MindSpore-AKG

7.6 MB

Author	SHA1	Message	Date
hanhuifeng2020	0fdfe435f0	For the transdata, auto_inline is not enabled by default	4 years ago
r1chardf1d0	f049354d88	stitch fusion code refactor, fix some bug	4 years ago
mindspore-ci-bot	611e8c34e2	!72 unify the calling of some global configs From: @ckey_dou Reviewed-by: @anyrenwei,@dylangeng Signed-off-by: @dylangeng	5 years ago
mindspore-ci-bot	b1f12ec2b2	!53 enable online tuning for composite ops on ascend From: @dabaiji Reviewed-by: @gaoxiong1,@anyrenwei Signed-off-by: @anyrenwei	5 years ago
dabaiji	4934596bad	add online tuner api	5 years ago
ckey_Dou	05c7f60df6	1. append pid after cuda_meta 2. replace XXX_META_PATH by get_XXX_meta_path 3. replace the calling of string 'MS_AKG_DUMP_xxx' by get_dump_xx_flag	5 years ago
lingyunli63	5f5125d442	enable autoinline for matmul	5 years ago
r1chardf1d0	6a01831f58	optimize stitch fusion	5 years ago
hanhuifeng2020	b061664070	refactor the code related to build of composite	5 years ago
hanhuifeng2020	1021d3cdb7	Some modifications about composite: 1. Support the composite of TransData operator 2. Add the enable_atomic_add attribute to Ascend backend 3. Adjust some codes related to the build of compoiste to make the build call and the compilewithjson call as consistent as possible	5 years ago
lingyunli63	fdb48f9308	set bypass attr from tuned repo for matmul	5 years ago
xsmq	46f4c28fcf	init	5 years ago