/* Register to Stack convert for GNU compiler.
Copyright (C) 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
- 2000, 2001, 2002, 2003, 2004 Free Software Foundation, Inc.
+ 2000, 2001, 2002, 2003, 2004, 2005 Free Software Foundation, Inc.
This file is part of GCC.
You should have received a copy of the GNU General Public License
along with GCC; see the file COPYING. If not, write to the Free
- Software Foundation, 59 Temple Place - Suite 330, Boston, MA
- 02111-1307, USA. */
+ Software Foundation, 51 Franklin Street, Fifth Floor, Boston, MA
+ 02110-1301, USA. */
/* This pass converts stack-like registers from the "flat register
file" model that gcc uses, to a stack convention that the 387 uses.
#include "varray.h"
#include "reload.h"
#include "ggc.h"
+#include "timevar.h"
+#include "tree-pass.h"
+#include "target.h"
+#include "vecprim.h"
+
+#ifdef STACK_REGS
/* We use this array to cache info about insns, because otherwise we
spend too much time in stack_regs_mentioned_p.
Indexed by insn UIDs. A value of zero is uninitialized, one indicates
the insn uses stack registers, two indicates the insn does not use
stack registers. */
-static GTY(()) varray_type stack_regs_mentioned_data;
-
-#ifdef STACK_REGS
+static VEC(char,heap) *stack_regs_mentioned_data;
#define REG_STACK_SIZE (LAST_STACK_REG - FIRST_STACK_REG + 1)
+int regstack_completed = 0;
+
/* This is the basic stack record. TOP is an index into REG[] such
that REG[TOP] is the top of stack. If TOP is -1 the stack is empty.
struct stack_def stack_out; /* Output stack configuration. */
HARD_REG_SET out_reg_set; /* Stack regs live on output. */
int done; /* True if block already converted. */
- int predecessors; /* Number of predecessors that needs
+ int predecessors; /* Number of predecessors that need
to be visited. */
} *block_info;
/* The block we're currently working on. */
static basic_block current_block;
+/* In the current_block, whether we're processing the first register
+ stack or call instruction, i.e. the regstack is currently the
+ same as BLOCK_INFO(current_block)->stack_in. */
+static bool starting_stack_p;
+
/* This is the register file for all register after conversion. */
static rtx
FP_mode_reg[LAST_STACK_REG+1-FIRST_STACK_REG][(int) MAX_MACHINE_MODE];
/* Forward declarations */
static int stack_regs_mentioned_p (rtx pat);
-static void straighten_stack (rtx, stack);
static void pop_stack (stack, int);
static rtx *get_true_reg (rtx *);
static void remove_regno_note (rtx, enum reg_note, unsigned int);
static int get_hard_regnum (stack, rtx);
static rtx emit_pop_insn (rtx, stack, rtx, enum emit_where);
-static void emit_swap_insn (rtx, stack, rtx);
static void swap_to_top(rtx, stack, rtx, rtx);
static bool move_for_stack_reg (rtx, stack, rtx);
static bool move_nan_for_stack_reg (rtx, stack, rtx);
static void subst_asm_stack_regs (rtx, stack);
static bool subst_stack_regs (rtx, stack);
static void change_stack (rtx, stack, stack, enum emit_where);
-static int convert_regs_entry (void);
-static void convert_regs_exit (void);
-static int convert_regs_1 (FILE *, basic_block);
-static int convert_regs_2 (FILE *, basic_block);
-static int convert_regs (FILE *);
static void print_stack (FILE *, stack);
static rtx next_flags_user (rtx);
-static bool compensate_edge (edge, FILE *);
\f
/* Return nonzero if any stack register is mentioned somewhere within PAT. */
return 0;
uid = INSN_UID (insn);
- max = VARRAY_SIZE (stack_regs_mentioned_data);
+ max = VEC_length (char, stack_regs_mentioned_data);
if (uid >= max)
{
+ char *p;
+ unsigned int old_max = max;
+
/* Allocate some extra size to avoid too many reallocs, but
do not grow too quickly. */
- max = uid + uid / 20;
- VARRAY_GROW (stack_regs_mentioned_data, max);
+ max = uid + uid / 20 + 1;
+ VEC_safe_grow (char, heap, stack_regs_mentioned_data, max);
+ p = VEC_address (char, stack_regs_mentioned_data);
+ memset (&p[old_max], 0,
+ sizeof (char) * (max - old_max));
}
- test = VARRAY_CHAR (stack_regs_mentioned_data, uid);
+ test = VEC_index (char, stack_regs_mentioned_data, uid);
if (test == 0)
{
/* This insn has yet to be examined. Do so now. */
test = stack_regs_mentioned_p (PATTERN (insn)) ? 1 : 2;
- VARRAY_CHAR (stack_regs_mentioned_data, uid) = test;
+ VEC_replace (char, stack_regs_mentioned_data, uid, test);
}
return test == 1;
return NULL_RTX;
}
\f
-/* Reorganize the stack into ascending numbers,
- after this insn. */
+/* Reorganize the stack into ascending numbers, before this insn. */
static void
straighten_stack (rtx insn, stack regstack)
for (top = temp_stack.top = regstack->top; top >= 0; top--)
temp_stack.reg[top] = FIRST_STACK_REG + temp_stack.top - top;
- change_stack (insn, regstack, &temp_stack, EMIT_AFTER);
+ change_stack (insn, regstack, &temp_stack, EMIT_BEFORE);
}
/* Pop a register from the stack. */
}
}
\f
-/* Convert register usage from "flat" register file usage to a "stack
- register file. FILE is the dump file, if used.
