/* 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"
+#include "timevar.h"
+#include "sbitmap.h"
+
+#if !defined FUNCTION_OK_FOR_SIBCALL
+#define FUNCTION_OK_FOR_SIBCALL(DECL) 1
+#endif
#if !defined PREFERRED_STACK_BOUNDARY && defined STACK_BOUNDARY
#define PREFERRED_STACK_BOUNDARY STACK_BOUNDARY
#ifdef PUSH_ROUNDING
#if defined (STACK_GROWS_DOWNWARD) != defined (ARGS_GROW_DOWNWARD)
-#define PUSH_ARGS_REVERSED /* If it's last to first */
+#define PUSH_ARGS_REVERSED PUSH_ARGS
+#endif
+
#endif
+#ifndef PUSH_ARGS_REVERSED
+#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. */
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;
differ from STACK if this arg pads downward. This location is known
to be aligned to FUNCTION_ARG_BOUNDARY. */
rtx stack_slot;
-#ifdef ACCUMULATE_OUTGOING_ARGS
/* Place that this stack area has been saved, if needed. */
rtx save_area;
-#endif
/* If an argument's alignment does not permit direct copying into registers,
copy in smaller-sized pieces into pseudos. These are stored in a
block pointed to by this field. The next field says how many
struct args_size alignment_pad;
};
-#ifdef ACCUMULATE_OUTGOING_ARGS
/* A vector of one char per byte of stack space. A byte if non-zero if
the corresponding stack location has been used.
This vector is used to prevent a function call within an argument from
/* 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
to make sure the object being constructed does not overlap the
argument list for the constructor call. */
int stack_arg_under_construction;
-#endif
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. */
+#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. */
+#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 a function that won't throw an exception. */
+#define ECF_NOTHROW 16
+/* Nonzero if this is a call to setjmp or a related function. */
+#define ECF_RETURNS_TWICE 32
+/* Nonzero if this is a call to `longjmp'. */
+#define ECF_LONGJMP 64
+/* Nonzero if this is a syscall that makes a new process in the image of
+ the current one. */
+#define ECF_FORK_OR_EXEC 128
+#define ECF_SIBCALL 256
+/* 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,
- rtx, int, rtx, int, int));
+ rtx, int, rtx, int));
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));
static int finalize_must_preallocate PARAMS ((int, int,
struct arg_data *,
struct args_size *));
-static void precompute_arguments PARAMS ((int, int, int,
- struct arg_data *,
- struct args_size *));
-static int compute_argument_block_size PARAMS ((int,
+static void precompute_arguments PARAMS ((int, int,
+ struct arg_data *));
+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));
+ 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));
-#if defined(ACCUMULATE_OUTGOING_ARGS) && defined(REG_PARM_STACK_SPACE)
+#ifdef REG_PARM_STACK_SPACE
static rtx save_fixed_argument_area PARAMS ((int, rtx, int *, int *));
static void restore_fixed_argument_area PARAMS ((rtx, rtx, int, int));
#endif
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)
return 1;
- /* Only expressions and references can contain calls. */
- if (type != 'e' && type != '<' && type != '1' && type != '2' && type != 'r'
- && type != 'b')
- return 0;
-
switch (code)
{
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);
+ == FUNCTION_DECL)
+ && (special_function_p (TREE_OPERAND (TREE_OPERAND (exp, 0), 0),
+ 0)
+ & ECF_MAY_BE_ALLOCA))
+ return 1;
+
+ break;
- if ((DECL_BUILT_IN (fndecl)
- && DECL_BUILT_IN_CLASS (fndecl) == BUILT_IN_NORMAL
- && DECL_FUNCTION_CODE (fndecl) == BUILT_IN_ALLOCA)
- || (DECL_SAVED_INSNS (fndecl)
- && DECL_SAVED_INSNS (fndecl)->calls_alloca))
+ 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;
- }
+ }
- /* Third operand is RTL. */
- length = 2;
- break;
+ return 0;
case SAVE_EXPR:
if (SAVE_EXPR_RTL (exp) != 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 0;
- case METHOD_CALL_EXPR:
- length = 3;
- break;
-
- case WITH_CLEANUP_EXPR:
- length = 1;
- break;
-
- case RTL_EXPR:
+ case TREE_LIST:
+ for (; exp != 0; exp = TREE_CHAIN (exp))
+ if (calls_function_1 (TREE_VALUE (exp), which))
+ return 1;
return 0;
-
+
default:
break;
}
+ /* 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++)
if (TREE_OPERAND (exp, i) != 0
&& calls_function_1 (TREE_OPERAND (exp, i), which))
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.
-
- IS_CONST is true if this is a `const' call. */
+ denote registers used by the called function. */
static void
emit_call_1 (funexp, fndecl, funtype, stack_size, rounded_stack_size,
struct_value_size, next_arg_reg, valreg, old_inhibit_defer_pop,
- call_fusage, is_const, nothrow)
+ call_fusage, ecf_flags)
rtx funexp;
tree fndecl ATTRIBUTE_UNUSED;
tree funtype ATTRIBUTE_UNUSED;
rtx valreg;
int old_inhibit_defer_pop;
rtx call_fusage;
- int is_const, nothrow;
+ 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;
-#ifndef ACCUMULATE_OUTGOING_ARGS
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,
if (GET_CODE (funexp) != SYMBOL_REF)
funexp = memory_address (FUNCTION_MODE, funexp);
-#ifndef ACCUMULATE_OUTGOING_ARGS
+#if defined (HAVE_sibcall_pop) && defined (HAVE_sibcall_value_pop)
+ if ((ecf_flags & ECF_SIBCALL)
+ && HAVE_sibcall_pop && HAVE_sibcall_value_pop
+ && (n_popped > 0 || stack_size == 0))
+ {
+ 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,
+ 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),
+ rounded_stack_size_rtx, next_arg_reg, n_pop);
+
+ emit_call_insn (pat);
+ already_popped = 1;
+ }
+ else
+#endif
+
#if defined (HAVE_call_pop) && defined (HAVE_call_value_pop)
/* If the target has "call" or "call_value" insns, then prefer them
if no arguments are actually popped. If the target does not have
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);
}
else
#endif
+
+#if defined (HAVE_sibcall) && defined (HAVE_sibcall_value)
+ if ((ecf_flags & ECF_SIBCALL)
+ && HAVE_sibcall && HAVE_sibcall_value)
+ {
+ if (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),
+ rounded_stack_size_rtx, next_arg_reg,
+ struct_value_size_rtx));
+ }
+ else
#endif
#if defined (HAVE_call) && defined (HAVE_call_value)
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));
}
if (! call_insn)
abort ();
+ /* Mark memory as used for "pure" function call. */
+ if (ecf_flags & ECF_PURE)
+ {
+ call_fusage = gen_rtx_EXPR_LIST (VOIDmode,
+ gen_rtx_USE (VOIDmode,
+ gen_rtx_MEM (BLKmode,
+ gen_rtx_SCRATCH (VOIDmode))), call_fusage);
+ }
+
/* Put the register usage information on the CALL. If there is already
some usage information, put ours at the end. */
if (CALL_INSN_FUNCTION_USAGE (call_insn))
CALL_INSN_FUNCTION_USAGE (call_insn) = call_fusage;
/* If this is a const call, then set the insn's unchanging bit. */
- if (is_const)
+ if (ecf_flags & (ECF_CONST | ECF_PURE))
CONST_CALL_P (call_insn) = 1;
/* If this call can't throw, attach a REG_EH_REGION reg note to that
effect. */
- if (nothrow)
+ if (ecf_flags & ECF_NOTHROW)
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 the context of the call as a whole permits. */
inhibit_defer_pop = old_inhibit_defer_pop;
-#ifndef ACCUMULATE_OUTGOING_ARGS
- /* If returning from the subroutine does not automatically pop the args,
- we need an instruction to pop them sooner or later.
- Perhaps do it now; perhaps just record how much space to pop later.
-
- If returning from the subroutine does pop the args, indicate that the
- stack pointer will be changed. */
-
- /* The space for the args is no longer waiting for the call; either it
- was popped by the call, or it'll be popped below. */
- arg_space_so_far -= rounded_stack_size;
-
if (n_popped > 0)
{
if (!already_popped)
CALL_INSN_FUNCTION_USAGE (call_insn));
rounded_stack_size -= n_popped;
rounded_stack_size_rtx = GEN_INT (rounded_stack_size);
+ stack_pointer_delta -= n_popped;
}
- if (rounded_stack_size != 0)
+ if (!ACCUMULATE_OUTGOING_ARGS)
{
- if (flag_defer_pop && inhibit_defer_pop == 0 && !is_const)
- pending_stack_adjust += rounded_stack_size;
- else
- adjust_stack (rounded_stack_size_rtx);
+ /* If returning from the subroutine does not automatically pop the args,
+ we need an instruction to pop them sooner or later.
+ Perhaps do it now; perhaps just record how much space to pop later.
+
+ If returning from the subroutine does pop the args, indicate that the
+ stack pointer will be changed. */
+
+ if (rounded_stack_size != 0 && ! (ecf_flags & ECF_SP_DEPRESSED))
+ {
+ if (flag_defer_pop && inhibit_defer_pop == 0
+ && ! (ecf_flags & (ECF_CONST | ECF_PURE)))
+ pending_stack_adjust += rounded_stack_size;
+ else
+ adjust_stack (rounded_stack_size_rtx);
+ }
}
-#endif
+ /* When we accumulate outgoing args, we must avoid any stack manipulations.
+ Restore the stack pointer to its original value now. Usually
+ ACCUMULATE_OUTGOING_ARGS targets don't get here, but there are exceptions.
