Col_lin
/
mindspore2022
forked from huawei/mindspore2022
0
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
22,898 Commits 18 Branches 22 Tags 3.6 GiB
Commit Graph

96 Commits

Author SHA1 Message Date
hanhuifeng2020 bc46d644fe [GraphKernel]Support reshape/elewise/broadcast+transdata fusion 2021-05-08 11:42:14 +08:00
mindspore-ci-bot 0887d35b1c !15951 [GraphKernel]add the attribute reduce_output_fuse to enable fuse for the reduce_output on Ascend 
From: @hanhuifeng2020
Reviewed-by: @gaoxiong1,@dylangeng
Signed-off-by: @dylangeng
 2021-05-07 11:38:06 +08:00
mindspore-ci-bot ed539597c2 !15415 [GraphKernel]adapt for logsoftmax in ascend 
From: @wenfangpei
Reviewed-by: @gaoxiong1,@ckey_dou,@gaoxiong1,@ckey_dou
Signed-off-by: @ckey_dou,@ckey_dou
 2021-05-07 11:19:58 +08:00
wenfangpei 4174a7b38f expanders of some fusion ops 2021-05-06 10:23:41 +08:00
wenfangpei db8256e61f adapt for logsoftmax in ascend 2021-05-06 10:13:28 +08:00
mindspore-ci-bot 1827697642 !15961 Eliminate recursion call in split model 
From: @gaoxiong1
Reviewed-by: @ckey_dou,@dylangeng
Signed-off-by: @dylangeng
 2021-05-06 10:06:35 +08:00
wenfangpei c41875b318 adapt expanders of some ops from gpu to ascend 2021-05-04 16:32:45 +08:00
Gaoxiong 4bc67f38de eliminate recursion call 2021-04-30 16:26:27 +08:00
hanhuifeng2020 425d401e85 [GraphKernel]add the attr reduce_output_fuse to enable fuse for reduce_output on Ascend 2021-04-30 11:48:23 +08:00
mindspore-ci-bot 52e7f51970 !15741 [GraphKernel] batchnorm expander supports when first input is float16 
From: @looop5
Reviewed-by: @gaoxiong1,@dylangeng
Signed-off-by: @dylangeng
 2021-04-27 16:27:29 +08:00
mindspore-ci-bot ff75ce8ac4 !15097 [graph kernel] clean code for expanders. 
From: @chenlei_autodiff
Reviewed-by: @gaoxiong1,@dylangeng
Signed-off-by: @dylangeng
 2021-04-27 16:26:18 +08:00
looop5 24f441ba33 batchnorm expander supports when first input is float16 2021-04-27 09:46:42 +08:00
mindspore-ci-bot d6f58cb765 !15658 Reduce recursion overhead of split model 
From: @gaoxiong1
Reviewed-by: @anyrenwei,@dylangeng
Signed-off-by: @dylangeng
 2021-04-26 16:39:04 +08:00
chenlei_autodiff fd227bb448 [graph kernel] clean code for expanders. 2021-04-26 16:28:37 +08:00
mindspore-ci-bot 168c64b60d !15648 [GraphKernel] negative axis in Squeeze expander. 
From: @chenlei_autodiff
Reviewed-by: @gaoxiong1,@anyrenwei
Signed-off-by: @anyrenwei
 2021-04-26 14:18:21 +08:00
Gaoxiong 71002ed19d reduce recursion overhead of split model 2021-04-25 21:27:09 +08:00
zengzitao 8dcff8d83c refactor tile op and in expander open on gpu 2021-04-25 17:42:37 +08:00
chenlei_autodiff b419f60b0d [GraphKernel] negative axis in Squeeze expander. 2021-04-25 17:31:04 +08:00
r1chardf1d0 5c5d125b1d optimize stitch fusion strategy 2021-04-25 10:58:19 +08:00
hanhuifeng2020 25505642ce enable GraphKernel for TransData 2021-04-16 16:23:15 +08:00
lingyunli63 c48c2430f0 fuse matmul and elementwise in graphkernel 2021-04-14 20:52:35 +08:00
mindspore-ci-bot cd002cb7f7 !14893 enable stitch fusion on bert 
From: @r1chardf1d0
Reviewed-by: @gaoxiong1,@ckey_dou
Signed-off-by: @ckey_dou
 2021-04-12 10:39:00 +08:00
mindspore-ci-bot 5b4685c5ea !14604 [GraphKernel] add some expander ops 
From: @chenlei_autodiff
Reviewed-by: 
Signed-off-by:
 2021-04-12 10:33:53 +08:00
r1chardf1d0 3b32995936 enable stitch fusion on bert 2021-04-09 18:29:55 +08:00
chenlei_autodiff 13fbfca6b9 [graph kernel] add expander ops. 2021-04-09 17:33:06 +08:00
wenfangpei b9715db358 bugfix in expanders of layernorm 2021-04-09 17:31:32 +08:00
wenfangpei 66d28af79e adapt for layernorm in ascend 2021-04-08 18:11:59 +08:00
mindspore-ci-bot b5bc938deb !12914 [GraphKernel]expander lamb_apply_weight_assign 
From: @wenfangpei
Reviewed-by: @anyrenwei,@gaoxiong1,@gaoxiong1
Signed-off-by: @anyrenwei
 2021-04-06 17:02:16 +08:00
wenfangpei 83399c1b8d adapt for layermorm C++ code 2021-04-06 11:06:50 +08:00
wenfangpei a4ad6066b1 expander lamb_apply_weight_assign 2021-04-02 17:28:59 +08:00
mindspore-ci-bot ddf75da542 !14085 [GraphKernel] add some expander ops 
From: @chenlei_autodiff
Reviewed-by: 
Signed-off-by:
 2021-03-30 15:33:06 +08:00
mindspore-ci-bot 75fdaaa6aa !14304 [GraphKernel] Dump GraphKernel split info as text; dump akg kernel launch fail message 
From: @dayschan
Reviewed-by: @gaoxiong1,@gaoxiong1,@anyrenwei
Signed-off-by: @anyrenwei
 2021-03-30 15:32:14 +08:00
chenlei_autodiff f4289d40f3 add graph kernel expander ops. 2021-03-30 11:19:40 +08:00
dayschan 3c6c30024c dump graph_kernel_split info 2021-03-29 17:15:02 +08:00
tronzhang 87bf1ec80f delete mark_interface_fusion and tensor reuse frontend pass for graph kernel 2021-03-27 11:45:04 +08:00
lingyunli63 4b966ed40d support matmul on D 2021-03-26 15:57:26 +08:00
huangbingjian 72ae1799f3 remove control_depend from py file 2021-03-22 20:45:45 +08:00
mindspore-ci-bot 5b95409022 !13512 add some expander ops 
From: @zengzitao
Reviewed-by: 
Signed-off-by:
 2021-03-19 15:00:23 +08:00
mindspore-ci-bot 2fadad0875 !13121 expander lamb_apply_optimizer_assign 
From: @wenfangpei
Reviewed-by: 
Signed-off-by:
 2021-03-19 14:59:37 +08:00
wenfangpei 043a558ae2 expander lamb_apply_optimizer_assign 2021-03-19 09:54:10 +08:00
zengzitao d0a656f3cd add some expander ops 2021-03-18 19:03:36 +08:00
dayschan a2967330ea Normalize the Reduce nodes' axis in GraphKernel 
changed all reduce axis to non-negative value
 2021-03-16 15:55:50 +08:00
tronzhang 7252ffb66b pipeline optimization for parallel fusion 2021-03-09 17:07:24 +08:00
mindspore-ci-bot d285692217 !12852 Change maketuple in coreops 
From: @liangzhibo
Reviewed-by: 
Signed-off-by:
 2021-03-05 19:13:42 +08:00
l00591931 680324f225 Change make tuple in core.ops 2021-03-05 10:54:33 +08:00
dayschan 454500309c add OpInfer for op Select 2021-03-05 10:21:28 +08:00
dayschan 7beca18f3c Refactor GraphKernelExpander (3rd submission) 
Rewrited the OpInfer as a class with functions "infer_shape","infer_type"
and "infer_format". Used op name to find the subclass that have implemented
these functions. Two common class "_Elemwise" and "_Reduce" was provided.