-
- Construct a CFG and run life analysis. Then convert each insn one
- by one. Run a last cleanup_cfg pass, if optimizing, to eliminate
- code duplication created when the converter inserts pop insns on
- the edges. */
-
-bool
-reg_to_stack (FILE *file)
-{
- basic_block bb;
- int i;
- int max_uid;
-
- /* Clean up previous run. */
- stack_regs_mentioned_data = 0;
-
- /* See if there is something to do. Flow analysis is quite
- expensive so we might save some compilation time. */
- for (i = FIRST_STACK_REG; i <= LAST_STACK_REG; i++)
- if (regs_ever_live[i])
- break;
- if (i > LAST_STACK_REG)
- return false;
-
- /* Ok, floating point instructions exist. If not optimizing,
- build the CFG and run life analysis.
- Also need to rebuild life when superblock scheduling is done
- as it don't update liveness yet. */
- if (!optimize
- || (flag_sched2_use_superblocks
- && flag_schedule_insns_after_reload))
- {
- count_or_remove_death_notes (NULL, 1);
- life_analysis (file, PROP_DEATH_NOTES);
- }
- mark_dfs_back_edges ();
-
- /* Set up block info for each basic block. */
- alloc_aux_for_blocks (sizeof (struct block_info_def));
- FOR_EACH_BB_REVERSE (bb)
- {
- edge e;
- edge_iterator ei;
-
- FOR_EACH_EDGE (e, ei, bb->preds)
- if (!(e->flags & EDGE_DFS_BACK)
- && e->src != ENTRY_BLOCK_PTR)
- BLOCK_INFO (bb)->predecessors++;
- }
-
- /* Create the replacement registers up front. */
- for (i = FIRST_STACK_REG; i <= LAST_STACK_REG; i++)
- {
- enum machine_mode mode;
- for (mode = GET_CLASS_NARROWEST_MODE (MODE_FLOAT);
- mode != VOIDmode;
- mode = GET_MODE_WIDER_MODE (mode))
- FP_MODE_REG (i, mode) = gen_rtx_REG (mode, i);
- for (mode = GET_CLASS_NARROWEST_MODE (MODE_COMPLEX_FLOAT);
- mode != VOIDmode;
- mode = GET_MODE_WIDER_MODE (mode))
- FP_MODE_REG (i, mode) = gen_rtx_REG (mode, i);
- }
-
- ix86_flags_rtx = gen_rtx_REG (CCmode, FLAGS_REG);
-
- /* A QNaN for initializing uninitialized variables.
-
- ??? We can't load from constant memory in PIC mode, because
- we're inserting these instructions before the prologue and
- the PIC register hasn't been set up. In that case, fall back
- on zero, which we can get from `ldz'. */
-
- if (flag_pic)
- not_a_num = CONST0_RTX (SFmode);
- else
- {
- not_a_num = gen_lowpart (SFmode, GEN_INT (0x7fc00000));
- not_a_num = force_const_mem (SFmode, not_a_num);
- }
-
- /* Allocate a cache for stack_regs_mentioned. */
- max_uid = get_max_uid ();
- VARRAY_CHAR_INIT (stack_regs_mentioned_data, max_uid + 1,
- "stack_regs_mentioned cache");
-
- convert_regs (file);
-
- free_aux_for_blocks ();
- return true;
-}
-
-\f
/* Return a pointer to the REG expression within PAT. If PAT is not a
REG, possible enclosed by a conversion rtx, return the inner part of
PAT that stopped the search. */
result = DECL_RTL_IF_SET (DECL_RESULT (decl));
if (result != 0)
- {
-#ifdef FUNCTION_OUTGOING_VALUE
- result
- = FUNCTION_OUTGOING_VALUE (TREE_TYPE (DECL_RESULT (decl)), decl);
-#else
- result = FUNCTION_VALUE (TREE_TYPE (DECL_RESULT (decl)), decl);
-#endif
- }
+ result = targetm.calls.function_value (TREE_TYPE (DECL_RESULT (decl)),
+ decl, true);
return result != 0 && STACK_REG_P (result) ? result : 0;
}
gcc_assert (regno <= LAST_STACK_REG);
gcc_assert (STACK_REG_P (*reg));
- gcc_assert (GET_MODE_CLASS (GET_MODE (*reg)) == MODE_FLOAT
+ gcc_assert (SCALAR_FLOAT_MODE_P (GET_MODE (*reg))
|| GET_MODE_CLASS (GET_MODE (*reg)) == MODE_COMPLEX_FLOAT);
*reg = FP_MODE_REG (regno, GET_MODE (*reg));
if (LABEL_P (tmp)
|| CALL_P (tmp)
|| NOTE_INSN_BASIC_BLOCK_P (tmp)
- || (NOTE_P (tmp)
- && NOTE_LINE_NUMBER (tmp) == NOTE_INSN_UNLIKELY_EXECUTED_CODE)
|| (NONJUMP_INSN_P (tmp)
&& stack_regs_mentioned (tmp)))
{
return;
}
+ /* Avoid emitting the swap if this is the first register stack insn
+ of the current_block. Instead update the current_block's stack_in
+ and let compensate edges take care of this for us. */
+ if (current_block && starting_stack_p)
+ {