+ On i386 ACCUMULATE_OUTGOING_ARGS can be enabled on demand, and
+ popping variants of functions exist as well.
+
+ ??? 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)
+ anti_adjust_stack (GEN_INT (n_popped));
}
/* Determine if the function identified by NAME and FNDECL is one with
For example, if the function might return more than one time (setjmp), then
set RETURNS_TWICE to a nonzero value.
- Similarly set IS_LONGJMP for if the function is in the longjmp family.
+ Similarly set LONGJMP for if the function is in the longjmp family.
- Set IS_MALLOC for any of the standard memory allocation functions which
+ Set MALLOC for any of the standard memory allocation functions which
allocate from the heap.
Set MAY_BE_ALLOCA for any memory allocation function that might allocate
space from the stack such as alloca. */
-void
-special_function_p (fndecl, returns_twice, is_longjmp, fork_or_exec,
- is_malloc, may_be_alloca)
+static int
+special_function_p (fndecl, flags)
tree fndecl;
- int *returns_twice;
- int *is_longjmp;
- int *fork_or_exec;
- int *is_malloc;
- int *may_be_alloca;
+ int flags;
{
- *returns_twice = 0;
- *is_longjmp = 0;
- *fork_or_exec = 0;
- *may_be_alloca = 0;
-
- /* The function decl may have the `malloc' attribute. */
- *is_malloc = fndecl && DECL_IS_MALLOC (fndecl);
-
- if (! *is_malloc
+ if (! (flags & ECF_MALLOC)
&& fndecl && DECL_NAME (fndecl)
&& IDENTIFIER_LENGTH (DECL_NAME (fndecl)) <= 17
/* Exclude functions not at the file scope, or not `extern',
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
anything that does not understand its behavior. */
- *may_be_alloca
- = (((IDENTIFIER_LENGTH (DECL_NAME (fndecl)) == 6
- && name[0] == 'a'
- && ! strcmp (name, "alloca"))
- || (IDENTIFIER_LENGTH (DECL_NAME (fndecl)) == 16
- && name[0] == '_'
- && ! strcmp (name, "__builtin_alloca"))));
+ if (((IDENTIFIER_LENGTH (DECL_NAME (fndecl)) == 6
+ && name[0] == 'a'
+ && ! strcmp (name, "alloca"))
+ || (IDENTIFIER_LENGTH (DECL_NAME (fndecl)) == 16
+ && name[0] == '_'
+ && ! strcmp (name, "__builtin_alloca"))))
+ flags |= ECF_MAY_BE_ALLOCA;
/* Disregard prefix _, __ or __x. */
if (name[0] == '_')
if (tname[0] == 's')
{
- *returns_twice
- = ((tname[1] == 'e'
- && (! strcmp (tname, "setjmp")
- || ! strcmp (tname, "setjmp_syscall")))
- || (tname[1] == 'i'
- && ! strcmp (tname, "sigsetjmp"))
- || (tname[1] == 'a'
- && ! strcmp (tname, "savectx")));
+ if ((tname[1] == 'e'
+ && (! strcmp (tname, "setjmp")
+ || ! strcmp (tname, "setjmp_syscall")))
+ || (tname[1] == 'i'
+ && ! strcmp (tname, "sigsetjmp"))
+ || (tname[1] == 'a'
+ && ! strcmp (tname, "savectx")))
+ flags |= ECF_RETURNS_TWICE;
+
if (tname[1] == 'i'
&& ! strcmp (tname, "siglongjmp"))
- *is_longjmp = 1;
+ flags |= ECF_LONGJMP;
}
else if ((tname[0] == 'q' && tname[1] == 's'
&& ! strcmp (tname, "qsetjmp"))
|| (tname[0] == 'v' && tname[1] == 'f'
&& ! strcmp (tname, "vfork")))
- *returns_twice = 1;
+ flags |= ECF_RETURNS_TWICE;
else if (tname[0] == 'l' && tname[1] == 'o'
&& ! strcmp (tname, "longjmp"))
- *is_longjmp = 1;
+ flags |= ECF_LONGJMP;
else if ((tname[0] == 'f' && tname[1] == 'o'
&& ! strcmp (tname, "fork"))
&& (tname[5] == '\0'
|| ((tname[5] == 'p' || tname[5] == 'e')
&& tname[6] == '\0'))))
- *fork_or_exec = 1;
+ flags |= ECF_FORK_OR_EXEC;
/* Do not add any more malloc-like functions to this list,
instead mark them as malloc functions using the malloc attribute.
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")
|| ! strcmp (tname, "strdup")))
- *is_malloc = 1;
+ flags |= ECF_MALLOC;
+ }
+ return flags;
+}
+
+/* Return nonzero when tree represent call to longjmp. */
+
+int
+setjmp_call_p (fndecl)
+ tree fndecl;
+{
+ return special_function_p (fndecl, 0) & ECF_RETURNS_TWICE;
+}
+
+/* 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))
+ {
+ /* The function exp may have the `malloc' attribute. */
+ if (DECL_P (exp) && DECL_IS_MALLOC (exp))
+ flags |= ECF_MALLOC;
+
+ /* The function exp may have the `pure' attribute. */
+ if (DECL_P (exp) && DECL_IS_PURE (exp))
+ flags |= ECF_PURE;
+
+ if (TREE_NOTHROW (exp))
+ flags |= ECF_NOTHROW;
}
+
+ if (TREE_READONLY (exp) && ! TREE_THIS_VOLATILE (exp))
+ flags |= ECF_CONST;
+
+ if (TREE_THIS_VOLATILE (exp))
+ flags |= ECF_NORETURN;
+
+ return flags;
}
/* Precompute all register parameters as described by ARGS, storing values
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);
}
}
-#if defined(ACCUMULATE_OUTGOING_ARGS) && defined(REG_PARM_STACK_SPACE)
+#ifdef REG_PARM_STACK_SPACE
/* 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,
if (save_mode == BLKmode)
{
save_area = assign_stack_temp (BLKmode, num_to_save, 0);
- /* Cannot use emit_block_move here because it can be done by a library
- call which in turn gets into this place again and deadly infinite
- recursion happens. */
+ /* Cannot use emit_block_move here because it can be done by a
+ library call which in turn gets into this place again and deadly
+ infinite recursion happens. */
move_by_pieces (validize_mem (save_area), stack_area, num_to_save,
- PARM_BOUNDARY / BITS_PER_UNIT);
+ PARM_BOUNDARY);
}
else
{
call which in turn gets into this place again and deadly infinite
recursion happens. */
move_by_pieces (stack_area, validize_mem (save_area),
- high_to_save - low_to_save + 1,
- PARM_BOUNDARY / BITS_PER_UNIT);
+ 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
bytes -= bitsize / BITS_PER_UNIT;
store_bit_field (reg, bitsize, big_endian_correction, word_mode,
- extract_bit_field (word, bitsize, 0, 1,
- NULL_RTX, word_mode,
- word_mode,
- bitalign / BITS_PER_UNIT,
+ extract_bit_field (word, bitsize, 0, 1, NULL_RTX,
+ word_mode, word_mode, bitalign,
BITS_PER_WORD),
- bitalign / BITS_PER_UNIT, BITS_PER_WORD);
+ bitalign, BITS_PER_WORD);
}
}
}
/* Fill in ARGS_SIZE and ARGS array based on the parameters found in
- ACTPARMS.
+ ACTPARMS.
NUM_ACTUALS is the total number of parameters.
OLD_STACK_LEVEL is a pointer to an rtx which olds the old stack level
and may be modified by this routine.
