i386 epilogue code tweek
Richard Henderson
rth@cygnus.com
Thu Feb 3 14:52:00 GMT 2000
On Thu, Feb 03, 2000 at 07:13:33PM +0100, Jan Hubicka wrote:
> I've changed epilogue expander in case only one register is popped. Now the
> register is popped by move insn and then esp is restored by movl %ebp, %esp.
> This is shorter than previous code involving lea instructions.
> This allows to use lea for all functions that save maximally one register.
> Now switching to SAVE_REGS_FIRST mode seems to be win overall even on the
> K6.
>
> The patch also attempts to make epilogue expander easier to understand.
Hrm.. I don't think it did a very good job of making the expander
easier to understand. Also, I think the lea used to remove the
frame needs kind of special dependancy help.
What do you think of this version?
r~
* i386.md (epilogue_deallocate_stack): Include an offset; emit
lea for a nonzero offset.
* i386.c (SAVE_REGS_FIRST): Remove.
(ix86_initial_elimination_offset): Handle only SAVE_REGS_FIRST mode.
(ix86_compute_frame_size): Likewise.
(ix86_expand_epilogue): Likewise; use mov instead of pop to restore
a register when profitable, emit leave when profitable.
(ix86_emit_restore_regs): Remove.
(ix86_attr_length_default): Handle epilogue_deallocate_stack
as a TYPE_LEA insn.
Index: i386.c
===================================================================
RCS file: /cvs/gcc/egcs/gcc/config/i386/i386.c,v
retrieving revision 1.132
diff -c -p -d -r1.132 i386.c
*** i386.c 2000/02/02 07:50:20 1.132
--- i386.c 2000/02/03 22:42:10
*************** Boston, MA 02111-1307, USA. */
*** 41,56 ****
#include "basic-block.h"
#include "ggc.h"
- /* True when we want to do pushes before allocating stack to get better
- scheduling.
-
- Saving registers first is win in the most cases except for LEAVE
- instruction. Macro is 0 iff we will use LEAVE. */
-
- #define SAVED_REGS_FIRST \
- (!frame_pointer_needed || (!TARGET_USE_LEAVE && !optimize_size))
-
-
#ifdef EXTRA_CONSTRAINT
/* If EXTRA_CONSTRAINT is defined, then the 'S'
constraint in REG_CLASS_FROM_LETTER will no longer work, and various
--- 41,46 ----
*************** static HOST_WIDE_INT ix86_compute_frame_
*** 411,417 ****
int *, int *, int *));
static int ix86_nsaved_regs PARAMS((void));
static void ix86_emit_save_regs PARAMS((void));
- static void ix86_emit_restore_regs PARAMS((void));
static void ix86_emit_epilogue_esp_adjustment PARAMS((int));
struct ix86_address
--- 401,406 ----
*************** ix86_initial_elimination_offset (from, t
*** 1705,1720 ****
saved frame pointer if frame_pointer_needed
<- HARD_FRAME_POINTER
! [saved regs if SAVED_REGS_FIRST]
[padding1] \
| <- FRAME_POINTER
[frame] > tsize
|
[padding2] /
-
- [saved regs if !SAVED_REGS_FIRST]
- <- STACK_POINTER
*/
if (from == ARG_POINTER_REGNUM && to == HARD_FRAME_POINTER_REGNUM)
--- 1694,1706 ----
saved frame pointer if frame_pointer_needed
<- HARD_FRAME_POINTER
! [saved regs]
[padding1] \
| <- FRAME_POINTER
[frame] > tsize
|
[padding2] /
*/
if (from == ARG_POINTER_REGNUM && to == HARD_FRAME_POINTER_REGNUM)
*************** ix86_initial_elimination_offset (from, t
*** 1725,1732 ****
&& to == HARD_FRAME_POINTER_REGNUM)
{
ix86_compute_frame_size (get_frame_size (), &nregs, &padding1, (int *)0);
! if (SAVED_REGS_FIRST)
! padding1 += nregs * UNITS_PER_WORD;
return -padding1;
}
else
--- 1711,1717 ----
&& to == HARD_FRAME_POINTER_REGNUM)
{
ix86_compute_frame_size (get_frame_size (), &nregs, &padding1, (int *)0);
! padding1 += nregs * UNITS_PER_WORD;
return -padding1;
}
else
*************** ix86_initial_elimination_offset (from, t
*** 1743,1752 ****
return tsize + nregs * UNITS_PER_WORD + frame_size;
else if (from != FRAME_POINTER_REGNUM)