+ BLOCK_INFO (current_block)->stack_in = *regstack;
+ starting_stack_p = false;
+ return;
+ }
+
swap_rtx = gen_swapxf (FP_MODE_REG (hard_regno, XFmode),
FP_MODE_REG (FIRST_STACK_REG, XFmode));
available. Push the source value here if the register
stack is not full, and then write the value to memory via
a pop. */
- rtx push_rtx, push_insn;
+ rtx push_rtx;
rtx top_stack_reg = FP_MODE_REG (FIRST_STACK_REG, GET_MODE (src));
push_rtx = gen_movxf (top_stack_reg, top_stack_reg);
- push_insn = emit_insn_before (push_rtx, insn);
+ emit_insn_before (push_rtx, insn);
REG_NOTES (insn) = gen_rtx_EXPR_LIST (REG_DEAD, top_stack_reg,
REG_NOTES (insn));
}
pat = PATTERN (insn);
}
- /* See if this is, or ends in, a fnstsw, aka unspec 9. If so, we're
- not doing anything with the cc value right now. We may be able to
- search for one though. */
+ /* See if this is, or ends in, a fnstsw. If so, we're not doing anything
+ with the cc value right now. We may be able to search for one
+ though. */
if (GET_CODE (pat) == SET
&& GET_CODE (SET_SRC (pat)) == UNSPEC
return 0;
}
+ /* We haven't found it. */
+ if (insn == BB_END (current_block))
+ return 0;
+
/* So we've found the insn using this value. If it is anything
- other than sahf, aka unspec 10, or the value does not die
- (meaning we'd have to search further), then we must give up. */
+ other than sahf or the value does not die (meaning we'd have
+ to search further), then we must give up. */
pat = PATTERN (insn);
if (GET_CODE (pat) != SET
|| GET_CODE (SET_SRC (pat)) != UNSPEC
{
rtx *src1, *src2;
rtx src1_note, src2_note;
- rtx flags_user;
src1 = get_true_reg (&XEXP (pat_src, 0));
src2 = get_true_reg (&XEXP (pat_src, 1));
- flags_user = next_flags_user (insn);
/* ??? If fxch turns out to be cheaper than fstp, give priority to
registers that die in this insn - move those to stack top first. */
if (pat != PATTERN (insn))
{
/* The fix_truncdi_1 pattern wants to be able to allocate
- it's own scratch register. It does this by clobbering
+ its own scratch register. It does this by clobbering
an fp reg so that it is assured of an empty reg-stack
register. If the register is live, kill it now.
Remove the DEAD/UNUSED note so we don't try to kill it
if (!note)
{
rtx t = *dest;
- if (get_hard_regnum (regstack, t) == -1)
- control_flow_insn_deleted
- |= move_nan_for_stack_reg (insn, regstack, t);
if (COMPLEX_MODE_P (GET_MODE (t)))
{
- t = FP_MODE_REG (REGNO (t) + 1, DFmode);
- if (get_hard_regnum (regstack, t) == -1)
- control_flow_insn_deleted
- |= move_nan_for_stack_reg (insn, regstack, t);
+ rtx u = FP_MODE_REG (REGNO (t) + 1, SFmode);
+ if (get_hard_regnum (regstack, u) == -1)
+ {
+ rtx pat2 = gen_rtx_CLOBBER (VOIDmode, u);
+ rtx insn2 = emit_insn_before (pat2, insn);
+ control_flow_insn_deleted
+ |= move_nan_for_stack_reg (insn2, regstack, u);
+ }
}
+ if (get_hard_regnum (regstack, t) == -1)
+ control_flow_insn_deleted
+ |= move_nan_for_stack_reg (insn, regstack, t);
}
}
}
case UNSPEC:
switch (XINT (pat_src, 1))
{
+ case UNSPEC_FIST:
+
+ case UNSPEC_FIST_FLOOR:
+ case UNSPEC_FIST_CEIL:
+
+ /* These insns only operate on the top of the stack. */
+
+ src1 = get_true_reg (&XVECEXP (pat_src, 0, 0));
+ emit_swap_insn (insn, regstack, *src1);
+
+ src1_note = find_regno_note (insn, REG_DEAD, REGNO (*src1));
+
+ if (STACK_REG_P (*dest))
+ replace_reg (dest, FIRST_STACK_REG);
+
+ if (src1_note)
+ {
+ replace_reg (&XEXP (src1_note, 0), FIRST_STACK_REG);
+ regstack->top--;
+ CLEAR_HARD_REG_BIT (regstack->reg_set, REGNO (*src1));
+ }
+
+ replace_reg (src1, FIRST_STACK_REG);
+ break;
+
case UNSPEC_SIN:
case UNSPEC_COS:
case UNSPEC_FRNDINT:
if (top >= 0)
{
- straighten_stack (PREV_INSN (insn), regstack);
+ straighten_stack (insn, regstack);
/* Now mark the arguments as dead after the call. */
if (NOTE_P (insn) || INSN_DELETED_P (insn))
return control_flow_insn_deleted;
+ /* If this a noreturn call, we can't insert pop insns after it.
