/* Convert function calls to rtl insns, for GNU C compiler.
Copyright (C) 1989, 1992, 1993, 1994, 1995, 1996, 1997, 1998
- 1999, 2000 Free Software Foundation, Inc.
+ 1999, 2000, 2001 Free Software Foundation, Inc.
This file is part of GNU CC.
#include "expr.h"
#include "function.h"
#include "regs.h"
-#include "insn-flags.h"
#include "toplev.h"
#include "output.h"
#include "tm_p.h"
-
-#ifndef ACCUMULATE_OUTGOING_ARGS
-#define ACCUMULATE_OUTGOING_ARGS 0
-#endif
-
-/* Supply a default definition for PUSH_ARGS. */
-#ifndef PUSH_ARGS
-#ifdef PUSH_ROUNDING
-#define PUSH_ARGS !ACCUMULATE_OUTGOING_ARGS
-#else
-#define PUSH_ARGS 0
-#endif
-#endif
+#include "timevar.h"
+#include "sbitmap.h"
#if !defined FUNCTION_OK_FOR_SIBCALL
#define FUNCTION_OK_FOR_SIBCALL(DECL) 1
#define PUSH_ARGS_REVERSED 0
#endif
+#ifndef STACK_POINTER_OFFSET
+#define STACK_POINTER_OFFSET 0
+#endif
+
/* Like PREFERRED_STACK_BOUNDARY but in units of bytes, not bits. */
#define STACK_BYTES (PREFERRED_STACK_BOUNDARY / BITS_PER_UNIT)
PARALLEL if the arg is to be copied into multiple non-contiguous
registers. */
rtx reg;
+ /* Register to pass this argument in when generating tail call sequence.
+ This is not the same register as for normal calls on machines with
+ register windows. */
+ rtx tail_call_reg;
/* If REG was promoted from the actual mode of the argument expression,
indicates whether the promotion is sign- or zero-extended. */
int unsignedp;
/* Size of STACK_USAGE_MAP. */
static int highest_outgoing_arg_in_use;
+/* A bitmap of virtual-incoming stack space. Bit is set if the corresponding
+ stack location's tail call argument has been already stored into the stack.
+ This bitmap is used to prevent sibling call optimization if function tries
+ to use parent's incoming argument slots when they have been already
+ overwritten with tail call arguments. */
+static sbitmap stored_args_map;
+
/* stack_arg_under_construction is nonzero when an argument may be
initialized with a constructor call (including a C function that
returns a BLKmode struct) and expand_call must take special action
static int calls_function PARAMS ((tree, int));
static int calls_function_1 PARAMS ((tree, int));
-/* Nonzero if this is a call to a `const' function. */
+/* Nonzero if this is a call to a `const' function. */
#define ECF_CONST 1
/* Nonzero if this is a call to a `volatile' function. */
#define ECF_NORETURN 2
-/* Nonzero if this is a call to malloc or a related function. */
+/* Nonzero if this is a call to malloc or a related function. */
#define ECF_MALLOC 4
/* Nonzero if it is plausible that this is a call to alloca. */
#define ECF_MAY_BE_ALLOCA 8
/* Nonzero if this is a call to "pure" function (like const function,
but may read memory. */
#define ECF_PURE 512
+/* Nonzero if this is a call to a function that returns with the stack
+ pointer depressed. */
+#define ECF_SP_DEPRESSED 1024
static void emit_call_1 PARAMS ((rtx, tree, tree, HOST_WIDE_INT,
HOST_WIDE_INT, HOST_WIDE_INT, rtx,
static void precompute_register_parameters PARAMS ((int,
struct arg_data *,
int *));
-static void store_one_arg PARAMS ((struct arg_data *, rtx, int, int,
+static int store_one_arg PARAMS ((struct arg_data *, rtx, int, int,
int));
static void store_unaligned_arguments_into_pseudos PARAMS ((struct arg_data *,
int));
struct args_size *));
static void precompute_arguments PARAMS ((int, int,
struct arg_data *));
-static int compute_argument_block_size PARAMS ((int,
+static int compute_argument_block_size PARAMS ((int,
struct args_size *,
int));
static void initialize_argument_information PARAMS ((int,
rtx, int));
static rtx rtx_for_function_call PARAMS ((tree, tree));
static void load_register_parameters PARAMS ((struct arg_data *,
- int, rtx *));
-static int libfunc_nothrow PARAMS ((rtx));
-static rtx emit_library_call_value_1 PARAMS ((int, rtx, rtx, int,
+ int, rtx *, int));
+static rtx emit_library_call_value_1 PARAMS ((int, rtx, rtx,
+ enum libcall_type,
enum machine_mode,
int, va_list));
static int special_function_p PARAMS ((tree, int));
static int flags_from_decl_or_type PARAMS ((tree));
static rtx try_to_integrate PARAMS ((tree, tree, rtx,
int, tree, rtx));
+static int check_sibcall_argument_overlap_1 PARAMS ((rtx));
+static int check_sibcall_argument_overlap PARAMS ((rtx, struct arg_data *));
+
static int combine_pending_stack_adjustment_and_call
PARAMS ((int, struct args_size *, int));
int which;
{
int val;
+
calls_function_save_exprs = 0;
val = calls_function_1 (exp, which);
calls_function_save_exprs = 0;
return val;
}
+/* Recursive function to do the work of above function. */
+
static int
calls_function_1 (exp, which)
tree exp;
{
register int i;
enum tree_code code = TREE_CODE (exp);
- int type = TREE_CODE_CLASS (code);
- int length = tree_code_length[(int) code];
+ int class = TREE_CODE_CLASS (code);
+ int length = first_rtl_op (code);
/* If this code is language-specific, we don't know what it will do. */
if ((int) code >= NUM_TREE_CODES)
case CALL_EXPR:
if (which == 0)
return 1;
+ else if ((TREE_CODE (TREE_TYPE (TREE_TYPE (TREE_OPERAND (exp, 0))))
+ == FUNCTION_TYPE)
+ && (TYPE_RETURNS_STACK_DEPRESSED
+ (TREE_TYPE (TREE_TYPE (TREE_OPERAND (exp, 0))))))
+ return 1;
else if (TREE_CODE (TREE_OPERAND (exp, 0)) == ADDR_EXPR
&& (TREE_CODE (TREE_OPERAND (TREE_OPERAND (exp, 0), 0))
- == FUNCTION_DECL))
- {
- tree fndecl = TREE_OPERAND (TREE_OPERAND (exp, 0), 0);
- int flags = special_function_p (fndecl, 0);
- if (flags & ECF_MAY_BE_ALLOCA)
- return 1;
- }
+ == FUNCTION_DECL)
+ && (special_function_p (TREE_OPERAND (TREE_OPERAND (exp, 0), 0),
+ 0)
+ & ECF_MAY_BE_ALLOCA))
+ return 1;
- /* Third operand is RTL. */
- length = 2;
break;
+ case CONSTRUCTOR:
+ {
+ tree tem;
+
+ for (tem = CONSTRUCTOR_ELTS (exp); tem != 0; tem = TREE_CHAIN (tem))
+ if (calls_function_1 (TREE_VALUE (tem), which))
+ return 1;
+ }
+
+ return 0;
+
case SAVE_EXPR:
if (SAVE_EXPR_RTL (exp) != 0)
return 0;
case BLOCK:
{
register tree local;
+ register tree subblock;
for (local = BLOCK_VARS (exp); local; local = TREE_CHAIN (local))
if (DECL_INITIAL (local) != 0
&& calls_function_1 (DECL_INITIAL (local), which))
return 1;
- }
- {
- register tree subblock;
for (subblock = BLOCK_SUBBLOCKS (exp);
subblock;
return 1;
}
return 0;
+
case TREE_LIST:
for (; exp != 0; exp = TREE_CHAIN (exp))
if (calls_function_1 (TREE_VALUE (exp), which))
return 1;
return 0;
- case METHOD_CALL_EXPR:
- length = 3;
- break;
-
- case WITH_CLEANUP_EXPR:
- length = 1;
- break;
-
- case RTL_EXPR:
- return 0;
-
default:
break;
}
- /* Only expressions and references can contain calls. */
- if (type != 'e' && type != '<' && type != '1' && type != '2' && type != 'r'
- && type != 'b')
+ /* Only expressions, references, and blocks can contain calls. */
+ if (! IS_EXPR_CODE_CLASS (class) && class != 'r' && class != 'b')
return 0;
for (i = 0; i < length; i++)
CALL_INSN_FUNCTION_USAGE information. */
rtx
-prepare_call_address (funexp, fndecl, call_fusage, reg_parm_seen)
+prepare_call_address (funexp, fndecl, call_fusage, reg_parm_seen, sibcallp)
rtx funexp;
tree fndecl;
rtx *call_fusage;
int reg_parm_seen;
+ int sibcallp;
{
rtx static_chain_value = 0;
funexp = ((SMALL_REGISTER_CLASSES && reg_parm_seen)
? force_not_mem (memory_address (FUNCTION_MODE, funexp))
: memory_address (FUNCTION_MODE, funexp));
- else
+ else if (! sibcallp)
{
#ifndef NO_FUNCTION_CSE
if (optimize && ! flag_no_function_cse)
We restore `inhibit_defer_pop' to that value.