- OLD_PENDING_ADJ, MUST_PREALLOCATE and IS_CONST are pointers to integer
+ OLD_PENDING_ADJ, MUST_PREALLOCATE and FLAGS are pointers to integer
flags which may may be modified by this routine. */
static void
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, is_const)
+ old_pending_adj, must_preallocate,
+ ecf_flags)
int num_actuals ATTRIBUTE_UNUSED;
struct arg_data *args;
struct args_size *args_size;
rtx *old_stack_level;
int *old_pending_adj;
int *must_preallocate;
- int *is_const;
+ int *ecf_flags;
{
/* 1 if scanning parms front to back, -1 if scanning back to front. */
int inc;
struct args_size alignment_pad;
int i;
tree p;
-
+
args_size->constant = 0;
args_size->var = 0;
We fill up ARGS from the front or from the back if necessary
so that in any case the first arg to be pushed ends up at the front. */
-#ifdef PUSH_ARGS_REVERSED
- i = num_actuals - 1, inc = -1;
- /* In this case, must reverse order of args
- so that we compute and push the last arg first. */
-#else
- i = 0, inc = 1;
-#endif
+ if (PUSH_ARGS_REVERSED)
+ {
+ i = num_actuals - 1, inc = -1;
+ /* In this case, must reverse order of args
+ so that we compute and push the last arg first. */
+ }
+ else
+ {
+ i = 0, inc = 1;
+ }
/* I counts args in order (to be) pushed; ARGPOS counts in order written. */
for (p = actparms, argpos = 0; p; p = TREE_CHAIN (p), i += inc, argpos++)
args[i].tree_value = TREE_VALUE (p);
/* Replace erroneous argument with constant zero. */
- if (type == error_mark_node || TYPE_SIZE (type) == 0)
+ if (type == error_mark_node || !COMPLETE_TYPE_P (type))
args[i].tree_value = integer_zero_node, type = integer_type_node;
/* 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
- && (TREE_CODE_CLASS (TREE_CODE (TREE_OPERAND
- (args[i].tree_value, 1)))
- == 'd')
+ && (DECL_P (TREE_OPERAND (args[i].tree_value, 1)))
&& ! REG_P (DECL_RTL (TREE_OPERAND (args[i].tree_value, 1))))
args[i].tree_value = TREE_OPERAND (args[i].tree_value, 1);
function being called. */
rtx copy;
- if (TYPE_SIZE (type) == 0
+ if (!COMPLETE_TYPE_P (type)
|| TREE_CODE (TYPE_SIZE (type)) != INTEGER_CST
|| (flag_stack_check && ! STACK_CHECK_BUILTIN
&& (0 < compare_tree_int (TYPE_SIZE_UNIT (type),
}
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);
- *is_const = 0;
+ *ecf_flags &= ~(ECF_CONST | ECF_PURE);
args[i].tree_value = build1 (ADDR_EXPR,
build_pointer_type (type),
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. */
+ 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
+
#ifdef FUNCTION_ARG_PARTIAL_NREGS
if (args[i].reg)
args[i].partial
/* If this is an addressable type, we cannot pre-evaluate it. Thus,
we cannot consider this function call constant. */
if (TREE_ADDRESSABLE (type))
- *is_const = 0;
+ *ecf_flags &= ~(ECF_CONST | ECF_PURE);
/* Compute the stack-size of this argument. */
if (args[i].reg == 0 || args[i].partial != 0
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;
{
int unadjusted_args_size = args_size->constant;
+ /* For accumulate outgoing args mode we don't need to align, since the frame
+ will be already aligned. Align to STACK_BOUNDARY in order to prevent
+ backends from generating missaligned frame sizes. */
+#ifdef STACK_BOUNDARY
+ if (ACCUMULATE_OUTGOING_ARGS && preferred_stack_boundary > STACK_BOUNDARY)
+ preferred_stack_boundary = STACK_BOUNDARY;
+#endif
+
/* 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. */
#ifdef PREFERRED_STACK_BOUNDARY
preferred_stack_boundary /= BITS_PER_UNIT;
if (preferred_stack_boundary > 1)
- args_size->var = round_up (args_size->var, preferred_stack_boundary);
+ {
+ /* We don't handle this case yet. To handle it correctly we have
+ 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 ();
+ args_size->var = round_up (args_size->var, preferred_stack_boundary);
+ }
#endif
if (reg_parm_stack_space > 0)
{
#ifdef PREFERRED_STACK_BOUNDARY
preferred_stack_boundary /= BITS_PER_UNIT;
+ if (preferred_stack_boundary < 1)
+ preferred_stack_boundary = 1;
args_size->constant = (((args_size->constant
- + arg_space_so_far
- + pending_stack_adjust
+ + stack_pointer_delta
+ preferred_stack_boundary - 1)
/ preferred_stack_boundary
* preferred_stack_boundary)
- - arg_space_so_far
- - pending_stack_adjust);
+ - stack_pointer_delta);
#endif
args_size->constant = MAX (args_size->constant,
/* Precompute parameters as needed for a function call.
- IS_CONST indicates the target function is a pure function.
-
- MUST_PREALLOCATE indicates that we must preallocate stack space for
- any stack arguments.
+ FLAGS is mask of ECF_* constants.
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_SIZE contains information about the size of the arg list. */
+ 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 (is_const, must_preallocate, num_actuals, args, args_size)
- int is_const;
- int must_preallocate;
+precompute_arguments (flags, num_actuals, args)
+ int flags;
int num_actuals;
struct arg_data *args;
- struct args_size *args_size;
{
int i;
on the stack, then we must precompute any parameter which contains a
function call which will store arguments on the stack.
Otherwise, evaluating the parameter may clobber previous parameters
- which have already been stored into the stack. */
+ which have already been stored into the stack. (we have code to avoid
+ such case by saving the ougoing stack arguments, but it results in
+ worse code) */
for (i = 0; i < num_actuals; i++)
- if (is_const
- || ((args_size->var != 0 || args_size->constant != 0)
- && calls_function (args[i].tree_value, 1))
- || (must_preallocate
- && (args_size->var != 0 || args_size->constant != 0)
- && calls_function (args[i].tree_value, 0)))
+ 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;
locations. The Irix 6 ABI has examples of this. */
if (GET_CODE (reg) == PARALLEL)
- {
- emit_group_load (reg, args[i].value,
- int_size_in_bytes (TREE_TYPE (args[i].tree_value)),
- (TYPE_ALIGN (TREE_TYPE (args[i].tree_value))
- / BITS_PER_UNIT));
- }
+ emit_group_load (reg, args[i].value,
+ int_size_in_bytes (TREE_TYPE (args[i].tree_value)),
+ TYPE_ALIGN (TREE_TYPE (args[i].tree_value)));
/* If simple case, just do move. If normal partial, store_one_arg
has already loaded the register for us. In all other cases,
}
}
-/* 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.
- If the value is stored in TARGET then TARGET is returned.
- If IGNORE is nonzero, then we ignore the value of the function call. */
+/* Try to integrate function. See expand_inline_function for documentation
+ about the parameters. */
-rtx
-expand_call (exp, target, ignore)
- tree exp;
+static rtx
+try_to_integrate (fndecl, actparms, target, ignore, type, structure_value_addr)
+ tree fndecl;
+ tree actparms;
rtx target;
int ignore;
+ tree type;
+ rtx structure_value_addr;
{
- /* List of actual parameters. */
- tree actparms = TREE_OPERAND (exp, 1);
- /* RTX for the function to be called. */
- rtx funexp;
- /* Data type of the function. */
- tree funtype;
- /* Declaration of the function being called,
- or 0 if the function is computed (not known by name). */
- tree fndecl = 0;
- char *name = 0;
+ rtx temp;
rtx before_call;
+ int i;
+ rtx old_stack_level = 0;
+ int reg_parm_stack_space = 0;
- /* Register in which non-BLKmode value will be returned,
- or 0 if no value or if value is BLKmode. */
- rtx valreg;
- /* Address where we should return a BLKmode value;
- 0 if value not BLKmode. */
- rtx structure_value_addr = 0;
- /* Nonzero if that address is being passed by treating it as
- an extra, implicit first parameter. Otherwise,
- it is passed by being copied directly into struct_value_rtx. */
- int structure_value_addr_parm = 0;
- /* Size of aggregate value wanted, or zero if none wanted
- or if we are using the non-reentrant PCC calling convention
- or expecting the value in registers. */
- HOST_WIDE_INT struct_value_size = 0;
- /* Nonzero if called function returns an aggregate in memory PCC style,
- by returning the address of where to find it. */
- int pcc_struct_value = 0;
-
- /* Number of actual parameters in this call, including struct value addr. */
- int num_actuals;
- /* Number of named args. Args after this are anonymous ones
- and they must all go on the stack. */
- int n_named_args;
-
- /* Vector of information about each argument.
- Arguments are numbered in the order they will be pushed,
- not the order they are written. */
- struct arg_data *args;
-
- /* Total size in bytes of all the stack-parms scanned so far. */
- struct args_size args_size;
- /* Size of arguments before any adjustments (such as rounding). */
- int unadjusted_args_size;
- /* Data on reg parms scanned so far. */
- CUMULATIVE_ARGS args_so_far;
- /* Nonzero if a reg parm has been scanned. */
- 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.
- 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
- ignore PUSH_ROUNDING in that case). */
-
-#ifdef PUSH_ROUNDING
- int must_preallocate = 0;
+#ifdef REG_PARM_STACK_SPACE
+#ifdef MAYBE_REG_PARM_STACK_SPACE
+ reg_parm_stack_space = MAYBE_REG_PARM_STACK_SPACE;
#else
- int must_preallocate = 1;
+ reg_parm_stack_space = REG_PARM_STACK_SPACE (fndecl);
#endif
-
- /* Size of the stack reserved for parameter registers. */
- int reg_parm_stack_space = 0;
-
- /* Address of space preallocated for stack parms
- (on machines that lack push insns), or 0 if space not preallocated. */
- rtx argblock = 0;
-
- /* Nonzero if it is plausible that this is a call to alloca. */
- int may_be_alloca;
- /* Nonzero if this is a call to malloc or a related function. */
- int is_malloc;
- /* Nonzero if this is a call to setjmp or a related function. */
- int returns_twice;
- /* Nonzero if this is a call to `longjmp'. */
- int is_longjmp;
- /* Nonzero if this is a syscall that makes a new process in the image of
- the current one. */
- int fork_or_exec;
- /* Nonzero if this is a call to an inline function. */
- int is_integrable = 0;
- /* Nonzero if this is a call to a `const' function.
- Note that only explicitly named functions are handled as `const' here. */
- int is_const = 0;
- /* Nonzero if this is a call to a `volatile' function. */
- int is_volatile = 0;
- /* Nonzero if this is a call to a function that won't throw an exception. */
- int nothrow = TREE_NOTHROW (exp);
-#if defined(ACCUMULATE_OUTGOING_ARGS) && defined(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;
- rtx save_area = 0; /* Place that it is saved */
#endif
-#ifdef ACCUMULATE_OUTGOING_ARGS
- int initial_highest_arg_in_use = highest_outgoing_arg_in_use;
- char *initial_stack_usage_map = stack_usage_map;
- int old_stack_arg_under_construction = 0;
-#endif
+ before_call = get_last_insn ();
- rtx old_stack_level = 0;
- int old_pending_adj = 0;
- int old_inhibit_defer_pop = inhibit_defer_pop;
- rtx call_fusage = 0;
- register tree p;
- register int i;
-#ifdef PREFERRED_STACK_BOUNDARY
- int preferred_stack_boundary = PREFERRED_STACK_BOUNDARY;
-#else
- /* In this case preferred_stack_boundary variable is meaningless.