+ Instead, reset the stack state to empty. */
+ if (CALL_P (insn)
+ && find_reg_note (insn, REG_NORETURN, NULL))
+ {
+ regstack->top = -1;
+ CLEAR_HARD_REG_SET (regstack->reg_set);
+ return control_flow_insn_deleted;
+ }
+
/* If there is a REG_UNUSED note on a stack register on this insn,
the indicated reg must be popped. The REG_UNUSED note is removed,
since the form of the newly emitted pop insn references the reg,
int reg;
int update_end = 0;
+ /* Stack adjustments for the first insn in a block update the
+ current_block's stack_in instead of inserting insns directly.
+ compensate_edges will add the necessary code later. */
+ if (current_block
+ && starting_stack_p
+ && where == EMIT_BEFORE)
+ {
+ BLOCK_INFO (current_block)->stack_in = *new;
+ starting_stack_p = false;
+ *old = *new;
+ return;
+ }
+
/* We will be inserting new insns "backwards". If we are to insert
after INSN, find the next insn, and insert before it. */
int inserted = 0;
edge e;
edge_iterator ei;
- basic_block block;
-
- FOR_EACH_BB_REVERSE (block)
- {
- block_info bi = BLOCK_INFO (block);
- int reg;
-
- /* Set current register status at last instruction `uninitialized'. */
- bi->stack_in.top = -2;
-
- /* Copy live_at_end and live_at_start into temporaries. */
- for (reg = FIRST_STACK_REG; reg <= LAST_STACK_REG; reg++)
- {
- if (REGNO_REG_SET_P (block->global_live_at_end, reg))
- SET_HARD_REG_BIT (bi->out_reg_set, reg);
- if (REGNO_REG_SET_P (block->global_live_at_start, reg))
- SET_HARD_REG_BIT (bi->stack_in.reg_set, reg);
- }
- }
/* Load something into each stack register live at function entry.
Such live registers can be caused by uninitialized variables or
}
}
-/* Adjust the stack of this block on exit to match the stack of the
- target block, or copy stack info into the stack of the successor
- of the successor hasn't been processed yet. */
+/* Copy the stack info from the end of edge E's source block to the
+ start of E's destination block. */
+
+static void
+propagate_stack (edge e)
+{
+ stack src_stack = &BLOCK_INFO (e->src)->stack_out;
+ stack dest_stack = &BLOCK_INFO (e->dest)->stack_in;
+ int reg;
+
+ /* Preserve the order of the original stack, but check whether
+ any pops are needed. */
+ dest_stack->top = -1;
+ for (reg = 0; reg <= src_stack->top; ++reg)
+ if (TEST_HARD_REG_BIT (dest_stack->reg_set, src_stack->reg[reg]))
+ dest_stack->reg[++dest_stack->top] = src_stack->reg[reg];
+}
+
+
+/* Adjust the stack of edge E's source block on exit to match the stack
+ of it's target block upon input. The stack layouts of both blocks
+ should have been defined by now. */
+
static bool
-compensate_edge (edge e, FILE *file)
+compensate_edge (edge e)
{
- basic_block block = e->src, target = e->dest;
- block_info bi = BLOCK_INFO (block);
- struct stack_def regstack, tmpstack;
+ basic_block source = e->src, target = e->dest;
stack target_stack = &BLOCK_INFO (target)->stack_in;
+ stack source_stack = &BLOCK_INFO (source)->stack_out;
+ struct stack_def regstack;
int reg;
- current_block = block;
- regstack = bi->stack_out;
- if (file)
- fprintf (file, "Edge %d->%d: ", block->index, target->index);
+ if (dump_file)
+ fprintf (dump_file, "Edge %d->%d: ", source->index, target->index);
+
+ gcc_assert (target_stack->top != -2);
- if (target_stack->top == -2)
+ /* Check whether stacks are identical. */
+ if (target_stack->top == source_stack->top)
{
- /* The target block hasn't had a stack order selected.
