Add -O[0123s] gcc optimization flags for gcc pass 4 speedup.

* session.h (struct systemtap_session): Add gcc_flags string field.
* main.cxx (usage): Document -O[0123s].
  (main): Default gcc_flags to -O0. Add O:: to getopt_long. Handle case
  'O' to set gcc_flags.
* buildrun.cxx (compile_pass): Add gcc_flags to EXTRA_CFLAGS.
* stap1.in: Add -O[0123s] documentation.
* testsuite/systemtap.base/cache.exp: Add tests for -O[0123s].

6 files changed, 53 insertions, 2 deletions

diff --git a/buildrun.cxx b/buildrun.cxx
index ccf1ca15..b54a0534 100644
--- a/buildrun.cxx
+++ b/buildrun.cxx
@@ -187,7 +187,7 @@ compile_pass (systemtap_session& s)
   // if (s.keep_tmpdir)
   // o << "CFLAGS += -fverbose-asm -save-temps" << endl;
 
-  o << "EXTRA_CFLAGS += -freorder-blocks" << endl; // improve on -Os
+  o << "EXTRA_CFLAGS += " << s.gcc_flags << endl; // Add -O[0123s]
 
   // o << "CFLAGS += -fno-unit-at-a-time" << endl;
 
diff --git a/hash.cxx b/hash.cxx
index 45ae05eb..4e85ed28 100644
--- a/hash.cxx
+++ b/hash.cxx
@@ -175,6 +175,7 @@ find_script_hash (systemtap_session& s, const string& script, const hash &base)
   h.add(s.ignore_dwarf);		// --ignore-dwarf
   h.add(s.consult_symtab);		// --kelf, --kmap
   h.add(s.skip_badvars);		// --skip-badvars
+  h.add(s.gcc_flags);                   // -O[0123s]
   if (!s.kernel_symtab_path.empty())	// --kmap
     {
       h.add(s.kernel_symtab_path);
diff --git a/main.cxx b/main.cxx
index 9dc658ff..2aeff580 100644
--- a/main.cxx
+++ b/main.cxx
@@ -111,6 +111,14 @@ usage (systemtap_session& s, int exitcode)
     << "              " << s.module_name << endl
     << "   -o FILE    send script output to file, instead of stdout. This supports" << endl
     << "              strftime(3) formats for FILE" << endl
+    << "   -O[0123s]  optimization to use for C code. Passed to gcc in pass 4." << endl
+    << "              -O  Alias for -O1" << endl
+    << "              -O0 Fast compilation" << endl
+    << "              -O1 Optimize, takes a bit more time" << endl
+    << "              -O2 Optimize more, takes more time" << endl
+    << "              -O3 Optimize even more, takes even more time" << endl
+    << "              -Os Optimize for size, like -O2 but tuned for small code size" << endl
+    << "              Default is -O0." << endl
     << "   -c CMD     start the probes, run CMD, and exit when it finishes" << endl
     << "   -x PID     sets target() to PID" << endl
     << "   -F         run as on-file flight recorder with -o." << endl
@@ -403,6 +411,7 @@ main (int argc, char * const argv [])
   s.ignore_dwarf = false;
   s.load_only = false;
   s.skip_badvars = false;
+  s.gcc_flags = "-O0";
 
   // Location of our signing certificate.
   // If we're root, use the database in SYSCONFDIR, otherwise
@@ -487,7 +496,7 @@ main (int argc, char * const argv [])
         { "vp", 1, &long_opt, LONG_OPT_VERBOSE_PASS },
         { NULL, 0, NULL, 0 }
       };
-      int grc = getopt_long (argc, argv, "hVMvtp:I:e:o:R:r:m:kgPc:x:D:bs:uqwl:d:L:FS:",
+      int grc = getopt_long (argc, argv, "hVMvtp:I:e:o:O::R:r:m:kgPc:x:D:bs:uqwl:d:L:FS:",
                                                           long_options, NULL);
       if (grc < 0)
         break;
@@ -551,6 +560,22 @@ main (int argc, char * const argv [])
           s.output_file = string (optarg);
           break;
 