CALL_FUSAGE is either empty or an EXPR_LIST of USE expressions that
- denote registers used by the called function. */
+ denote registers used by the called function. */
static void
emit_call_1 (funexp, fndecl, funtype, stack_size, rounded_stack_size,
int ecf_flags;
{
rtx rounded_stack_size_rtx = GEN_INT (rounded_stack_size);
-#if defined (HAVE_call) && defined (HAVE_call_value)
- rtx struct_value_size_rtx = GEN_INT (struct_value_size);
-#endif
rtx call_insn;
int already_popped = 0;
HOST_WIDE_INT n_popped = RETURN_POPS_ARGS (fndecl, funtype, stack_size);
+#if defined (HAVE_call) && defined (HAVE_call_value)
+ rtx struct_value_size_rtx;
+ struct_value_size_rtx = GEN_INT (struct_value_size);
+#endif
/* Ensure address is valid. SYMBOL_REF is already valid, so no need,
and we don't want to load it into a register as an optimization,
#if defined (HAVE_sibcall_pop) && defined (HAVE_sibcall_value_pop)
if ((ecf_flags & ECF_SIBCALL)
&& HAVE_sibcall_pop && HAVE_sibcall_value_pop
- && (RETURN_POPS_ARGS (fndecl, funtype, stack_size) > 0
- || stack_size == 0))
+ && (n_popped > 0 || stack_size == 0))
{
- rtx n_pop = GEN_INT (RETURN_POPS_ARGS (fndecl, funtype, stack_size));
+ rtx n_pop = GEN_INT (n_popped));
rtx pat;
/* If this subroutine pops its own args, record that in the call insn
if possible, for the sake of frame pointer elimination. */
if (valreg)
- pat = gen_sibcall_value_pop (valreg,
+ pat = GEN_SIBCALL_VALUE_POP (valreg,
gen_rtx_MEM (FUNCTION_MODE, funexp),
rounded_stack_size_rtx, next_arg_reg,
n_pop);
else
- pat = gen_sibcall_pop (gen_rtx_MEM (FUNCTION_MODE, funexp),
+ pat = GEN_SIBCALL_POP (gen_rtx_MEM (FUNCTION_MODE, funexp),
rounded_stack_size_rtx, next_arg_reg, n_pop);
emit_call_insn (pat);
even if the call has no arguments to pop. */
#if defined (HAVE_call) && defined (HAVE_call_value)
if (HAVE_call && HAVE_call_value && HAVE_call_pop && HAVE_call_value_pop
- && n_popped > 0)
+ && n_popped > 0 && ! (ecf_flags & ECF_SP_DEPRESSED))
#else
if (HAVE_call_pop && HAVE_call_value_pop)
#endif
if possible, for the sake of frame pointer elimination. */
if (valreg)
- pat = gen_call_value_pop (valreg,
+ pat = GEN_CALL_VALUE_POP (valreg,
gen_rtx_MEM (FUNCTION_MODE, funexp),
rounded_stack_size_rtx, next_arg_reg, n_pop);
else
- pat = gen_call_pop (gen_rtx_MEM (FUNCTION_MODE, funexp),
+ pat = GEN_CALL_POP (gen_rtx_MEM (FUNCTION_MODE, funexp),
rounded_stack_size_rtx, next_arg_reg, n_pop);
emit_call_insn (pat);
&& HAVE_sibcall && HAVE_sibcall_value)
{
if (valreg)
- emit_call_insn (gen_sibcall_value (valreg,
+ emit_call_insn (GEN_SIBCALL_VALUE (valreg,
gen_rtx_MEM (FUNCTION_MODE, funexp),
rounded_stack_size_rtx,
next_arg_reg, NULL_RTX));
else
- emit_call_insn (gen_sibcall (gen_rtx_MEM (FUNCTION_MODE, funexp),
+ emit_call_insn (GEN_SIBCALL (gen_rtx_MEM (FUNCTION_MODE, funexp),
rounded_stack_size_rtx, next_arg_reg,
struct_value_size_rtx));
}
if (HAVE_call && HAVE_call_value)
{
if (valreg)
- emit_call_insn (gen_call_value (valreg,
+ emit_call_insn (GEN_CALL_VALUE (valreg,
gen_rtx_MEM (FUNCTION_MODE, funexp),
rounded_stack_size_rtx, next_arg_reg,
NULL_RTX));
else
- emit_call_insn (gen_call (gen_rtx_MEM (FUNCTION_MODE, funexp),
+ emit_call_insn (GEN_CALL (gen_rtx_MEM (FUNCTION_MODE, funexp),
rounded_stack_size_rtx, next_arg_reg,
struct_value_size_rtx));
}
REG_NOTES (call_insn) = gen_rtx_EXPR_LIST (REG_EH_REGION, const0_rtx,
REG_NOTES (call_insn));
+ if (ecf_flags & ECF_NORETURN)
+ REG_NOTES (call_insn) = gen_rtx_EXPR_LIST (REG_NORETURN, const0_rtx,
+ REG_NOTES (call_insn));
+
SIBLING_CALL_P (call_insn) = ((ecf_flags & ECF_SIBCALL) != 0);
/* Restore this now, so that we do defer pops for this call's args
If returning from the subroutine does pop the args, indicate that the
stack pointer will be changed. */
- if (rounded_stack_size != 0)
+ if (rounded_stack_size != 0 && ! (ecf_flags & ECF_SP_DEPRESSED))
{
if (flag_defer_pop && inhibit_defer_pop == 0
- && !(ecf_flags & (ECF_CONST | ECF_PURE)))
+ && ! (ecf_flags & (ECF_CONST | ECF_PURE)))
pending_stack_adjust += rounded_stack_size;
else
adjust_stack (rounded_stack_size_rtx);
??? We may optimize similar to defer_pop above, but it is
probably not worthwhile.
-
+
??? It will be worthwhile to enable combine_stack_adjustments even for
such machines. */
else if (n_popped)
think they are. */
&& DECL_CONTEXT (fndecl) == NULL_TREE && TREE_PUBLIC (fndecl))
{
- char *name = IDENTIFIER_POINTER (DECL_NAME (fndecl));
- char *tname = name;
+ const char *name = IDENTIFIER_POINTER (DECL_NAME (fndecl));
+ const char *tname = name;
/* We assume that alloca will always be called by name. It
makes no sense to pass it as a pointer-to-function to
it may return the same address across multiple calls.
C++ operator new is not suitable because it is not required
to return a unique pointer; indeed, the standard placement new
- just returns its argument. */
+ just returns its argument. */
else if (TYPE_MODE (TREE_TYPE (TREE_TYPE (fndecl))) == Pmode
&& (! strcmp (tname, "malloc")
|| ! strcmp (tname, "calloc")
}
/* Return nonzero when tree represent call to longjmp. */
+
int
setjmp_call_p (fndecl)
tree fndecl;
}
/* Detect flags (function attributes) from the function type node. */
+
static int
flags_from_decl_or_type (exp)
tree exp;
{
int flags = 0;
+
/* ??? We can't set IS_MALLOC for function types? */
if (DECL_P (exp))
{
flags |= ECF_NOTHROW;
}
- if (TREE_READONLY (exp) && !TREE_THIS_VOLATILE (exp))
+ if (TREE_READONLY (exp) && ! TREE_THIS_VOLATILE (exp))
flags |= ECF_CONST;
if (TREE_THIS_VOLATILE (exp))
return flags;
}
-
/* Precompute all register parameters as described by ARGS, storing values
into fields within the ARGS array.
TYPE_MODE (TREE_TYPE (args[i].tree_value)),
args[i].value, args[i].unsignedp);
- /* If the value is expensive, and we are inside an appropriately
+ /* If the value is expensive, and we are inside an appropriately
short loop, put the value into a pseudo and then put the pseudo
into the hard reg.
|| (GET_CODE (args[i].value) == SUBREG
&& GET_CODE (SUBREG_REG (args[i].value)) == REG)))
&& args[i].mode != BLKmode
- && rtx_cost (args[i].value, SET) > 2
+ && rtx_cost (args[i].value, SET) > COSTS_N_INSNS (1)
&& ((SMALL_REGISTER_CLASSES && *reg_parm_seen)
|| preserve_subexpressions_p ()))
args[i].value = copy_to_mode_reg (args[i].mode, args[i].value);
/* The argument list is the property of the called routine and it
may clobber it. If the fixed area has been used for previous
parameters, we must save and restore it. */
+
static rtx
save_fixed_argument_area (reg_parm_stack_space, argblock,
low_to_save, high_to_save)
/* If we don't have the required alignment, must do this in BLKmode. */
if ((*low_to_save & (MIN (GET_MODE_SIZE (save_mode),
- BIGGEST_ALIGNMENT / UNITS_PER_WORD) - 1)))
+ BIGGEST_ALIGNMENT / UNITS_PER_WORD) - 1)))
save_mode = BLKmode;
#ifdef ARGS_GROW_DOWNWARD
- stack_area = gen_rtx_MEM (save_mode,
- memory_address (save_mode,
- plus_constant (argblock,
- - *high_to_save)));
+ stack_area
+ = gen_rtx_MEM (save_mode,
+ memory_address (save_mode,
+ plus_constant (argblock,
+ - *high_to_save)));
#else
stack_area = gen_rtx_MEM (save_mode,
memory_address (save_mode,
high_to_save - low_to_save + 1, PARM_BOUNDARY);
}
#endif
-
+
/* If any elements in ARGS refer to parameters that are to be passed in
registers, but not in memory, and whose alignment does not permit a
direct copy into registers. Copy the values into a group of pseudos
- which we will later copy into the appropriate hard registers.
+ which we will later copy into the appropriate hard registers.
Pseudos for each unaligned argument will be stored into the array
args[argnum].aligned_regs. The caller is responsible for deallocating
int num_actuals;
{
int i, j;
-
+
for (i = 0; i < num_actuals; i++)
if (args[i].reg != 0 && ! args[i].pass_on_stack
&& args[i].mode == BLKmode
}
/* Fill in ARGS_SIZE and ARGS array based on the parameters found in
- ACTPARMS.
+ ACTPARMS.
NUM_ACTUALS is the total number of parameters.
struct args_size alignment_pad;
int i;
tree p;
-
+
args_size->constant = 0;
args_size->var = 0;
/* If TYPE is a transparent union, pass things the way we would
pass the first field of the union. We have already verified that
the modes are the same. */
- if (TYPE_TRANSPARENT_UNION (type))
+ if (TREE_CODE (type) == UNION_TYPE && TYPE_TRANSPARENT_UNION (type))
type = TREE_TYPE (TYPE_FIELDS (type));
/* Decide where to pass this arg.
but it is safe in the only case where this is a useful
optimization; namely, when the argument is a plain object.
In that case, the frontend is just asking the backend to
- make a bitwise copy of the argument. */
-
+ make a bitwise copy of the argument. */
+
if (TREE_CODE (args[i].tree_value) == TARGET_EXPR
&& (DECL_P (TREE_OPERAND (args[i].tree_value, 1)))
&& ! REG_P (DECL_RTL (TREE_OPERAND (args[i].tree_value, 1))))
}
copy = gen_rtx_MEM (BLKmode,
- allocate_dynamic_stack_space (size_rtx,
- NULL_RTX,
- TYPE_ALIGN (type)));
+ allocate_dynamic_stack_space
+ (size_rtx, NULL_RTX, TYPE_ALIGN (type)));
+ set_mem_attributes (copy, type, 1);
}
else
- {
- int size = int_size_in_bytes (type);
- copy = assign_stack_temp (TYPE_MODE (type), size, 0);
- }
-
- MEM_SET_IN_STRUCT_P (copy, AGGREGATE_TYPE_P (type));
+ copy = assign_temp (type, 0, 1, 0);
store_expr (args[i].tree_value, copy, 0);
*ecf_flags &= ~(ECF_CONST | ECF_PURE);
args[i].unsignedp = unsignedp;
args[i].mode = mode;
+ args[i].reg = FUNCTION_ARG (*args_so_far, mode, type,
+ argpos < n_named_args);
#ifdef FUNCTION_INCOMING_ARG
/* If this is a sibling call and the machine has register windows, the
register window has to be unwinded before calling the routine, so
arguments have to go into the incoming registers. */
- if (*ecf_flags & ECF_SIBCALL)
- args[i].reg = FUNCTION_INCOMING_ARG (*args_so_far, mode, type,
- argpos < n_named_args);
- else
+ args[i].tail_call_reg = FUNCTION_INCOMING_ARG (*args_so_far, mode, type,
+ argpos < n_named_args);
+#else
+ args[i].tail_call_reg = args[i].reg;
#endif
- args[i].reg = FUNCTION_ARG (*args_so_far, mode, type,
- argpos < n_named_args);
#ifdef FUNCTION_ARG_PARTIAL_NREGS
if (args[i].reg)
args[i].size.constant -= ((args[i].partial * UNITS_PER_WORD)
/ (PARM_BOUNDARY / BITS_PER_UNIT)
* (PARM_BOUNDARY / BITS_PER_UNIT));
-
+
/* Update ARGS_SIZE, the total stack space for args so far. */
args_size->constant += args[i].size.constant;
static int
compute_argument_block_size (reg_parm_stack_space, args_size,
- preferred_stack_boundary)
+ preferred_stack_boundary)
int reg_parm_stack_space;
struct args_size *args_size;
int preferred_stack_boundary ATTRIBUTE_UNUSED;
if (preferred_stack_boundary > 1)
{
/* We don't handle this case yet. To handle it correctly we have
- to add the delta, round and substract the delta.
+ to add the delta, round and substract the delta.
Currently no machine description requires this support. */
if (stack_pointer_delta & (preferred_stack_boundary - 1))
- abort();
+ abort ();
args_size->var = round_up (args_size->var, preferred_stack_boundary);
}
#endif
NUM_ACTUALS is the number of arguments.
- ARGS is an array containing information for each argument; this routine
- fills in the INITIAL_VALUE and VALUE fields for each precomputed argument.