- We need merely ensure that no pops are needed. */
- for (reg = regstack.top; reg >= 0; --reg)
- if (!TEST_HARD_REG_BIT (target_stack->reg_set, regstack.reg[reg]))
+ for (reg = target_stack->top; reg >= 0; --reg)
+ if (target_stack->reg[reg] != source_stack->reg[reg])
break;
if (reg == -1)
{
- if (file)
- fprintf (file, "new block; copying stack position\n");
-
- /* change_stack kills values in regstack. */
- tmpstack = regstack;
-
- change_stack (BB_END (block), &tmpstack, target_stack, EMIT_AFTER);
+ if (dump_file)
+ fprintf (dump_file, "no changes needed\n");
return false;
}
-
- if (file)
- fprintf (file, "new block; pops needed\n");
}
- else
- {
- if (target_stack->top == regstack.top)
- {
- for (reg = target_stack->top; reg >= 0; --reg)
- if (target_stack->reg[reg] != regstack.reg[reg])
- break;
-
- if (reg == -1)
- {
- if (file)
- fprintf (file, "no changes needed\n");
- return false;
- }
- }
- if (file)
- {
- fprintf (file, "correcting stack to ");
- print_stack (file, target_stack);
- }
+ if (dump_file)
+ {
+ fprintf (dump_file, "correcting stack to ");
+ print_stack (dump_file, target_stack);
}
- /* Care for non-call EH edges specially. The normal return path have
- values in registers. These will be popped en masse by the unwind
- library. */
- if ((e->flags & (EDGE_EH | EDGE_ABNORMAL_CALL)) == EDGE_EH)
- target_stack->top = -1;
-
- /* Other calls may appear to have values live in st(0), but the
+ /* Abnormal calls may appear to have values live in st(0), but the
abnormal return path will not have actually loaded the values. */
- else if (e->flags & EDGE_ABNORMAL_CALL)
+ if (e->flags & EDGE_ABNORMAL_CALL)
{
/* Assert that the lifetimes are as we expect -- one value
live at st(0) on the end of the source block, and no
- values live at the beginning of the destination block. */
- HARD_REG_SET tmp;
-
- CLEAR_HARD_REG_SET (tmp);
- GO_IF_HARD_REG_EQUAL (target_stack->reg_set, tmp, eh1);
- gcc_unreachable ();
- eh1:
-
- /* We are sure that there is st(0) live, otherwise we won't compensate.
+ values live at the beginning of the destination block.
For complex return values, we may have st(1) live as well. */
- SET_HARD_REG_BIT (tmp, FIRST_STACK_REG);
- if (TEST_HARD_REG_BIT (regstack.reg_set, FIRST_STACK_REG + 1))
- SET_HARD_REG_BIT (tmp, FIRST_STACK_REG + 1);
- GO_IF_HARD_REG_EQUAL (regstack.reg_set, tmp, eh2);
- gcc_unreachable ();
- eh2:
+ gcc_assert (source_stack->top == 0 || source_stack->top == 1);
+ gcc_assert (target_stack->top == -1);
+ return false;
+ }
- target_stack->top = -1;
+ /* Handle non-call EH edges specially. The normal return path have
+ values in registers. These will be popped en masse by the unwind
+ library. */
+ if (e->flags & EDGE_EH)
+ {
+ gcc_assert (target_stack->top == -1);
+ return false;
}
+ /* We don't support abnormal edges. Global takes care to
+ avoid any live register across them, so we should never
+ have to insert instructions on such edges. */
+ gcc_assert (! (e->flags & EDGE_ABNORMAL));
+
+ /* Make a copy of source_stack as change_stack is destructive. */
+ regstack = *source_stack;
+
/* It is better to output directly to the end of the block
instead of to the edge, because emit_swap can do minimal
insn scheduling. We can do this when there is only one
edge out, and it is not abnormal. */
- else if (EDGE_COUNT (block->succs) == 1 && !(e->flags & EDGE_ABNORMAL))
+ if (EDGE_COUNT (source->succs) == 1)
{
- /* change_stack kills values in regstack. */
- tmpstack = regstack;
-
- change_stack (BB_END (block), &tmpstack, target_stack,
- (JUMP_P (BB_END (block))
- ? EMIT_BEFORE : EMIT_AFTER));
+ current_block = source;
+ change_stack (BB_END (source), ®stack, target_stack,
+ (JUMP_P (BB_END (source)) ? EMIT_BEFORE : EMIT_AFTER));
}
else
{
rtx seq, after;
- /* We don't support abnormal edges. Global takes care to
- avoid any live register across them, so we should never
- have to insert instructions on such edges. */
- gcc_assert (!(e->flags & EDGE_ABNORMAL));
-
current_block = NULL;
start_sequence ();
/* ??? change_stack needs some point to emit insns after. */
after = emit_note (NOTE_INSN_DELETED);
- tmpstack = regstack;
- change_stack (after, &tmpstack, target_stack, EMIT_BEFORE);
