[perf-training] Make training data location configurable

[smeenai]

Summary

We may wish to keep the PGO training data outside the repository. Add a
CMake variable to allow referencing an external lit testsuite.

[smeenai]

I'm not sure where best to document this. We have https://llvm.org/docs/HowToBuildWithPGO.html, but that doesn't mention the build system's built-in PGO support as far as I can tell.

[phosek]

LGTM

[Amir]

Hi @smeenai,

Sorry for asking on this diff, but... Do you have any pointers for training sets?

Thanks!

clang/utils/perf-training/CMakeLists.txt

 
 # All test suites added here should be excuded from check-all
 set(EXCLUDE_FROM_ALL On)
 
 if (CMAKE_CFG_INTDIR STREQUAL ".")
   set(LLVM_BUILD_MODE ".")
 else ()
   set(LLVM_BUILD_MODE "%(build_mode)s")
 endif ()
 
 string(REPLACE ${CMAKE_CFG_INTDIR} ${LLVM_BUILD_MODE} CLANG_TOOLS_DIR ${LLVM_RUNTIME_OUTPUT_INTDIR})
 
+set(CLANG_PGO_TRAINING_DATA "${CMAKE_CURRENT_SOURCE_DIR}" CACHE PATH
+  "The path to a lit testsuite containing samples for PGO and order file generation"
+  )
+
 if(LLVM_BUILD_INSTRUMENTED)
   configure_lit_site_cfg(
     ${CMAKE_CURRENT_SOURCE_DIR}/lit.site.cfg.in
     ${CMAKE_CURRENT_BINARY_DIR}/pgo-data/lit.site.cfg
     )
 
   add_lit_testsuite(generate-profraw "Generating clang PGO data"
     ${CMAKE_CURRENT_BINARY_DIR}/pgo-data/

clang/utils/perf-training/lit.cfg

 
 sysroot_flags = getSysrootFlagsOnDarwin(config, lit_config)
 
 config.clang = lit.util.which('clang', config.clang_tools_dir).replace('\\', '/')
 
 config.name = 'Clang Perf Training'
 config.suffixes = ['.c', '.cpp', '.m', '.mm', '.cu', '.ll', '.cl', '.s', '.S', '.modulemap']
 
-cc1_wrapper = '%s %s/perf-helper.py cc1' % (config.python_exe, config.test_source_root)
+cc1_wrapper = '%s %s/perf-helper.py cc1' % (config.python_exe, config.perf_helper_dir)
 
 use_lit_shell = os.environ.get("LIT_USE_INTERNAL_SHELL")
 config.test_format = lit.formats.ShTest(use_lit_shell == "0")
 config.substitutions.append( ('%clang_cpp_skip_driver', ' %s %s %s ' % (cc1_wrapper, config.clang, sysroot_flags)))
 config.substitutions.append( ('%clang_cpp', ' %s --driver-mode=g++ %s ' % (config.clang, sysroot_flags)))
 config.substitutions.append( ('%clang_skip_driver', ' %s %s %s ' % (cc1_wrapper, config.clang, sysroot_flags)))
 config.substitutions.append( ('%clang', ' %s %s ' % (config.clang, sysroot_flags) ) )
 config.substitutions.append( ('%test_root', config.test_exec_root ) )
 
 config.environment['LLVM_PROFILE_FILE'] = 'perf-training-%p.profraw'
 

clang/utils/perf-training/lit.site.cfg.in

 @LIT_SITE_CFG_IN_HEADER@
 
 import sys
 
 config.clang_tools_dir = "@CLANG_TOOLS_DIR@"
+config.perf_helper_dir = "@CMAKE_CURRENT_SOURCE_DIR@"
 config.test_exec_root = "@CMAKE_CURRENT_BINARY_DIR@"
-config.test_source_root = "@CMAKE_CURRENT_SOURCE_DIR@"
+config.test_source_root = "@CLANG_PGO_TRAINING_DATA@"
 config.target_triple = "@TARGET_TRIPLE@"
 config.python_exe = "@PYTHON_EXECUTABLE@"
 
