azalea/kotlin-fork
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Pass CPU information to codegenerator, to fix mingw_x86 opt. (#2870)

Browse Source
This commit is contained in:
Nikolay Igotti
2019-04-16 08:29:51 +03:00
committed by GitHub
parent bfd30c626d
commit fd4adaaeaf
1 changed files with 6 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
konan/konan.properties
+6 -2
View File
@@ -390,10 +390,14 @@ dependencies.macos_x64-mingw_x86 = \
quadruple.mingw_x86 = i686-w64-mingw32
targetSysRoot.mingw_x86 = msys2-mingw-w64-i686-gcc-7.4.0-clang-llvm-6.0.1
# For using with Universal Windows Platform (UWP) we need to use this slower option.
# For using with Universal Windows Platform (UWP) we need to use slower -emulated-tls option.
# See https://youtrack.jetbrains.com/issue/KT-27654.
# TODO: remove, once fixed in mingw, check with the bug testcase.
llvmLtoFlags.mingw_x86 = -emulated-tls
# -mcpu=sandybridge makes sure that cogenerator knows which CPU flavour to emit code for.
# Can be seen on optimized build and FP tests, where value is passed in coprocessor registers
# by Kotlin and taken from SSE registers by C code.
# TODO: once this information is available in function attributes, we can remove this flag.
llvmLtoFlags.mingw_x86 = -emulated-tls -mcpu=sandybridge
llvmLtoOptFlags.mingw_x86 = -O3 -function-sections
llvmLtoNooptFlags.mingw_x86 = -O1
linkerNoDebugFlags.mingw_x86 = -Wl,-S