This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
patch to improve i386 epilogue (version 6)
- To: egcs-patches at cygnus dot com
- Subject: patch to improve i386 epilogue (version 6)
- From: John Wehle <john at feith dot com>
- Date: Sun, 11 Oct 1998 18:56:22 -0400 (EDT)
- Cc: law at cygnus dot com
[ This version is in principle identical to version 5. It has
been updated to match the current source tree and to include
some suggestions by Jeff. ]
This change allows gcc to skip the loading of the stack pointer
when it is already known to contain the proper value.
ChangeLog:
Sun Oct 11 16:49:15 EDT 1998 John Wehle (john@feith.com)
* flow.c: Update comment.
(notice_stack_pointer_modification): New static function.
(record_volatile_insns): Use it.
(mark_regs_live_at_end): Mark the stack pointer as alive
at the end of the function if current_function_sp_is_unchanging
is set.
(life_analysis_1): Set current_function_sp_is_unchanging.
* function.c: Define it.
(init_function_start): Initialize it.
* output.h: Declare it.
* reorg.c (fill_simple_delay_slots, dbr_schedule): Mark
the stack pointer as alive at the end of the function if
current_function_sp_is_unchanging is set.
* i386.c (ix86_epilogue): Optimize the restoring
of the stack pointer.
Enjoy!
-- John Wehle
------------------8<------------------------8<------------------------
*** gcc/flow.c.ORIGINAL Sun Oct 11 13:56:47 1998
--- gcc/flow.c Sun Oct 11 16:48:22 1998
*************** Boston, MA 02111-1307, USA. */
*** 106,112 ****
life_analysis fills in certain vectors containing information about
register usage: reg_n_refs, reg_n_deaths, reg_n_sets, reg_live_length,
! reg_n_calls_crosses and reg_basic_block. */
#include "config.h"
#include "system.h"
--- 106,115 ----
life_analysis fills in certain vectors containing information about
register usage: reg_n_refs, reg_n_deaths, reg_n_sets, reg_live_length,
! reg_n_calls_crosses and reg_basic_block.
!
! life_analysis sets current_function_sp_is_unchanging if the function
! doesn't modify the stack pointer. */
#include "config.h"
#include "system.h"
*************** noop_move_p (insn)
*** 1221,1229 ****
return 0;
}
/* Record which insns refer to any volatile memory
or for any reason can't be deleted just because they are dead stores.
! Also, delete any insns that copy a register to itself. */
static void
record_volatile_insns (f)
rtx f;
--- 1224,1251 ----
return 0;
}
+ static void
+ notice_stack_pointer_modification (x, pat)
+ rtx x;
+ rtx pat ATTRIBUTE_UNUSED;
+ {
+ if (x == stack_pointer_rtx
+ /* The stack pointer is only modified indirectly as the result
+ of a push until later in flow. See the comments in rtl.texi
+ regarding Embedded Side-Effects on Addresses. */
+ || (GET_CODE (x) == MEM
+ && (GET_CODE (XEXP (x, 0)) == PRE_DEC
+ || GET_CODE (XEXP (x, 0)) == PRE_INC
+ || GET_CODE (XEXP (x, 0)) == POST_DEC
+ || GET_CODE (XEXP (x, 0)) == POST_INC)
+ && XEXP (XEXP (x, 0), 0) == stack_pointer_rtx))
+ current_function_sp_is_unchanging = 0;
+ }
+
/* Record which insns refer to any volatile memory
or for any reason can't be deleted just because they are dead stores.
! Also, delete any insns that copy a register to itself.
! And see if the stack pointer is modified. */
static void
record_volatile_insns (f)
rtx f;
*************** record_volatile_insns (f)
*** 1264,1269 ****
--- 1286,1296 ----
NOTE_SOURCE_FILE (insn) = 0;
}
}
+
+ /* Check if insn modifies the stack pointer. */
+ if ( current_function_sp_is_unchanging
+ && GET_RTX_CLASS (GET_CODE (insn)) == 'i')
+ note_stores (PATTERN (insn), notice_stack_pointer_modification);
}
}
*************** mark_regs_live_at_end (set)
*** 1279,1285 ****
if (! EXIT_IGNORE_STACK
|| (! FRAME_POINTER_REQUIRED
&& ! current_function_calls_alloca
! && flag_omit_frame_pointer))
#endif
/* If exiting needs the right stack value,
consider the stack pointer live at the end of the function. */
--- 1306,1313 ----
if (! EXIT_IGNORE_STACK
|| (! FRAME_POINTER_REQUIRED
&& ! current_function_calls_alloca
! && flag_omit_frame_pointer)
! || current_function_sp_is_unchanging)
#endif
/* If exiting needs the right stack value,
consider the stack pointer live at the end of the function. */
*************** life_analysis_1 (f, nregs)
*** 1375,1380 ****
--- 1403,1413 ----
basic_block_significant
= (regset *) alloca (n_basic_blocks * sizeof (regset));
init_regset_vector (basic_block_significant, n_basic_blocks, &flow_obstack);
+
+ /* Assume that the stack pointer is unchanging if alloca hasn't been used.
