Commit Graph

45 Commits

Author SHA1 Message Date
liangyongxiong 9f6b015032 RDR adapts for CPU dynamic memory allocation 2021-08-26 11:13:19 +08:00
louie5 b1fe84faec release resource reference by RDR at the beginning of C++ interface 2021-07-22 09:37:12 +08:00
wangjunbao 2367337c2e Modify the warning message when not enabling RDR 2021-06-22 16:01:49 +08:00
wangjunbao 62e5e5bdb5 Modify the warning message when not enabling RDR 2021-06-22 14:17:14 +08:00
i-robot 7da28cef6f !17873 [MS][LOG] Make directorys when the path set by GLOG_log_dir is not exist
Merge pull request !17873 from louie5/master
2021-06-09 20:57:45 +08:00
louei5 5f51d03962 Make directory when the path set by GLOG_log_dir is not exist. 2021-06-07 17:29:15 +08:00
zhoufeng 658c2072fc owners config
Signed-off-by: zhoufeng <zhoufeng54@huawei.com>
2021-06-07 11:03:33 +08:00
louei5 44ff50f6aa make rdr support recording gpu memory info in Actor framwork 2021-06-05 17:35:28 +08:00
louei5 6f82ec5fc9 fix pclint & codecheck warnings 2021-05-29 09:41:40 +08:00
Margaret_wangrui a7859ecc45 clean code check 2021-05-27 14:40:42 +08:00
luopengting 786eb319a1 fix ci warning:
1. Use 'const' for member functions that do not modify member variables
2. Use the 'override' keyword when overriding virtual functions
2021-05-11 09:26:50 +08:00
luopengting 61f2f19d9a make GetSubModuleName() as common function
Fix wrong module name of rdr files, make GetSubModuleName() as a common
function.
If rdr also has SUB_MODULE_NAMES, then if log_adapter modifies the name
of the module, the module name obtained by the rdr module will be wrong.

Mainly involves the following modifications:
1. move GetSubModuleName() from log_adapter.cc to log_adapter_common.cc
2. delete GetSubModuleName() in rdr module
2021-04-21 16:13:43 +08:00
luopengting 727dc08bfa fix TAINTED_SCALAR and capitalize constants
1. fix the tainted variable 'path'
2. capitalize constants in env_config_parser
3. move constants about attribute to common utils.h
2021-04-08 14:22:12 +08:00
luopengting facff39ff8 fix codedex in RDR
1. check whether path is empty in IsEveryFilenameValid().
2. do not print the string in IsStrLengthValid(), it's not safe.
3. make the constants about hash in recorder_manager.h.
4. initialize some variables.
2021-04-02 14:39:14 +08:00
louei5 f23ce6c7d9 optimize record gpu memory information 2021-03-26 11:08:35 +08:00
mindspore-ci-bot 0da0f72c6c !13329 [MS][RDR] optimize saving FuncGraph to using same params with save_graphs
From: @louie5
Reviewed-by: @zh_qh
Signed-off-by:
2021-03-22 08:51:30 +08:00
louei5 fd51088da8 fix rdr saving FuncGraph and KernelGraph different with save_graphs 2021-03-20 11:46:16 +08:00
luopengting 23ba47df6c add MS_RDR_ENABLE and MS_RDR_PATH
MS_RDR_ENABLE should be in {0, 1}. if it is set as other value, it will
be 0 and WARNING information will be printed to tell user.
The settings in config file set by the user takes precedence over the
environment variables. If 'rdr_enable' is 0 in environment variable,
and then 'enable' is true in config file, the RDR will turned on.
2021-03-18 15:48:59 +08:00
luopengting c8ba7694c5 refactor RDR to support single name
1. support single name
2. add hash method for pair
3. move constructor and destructor of MemAddressInfo as public
4. remove graph_id
5. modify interval for somas info
2021-03-17 11:34:05 +08:00
louei5 46d78efb0c fix saving too much gpu memory info and not skip saving kernel graph in protobuff type 2021-03-11 15:11:32 +08:00
luopengting 027edbb679 fix rdr log when env_config_path is empty 2021-03-09 17:32:06 +08:00
mindspore-ci-bot 68eb5125c6 !12905 reduce unnecessary blank lines, add destructor for recorders
From: @luopengting
Reviewed-by: @wenkai_dist,@ouwenchang
Signed-off-by: @wenkai_dist
2021-03-09 14:18:56 +08:00
luopengting e4e04c81ef reduce unnecessary blank lines, add destructor for recorders 2021-03-09 11:09:54 +08:00
mindspore-ci-bot 52909e6d64 !12894 [MS][RDR] Fix bugs in recording gpu memory information by RDR
From: @louie5
Reviewed-by: 
Signed-off-by:
2021-03-08 11:31:04 +08:00
louei5 6ed2ab2976 fix bugs in recording gpu memory information by RDR 2021-03-04 22:48:48 +08:00
luopengting db5de7fd28 add trigger point for distribute task failed
1. add trigger point
2. reduce unuse import
3. modify GetRealFilePath
2021-03-04 16:52:28 +08:00
louei5 99203038a5 support recording gpu memory information and graph execute order 2021-03-03 22:04:31 +08:00
luopengting 7914363d25 optimize rdr mainly to reduce the redundant code
1. move common function to base recorder
2. add UpdataRdrEnable() for recorder manager, and update once
3. add delimiter virable for base recorder, set it as '.'
4. modify recorders to use the suffix parameter of GetFileRealPath()
5. improve the text in ms_context

