52 Commits (32211fe9c4c22168dfb73f19763b17ac9191341a)

Author	SHA1	Message	Date
Qi Li	34f1628ce8	[ROCM] update fluid platform for rocm39 (part2), test=develop (#30774)	4 years ago
WangXi	572c466d19	[Prepare for MultiProcess xpu] unified gen nccl id, refine imperative reducer (#30455)	4 years ago
Leo Chen	81217a94d8	unify calling cudaSetDevice (#30470) ... * unify calling cudaSetDevice * fix compile	4 years ago
Huihuang Zheng	acc11c2a62	Retry CUDA Initialization to Fix Random Failure, test=develop (#28323) ... This PR is follow up of #28213. On that PR we tried to decrease GPU usage, however the CI still randomly failed. So I added retry logic for the initialization of nccl and cusolver. If the initialization failed, we can retry to avoid the random failure.	4 years ago
GaoWei8	c10dcff12d	refine PADDLE_ENFORCE (#25456) ... * Refine PADDLE_ENFORCE in paddle/fluid/platform test=develop	5 years ago
GaoWei8	ea7e532598	Refine PADDLE_ENFORCE (#25369) ... * refine PADDLE_ENFORCE test=develop	5 years ago
Chen Weihang	aa0f254fbe	Add macro BOOST_GET to enrich the error information of boost :: get (#24175) ... * add new macro BOOST_GET_SAFELY & unittests, test=develop * add different macro type, test=develop * fix get macro type in executor, test=develop * four macro part change backup * using one macro for all case, test=develop * revert attribute change, test=develop * change to three func to solve gcc4.8 bug, test=develop * polish some details, test=develop	5 years ago
Yi Liu	2169e6fb58	Initialize global nccl_comm in PE (#23275)	5 years ago
Wilber	de009152a7	Compile without nccl deps. [2/2] (#22484) ... Compile without nccl deps. [1/2] Co-authored-by: 石晓伟 <39303645+Shixiaowei02@users.noreply.github.com>	5 years ago
Zeng Jinle	cdb3d27985	Fix warn of gcc8 (#21205) ... * fix warnings oof gcc 8 compilation, test=develop * fix boost::bad_get, test=develop * refine PADDLE_ENFORCE, test=develop	5 years ago
Tao Luo	75d1571995	refine PADDLE_ENFORCE codes for unify PADDLE_ASSERT_MSG (#19603) ... test=develop	6 years ago
gongweibao	c0a82748cf	Polish backwards optimizer dependency codes and use more default values. (#18255)	6 years ago
gongweibao	f5caf3443c	Fix reinitialized ncclid error! (#18025)	6 years ago
gongweibao	0d561ef442	fix 2dconn test=develop (#17681)	6 years ago
gongweibao	65bbf950ee	Add multi-ncclcomm and 2D ncclallreduce support. (#17263)	6 years ago
Wu Yi	6382b62f6b	Collective ops (#15572) ... * wip allreduce in op * wip * wip * wip * wip adding test * wip for conflict with mp mode * fix tests test=develop * fix cpu build test=develop * fix travis clang format test=develop * fix cpu build test=develop * update api.spec test=develop * delete comment test=develop * fix cpplint test=develop * fix test=develop * follow comment test=develop * add file test=develop * fix build test=develop * update test=develop * to be compatible with sync_bn, and fix mp mode in develop test=develop	6 years ago
qingqing01	8ad672a287	Support sync batch norm. (#16121) ... * Support Sync Batch Norm. * Note, do not enable it in one device. Usage: build_strategy = fluid.BuildStrategy() build_strategy.sync_batch_norm = True binary = fluid.compiler.CompiledProgram(tp).with_data_parallel( loss_name=loss_mean.name, build_strategy=build_strategy)	6 years ago
sneaxiy	ba4f43fd62	fix compile error in distributed mode ... test=develop	6 years ago
Yancey1989	86bb583881	Merge branch 'develop' of github.com:PaddlePaddle/Paddle into parallel_graph_mode	6 years ago
Yancey1989	41a64f6a2a	Merge branch 'develop' of github.com:PaddlePaddle/Paddle into parallel_graph_mode	6 years ago