- */
+ ARGS is an array containing information for each argument; this
+ routine fills in the INITIAL_VALUE and VALUE fields for each
+ precomputed argument. */
static void
precompute_arguments (flags, num_actuals, args)
if ((flags & (ECF_CONST | ECF_PURE))
|| calls_function (args[i].tree_value, !ACCUMULATE_OUTGOING_ARGS))
{
+ enum machine_mode mode;
+
/* If this is an addressable type, we cannot pre-evaluate it. */
if (TREE_ADDRESSABLE (TREE_TYPE (args[i].tree_value)))
abort ();
args[i].initial_value = args[i].value
= protect_from_queue (args[i].value, 0);
- if (TYPE_MODE (TREE_TYPE (args[i].tree_value)) != args[i].mode)
+ mode = TYPE_MODE (TREE_TYPE (args[i].tree_value));
+ if (mode != args[i].mode)
{
args[i].value
- = convert_modes (args[i].mode,
- TYPE_MODE (TREE_TYPE (args[i].tree_value)),
+ = convert_modes (args[i].mode, mode,
args[i].value, args[i].unsignedp);
#ifdef PROMOTE_FOR_CALL_ONLY
/* CSE will replace this only if it contains args[i].value
&& GET_MODE_CLASS (args[i].mode) == MODE_INT)
{
args[i].initial_value
- = gen_rtx_SUBREG (TYPE_MODE (TREE_TYPE (args[i].tree_value)),
- args[i].value, 0);
+ = gen_lowpart_SUBREG (mode, args[i].value);
SUBREG_PROMOTED_VAR_P (args[i].initial_value) = 1;
SUBREG_PROMOTED_UNSIGNED_P (args[i].initial_value)
= args[i].unsignedp;
/* If we preallocated stack space, compute the address of each argument
and store it into the ARGS array.
- We need not ensure it is a valid memory address here; it will be
+ We need not ensure it is a valid memory address here; it will be
validized when it is used.
ARGBLOCK is an rtx for the address of the outgoing arguments. */
addr = plus_constant (addr, arg_offset);
args[i].stack = gen_rtx_MEM (args[i].mode, addr);
- MEM_SET_IN_STRUCT_P
- (args[i].stack,
- AGGREGATE_TYPE_P (TREE_TYPE (args[i].tree_value)));
+ set_mem_attributes (args[i].stack,
+ TREE_TYPE (args[i].tree_value), 1);
if (GET_CODE (slot_offset) == CONST_INT)
addr = plus_constant (arg_reg, INTVAL (slot_offset));
addr = plus_constant (addr, arg_offset);
args[i].stack_slot = gen_rtx_MEM (args[i].mode, addr);
+ set_mem_attributes (args[i].stack_slot,
+ TREE_TYPE (args[i].tree_value), 1);
+
+ /* Function incoming arguments may overlap with sibling call
+ outgoing arguments and we cannot allow reordering of reads
+ from function arguments with stores to outgoing arguments
+ of sibling calls. */
+ set_mem_alias_set (args[i].stack, 0);
+ set_mem_alias_set (args[i].stack_slot, 0);
}
}
}
-
+
/* Given a FNDECL and EXP, return an rtx suitable for use as a target address
in a call instruction.
{
rtx funaddr;
push_temp_slots ();
- funaddr = funexp =
- expand_expr (TREE_OPERAND (exp, 0), NULL_RTX, VOIDmode, 0);
- pop_temp_slots (); /* FUNEXP can't be BLKmode */
+ funaddr = funexp =
+ expand_expr (TREE_OPERAND (exp, 0), NULL_RTX, VOIDmode, 0);
+ pop_temp_slots (); /* FUNEXP can't be BLKmode. */
/* Check the function is executable. */
if (current_function_check_memory_usage)
#ifdef POINTERS_EXTEND_UNSIGNED
/* It might be OK to convert funexp in place, but there's
a lot going on between here and when it happens naturally
- that this seems safer. */
- funaddr = convert_memory_address (Pmode, funexp);
+ that this seems safer. */
+ funaddr = convert_memory_address (Pmode, funexp);
#endif
- emit_library_call (chkr_check_exec_libfunc, 1,
- VOIDmode, 1,
- funaddr, Pmode);
+ emit_library_call (chkr_check_exec_libfunc, LCT_CONST_MAKE_BLOCK,
+ VOIDmode, 1, funaddr, Pmode);
}
emit_queue ();
}
/* Do the register loads required for any wholly-register parms or any
parms which are passed both on the stack and in a register. Their
- expressions were already evaluated.
+ expressions were already evaluated.
Mark all register-parms as living through the call, putting these USE
insns in the CALL_INSN_FUNCTION_USAGE field. */
static void
-load_register_parameters (args, num_actuals, call_fusage)
+load_register_parameters (args, num_actuals, call_fusage, flags)
struct arg_data *args;
int num_actuals;
rtx *call_fusage;
+ int flags;
{
int i, j;
for (i = 0; i < num_actuals; i++)
#endif
{
- rtx reg = args[i].reg;
+ rtx reg = ((flags & ECF_SIBCALL)
+ ? args[i].tail_call_reg : args[i].reg);
int partial = args[i].partial;
int nregs;
}
}
-/* Try to integreate function. See expand_inline_function for documentation
+/* Try to integrate function. See expand_inline_function for documentation
about the parameters. */
static rtx
before_call = get_last_insn ();
+ timevar_push (TV_INTEGRATION);
+
temp = expand_inline_function (fndecl, actparms, target,
ignore, type,
structure_value_addr);
+ timevar_pop (TV_INTEGRATION);
+
/* If inlining succeeded, return. */
if (temp != (rtx) (HOST_WIDE_INT) - 1)
{
outgoing argument list in addition to the requested
space, but there is no way to ask for stack space such
that an argument list of a certain length can be
- safely constructed.
+ safely constructed.
Add the stack space reserved for register arguments, if
any, in the inline function. What is really needed is the
just pushed the arguments without adjust the stack here. */
HOST_WIDE_INT unadjusted_alignment;
- unadjusted_alignment
+ unadjusted_alignment
= ((stack_pointer_delta + unadjusted_args_size)
% preferred_unit_stack_boundary);
-UNADJUSTED_ALIGNMENT modulo the PREFERRED_UNIT_STACK_BOUNDARY. */
/* Begin by trying to pop all the bytes. */
- unadjusted_alignment
- = (unadjusted_alignment
+ unadjusted_alignment
+ = (unadjusted_alignment
- (pending_stack_adjust % preferred_unit_stack_boundary));
adjustment = pending_stack_adjust;
/* Push enough additional bytes that the stack will be aligned
after the arguments are pushed. */
- if (unadjusted_alignment >= 0)
- adjustment -= preferred_unit_stack_boundary - unadjusted_alignment;
- else
- adjustment += unadjusted_alignment;
-
+ if (preferred_unit_stack_boundary > 1)
+ {
+ if (unadjusted_alignment > 0)
+ adjustment -= preferred_unit_stack_boundary - unadjusted_alignment;
+ else
+ adjustment += unadjusted_alignment;
+ }
+
/* Now, sets ARGS_SIZE->CONSTANT so that we pop the right number of
bytes after the call. The right number is the entire
PENDING_STACK_ADJUST less our ADJUSTMENT plus the amount required
by the arguments in the first place. */
- args_size->constant
+ args_size->constant
= pending_stack_adjust - adjustment + unadjusted_args_size;
return adjustment;
}
+/* Scan X expression if it does not dereference any argument slots
+ we already clobbered by tail call arguments (as noted in stored_args_map
+ bitmap).
+ Return non-zero if X expression dereferences such argument slots,
+ zero otherwise. */
+
+static int
+check_sibcall_argument_overlap_1 (x)
+ rtx x;
+{
+ RTX_CODE code;
+ int i, j;
+ unsigned int k;
+ const char *fmt;
+
+ if (x == NULL_RTX)
+ return 0;
+
+ code = GET_CODE (x);
+
+ if (code == MEM)
+ {
+ if (XEXP (x, 0) == current_function_internal_arg_pointer)
+ i = 0;
+ else if (GET_CODE (XEXP (x, 0)) == PLUS
+ && XEXP (XEXP (x, 0), 0) ==
+ current_function_internal_arg_pointer
+ && GET_CODE (XEXP (XEXP (x, 0), 1)) == CONST_INT)
+ i = INTVAL (XEXP (XEXP (x, 0), 1));
+ else
+ return 0;
+
+#ifdef ARGS_GROW_DOWNWARD
+ i = -i - GET_MODE_SIZE (GET_MODE (x));
+#endif
+
+ for (k = 0; k < GET_MODE_SIZE (GET_MODE (x)); k++)
+ if (i + k < stored_args_map->n_bits
+ && TEST_BIT (stored_args_map, i + k))
+ return 1;
+
+ return 0;
+ }
+
+ /* Scan all subexpressions. */
+ fmt = GET_RTX_FORMAT (code);
+ for (i = 0; i < GET_RTX_LENGTH (code); i++, fmt++)
+ {
+ if (*fmt == 'e')
+ {
+ if (check_sibcall_argument_overlap_1 (XEXP (x, i)))
+ return 1;
+ }
+ else if (*fmt == 'E')
+ {
+ for (j = 0; j < XVECLEN (x, i); j++)
+ if (check_sibcall_argument_overlap_1 (XVECEXP (x, i, j)))
+ return 1;
+ }
+ }
+ return 0;
+}
+
+/* Scan sequence after INSN if it does not dereference any argument slots
+ we already clobbered by tail call arguments (as noted in stored_args_map
+ bitmap). Add stack slots for ARG to stored_args_map bitmap afterwards.
+ Return non-zero if sequence after INSN dereferences such argument slots,
+ zero otherwise. */
+
+static int
+check_sibcall_argument_overlap (insn, arg)
+ rtx insn;
+ struct arg_data *arg;
+{
+ int low, high;
+
+ if (insn == NULL_RTX)
+ insn = get_insns ();
+ else
+ insn = NEXT_INSN (insn);
+
+ for (; insn; insn = NEXT_INSN (insn))
+ if (INSN_P (insn)
+ && check_sibcall_argument_overlap_1 (PATTERN (insn)))
+ break;
+
+#ifdef ARGS_GROW_DOWNWARD
+ low = -arg->offset.constant - arg->size.constant;
+#else
+ low = arg->offset.constant;
+#endif
+
+ for (high = low + arg->size.constant; low < high; low++)
+ SET_BIT (stored_args_map, low);
+ return insn != NULL_RTX;
+}
+
/* Generate all the code for a function call
and return an rtx for its value.
Store the value in TARGET (specified as an rtx) if convenient.
/* Declaration of the function being called,
or 0 if the function is computed (not known by name). */
tree fndecl = 0;
- char *name = 0;
rtx insn;
- int try_tail_call;
+ int try_tail_call = 1;
+ int try_tail_recursion = 1;
int pass;
/* Register in which non-BLKmode value will be returned,
/* Total size in bytes of all the stack-parms scanned so far. */
struct args_size args_size;
+ struct args_size adjusted_args_size;
/* Size of arguments before any adjustments (such as rounding). */
int unadjusted_args_size;
/* Data on reg parms scanned so far. */
int reg_parm_seen;
/* Nonzero if this is an indirect function call. */
- /* Nonzero if we must avoid push-insns in the args for this call.
+ /* Nonzero if we must avoid push-insns in the args for this call.
If stack space is allocated for register parameters, but not by the
caller, then it is preallocated in the fixed part of the stack frame.
