[CMake] Support a simple case for bootstrap builds to generate PGO data

[beanz]

Summary

This patch adds support for the clang multi-stage bootstrapping to support PGO profdata generation, and can build a 2 or 3 stage compiler.

With this patch applied you can configure your build directory with the following invocation of CMake:

cmake -G <generator> -C <path_to_clang>/cmake/caches/PGO-stage1.cmake <source dir>

After configuration the following additional targets will be generated:

stage2:
Builds a stage1 x86 compiler, runtime, and required tools (llvm-config, llvm-profdata) then uses that compiler to build an instrumented stage2 compiler.

stage2-generate-profdata:
Depends on "stage2" and will use the stage2 compiler to generate profdata based on the training files in <clang>/utils/perf-training

stage3:
Depends on "stage2-generate-profdata" and will use the stage1 compiler with the stage2 profdata to build a PGO-optimized compiler.

stage3-check-llvm:
Depends on stage3 and runs check-llvm using the stage3 compiler.

stage3-check-clang:
Depends on stage3 and runs check-clang using the stage3 compiler.

stage3-check-all:
Depends on stage3 and runs check-all using the stage3 compiler.

stage3-test-suite:
Depends on stage3 and runs the test-suite using the stage3 compiler (requires in-tree test-suite).

[beanz]

Updates based on bogner's feedback.

• Target stages are now stage2-instrumented and stage2 instead of stage2 and stage3.
• Renamed PGO-stage1.cmake to PGO.cmake

[beanz]

Updating to fix a bug in my CMake regex handling that caused the targets to not be mapped up correctly.

[beanz]

One more dependency hookup fix, this one makes it so that stage2 doesn't depend on the stage2-instrumented compiler-rt, and avoids building it when you invoke the 'stage2' target.

[chandlerc]

Uh, did you intend to commit this prior to Justin or others giving an LGTM?

cfe/trunk/CMakeLists.txt

if (CLANG_ENABLE_BOOTSTRAP)
 
   if(NOT CLANG_STAGE)
     set(CLANG_STAGE stage1)
     message(STATUS "Setting current clang stage to: ${CLANG_STAGE}")
   endif()
 
   string(REGEX MATCH "stage([0-9]*)" MATCHED_STAGE "${CLANG_STAGE}")
   if(MATCHED_STAGE)
-    math(EXPR STAGE_NUM "${MATCHED_STAGE} + 1")
+    if(NOT LLVM_BUILD_INSTRUMENTED)
+      math(EXPR STAGE_NUM "${CMAKE_MATCH_1} + 1")
       set(NEXT_CLANG_STAGE stage${STAGE_NUM})
     else()
+      set(NEXT_CLANG_STAGE stage${CMAKE_MATCH_1})
+    endif()
+  else()
     set(NEXT_CLANG_STAGE bootstrap)
   endif()
+
+  if(BOOTSTRAP_LLVM_BUILD_INSTRUMENTED)
+    set(NEXT_CLANG_STAGE ${NEXT_CLANG_STAGE}-instrumented)
+  endif()
   message(STATUS "Setting next clang stage to: ${NEXT_CLANG_STAGE}")
   
   
   set(STAMP_DIR ${CMAKE_CURRENT_BINARY_DIR}/${NEXT_CLANG_STAGE}-stamps/)
   set(BINARY_DIR ${CMAKE_CURRENT_BINARY_DIR}/${NEXT_CLANG_STAGE}-bins/)
 
   # If on Darwin we need to make bootstrap depend on LTO and pass
   # DARWIN_LTO_LIBRARY so that -flto will work using the just-built compiler
set(BOOTSTRAP_DEFAULT_PASSTHROUGH
     LLVM_VERSION_SUFFIX
     CLANG_REPOSITORY_STRING
     CMAKE_MAKE_PROGRAM)
 
   if(TARGET compiler-rt)
     set(RUNTIME_DEP compiler-rt)
   endif()
 
+  set(COMPILER_OPTIONS
+    -DCMAKE_CXX_COMPILER=${LLVM_RUNTIME_OUTPUT_INTDIR}/clang++
+    -DCMAKE_C_COMPILER=${LLVM_RUNTIME_OUTPUT_INTDIR}/clang
+    -DCMAKE_ASM_COMPILER=${LLVM_RUNTIME_OUTPUT_INTDIR}/clang)
+
+  if(BOOTSTRAP_LLVM_BUILD_INSTRUMENTED)
+    set(PGO_DEP llvm-profdata)
+    set(PGO_OPT -DLLVM_PROFDATA=${LLVM_RUNTIME_OUTPUT_INTDIR}/llvm-profdata)
+  endif()
+
+  if(LLVM_BUILD_INSTRUMENTED)
+    set(PGO_DEP generate-profdata)
+    set(PGO_OPT -DLLVM_PROFDATA_FILE=${CMAKE_CURRENT_BINARY_DIR}/utils/perf-training/clang.profdata)
+    set(COMPILER_OPTIONS
+      -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}
+      -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}
+      -DCMAKE_ASM_COMPILER=${CMAKE_ASM_COMPILER})
+    set(RUNTIME_DEP) # Don't set runtime dependencies
+  endif()
+
   # Find all variables that start with BOOTSTRAP_ and populate a variable with
   # them.
   get_cmake_property(variableNames VARIABLES)
   foreach(variableName ${variableNames})
     if(variableName MATCHES "^BOOTSTRAP_")
       string(SUBSTRING ${variableName} 10 -1 varName)
       string(REPLACE ";" "\;" value "${${variableName}}")
       list(APPEND PASSTHROUGH_VARIABLES
         -D${varName}=${value})
     endif()
   endforeach()
 
