Commit Graph

28 Commits

Author SHA1 Message Date
zhoufeng b7bc99debf move head files to include
Signed-off-by: zhoufeng <zhoufeng54@huawei.com>
2023-03-08 10:44:43 +08:00
wYann cdd61d758a fix some bugs of kernel object type selection 2023-01-17 20:52:53 +08:00
gaoyong10 9eacdb70c7 Mv GetOutputTensorNum to backend anf_runtime_.cc 2022-12-13 10:36:59 +08:00
jjfeing 82f04825f3 remove enum fusion type 2022-12-08 11:19:48 +08:00
lby 1aa76afba5 decoupling plugin and runtime 2022-07-20 10:48:18 +08:00
ckey_Dou 0f4901b3ed shape function refactor: 2022-06-17 23:36:41 +08:00
gaoyong10 c2d67a1ba9 Fix multi collect output for construct of tuple-tuple-getitem 2022-04-22 17:22:14 +08:00
zhoufeng f49b195c39 extract common as an independent shared library
Signed-off-by: zhoufeng <zhoufeng54@huawei.com>
2022-02-26 10:55:51 +08:00
marui d47e79b04c Refactor ccsrc directories and CMakeLists files 2022-02-13 14:37:51 +08:00
dayschan cbb84ff580 Move IsRealKernel and IsRealCNodeKernel from AnfAlgo to AnfUtils
the function IsOneOfPrimitive and IsOneOfPrimitiveCNode is useful,
we can move them into anf.cc
2021-11-09 14:34:21 +08:00
LaiYongqiang 898a48d6ea fix fusion type 2021-07-13 10:44:50 +08:00
dingpeifei 87e41aaeee IR operators of GPU and CPU are unified as batchnorm 2021-03-18 19:02:28 +08:00
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)
```
2021-02-08 09:01:15 +08:00
mindspore-ci-bot e897eb4c41 !11915 Change TensorAdd to Add, merge from r1.1 to master
From: @liangzhibo
Reviewed-by: @ginfung,@zh_qh
Signed-off-by: @zh_qh
2021-02-02 09:16:08 +08:00
l00591931 9ec100d069 Change TensorAdd to Add, from r1.1 to master 2021-02-01 17:53:52 +08:00
liubuyu 4a3cbda322 get cube size by dtype 2021-01-29 20:08:56 +08:00
Yi Huaijie d7faa77b5e support int64 shape 2020-11-05 21:15:49 +08:00
Wei Luning c1c30a44f1 rename param_value -> param_info 2020-08-18 11:16:28 +08:00
Wei Luning a05c38bb63 make python Parameter inherit from Tensor 2020-08-03 09:12:24 +08:00
WilliamLian d10d1a17f0 spilt valuenode & parameter's tuple output to maketuple 2020-07-20 09:48:03 +08:00
He Wei ee1510da41 Eliminate circular dependency between 'ir' and 'device/kernel' 2020-07-14 19:56:22 +08:00
liubuyu 43c79eb853 mindspore path adjust 2020-07-14 18:07:28 +08:00
He Wei f337c6bc14 Decouple ParamValue from python 2020-07-09 16:16:22 +08:00
zhoufeng 1f8b00dff2 Fix empty graph dump ir
Signed-off-by: zhoufeng <zhoufeng54@huawei.com>
2020-07-04 14:26:52 +08:00
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.
2020-06-20 22:31:54 +08:00
leopz 40e15996b0 move default_param out of parameter and remove pybind11 in anf define 2020-05-19 19:11:12 +08:00
lianliguang 5d225f934f change the padding strategy & refactor insert transdata 2020-04-14 12:57:40 +08:00
zhunaipan 930a1fb0a8 initial version
Signed-off-by: leonwanghui <leon.wanghui@huawei.com>
2020-03-27 22:54:54 +08:00