So the entire argument block must then be preallocated (i.e., we
int old_inhibit_defer_pop = inhibit_defer_pop;
int old_stack_allocated;
rtx call_fusage;
- register tree p;
+ register tree p = TREE_OPERAND (exp, 0);
register int i;
/* The alignment of the stack, in bits. */
HOST_WIDE_INT preferred_stack_boundary;
flags |= flags_from_decl_or_type (fndecl);
}
- /* If we don't have specific function to call, see if we have a
+ /* If we don't have specific function to call, see if we have a
attributes set in the type. */
else
+ flags |= flags_from_decl_or_type (TREE_TYPE (TREE_TYPE (p)));
+
+ /* Mark if the function returns with the stack pointer depressed. */
+ if (TREE_CODE (TREE_TYPE (TREE_TYPE (p))) == FUNCTION_TYPE
+ && TYPE_RETURNS_STACK_DEPRESSED (TREE_TYPE (TREE_TYPE (p))))
{
- p = TREE_OPERAND (exp, 0);
- flags |= flags_from_decl_or_type (TREE_TYPE (TREE_TYPE (p)));
+ flags |= ECF_SP_DEPRESSED;
+ flags &= ~(ECF_PURE | ECF_CONST);
}
#ifdef REG_PARM_STACK_SPACE
structure_value_addr = XEXP (target, 0);
else
{
- /* Assign a temporary to hold the value. */
- tree d;
-
/* For variable-sized objects, we must be called with a target
specified. If we were to allocate space on the stack here,
we would have no way of knowing when to free it. */
+ rtx d = assign_temp (TREE_TYPE (exp), 1, 1, 1);
- if (struct_value_size < 0)
- abort ();
-
- /* This DECL is just something to feed to mark_addressable;
- it doesn't get pushed. */
- d = build_decl (VAR_DECL, NULL_TREE, TREE_TYPE (exp));
- DECL_RTL (d) = assign_temp (TREE_TYPE (exp), 1, 0, 1);
- mark_addressable (d);
- mark_temp_addr_taken (DECL_RTL (d));
- structure_value_addr = XEXP (DECL_RTL (d), 0);
- TREE_USED (d) = 1;
+ mark_temp_addr_taken (d);
+ structure_value_addr = XEXP (d, 0);
target = 0;
}
}
return temp;
}
- currently_expanding_call++;
+ /* Figure out the amount to which the stack should be aligned. */
+#ifdef PREFERRED_STACK_BOUNDARY
+ preferred_stack_boundary = PREFERRED_STACK_BOUNDARY;
+#else
+ preferred_stack_boundary = STACK_BOUNDARY;
+#endif
+
+ /* Operand 0 is a pointer-to-function; get the type of the function. */
+ funtype = TREE_TYPE (TREE_OPERAND (exp, 0));
+ if (! POINTER_TYPE_P (funtype))
+ abort ();
+ funtype = TREE_TYPE (funtype);
+
+ /* See if this is a call to a function that can return more than once
+ or a call to longjmp or malloc. */
+ flags |= special_function_p (fndecl, flags);
+
+ if (flags & ECF_MAY_BE_ALLOCA)
+ current_function_calls_alloca = 1;
+
+ /* If struct_value_rtx is 0, it means pass the address
+ as if it were an extra parameter. */
+ if (structure_value_addr && struct_value_rtx == 0)
+ {
+ /* If structure_value_addr is a REG other than
+ virtual_outgoing_args_rtx, we can use always use it. If it
+ is not a REG, we must always copy it into a register.
+ If it is virtual_outgoing_args_rtx, we must copy it to another
+ register in some cases. */
+ rtx temp = (GET_CODE (structure_value_addr) != REG
+ || (ACCUMULATE_OUTGOING_ARGS
+ && stack_arg_under_construction
+ && structure_value_addr == virtual_outgoing_args_rtx)
+ ? copy_addr_to_reg (structure_value_addr)
+ : structure_value_addr);
+
+ actparms
+ = tree_cons (error_mark_node,
+ make_tree (build_pointer_type (TREE_TYPE (funtype)),
+ temp),
+ actparms);
+ structure_value_addr_parm = 1;
+ }
+
+ /* Count the arguments and set NUM_ACTUALS. */
+ for (p = actparms, num_actuals = 0; p; p = TREE_CHAIN (p))
+ num_actuals++;
+
+ /* Compute number of named args.
+ Normally, don't include the last named arg if anonymous args follow.
+ We do include the last named arg if STRICT_ARGUMENT_NAMING is nonzero.
+ (If no anonymous args follow, the result of list_length is actually
+ one too large. This is harmless.)
+
+ If PRETEND_OUTGOING_VARARGS_NAMED is set and STRICT_ARGUMENT_NAMING is
+ zero, this machine will be able to place unnamed args that were
+ passed in registers into the stack. So treat all args as named.
+ This allows the insns emitting for a specific argument list to be
+ independent of the function declaration.
+
+ If PRETEND_OUTGOING_VARARGS_NAMED is not set, we do not have any
+ reliable way to pass unnamed args in registers, so we must force
+ them into memory. */
+
+ if ((STRICT_ARGUMENT_NAMING
+ || ! PRETEND_OUTGOING_VARARGS_NAMED)
+ && TYPE_ARG_TYPES (funtype) != 0)
+ n_named_args
+ = (list_length (TYPE_ARG_TYPES (funtype))
+ /* Don't include the last named arg. */
+ - (STRICT_ARGUMENT_NAMING ? 0 : 1)
+ /* Count the struct value address, if it is passed as a parm. */
+ + structure_value_addr_parm);
+ else
+ /* If we know nothing, treat all args as named. */
+ n_named_args = num_actuals;
+
+ /* Start updating where the next arg would go.
+
+ On some machines (such as the PA) indirect calls have a different
+ calling convention than normal calls. The last argument in
+ INIT_CUMULATIVE_ARGS tells the backend if this is an indirect call
+ or not. */
+ INIT_CUMULATIVE_ARGS (args_so_far, funtype, NULL_RTX, (fndecl == 0));
+
+ /* Make a vector to hold all the information about each arg. */
+ args = (struct arg_data *) alloca (num_actuals * sizeof (struct arg_data));
+ memset ((char *) args, 0, num_actuals * sizeof (struct arg_data));
+
+ /* Build up entries in the ARGS array, compute the size of the
+ arguments into ARGS_SIZE, etc. */
+ initialize_argument_information (num_actuals, args, &args_size,
+ n_named_args, actparms, fndecl,
+ &args_so_far, reg_parm_stack_space,
+ &old_stack_level, &old_pending_adj,
+ &must_preallocate, &flags);
+
+ if (args_size.var)
+ {
+ /* If this function requires a variable-sized argument list, don't
+ try to make a cse'able block for this call. We may be able to
+ do this eventually, but it is too complicated to keep track of
+ what insns go in the cse'able block and which don't. */
+
+ flags &= ~(ECF_CONST | ECF_PURE);
+ must_preallocate = 1;
+ }
+
+ /* Now make final decision about preallocating stack space. */
+ must_preallocate = finalize_must_preallocate (must_preallocate,
+ num_actuals, args,
+ &args_size);
+
+ /* If the structure value address will reference the stack pointer, we
+ must stabilize it. We don't need to do this if we know that we are
+ not going to adjust the stack pointer in processing this call. */
+
+ if (structure_value_addr
+ && (reg_mentioned_p (virtual_stack_dynamic_rtx, structure_value_addr)
+ || reg_mentioned_p (virtual_outgoing_args_rtx,
+ structure_value_addr))
+ && (args_size.var
+ || (!ACCUMULATE_OUTGOING_ARGS && args_size.constant)))
+ structure_value_addr = copy_to_reg (structure_value_addr);
/* Tail calls can make things harder to debug, and we're traditionally
pushed these optimizations into -O2. Don't try if we're already
- expanding a call, as that means we're an argument. Similarly, if
- there's pending loops or cleanups we know there's code to follow
- the call.
+ expanding a call, as that means we're an argument. Don't try if
+ there's cleanups, as we know there's code to follow the call.
- If rtx_equal_function_value_matters is false, that means we've
+ If rtx_equal_function_value_matters is false, that means we've
finished with regular parsing. Which means that some of the
machinery we use to generate tail-calls is no longer in place.
This is most often true of sjlj-exceptions, which we couldn't
tail-call to anyway. */
+ if (currently_expanding_call++ != 0
+ || !flag_optimize_sibling_calls
+ || !rtx_equal_function_value_matters
+ || any_pending_cleanups (1)
+ || args_size.var)
+ try_tail_call = try_tail_recursion = 0;
+
+ /* Tail recursion fails, when we are not dealing with recursive calls. */
+ if (!try_tail_recursion
+ || TREE_CODE (TREE_OPERAND (exp, 0)) != ADDR_EXPR
+ || TREE_OPERAND (TREE_OPERAND (exp, 0), 0) != current_function_decl)
+ try_tail_recursion = 0;
+
+ /* Rest of purposes for tail call optimizations to fail. */
+ if (
+#ifdef HAVE_sibcall_epilogue
+ !HAVE_sibcall_epilogue
+#else
+ 1
+#endif
+ || !try_tail_call
+ /* Doing sibling call optimization needs some work, since
+ structure_value_addr can be allocated on the stack.
+ It does not seem worth the effort since few optimizable
+ sibling calls will return a structure. */
+ || structure_value_addr != NULL_RTX
+ /* If the register holding the address is a callee saved
+ register, then we lose. We have no way to prevent that,
+ so we only allow calls to named functions. */
+ /* ??? This could be done by having the insn constraints
+ use a register class that is all call-clobbered. Any
+ reload insns generated to fix things up would appear
+ before the sibcall_epilogue. */
+ || fndecl == NULL_TREE
+ || (flags & (ECF_RETURNS_TWICE | ECF_LONGJMP))
+ || TREE_THIS_VOLATILE (fndecl)
+ || !FUNCTION_OK_FOR_SIBCALL (fndecl)
+ /* If this function requires more stack slots than the current
+ function, we cannot change it into a sibling call. */
+ || args_size.constant > current_function_args_size
+ /* If the callee pops its own arguments, then it must pop exactly
+ the same number of arguments as the current function. */
+ || RETURN_POPS_ARGS (fndecl, funtype, args_size.constant)
+ != RETURN_POPS_ARGS (current_function_decl,
+ TREE_TYPE (current_function_decl),
+ current_function_args_size))
try_tail_call = 0;
- if (flag_optimize_sibling_calls
- && currently_expanding_call == 1
- && rtx_equal_function_value_matters
- && stmt_loop_nest_empty ()
- && ! any_pending_cleanups (1))
- {
- tree new_actparms = NULL_TREE;
+ if (try_tail_call || try_tail_recursion)
+ {
+ int end, inc;
+ actparms = NULL_TREE;
/* Ok, we're going to give the tail call the old college try.
This means we're going to evaluate the function arguments
up to three times. There are two degrees of badness we can
(See unsafe_for_reeval commentary for details.)
Generate a new argument list. Pass safe arguments through
- unchanged. For the easy badness wrap them in UNSAVE_EXPRs.
+ unchanged. For the easy badness wrap them in UNSAVE_EXPRs.
For hard badness, evaluate them now and put their resulting
- rtx in a temporary VAR_DECL. */
+ rtx in a temporary VAR_DECL.
