Re: Incorrect optimized (-O2) linked list code with 4.3.2

[Richard Guenther <richard.guenther@gmail.com>]

On Mon, Sep 12, 2011 at 10:10 AM, pavan tc <pavan.tc@gmail.com> wrote:
> Hi,
>
> I would like to know if there have been issues with optimized linked
> list code with GCC 4.3.2. [optiimization flag : -O2]
>
> The following is the inlined code that has the problem:
>
> static inline void
> list_add_tail (struct list_head *new, struct list_head *head)
> {
>        new->next = head;
>        new->prev = head->prev;
>
>        new->prev->next = new;
>        new->next->prev = new;
> }
>
> The above code has been used in the loop as below:
>
>        pool = GF_CALLOC (count, padded_sizeof_type, gf_common_mt_long);
>        if (!pool) {
>                GF_FREE (mem_pool);
>                return NULL;
>        }
>
>        for (i = 0; i < count; i++) {
>                list = pool + (i * (padded_sizeof_type));
>                INIT_LIST_HEAD (list);
>                list_add_tail (list, &mem_pool->list);
>                ^^^^^^^^^^^^
>        }
>
> '&mem_pool-> list' is used as the list head. mem_pool is a pointer to type :
> struct mem_pool {
>        struct list_head  list;
>        int               hot_count;
>        int               cold_count;
>        gf_lock_t         lock;
>        unsigned long     padded_sizeof_type;
>        void             *pool;
>        void             *pool_end;
>        int               real_sizeof_type;
> };
>
> 'list' is the new member being added to the tail of the list pointed to by head.
> It is a pointer to type:
> struct list_head {
>        struct list_head *next;
>        struct list_head *prev;
> };
>
> The generated assembly for the loop (with the linined list_add_tail())
> is as below:
>
>   40a1c:       e8 0f 03 fd ff         callq  10d30 <__gf_calloc@plt>
>   40a21:       48 85 c0               test   %rax,%rax
>   40a24:       48 89 c7               mov    %rax,%rdi
>   40a27:       0f 84 bf 00 00 00   je     40aec <mem_pool_new_fn+0x14c>
>   40a2d:       48 8b 73 08          mov    0x8(%rbx),%rsi
>   40a31:       4d 8d 44 24 01      lea    0x1(%r12),%r8
>   40a36:       31 c0                   xor    %eax,%eax
>   40a38:       b9 01 00 00 00     mov    $0x1,%ecx
>   40a3d:       0f 1f 00                nopl   (%rax)
>   40a40:       49 0f af c5            imul   %r13,%rax
>       <=== loop start
>   40a44:       48 8d 04 07          lea    (%rdi,%rax,1),%rax
>   40a48:       48 89 18              mov    %rbx,(%rax)
> # list->next = head
>   40a4b:       48 89 06              mov    %rax,(%rsi)
> #  <!!> head->prev->next = list
>   40a4e:       48 8b 10              mov    (%rax),%rdx
> # rdx holds list->next
>   40a51:       48 89 70 08         mov    %rsi,0x8(%rax)          #
> list->prev = head->prev;
>   40a55:       48 89 42 08         mov    %rax,0x8(%rdx)         #
> list->next->prev = list
>   40a59:       48 89 c8              mov    %rcx,%rax
>   40a5c:       48 83 c1 01          add    $0x1,%rcx
>   40a60:       4c 39 c1               cmp    %r8,%rcx
>   40a63:       75 db                   jne    40a40 <mem_pool_new_fn+0xa0>
>
> In the assembly above, %rbx  holds the address of 'head'.
> %rsi holds the value of head->prev. This is assigned outside the loop and the
> compiler classifies it as a loop invariant, which is where, I think,
> the problem is.
> This line of code should have been inside the loop.
> <!!> - %rsi still holds the value of head->prev that was assigned
> outside the loop.
>
> The following experiments eliminate the problem:
>
> 1. Using 'volatile' on the address that 'head' points to.
> 2. Using a function call (logging calls, for example) inside the loop.
> 3. Using the direct libc calloc instead of the GF_CALLOC.
> [GF_CALLOC does some accounting when accounting is enabled. Calls vanilla
> libc calloc() otherwise].
>
> So, anything that necessitates a different usage of %rsi seems to be correcting
> the behaviour.
>
> 4. Using gcc 4.4.3 [ The obvious solution would then be to use 4.4.3,
> but I would
> like to understand if this is a known problem with 4.3.2. Small
> programs written to
> emulate this problem do not exhibit the erroneous behaviour.]
>
> Please let me know if any more details about this behaviour are required.
> I'll be glad to provide them.

Use -fno-strict-aliasing.  Your code invokes undefined behavior.

> TIA,
> Pavan
>