+ change_stack (after, ®stack, target_stack, EMIT_BEFORE);
seq = get_insns ();
end_sequence ();
return false;
}
+/* Traverse all non-entry edges in the CFG, and emit the necessary
+ edge compensation code to change the stack from stack_out of the
+ source block to the stack_in of the destination block. */
+
+static bool
+compensate_edges (void)
+{
+ bool inserted = false;
+ basic_block bb;
+
+ starting_stack_p = false;
+
+ FOR_EACH_BB (bb)
+ if (bb != ENTRY_BLOCK_PTR)
+ {
+ edge e;
+ edge_iterator ei;
+
+ FOR_EACH_EDGE (e, ei, bb->succs)
+ inserted |= compensate_edge (e);
+ }
+ return inserted;
+}
+
+/* Select the better of two edges E1 and E2 to use to determine the
+ stack layout for their shared destination basic block. This is
+ typically the more frequently executed. The edge E1 may be NULL
+ (in which case E2 is returned), but E2 is always non-NULL. */
+
+static edge
+better_edge (edge e1, edge e2)
+{
+ if (!e1)
+ return e2;
+
+ if (EDGE_FREQUENCY (e1) > EDGE_FREQUENCY (e2))
+ return e1;
+ if (EDGE_FREQUENCY (e1) < EDGE_FREQUENCY (e2))
+ return e2;
+
+ if (e1->count > e2->count)
+ return e1;
+ if (e1->count < e2->count)
+ return e2;
+
+ /* Prefer critical edges to minimize inserting compensation code on
+ critical edges. */
+
+ if (EDGE_CRITICAL_P (e1) != EDGE_CRITICAL_P (e2))
+ return EDGE_CRITICAL_P (e1) ? e1 : e2;
+
+ /* Avoid non-deterministic behavior. */
+ return (e1->src->index < e2->src->index) ? e1 : e2;
+}
+
/* Convert stack register references in one block. */
-static int
-convert_regs_1 (FILE *file, basic_block block)
+static void
+convert_regs_1 (basic_block block)
{
struct stack_def regstack;
block_info bi = BLOCK_INFO (block);
- int deleted, inserted, reg;
+ int reg;
rtx insn, next;
- edge e, beste = NULL;
bool control_flow_insn_deleted = false;
- edge_iterator ei;
- inserted = 0;
- deleted = 0;
any_malformed_asm = false;
- /* Find the edge we will copy stack from. It should be the most frequent
- one as it will get cheapest after compensation code is generated,
- if multiple such exists, take one with largest count, prefer critical
- one (as splitting critical edges is more expensive), or one with lowest
- index, to avoid random changes with different orders of the edges. */
- FOR_EACH_EDGE (e, ei, block->preds)
- {
- if (e->flags & EDGE_DFS_BACK)
- ;
- else if (! beste)
- beste = e;
- else if (EDGE_FREQUENCY (beste) < EDGE_FREQUENCY (e))
- beste = e;
- else if (EDGE_FREQUENCY (beste) > EDGE_FREQUENCY (e))
- ;
- else if (beste->count < e->count)
- beste = e;
- else if (beste->count > e->count)
- ;
- else if ((EDGE_CRITICAL_P (e) != 0)
- != (EDGE_CRITICAL_P (beste) != 0))
- {
- if (EDGE_CRITICAL_P (e))
- beste = e;
- }
- else if (e->src->index < beste->src->index)
- beste = e;
- }
-
- /* Initialize stack at block entry. */
+ /* Choose an initial stack layout, if one hasn't already been chosen. */
if (bi->stack_in.top == -2)
{
+ edge e, beste = NULL;
+ edge_iterator ei;
+
+ /* Select the best incoming edge (typically the most frequent) to
+ use as a template for this basic block. */
+ FOR_EACH_EDGE (e, ei, block->preds)
+ if (BLOCK_INFO (e->src)->done)
+ beste = better_edge (beste, e);
+
if (beste)
- inserted |= compensate_edge (beste, file);
+ propagate_stack (beste);
else
{
/* No predecessors. Create an arbitrary input stack. */
- int reg;
-
bi->stack_in.top = -1;
for (reg = LAST_STACK_REG; reg >= FIRST_STACK_REG; --reg)
if (TEST_HARD_REG_BIT (bi->stack_in.reg_set, reg))
bi->stack_in.reg[++bi->stack_in.top] = reg;
}
}
- else
- /* Entry blocks do have stack already initialized. */
- beste = NULL;
-
- current_block = block;
- if (file)
+ if (dump_file)
{
- fprintf (file, "\nBasic block %d\nInput stack: ", block->index);
- print_stack (file, &bi->stack_in);
+ fprintf (dump_file, "\nBasic block %d\nInput stack: ", block->index);
+ print_stack (dump_file, &bi->stack_in);
}
/* Process all insns in this block. Keep track of NEXT so that we
don't process insns emitted while substituting in INSN. */
+ current_block = block;
next = BB_HEAD (block);
regstack = bi->stack_in;
+ starting_stack_p = true;
+
do
{
insn = next;
if (stack_regs_mentioned (insn)
|| CALL_P (insn))
{
- if (file)
+ if (dump_file)
{
