Linker error truncating from float

[MinliLiao]

I got the following linker error when trying with the example from README:

1. ld.lld: error: undefined symbol: __raptor_fprt_ieee_32_func_sqrtf referenced by __raptor_done_truncate_op_func_ieee_32_to_mpfr_5_8_0_0_1__ZSt4sqrtf
2. ld.lld: error: undefined symbol: __raptor_fprt_ieee_32_binop_fadd referenced by __raptor_done_truncate_op_func_ieee_32_to_mpfr_5_8_0_0_1__Z3barff.

• Source code that resulted in the linker error:

  #include <math.h>
  
  template <typename fty> fty *__raptor_truncate_op_func(fty *, int, int, int, int);
  
  float bar(float a, float b) {
    return a + b;
  }
  float foo(float *a, float b) {
    a[0] = sqrt(b);
    return bar(a[1], b);
  }
  
  int main() {
    float a[2], b;
    auto f = __raptor_truncate_op_func(
      /* function */    foo,
      /* from_type */   32,
      /* to_type: 0 for builtin IEEE type, 1 for MPFR */   1,
      /* to_exponent */ 5,
      /* to_mantissa */ 8);
  
    f(a,b);
  
    return 0;
  }

• Compiled with raptor-clang++ <sourcefile>

• LLVM was installed with spack install llvm+clang+flang+lld+mlir@20, the spec in my spack env is as follows:

  ==> 2 root specs
  -- no arch / no compilers ---------------------------------------
  [+] llvm@20+clang+flang+lld+mlir  [+] py-lit
  
  -- linux-fedora42-x86_64 / no compilers -------------------------
  gcc@15.2.1  glibc@2.41
  
  -- linux-fedora42-zen2 / %c,cxx=gcc@15.2.1 ----------------------
  berkeley-db@18.1.40  cmake@3.31.11  expat@2.7.4  hwloc@2.12.2  llvm@20.1.8           nghttp2@1.67.1  openssl@3.6.1  re2c@4.4    zlib-ng@2.3.2
  binutils@2.43.1      curl@8.18.0    gettext@1.0  libffi@3.5.2  ncurses@6.5-20250705  ninja@1.13.2    python@3.14.3  swig@4.1.1  zstd@1.5.7
  
  -- linux-fedora42-zen2 / %c=gcc@15.2.1 --------------------------
  bzip2@1.0.8     gmake@4.4.1           libiconv@1.18      libxml2@2.15.1  perl@5.42.0    readline@8.3   unzip@6.0
  diffutils@3.12  libbsd@0.12.2         libmd@1.1.0        lua@5.3.6       pigz@2.8       sqlite@3.51.2  util-linux-uuid@2.41
  gdbm@1.26       libedit@3.1-20240808  libpciaccess@0.17  pcre2@10.44     pkgconf@2.5.1  tar@1.35       xz@5.6.3
  
  -- linux-fedora42-zen2 / no compilers ---------------------------
  ca-certificates-mozilla@2025-08-12  gcc-runtime@15.2.1      py-lit@18.1.8  py-setuptools@80.9.0  python-venv@1.0
  compiler-wrapper@1.0                perl-data-dumper@2.173  py-pip@25.1.1  py-wheel@0.45.1       util-macros@1.20.2
  

• Other details:
  There was no problem when truncating from double instead of float.
  Same error occurs when using #include <cmath> and std::sqrt, looks like sqrtf is used for float from both cmath and math.h.

PS: The "TYPE" on line 38 of runtime/ir/Flops.def may be a bug.