Move from hipcc to amdclang #1920
Conversation
| if os.name == "nt": | ||
| globalParameters["AssemblerPath"] = locateExe(globalParameters["ROCmBinPath"], "clang++.exe") |
There was a problem hiding this comment.
What about amdclang on windows?
| @@ -40,7 +40,7 @@ set(TENSILE_USE_OPENMP ON CACHE BOOL "Use OpenMP to improve performance.") | |||
| set(TENSILE_STATIC_ONLY ON CACHE BOOL "Disable exposing Tensile symbols in a shared library.") | |||
|
|
|||
| if(NOT DEFINED CXX_VERSION_STRING) | |||
| if(CMAKE_CXX_COMPILER MATCHES ".*/hipcc$") | |||
| if(CMAKE_CXX_COMPILER MATCHES ".*/hipcc$" OR CMAKE_CXX_COMPILER MATCHES ".*clang\\+\\+") | |||
There was a problem hiding this comment.
Consider using CMAKE_<LANG>_COMPILER_VERSION (new ticket):
https://cmake.org/cmake/help/v3.0/variable/CMAKE_LANG_COMPILER_VERSION.html
Tensile/Common.py
Outdated
| @@ -2230,8 +2231,8 @@ def capRow(caps, cap): | |||
| def which(p): | |||
| exes = [p+x for x in ['', '.exe', '.bat']] | |||
| system_path = os.environ['PATH'].split(os.pathsep) | |||
| if p == 'hipcc' and 'CMAKE_CXX_COMPILER' in os.environ and os.path.isfile(os.environ['CMAKE_CXX_COMPILER']): | |||
| return os.environ['CMAKE_CXX_COMPILER'] | |||
| if (p == 'hipcc' or p == 'amdclang++') and 'CXX' in os.environ and os.path.isfile(os.environ['CXX']): | |||
There was a problem hiding this comment.
This needs testing on windows
| @@ -466,7 +466,7 @@ def functionPrefix(self, kernel): | |||
| kStr += " } while (assumed != old);%s" % (self.endLine) | |||
| kStr += "}%s" % (self.endLine) | |||
| """ | |||
| if globalParameters["CxxCompiler"] == "hipcc": | |||
| if globalParameters["CxxCompiler"] == "hipcc" or globalParameters["CxxCompiler"] == "amdclang++": | |||
There was a problem hiding this comment.
Look into removing OCL ticket
| hipccver = globalParameters['HipClangVersion'].split(".") | ||
| hipccMaj = int(hipccver[0]) | ||
| hipccMin = int(hipccver[1]) | ||
|
|
There was a problem hiding this comment.
When using amdclang we will always pass the if test
There was a problem hiding this comment.
Maybe a ticket for dropping 5.2
| @@ -40,7 +40,7 @@ set(TENSILE_USE_OPENMP ON CACHE BOOL "Use OpenMP to improve performance.") | |||
| set(TENSILE_STATIC_ONLY ON CACHE BOOL "Disable exposing Tensile symbols in a shared library.") | |||
|
|
|||
| if(NOT DEFINED CXX_VERSION_STRING) | |||
| if(CMAKE_CXX_COMPILER MATCHES ".*/hipcc$") | |||
| if(CMAKE_CXX_COMPILER MATCHES ".*/hipcc$" OR CMAKE_CXX_COMPILER MATCHES ".*clang\\+\\+") | |||
There was a problem hiding this comment.
Does this check match version-suffixed compiler names? If it needs that .* at the start to match amdclang++, I think it might need a .* on the end to match clang++-17.
There was a problem hiding this comment.
I am exploring removing this logic altogether in favor of CMAKE_<LANG>_COMPILER_VERSION
ellosel
force-pushed
the
clang-migration
branch
3 times, most recently
from
fe3d527
to
a7ef0fc
Compare
The goal of this change set is to move the default compiler from hipcc to amdclang++. The most significant difference between hipcc and amdclang are the flags used when invoking the compiler wrapped by hipcc. We attempt to delineate those differences below by noting if a flag is defaulted (d), set manually (m) or unset (blank):
-O3O2by default-x hip-D__HIP_HCC_COMPAT_MODE__=1-mrelax-all-mframe-pointer=all-mframe-pointer=none-Wno-format-nonliteral-fallow-half-arguments-and-returns-mllvm -amdgpu-early-inline-all=true-mllvm -amdgpu-function-calls=false--genco--cuda-device-onlygencois passedTo reproduce one can do the following:
Configure the HostTestLibrary directory
Run the build
Remove DecisionTree test binaries, rebuild with verbosity and review the history for build command
Run the build command from above with
-vThe command above will output the detailed compilation commands and arguments used to determine differences in compilation with hipcc and amdclang++.