- fprintf (file, " insn %d input stack: ",
+ fprintf (dump_file, " insn %d input stack: ",
INSN_UID (insn));
- print_stack (file, ®stack);
+ print_stack (dump_file, ®stack);
}
control_flow_insn_deleted |= subst_stack_regs (insn, ®stack);
+ starting_stack_p = false;
}
}
while (next);
- if (file)
+ if (dump_file)
{
- fprintf (file, "Expected live registers [");
+ fprintf (dump_file, "Expected live registers [");
for (reg = FIRST_STACK_REG; reg <= LAST_STACK_REG; ++reg)
if (TEST_HARD_REG_BIT (bi->out_reg_set, reg))
- fprintf (file, " %d", reg);
- fprintf (file, " ]\nOutput stack: ");
- print_stack (file, ®stack);
+ fprintf (dump_file, " %d", reg);
+ fprintf (dump_file, " ]\nOutput stack: ");
+ print_stack (dump_file, ®stack);
}
insn = BB_END (block);
{
rtx set;
- if (file)
- fprintf (file, "Emitting insn initializing reg %d\n", reg);
+ if (dump_file)
+ fprintf (dump_file, "Emitting insn initializing reg %d\n", reg);
set = gen_rtx_SET (VOIDmode, FP_MODE_REG (reg, SFmode), not_a_num);
insn = emit_insn_after (set, insn);
gcc_assert (any_malformed_asm);
win:
bi->stack_out = regstack;
-
- /* Compensate the back edges, as those wasn't visited yet. */
- FOR_EACH_EDGE (e, ei, block->succs)
- {
- if (e->flags & EDGE_DFS_BACK
- || (e->dest == EXIT_BLOCK_PTR))
- {
- gcc_assert (BLOCK_INFO (e->dest)->done
- || e->dest == block);
- inserted |= compensate_edge (e, file);
- }
- }
- FOR_EACH_EDGE (e, ei, block->preds)
- {
- if (e != beste && !(e->flags & EDGE_DFS_BACK)
- && e->src != ENTRY_BLOCK_PTR)
- {
- gcc_assert (BLOCK_INFO (e->src)->done);
- inserted |= compensate_edge (e, file);
- }
- }
-
- return inserted;
+ bi->done = true;
}
/* Convert registers in all blocks reachable from BLOCK. */
-static int
-convert_regs_2 (FILE *file, basic_block block)
+static void
+convert_regs_2 (basic_block block)
{
basic_block *stack, *sp;
- int inserted;
/* We process the blocks in a top-down manner, in a way such that one block
is only processed after all its predecessors. The number of predecessors
of every block has already been computed. */
- stack = xmalloc (sizeof (*stack) * n_basic_blocks);
+ stack = XNEWVEC (basic_block, n_basic_blocks);
sp = stack;
*sp++ = block;
- inserted = 0;
do
{
edge e;
*sp++ = e->dest;
}
- inserted |= convert_regs_1 (file, block);
- BLOCK_INFO (block)->done = 1;
+ convert_regs_1 (block);
}
while (sp != stack);
free (stack);
-
- return inserted;
}
/* Traverse all basic blocks in a function, converting the register
references in each insn from the "flat" register file that gcc uses,
to the stack-like registers the 387 uses. */
-static int
-convert_regs (FILE *file)
+static void
+convert_regs (void)
{
int inserted;
basic_block b;
/* Process all blocks reachable from all entry points. */
FOR_EACH_EDGE (e, ei, ENTRY_BLOCK_PTR->succs)
- inserted |= convert_regs_2 (file, e->dest);
+ convert_regs_2 (e->dest);
/* ??? Process all unreachable blocks. Though there's no excuse
for keeping these even when not optimizing. */
block_info bi = BLOCK_INFO (b);
if (! bi->done)
- inserted |= convert_regs_2 (file, b);
+ convert_regs_2 (b);
}
+
+ inserted |= compensate_edges ();
+
clear_aux_for_blocks ();
fixup_abnormal_edges ();
if (inserted)
commit_edge_insertions ();
- if (file)
- fputc ('\n', file);
+ if (dump_file)
+ fputc ('\n', dump_file);
+}
+\f
+/* Convert register usage from "flat" register file usage to a "stack
+ register file. FILE is the dump file, if used.
- return inserted;
+ Construct a CFG and run life analysis. Then convert each insn one
+ by one. Run a last cleanup_cfg pass, if optimizing, to eliminate
+ code duplication created when the converter inserts pop insns on
+ the edges. */
+
+static bool
+reg_to_stack (void)
+{
+ basic_block bb;
+ int i;
+ int max_uid;
+
+ /* Clean up previous run. */
+ if (stack_regs_mentioned_data != NULL)
+ VEC_free (char, heap, stack_regs_mentioned_data);
+
+ /* See if there is something to do. Flow analysis is quite
+ expensive so we might save some compilation time. */
+ for (i = FIRST_STACK_REG; i <= LAST_STACK_REG; i++)
+ if (regs_ever_live[i])
+ break;
+ if (i > LAST_STACK_REG)
+ return false;
+
+ /* Ok, floating point instructions exist. If not optimizing,
+ build the CFG and run life analysis.