- It is used only in order to keep ifdef noise down when calling
- compute_argument_block_size. */
- int preferred_stack_boundary = 0;
-#endif
+ timevar_push (TV_INTEGRATION);
- /* The value of the function call can be put in a hard register. But
- if -fcheck-memory-usage, code which invokes functions (and thus
- damages some hard registers) can be inserted before using the value.
- So, target is always a pseudo-register in that case. */
- if (current_function_check_memory_usage)
- target = 0;
+ temp = expand_inline_function (fndecl, actparms, target,
+ ignore, type,
+ structure_value_addr);
- /* See if we can find a DECL-node for the actual function.
- As a result, decide whether this is a call to an integrable function. */
+ timevar_pop (TV_INTEGRATION);
- p = TREE_OPERAND (exp, 0);
- if (TREE_CODE (p) == ADDR_EXPR)
+ /* If inlining succeeded, return. */
+ if (temp != (rtx) (HOST_WIDE_INT) - 1)
{
- fndecl = TREE_OPERAND (p, 0);
- if (TREE_CODE (fndecl) != FUNCTION_DECL)
- fndecl = 0;
- else
+ if (ACCUMULATE_OUTGOING_ARGS)
{
- if (!flag_no_inline
- && fndecl != current_function_decl
- && DECL_INLINE (fndecl)
- && DECL_SAVED_INSNS (fndecl)
- && DECL_SAVED_INSNS (fndecl)->inlinable)
- is_integrable = 1;
- else if (! TREE_ADDRESSABLE (fndecl))
- {
- /* In case this function later becomes inlinable,
- record that there was already a non-inline call to it.
-
- Use abstraction instead of setting TREE_ADDRESSABLE
- directly. */
- if (DECL_INLINE (fndecl) && warn_inline && !flag_no_inline
- && optimize > 0)
- {
- warning_with_decl (fndecl, "can't inline call to `%s'");
- warning ("called from here");
- }
- mark_addressable (fndecl);
- }
-
- if (TREE_READONLY (fndecl) && ! TREE_THIS_VOLATILE (fndecl)
- && TYPE_MODE (TREE_TYPE (exp)) != VOIDmode)
- is_const = 1;
-
- if (TREE_THIS_VOLATILE (fndecl))
- is_volatile = 1;
-
- if (TREE_NOTHROW (fndecl))
- nothrow = 1;
- }
- }
-
- /* If we don't have specific function to call, see if we have a
- constant or `noreturn' function from the type. */
- if (fndecl == 0)
- {
- is_const = TREE_READONLY (TREE_TYPE (TREE_TYPE (p)));
- is_volatile = TREE_THIS_VOLATILE (TREE_TYPE (TREE_TYPE (p)));
- }
-
-#ifdef REG_PARM_STACK_SPACE
-#ifdef MAYBE_REG_PARM_STACK_SPACE
- reg_parm_stack_space = MAYBE_REG_PARM_STACK_SPACE;
-#else
- reg_parm_stack_space = REG_PARM_STACK_SPACE (fndecl);
-#endif
-#endif
-
-#if defined(PUSH_ROUNDING) && ! defined(OUTGOING_REG_PARM_STACK_SPACE)
- if (reg_parm_stack_space > 0)
- must_preallocate = 1;
-#endif
-
- /* Warn if this value is an aggregate type,
- regardless of which calling convention we are using for it. */
- if (warn_aggregate_return && AGGREGATE_TYPE_P (TREE_TYPE (exp)))
- warning ("function call has aggregate value");
-
- /* Set up a place to return a structure. */
-
- /* Cater to broken compilers. */
- if (aggregate_value_p (exp))
- {
- /* This call returns a big structure. */
- is_const = 0;
-
-#ifdef PCC_STATIC_STRUCT_RETURN
- {
- pcc_struct_value = 1;
- /* Easier than making that case work right. */
- if (is_integrable)
- {
- /* In case this is a static function, note that it has been
- used. */
- if (! TREE_ADDRESSABLE (fndecl))
- mark_addressable (fndecl);
- is_integrable = 0;
- }
- }
-#else /* not PCC_STATIC_STRUCT_RETURN */
- {
- struct_value_size = int_size_in_bytes (TREE_TYPE (exp));
-
- if (target && GET_CODE (target) == MEM)
- 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. */
-
- 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;
- target = 0;
- }
- }
-#endif /* not PCC_STATIC_STRUCT_RETURN */
- }
-
- /* If called function is inline, try to integrate it. */
-
- if (is_integrable)
- {
- rtx temp;
-
-#ifdef ACCUMULATE_OUTGOING_ARGS
- before_call = get_last_insn ();
-#endif
-
- temp = expand_inline_function (fndecl, actparms, target,
- ignore, TREE_TYPE (exp),
- structure_value_addr);
-
- /* If inlining succeeded, return. */
- if (temp != (rtx) (HOST_WIDE_INT) -1)
- {
-#ifdef ACCUMULATE_OUTGOING_ARGS
/* If the outgoing argument list must be preserved, push
the stack before executing the inlined function if it
makes any calls. */
rtx insn = NULL_RTX, seq;
/* Look for a call in the inline function code.
- If DECL_SAVED_INSNS (fndecl)->outgoing_args_size is
- nonzero then there is a call and it is not necessary
- to scan the insns. */
+ If DECL_SAVED_INSNS (fndecl)->outgoing_args_size is
+ nonzero then there is a call and it is not necessary
+ to scan the insns. */
if (DECL_SAVED_INSNS (fndecl)->outgoing_args_size == 0)
for (insn = first_insn; insn; insn = NEXT_INSN (insn))
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
emit_stack_restore (SAVE_BLOCK, old_stack_level, NULL_RTX);
}
}
-#endif
-
- /* If the result is equivalent to TARGET, return TARGET to simplify
- checks in store_expr. They can be equivalent but not equal in the
- case of a function that returns BLKmode. */
- if (temp != target && rtx_equal_p (temp, target))
- return target;
- return temp;
}
- /* If inlining failed, mark FNDECL as needing to be compiled
- separately after all. If function was declared inline,
- give a warning. */
- if (DECL_INLINE (fndecl) && warn_inline && !flag_no_inline
- && optimize > 0 && ! TREE_ADDRESSABLE (fndecl))
- {
- warning_with_decl (fndecl, "inlining failed in call to `%s'");
- warning ("called from here");
- }
- mark_addressable (fndecl);
+ /* If the result is equivalent to TARGET, return TARGET to simplify
+ checks in store_expr. They can be equivalent but not equal in the
+ case of a function that returns BLKmode. */
+ if (temp != target && rtx_equal_p (temp, target))
+ return target;
+ return temp;
}
- function_call_count++;
-
- if (fndecl && DECL_NAME (fndecl))
- name = IDENTIFIER_POINTER (DECL_NAME (fndecl));
-
- /* Ensure current function's preferred stack boundary is at least
- what we need. We don't have to increase alignment for recursive
- functions. */
- if (cfun->preferred_stack_boundary < preferred_stack_boundary
- && 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. */
- special_function_p (fndecl, &returns_twice, &is_longjmp, &fork_or_exec,
- &is_malloc, &may_be_alloca);
-
- if (may_be_alloca)
- current_function_calls_alloca = 1;
-
- /* 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);
-
- /* 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 (is_const)
- NO_DEFER_POP;
-
- /* Don't let pending stack adjusts add up to too much.
- Also, do all pending adjustments now
- if there is any chance this might be a call to alloca. */
-
- if (pending_stack_adjust >= 32
- || (pending_stack_adjust > 0 && may_be_alloca))
- do_pending_stack_adjust ();
-
- if (profile_arc_flag && fork_or_exec)
+ /* If inlining failed, mark FNDECL as needing to be compiled
+ separately after all. If function was declared inline,
+ give a warning. */
+ if (DECL_INLINE (fndecl) && warn_inline && !flag_no_inline
+ && optimize > 0 && !TREE_ADDRESSABLE (fndecl))
{
- /* 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. */
- emit_library_call (gen_rtx_SYMBOL_REF (Pmode, "__bb_fork_func"), 0,
- VOIDmode, 0);
-
- /* ??? When __clone is called with CLONE_VM set, profiling is
- subject to race conditions, just as with multithreaded programs. */
+ warning_with_decl (fndecl, "inlining failed in call to `%s'");
+ warning ("called from here");
}
+ mark_addressable (fndecl);
+ return (rtx) (HOST_WIDE_INT) - 1;
+}
- /* 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));
+/* We need to pop PENDING_STACK_ADJUST bytes. But, if the arguments
+ wouldn't fill up an even multiple of PREFERRED_UNIT_STACK_BOUNDARY
+ bytes, then we would need to push some additional bytes to pad the
+ arguments. So, we compute an adjust to the stack pointer for an
+ amount that will leave the stack under-aligned by UNADJUSTED_ARGS_SIZE
+ bytes. Then, when the arguments are pushed the stack will be perfectly
+ aligned. ARGS_SIZE->CONSTANT is set to the number of bytes that should
+ be popped after the call. Returns the adjustment. */
- /* 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)
+static int
+combine_pending_stack_adjustment_and_call (unadjusted_args_size,
+ args_size,
+ preferred_unit_stack_boundary)
+ int unadjusted_args_size;
+ struct args_size *args_size;
+ int preferred_unit_stack_boundary;
+{
+ /* The number of bytes to pop so that the stack will be
+ under-aligned by UNADJUSTED_ARGS_SIZE bytes. */
+ HOST_WIDE_INT adjustment;
+ /* The alignment of the stack after the arguments are pushed, if we
+ just pushed the arguments without adjust the stack here. */
+ HOST_WIDE_INT unadjusted_alignment;
+
+ unadjusted_alignment
+ = ((stack_pointer_delta + unadjusted_args_size)
+ % preferred_unit_stack_boundary);
+
+ /* We want to get rid of as many of the PENDING_STACK_ADJUST bytes
+ as possible -- leaving just enough left to cancel out the
+ UNADJUSTED_ALIGNMENT. In other words, we want to ensure that the
+ PENDING_STACK_ADJUST is non-negative, and congruent to
+ -UNADJUSTED_ALIGNMENT modulo the PREFERRED_UNIT_STACK_BOUNDARY. */
+
+ /* Begin by trying to pop all the bytes. */
+ 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 (preferred_unit_stack_boundary > 1)
{
- /* 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
-#ifdef ACCUMULATE_OUTGOING_ARGS
- || (stack_arg_under_construction
- && structure_value_addr == virtual_outgoing_args_rtx)
-#endif
- ? 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;
+ if (unadjusted_alignment > 0)
+ adjustment -= preferred_unit_stack_boundary - unadjusted_alignment;
+ else
+ adjustment += unadjusted_alignment;
}
- /* 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.)
