Executing on host: /test/gnu/gcc/objdir/gcc/xgcc -B/test/gnu/gcc/objdir/gcc/ /te st/gnu/gcc/gcc/gcc/testsuite/gcc.dg/builtins-20.c -O2 -ffast-math -fno-show-co lumn -lm -o builtins-20.exe (timeout = 300) ld: Unsatisfied symbol "cargf" in file /var/tmp//ccKMBpML.o ld: Unsatisfied symbol "cargl" in file /var/tmp//ccKMBpML.o ld: Unsatisfied symbol "atan2l" in file /var/tmp//ccKMBpML.o ld: Unsatisfied symbol "link_error" in file /var/tmp//ccKMBpML.o 4 errors. collect2: ld returned 1 exit status compiler exited with status 1 output is: ld: Unsatisfied symbol "cargf" in file /var/tmp//ccKMBpML.o ld: Unsatisfied symbol "cargl" in file /var/tmp//ccKMBpML.o ld: Unsatisfied symbol "atan2l" in file /var/tmp//ccKMBpML.o ld: Unsatisfied symbol "link_error" in file /var/tmp//ccKMBpML.o 4 errors. collect2: ld returned 1 exit status
Subject: Re: New: FAIL: gcc.dg/builtins-20.c (test for excess errors) test3l and test3f fail. Attached .s file. Dave
Created attachment 14115 [details] builtins-20.s
I believe that this test started failing when mode cases were added in the "Handle CONJ_EXPR in negate_expr_p and fold_negate_expr". Test doesn't fail on linux, so this has something to do with missing C99 functions.
Sorry, comment #3 is wrong. This change was introduced by the additions to builtins-20.c for the carg(z) -> atan2(z.i, z.r). There we have: +static tree +fold_builtin_carg(tree arglist, tree type) +{ + if (validate_arglist (arglist, COMPLEX_TYPE, VOID_TYPE)) + { + tree atan2_fn = mathfn_built_in (type, BUILT_IN_ATAN2); + + if (atan2_fn) + { + tree arg = builtin_save_expr (TREE_VALUE (arglist)); + tree r_arg = fold_build1 (REALPART_EXPR, type, arg); + tree i_arg = fold_build1 (IMAGPART_EXPR, type, arg); + tree newarglist = tree_cons (NULL_TREE, i_arg, + build_tree_list (NULL_TREE, r_arg)); + return build_function_call_expr (atan2_fn, newarglist); + } + } + + return NULL_TREE; +} PA-RISC HP-UX has atan2 and atan2f, but not atan2l. We have in builtins.def: DEF_C99_C90RES_BUILTIN (BUILT_IN_ATAN2F, "atan2f", BT_FN_FLOAT_FLOAT_FLOAT, ATTR _MATHFN_FPROUNDING_ERRNO) and /* Builtin that is specified by C99 and C90 reserve the name for future use. We can still recognize the builtin in C90 mode but we can't produce it implicitly. */ #undef DEF_C99_C90RES_BUILTIN #define DEF_C99_C90RES_BUILTIN(ENUM, NAME, TYPE, ATTRS) \ DEF_BUILTIN (ENUM, "__builtin_" NAME, BUILT_IN_NORMAL, TYPE, TYPE, \ true, true, !flag_isoc99, ATTRS, TARGET_C99_FUNCTIONS, true) Think we need to do something about atan2f in the backend. However, we don't have atan2l, so the test will have to be xfailed...
There doesn't seem to be any difference between the definitions for DEF_C99_BUILTIN and DEF_C99_C90RES_BUILTIN.
It looks like this test is passing since the test was changed to include the #ifdef HAVE_C99_RUNTIME. It looks OK on the 4.3 branch and trunk so I think we can close it. Any objections?
Subject: Re: FAIL: gcc.dg/builtins-20.c (test for excess errors) > I think we can close it. Any objections? No. Dave
Fixed with change to test.