[Patch]: Fix PR middle-end/25158
Kaveh R. Ghazi
ghazi@caipclassic.rutgers.edu
Wed Nov 30 17:34:00 GMT 2005
I believe this patch fixes PR middle-end/25158. We're not eliminating
"unlocked" stdio calls with empty output strings on hpux.
Hpux doesn't have fputc_unlocked, and clears out that as a possible
replacement function. As I noted in the PR, we have to defer the
checks in fold_builtin_fputs for missing replacement functions in
order to reach the code which eliminates empty output strings. (All
other builtins seem to defer the check this way also.)
Tested with a full bootstrap on i686-unknown-linux-gnu, no
regressions. I also built a cross-compiler targetted to
hppa2.0w-hp-hpux11.11 and compiled a testcase with empty strings
passed to fprintf_unlocked and fputs_unlocked and ensured they were
optimized away.
I'm going to wait for confirmation from Dave that everything's okay on
a native hpux test.
Pending that, okay for mainline and 4.1?
Thanks,
--Kaveh
2005-11-29 Kaveh R. Ghazi <ghazi@caip.rutgers.edu>
PR middle-end/25158
* builtins.c (fold_builtin_fputs): Defer check for missing
replacement functions.
diff -rup orig/egcc-SVN20051129/gcc/builtins.c egcc-SVN20051129/gcc/builtins.c
--- orig/egcc-SVN20051129/gcc/builtins.c 2005-11-29 00:18:53.000000000 -0500
+++ egcc-SVN20051129/gcc/builtins.c 2005-11-29 16:57:20.000000000 -0500
@@ -9639,9 +9639,8 @@ fold_builtin_fputs (tree arglist, bool i
tree const fn_fwrite = unlocked ? built_in_decls[BUILT_IN_FWRITE_UNLOCKED]
: implicit_built_in_decls[BUILT_IN_FWRITE];
- /* If the return value is used, or the replacement _DECL isn't
- initialized, don't do the transformation. */
- if (!ignore || !fn_fputc || !fn_fwrite)
+ /* If the return value is used, don't do the transformation. */
+ if (!ignore)
return 0;
/* Verify the arguments in the original call. */
@@ -9703,6 +9702,11 @@ fold_builtin_fputs (tree arglist, bool i
gcc_unreachable ();
}
+ /* If the replacement _DECL isn't initialized, don't do the
+ transformation. */
+ if (!fn)
+ return 0;
+
/* These optimizations are only performed when the result is ignored,
hence there's no need to cast the result to integer_type_node. */
return build_function_call_expr (fn, arglist);
More information about the Gcc-patches
mailing list