+ /* 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
+ = pending_stack_adjust - adjustment + unadjusted_args_size;
- 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.
+ return adjustment;
+}
- 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. */
+/* 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. */
- 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;
+static int
+check_sibcall_argument_overlap_1 (x)
+ rtx x;
+{
+ RTX_CODE code;
+ int i, j;
+ unsigned int k;
+ const char *fmt;
- /* 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));
+ if (x == NULL_RTX)
+ return 0;
- /* 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,
- &is_const);
+ code = GET_CODE (x);
-#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 (code == MEM)
{
- /* 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. */
-
- is_const = 0;
- must_preallocate = 1;
- }
-
- /* 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. */
- unadjusted_args_size
- = compute_argument_block_size (reg_parm_stack_space, &args_size,
- preferred_stack_boundary);
-
- /* 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 (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;
- 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
-#ifndef ACCUMULATE_OUTGOING_ARGS
- || args_size.constant
+#ifdef ARGS_GROW_DOWNWARD
+ i = -i - GET_MODE_SIZE (GET_MODE (x));
#endif
- ))
- structure_value_addr = copy_to_reg (structure_value_addr);
- /* Precompute any arguments as needed. */
- precompute_arguments (is_const, must_preallocate, num_actuals,
- args, &args_size);
-
- /* Now we are about to start emitting insns that can be deleted
- if a libcall is deleted. */
- if (is_const || is_malloc)
- start_sequence ();
+ 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;
- /* If we have no actual push instructions, or shouldn't use them,
- make space for all args right now. */
+ return 0;
+ }
- if (args_size.var != 0)
+ /* Scan all subexpressions. */
+ fmt = GET_RTX_FORMAT (code);
+ for (i = 0; i < GET_RTX_LENGTH (code); i++, fmt++)
{
- if (old_stack_level == 0)
+ if (*fmt == 'e')
{
- emit_stack_save (SAVE_BLOCK, &old_stack_level, NULL_RTX);
- old_pending_adj = pending_stack_adjust;
- pending_stack_adjust = 0;
-#ifdef ACCUMULATE_OUTGOING_ARGS
- /* stack_arg_under_construction says whether a stack arg is
- being constructed at the old stack level. Pushing the stack
- gets a clean outgoing argument block. */
- old_stack_arg_under_construction = stack_arg_under_construction;
- stack_arg_under_construction = 0;
-#endif
+ 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;
}
- argblock = push_block (ARGS_SIZE_RTX (args_size), 0, 0);
}
- else
- {
- /* Note that we must go through the motions of allocating an argument
- block even if the size is zero because we may be storing args
- in the area reserved for register arguments, which may be part of
- the stack frame. */
-
- int needed = args_size.constant;
-
- /* Store the maximum argument space used. It will be pushed by
- the prologue (if ACCUMULATE_OUTGOING_ARGS, or stack overflow
- checking). */
-
- if (needed > current_function_outgoing_args_size)
- current_function_outgoing_args_size = needed;
+ return 0;
+}
- if (must_preallocate)
- {
-#ifdef ACCUMULATE_OUTGOING_ARGS
- /* Since the stack pointer will never be pushed, it is possible for
- the evaluation of a parm to clobber something we have already
- written to the stack. Since most function calls on RISC machines
- do not use the stack, this is uncommon, but must work correctly.
+/* 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. */
- 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.
+static int
+check_sibcall_argument_overlap (insn, arg)
+ rtx insn;
+ struct arg_data *arg;
+{
+ int low, high;
- Another approach might be to try to reorder the argument
- evaluations to avoid this conflicting stack usage. */
+ if (insn == NULL_RTX)
+ insn = get_insns ();
+ else
+ insn = NEXT_INSN (insn);
-#ifndef OUTGOING_REG_PARM_STACK_SPACE
- /* Since we will be writing into the entire argument area, the
- map must be allocated for its entire size, not just the part that
- is the responsibility of the caller. */
- needed += reg_parm_stack_space;
-#endif
+ for (; insn; insn = NEXT_INSN (insn))
+ if (INSN_P (insn)
+ && check_sibcall_argument_overlap_1 (PATTERN (insn)))
+ break;
#ifdef ARGS_GROW_DOWNWARD
- highest_outgoing_arg_in_use = MAX (initial_highest_arg_in_use,
- needed + 1);
+ low = -arg->offset.constant - arg->size.constant;
#else
- highest_outgoing_arg_in_use = MAX (initial_highest_arg_in_use,
- needed);
+ low = arg->offset.constant;
#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);
- if (initial_highest_arg_in_use != highest_outgoing_arg_in_use)
- bzero (&stack_usage_map[initial_highest_arg_in_use],
- highest_outgoing_arg_in_use - initial_highest_arg_in_use);
- needed = 0;
+ for (high = low + arg->size.constant; low < high; low++)
+ SET_BIT (stored_args_map, low);
+ return insn != NULL_RTX;
+}
- /* 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.
- */
+/* 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.
+ If the value is stored in TARGET then TARGET is returned.
+ If IGNORE is nonzero, then we ignore the value of the function call. */
- argblock = virtual_outgoing_args_rtx;
+rtx
+expand_call (exp, target, ignore)
+ tree exp;
+ rtx target;
+ int ignore;
+{
+ /* Nonzero if we are currently expanding a call. */
+ static int currently_expanding_call = 0;
-#else /* not ACCUMULATE_OUTGOING_ARGS */
- if (inhibit_defer_pop == 0)
- {
- /* Try to reuse some or all of the pending_stack_adjust
- to get this space. Maybe we can avoid any pushing. */
- if (needed > pending_stack_adjust)
- {
- needed -= pending_stack_adjust;
- pending_stack_adjust = 0;
- }
- else
- {
- pending_stack_adjust -= needed;
- needed = 0;
- }
- }
- /* Special case this because overhead of `push_block' in this
- case is non-trivial. */
- if (needed == 0)
- argblock = virtual_outgoing_args_rtx;
- else
- argblock = push_block (GEN_INT (needed), 0, 0);
-
- /* We only really need to call `copy_to_reg' in the case where push
- insns are going to be used to pass ARGBLOCK to a function
- call in ARGS. In that case, the stack pointer changes value
- from the allocation point to the call point, and hence
- the value of VIRTUAL_OUTGOING_ARGS_RTX changes as well.
- But might as well always do it. */
- argblock = copy_to_reg (argblock);
-#endif /* not ACCUMULATE_OUTGOING_ARGS */
- }
- }
+ /* List of actual parameters. */
+ tree actparms = TREE_OPERAND (exp, 1);
+ /* RTX for the function to be called. */
+ rtx funexp;
+ /* Sequence of insns to perform a tail recursive "call". */
+ rtx tail_recursion_insns = NULL_RTX;
+ /* Sequence of insns to perform a normal "call". */
+ rtx normal_call_insns = NULL_RTX;
+ /* Sequence of insns to perform a tail recursive "call". */
+ rtx tail_call_insns = NULL_RTX;
+ /* Data type of the function. */
+ tree funtype;
+ /* Declaration of the function being called,
+ or 0 if the function is computed (not known by name). */
+ tree fndecl = 0;
+ rtx insn;
+ int try_tail_call = 1;
+ int try_tail_recursion = 1;
+ int pass;
-#ifdef ACCUMULATE_OUTGOING_ARGS
- /* The save/restore code in store_one_arg handles all cases except one:
- a constructor call (including a C function returning a BLKmode struct)
- to initialize an argument. */
- if (stack_arg_under_construction)
- {
-#ifndef OUTGOING_REG_PARM_STACK_SPACE
- rtx push_size = GEN_INT (reg_parm_stack_space + args_size.constant);
-#else
- rtx push_size = GEN_INT (args_size.constant);
-#endif
- if (old_stack_level == 0)
- {
- emit_stack_save (SAVE_BLOCK, &old_stack_level, NULL_RTX);
- old_pending_adj = pending_stack_adjust;
- pending_stack_adjust = 0;
- /* stack_arg_under_construction says whether a stack arg is
- being constructed at the old stack level. Pushing the stack
- gets a clean outgoing argument block. */
- old_stack_arg_under_construction = stack_arg_under_construction;
- stack_arg_under_construction = 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);
- highest_outgoing_arg_in_use = 0;
- }
- allocate_dynamic_stack_space (push_size, NULL_RTX, BITS_PER_UNIT);
- }
- /* If argument evaluation might modify the stack pointer, copy the
- address of the argument list to a register. */
- for (i = 0; i < num_actuals; i++)
- if (args[i].pass_on_stack)
- {
- argblock = copy_addr_to_reg (argblock);
- break;
- }
+ /* Register in which non-BLKmode value will be returned,
+ or 0 if no value or if value is BLKmode. */
+ rtx valreg;
+ /* Address where we should return a BLKmode value;
+ 0 if value not BLKmode. */
+ rtx structure_value_addr = 0;
+ /* Nonzero if that address is being passed by treating it as
+ an extra, implicit first parameter. Otherwise,
+ it is passed by being copied directly into struct_value_rtx. */
+ int structure_value_addr_parm = 0;
+ /* Size of aggregate value wanted, or zero if none wanted
+ or if we are using the non-reentrant PCC calling convention
+ or expecting the value in registers. */
+ HOST_WIDE_INT struct_value_size = 0;
+ /* Nonzero if called function returns an aggregate in memory PCC style,
+ by returning the address of where to find it. */
+ int pcc_struct_value = 0;
+
+ /* Number of actual parameters in this call, including struct value addr. */
+ int num_actuals;
+ /* Number of named args. Args after this are anonymous ones
+ and they must all go on the stack. */
+ int n_named_args;
+
+ /* Vector of information about each argument.
