未验证 提交 60e3e350 编写于 作者: Y Yichen Zhang 提交者: GitHub

fix the timeout bug of some communication api on A100 (#54513)

上级 9857bfc7
...@@ -79,8 +79,13 @@ if((WITH_GPU OR WITH_ROCM) AND (LINUX)) ...@@ -79,8 +79,13 @@ if((WITH_GPU OR WITH_ROCM) AND (LINUX))
py_test_modules( py_test_modules(
test_collective_alltoall_api MODULES test_collective_alltoall_api ENVS test_collective_alltoall_api MODULES test_collective_alltoall_api ENVS
"http_proxy=;https_proxy=;PYTHONPATH=..:${PADDLE_BINARY_DIR}/python") "http_proxy=;https_proxy=;PYTHONPATH=..:${PADDLE_BINARY_DIR}/python")
set_tests_properties(test_collective_alltoall_api if(${CUDA_ARCH_NAME} STREQUAL "Ampere")
PROPERTIES TIMEOUT "120" LABELS "RUN_TYPE=DIST") set_tests_properties(test_collective_alltoall_api
PROPERTIES TIMEOUT "160" LABELS "RUN_TYPE=DIST")
else()
set_tests_properties(test_collective_alltoall_api
PROPERTIES TIMEOUT "120" LABELS "RUN_TYPE=DIST")
endif()
endif() endif()
if((WITH_GPU OR WITH_ROCM) AND (LINUX)) if((WITH_GPU OR WITH_ROCM) AND (LINUX))
bash_test_modules( bash_test_modules(
...@@ -130,8 +135,13 @@ if((WITH_GPU OR WITH_ROCM) AND (LINUX)) ...@@ -130,8 +135,13 @@ if((WITH_GPU OR WITH_ROCM) AND (LINUX))
py_test_modules( py_test_modules(
test_collective_broadcast_api MODULES test_collective_broadcast_api ENVS test_collective_broadcast_api MODULES test_collective_broadcast_api ENVS
"http_proxy=;https_proxy=;PYTHONPATH=..:${PADDLE_BINARY_DIR}/python") "http_proxy=;https_proxy=;PYTHONPATH=..:${PADDLE_BINARY_DIR}/python")
set_tests_properties(test_collective_broadcast_api if(${CUDA_ARCH_NAME} STREQUAL "Ampere")
PROPERTIES TIMEOUT "300" LABELS "RUN_TYPE=DIST") set_tests_properties(test_collective_broadcast_api
PROPERTIES TIMEOUT "360" LABELS "RUN_TYPE=DIST")
else()
set_tests_properties(test_collective_broadcast_api
PROPERTIES TIMEOUT "300" LABELS "RUN_TYPE=DIST")
endif()
endif() endif()
if((WITH_GPU OR WITH_ROCM) AND (LINUX)) if((WITH_GPU OR WITH_ROCM) AND (LINUX))
py_test_modules( py_test_modules(
...@@ -228,8 +238,13 @@ if((WITH_GPU OR WITH_ROCM) AND (LINUX)) ...@@ -228,8 +238,13 @@ if((WITH_GPU OR WITH_ROCM) AND (LINUX))
test_collective_reduce_scatter_api MODULES test_collective_reduce_scatter_api MODULES
test_collective_reduce_scatter_api ENVS test_collective_reduce_scatter_api ENVS
"http_proxy=;https_proxy=;PYTHONPATH=..:${PADDLE_BINARY_DIR}/python") "http_proxy=;https_proxy=;PYTHONPATH=..:${PADDLE_BINARY_DIR}/python")
set_tests_properties(test_collective_reduce_scatter_api if(${CUDA_ARCH_NAME} STREQUAL "Ampere")
PROPERTIES TIMEOUT "150" LABELS "RUN_TYPE=DIST") set_tests_properties(test_collective_reduce_scatter_api
PROPERTIES TIMEOUT "210" LABELS "RUN_TYPE=DIST")
else()
set_tests_properties(test_collective_reduce_scatter_api
PROPERTIES TIMEOUT "150" LABELS "RUN_TYPE=DIST")
endif()
endif() endif()
if((WITH_GPU OR WITH_ROCM) AND (LINUX)) if((WITH_GPU OR WITH_ROCM) AND (LINUX))
py_test_modules( py_test_modules(
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册