On Thursday 28 June 2007 12:42:57 Joel Reymont wrote:
> Where does the 65% speed-up come from?

Good question.

> Just from using match?

Yes, or you can reorder the branches of the "if" statement, putting the common 
branch first.

My code gives:

camlTest__work_58:
.L101:
        cmpq    $1, %rbx
        je      .L100
        movq    %rbx, %rdi
        addq    $-2, %rdi
        leaq    -1(%rax, %rbx), %rax
        movq    %rdi, %rbx
        jmp     .L101
        .align  4
.L100:
        ret
        .text
        .align  16
        .globl  camlTest__sum_nums3_61
camlTest__sum_nums3_61:
.L102:
        movq    %rax, %rbx
        movq    $1, %rax
        jmp     camlTest__work_58
        .text
        .align  16
        .globl  camlTest__entry

So it branches out of the loop when todo=0 and does one branch per loop.

Both of Thomas' implementations give:

camlTest__work_60:
.L101:
        cmpq    $1, %rbx
        jne     .L100
        ret
        .align  4
.L100:
        movq    %rbx, %rdi
        addq    $-2, %rdi
        leaq    -1(%rax, %rbx), %rax
        movq    %rdi, %rbx
        jmp     .L101
        .text
        .align  16
        .globl  camlTest__sum_nums_58
camlTest__sum_nums_58:
.L102:
        movq    %rax, %rbx
        leaq    camlTest__2(%rip), %rax
        movq    $1, %rax
        jmp     camlTest__work_60
        .text
        .align  16
        .globl  camlTest__entry

which branches within the loop if todo<>0 and then back to the start of the 
loop. So this branches twice per loop.

PS: This has nothing to do with consing or continuations.
-- 
Dr Jon D Harrop, Flying Frog Consultancy Ltd.
The OCaml Journal
http://www.ffconsultancy.com/products/ocaml_journal/?e