   # Populate the passthrough variables
   foreach(variableName ${CLANG_BOOTSTRAP_PASSTHROUGH} ${BOOTSTRAP_DEFAULT_PASSTHROUGH})
     if(${variableName})
       string(REPLACE ";" "\;" value ${${variableName}})
       list(APPEND PASSTHROUGH_VARIABLES
         -D${variableName}=${value})
     endif()
   endforeach()
 
   ExternalProject_Add(${NEXT_CLANG_STAGE}
-    DEPENDS clang ${LTO_DEP} ${RUNTIME_DEP}
+    DEPENDS clang ${LTO_DEP} ${RUNTIME_DEP} ${PGO_DEP}
     PREFIX ${NEXT_CLANG_STAGE}
     SOURCE_DIR ${CMAKE_SOURCE_DIR}
     STAMP_DIR ${STAMP_DIR}
     BINARY_DIR ${BINARY_DIR}
     ${cmake_3_1_EXCLUDE_FROM_ALL}
     CMAKE_ARGS
                 # We shouldn't need to set this here, but INSTALL_DIR doesn't
                 # seem to work, so instead I'm passing this through
                 -DCMAKE_INSTALL_PREFIX=${CMAKE_INSTALL_PREFIX}
                 ${CLANG_BOOTSTRAP_CMAKE_ARGS}
                 ${PASSTHROUGH_VARIABLES}
-                -DCMAKE_CXX_COMPILER=${LLVM_RUNTIME_OUTPUT_INTDIR}/clang++
-                -DCMAKE_C_COMPILER=${LLVM_RUNTIME_OUTPUT_INTDIR}/clang
-                -DCMAKE_ASM_COMPILER=${LLVM_RUNTIME_OUTPUT_INTDIR}/clang
                  -DCLANG_STAGE=${NEXT_CLANG_STAGE}
-                ${LTO_LIBRARY} ${LTO_AR} ${LTO_RANLIB} ${verbose}
+                ${COMPILER_OPTIONS}
+                ${LTO_LIBRARY} ${LTO_AR} ${LTO_RANLIB} ${verbose} ${PGO_OPT}
     INSTALL_COMMAND ""
     STEP_TARGETS configure build
     ${cmake_3_4_USES_TERMINAL_OPTIONS}
     )
 
   # exclude really-install from main target
   set_target_properties(${NEXT_CLANG_STAGE} PROPERTIES _EP_really-install_EXCLUDE_FROM_MAIN On)
   ExternalProject_Add_Step(${NEXT_CLANG_STAGE} really-install

cfe/trunk/cmake/caches/PGO-stage2.cmake

+set(CMAKE_BUILD_TYPE RELEASE CACHE STRING "")
+set(CLANG_ENABLE_BOOTSTRAP ON CACHE BOOL "")
+set(LLVM_BUILD_EXTERNAL_COMPILER_RT ON CACHE BOOL "")
+
+set(CLANG_BOOTSTRAP_TARGETS check-all check-llvm check-clang test-suite CACHE STRING "")
+
+set(CLANG_BOOTSTRAP_CMAKE_ARGS
+  -C ${CMAKE_CURRENT_LIST_DIR}/PGO-stage3.cmake
+  CACHE STRING "")

cfe/trunk/cmake/caches/PGO-stage3.cmake

+set(CMAKE_BUILD_TYPE RELEASE CACHE STRING "")
+set(LLVM_BUILD_EXTERNAL_COMPILER_RT ON CACHE BOOL "")

cfe/trunk/cmake/caches/PGO.cmake

+set(CMAKE_BUILD_TYPE RELEASE CACHE STRING "")
+set(CLANG_ENABLE_BOOTSTRAP ON CACHE BOOL "")
+set(LLVM_BUILD_EXTERNAL_COMPILER_RT ON CACHE BOOL "")
+
+set(LLVM_TARGETS_TO_BUILD X86 CACHE STRING "")
+set(BOOTSTRAP_LLVM_BUILD_INSTRUMENTED ON CACHE BOOL "")
+set(CLANG_BOOTSTRAP_TARGETS
+  generate-profdata
+  stage2
+  stage2-check-all
+  stage2-check-llvm
+  stage2-check-clang
+  stage2-test-suite CACHE STRING "")
+
+set(CLANG_BOOTSTRAP_CMAKE_ARGS
+  -C ${CMAKE_CURRENT_LIST_DIR}/PGO-stage2.cmake
+  CACHE STRING "")