 # Support substitution of the tools and libs dirs with user parameters. This is
 # used when we can't determine the tool dir at configuration time.
 try:
     config.clang_tools_dir = config.clang_tools_dir % lit_config.params
 except KeyError:
     e = sys.exc_info()[1]
     key, = e.args
     lit_config.fatal("unable to find %r parameter, use '--param=%s=VALUE'" % (key,key))
 
 # Let the main config do the real work.
 lit_config.load_config(config, "@CLANG_SOURCE_DIR@/utils/perf-training/lit.cfg")

clang/utils/perf-training/order-files.lit.cfg

 
 sysroot_flags = getSysrootFlagsOnDarwin(config, lit_config)
 
 config.clang = os.path.realpath(lit.util.which('clang', config.clang_tools_dir)).replace('\\', '/')
 
 config.name = 'Clang Perf Training'
 config.suffixes = ['.c', '.cpp', '.m', '.mm', '.cu', '.ll', '.cl', '.s', '.S', '.modulemap']
 
-dtrace_wrapper = '%s %s/perf-helper.py dtrace' % (config.python_exe, config.test_source_root)
-dtrace_wrapper_cc1 = '%s %s/perf-helper.py dtrace --cc1' % (config.python_exe, config.test_source_root)
+dtrace_wrapper = '%s %s/perf-helper.py dtrace' % (config.python_exe, config.perf_helper_dir)
+dtrace_wrapper_cc1 = '%s %s/perf-helper.py dtrace --cc1' % (config.python_exe, config.perf_helper_dir)
 
 if 'darwin' in config.target_triple:
     lit_config.note('using DTrace oneshot probe')
     dtrace_wrapper = '%s --use-oneshot' % dtrace_wrapper
     dtrace_wrapper_cc1 = '%s --use-oneshot' % dtrace_wrapper_cc1
 
 use_lit_shell = os.environ.get("LIT_USE_INTERNAL_SHELL")
 config.test_format = lit.formats.ShTest(use_lit_shell == "0")
 config.substitutions.append( ('%clang_cpp_skip_driver', ' %s %s --driver-mode=g++ %s ' % (dtrace_wrapper_cc1, config.clang, sysroot_flags)))
 config.substitutions.append( ('%clang_cpp', ' %s %s --driver-mode=g++ %s ' % (dtrace_wrapper, config.clang, sysroot_flags)))
 config.substitutions.append( ('%clang_skip_driver', ' %s %s %s ' % (dtrace_wrapper_cc1, config.clang, sysroot_flags)))
 config.substitutions.append( ('%clang', ' %s %s %s ' % (dtrace_wrapper, config.clang, sysroot_flags) ) )
 config.substitutions.append( ('%test_root', config.test_exec_root ) )
 

clang/utils/perf-training/order-files.lit.site.cfg.in

 @LIT_SITE_CFG_IN_HEADER@
 
 import sys
 
 config.clang_tools_dir = "@CLANG_TOOLS_DIR@"
+config.perf_helper_dir = "@CMAKE_CURRENT_SOURCE_DIR@"
 config.test_exec_root = "@CMAKE_CURRENT_BINARY_DIR@"
-config.test_source_root = "@CMAKE_CURRENT_SOURCE_DIR@"
+config.test_source_root = "@CLANG_PGO_TRAINING_DATA@"
 config.target_triple = "@TARGET_TRIPLE@"
 config.python_exe = "@PYTHON_EXECUTABLE@"
 
 # Support substitution of the tools and libs dirs with user parameters. This is
 # used when we can't determine the tool dir at configuration time.
 try:
     config.clang_tools_dir = config.clang_tools_dir % lit_config.params
 except KeyError:
     e = sys.exc_info()[1]
     key, = e.args
     lit_config.fatal("unable to find %r parameter, use '--param=%s=VALUE'" % (key,key))
 
 # Let the main config do the real work.
 lit_config.load_config(config, "@CLANG_SOURCE_DIR@/utils/perf-training/order-files.lit.cfg")