LLVM's x86 code generation from the IR is very impressive in that it recognizes sub-blocks in common and does jumps to them.

gistfile1.md

Below is the handwritten LLVM IR for SP_2SWAP and SP_ROT

SP_2SWAP:

kernel.SP_2SWAP:
    ; swap top two pairs of elements on the stack
    ;   pop %eax
    ;   pop %ebx
    ;   pop %ecx
    ;   pop %edx
    ;   push %ebx
    ;   push %eax
    ;   push %edx
    ;   push %ecx
    %SP.ptr.SP_2SWAP = getelementptr %cell.ptr* %SP.ptr.ptr, i32 0
    %SP.SP_2SWAP = load %cell.ptr* %SP.ptr.SP_2SWAP

    %A.addr.ptr.SP_2SWAP = getelementptr %cell.ptr %SP.SP_2SWAP, i32 0
    %A.addr.int.SP_2SWAP = ptrtoint %cell.ptr %A.addr.ptr.SP_2SWAP to %addr
    %A.cell.SP_2SWAP = load %cell* %A.addr.ptr.SP_2SWAP

    %B.addr.int.SP_2SWAP = add %addr %A.addr.int.SP_2SWAP, 8
    %B.addr.ptr.SP_2SWAP = inttoptr %addr %B.addr.int.SP_2SWAP to %cell*
    %B.cell.SP_2SWAP = load %cell* %B.addr.ptr.SP_2SWAP

    %C.addr.int.SP_2SWAP = add %addr %B.addr.int.SP_2SWAP, 8
    %C.addr.ptr.SP_2SWAP = inttoptr %addr %C.addr.int.SP_2SWAP to %cell*
    %C.cell.SP_2SWAP = load %cell* %C.addr.ptr.SP_2SWAP

    %D.addr.int.SP_2SWAP = add %addr %C.addr.int.SP_2SWAP, 8
    %D.addr.ptr.SP_2SWAP = inttoptr %addr %D.addr.int.SP_2SWAP to %cell*
    %D.cell.SP_2SWAP = load %cell* %D.addr.ptr.SP_2SWAP

    store %cell %B.cell.SP_2SWAP, %cell* %D.addr.ptr.SP_2SWAP  ; %(edx)
    store %cell %A.cell.SP_2SWAP, %cell* %C.addr.ptr.SP_2SWAP  ; %(ecx)
    store %cell %D.cell.SP_2SWAP, %cell* %B.addr.ptr.SP_2SWAP  ; %(ebx)
    store %cell %C.cell.SP_2SWAP, %cell* %A.addr.ptr.SP_2SWAP  ; %(eax)

    br label %kernel.NEXT

SP_ROT:

kernel.SP_ROT:
    ; rotate the first three elements at the top of the stack
    ;   pop %eax
    ;   pop %ebx
    ;   pop %ecx
    ;   push %ebx
    ;   push %eax
    ;   push %ecx

    ; load the memory address that %SP.ptr.ptr resolves to
    %SP.ptr.SP_ROT = getelementptr %cell.ptr* %SP.ptr.ptr, i32 0
    %SP.SP_ROT = load %cell.ptr* %SP.ptr.SP_ROT
    %SP.addr.ptr.SP_ROT = getelementptr %cell.ptr %SP.SP_ROT, i32 0
    %SP.addr.int.SP_ROT = ptrtoint %cell.ptr %SP.addr.ptr.SP_ROT
                                 to %addr

    ; load %eax
    %A.int.SP_ROT = load %cell* %SP.addr.ptr.SP_ROT

    ; load %ebx
    %SP.addr.incr.int.SP_ROT = add %addr %SP.addr.int.SP_ROT, 8
    %SP.addr.incr.ptr.SP_ROT = inttoptr %addr %SP.addr.incr.int.SP_ROT
                                      to %cell.ptr

    %B.int.SP_ROT = load %cell* %SP.addr.incr.ptr.SP_ROT

    ; load %ecx
    %SP.addr.incr.incr.int.SP_ROT = add %addr %SP.addr.incr.int.SP_ROT, 8
    %SP.addr.incr.incr.ptr.SP_ROT = inttoptr %addr %SP.addr.incr.incr.int.SP_ROT
                                      to %cell.ptr

    %C.int.SP_ROT = load %cell* %SP.addr.incr.incr.ptr.SP_ROT

    ; directly store %eax, %ebx, and %ecx in the appropriate pointers
    store %cell %B.int.SP_ROT, %cell* %SP.addr.incr.incr.ptr.SP_ROT ; %(ecx)
    store %cell %A.int.SP_ROT, %cell* %SP.addr.incr.ptr.SP_ROT      ; %(ebx)
    store %cell %C.int.SP_ROT, %cell* %SP.addr.ptr.SP_ROT           ; %(eax)

    br label %kernel.NEXT

x86 assembler output of the 2SWAP and ROT functions -- below, we can see that SP_2SWAP jumps to a subblock of SP_ROT:

LBB13_11:                               ## %kernel.SP_2SWAP
                                        ##   in Loop: Header=BB13_1 Depth=1
    movq    (%rbx), %rax
    movq    16(%rax), %rcx
    movq    24(%rax), %rdx
    movq    (%rax), %rsi
    movq    8(%rax), %rdi
    movq    %rdi, 24(%rax)
    jmp LBB13_18

LBB13_17:                               ## %kernel.SP_ROT
                                        ##   in Loop: Header=BB13_1 Depth=1
    movq    (%rbx), %rax
    movq    16(%rax), %rcx
    movq    (%rax), %rdx
    movq    8(%rax), %rsi
LBB13_18:                               ## %kernel.SP_ROT
                                        ##   in Loop: Header=BB13_1 Depth=1
    movq    %rsi, 16(%rax)
    movq    %rdx, 8(%rax)
    movq    %rcx, (%rax)
    jmp LBB13_1