[flang-commits] [clang] [flang] [llvm] [flang][cmake][perf-training] Optimize flang with PGO and BOLT (PR #198863)

[Paul Osmialowski via flang-commits]

================
@@ -30,8 +30,17 @@ config.default_sysroot = "@DEFAULT_SYSROOT@"
 config.flang_runtime_f128_math_lib = "@FLANG_RUNTIME_F128_MATH_LIB@"
 config.have_ldbl_mant_dig_113 = "@HAVE_LDBL_MANT_DIG_113@"
 
+config.perf_helper_dir = "@LLVM_MAIN_SRC_DIR@/../clang/utils/perf-training"
+config.python3_exe = "@Python3_EXECUTABLE@"
+config.flang_bolt_mode = "@FLANG_BOLT@"
+config.flang_bolt_name = "@FLANG_BOLT_INSTRUMENTED@"
+config.flang_opt_tools_dir = lit_config.substitute("@CURRENT_TOOLS_DIR@")
----------------
pawosm-arm wrote:

That's the key difference here. when we use `config.flang_opt_tools_dir` we want to specifically access/run the instrumented flang binary. We run this during flang optimization, hence the `opt` infix.

https://github.com/llvm/llvm-project/pull/198863