Can not add RdrEnablePlatform because linux is supported only.
If os is linux, (!is_rdr_supported && rdr_enable) will be false always,
else, (is_rdr_supported && !rdr_enable) will be false always.
2021-03-01 09:32:57 +08:00
louei5 9a48405a41 recording func_graph in pipeline and task debug information 2021-02-26 19:20:05 +08:00
louei5 3d540a515a add task_debug_info recorder 2021-02-24 14:19:24 +08:00
mindspore-ci-bot efbda3bec5 !12534 Ignore renormalization if no changes made after last renormalization, and release all data after finished RDR checking.
From: @zh_qh
Reviewed-by: @ginfung,@hwhewei
Signed-off-by: @ginfung
2021-02-23 18:32:43 +08:00
Zhang Qinghua 88e4dbeb04 Ignore renormalization if no changes made after last renormalization.
Release all data after finished RDR checking.
2021-02-23 15:26:23 +08:00
mindspore-ci-bot b4a3f3c1ba !12490 support recording stream execute order
From: @luopengting
Reviewed-by: 
Signed-off-by:
2021-02-23 15:01:58 +08:00
mindspore-ci-bot 590e889cba !12265 [MS][RDR] optimize saving FuncGraph
From: @louie5
Reviewed-by: 
Signed-off-by:
2021-02-23 14:26:14 +08:00
louei5 d1d5976ca2 using rdr module in proper position 2021-02-23 09:58:41 +08:00
luopengting 7fe901491a support recording stream execute order
1. add ExecNode and StreamRxecOrderRecorer
2. add json format and json dump for stream exec order
3. add stream exec order api
4. add in CMakeLists
2021-02-22 11:44:36 +08:00
xuxs c70049b190 mindquantum
mindquantum

cmake lint

projectq patch

test pqc

space

pylint

pqc python ops

doc

check

bug

bug

bug

bug

check boundary condition

mindquantum in log_adapter

mindquantum in adapter and recorder cc

substract to subtract

public variable to private variable

public variable to private variable

commit

debug

debug

const in utils

delete dtol

clint
2021-02-20 12:33:51 +08:00
luopengting b4d56f6358 add functions to check for directory and filename
1. add validation for directory and filename
2. refactor the file operations about path
3. add in CMakeLists
4. add logger in nn_ops
2021-02-08 11:50:02 +08:00
louei5 72e567d06b change saving filename in RDR module 2021-02-07 15:31:27 +08:00
louei5 762feb591c Add RDR module compiling switch and check saving filename 2021-01-30 11:29:27 +08:00
mindspore-ci-bot 5b42996bc5 !11698 Add timestamp for RDR module
From: @louie5
Reviewed-by: 
Signed-off-by:
2021-01-29 14:40:47 +08:00
luopengting 66961f2f1d add StringRecorder for RDR, record somas info:
1. add string recorder
2. record somas
3. remove SomasInfoRecorder
4. fix clang and cpplint
2021-01-29 09:46:51 +08:00
louei5 f00032145f Add timestamp into filename of saving recorder file and change DumpIR method 2021-01-28 19:50:42 +08:00
luopengting 4683ecc06b support env_config_path to set the config file for DFX
1. add env_config_path for ms_context
2. add EnvConfigParser to parse config
3. add in CMakeLists
4. clang-format, cpplint
5. fix spelling in ms_context

Use EnvConfigParser in the constructed function of base_recorder,
when a recorder is initialized, its direcctory will be set as the
value from config file. In addition, recorder_manager will get the
'enable' from config file. If 'enable' is true and the path is not a
valid value, it will be the default value: './rdr'.
2021-01-28 15:10:23 +08:00
louei5 ac02213521 Adding RDR tool for locating bugs, e.g. run task error. 2021-01-25 10:46:47 +08:00