+ Arguments are numbered in the order they will be pushed,
+ not the order they are written. */
+ struct arg_data *args;
+
+ /* 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. */
+ CUMULATIVE_ARGS args_so_far;
+ /* Nonzero if a reg parm has been scanned. */
+ 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.
+ 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
+ ignore PUSH_ROUNDING in that case). */
+
+ int must_preallocate = !PUSH_ARGS;
+
+ /* Size of the stack reserved for parameter registers. */
+ int reg_parm_stack_space = 0;
+
+ /* Address of space preallocated for stack parms
+ (on machines that lack push insns), or 0 if space not preallocated. */
+ rtx argblock = 0;
+
+ /* Mask of ECF_ flags. */
+ int flags = 0;
+ /* Nonzero if this is a call to an inline function. */
+ int is_integrable = 0;
+#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;
+ rtx save_area = 0; /* Place that it is saved */
#endif
- compute_argument_addresses (args, argblock, num_actuals);
+ int initial_highest_arg_in_use = highest_outgoing_arg_in_use;
+ char *initial_stack_usage_map = stack_usage_map;
+ int old_stack_arg_under_construction = 0;
-#ifdef PUSH_ARGS_REVERSED
-#ifdef PREFERRED_STACK_BOUNDARY
- /* If we push args individually in reverse order, perform stack alignment
- before the first push (the last arg). */
- if (args_size.constant != unadjusted_args_size)
+ rtx old_stack_level = 0;
+ int old_pending_adj = 0;
+ int old_inhibit_defer_pop = inhibit_defer_pop;
+ int old_stack_allocated;
+ rtx call_fusage;
+ register tree p = TREE_OPERAND (exp, 0);
+ register int i;
+ /* The alignment of the stack, in bits. */
+ HOST_WIDE_INT preferred_stack_boundary;
+ /* The alignment of the stack, in bytes. */
+ HOST_WIDE_INT preferred_unit_stack_boundary;
+
+ /* The value of the function call can be put in a hard register. But
+ if -fcheck-memory-usage, code which invokes functions (and thus
+ damages some hard registers) can be inserted before using the value.
+ So, target is always a pseudo-register in that case. */
+ if (current_function_check_memory_usage)
+ target = 0;
+
+ /* See if this is "nothrow" function call. */
+ if (TREE_NOTHROW (exp))
+ flags |= ECF_NOTHROW;
+
+ /* See if we can find a DECL-node for the actual function.
+ As a result, decide whether this is a call to an integrable function. */
+
+ fndecl = get_callee_fndecl (exp);
+ if (fndecl)
{
- /* When the stack adjustment is pending,
- we get better code by combining the adjustments. */
- if (pending_stack_adjust && !is_const
- && !inhibit_defer_pop)
+ if (!flag_no_inline
+ && fndecl != current_function_decl
+ && DECL_INLINE (fndecl)
+ && DECL_SAVED_INSNS (fndecl)
+ && DECL_SAVED_INSNS (fndecl)->inlinable)
+ is_integrable = 1;
+ else if (! TREE_ADDRESSABLE (fndecl))
{
- args_size.constant = (unadjusted_args_size
- + ((pending_stack_adjust + args_size.constant
- + arg_space_so_far
- - unadjusted_args_size)
- % (preferred_stack_boundary / BITS_PER_UNIT)));
- pending_stack_adjust -= args_size.constant - unadjusted_args_size;
- do_pending_stack_adjust ();
+ /* In case this function later becomes inlinable,
+ record that there was already a non-inline call to it.
+
+ Use abstraction instead of setting TREE_ADDRESSABLE
+ directly. */
+ if (DECL_INLINE (fndecl) && warn_inline && !flag_no_inline
+ && optimize > 0)
+ {
+ warning_with_decl (fndecl, "can't inline call to `%s'");
+ warning ("called from here");
+ }
+ mark_addressable (fndecl);
}
- else if (argblock == 0)
- anti_adjust_stack (GEN_INT (args_size.constant - unadjusted_args_size));
- arg_space_so_far += args_size.constant - unadjusted_args_size;
- /* Now that the stack is properly aligned, pops can't safely
- be deferred during the evaluation of the arguments. */
- NO_DEFER_POP;
+ flags |= flags_from_decl_or_type (fndecl);
}
+
+ /* 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))))
+ {
+ flags |= ECF_SP_DEPRESSED;
+ flags &= ~(ECF_PURE | ECF_CONST);
+ }
+
+#ifdef REG_PARM_STACK_SPACE
+#ifdef MAYBE_REG_PARM_STACK_SPACE
+ reg_parm_stack_space = MAYBE_REG_PARM_STACK_SPACE;
+#else
+ reg_parm_stack_space = REG_PARM_STACK_SPACE (fndecl);
#endif
#endif
- /* Don't try to defer pops if preallocating, not even from the first arg,
- since ARGBLOCK probably refers to the SP. */
- if (argblock)
- NO_DEFER_POP;
+#ifndef OUTGOING_REG_PARM_STACK_SPACE
+ if (reg_parm_stack_space > 0 && PUSH_ARGS)
+ must_preallocate = 1;
+#endif
- funexp = rtx_for_function_call (fndecl, exp);
+ /* Warn if this value is an aggregate type,
+ regardless of which calling convention we are using for it. */
+ if (warn_aggregate_return && AGGREGATE_TYPE_P (TREE_TYPE (exp)))
+ warning ("function call has aggregate value");
- /* Figure out the register where the value, if any, will come back. */
- valreg = 0;
- if (TYPE_MODE (TREE_TYPE (exp)) != VOIDmode
- && ! structure_value_addr)
+ /* Set up a place to return a structure. */
+
+ /* Cater to broken compilers. */
+ if (aggregate_value_p (exp))
{
- if (pcc_struct_value)
- valreg = hard_function_value (build_pointer_type (TREE_TYPE (exp)),
- fndecl, 0);
- else
- valreg = hard_function_value (TREE_TYPE (exp), fndecl, 0);
- }
+ /* This call returns a big structure. */
+ flags &= ~(ECF_CONST | ECF_PURE);
- /* Precompute all register parameters. It isn't safe to compute anything
- once we have started filling any specific hard regs. */
- precompute_register_parameters (num_actuals, args, ®_parm_seen);
+#ifdef PCC_STATIC_STRUCT_RETURN
+ {
+ pcc_struct_value = 1;
+ /* Easier than making that case work right. */
+ if (is_integrable)
+ {
+ /* In case this is a static function, note that it has been
+ used. */
+ if (! TREE_ADDRESSABLE (fndecl))
+ mark_addressable (fndecl);
+ is_integrable = 0;
+ }
+ }
+#else /* not PCC_STATIC_STRUCT_RETURN */
+ {
+ struct_value_size = int_size_in_bytes (TREE_TYPE (exp));
-#if defined(ACCUMULATE_OUTGOING_ARGS) && defined(REG_PARM_STACK_SPACE)
+ if (target && GET_CODE (target) == MEM)
+ structure_value_addr = XEXP (target, 0);
+ else
+ {
+ /* 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);
- /* Save the fixed argument area if it's part of the caller's frame and
- is clobbered by argument setup for this call. */
- save_area = save_fixed_argument_area (reg_parm_stack_space, argblock,
- &low_to_save, &high_to_save);
-#endif
-
+ mark_temp_addr_taken (d);
+ structure_value_addr = XEXP (d, 0);
+ target = 0;
+ }
+ }
+#endif /* not PCC_STATIC_STRUCT_RETURN */
+ }
- /* Now store (and compute if necessary) all non-register parms.
- These come before register parms, since they can require block-moves,
- which could clobber the registers used for register parms.
- Parms which have partial registers are not stored here,
- but we do preallocate space here if they want that. */
+ /* If called function is inline, try to integrate it. */
- for (i = 0; i < num_actuals; i++)
- if (args[i].reg == 0 || args[i].pass_on_stack)
- store_one_arg (&args[i], argblock, may_be_alloca,
- args_size.var != 0, reg_parm_stack_space);
-
- /* 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,
- make a group of pseudos that correspond to each register that we
- will later fill. */
- if (STRICT_ALIGNMENT)
- store_unaligned_arguments_into_pseudos (args, num_actuals);
-
- /* Now store any partially-in-registers parm.
