tensorflow/third_party/cudnn_frontend.BUILD

25 lines
492 B
Plaintext
Raw Permalink Normal View History

PR #46965: Integrate CUDNN v8 frontend API for convolution Imported from GitHub PR https://github.com/tensorflow/tensorflow/pull/46965 This PR integrates the new cuDNN Frontend API (https://github.com/NVIDIA/cudnn-frontend) for the convolution backends. FYI. @nluehr Copybara import of the project: -- 61ec89a3e86b6001b2b7ea4f5317df19aa79b04c by Kaixi Hou <kaixih@nvidia.com>: CUDNNv8: Add configs for headers -- d3c34717fede54f2e77fc43a3b42a08c1bda483b by Kaixi Hou <kaixih@nvidia.com>: CUDNNv8: Add support in stream executor -- 5d54a49ecc844c70dfdd50c403b561ee6352f96b by Kaixi Hou <kaixih@nvidia.com>: CUDNNv8: Add support in autotuner -- 1ee715d6ee2880abc56936d659663b5060ff4582 by Kaixi Hou <kaixih@nvidia.com>: CUDNNv8: Add support in Conv2D -- 2badb9e39e1101dcdf1287ceb1df0b4ac376dfe5 by Kaixi Hou <kaixih@nvidia.com>: CUDNNv8: Add support in Conv3D -- 76b4645547e32902fc55d5fddae836a1aa1349e7 by Kaixi Hou <kaixih@nvidia.com>: CUDNNv8: Fix a failed test -- 322e858bd8e0a6bc2f39aed8390fea8ff94fcb9e by Kaixi Hou <kaixih@nvidia.com>: Cleanup codes -- 075baf8fd6165b5f2d25b7384eb6e15853bf3bfa by Kaixi Hou <kaixih@nvidia.com>: Merge AutotuneExecutionPlanResult into AutotuneResult -- fd2335ed63e620a34b27133771e7dbc5a712715c by Kaixi Hou <kaixih@nvidia.com>: Use AlgorithmXXX instead of ExecutionPlanXXX -- aa6f984dc73a025d1bc84df96b3ef26e66ca741a by Kaixi Hou <kaixih@nvidia.com>: Remove ProfileExecutionPlanResults -- 994af73b612bf8dbf865e163a4d2713a3a2f6721 by Kaixi Hou <kaixih@nvidia.com>: Separate the cudnn-specific DoConvolve -- ed22ef7896f25c34ba3a0018c6177be5f9330dd2 by Kaixi Hou <kaixih@nvidia.com>: Cleanup code -- a047609b552b7c950a0f590f6aeedf09db503bc6 by Kaixi Hou <kaixih@nvidia.com>: Add env var to switch on cudnn frontend -- d5470a401824b292d96076255def2af06c309236 by Kaixi Hou <kaixih@nvidia.com>: Cleanup code -- 558472bcda4d59e803394a5e863b6310a4e5444b by Kaixi Hou <kaixih@nvidia.com>: Unify autotuning -- 4e780fed915607eecffb097e3790c443630a55d3 by Kaixi Hou <kaixih@nvidia.com>: Remove some macros -- c865c2d8822ecc663f95a01e08b2f465d478f4e9 by Kaixi Hou <kaixih@nvidia.com>: Combine BestCudnnConvAlgorithm -- a51a1ea7f13d8310ae9aa3c9e5339fb8bcb557a7 by Kaixi Hou <kaixih@nvidia.com>: Use original autotuning map -- a114edefdbb65fed5f9f9e4bbb30d9f8c2f8e226 by Kaixi Hou <kaixih@nvidia.com>: Use shared ptr -- 7e8898c42fa9e02b3bc921b0d26619e87c284512 by Kaixi Hou <kaixih@nvidia.com>: Remove some macros -- 5ffdb1f896c40dd5273a47aefdb1e3b714f7f06f by Kaixi Hou <kaixih@nvidia.com>: Use if macros -- 6838420da8ce4672d23d497c28932528d022fe2a by Kaixi Hou <kaixih@nvidia.com>: Remove if macros PiperOrigin-RevId: 364409851 Change-Id: I829affdf6a2fb344a16e82ab201ff76b7421eef1
2021-03-22 21:32:07 +00:00
# Description:
# The cuDNN Frontend API is a C++ header-only library that demonstrates how
# to use the cuDNN C backend API.
package(
default_visibility = ["//visibility:public"],
)
licenses(["notice"]) # MIT
exports_files(["LICENSE.txt"])
filegroup(
name = "cudnn_frontend_header_files",
srcs = glob([
"include/**",
]),
)
cc_library(
name = "cudnn_frontend",
hdrs = [":cudnn_frontend_header_files"],
include_prefix = "third_party/cudnn_frontend",
)