abort ();
- else if (SAVED_REGS_FIRST)
- return tsize - padding1;
else
! return tsize + nregs * UNITS_PER_WORD - padding1;
}
}
--- 1728,1735 ----
return tsize + nregs * UNITS_PER_WORD + frame_size;
else if (from != FRAME_POINTER_REGNUM)
abort ();
else
! return tsize - padding1;
}
}
*************** ix86_compute_frame_size (size, nregs_on_
*** 1788,1800 ****
if (stack_alignment_needed < 4)
stack_alignment_needed = 4;
! if (stack_alignment_needed > preferred_alignment)
abort ();
! if (SAVED_REGS_FIRST)
! offset += nregs * UNITS_PER_WORD;
! else
! total_size += nregs * UNITS_PER_WORD;
total_size += offset;
--- 1771,1780 ----
if (stack_alignment_needed < 4)
stack_alignment_needed = 4;
! if (stack_alignment_needed > PREFERRED_STACK_BOUNDARY / BITS_PER_UNIT)
abort ();
! offset += nregs * UNITS_PER_WORD;
total_size += offset;
*************** ix86_compute_frame_size (size, nregs_on_
*** 1807,1815 ****
}
/* Align stack boundary. */
! if (!current_function_is_leaf)
! padding2 = ((total_size + preferred_alignment - 1)
! & -preferred_alignment) - total_size;
}
#endif
--- 1787,1794 ----
}
/* Align stack boundary. */
! padding2 = ((total_size + preferred_alignment - 1)
! & -preferred_alignment) - total_size;
}
#endif
*************** ix86_expand_prologue ()
*** 1868,1875 ****
RTX_FRAME_RELATED_P (insn) = 1;
}
! if (SAVED_REGS_FIRST)
! ix86_emit_save_regs ();
if (tsize == 0)
;
--- 1847,1853 ----
RTX_FRAME_RELATED_P (insn) = 1;
}
! ix86_emit_save_regs ();
if (tsize == 0)
;
*************** ix86_expand_prologue ()
*** 1903,1911 ****
CALL_INSN_FUNCTION_USAGE (insn));
}
- if (!SAVED_REGS_FIRST)
- ix86_emit_save_regs ();
-
#ifdef SUBTARGET_PROLOGUE
SUBTARGET_PROLOGUE;
#endif
--- 1881,1886 ----
*************** ix86_expand_prologue ()
*** 1920,1944 ****
emit_insn (gen_blockage ());
}
- /* Emit code to pop all registers from stack. */
-
- static void
- ix86_emit_restore_regs ()
- {
- int pic_reg_used = flag_pic && (current_function_uses_pic_offset_table
- || current_function_uses_const_pool);
- int limit = (frame_pointer_needed
- ? HARD_FRAME_POINTER_REGNUM : STACK_POINTER_REGNUM);
- int regno;
-
- for (regno = 0; regno < limit; regno++)
- if ((regs_ever_live[regno] && !call_used_regs[regno])
- || (regno == PIC_OFFSET_TABLE_REGNUM && pic_reg_used))
- {
- emit_insn (gen_popsi1 (gen_rtx_REG (SImode, regno)));
- }
- }
-
/* Emit code to add TSIZE to esp value. Use POP instruction when
profitable. */
--- 1895,1900 ----
*************** ix86_emit_epilogue_esp_adjustment (tsize
*** 1991,2050 ****
void
ix86_expand_epilogue ()
{
- int regno;
int nregs;
! int limit;
int pic_reg_used = flag_pic && (current_function_uses_pic_offset_table
|| current_function_uses_const_pool);
int sp_valid = !frame_pointer_needed || current_function_sp_is_unchanging;
HOST_WIDE_INT offset;
! HOST_WIDE_INT tsize = ix86_compute_frame_size (get_frame_size (), &nregs, (int *)0,
! (int *)0);
- /* SP is often unreliable so we may have to go off the frame pointer. */
! offset = -(tsize + nregs * UNITS_PER_WORD);
! if (SAVED_REGS_FIRST)
{
! if (!sp_valid)
! {
! if (nregs)
! emit_insn (gen_rtx_SET (VOIDmode, stack_pointer_rtx,
! gen_rtx_PLUS (SImode, hard_frame_pointer_rtx,
! GEN_INT (- nregs * UNITS_PER_WORD))));
! else
! emit_insn (gen_epilogue_deallocate_stack (stack_pointer_rtx,
! hard_frame_pointer_rtx));
! }
! else if (tsize)
! ix86_emit_epilogue_esp_adjustment (tsize);
! ix86_emit_restore_regs ();
}
/* If we're only restoring one register and sp is not valid then
using a move instruction to restore the register since it's
! less work than reloading sp and popping the register. Otherwise,
! restore sp (if necessary) and pop the registers. */
!
