When optimizing, GCC converts calls to cos() and sin() (and probably others,
these were the ones I saw) into calls to cosf() and sinf() when cos() and sin()
are called with a "float" argument.  The problem with this is that neither
libgcc nor uClibc provide cosf and sinf!  The bug can be worked around by
passing -fno-builtin-cos and -fno-builtin-sin.

The trivial code:
extern double cos(double);
float do_cos(float f)
{ return cos(f); }

Produces this assembly code output:
        .file   "test.c"
        .text
        .align  2
        .global do_cos
        .type   do_cos, %function
do_cos:
        @ args = 0, pretend = 0, frame = 0
        @ frame_needed = 0, uses_anonymous_args = 0
        @ link register save eliminated.
        @ lr needed for prologue
        b       cosf
        .size   do_cos, .-do_cos
        .ident  "GCC: (GNU) 3.4.2"

When run as "arm-linux-gcc -S -O2 test.c"

Output of "arm-linux-gcc -v:"
Reading specs from
/home/jonathan/workspace/gumstix/gumstix-buildroot/build_arm_nofpu/staging_dir/bin-ccache/../lib/gcc/arm-linux-uclibc/3.4.2/specs
Configured with:
/home/jonathan/workspace/gumstix/gumstix-buildroot/toolchain_build_arm_nofpu/gcc-3.4.2/configure
--prefix=/home/jonathan/workspace/gumstix/gumstix-buildroot/build_arm_nofpu/staging_dir
--build=i386-pc-linux-gnu --host=i386-pc-linux-gnu --target=arm-linux-uclibc
--enable-languages=c,c++ --enable-shared --disable-__cxa_atexit
--enable-target-optspace --with-gnu-ld --disable-nls --with-float=soft
--enable-sjlj-exceptions
Thread model: posix
gcc version 3.4.2

-- 
           Summary: GCC incorrectly issues calls to sinf, cosf
           Product: gcc
           Version: 3.4.2
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: c
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: jbrandmeyer at earthlink dot net
                CC: gcc-bugs at gcc dot gnu dot org
  GCC host triplet: i386-pc-linux-gnu
GCC target triplet: arm-linux-uclibc


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20353

Reply via email to