+ This will be cleared by record_volatile_insns if it encounters an insn
+ which modifies the stack pointer. */
+ current_function_sp_is_unchanging = !current_function_calls_alloca;
record_volatile_insns (f);
*** gcc/function.c.ORIGINAL Sun Oct 11 13:56:47 1998
--- gcc/function.c Sun Oct 11 15:50:08 1998
*************** int current_function_has_computed_jump;
*** 138,143 ****
--- 138,149 ----
int current_function_contains_functions;
+ /* Nonzero if function being compiled doesn't modify the stack pointer
+ (ignoring the prologue and epilogue). This is only valid after
+ life_analysis has run. */
+
+ int current_function_sp_is_unchanging;
+
/* Nonzero if the current function is a thunk (a lightweight function that
just adjusts one of its arguments and forwards to another function), so
we should try to cut corners where we can. */
*************** init_function_start (subr, filename, lin
*** 5426,5431 ****
--- 5432,5438 ----
current_function_has_nonlocal_label = 0;
current_function_has_nonlocal_goto = 0;
current_function_contains_functions = 0;
+ current_function_sp_is_unchanging = 0;
current_function_is_thunk = 0;
current_function_returns_pcc_struct = 0;
*** gcc/output.h.ORIGINAL Sun Oct 11 13:56:48 1998
--- gcc/output.h Sun Oct 11 15:49:58 1998
*************** extern int current_function_has_nonlocal
*** 380,385 ****
--- 380,391 ----
extern int current_function_contains_functions;
+ /* Nonzero if function being compiled doesn't modify the stack pointer
+ (ignoring the prologue and epilogue). This is only valid after
+ life_analysis has run. */
+
+ extern int current_function_sp_is_unchanging;
+
/* Nonzero if the current function returns a pointer type */
extern int current_function_returns_pointer;
*** gcc/reorg.c.ORIGINAL Fri Oct 9 22:41:36 1998
--- gcc/reorg.c Sun Oct 11 15:56:56 1998
*************** fill_simple_delay_slots (non_jumps_p)
*** 3452,3458 ****
SET_HARD_REG_BIT (needed.regs, HARD_FRAME_POINTER_REGNUM);
#endif
#ifdef EXIT_IGNORE_STACK
! if (! EXIT_IGNORE_STACK)
#endif
SET_HARD_REG_BIT (needed.regs, STACK_POINTER_REGNUM);
}
--- 3452,3459 ----
SET_HARD_REG_BIT (needed.regs, HARD_FRAME_POINTER_REGNUM);
#endif
#ifdef EXIT_IGNORE_STACK
! if (! EXIT_IGNORE_STACK
! || current_function_sp_is_unchanging)
#endif
SET_HARD_REG_BIT (needed.regs, STACK_POINTER_REGNUM);
}
*************** dbr_schedule (first, file)
*** 4602,4608 ****
SET_HARD_REG_BIT (end_of_function_needs.regs, HARD_FRAME_POINTER_REGNUM);
#endif
#ifdef EXIT_IGNORE_STACK
! if (! EXIT_IGNORE_STACK)
#endif
SET_HARD_REG_BIT (end_of_function_needs.regs, STACK_POINTER_REGNUM);
}
--- 4603,4610 ----
SET_HARD_REG_BIT (end_of_function_needs.regs, HARD_FRAME_POINTER_REGNUM);
#endif
#ifdef EXIT_IGNORE_STACK
! if (! EXIT_IGNORE_STACK
! || current_function_sp_is_unchanging)
#endif
SET_HARD_REG_BIT (end_of_function_needs.regs, STACK_POINTER_REGNUM);
}
*** gcc/config/i386/i386.c.ORIGINAL Fri Oct 9 22:42:20 1998
--- gcc/config/i386/i386.c Sun Oct 11 15:49:22 1998
*************** ix86_epilogue (do_rtl)
*** 2294,2299 ****
--- 2294,2300 ----
rtx xops[3];
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;
long tsize = get_frame_size ();
/* Compute the number of registers to pop */
*************** ix86_epilogue (do_rtl)
*** 2307,2318 ****
|| (regno == PIC_OFFSET_TABLE_REGNUM && pic_reg_used))
nregs++;
! /* sp is often unreliable so we must go off the frame pointer.
!
! In reality, we may not care if sp is unreliable, because we can restore
! the register relative to the frame pointer. In theory, since each move
! is the same speed as a pop, and we don't need the leal, this is faster.
! For now restore multiple registers the old way. */
offset = - tsize - (nregs * UNITS_PER_WORD);
--- 2308,2314 ----
|| (regno == PIC_OFFSET_TABLE_REGNUM && pic_reg_used))
nregs++;
! /* sp is often unreliable so we may have to go off the frame pointer. */
offset = - tsize - (nregs * UNITS_PER_WORD);
*************** ix86_epilogue (do_rtl)
*** 2329,2337 ****
if (flag_pic || profile_flag || profile_block_flag)
emit_insn (gen_blockage ());
! if (nregs > 1 || ! frame_pointer_needed)
{
! if (frame_pointer_needed)
{
xops[0] = adj_offsettable_operand (AT_BP (QImode), offset);
if (do_rtl)
--- 2325,2338 ----
if (flag_pic || profile_flag || profile_block_flag)
emit_insn (gen_blockage ());
! /* 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. */
!
! if (nregs > 1 || sp_valid)
{
! if ( !sp_valid )
{
xops[0] = adj_offsettable_operand (AT_BP (QImode), offset);
if (do_rtl)
-------------------------------------------------------------------------
| Feith Systems | Voice: 1-215-646-8000 | Email: john@feith.com |
| John Wehle | Fax: 1-215-540-5495 | |
-------------------------------------------------------------------------