- This is the last place a block-move can happen. */
- 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, may_be_alloca,
- args_size.var != 0, reg_parm_stack_space);
+ if (is_integrable)
+ {
+ rtx temp = try_to_integrate (fndecl, actparms, target,
+ ignore, TREE_TYPE (exp),
+ structure_value_addr);
+ if (temp != (rtx) (HOST_WIDE_INT) - 1)
+ return temp;
+ }
-#ifndef PUSH_ARGS_REVERSED
+ /* Figure out the amount to which the stack should be aligned. */
#ifdef PREFERRED_STACK_BOUNDARY
- /* If we pushed args in forward order, perform stack alignment
- after pushing the last arg. */
- if (argblock == 0)
- anti_adjust_stack (GEN_INT (args_size.constant - unadjusted_args_size));
-#endif
-#endif
-
- /* If register arguments require space on the stack and stack space
- was not preallocated, allocate stack space here for arguments
- passed in registers. */
-#if ! defined(ACCUMULATE_OUTGOING_ARGS) && defined(OUTGOING_REG_PARM_STACK_SPACE)
- if (must_preallocate == 0 && reg_parm_stack_space > 0)
- anti_adjust_stack (GEN_INT (reg_parm_stack_space));
+ preferred_stack_boundary = PREFERRED_STACK_BOUNDARY;
+#else
+ preferred_stack_boundary = STACK_BOUNDARY;
#endif
- /* Pass the function the address in which to return a structure value. */
- if (structure_value_addr && ! structure_value_addr_parm)
- {
- emit_move_insn (struct_value_rtx,
- force_reg (Pmode,
- force_operand (structure_value_addr,
- NULL_RTX)));
+ /* 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);
- /* Mark the memory for the aggregate as write-only. */
- if (current_function_check_memory_usage)
- emit_library_call (chkr_set_right_libfunc, 1,
- VOIDmode, 3,
- structure_value_addr, Pmode,
- GEN_INT (struct_value_size), TYPE_MODE (sizetype),
- GEN_INT (MEMORY_USE_WO),
- TYPE_MODE (integer_type_node));
+ /* 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 (GET_CODE (struct_value_rtx) == REG)
- use_reg (&call_fusage, struct_value_rtx);
- }
+ if (flags & ECF_MAY_BE_ALLOCA)
+ current_function_calls_alloca = 1;
- funexp = prepare_call_address (funexp, fndecl, &call_fusage, reg_parm_seen);
+ /* 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);
- load_register_parameters (args, num_actuals, &call_fusage);
+ actparms
+ = tree_cons (error_mark_node,
+ make_tree (build_pointer_type (TREE_TYPE (funtype)),
+ temp),
+ actparms);
+ structure_value_addr_parm = 1;
+ }
- /* Perform postincrements before actually calling the function. */
- emit_queue ();
+ /* Count the arguments and set NUM_ACTUALS. */
+ for (p = actparms, num_actuals = 0; p; p = TREE_CHAIN (p))
+ num_actuals++;
- /* Save a pointer to the last insn before the call, so that we can
- later safely search backwards to find the CALL_INSN. */
- before_call = get_last_insn ();
+ /* 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.)
- /* All arguments and registers used for the call must be set up by now! */
+ 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.
- /* Generate the actual call instruction. */
- emit_call_1 (funexp, fndecl, funtype, unadjusted_args_size,
- args_size.constant, struct_value_size,
- FUNCTION_ARG (args_so_far, VOIDmode, void_type_node, 1),
- valreg, old_inhibit_defer_pop, call_fusage, is_const, nothrow);
+ 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 call is cse'able, make appropriate pair of reg-notes around it.
- Test valreg so we don't crash; may safely ignore `const'
- if return type is void. Disable for PARALLEL return values, because
- we have no way to move such values into a pseudo register. */
- if (is_const && valreg != 0 && GET_CODE (valreg) != PARALLEL)
- {
- rtx note = 0;
- rtx temp = gen_reg_rtx (GET_MODE (valreg));
- rtx insns;
+ 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;
- /* Mark the return value as a pointer if needed. */
- if (TREE_CODE (TREE_TYPE (exp)) == POINTER_TYPE)
- {
- tree pointed_to = TREE_TYPE (TREE_TYPE (exp));
- mark_reg_pointer (temp, TYPE_ALIGN (pointed_to) / BITS_PER_UNIT);
- }
+ /* Start updating where the next arg would go.
- /* Construct an "equal form" for the value which mentions all the
- arguments in order as well as the function name. */
-#ifdef PUSH_ARGS_REVERSED
- for (i = 0; i < num_actuals; i++)
- note = gen_rtx_EXPR_LIST (VOIDmode, args[i].initial_value, note);
-#else
- for (i = num_actuals - 1; i >= 0; i--)
- note = gen_rtx_EXPR_LIST (VOIDmode, args[i].initial_value, note);
-#endif
- note = gen_rtx_EXPR_LIST (VOIDmode, funexp, note);
+ 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));
- insns = get_insns ();
- end_sequence ();
+ /* 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));
- emit_libcall_block (insns, temp, valreg, note);
+ /* 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);
- valreg = temp;
- }
- else if (is_const)
+ if (args_size.var)
{
- /* Otherwise, just write out the sequence without a note. */
- rtx insns = get_insns ();
+ /* 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. */
- end_sequence ();
- emit_insns (insns);
+ flags &= ~(ECF_CONST | ECF_PURE);
+ must_preallocate = 1;
}
- else if (is_malloc)
- {
- rtx temp = gen_reg_rtx (GET_MODE (valreg));
- rtx last, insns;
-
- /* 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 / BITS_PER_UNIT);
-
- emit_move_insn (temp, valreg);
- /* The return value from a malloc-like function can not alias
- anything else. */
- last = get_last_insn ();
- REG_NOTES (last) =
- gen_rtx_EXPR_LIST (REG_NOALIAS, temp, REG_NOTES (last));
+ /* Now make final decision about preallocating stack space. */
+ must_preallocate = finalize_must_preallocate (must_preallocate,
+ num_actuals, args,
+ &args_size);
- /* Write out the sequence. */
- insns = get_insns ();
- end_sequence ();
- emit_insns (insns);
- valreg = temp;
- }
+ /* 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. */
- /* For calls to `setjmp', etc., inform flow.c it should complain
- if nonvolatile values are live. */
+ 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);
- if (returns_twice)
+ /* 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. 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
+ 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 (try_tail_call || try_tail_recursion)
{
- /* The 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)
+ 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
+ encounter, those that can be unsaved and those that can't.
+ (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.
+ For hard badness, evaluate them now and put their resulting
+ rtx in a temporary VAR_DECL.
+
+ 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)
{
- last = PREV_INSN (last);
- /* There was no CALL_INSN? */
- if (last == before_call)
- abort ();
+ inc = 1;
+ i = 0;
+ end = num_actuals;
+ }
+ else
+ {
+ inc = -1;
+ i = num_actuals - 1;
+ end = -1;
}
- emit_note_after (NOTE_INSN_SETJMP, last);
- current_function_calls_setjmp = 1;
- }
-
- if (is_longjmp)
- current_function_calls_longjmp = 1;
-
- /* Notice functions that cannot return.
- If optimizing, insns emitted below will be dead.
- If not optimizing, they will exist, which is useful
- if the user uses the `return' command in the debugger. */
- if (is_volatile || is_longjmp)
- emit_barrier ();
+ for (; i != end; i += inc)
+ {
+ switch (unsafe_for_reeval (args[i].tree_value))
+ {
+ case 0: /* Safe. */
+ break;
- /* If value type not void, return an rtx for the value. */
+ case 1: /* Mildly unsafe. */
+ args[i].tree_value = unsave_expr (args[i].tree_value);
+ break;
- /* If there are cleanups to be called, don't use a hard reg as target.
- We need to double check this and see if it matters anymore. */
- if (any_pending_cleanups (1)
- && target && REG_P (target)
- && REGNO (target) < FIRST_PSEUDO_REGISTER)
- target = 0;
+ 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;
- if (TYPE_MODE (TREE_TYPE (exp)) == VOIDmode
- || ignore)
- {
- target = const0_rtx;
- }
- else if (structure_value_addr)
- {
- 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)));
+ 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. */
+ if (any_pending_cleanups (1))
+ try_tail_call = try_tail_recursion = 0;
}
- else if (pcc_struct_value)
- {
- /* This is the special C++ case where we need to
- know what the true target was. We take care to
- 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)));
- }
- /* Handle calls that return values in multiple non-contiguous locations.
- The Irix 6 ABI has examples of this. */
- else if (GET_CODE (valreg) == PARALLEL)
+
+ /* Generate a tail recursion sequence when calling ourselves. */
+
+ if (try_tail_recursion)
{
- int bytes = int_size_in_bytes (TREE_TYPE (exp));
+ /* We want to emit any pending stack adjustments before the tail
+ 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 = 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 ();
- if (target == 0)
+ /* 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 ();
+ /* 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 ()))
{
- target = assign_stack_temp (TYPE_MODE (TREE_TYPE (exp)), bytes, 0);
- MEM_SET_IN_STRUCT_P (target, AGGREGATE_TYPE_P (TREE_TYPE (exp)));
- preserve_temp_slots (target);
+ if (any_pending_cleanups (1))
+ try_tail_call = try_tail_recursion = 0;
+ else
+ tail_recursion_insns = get_insns ();
}
+ expand_end_target_temps ();
+ end_sequence ();
- if (! rtx_equal_p (target, valreg))
- emit_group_store (target, valreg, bytes,
- TYPE_ALIGN (TREE_TYPE (exp)) / BITS_PER_UNIT);
+ /* Restore the original pending stack adjustment for the sibling and
+ normal call cases below. */
+ pending_stack_adjust = save_pending_stack_adjust;
+ stack_pointer_delta = save_stack_pointer_delta;
}
- else if (target && GET_MODE (target) == TYPE_MODE (TREE_TYPE (exp))
- && GET_MODE (target) == GET_MODE (valreg))
- /* TARGET and VALREG cannot be equal at this point because the latter
- would not have REG_FUNCTION_VALUE_P true, while the former would if
- it were referring to the same register.