+ Also need to rebuild life when superblock scheduling is done
+ as it don't update liveness yet. */
+ if (!optimize
+ || ((flag_sched2_use_superblocks || flag_sched2_use_traces)
+ && flag_schedule_insns_after_reload))
+ {
+ count_or_remove_death_notes (NULL, 1);
+ life_analysis (PROP_DEATH_NOTES);
+ }
+ mark_dfs_back_edges ();
+
+ /* Set up block info for each basic block. */
+ alloc_aux_for_blocks (sizeof (struct block_info_def));
+ FOR_EACH_BB (bb)
+ {
+ block_info bi = BLOCK_INFO (bb);
+ edge_iterator ei;
+ edge e;
+ int reg;
+
+ FOR_EACH_EDGE (e, ei, bb->preds)
+ if (!(e->flags & EDGE_DFS_BACK)
+ && e->src != ENTRY_BLOCK_PTR)
+ bi->predecessors++;
+
+ /* Set current register status at last instruction `uninitialized'. */
+ bi->stack_in.top = -2;
+
+ /* Copy live_at_end and live_at_start into temporaries. */
+ for (reg = FIRST_STACK_REG; reg <= LAST_STACK_REG; reg++)
+ {
+ if (REGNO_REG_SET_P (bb->il.rtl->global_live_at_end, reg))
+ SET_HARD_REG_BIT (bi->out_reg_set, reg);
+ if (REGNO_REG_SET_P (bb->il.rtl->global_live_at_start, reg))
+ SET_HARD_REG_BIT (bi->stack_in.reg_set, reg);
+ }
+ }
+
+ /* Create the replacement registers up front. */
+ for (i = FIRST_STACK_REG; i <= LAST_STACK_REG; i++)
+ {
+ enum machine_mode mode;
+ for (mode = GET_CLASS_NARROWEST_MODE (MODE_FLOAT);
+ mode != VOIDmode;
+ mode = GET_MODE_WIDER_MODE (mode))
+ FP_MODE_REG (i, mode) = gen_rtx_REG (mode, i);
+ for (mode = GET_CLASS_NARROWEST_MODE (MODE_COMPLEX_FLOAT);
+ mode != VOIDmode;
+ mode = GET_MODE_WIDER_MODE (mode))
+ FP_MODE_REG (i, mode) = gen_rtx_REG (mode, i);
+ }
+
+ ix86_flags_rtx = gen_rtx_REG (CCmode, FLAGS_REG);
+
+ /* A QNaN for initializing uninitialized variables.
+
+ ??? We can't load from constant memory in PIC mode, because
+ we're inserting these instructions before the prologue and
+ the PIC register hasn't been set up. In that case, fall back
+ on zero, which we can get from `ldz'. */
+
+ if (flag_pic)
+ not_a_num = CONST0_RTX (SFmode);
+ else
+ {
+ not_a_num = gen_lowpart (SFmode, GEN_INT (0x7fc00000));
+ not_a_num = force_const_mem (SFmode, not_a_num);
+ }
+
+ /* Allocate a cache for stack_regs_mentioned. */
+ max_uid = get_max_uid ();
+ stack_regs_mentioned_data = VEC_alloc (char, heap, max_uid + 1);
+ memset (VEC_address (char, stack_regs_mentioned_data),
+ 0, sizeof (char) * max_uid + 1);
+
+ convert_regs ();
+
+ free_aux_for_blocks ();
+ return true;
}
#endif /* STACK_REGS */
+\f
+static bool
+gate_handle_stack_regs (void)
+{
+#ifdef STACK_REGS
+ return 1;
+#else
+ return 0;
+#endif
+}
-#include "gt-reg-stack.h"
+/* Convert register usage from flat register file usage to a stack
+ register file. */
+static unsigned int
+rest_of_handle_stack_regs (void)
+{
+#ifdef STACK_REGS
+ if (reg_to_stack () && optimize)
+ {
+ regstack_completed = 1;
+ if (cleanup_cfg (CLEANUP_EXPENSIVE | CLEANUP_POST_REGSTACK
+ | (flag_crossjumping ? CLEANUP_CROSSJUMP : 0))
+ && (flag_reorder_blocks || flag_reorder_blocks_and_partition))
+ {
+ reorder_basic_blocks (0);
+ cleanup_cfg (CLEANUP_EXPENSIVE | CLEANUP_POST_REGSTACK);
+ }
+ }
+ else
+ regstack_completed = 1;
+#endif
+ return 0;
+}
+
+struct tree_opt_pass pass_stack_regs =
+{
+ "stack", /* name */
+ gate_handle_stack_regs, /* gate */
+ rest_of_handle_stack_regs, /* execute */
+ NULL, /* sub */
+ NULL, /* next */
+ 0, /* static_pass_number */
+ TV_REG_STACK, /* tv_id */
+ 0, /* properties_required */
+ 0, /* properties_provided */
+ 0, /* properties_destroyed */
+ 0, /* todo_flags_start */
+ TODO_dump_func |
+ TODO_ggc_collect, /* todo_flags_finish */
+ 'k' /* letter */
+};