- for (p = actparms; p; p = TREE_CHAIN (p))
- switch (unsafe_for_reeval (TREE_VALUE (p)))
- {
- case 0: /* Safe. */
- new_actparms = tree_cons (TREE_PURPOSE (p), TREE_VALUE (p),
- new_actparms);
- break;
-
- case 1: /* Mildly unsafe. */
- new_actparms = tree_cons (TREE_PURPOSE (p),
- unsave_expr (TREE_VALUE (p)),
- new_actparms);
- break;
-
- case 2: /* Wildly unsafe. */
+ initialize_argument_information has ordered the array for the
+ order to be pushed, and we must remember this when reconstructing
+ the original argument orde. */
+
+ if (PUSH_ARGS_REVERSED)
+ {
+ inc = 1;
+ i = 0;
+ end = num_actuals;
+ }
+ else
+ {
+ inc = -1;
+ i = num_actuals - 1;
+ end = -1;
+ }
+
+ for (; i != end; i += inc)
+ {
+ switch (unsafe_for_reeval (args[i].tree_value))
{
- tree var = build_decl (VAR_DECL, NULL_TREE,
- TREE_TYPE (TREE_VALUE (p)));
- DECL_RTL (var) = expand_expr (TREE_VALUE (p), NULL_RTX,
- VOIDmode, EXPAND_NORMAL);
- new_actparms = tree_cons (TREE_PURPOSE (p), var, new_actparms);
- }
- break;
+ case 0: /* Safe. */
+ break;
- default:
- abort ();
- }
+ case 1: /* Mildly unsafe. */
+ args[i].tree_value = unsave_expr (args[i].tree_value);
+ break;
- /* We built the new argument chain backwards. */
- actparms = nreverse (new_actparms);
+ case 2: /* Wildly unsafe. */
+ {
+ tree var = build_decl (VAR_DECL, NULL_TREE,
+ TREE_TYPE (args[i].tree_value));
+ SET_DECL_RTL (var,
+ expand_expr (args[i].tree_value, NULL_RTX,
+ VOIDmode, EXPAND_NORMAL));
+ args[i].tree_value = var;
+ }
+ break;
+ default:
+ abort ();
+ }
+ /* We need to build actparms for optimize_tail_recursion. We can
+ safely trash away TREE_PURPOSE, since it is unused by this
+ function. */
+ if (try_tail_recursion)
+ actparms = tree_cons (NULL_TREE, args[i].tree_value, actparms);
+ }
/* Expanding one of those dangerous arguments could have added
cleanups, but otherwise give it a whirl. */
- try_tail_call = ! any_pending_cleanups (1);
+ if (any_pending_cleanups (1))
+ try_tail_call = try_tail_recursion = 0;
}
/* Generate a tail recursion sequence when calling ourselves. */
- if (try_tail_call
- && TREE_CODE (TREE_OPERAND (exp, 0)) == ADDR_EXPR
- && TREE_OPERAND (TREE_OPERAND (exp, 0), 0) == current_function_decl)
+ if (try_tail_recursion)
{
/* We want to emit any pending stack adjustments before the tail
recursion "call". That way we know any adjustment after the tail
int save_pending_stack_adjust = pending_stack_adjust;
int save_stack_pointer_delta = stack_pointer_delta;
+ /* Emit any queued insns now; otherwise they would end up in
+ only one of the alternates. */
+ emit_queue ();
+
/* Use a new sequence to hold any RTL we generate. We do not even
know if we will use this RTL yet. The final decision can not be
made until after RTL generation for the entire function is
complete. */
start_sequence ();
-
- /* Emit the pending stack adjustments before we expand any arguments. */
- do_pending_stack_adjust ();
-
+ /* If expanding any of the arguments creates cleanups, we can't
+ do a tailcall. So, we'll need to pop the pending cleanups
+ list. If, however, all goes well, and there are no cleanups
+ then the call to expand_start_target_temps will have no
+ effect. */
+ expand_start_target_temps ();
if (optimize_tail_recursion (actparms, get_last_insn ()))
- tail_recursion_insns = get_insns ();
+ {
+ if (any_pending_cleanups (1))
+ try_tail_call = try_tail_recursion = 0;
+ else
+ tail_recursion_insns = get_insns ();
+ }
+ expand_end_target_temps ();
end_sequence ();
/* Restore the original pending stack adjustment for the sibling and
stack_pointer_delta = save_stack_pointer_delta;
}
- function_call_count++;
-
- if (fndecl && DECL_NAME (fndecl))
- name = IDENTIFIER_POINTER (DECL_NAME (fndecl));
-
- /* Figure out the amount to which the stack should be aligned. */
-#ifdef PREFERRED_STACK_BOUNDARY
- preferred_stack_boundary = PREFERRED_STACK_BOUNDARY;
-#else
- preferred_stack_boundary = STACK_BOUNDARY;
-#endif
- preferred_unit_stack_boundary = preferred_stack_boundary / BITS_PER_UNIT;
+ if (profile_arc_flag && (flags & ECF_FORK_OR_EXEC))
+ {
+ /* A fork duplicates the profile information, and an exec discards
+ it. We can't rely on fork/exec to be paired. So write out the
+ profile information we have gathered so far, and clear it. */
+ /* ??? When Linux's __clone is called with CLONE_VM set, profiling
+ is subject to race conditions, just as with multithreaded
+ programs. */
+
+ emit_library_call (gen_rtx_SYMBOL_REF (Pmode, "__bb_fork_func"), 0,
+ VOIDmode, 0);
+ }
/* Ensure current function's preferred stack boundary is at least
what we need. We don't have to increase alignment for recursive
&& fndecl != current_function_decl)
cfun->preferred_stack_boundary = preferred_stack_boundary;
- /* See if this is a call to a function that can return more than once
- or a call to longjmp or malloc. */
- flags |= special_function_p (fndecl, flags);
-
- if (flags & ECF_MAY_BE_ALLOCA)
- current_function_calls_alloca = 1;
+ preferred_unit_stack_boundary = preferred_stack_boundary / BITS_PER_UNIT;
- /* Operand 0 is a pointer-to-function; get the type of the function. */
- funtype = TREE_TYPE (TREE_OPERAND (exp, 0));
- if (! POINTER_TYPE_P (funtype))
- abort ();
- funtype = TREE_TYPE (funtype);
+ function_call_count++;
/* We want to make two insn chains; one for a sibling call, the other
for a normal call. We will select one of the two chains after
recursion "call". That way we know any adjustment after the tail
recursion call can be ignored if we indeed use the tail recursion
call expansion. */
- int save_pending_stack_adjust;
- int save_stack_pointer_delta;
+ int save_pending_stack_adjust = 0;
+ int save_stack_pointer_delta = 0;
rtx insns;
rtx before_call, next_arg_reg;
if (pass == 0)
{
- /* Various reasons we can not use a sibling call. */
- if (! try_tail_call
-#ifdef HAVE_sibcall_epilogue
- || ! HAVE_sibcall_epilogue
-#else
- || 1
-#endif
- /* The structure value address is used and modified in the
- loop below. It does not seem worth the effort to save and
- restore it as a state variable since few optimizable
- sibling calls will return a structure. */
- || structure_value_addr != NULL_RTX
- /* If the register holding the address is a callee saved
- register, then we lose. We have no way to prevent that,
- so we only allow calls to named functions. */
- /* ??? This could be done by having the insn constraints
- use a register class that is all call-clobbered. Any
- reload insns generated to fix things up would appear
- before the sibcall_epilogue. */
- || fndecl == NULL_TREE
- || ! FUNCTION_OK_FOR_SIBCALL (fndecl))
+ if (! try_tail_call)
continue;
/* Emit any queued insns now; otherwise they would end up in
only one of the alternates. */
emit_queue ();
- /* We know at this point that there are not currently any
- pending cleanups. If, however, in the process of evaluating
- the arguments we were to create some, we'll need to be
- able to get rid of them. */
- expand_start_target_temps ();
-
/* State variables we need to save and restore between
iterations. */
save_pending_stack_adjust = pending_stack_adjust;
argblock = 0;
call_fusage = 0;
- /* Start a new sequence for the normal call case.
+ /* Start a new sequence for the normal call case.
From this point on, if the sibling call fails, we want to set
sibcall_failure instead of continuing the loop. */
start_sequence ();
- /* When calling a const function, we must pop the stack args right away,
- so that the pop is deleted or moved with the call. */
- if (flags & (ECF_CONST | ECF_PURE))
- NO_DEFER_POP;
+ if (pass == 0)
+ {
+ /* We know at this point that there are not currently any
+ pending cleanups. If, however, in the process of evaluating
+ the arguments we were to create some, we'll need to be
+ able to get rid of them. */
+ expand_start_target_temps ();
+ }
/* Don't let pending stack adjusts add up to too much.
Also, do all pending adjustments now if there is any chance
|| pass == 0)
do_pending_stack_adjust ();
- if (profile_arc_flag && (flags & ECF_FORK_OR_EXEC))
- {
- /* A fork duplicates the profile information, and an exec discards
- it. We can't rely on fork/exec to be paired. So write out the
- profile information we have gathered so far, and clear it. */
- /* ??? When Linux's __clone is called with CLONE_VM set, profiling
- is subject to race conditions, just as with multithreaded
- programs. */
-
- emit_library_call (gen_rtx_SYMBOL_REF (Pmode, "__bb_fork_func"), 0,
- VOIDmode, 0);
- }
+ /* When calling a const function, we must pop the stack args right away,
+ so that the pop is deleted or moved with the call. */
+ if (flags & (ECF_CONST | ECF_PURE))
+ NO_DEFER_POP;
/* Push the temporary stack slot level so that we can free any
temporaries we make. */
push_temp_slots ();
- /* Start updating where the next arg would go.
-
- On some machines (such as the PA) indirect calls have a different
- calling convention than normal calls. The last argument in
- INIT_CUMULATIVE_ARGS tells the backend if this is an indirect call
- or not. */
- INIT_CUMULATIVE_ARGS (args_so_far, funtype, NULL_RTX, (fndecl == 0));
-
- /* If struct_value_rtx is 0, it means pass the address
- as if it were an extra parameter. */
- if (structure_value_addr && struct_value_rtx == 0)
- {
- /* If structure_value_addr is a REG other than
- virtual_outgoing_args_rtx, we can use always use it. If it
- is not a REG, we must always copy it into a register.
- If it is virtual_outgoing_args_rtx, we must copy it to another
- register in some cases. */
- rtx temp = (GET_CODE (structure_value_addr) != REG
- || (ACCUMULATE_OUTGOING_ARGS
- && stack_arg_under_construction
- && structure_value_addr == virtual_outgoing_args_rtx)
- ? copy_addr_to_reg (structure_value_addr)
- : structure_value_addr);
-
- actparms
- = tree_cons (error_mark_node,
- make_tree (build_pointer_type (TREE_TYPE (funtype)),
- temp),
- actparms);
- structure_value_addr_parm = 1;
- }
-
- /* Count the arguments and set NUM_ACTUALS. */
- for (p = actparms, i = 0; p; p = TREE_CHAIN (p)) i++;
- num_actuals = i;
-
- /* Compute number of named args.
- Normally, don't include the last named arg if anonymous args follow.
- We do include the last named arg if STRICT_ARGUMENT_NAMING is nonzero.
- (If no anonymous args follow, the result of list_length is actually
- one too large. This is harmless.)
-
- If PRETEND_OUTGOING_VARARGS_NAMED is set and STRICT_ARGUMENT_NAMING is
- zero, this machine will be able to place unnamed args that were
- passed in registers into the stack. So treat all args as named.
- This allows the insns emitting for a specific argument list to be
- independent of the function declaration.
-
- If PRETEND_OUTGOING_VARARGS_NAMED is not set, we do not have any
- reliable way to pass unnamed args in registers, so we must force
- them into memory. */
-
- if ((STRICT_ARGUMENT_NAMING
- || ! PRETEND_OUTGOING_VARARGS_NAMED)
- && TYPE_ARG_TYPES (funtype) != 0)
- n_named_args
- = (list_length (TYPE_ARG_TYPES (funtype))
- /* Don't include the last named arg. */
- - (STRICT_ARGUMENT_NAMING ? 0 : 1)
- /* Count the struct value address, if it is passed as a parm. */
- + structure_value_addr_parm);
- else
- /* If we know nothing, treat all args as named. */
- n_named_args = num_actuals;
-
- /* Make a vector to hold all the information about each arg. */
- args = (struct arg_data *) alloca (num_actuals
- * sizeof (struct arg_data));
- bzero ((char *) args, num_actuals * sizeof (struct arg_data));
-
- /* Build up entries inthe ARGS array, compute the size of the arguments
- into ARGS_SIZE, etc. */
- initialize_argument_information (num_actuals, args, &args_size,
- n_named_args, actparms, fndecl,
- &args_so_far, reg_parm_stack_space,
- &old_stack_level, &old_pending_adj,
- &must_preallocate, &flags);
-
#ifdef FINAL_REG_PARM_STACK_SPACE
reg_parm_stack_space = FINAL_REG_PARM_STACK_SPACE (args_size.constant,
args_size.var);
#endif
-
- if (args_size.var)
- {
- /* If this function requires a variable-sized argument list, don't
- try to make a cse'able block for this call. We may be able to
- do this eventually, but it is too complicated to keep track of
- what insns go in the cse'able block and which don't.