+        case 'O':
+	  if (optarg == NULL)
+	    s.gcc_flags = "-O1";
+	  else
+	    s.gcc_flags = "-O" + string (optarg);
+
+	  if (s.gcc_flags != "-O0" && s.gcc_flags != "-O1"
+	      && s.gcc_flags != "-O2" && s.gcc_flags != "-O3"
+	      && s.gcc_flags != "-Os")
+	    {
+	      cerr << "'" << s.gcc_flags << "'"
+		   << " isn't a valid optimization option." << endl;
+	      usage (s, 1);
+	    }
+          break;
+
         case 'R':
           s.runtime_path = string (optarg);
           break;
diff --git a/session.h b/session.h
index a617e47f..32004410 100644
--- a/session.h
+++ b/session.h
@@ -135,6 +135,9 @@ struct systemtap_session
   // Skip bad $ vars
   bool skip_badvars;
 
+  // Optimization flag passed to gcc -O[0123s]
+  std::string gcc_flags;
+
   // NB: It is very important for all of the above (and below) fields
   // to be cleared in the systemtap_session ctor (elaborate.cxx)
   // and/or main.cxx(main).
diff --git a/stap.1.in b/stap.1.in
index 7736a612..1a287a3e 100644
--- a/stap.1.in
+++ b/stap.1.in
@@ -183,6 +183,13 @@ Send standard output to named file. In bulk mode, percpu files will
 start with FILE_ (FILE_cpu with -F) followed by the cpu number.
 This supports strftime(3) formats for FILE.
 .TP
+.BI \-O[0123s]
+Optimization to use for C code. Passed to gcc in pass 4. -O is an
+alias for -O1, -O0 enables fast compilation, -O1 optimizes the code and
+takes a bit more time, -O2 optimizes more and takes more time, -O3
+optimizes even more and takes even more time, -Os optimizes for size,
+like -O2 but tuned for small code size. The default is -O0.
+.TP
 .BI \-c " CMD"
 Start the probes, run CMD, and exit when CMD finishes.
 .TP
diff --git a/testsuite/systemtap.base/cache.exp b/testsuite/systemtap.base/cache.exp
index 5c5c1441..e70f3334 100644
--- a/testsuite/systemtap.base/cache.exp
+++ b/testsuite/systemtap.base/cache.exp
@@ -102,6 +102,21 @@ stap_compile MERGE2 [F_CACHED_COMPILE] $basic_script1 -b -M
 stap_compile TIMING1 [F_UNCACHED_COMPILE] $basic_script1 -t
 stap_compile TIMING2 [F_CACHED_COMPILE] $basic_script1 -t
 
+# Using '-O[0123s]' (C compile optimizations) should change the hash
+# unless they are the same
+stap_compile OPTO [F_UNCACHED_COMPILE] $basic_script1 -O
+# -O0 is the default, so will have been cached by BASIC1
+stap_compile OPTZERO [F_CACHED_COMPILE] $basic_script1 -O0
+# -O is an alias for -O1, so already cached
+stap_compile OPTONE [F_CACHED_COMPILE] $basic_script1 -O1
+stap_compile OPTTWO [F_UNCACHED_COMPILE] $basic_script1 -O2
+stap_compile OPTTHREE [F_UNCACHED_COMPILE] $basic_script1 -O3
+stap_compile OPTSMALL [F_UNCACHED_COMPILE] $basic_script1 -Os
+# Bad -O argument should be rejected
+stap_compile OPTBAD [F_UNCACHED_NO_COMPILE] $basic_script1 -OX
+
+stap_compile TIMING2 [F_CACHED_COMPILE] $basic_script1 -t
+
 # Changing the runtime directory should change the hash
 set new_runtime [exec pwd]/.cache_test_runtime
 exec /bin/rm -f $new_runtime