Wu Yi	856f0da0fe	Fp16 training (#14992) ... * wip * wip * wip * wip for test * add fp16 tests test=develop * fix cpu build test=develop * fix test=develop * fix py3 tests test=develop * fix lr_scheduler dtype test=develop * fix test=dvelop * test fix ci compile test=develop * fix build and merge test=develop * fallback momentumop change to general test=develop * make fp16 lr schedule simple test=develop * fix ut test=develop * fix tests test=develop * remove fp16 learning rate cast test=develop	6 years ago
typhoonzero	da87f7a698	Revert "[Feature] Fp16 training for resnet50 (#14850)" ... This reverts commit 3d750f9c5a.	6 years ago
Wu Yi	3d750f9c5a	[Feature] Fp16 training for resnet50 (#14850) ... * wip * wip * wip * wip for test * add fp16 tests test=develop * fix cpu build test=develop * fix test=develop * fix py3 tests test=develop * fix lr_scheduler dtype test=develop * fix test=dvelop * test fix ci compile test=develop * fix build and merge test=develop * fallback momentumop change to general test=develop	6 years ago
Yancey1989	4a4ccac1d0	update by comment test=develop	6 years ago
Yu Yang	9bd70a1e04	Change tensor uses proto::VarType::type ... test=develop	6 years ago
Yancey1989	cb8a24be14	clean code	6 years ago
Yancey1989	c9de6f1b05	init parallel graph mode	6 years ago
Wu Yi	29d9fb53fc	[Feature] multi process multi gpu dist training, boost v100 performance by 20% (#14661) ... * wip multi process multi gpu dist training * workable for p2p * update test=develop * change back env name test=develop * fix alloc init * fix cpu build test=devlop * fix mac tests test=develop * refine code * refine test=develop	6 years ago
minqiyang	53433d7f2e	Revert the changes of VLOG ... test=develop	6 years ago
peizhilin	7840d181c9	fix style issue	6 years ago
peizhilin	ca60e1d34d	Merge remote-tracking branch 'upstream/develop' into windows/build	6 years ago
minqiyang	0c3227a523	Change the origin VLOG level to 10 times ... Fix code to support cpplint syntax check test=develop	6 years ago
peizhilin	9d67c1fb69	cpu build support	6 years ago
Wu Yi	f90c7865f0	Benchmark tool for imgnet (#12305) ... * support test using executor without reader * run imgnet * update fluid benchmark * wip * update * update all models * support pyreader * update * clean up * make profile batches contollable * update API.spec * update scripts * clean dockerfile * update * clean comments * add scope argument docstring * use num_trainers to determine nccl init comms	7 years ago
Qiyang Min	046bb5c8cb	Fix NCCLBcast hang up bug in Parallel Executor (#11377) ... * 1. Create buddy allocator in each places before NcclBcast the variables 2. Check the memory usage of ALL gpus rather than the first one * 1. Make NCCLGroupGuard guards only the ncclBcast part, which avoid ncclGroupEnd blocking the exception throwing 2. NOTE the usage of NCCLGroupGuard * Remove the memory usage check of gpus * Fix code style	7 years ago
gongweibao	4fb7cc7f5e	Move sync_mode device ctx from grpc server (#10881)	7 years ago
yuyang18	7c777dd549	Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into feature/exec_strategy	7 years ago
yuyang18	08295f9877	Add build strategy	7 years ago
typhoonzero	7b0c0273f4	update by comments	7 years ago
typhoonzero	f5840d8925	follow comments	7 years ago
typhoonzero	17009d0627	workable version	7 years ago
typhoonzero	3667578ec2	testing	7 years ago
typhoonzero	d9320dcd94	complete code	7 years ago
Yu Yang	093d227a77	Use mutex to stablize ncclCtxMap	7 years ago
Yu Yang	c64190ecbb	Polish NCCLHelper	7 years ago
qingqing01	129859e732	Support data type int64 in NCCL. (#9818)	7 years ago
Yu Yang	7dcb217e31	Refine allreduce op	7 years ago
Yu Yang	c0c2e15920	NCCL AllReduce	7 years ago
Yu Yang	fe7ed285d1	Extract NCCLCtxMap	7 years ago
Yu Yang	6ebc6bf533	ReorganizeCode	7 years ago