18 Commits (8c19d7aa2f89a38b3a68e53c73d88af16a3de8ce)

Author	SHA1	Message	Date
Zhang Ting	f0ee159280	enable exhaustive_search for forward and backward algos when dtype is float16 (#30959) ... * enable exhaustive_search for input_grad when dtype is float16 * enable exhaustive_search for forward algos	4 years ago
AshburnLee	924aac2216	Add tf32 switch for cuDNN (#29192)	4 years ago
wangchaochaohu	8b853b3030	fix the number of perf algo for conv cudnn in exhaustive mode (#28694)	4 years ago
Leo Chen	f962bd3432	Fix cudnn workspace limit in cudnn-8 (#28611)	4 years ago
Zhang Ting	d5cc144c60	tune backward filter algorithm for float16 (#27529) ... * use exhaustive_search for float16 * tune algo only when dtype is float16	4 years ago
Shang Zhizhou	c17f9cf25f	[bug fix]:Memory increases after adapting the cudnn version to cudnn8 (#27436) ... * [bug fix]:Memory increases after adapting the cudnn version to 8 * [bug fix]cudnnGetConvolutionForwardAlgorithm not defined	4 years ago
Zhaolong Xing	358bc06c72	[CUDNN8 support] : support CUDNN8 (#25664) ... * cunn8 support test=develop * fix ci error test=develop	5 years ago
wangchaochaohu	355caee18b	fix conv_transpose Op fp16 error test=develop (#24695)	5 years ago
zhongpu	b4b6763ab2	fix bug for exhaustive_search in conv_fusion_op, test=develop (#23727)	5 years ago
zhongpu	dbfbd7eac4	support Exhaustive search in dygraph (#23415) ... * use global conv cache; test=develop * use singleton cache; test=develop * fix format error; test=develop * add cudnn helper header; test=develop * fix header error; test=develop * fix mac unitest; test=develop * fix mac unitest; test=develop * fix file format; test=develop * fix include file error, test=develop * remove kernel_configs_ in class ExecutionContext and kernel_configs_map_ in class OperatorWithKernel, test=develop * fix test_elementwise_mul_op_dim, test=develop * fix compile error, test=develop Co-authored-by: phlrain <phliuhongyu@126.com>	5 years ago
zhongpu	bfb07aafe8	Revert "Exhaustive search (#22821)", test=develop (#23401) ... This reverts commit 48144e4099.	5 years ago
zhongpu	48144e4099	Exhaustive search (#22821) ... * use global conv cache; test=develop * use singleton cache; test=develop * fix format error; test=develop * add cudnn helper header; test=develop * fix header error; test=develop * fix mac unitest; test=develop * fix mac unitest; test=develop * fix file format; test=develop * fix include file error, test=develop * remove kernel_configs_ in class ExecutionContext and kernel_configs_map_ in class OperatorWithKernel, test=develop * fix test_elementwise_mul_op_dim, test=develop Co-authored-by: phlrain <phliuhongyu@126.com>	5 years ago
Chen Weihang	ba8414d3a5	replace CUDNN_ENFORCE with PADDLE_ENFORCE_CUDA_SUCCESS, test=develop (#22109)	5 years ago
wangchaochaohu	bf379fef96	refine code for code reuse test=develop (#20988)	5 years ago
liuwei1031	e03c1d8a9e	fix conv_op compilation issue on windows (#20230)	5 years ago
gongweibao	abaf87be2b	Change backward_guard to optimize_guard to maximize the allreduce overlap. (#19506) ... Change backward_guard to optimize_guard to maximize the allreduce overlap	6 years ago
wangchaochaohu	6b78e00da4	Cudnn convolution reconstruction (#18284) ... * rewrite the conv_op using cudnn_conv_helper * add workspace limit for v7 test=develop * fix test=develop * add half float test=develop * fix test=develop * fix test=develop * revise code style test=develop * fix test=develop	6 years ago
qingqing01	e32c9888f5	Double backward of conv2d. (#17211) ... * Add conv2d_grad_grad_op * Extracte the cuDNN conv algo searching code in conv_cudnn_helper.h. - Now use it in conv2d_grad_grad. - Will simply the searching code in conv2d and conv2d_grad in next PR. * Enhance and fix bug in unit testing of gradient_checker. * Support to fetch empty variables，return None in Python.	6 years ago