[cmake] Add LLVM_USE_GDB_INDEX option for gold/lld linkers

[ddcc]

Summary

Add option for -Wl,--gdb-index and update documentation

[MaskRay]

This seems fine. I have -DCMAKE_EXE_LINKER_FLAGS=-Wl,--gdb-index -DCMAKE_SHARED_LINKER_FLAGS=-Wl,--gdb-index -DLLVM_USE_SPLIT_DWARF=On in my long cmake command line. I am just wondering whether this new option will be discoverable.

Some people don't use -DLLVM_USE_LINKER=lld, but use -DLLVM_ENABLE_LLD=On.

[ddcc]

In D77154#1953423, @MaskRay wrote:

Some people don't use -DLLVM_USE_LINKER=lld, but use -DLLVM_ENABLE_LLD=On.

It's a bit confusing, but earlier in HandleLLVMOptions.cmake, LLVM_ENABLE_LLD is normalized into LLVM_USE_LINKER. There's also an internal variable LINKER_IS_LLD_LINK that is set by either.

[hliao]

The same patch posted previously, https://reviews.llvm.org/D58628

[hliao]

In D77154#1953423, @MaskRay wrote:

This seems fine. I have -DCMAKE_EXE_LINKER_FLAGS=-Wl,--gdb-index -DCMAKE_SHARED_LINKER_FLAGS=-Wl,--gdb-index -DLLVM_USE_SPLIT_DWARF=On in my long cmake command line. I am just wondering whether this new option will be discoverable.

Some people don't use -DLLVM_USE_LINKER=lld, but use -DLLVM_ENABLE_LLD=On.

As I mentioned in https://reviews.llvm.org/D58628, if your system's default linker doesn't recognize --gdb-index, adding option into CMAKE_EXE_LINKER_FLAGS fails the make detection stage, where the default linker is used.

[ddcc]

In D77154#1953470, @hliao wrote:

The same patch posted previously, https://reviews.llvm.org/D58628

Oops, I saw your patch to LLVM_USE_SPLIT_DWARF that landed, but missed that one.

In D77154#1953480, @hliao wrote:

As I mentioned in https://reviews.llvm.org/D58628, if your system's default linker doesn't recognize --gdb-index, adding option into CMAKE_EXE_LINKER_FLAGS fails the make detection stage, where the default linker is used.

It seems like -fuse-ld=gold/lld would be the simplest way forward? Do you plan to revise your patch to address that, lld, and @smeenai comment about -ggnu-pubnames? My only other feedback would be to update the docs, since even LLVM_USE_SPLIT_DWARF isn't currently listed, and that incorporating objcopy would make things more complicated.

[dblaikie]

In D77154#1953480, @hliao wrote:

In D77154#1953423, @MaskRay wrote:

This seems fine. I have -DCMAKE_EXE_LINKER_FLAGS=-Wl,--gdb-index -DCMAKE_SHARED_LINKER_FLAGS=-Wl,--gdb-index -DLLVM_USE_SPLIT_DWARF=On in my long cmake command line. I am just wondering whether this new option will be discoverable.

Some people don't use -DLLVM_USE_LINKER=lld, but use -DLLVM_ENABLE_LLD=On.

As I mentioned in https://reviews.llvm.org/D58628, if your system's default linker doesn't recognize --gdb-index, adding option into CMAKE_EXE_LINKER_FLAGS fails the make detection stage, where the default linker is used.

Perhaps that issue should be fixed (so that LLVM_ENABLE_LLD/LLVM_USE_LINKER kicks in early enough (same as putting -fuse-ld in CMAKE_EXE_LINKER_FLAGS) - seems a bit problematic to have different compilation modes at different points along that (but I guess make those LLD/linker enabling features are designed to use make detection to silently not do the thing you asked for if it's unsupported)

To my mind this all seems a bit complicated - and I wonder what the point of all these cmake supported features is compared to users specifying the flags in the flag variables directly?

llvm/CMakeLists.txt

set(LLVM_USE_SANITIZER "" CACHE STRING
   "Define the sanitizer used to build binaries and tests.")
 option(LLVM_OPTIMIZE_SANITIZED_BUILDS "Pass -O1 on debug sanitizer builds" ON)
 set(LLVM_LIB_FUZZING_ENGINE "" CACHE PATH
   "Path to fuzzing library for linking with fuzz targets")
 
 option(LLVM_USE_SPLIT_DWARF
   "Use -gsplit-dwarf when compiling llvm." OFF)
 
+option(LLVM_USE_GDB_INDEX
+  "Use -Wl,--gdb-index when compiling llvm." OFF)
+
 # Define an option controlling whether we should build for 32-bit on 64-bit
 # platforms, where supported.
 if( CMAKE_SIZEOF_VOID_P EQUAL 8 AND NOT (WIN32 OR ${CMAKE_SYSTEM_NAME} MATCHES "AIX"))
   # TODO: support other platforms and toolchains.
   option(LLVM_BUILD_32_BITS "Build 32 bits executables and libraries." OFF)
 endif()
 
 # Define the default arguments to use with 'lit', and an option for the user to

llvm/cmake/modules/HandleLLVMOptions.cmake

 # Turn on -gsplit-dwarf if requested in debug builds.
 if (LLVM_USE_SPLIT_DWARF AND
     ((uppercase_CMAKE_BUILD_TYPE STREQUAL "DEBUG") OR
      (uppercase_CMAKE_BUILD_TYPE STREQUAL "RELWITHDEBINFO")))
   # Limit to clang and gcc so far. Add compilers supporting this option.
   if (CMAKE_CXX_COMPILER_ID MATCHES "Clang" OR
       CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
     add_compile_options(-gsplit-dwarf)
+
+    # Turn on -Wl,--gdb-index if requested and supported
+    if (LLVM_USE_GDB_INDEX AND
+        (LLVM_USE_LINKER STREQUAL "gold" OR LLVM_USE_LINKER STREQUAL "lld"))
+      append("-Wl,--gdb-index" CMAKE_EXE_LINKER_FLAGS CMAKE_SHARED_LINKER_FLAGS)
+    endif()
   endif()
 endif()
 
 add_definitions( -D__STDC_CONSTANT_MACROS )
 add_definitions( -D__STDC_FORMAT_MACROS )
 add_definitions( -D__STDC_LIMIT_MACROS )
 
 # clang and gcc don't default-print colored diagnostics when invoked from Ninja.

llvm/docs/CMake.rst

 **LLVM_USE_RELATIVE_PATHS_IN_DEBUG_INFO**:BOOL
   Rewrite absolute source paths in debug info to relative ones. The source prefix
   can be adjusted via the LLVM_SOURCE_PREFIX variable.
 
 **LLVM_USE_RELATIVE_PATHS_IN_FILES**:BOOL
   Rewrite absolute source paths in sources and debug info to relative ones. The
   source prefix can be adjusted via the LLVM_SOURCE_PREFIX variable.
 
+**LLVM_USE_SPLIT_DWARF**:BOOL
+  If enabled, store DWARF debug information separately from object files. Enabling
+  this option can speed up build times in Debug configurations.
+
+**LLVM_USE_GDB_INDEX**:BOOL
+  If enabled, generate GDB index tables for locating split DWARF debug
+  information at link time. Enabling this option can speed up GDB debugging
+  startup times in Debug configurations with LLVM_USE_SPLIT_DWARF enabled.
+
 CMake Caches
 ============
 
 Recently LLVM and Clang have been adding some more complicated build system
 features. Utilizing these new features often involves a complicated chain of
 CMake variables passed on the command line. Clang provides a collection of CMake
 cache scripts to make these features more approachable.