[Bug target/43513] New: The stack pointer is adjusted twice

[Bug target/43513] New: The stack pointer is adjusted twice

[carrot at google dot com]

Compile the attached code with options -march=armv7-a -mthumb -Os -fpic, gcc
generates:

foo3:
        push    {r4, r5, r6, r7, lr}
        sub     sp, sp, #12      // A
        ldr     r3, .L4
        add     r7, sp, #0       // B
        sub     sp, sp, #48      // C
        mov     r0, sp
         ...

Instruction A is used to allocate space for spilled variables. Instruction B
setups a frame pointer. Instruction C allocate space for local variables.
Instruction A and C can be merged into one, so ABC can be rewritten as:

        sub     sp, sp, 60
        add     r7, sp, r8

Instruction C is emitted at the expand phase. But instructions AB is emitted
from function arm_expand_prologue in pro_and_epilogue pass. So one possible
solution is stop emitting code to allocate space for local variables. Instead
record the local variables' space size and allocate the space with spill space
together in pro_and_epilogue pass.


-- 
           Summary: The stack pointer is adjusted twice
           Product: gcc
           Version: 4.5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: carrot at google dot com
 GCC build triplet: i686-linux
  GCC host triplet: i686-linux
GCC target triplet: arm-eabi


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43513

[Bug target/43513] The stack pointer is adjusted twice

[carrot at google dot com]

------- Comment #1 from carrot at google dot com  2010-03-25 09:20 -------
Created an attachment (id=20195)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=20195&action=view)
test case


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43513

[Bug target/43513] The stack pointer is adjusted twice

[steven at gcc dot gnu dot org]

------- Comment #2 from steven at gcc dot gnu dot org  2010-03-25 09:53 -------
Any reason why combine-stack-adj.c doesn't perform this optimization for you?


-- 

steven at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2010-03-25 09:53:22
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43513

[Bug target/43513] The stack pointer is adjusted twice

[carrot at google dot com]

------- Comment #3 from carrot at google dot com  2010-03-26 06:42 -------
There are 2 problems with csa for arm:

1. In function rest_of_handle_stack_adjustments:

  if (!ACCUMULATE_OUTGOING_ARGS)
    {
      ...
    }

   ACCUMULATE_OUTGOING_ARGS is defined to 1 in arm.h, so this pass is not
applied to arm back end.

2. Even if I change the condition to enable it, the stack adjustments are still
not combined. 

   Following are related insns:

        ...

(insn/f 117 116 118 2 src/t9.c:5 (set (reg/f:SI 13 sp)
        (plus:SI (reg/f:SI 13 sp)
            (const_int -12 [0xfffffffffffffff4]))) -1 (nil))

(insn 118 117 119 2 src/t9.c:5 (set (mem:BLK (scratch) [0 A8])
        (unspec:BLK [
                (reg/f:SI 13 sp)
                (reg/f:SI 7 r7)
            ] 5)) -1 (nil))

(insn/f 119 118 120 2 src/t9.c:5 (set (reg/f:SI 7 r7)
        (plus:SI (reg/f:SI 13 sp)
            (const_int 0 [0x0]))) -1 (nil))

       ...

(insn 55 50 57 2 src/t9.c:7 (set (reg/f:SI 13 sp)
        (plus:SI (reg/f:SI 13 sp)
            (const_int -48 [0xffffffffffffffd0]))) 4 {*arm_addsi3} (nil))

       ...

    Among them insn 118 trigger the following code in function
combine_stack_adjustments_for_block, and prevent the combine of stack
adjustment codes. It is emitted along with insn 117.

      /* Otherwise, we were not able to process the instruction.
         Do not continue collecting data across such a one.  */
      if (last_sp_set
          && (CALL_P (insn)
              || reg_mentioned_p (stack_pointer_rtx, PATTERN (insn))))
        {
           /* Clear the collected stack references. */
            ...
        } 


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43513

[Bug middle-end/43513] The stack pointer is adjusted twice

[rearnsha at gcc dot gnu dot org]

------- Comment #4 from rearnsha at gcc dot gnu dot org  2010-03-26 09:05 -------
This is a case where we have a 'variable length array' declaration but where
the variable is really const and well-known at compile time.  However, we still
end up with it being 'variably sized' at rtl expand time.

I think it would be much better if the generic optimizers could spot this case
and remove it entirely, turning it into a natural array declartion.  Then we
could eliminate even more from this function (for example, it doesn't really
need a frame pointer).

So changing to 'middle-end'.


-- 

rearnsha at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rearnsha at gcc dot gnu dot
                   |                            |org
           Severity|normal                      |enhancement
          Component|target                      |middle-end


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43513