[LLVMdev] getting closer!

[Terence Parr]

Ok, I *might* be getting this from the assembly code.  The assembly  
code has:

L_llvm_gc_root_chain$non_lazy_ptr:
         .indirect_symbol _llvm_gc_root_chain
         .long   0

and I see it being used in the function preamble.  Is that a ref to an  
extern symbol or the def?  I.e., is it referring to

StackEntry *llvm_gc_root_chain;

that I must have in my GC C code? (semispace.c has it)

SO! I might be getting this.  The shadow stack plugin assumes I have

struct StackEntry {
   StackEntry *Next;       // Caller's stack entry.
   const FrameMap *Map;    // Pointer to constant FrameMap.
   void *Roots[];          // Stack roots (in-place array).
};

as my stack item layout and I must provide a shadow stack head.  From  
that, it will push/pop in functions?  If so, that's easy enough. :)   
What I was/am missing is the explicit link between types and variables  
in a GC.c file and the generated machine code.  If I can get that last  
explicit link, I'm off to the races.  Anybody?  My IR doesn't seem to  
have any roots, even though I've allocated an int and declared a ptr  
on the stack.

declare void @llvm.gcroot(i8 **, i8*)
declare void @llvm_gc_collect()
declare i32* @llvm_gc_allocate(i32)
declare void @llvm_gc_initialize(i32)

define void @foo() gc "shadow-stack" {
; int *pa = malloc(sizeof(int));
     %a = call i32* @llvm_gc_allocate(i32 4)
     %pa = alloca i32*
     store i32* %a, i32** %pa

     %c = bitcast i32** %pa to i8**
     call void @llvm.gcroot(i8** %c, i8* null)
; *pa = 99;
     %t0 = add i32 99,0
     %t1 = load i32** %pa
     ;%t2 = getelementptr i32** %t1, i32 0
     store i32 %t0, i32* %t1

     store i32* null, i32** %pa; say it's dead
     ret void
}

define void @main() {
     call void @llvm_gc_initialize(i32 1024)
     call void @foo()
     call void @llvm_gc_collect()
     ret void
}

I get llvm_gc_root_chain as null when I try to walk roots.

Ter