! else if (nregs > 1 || sp_valid)
{
! if (!sp_valid)
! {
! rtx addr_offset;
! addr_offset = adj_offsettable_operand (AT_BP (QImode), offset);
! addr_offset = XEXP (addr_offset, 0);
!
! emit_insn (gen_rtx_SET (VOIDmode, stack_pointer_rtx, addr_offset));
! }
! ix86_emit_restore_regs ();
! }
! else
! {
! limit = (frame_pointer_needed
! ? HARD_FRAME_POINTER_REGNUM : STACK_POINTER_REGNUM);
! for (regno = 0; regno < limit; regno++)
if ((regs_ever_live[regno] && ! call_used_regs[regno])
|| (regno == PIC_OFFSET_TABLE_REGNUM && pic_reg_used))
{
--- 1947,1982 ----
void
ix86_expand_epilogue ()
{
int nregs;
! int regno;
!
int pic_reg_used = flag_pic && (current_function_uses_pic_offset_table
|| current_function_uses_const_pool);
int sp_valid = !frame_pointer_needed || current_function_sp_is_unchanging;
HOST_WIDE_INT offset;
! HOST_WIDE_INT tsize = ix86_compute_frame_size (get_frame_size (), &nregs,
! (int *)0, (int *)0);
! /* Calculate start of saved registers relative to ebp. */
! offset = -nregs * UNITS_PER_WORD;
! #ifdef FUNCTION_BLOCK_PROFILER_EXIT
! if (profile_block_flag == 2)
{
! FUNCTION_BLOCK_PROFILER_EXIT;
}
+ #endif
/* If we're only restoring one register and sp is not valid then
using a move instruction to restore the register since it's
! less work than reloading sp and popping the register. */
! if (!sp_valid && nregs <= 1)
{
! if (!frame_pointer_needed)
! abort();
! for (regno = 0; regno < HARD_FRAME_POINTER_REGNUM; regno++)
if ((regs_ever_live[regno] && ! call_used_regs[regno])
|| (regno == PIC_OFFSET_TABLE_REGNUM && pic_reg_used))
{
*************** ix86_expand_epilogue ()
*** 2052,2081 ****
adj_offsettable_operand (AT_BP (Pmode), offset));
offset += 4;
}
- }
- if (frame_pointer_needed)
- {
/* If not an i386, mov & pop is faster than "leave". */
if (TARGET_USE_LEAVE || optimize_size)
emit_insn (gen_leave ());
else
{
! if (!SAVED_REGS_FIRST)
! emit_insn (gen_epilogue_deallocate_stack (stack_pointer_rtx,
! hard_frame_pointer_rtx));
emit_insn (gen_popsi1 (hard_frame_pointer_rtx));
}
}
! else if (!SAVED_REGS_FIRST && tsize)
! ix86_emit_epilogue_esp_adjustment (tsize);
!
