crash when linking swc with Alchemy

Question

I have a project I'm trying to compile with alchemy. It will compile .o and .a files, but when trying to create a .swc, it will fail. It appears to crash with this error:

g++ -swc -o mylib.swc my-flex-interface.cpp mylib.a
    Cannot yet select: 0x279c810: ch,flag = AVM2ISD::CALL - A call instruction 0x279c7a0, 0x29c4350
    0   llc                                 0x00636dfe _ZNSt8_Rb_treeIN4llvm3sys4PathES2_St9_IdentityIS2_ESt4lessIS2_ESaIS2_EE13insert_uniqueERKS2_ + 6078
    1   llc                                 0x006373a2 _ZNSt8_Rb_treeIN4llvm3sys4PathES2_St9_IdentityIS2_ESt4lessIS2_ESaIS2_EE13insert_uniqueERKS2_ + 7522
    2   libSystem.B.dylib                   0x9530942b _sigtramp + 43
    3   ???                                 0xffffffff 0x0 + 4294967295
    4   libSystem.B.dylib                   0x953968e5 raise + 26
    5   libSystem.B.dylib                   0x953ac99c abort + 93
    6   llc                                 0x002f4fe0 _ZN98_GLOBAL__N__Volumes_data_dev_FlaCC_llvm_2.1_lib_Target_AVM2_AVM2ISelDAGToDAG.cpp_00000000_F04616B616AVM2DAGToDAGISel6Emit_7ERKN4llvm9SDOperandEj + 0
    7   llc                                 0x002f8e1b _ZN98_GLOBAL__N__Volumes_data_dev_FlaCC_llvm_2.1_lib_Target_AVM2_AVM2ISelDAGToDAG.cpp_00000000_F04616B616AVM2DAGToDAGISel10SelectCodeEN4llvm9SDOperandE + 2219
    8   llc                                 0x002fa193 _ZN98_GLOBAL__N__Volumes_data_dev_FlaCC_llvm_2.1_lib_Target_AVM2_AVM2ISelDAGToDAG.cpp_00000000_F04616B616AVM2DAGToDAGISel10SelectRootEN4llvm9SDOperandE + 819
    9   llc                                 0x002e6a2c _ZN4llvm19X86_64TargetMachineD0Ev + 65116
    10  llc                                 0x003de4ca _ZN4llvm11StoreSDNodeD1Ev + 1610
    11  llc                                 0x0040d3fe _ZN4llvm11StoreSDNodeD1Ev + 193918
    12  llc                                 0x0040f92e _ZN4llvm11StoreSDNodeD1Ev + 203438
    13  llc                                 0x005d1926 _ZN4llvm12FunctionPassD1Ev + 20998
    14  llc                                 0x005d1f3a _ZN4llvm12FunctionPassD1Ev + 22554
    15  llc                                 0x005d20c5 _ZN4llvm12FunctionPassD1Ev + 22949
    16  llc                                 0x00002e44 0x0 + 11844
    17  llc                                 0x00001f36 0x0 + 7990
    18  ???                                 0x00000006 0x0 + 6
    make[2]: *** [src/app/alchemy/sonic.swc] Error 6
    make[1]: *** [src/app/alchemy/CMakeFiles/alchemy.dir/all] Error 2
    make: *** [all] Error 2

I'm not familiar enough with LLVM (which Alchemy uses under the hood) to figure out what this error means. Any ideas?

Answer 1

As of Alchemy v0.5a, when compiling memcpy/memmove/memset the LLVM frontend will sometimes (on Mac OS X) emit intermediate code that the Alchemy backend linker can't handle. It will crash with the "cannot yet select" error.

The workaround is either to compile on a different platform (Ubuntu, Windows) or to override the implementations of memcpy/memmove/memset:

static void * custom_memmove( void * destination, const void * source, size_t num ) {
  void *result;
  __asm__("%0 memmove(%1, %2, %3)\n" : "=r"(result) : "r"(destination), "r"(source), "r"(num));
  return result;
}
static void * custom_memcpy ( void * destination, const void * source, size_t num ) {
  void *result;
  __asm__("%0 memcpy(%1, %2, %3)\n" : "=r"(result) : "r"(destination), "r"(source), "r"(num));
  return result;
}
static void * custom_memset ( void * ptr, int value, size_t num ) {
  void *result;
  __asm__("%0 memset(%1, %2, %3)\n" : "=r"(result) : "r"(ptr), "r"(value), "r"(num));
  return result;
}
#define memmove custom_memmove
#define memcpy custom_memcpy
#define memset custom_memset

Unfortunately this workaround requires putting the above code (via a #include) into every module that uses memcpy.

Ultimately, I injected the memcpy/memset/memmove workarounds by overriding string.h (using -I and #include_next), but this required hacking the achacks/gcc script so that the -I flags are processed in the right order:

sub dash_I
{
  my $arg = (shift);
  my $path;

  if($arg =~ /-I=?(.*)/)
    { $path = $1 } 
  else
    { $path = shift(ARGV) }
  my $ppath = $incmap{$path};
  my @pathes = $ppath ? @$ppath : ($path);

  # BUGFIX: insert user -I header search paths
  # early so that they override the system search paths
  #
  #push(@compile, map { "-I$_" } @pathes);
  splice(@compile, 1, 0, map { "-I$_" } @pathes);

  0
}