[flang-commits] [flang] [flang][cmake][bolt] Do not train on non-strictly-fortran test cases (PR #206130)

[Kiran Chandramohan via flang-commits]

================
@@ -29,3 +29,27 @@ config.name = "Flang BOLT Training"
 config.flang_exe = perf_wrapper + flang_nowrapper
 
 config.llvm_profile_file = ''
+
+# suffixes: A list of file extensions to treat as test files.
+config.suffixes = [
+    ".f",
+    ".F",
+    ".ff",
+    ".FOR",
+    ".for",
+    ".f77",
+    ".f90",
+    ".F90",
+    ".ff90",
+    ".f95",
+    ".F95",
+    ".ff95",
+    ".fpp",
+    ".FPP",
+    ".f18",
+    ".F18",
+    ".f03",
+    ".F03",
+    ".f08",
+    ".F08",
+]
----------------
kiranchandramohan wrote:

@pawosm-arm Could you remind me what exactly we are focusing on improving with pgo and bolt here? Not requesting any change here but just trying to understand this a bit better.

Tests in flang/test/Driver would only exercise the Driver.
Tests in flang/test/Parser would only exercise the Driver, Parser.
Tests in flang/test/ Semantics/Evaluate would exercise the Driver, Parser and Semantics.
Tests in flang/test/Lower would exercise Driver, Parser, Semantics and Lowering
Tests in flang/test/Integration would exercise everything upto LLVM IR generation. But there are only 51 of these.

Most of the tests in Fir/HLFIR that tests transforming or optimizing Fir/HFIR would be ignored by this change since you are not including the *.fir/*.mlir tests.


https://github.com/llvm/llvm-project/pull/206130