-
- Also do not make a sibling call. */
-
- flags &= ~(ECF_CONST | ECF_PURE);
- must_preallocate = 1;
- sibcall_failure = 1;
- }
-
- if (args_size.constant > current_function_args_size)
- {
- /* If this function requires more stack slots than the current
- function, we cannot change it into a sibling call. */
- sibcall_failure = 1;
- }
-
- /* If the callee pops its own arguments, then it must pop exactly
- the same number of arguments as the current function. */
- if (RETURN_POPS_ARGS (fndecl, funtype, args_size.constant)
- != RETURN_POPS_ARGS (current_function_decl,
- TREE_TYPE (current_function_decl),
- current_function_args_size))
- sibcall_failure = 1;
-
- /* Now make final decision about preallocating stack space. */
- must_preallocate = finalize_must_preallocate (must_preallocate,
- num_actuals, args,
- &args_size);
-
- /* If the structure value address will reference the stack pointer, we
- must stabilize it. We don't need to do this if we know that we are
- not going to adjust the stack pointer in processing this call. */
-
- if (structure_value_addr
- && (reg_mentioned_p (virtual_stack_dynamic_rtx, structure_value_addr)
- || reg_mentioned_p (virtual_outgoing_args_rtx,
- structure_value_addr))
- && (args_size.var
- || (!ACCUMULATE_OUTGOING_ARGS && args_size.constant)))
- structure_value_addr = copy_to_reg (structure_value_addr);
-
/* Precompute any arguments as needed. */
if (pass)
precompute_arguments (flags, num_actuals, args);
if (flags & (ECF_CONST | ECF_PURE | ECF_MALLOC))
start_sequence ();
+ adjusted_args_size = args_size;
/* Compute the actual size of the argument block required. The variable
and constant sizes must be combined, the size may have to be rounded,
and there may be a minimum required size. When generating a sibcall
pattern, do not round up, since we'll be re-using whatever space our
caller provided. */
unadjusted_args_size
- = compute_argument_block_size (reg_parm_stack_space, &args_size,
+ = compute_argument_block_size (reg_parm_stack_space,
+ &adjusted_args_size,
(pass == 0 ? 0
: preferred_stack_boundary));
- old_stack_allocated = stack_pointer_delta - pending_stack_adjust;
+ old_stack_allocated = stack_pointer_delta - pending_stack_adjust;
/* The argument block when performing a sibling call is the
incoming argument block. */
if (pass == 0)
- argblock = virtual_incoming_args_rtx;
+ {
+ argblock = virtual_incoming_args_rtx;
+ stored_args_map = sbitmap_alloc (args_size.constant);
+ sbitmap_zero (stored_args_map);
+ }
/* If we have no actual push instructions, or shouldn't use them,
make space for all args right now. */
- else if (args_size.var != 0)
+ else if (adjusted_args_size.var != 0)
{
if (old_stack_level == 0)
{
old_stack_arg_under_construction = stack_arg_under_construction;
stack_arg_under_construction = 0;
}
- argblock = push_block (ARGS_SIZE_RTX (args_size), 0, 0);
+ argblock = push_block (ARGS_SIZE_RTX (adjusted_args_size), 0, 0);
}
else
{
in the area reserved for register arguments, which may be part of
the stack frame. */
- int needed = args_size.constant;
+ int needed = adjusted_args_size.constant;
/* Store the maximum argument space used. It will be pushed by
the prologue (if ACCUMULATE_OUTGOING_ARGS, or stack overflow
Therefore, we save any area of the stack that was already
written and that we are using. Here we set up to do this
by making a new stack usage map from the old one. The
- actual save will be done by store_one_arg.
+ actual save will be done by store_one_arg.
Another approach might be to try to reorder the argument
evaluations to avoid this conflicting stack usage. */
= (char *) alloca (highest_outgoing_arg_in_use);
if (initial_highest_arg_in_use)
- bcopy (initial_stack_usage_map, stack_usage_map,
- initial_highest_arg_in_use);
+ memcpy (stack_usage_map, initial_stack_usage_map,
+ initial_highest_arg_in_use);
if (initial_highest_arg_in_use != highest_outgoing_arg_in_use)
- bzero (&stack_usage_map[initial_highest_arg_in_use],
+ memset (&stack_usage_map[initial_highest_arg_in_use], 0,
(highest_outgoing_arg_in_use
- initial_highest_arg_in_use));
needed = 0;
/* The address of the outgoing argument list must not be
copied to a register here, because argblock would be left
pointing to the wrong place after the call to
- allocate_dynamic_stack_space below. */
+ allocate_dynamic_stack_space below. */
argblock = virtual_outgoing_args_rtx;
- }
+ }
else
{
if (inhibit_defer_pop == 0)
/* Try to reuse some or all of the pending_stack_adjust
to get this space. */
needed
- = (combine_pending_stack_adjustment_and_call
+ = (combine_pending_stack_adjustment_and_call
(unadjusted_args_size,
- &args_size,
+ &adjusted_args_size,
preferred_unit_stack_boundary));
/* combine_pending_stack_adjustment_and_call computes
do_pending_stack_adjust ();
needed = 0;
}
- else
+ else
/* We need to allocate space. We'll do that in
push_block below. */
pending_stack_adjust = 0;
{
#ifndef OUTGOING_REG_PARM_STACK_SPACE
rtx push_size = GEN_INT (reg_parm_stack_space
- + args_size.constant);
+ + adjusted_args_size.constant);
#else
- rtx push_size = GEN_INT (args_size.constant);
+ rtx push_size = GEN_INT (adjusted_args_size.constant);
#endif
if (old_stack_level == 0)
{
/* Make a new map for the new argument list. */
stack_usage_map = (char *)
alloca (highest_outgoing_arg_in_use);
- bzero (stack_usage_map, highest_outgoing_arg_in_use);
+ memset (stack_usage_map, 0, highest_outgoing_arg_in_use);
highest_outgoing_arg_in_use = 0;
}
allocate_dynamic_stack_space (push_size, NULL_RTX,
/* If we push args individually in reverse order, perform stack alignment
before the first push (the last arg). */
if (PUSH_ARGS_REVERSED && argblock == 0
- && args_size.constant != unadjusted_args_size)
+ && adjusted_args_size.constant != unadjusted_args_size)
{
/* When the stack adjustment is pending, we get better code
by combining the adjustments. */
- if (pending_stack_adjust
+ if (pending_stack_adjust
&& ! (flags & (ECF_CONST | ECF_PURE))
&& ! inhibit_defer_pop)
{
pending_stack_adjust
- = (combine_pending_stack_adjustment_and_call
+ = (combine_pending_stack_adjustment_and_call
(unadjusted_args_size,
- &args_size,
+ &adjusted_args_size,
preferred_unit_stack_boundary));
do_pending_stack_adjust ();
}
else if (argblock == 0)
- anti_adjust_stack (GEN_INT (args_size.constant
+ anti_adjust_stack (GEN_INT (adjusted_args_size.constant
- unadjusted_args_size));
}
/* Now that the stack is properly aligned, pops can't safely
for (i = 0; i < num_actuals; i++)
if (args[i].reg == 0 || args[i].pass_on_stack)
- store_one_arg (&args[i], argblock, flags,
- args_size.var != 0, reg_parm_stack_space);
+ {
+ rtx before_arg = get_last_insn ();
+
+ if (store_one_arg (&args[i], argblock, flags,
+ adjusted_args_size.var != 0,
+ reg_parm_stack_space)
+ || (pass == 0
+ && check_sibcall_argument_overlap (before_arg,
+ &args[i])))
+ sibcall_failure = 1;
+ }
/* If we have a parm that is passed in registers but not in memory
and whose alignment does not permit a direct copy into registers,
if (reg_parm_seen)
for (i = 0; i < num_actuals; i++)
if (args[i].partial != 0 && ! args[i].pass_on_stack)
- store_one_arg (&args[i], argblock, flags,
- args_size.var != 0, reg_parm_stack_space);
+ {
+ rtx before_arg = get_last_insn ();
+
+ if (store_one_arg (&args[i], argblock, flags,
+ adjusted_args_size.var != 0,
+ reg_parm_stack_space)
+ || (pass == 0
+ && check_sibcall_argument_overlap (before_arg,
+ &args[i])))
+ sibcall_failure = 1;
+ }
#ifdef PREFERRED_STACK_BOUNDARY
/* If we pushed args in forward order, perform stack alignment
after pushing the last arg. */
if (!PUSH_ARGS_REVERSED && argblock == 0)
- anti_adjust_stack (GEN_INT (args_size.constant
+ anti_adjust_stack (GEN_INT (adjusted_args_size.constant
- unadjusted_args_size));
#endif
passed in registers. */
#ifdef OUTGOING_REG_PARM_STACK_SPACE
if (!ACCUMULATE_OUTGOING_ARGS
- && must_preallocate == 0 && reg_parm_stack_space > 0)
+ && must_preallocate == 0 && reg_parm_stack_space > 0)
anti_adjust_stack (GEN_INT (reg_parm_stack_space));
#endif
/* Mark the memory for the aggregate as write-only. */
if (current_function_check_memory_usage)
- emit_library_call (chkr_set_right_libfunc, 1,
+ emit_library_call (chkr_set_right_libfunc, LCT_CONST_MAKE_BLOCK,
VOIDmode, 3,
- structure_value_addr, ptr_mode,
+ structure_value_addr, ptr_mode,
GEN_INT (struct_value_size),
TYPE_MODE (sizetype),
GEN_INT (MEMORY_USE_WO),
}
funexp = prepare_call_address (funexp, fndecl, &call_fusage,
- reg_parm_seen);
+ reg_parm_seen, pass == 0);
+
+ load_register_parameters (args, num_actuals, &call_fusage, flags);
- load_register_parameters (args, num_actuals, &call_fusage);
-
/* Perform postincrements before actually calling the function. */
emit_queue ();
/* Generate the actual call instruction. */
emit_call_1 (funexp, fndecl, funtype, unadjusted_args_size,
- args_size.constant, struct_value_size,
+ adjusted_args_size.constant, struct_value_size,
next_arg_reg, valreg, old_inhibit_defer_pop, call_fusage,
flags);
/* Verify that we've deallocated all the stack we used. */
if (pass
- && old_stack_allocated != stack_pointer_delta - pending_stack_adjust)
- abort();
+ && old_stack_allocated != stack_pointer_delta - pending_stack_adjust)
+ abort ();
/* If call is cse'able, make appropriate pair of reg-notes around it.
Test valreg so we don't crash; may safely ignore `const'
gen_rtx_SCRATCH (VOIDmode))), note);
emit_libcall_block (insns, temp, valreg, note);
-
+
valreg = temp;
}
else if (flags & (ECF_CONST | ECF_PURE))
rtx temp = gen_reg_rtx (GET_MODE (valreg));
rtx last, insns;
- /* The return value from a malloc-like function is a pointer. */
+ /* The return value from a malloc-like function is a pointer. */
if (TREE_CODE (TREE_TYPE (exp)) == POINTER_TYPE)
mark_reg_pointer (temp, BIGGEST_ALIGNMENT);
/* The return value from a malloc-like function can not alias
anything else. */
last = get_last_insn ();
- REG_NOTES (last) =
+ REG_NOTES (last) =
gen_rtx_EXPR_LIST (REG_NOALIAS, temp, REG_NOTES (last));
/* Write out the sequence. */
{
emit_note_after (NOTE_INSN_SETJMP, last);
current_function_calls_setjmp = 1;
- sibcall_failure = 1;
}
else
emit_barrier_after (last);
}
if (flags & ECF_LONGJMP)
- current_function_calls_longjmp = 1, sibcall_failure = 1;
+ current_function_calls_longjmp = 1;
/* If this function is returning into a memory location marked as
readonly, it means it is initializing that location. But we normally
{
if (target == 0 || GET_CODE (target) != MEM)
{
- target = gen_rtx_MEM (TYPE_MODE (TREE_TYPE (exp)),
- memory_address (TYPE_MODE (TREE_TYPE (exp)),
- structure_value_addr));
- MEM_SET_IN_STRUCT_P (target,
- AGGREGATE_TYPE_P (TREE_TYPE (exp)));
+ target
+ = gen_rtx_MEM (TYPE_MODE (TREE_TYPE (exp)),
+ memory_address (TYPE_MODE (TREE_TYPE (exp)),
+ structure_value_addr));
+ set_mem_attributes (target, exp, 1);
}
}
else if (pcc_struct_value)
never use this value more than once in one expression. */
target = gen_rtx_MEM (TYPE_MODE (TREE_TYPE (exp)),
copy_to_reg (valreg));
- MEM_SET_IN_STRUCT_P (target, AGGREGATE_TYPE_P (TREE_TYPE (exp)));
+ set_mem_attributes (target, exp, 1);
}
/* Handle calls that return values in multiple non-contiguous locations.
