saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-05-29 23:06:41 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Improve cache hit rate in windows build (#15538)

Browse source 
### Description
1.  Update /Zi to /Z7 in abseil project while using cache
2.  Skip target_precompile_headers while using cache


### Motivation and Context
There're about 1/4 uncacheable calls in Windows GPU compilation with
cache.
```
Uncacheable calls:                   441 / 1641 (26.87%)
  Could not use precompiled header:  361 /  441 (81.86%)
  Preprocessing failed:                1 /  441 ( 0.23%)
  Unsupported compiler option:        79 /  441 (17.91%)
```

https://dev.azure.com/onnxruntime/onnxruntime/_build/results?buildId=961916&view=logs&j=5076e696-f193-5f12-2d8a-703dda41a79b&t=9b927034-e3ef-5e25-c6df-387bc37acd63&l=21

The root cause of `Unsupported compiler option` is that /Zi in Abseil
isn't updated to /Z7.
The root cause of `Could not use precompiled header` is the
`target_precompile_headers` creates cmake_pch.pch every time and it's
hash value is changed too.

### Result
It could reduce compilation time by another 20%. 
For example:
It took 16m43 in CUDA training compilation on Windows.
It takes 13m32 after the change.

https://dev.azure.com/onnxruntime/onnxruntime/_build/results?buildId=964002&view=logs&s=959c6b43-5937-53e5-5f36-e53cb0249117


### N.B.
In winml project, it's using own target_precompile**d**_header
https://github.com/microsoft/onnxruntime/blob/main/cmake/precompiled_header.cmake.
Just let it be.
This commit is contained in:
Yi Zhang 2023-04-19 00:31:35 +08:00 committed by GitHub
parent cf19c3697d
commit 698e9f71cd
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
3 changed files with 16 additions and 29 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

23
cmake/CMakeLists.txt
View file

@ -12,6 +12,11 @@ cmake_policy(SET CMP0117 NEW)
# Don't let cmake set a default value for CMAKE_CUDA_ARCHITECTURES
cmake_policy(SET CMP0104 OLD)
# Enable Hot Reload for MSVC compilers if supported.
if (POLICY CMP0141)
cmake_policy(SET CMP0141 NEW)
endif()
# Project
project(onnxruntime C CXX ASM)
@ -332,6 +337,11 @@ function(set_msvc_c_cpp_compiler_warning_level warning_level)
endif()
endfunction()
if(MSVC_Z7_OVERRIDE)
set(CMAKE_MSVC_DEBUG_INFORMATION_FORMAT "$<$<CONFIG:Debug,RelWithDebInfo>:Embedded>")
endif(MSVC_Z7_OVERRIDE)
# set default MSVC warning level to 3 for external dependencies
set_msvc_c_cpp_compiler_warning_level(3)
@ -722,19 +732,6 @@ if (onnxruntime_ENABLE_LAZY_TENSOR)
# cleaner.
endif()
if(MSVC)
# Replace /Zi and /ZI with /Z7
if(MSVC_Z7_OVERRIDE)
foreach(flag_var
CMAKE_C_FLAGS CMAKE_C_FLAGS_DEBUG CMAKE_C_FLAGS_RELWITHDEBINFO
CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_DEBUG CMAKE_CXX_FLAGS_RELWITHDEBINFO)
if(${flag_var} MATCHES "/Z[iI]")
string(REGEX REPLACE "/Z[iI]" "/Z7" ${flag_var} "${${flag_var}}")
endif(${flag_var} MATCHES "/Z[iI]")
endforeach(flag_var)
endif(MSVC_Z7_OVERRIDE)
endif()
function(onnxruntime_set_compile_flags target_name)
if (CPUINFO_SUPPORTED)
onnxruntime_add_include_to_target(${target_name} cpuinfo::cpuinfo)

12
cmake/external/helper_functions.cmake vendored
View file

@ -176,18 +176,6 @@ macro(onnxruntime_fetchcontent_makeavailable)
set_target_properties(${subdir_target} PROPERTIES COMPILE_WARNING_AS_ERROR OFF)
endif()
endforeach()
if(MSVC)
# Replace /Zi and /ZI with /Z7
if(MSVC_Z7_OVERRIDE)
foreach(flag_var
CMAKE_C_FLAGS CMAKE_C_FLAGS_DEBUG CMAKE_C_FLAGS_RELWITHDEBINFO
CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_DEBUG CMAKE_CXX_FLAGS_RELWITHDEBINFO)
if(${flag_var} MATCHES "/Z[iI]")
string(REGEX REPLACE "/Z[iI]" "/Z7" ${flag_var} "${${flag_var}}")
endif(${flag_var} MATCHES "/Z[iI]")
endforeach(flag_var)
endif(MSVC_Z7_OVERRIDE)
endif()
endif()
unset(__cmake_srcdir)

10
cmake/onnxruntime_providers.cmake
View file

@ -529,10 +529,12 @@ if (onnxruntime_USE_CUDA)
if (WIN32)
# *.cu cannot use PCH
target_precompile_headers(onnxruntime_providers_cuda PUBLIC
"${ONNXRUNTIME_ROOT}/core/providers/cuda/cuda_pch.h"
"${ONNXRUNTIME_ROOT}/core/providers/cuda/cuda_pch.cc"
)
if (NOT onnxruntime_BUILD_CACHE)
target_precompile_headers(onnxruntime_providers_cuda PUBLIC
"${ONNXRUNTIME_ROOT}/core/providers/cuda/cuda_pch.h"
"${ONNXRUNTIME_ROOT}/core/providers/cuda/cuda_pch.cc"
)
endif()
# minimize the Windows includes.
# this avoids an issue with CUDA 11.6 where 'small' is defined in the windows and cuda headers.