[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

API modification request



Valette Eric wrote:
> Till Straumann wrote:
> 
>> 3/4) I claimed that table_entry->hdl(table_entry->arg) is only one
>>    extra instruction on x86/PPC. If the hardware gives you just one
>>    pointer, there are one more instructions needed. The worst case
>>    is hardware jumping directly to some vector table entry. In this
>>    case, you'd probably need in the order of 5 instructions.
> 
> 
> I still do not follow the one instruction argument (or is it C 
> instrcutions) at least on PPC/ix86 : pushing table[vector].arg in C is 
> more instructions in ASM. The push on Ix86 is already one and the 
> computation is probably two and may leed to push a register to store 
> intermediate result. On PPC, the computation to get the void* arg is 
> worse

No, it is not - because the pointer to the 'handler' must be computed
anyways and accessing the 'arg' is just one instruction away.

> but you do not need to push the argument (except that r3 may be 
> already used).
> 

OK, I'll just answer this; The current code [written by yourself] (PPC) 
does (disassembled C code):

   lis    r9,  rtems_hdl_tbl at h
   mulli  r12, r30, 24    /* r12 = vector * sizeof(connect_data) */
   lwz    r5,  rtems_hdl_tbl at l(r9)  /* r5 = rtems_hdl_tbl */
   add    r10, r12, r5    /* r10 = &rtems_hdl_tbl[vector] */
   lwz    r8,  8(r10)     /* r8  = rtems_hdl_tbl[vector].hdl */
   /* HERE's MY EXTRA INSTRUCTION (no need to save r3) */
   lwz    r3,  20(r10)    /* r3  = rtems_hdl_tbl[vector].arg */
   /* END OF EXTRA CODE */
   mtctr  r8
   bctrl

Note that the entire C_dispatch_irq_handler() instruction flow
taken normally is ~ 50 instructions.

As for X86: attached is the C code for two versions of a dispatcher
(with vs. without argument). If you compile '-O -fomit-frame-pointer'
and disassemble, you'll see that they differ by 1 instruction. I admit 
that the real lowlevel handler may need to pop the argument after the
user handler returns...

-- Till
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: dummy_dispatch.c
Url: http://rtems.rtems.org/pipermail/rtems-users/attachments/20030203/1620e2f4/attachment.c