The Irix 6 ABI has examples of this. */
else if (GET_CODE (valreg) == PARALLEL)
{
- int bytes = int_size_in_bytes (TREE_TYPE (exp));
-
if (target == 0)
{
- target = assign_stack_temp (TYPE_MODE (TREE_TYPE (exp)),
- bytes, 0);
- MEM_SET_IN_STRUCT_P (target, AGGREGATE_TYPE_P (TREE_TYPE (exp)));
+ /* This will only be assigned once, so it can be readonly. */
+ tree nt = build_qualified_type (TREE_TYPE (exp),
+ (TYPE_QUALS (TREE_TYPE (exp))
+ | TYPE_QUAL_CONST));
+
+ target = assign_temp (nt, 0, 1, 1);
preserve_temp_slots (target);
}
if (! rtx_equal_p (target, valreg))
- emit_group_store (target, valreg, bytes,
+ emit_group_store (target, valreg,
+ int_size_in_bytes (TREE_TYPE (exp)),
TYPE_ALIGN (TREE_TYPE (exp)));
/* We can not support sibling calls for this case. */
emit_move_insn (target, valreg);
}
else if (TYPE_MODE (TREE_TYPE (exp)) == BLKmode)
- target = copy_blkmode_from_reg (target, valreg, TREE_TYPE (exp));
+ {
+ target = copy_blkmode_from_reg (target, valreg, TREE_TYPE (exp));
+
+ /* We can not support sibling calls for this case. */
+ sibcall_failure = 1;
+ }
else
target = copy_to_reg (valreg);
{
tree type = TREE_TYPE (exp);
int unsignedp = TREE_UNSIGNED (type);
+ int offset = 0;
/* If we don't promote as expected, something is wrong. */
if (GET_MODE (target)
!= promote_mode (type, TYPE_MODE (type), &unsignedp, 1))
abort ();
- target = gen_rtx_SUBREG (TYPE_MODE (type), target, 0);
+ if ((WORDS_BIG_ENDIAN || BYTES_BIG_ENDIAN)
+ && GET_MODE_SIZE (GET_MODE (target))
+ > GET_MODE_SIZE (TYPE_MODE (type)))
+ {
+ offset = GET_MODE_SIZE (GET_MODE (target))
+ - GET_MODE_SIZE (TYPE_MODE (type));
+ if (! BYTES_BIG_ENDIAN)
+ offset = (offset / UNITS_PER_WORD) * UNITS_PER_WORD;
+ else if (! WORDS_BIG_ENDIAN)
+ offset %= UNITS_PER_WORD;
+ }
+ target = gen_rtx_SUBREG (TYPE_MODE (type), target, offset);
SUBREG_PROMOTED_VAR_P (target) = 1;
SUBREG_PROMOTED_UNSIGNED_P (target) = unsignedp;
}
/* If size of args is variable or this was a constructor call for a stack
argument, restore saved stack-pointer value. */
- if (old_stack_level)
+ if (old_stack_level && ! (flags & ECF_SP_DEPRESSED))
{
emit_stack_restore (SAVE_BLOCK, old_stack_level, NULL_RTX);
pending_stack_adjust = old_pending_adj;
stack_usage_map = initial_stack_usage_map;
}
- /* If this was alloca, record the new stack level for nonlocal gotos.
+ /* If this was alloca, record the new stack level for nonlocal gotos.
Check for the handler slots since we might not have a save area
for non-local gotos. */
{
tail_call_insns = insns;
- /* If something prevents making this a sibling call,
- zero out the sequence. */
- if (sibcall_failure)
- tail_call_insns = NULL_RTX;
/* Restore the pending stack adjustment now that we have
finished generating the sibling call sequence. */
pending_stack_adjust = save_pending_stack_adjust;
stack_pointer_delta = save_stack_pointer_delta;
+
+ /* Prepare arg structure for next iteration. */
+ for (i = 0; i < num_actuals; i++)
+ {
+ args[i].value = 0;
+ args[i].aligned_regs = 0;
+ args[i].stack = 0;
+ }
+
+ sbitmap_free (stored_args_map);
}
else
normal_call_insns = insns;
+
+ /* If something prevents making this a sibling call,
+ zero out the sequence. */
+ if (sibcall_failure)
+ tail_call_insns = NULL_RTX;
}
/* The function optimize_sibling_and_tail_recursive_calls doesn't
function who's expansion contains another CALL_PLACEHOLDER.
If there are any C_Ps in any of these sequences, replace them
- with their normal call. */
+ with their normal call. */
for (insn = normal_call_insns; insn; insn = NEXT_INSN (insn))
if (GET_CODE (insn) == CALL_INSN
currently_expanding_call--;
- return target;
-}
-\f
-/* Returns nonzero if FUN is the symbol for a library function which can
- not throw. */
-
-static int
-libfunc_nothrow (fun)
- rtx fun;
-{
- if (fun == throw_libfunc
- || fun == rethrow_libfunc
- || fun == sjthrow_libfunc
- || fun == sjpopnthrow_libfunc)
- return 0;
+ /* If this function returns with the stack pointer depressed, ensure
+ this block saves and restores the stack pointer, show it was
+ changed, and adjust for any outgoing arg space. */
+ if (flags & ECF_SP_DEPRESSED)
+ {
+ clear_pending_stack_adjust ();
+ emit_insn (gen_rtx (CLOBBER, VOIDmode, stack_pointer_rtx));
+ emit_move_insn (virtual_stack_dynamic_rtx, stack_pointer_rtx);
+ save_stack_pointer ();
+ }
- return 1;
+ return target;
}
\f
/* Output a library call to function FUN (a SYMBOL_REF rtx).
- The RETVAL parameter specifies whether return value needs to be saved, other
- parameters are documented in the emit_library_call function bellow. */
+ The RETVAL parameter specifies whether return value needs to be saved, other
+ parameters are documented in the emit_library_call function below. */
static rtx
emit_library_call_value_1 (retval, orgfun, value, fn_type, outmode, nargs, p)
int retval;
rtx orgfun;
rtx value;
- int fn_type;
+ enum libcall_type fn_type;
enum machine_mode outmode;
int nargs;
va_list p;
struct args_size alignment_pad;
rtx argblock = 0;
CUMULATIVE_ARGS args_so_far;
- struct arg { rtx value; enum machine_mode mode; rtx reg; int partial;
- struct args_size offset; struct args_size size; rtx save_area; };
+ struct arg
+ {
+ rtx value;
+ enum machine_mode mode;
+ rtx reg;
+ int partial;
+ struct args_size offset;
+ struct args_size size;
+ rtx save_area;
+ };
struct arg *argvec;
int old_inhibit_defer_pop = inhibit_defer_pop;
rtx call_fusage = 0;
rtx valreg;
int pcc_struct_value = 0;
int struct_value_size = 0;
- int flags = 0;
+ int flags;
int reg_parm_stack_space = 0;
int needed;
+ rtx before_call;
#ifdef REG_PARM_STACK_SPACE
/* Define the boundary of the register parm stack space that needs to be
save, if any. */
int low_to_save = -1, high_to_save = 0;
- rtx save_area = 0; /* Place that it is saved */
+ rtx save_area = 0; /* Place that it is saved. */
#endif
/* Size of the stack reserved for parameter registers. */
#endif
#endif
- if (fn_type == 1)
- flags |= ECF_CONST;
- else if (fn_type == 2)
- flags |= ECF_PURE;
- fun = orgfun;
+ /* By default, library functions can not throw. */
+ flags = ECF_NOTHROW;
- if (libfunc_nothrow (fun))
- flags |= ECF_NOTHROW;
+ switch (fn_type)
+ {
+ case LCT_NORMAL:
+ case LCT_CONST:
+ case LCT_PURE:
+ /* Nothing to do here. */
+ break;
+ case LCT_CONST_MAKE_BLOCK:
+ flags |= ECF_CONST;
+ break;
+ case LCT_PURE_MAKE_BLOCK:
+ flags |= ECF_PURE;
+ break;
+ case LCT_NORETURN:
+ flags |= ECF_NORETURN;
+ break;
+ case LCT_THROW:
+ flags = ECF_NORETURN;
+ break;
+ }
+ fun = orgfun;
#ifdef PREFERRED_STACK_BOUNDARY
/* Ensure current function's preferred stack boundary is at least
if (value != 0 && GET_CODE (value) == MEM)
mem_value = value;
else
- mem_value = assign_stack_temp (outmode, GET_MODE_SIZE (outmode), 0);
+ mem_value = assign_temp (type_for_mode (outmode, 0), 0, 1, 1);
#endif
/* This call returns a big structure. */
library functions shouldn't have many args. */
argvec = (struct arg *) alloca ((nargs + 1) * sizeof (struct arg));
- bzero ((char *) argvec, (nargs + 1) * sizeof (struct arg));
+ memset ((char *) argvec, 0, (nargs + 1) * sizeof (struct arg));
+#ifdef INIT_CUMULATIVE_LIBCALL_ARGS
+ INIT_CUMULATIVE_LIBCALL_ARGS (args_so_far, outmode, fun);
+#else
INIT_CUMULATIVE_ARGS (args_so_far, NULL_TREE, fun, 0);
+#endif
args_size.constant = 0;
args_size.var = 0;
NULL_TREE, &args_size, &argvec[count].offset,
&argvec[count].size, &alignment_pad);
-
if (argvec[count].reg == 0 || argvec[count].partial != 0
|| reg_parm_stack_space > 0)
args_size.constant += argvec[count].size.constant;
#ifdef FUNCTION_ARG_PASS_BY_REFERENCE
if (FUNCTION_ARG_PASS_BY_REFERENCE (args_so_far, mode, NULL_TREE, 1))
{
- /* We do not support FUNCTION_ARG_CALLEE_COPIES here since it can
- be viewed as just an efficiency improvement. */
- rtx slot = assign_stack_temp (mode, GET_MODE_SIZE (mode), 0);
- emit_move_insn (slot, val);
- val = force_operand (XEXP (slot, 0), NULL_RTX);
+ rtx slot;
+ int must_copy = 1
+#ifdef FUNCTION_ARG_CALLEE_COPIES
+ && ! FUNCTION_ARG_CALLEE_COPIES (args_so_far, mode,
+ NULL_TREE, 1)
+#endif
+ ;
+
+ if (GET_MODE (val) == MEM && ! must_copy)
+ slot = val;
+ else if (must_copy)
+ {
+ slot = assign_temp (type_for_mode (mode, 0), 0, 1, 1);
+ emit_move_insn (slot, val);
+ }
+ else
+ {
+ tree type = type_for_mode (mode, 0);
+
+ slot = gen_rtx_MEM (mode,
+ expand_expr (build1 (ADDR_EXPR,
+ build_pointer_type
+ (type),
+ make_tree (type, val)),
+ NULL_RTX, VOIDmode, 0));
+ }
+
+ call_fusage = gen_rtx_EXPR_LIST (VOIDmode,
+ gen_rtx_USE (VOIDmode, slot),
+ call_fusage);
+ if (must_copy)
+ call_fusage = gen_rtx_EXPR_LIST (VOIDmode,
+ gen_rtx_CLOBBER (VOIDmode,
+ slot),
+ call_fusage);
+
mode = Pmode;
+ val = force_operand (XEXP (slot, 0), NULL_RTX);
}
#endif
locate_and_pad_parm (mode, NULL_TREE,
#ifdef STACK_PARMS_IN_REG_PARM_AREA
- 1,
+ 1,
#else
argvec[count].reg != 0,
#endif
stack_usage_map = (char *) alloca (highest_outgoing_arg_in_use);
if (initial_highest_arg_in_use)
- bcopy (initial_stack_usage_map, stack_usage_map,
- initial_highest_arg_in_use);
+ memcpy (stack_usage_map, initial_stack_usage_map,
+ initial_highest_arg_in_use);
if (initial_highest_arg_in_use != highest_outgoing_arg_in_use)
- bzero (&stack_usage_map[initial_highest_arg_in_use],
+ memset (&stack_usage_map[initial_highest_arg_in_use], 0,
highest_outgoing_arg_in_use - initial_highest_arg_in_use);
needed = 0;
- /* The address of the outgoing argument list must not be copied to a
- register here, because argblock would be left pointing to the
- wrong place after the call to allocate_dynamic_stack_space below.