! #ifdef FUNCTION_BLOCK_PROFILER_EXIT
! if (profile_block_flag == 2)
{
! FUNCTION_BLOCK_PROFILER_EXIT;
}
- #endif
if (current_function_pops_args && current_function_args_size)
{
--- 1984,2021 ----
adj_offsettable_operand (AT_BP (Pmode), offset));
offset += 4;
}
/* If not an i386, mov & pop is faster than "leave". */
if (TARGET_USE_LEAVE || optimize_size)
emit_insn (gen_leave ());
else
{
! emit_insn (gen_epilogue_deallocate_stack (stack_pointer_rtx,
! hard_frame_pointer_rtx,
! const0_rtx));
emit_insn (gen_popsi1 (hard_frame_pointer_rtx));
}
}
! else
{
! /* First step is to deallocate the stack frame so that we can
! pop the registers. */
! if (!sp_valid)
! {
! if (!frame_pointer_needed)
! abort ();
! emit_insn (gen_epilogue_deallocate_stack (stack_pointer_rtx,
! hard_frame_pointer_rtx,
! GEN_INT (offset)));
! }
! else if (tsize)
! ix86_emit_epilogue_esp_adjustment (tsize);
!
! for (regno = 0; regno < STACK_POINTER_REGNUM; regno++)
! if ((regs_ever_live[regno] && !call_used_regs[regno])
! || (regno == PIC_OFFSET_TABLE_REGNUM && pic_reg_used))
! emit_insn (gen_popsi1 (gen_rtx_REG (SImode, regno)));
}
if (current_function_pops_args && current_function_args_size)
{
*************** ix86_attr_length_default (insn)
*** 5853,5866 ****
{
/* Irritatingly, single_set doesn't work with REG_UNUSED present,
as we'll get from running life_analysis during reg-stack when
! not optimizing. */
rtx set = PATTERN (insn);
if (GET_CODE (set) == SET)
;
else if (GET_CODE (set) == PARALLEL
&& XVECLEN (set, 0) == 2
! && GET_CODE (XVECEXP (set, 0, 0)) == SET
! && GET_CODE (XVECEXP (set, 0, 1)) == CLOBBER)
set = XVECEXP (set, 0, 0);
else
abort ();
--- 5793,5807 ----
{
/* Irritatingly, single_set doesn't work with REG_UNUSED present,
as we'll get from running life_analysis during reg-stack when
! not optimizing. Not that it matters anyway, now that
! epilogue_deallocate_stack uses lea, and is by design not
! single_set. */
rtx set = PATTERN (insn);
if (GET_CODE (set) == SET)
;
else if (GET_CODE (set) == PARALLEL
&& XVECLEN (set, 0) == 2
! && GET_CODE (XVECEXP (set, 0, 0)) == SET)
set = XVECEXP (set, 0, 0);
else
abort ();
Index: i386.md
===================================================================
RCS file: /cvs/gcc/egcs/gcc/config/i386/i386.md,v
retrieving revision 1.133
diff -c -p -d -r1.133 i386.md
*** i386.md 2000/02/03 14:10:02 1.133
--- i386.md 2000/02/03 22:42:11
***************
*** 8831,8841 ****
(define_insn "epilogue_deallocate_stack"
[(set (match_operand:SI 0 "register_operand" "=r")
! (match_operand:SI 1 "register_operand" "+r"))
(set (match_dup 1) (match_dup 1))]
""
! "mov{l}\\t{%1, %0|%0, %1}"
! [(set_attr "type" "imov")])
(define_insn "allocate_stack_worker"
[(unspec:SI [(match_operand:SI 0 "register_operand" "a")] 3)
--- 8831,8854 ----
(define_insn "epilogue_deallocate_stack"
[(set (match_operand:SI 0 "register_operand" "=r")
! (plus:SI (match_operand:SI 1 "register_operand" "+r")
! (match_operand:SI 2 "immediate_operand" "i")))
(set (match_dup 1) (match_dup 1))]
""
! "*
! {
! if (get_attr_type (insn) == TYPE_IMOV)
! return \"mov{l}\\t{%1, %0|%0, %1}\";
! else
! {
! operands[2] = SET_SRC (XVECEXP (PATTERN (insn), 0, 0));
! return \"lea{l}\\t{%a2, %0|%0, %a2}\";
! }
! }"
! [(set (attr "type")
! (if_then_else (match_operand:SI 2 "const0_operand" "")
! (const_string "imov")
! (const_string "lea")))])
(define_insn "allocate_stack_worker"
[(unspec:SI [(match_operand:SI 0 "register_operand" "a")] 3)
More information about the Gcc-patches
mailing list