A possible approach to writing an LLVM backend

Playing around with dragonegg, I've found that I can get pretty detailed llvm-ir/asm output, and I can even make sense of how the C output by Irken is getting translated. Seems like this could be a handy way of getting started on writing an llvm backend. Not that I'm thinking of doing that. Must... resist...

Here's the count-to-a-million test code in Scheme/Irken:

(let loop ((n 1000000))
  (if (zero? n)
      "done"
      (loop (- n 1))))

Here's the C output:

FUN_loop_8:
    r0 = varref (0,0);
    if PXLL_IS_TRUE(PXLL_TEST(unbox(r0)==0)) {
      r0 = (object*) &constructed_0;
      PXLL_RETURN(0);
    } else {
      r0 = varref (0,0);
      r1 = (object *) 3;
      r0 = box(unbox(r0)-unbox(r1));
      lenv[2] = r0;
      goto FUN_loop_8;
    }
    PXLL_RETURN(0);
  L1:
  r1 = allocate (TC_CLOSURE, 2);
  r1[1] = &&FUN_loop_8; r1[2] = lenv;
  r0[2] = r1;
  r0 = allocate (TC_TUPLE, 2);
  r1 = (object *) 2000001;
  r0[2] = r1;
  r1 = top[2];
  r0[1] = r1[2]; lenv = r0; goto FUN_loop_8;

And the relevant LLVM IR:

"<bb 12>":                                        ; preds = %"<bb 17>", %"<bb 13>", %"<bb 24>"
  %D.5560_160 = phi i8* [ inttoptr (i64 2000001 to i8*), %"<bb 24>" ], [ %D.5560_160, %"<bb 13>" ], [ %4, %"<bb 17>" ] ; <i8*> [#uses=3]
  %3 = icmp ult i8* %D.5560_160, inttoptr (i64 2 to i8*) ; <i1> [#uses=1]
  br i1 %3, label %"<bb 13>", label %"<bb 17>"

"<bb 13>":                                        ; preds = %"<bb 12>"
  %D.4871_48 = load i8** %2, align 8              ; <i8*> [#uses=1]
  indirectbr i8* %D.4871_48, [label %"<bb 12>", label %Lreturn]

"<bb 17>":                                        ; preds = %"<bb 12>"
  %n.85_162 = ptrtoint i8* %D.5560_160 to i64     ; <i64> [#uses=1]
  %D.5572_16323 = add i64 %n.85_162, -2           ; <i64> [#uses=1]
  %D.5579_167 = or i64 %D.5572_16323, 1           ; <i64> [#uses=1]
  %4 = inttoptr i64 %D.5579_167 to i8*            ; <i8*> [#uses=2]
  store i8* %4, i8** %11, align 8
  br label %"<bb 12>"