- */
+ /* We must be careful to use virtual regs before they're instantiated,
+ and real regs afterwards. Loop optimization, for example, can create
+ new libcalls after we've instantiated the virtual regs, and if we
+ use virtuals anyway, they won't match the rtl patterns. */
- argblock = virtual_outgoing_args_rtx;
+ if (virtuals_instantiated)
+ argblock = plus_constant (stack_pointer_rtx, STACK_POINTER_OFFSET);
+ else
+ argblock = virtual_outgoing_args_rtx;
}
else
{
for (count = 0; count < reg_parm_stack_space; count++)
#endif
{
- if (count >= highest_outgoing_arg_in_use
+ if (count >= highest_outgoing_arg_in_use
|| stack_usage_map[count] == 0)
continue;
stack_area = gen_rtx_MEM (save_mode,
memory_address (save_mode,
plus_constant (argblock,
- - high_to_save)));
+ -high_to_save)));
#else
stack_area = gen_rtx_MEM (save_mode,
memory_address (save_mode,
}
}
#endif
-
+
/* Push the args that need to be pushed. */
/* ARGNUM indexes the ARGVEC array in the order in which the arguments
if (i != upper_bound)
{
/* We need to make a save area. See what mode we can make
- it. */
+ it. */
enum machine_mode save_mode
= mode_for_size (argvec[argnum].size.constant
* BITS_PER_UNIT,
else
argnum = 0;
- fun = prepare_call_address (fun, NULL_TREE, &call_fusage, 0);
+ fun = prepare_call_address (fun, NULL_TREE, &call_fusage, 0, 0);
/* Now load any reg parms into their regs. */
force_operand (XEXP (mem_value, 0),
NULL_RTX)));
if (GET_CODE (struct_value_rtx) == REG)
- use_reg (&call_fusage, struct_value_rtx);
+ use_reg (&call_fusage, struct_value_rtx);
}
/* Don't allow popping to be deferred, since then
#ifdef PREFERRED_STACK_BOUNDARY
/* Stack must be properly aligned now. */
if (stack_pointer_delta & (PREFERRED_STACK_BOUNDARY / BITS_PER_UNIT - 1))
- abort();
+ abort ();
#endif
+ before_call = get_last_insn ();
+
/* We pass the old value of inhibit_defer_pop + 1 to emit_call_1, which
will set inhibit_defer_pop to that value. */
/* The return type is needed to decide how many bytes the function pops.
always signed. We also assume that the list of arguments passed has
no impact, so we pretend it is unknown. */
- emit_call_1 (fun,
- get_identifier (XSTR (orgfun, 0)),
+ emit_call_1 (fun,
+ get_identifier (XSTR (orgfun, 0)),
build_function_type (outmode == VOIDmode ? void_type_node
: type_for_mode (outmode, 0), NULL_TREE),
- original_args_size.constant, args_size.constant,
+ original_args_size.constant, args_size.constant,
struct_value_size,
FUNCTION_ARG (args_so_far, VOIDmode, void_type_node, 1),
valreg,
old_inhibit_defer_pop + 1, call_fusage, flags);
+ /* For calls to `setjmp', etc., inform flow.c it should complain
+ if nonvolatile values are live. For functions that cannot return,
+ inform flow that control does not fall through. */
+
+ if (flags & (ECF_RETURNS_TWICE | ECF_NORETURN | ECF_LONGJMP))
+ {
+ /* The barrier or NOTE_INSN_SETJMP note must be emitted
+ immediately after the CALL_INSN. Some ports emit more than
+ just a CALL_INSN above, so we must search for it here. */
+
+ rtx last = get_last_insn ();
+ while (GET_CODE (last) != CALL_INSN)
+ {
+ last = PREV_INSN (last);
+ /* There was no CALL_INSN? */
+ if (last == before_call)
+ abort ();
+ }
+
+ if (flags & ECF_RETURNS_TWICE)
+ {
+ emit_note_after (NOTE_INSN_SETJMP, last);
+ current_function_calls_setjmp = 1;
+ }
+ else
+ emit_barrier_after (last);
+ }
+
/* Now restore inhibit_defer_pop to its actual original value. */
OK_DEFER_POP;
PARM_BOUNDARY);
}
#endif
-
+
/* If we saved any argument areas, restore them. */
for (count = 0; count < nargs; count++)
if (argvec[count].save_area)
(use (memory (scratch)). */
void
-emit_library_call VPARAMS((rtx orgfun, int fn_type, enum machine_mode outmode,
- int nargs, ...))
+emit_library_call VPARAMS((rtx orgfun, enum libcall_type fn_type,
+ enum machine_mode outmode, int nargs, ...))
{
#ifndef ANSI_PROTOTYPES
rtx orgfun;
If VALUE is nonzero, VALUE is returned. */
rtx
-emit_library_call_value VPARAMS((rtx orgfun, rtx value, int fn_type,
+emit_library_call_value VPARAMS((rtx orgfun, rtx value,
+ enum libcall_type fn_type,
enum machine_mode outmode, int nargs, ...))
{
#ifndef ANSI_PROTOTYPES
or 0 on a machine where arguments are pushed individually.
MAY_BE_ALLOCA nonzero says this could be a call to `alloca'
- so must be careful about how the stack is used.
+ so must be careful about how the stack is used.
VARIABLE_SIZE nonzero says that this was a variable-sized outgoing
argument stack. This is used if ACCUMULATE_OUTGOING_ARGS to indicate
that we need not worry about saving and restoring the stack.
- FNDECL is the declaration of the function we are calling. */
+ FNDECL is the declaration of the function we are calling.
-static void
-store_one_arg (arg, argblock, flags, variable_size,
- reg_parm_stack_space)
+ Return non-zero if this arg should cause sibcall failure,
+ zero otherwise. */
+
+static int
+store_one_arg (arg, argblock, flags, variable_size, reg_parm_stack_space)
struct arg_data *arg;
rtx argblock;
int flags;
int partial = 0;
int used = 0;
int i, lower_bound = 0, upper_bound = 0;
+ int sibcall_failure = 0;
if (TREE_CODE (pval) == ERROR_MARK)
- return;
+ return 1;
/* Push a new temporary level for any temporaries we make for
this argument. */
if (save_mode == BLKmode)
{
- arg->save_area = assign_stack_temp (BLKmode,
- arg->size.constant, 0);
- MEM_SET_IN_STRUCT_P (arg->save_area,
- AGGREGATE_TYPE_P (TREE_TYPE
- (arg->tree_value)));
+ tree ot = TREE_TYPE (arg->tree_value);
+ tree nt = build_qualified_type (ot, (TYPE_QUALS (ot)
+ | TYPE_QUAL_CONST));
+
+ arg->save_area = assign_temp (nt, 0, 1, 1);
preserve_temp_slots (arg->save_area);
emit_block_move (validize_mem (arg->save_area), stack_area,
- GEN_INT (arg->size.constant),
- PARM_BOUNDARY);
+ expr_size (arg->tree_value),
+ MIN (PARM_BOUNDARY, TYPE_ALIGN (nt)));
}
else
{
here. */
if (arg->n_aligned_regs != 0)
reg = 0;
-
+
/* If this is being passed partially in a register, we can't evaluate
it directly into its stack slot. Otherwise, we can. */
if (arg->value == 0)
/* If the value is already in the stack slot, we are done. */
if (current_function_check_memory_usage && GET_CODE (arg->stack) == MEM)
{
- emit_library_call (chkr_set_right_libfunc, 1, VOIDmode, 3,
- XEXP (arg->stack, 0), Pmode,
+ emit_library_call (chkr_set_right_libfunc, LCT_CONST_MAKE_BLOCK,
+ VOIDmode, 3, XEXP (arg->stack, 0), Pmode,
ARGS_SIZE_RTX (arg->size),
TYPE_MODE (sizetype),
GEN_INT (MEMORY_USE_RW),
/* Argument is a scalar, not entirely passed in registers.
(If part is passed in registers, arg->partial says how much
and emit_push_insn will take care of putting it there.)
-
+
Push it, and if its size is less than the
amount of space allocated to it,
also bump stack pointer by the additional space.
size_rtx = expr_size (pval);
}
+ if ((flags & ECF_SIBCALL) && GET_CODE (arg->value) == MEM)
+ {
+ /* emit_push_insn might not work properly if arg->value and
+ argblock + arg->offset areas overlap. */
+ rtx x = arg->value;
+ int i = 0;
+
+ if (XEXP (x, 0) == current_function_internal_arg_pointer
+ || (GET_CODE (XEXP (x, 0)) == PLUS
+ && XEXP (XEXP (x, 0), 0) ==
+ current_function_internal_arg_pointer
+ && GET_CODE (XEXP (XEXP (x, 0), 1)) == CONST_INT))
+ {
+ if (XEXP (x, 0) != current_function_internal_arg_pointer)
+ i = INTVAL (XEXP (XEXP (x, 0), 1));
+
+ /* expand_call should ensure this */
+ if (arg->offset.var || GET_CODE (size_rtx) != CONST_INT)
+ abort ();
+
+ if (arg->offset.constant > i)
+ {
+ if (arg->offset.constant < i + INTVAL (size_rtx))
+ sibcall_failure = 1;
+ }
+ else if (arg->offset.constant < i)
+ {
+ if (i < arg->offset.constant + INTVAL (size_rtx))
+ sibcall_failure = 1;
+ }
+ }
+ }
+
+ /* If parm is passed both in stack and in register and offset is
+ greater than reg_parm_stack_space, split the offset. */
+ if (arg->reg && arg->pass_on_stack)
+ {
+ if (arg->offset.constant < reg_parm_stack_space && arg->offset.var)
+ error ("variable offset is passed paritially in stack and in reg");
+ else if (arg->offset.constant < reg_parm_stack_space && arg->size.var)
+ error ("variable size is passed partially in stack and in reg");
+ else if (arg->offset.constant < reg_parm_stack_space
+ && ((arg->offset.constant + arg->size.constant)
+ > reg_parm_stack_space))
+ {
+ rtx size_rtx1 = GEN_INT (reg_parm_stack_space - arg->offset.constant);
+ emit_push_insn (arg->value, arg->mode, TREE_TYPE (pval), size_rtx1,
+ TYPE_ALIGN (TREE_TYPE (pval)), partial, reg,
+ excess, argblock, ARGS_SIZE_RTX (arg->offset),
+ reg_parm_stack_space,
+ ARGS_SIZE_RTX (arg->alignment_pad));
+
+ size_rtx = GEN_INT (INTVAL(size_rtx) - reg_parm_stack_space);
+ }
+ }
+
+
emit_push_insn (arg->value, arg->mode, TREE_TYPE (pval), size_rtx,
TYPE_ALIGN (TREE_TYPE (pval)), partial, reg, excess,
argblock, ARGS_SIZE_RTX (arg->offset),
ARGS_SIZE_RTX (arg->alignment_pad));
}
-
/* Unless this is a partially-in-register argument, the argument is now
- in the stack.
+ in the stack.
??? Note that this can change arg->value from arg->stack to
arg->stack_slot and it matters when they are not the same.
preserve_temp_slots (NULL_RTX);
free_temp_slots ();
pop_temp_slots ();
+
+ return sibcall_failure;
}
OSDN Git Service
