Zhang Qinghua
85f8290dce
Optimize TopoSort() and Infer() performance.
5 years ago
Zhang Qinghua
a456415cc1
Modify for infer performance. Should add more optimization later.
5 years ago
mindspore-ci-bot
dbcdda18ed
!8293 Resolve specialize error during transforming after block in PyNative mode.
Merge pull request !8293 from 张清华/grad_opt2
5 years ago
Zhang Qinghua
4e6e68f187
Resolve specialize error during transforming after block in PyNative mode.
5 years ago
Yi Huaijie
d7faa77b5e
support int64 shape
5 years ago
Zhang Qinghua
f1a851acb3
Resolve GPT net stack overflow problem, due to infer process.
5 years ago
Zhang Qinghua
077bde0767
Return a new abstract without tracking_id for fg ValueNode in CSE.
5 years ago
caifubi
d3b978147f
Ascend Dynamic Shape
5 years ago
buxue
483c8a179a
improve the recognition of Parameter object and raise error when convert keywordarg to pydata
5 years ago
mindspore-ci-bot
c57a472748
!5918 optimizer pynative memory
Merge pull request !5918 from flywind/optimizer_pynative_memory
5 years ago
buxue
458498900c
support not in and add check for grad_with_sens with no sense provide.
5 years ago
kpy
4338dd266e
optimizer pynative graph memory
5 years ago
buxue
3fd73f9d08
raise exception when use HookBackward in graph mode
5 years ago
fary86
b0f89685b4
Add device specific config check
5 years ago
buxue
a087ef7c34
raise exception when UniformPrimitive has wrong number args
5 years ago
Wei Luning
879a519136
updata signature
5 years ago
fary86
144a35b17e
Adapt GatherV2 for dynamic shape
5 years ago
fary86
fcbb3e0edc
Refactor ms_context implementation
5 years ago
zhousiyi
c25e37e7bf
make backend/optimizer pybind free
5 years ago
Wei Luning
1a2eb5c8f7
fix time cost
5 years ago
mindspore-ci-bot
5b1cf18cb9
!5055 prepare to support int64
Merge pull request !5055 from lirongzhen1/int64
5 years ago
lirongzhen1
531ad4df70
prepare to support int64
5 years ago
fary86
947e19b839
Fix bug of switch layer join
5 years ago
fary86
04524b6bd3
Fix coredump caused by function call depth too large
5 years ago
zhousiyi
d0e58dd765
remove ccsrc/common.h
replace frontend/operator/ops.h in backend with base/core_ops.h as
backend should not use any frontend-only primitive
5 years ago
zhoufeng
663278112f
optimize code compile performance
Signed-off-by: zhoufeng <zhoufeng54@huawei.com>
5 years ago
huangdongrun
2a6d346d2f
support if by if grad parameter
add join for ref
adjust env eliminate to eliminate all env ops
add partial app cache
resolve while endless
fix env eliminate
support for "for while" cases
fix join shape error
5 years ago
liubuyu
45b6600001
mv cc file that end with py to pybind
5 years ago
WilliamLian
1f5441d73a
remove inferfunction to core
5 years ago
buxue
02d6e3a43a
fix bugs
5 years ago
Wei Luning
ea3ddea3d4
remove ref origin
5 years ago
panyifeng
34e50e5d6e
fix cell bprop
5 years ago
fary86
7602054acd
Fix do concat in while loop specialize error
5 years ago
fary86
8ac5672abb
Add support for dynamic shape
5 years ago
panyifeng
9927e6eb5c
eager mode sparse
5 years ago
liubuyu
d81862a916
decoupling core and context
5 years ago
mindspore-ci-bot
2da29bce66
!3370 same graph used in different context should be treat differently.
Merge pull request !3370 from xychow/fix-context-dup
5 years ago
buxue
b075674cf2
support tensor attr shape and dtype in graph mode
5 years ago
zhousiyi
f926650c64
if AbstractFunction comparison succeed in NewContext, then the evaluator should use the same one, otherwise one of the evaluator will not be evaluated.
if funcgraph or metafuncgraph call it recursively, then anf_node should be used as tracking_id to discriminate the first occurcance and the
recursive occurance.
add anf_node to PrimitiveAbstractClosure hash() to reduce cost of GetEvaluatorFor().
ignore the tracking_id to make cse work.
5 years ago