http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45890
--- Comment #2 from Sharad Singhai <singhai at gcc dot gnu.org> 2011-10-01
19:42:48 UTC ---
I don't know if it is related. But the coverage is imprecise in case of
attached source file foo.c. Build it with O2, as
gcc --coverage -O2 foo.c
./a.out
gcov -b foo.c
Now the coverage data for the 'if' condition in 'sum' looks like this:
(in attached file foo.c.gcov)
8: 8: if (v[i]) total += 1;
branch 0 never executed
branch 1 never executed
branch 2 taken 75% (fallthrough)
branch 3 taken 25%
Thus a simple two-way conditional looks like a four-way branch. It is due to
early inlining of 'sum' into 'main where a couple of bb's get eliminated/merged
so that the branch coverage incorrectly gets attributed to the conditional.
Similarly the coverage data for the for-loop in 'sum' looks like this
9: 7: for (i = 0; i < N; ++i) {
branch 0 never executed
branch 1 never executed
branch 2 taken 89%
branch 3 taken 11% (fallthrough)
After disabling early inlining, the coverage data looks saner. (Attached in
foo.c.gcov.no-early-inlining.)
gcc --coverage -O2 -fno-early-inlining foo.c
./a.out
gcov -b foo.c
Adding the following patch mitigates the issue.
2011-09-30 Sharad Singhai <sing...@google.com>
* gcc.c (cc1_options): Added -fno-early-inlining for coverage.
Index: gcc.c
===================================================================
--- gcc.c (revision 179402)
+++ gcc.c (working copy)
@@ -776,7 +776,7 @@
%{!fsyntax-only:%{S:%W{o*}%{!o*:-o %b.s}}}\
%{fsyntax-only:-o %j} %{-param*}\
%{fmudflap|fmudflapth:-fno-builtin -fno-merge-constants}\
- %{coverage:-fprofile-arcs -ftest-coverage}";
+ %{coverage:-fprofile-arcs -ftest-coverage -fno-early-inlining}";
/* If an assembler wrapper is used to invoke post-assembly tools
like MAO, --save-temps need to be passed to save the output of