Op BiasAddGrad supported "FractalNZ" format.
 2021-02-25 19:14:46 +08:00
dayschan 9d572f3963 Refactor GraphKernelExpander (2nd submission) 
Changed the expander to class, and used the class name to identify the operator.
Moved the original op logic into the `_expand` function; added a `_check` function to check op inputs.

Use decorator to register the whitelist formats that operator supports,
and the decorator will change the `_check` function to check the formats.
 2021-02-22 14:49:21 +08:00
dayschan e0e6c39eae Refactor GraphKernelExpander (1st submission) 
Decoupled from the front-end interfaces.
1. Removed the call to "Renormalize".
   Completed the infer-format in model_builder.
   Only used the device shape and device format to
   infer an abstract shape without considering padding.
2. Removed the call to python's Primitive interfaces.
   The "Renormalize" relies on the PrimitivePy, so they can be
   removed together. After that, the functions "ConstAttrToInput",
   "DeleteAttrInInput" and related can be removed.
3. Reuse the AkgKernelJsonGenerator in GraphKernelExpander.
   1) set the attribute "extract_opinfo_from_anf" to true, so that
   the generator can handle the basic operator with anf info.
   2) added a function "extract_expand_info" in expander.py
   to convert the json into a more friendly format. The attrs
   was converted to a dict instead of a list.
4. Scalars only support DefaultFormat.
   Removed the argument "format" from graph_builder.value
5. Moved the expander op list from graph_kernel_helper.cc to graph_kernel_expander.cc
 2021-02-07 10:48:23 +08:00
jinyaohui 30a27b2adb modify Gelu、FastGelu to GeLU and FastGeLU 2021-02-05 17:19:52 +08:00