-
- If they refer to the same register, this move will be a no-op, except
- when function inlining is being done. */
- emit_move_insn (target, valreg);
- else if (TYPE_MODE (TREE_TYPE (exp)) == BLKmode)
- target = copy_blkmode_from_reg (target, valreg, TREE_TYPE (exp));
- else
- target = copy_to_reg (valreg);
-#ifdef PROMOTE_FUNCTION_RETURN
- /* If we promoted this return value, make the proper SUBREG. TARGET
- might be const0_rtx here, so be careful. */
- if (GET_CODE (target) == REG
- && TYPE_MODE (TREE_TYPE (exp)) != BLKmode
- && GET_MODE (target) != TYPE_MODE (TREE_TYPE (exp)))
+ if (profile_arc_flag && (flags & ECF_FORK_OR_EXEC))
{
- tree type = TREE_TYPE (exp);
- int unsignedp = TREE_UNSIGNED (type);
+ /* 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. */
- /* 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);
- SUBREG_PROMOTED_VAR_P (target) = 1;
- SUBREG_PROMOTED_UNSIGNED_P (target) = unsignedp;
+ emit_library_call (gen_rtx_SYMBOL_REF (Pmode, "__bb_fork_func"), 0,
+ VOIDmode, 0);
}
-#endif
- /* If size of args is variable or this was a constructor call for a stack
- argument, restore saved stack-pointer value. */
+ /* Ensure current function's preferred stack boundary is at least
+ what we need. We don't have to increase alignment for recursive
+ functions. */
+ if (cfun->preferred_stack_boundary < preferred_stack_boundary
+ && fndecl != current_function_decl)
+ cfun->preferred_stack_boundary = preferred_stack_boundary;
- if (old_stack_level)
- {
- emit_stack_restore (SAVE_BLOCK, old_stack_level, NULL_RTX);
- pending_stack_adjust = old_pending_adj;
-#ifdef ACCUMULATE_OUTGOING_ARGS
- stack_arg_under_construction = old_stack_arg_under_construction;
- highest_outgoing_arg_in_use = initial_highest_arg_in_use;
- stack_usage_map = initial_stack_usage_map;
-#endif
- }
-#ifdef ACCUMULATE_OUTGOING_ARGS
- else
+ preferred_unit_stack_boundary = preferred_stack_boundary / BITS_PER_UNIT;
+
+ 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
+ initial RTL generation is complete. */
+ for (pass = 0; pass < 2; pass++)
{
-#ifdef REG_PARM_STACK_SPACE
- if (save_area)
- restore_fixed_argument_area (save_area, argblock,
- high_to_save, low_to_save);
-#endif
+ int sibcall_failure = 0;
+ /* We want to emit ay pending stack adjustments before the tail
+ 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 = 0;
+ int save_stack_pointer_delta = 0;
+ rtx insns;
+ rtx before_call, next_arg_reg;
- /* If we saved any argument areas, restore them. */
- for (i = 0; i < num_actuals; i++)
- if (args[i].save_area)
- {
- enum machine_mode save_mode = GET_MODE (args[i].save_area);
- rtx stack_area
- = gen_rtx_MEM (save_mode,
- memory_address (save_mode,
- XEXP (args[i].stack_slot, 0)));
-
- if (save_mode != BLKmode)
- emit_move_insn (stack_area, args[i].save_area);
- else
- emit_block_move (stack_area, validize_mem (args[i].save_area),
- GEN_INT (args[i].size.constant),
- PARM_BOUNDARY / BITS_PER_UNIT);
- }
+ if (pass == 0)
+ {
+ if (! try_tail_call)
+ continue;
- highest_outgoing_arg_in_use = initial_highest_arg_in_use;
- stack_usage_map = initial_stack_usage_map;
- }
-#endif
+ /* Emit any queued insns now; otherwise they would end up in
+ only one of the alternates. */
+ emit_queue ();
+
+ /* State variables we need to save and restore between
+ iterations. */
+ save_pending_stack_adjust = pending_stack_adjust;
+ save_stack_pointer_delta = stack_pointer_delta;
+ }
+ if (pass)
+ flags &= ~ECF_SIBCALL;
+ else
+ flags |= ECF_SIBCALL;
- /* 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. */
+ /* Other state variables that we must reinitialize each time
+ through the loop (that are not initialized by the loop itself). */
+ argblock = 0;
+ call_fusage = 0;
- if (may_be_alloca && nonlocal_goto_handler_slots != 0)
- emit_stack_save (SAVE_NONLOCAL, &nonlocal_goto_stack_level, NULL_RTX);
+ /* Start a new sequence for the normal call case.
- pop_temp_slots ();
+ From this point on, if the sibling call fails, we want to set
+ sibcall_failure instead of continuing the loop. */
+ start_sequence ();
- /* Free up storage we no longer need. */
- for (i = 0; i < num_actuals; ++i)
- if (args[i].aligned_regs)
- free (args[i].aligned_regs);
+ 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 ();
+ }
- return target;
-}
-\f
-/* Returns nonzero if FUN is the symbol for a library function which can
- not throw. */
+ /* Don't let pending stack adjusts add up to too much.
+ Also, do all pending adjustments now if there is any chance
+ this might be a call to alloca or if we are expanding a sibling
+ call sequence. */
+ if (pending_stack_adjust >= 32
+ || (pending_stack_adjust > 0 && (flags & ECF_MAY_BE_ALLOCA))
+ || pass == 0)
+ do_pending_stack_adjust ();
+
+ /* 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 ();
-static int
-libfunc_nothrow (fun)
- rtx fun;
-{
- if (fun == throw_libfunc
- || fun == rethrow_libfunc
- || fun == sjthrow_libfunc
- || fun == sjpopnthrow_libfunc)
- return 0;
+#ifdef FINAL_REG_PARM_STACK_SPACE
+ reg_parm_stack_space = FINAL_REG_PARM_STACK_SPACE (args_size.constant,
+ args_size.var);
+#endif
+ /* Precompute any arguments as needed. */
+ if (pass)
+ precompute_arguments (flags, num_actuals, args);
+
+ /* Now we are about to start emitting insns that can be deleted
+ if a libcall is deleted. */
+ 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,
+ &adjusted_args_size,
+ (pass == 0 ? 0
+ : preferred_stack_boundary));
+
+ 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;
+ stored_args_map = sbitmap_alloc (args_size.constant);
+ sbitmap_zero (stored_args_map);
+ }
- return 1;
-}
+ /* If we have no actual push instructions, or shouldn't use them,
+ make space for all args right now. */
+ else if (adjusted_args_size.var != 0)
+ {
+ if (old_stack_level == 0)
+ {
+ emit_stack_save (SAVE_BLOCK, &old_stack_level, NULL_RTX);
+ old_pending_adj = pending_stack_adjust;
+ pending_stack_adjust = 0;
+ /* stack_arg_under_construction says whether a stack arg is
+ being constructed at the old stack level. Pushing the stack
+ gets a clean outgoing argument block. */
+ old_stack_arg_under_construction = stack_arg_under_construction;
+ stack_arg_under_construction = 0;
+ }
+ argblock = push_block (ARGS_SIZE_RTX (adjusted_args_size), 0, 0);
+ }
+ else
+ {
+ /* Note that we must go through the motions of allocating an argument
+ block even if the size is zero because we may be storing args
+ in the area reserved for register arguments, which may be part of
+ the stack frame. */
-/* Output a library call to function FUN (a SYMBOL_REF rtx)
- (emitting the queue unless NO_QUEUE is nonzero),
- for a value of mode OUTMODE,
- with NARGS different arguments, passed as alternating rtx values
- and machine_modes to convert them to.
- The rtx values should have been passed through protect_from_queue already.
+ int needed = adjusted_args_size.constant;
- NO_QUEUE will be true if and only if the library call is a `const' call
- which will be enclosed in REG_LIBCALL/REG_RETVAL notes; it is equivalent
- to the variable is_const in expand_call.
+ /* Store the maximum argument space used. It will be pushed by
+ the prologue (if ACCUMULATE_OUTGOING_ARGS, or stack overflow
+ checking). */
- NO_QUEUE must be true for const calls, because if it isn't, then
- any pending increment will be emitted between REG_LIBCALL/REG_RETVAL notes,
- and will be lost if the libcall sequence is optimized away.
+ if (needed > current_function_outgoing_args_size)
+ current_function_outgoing_args_size = needed;
- NO_QUEUE must be false for non-const calls, because if it isn't, the
- call insn will have its CONST_CALL_P bit set, and it will be incorrectly
- optimized. For instance, the instruction scheduler may incorrectly
- move memory references across the non-const call. */
+ if (must_preallocate)
+ {
+ if (ACCUMULATE_OUTGOING_ARGS)
+ {
+ /* Since the stack pointer will never be pushed, it is
+ possible for the evaluation of a parm to clobber
+ something we have already written to the stack.
+ Since most function calls on RISC machines do not use
+ the stack, this is uncommon, but must work correctly.
-void
-emit_library_call VPARAMS((rtx orgfun, int no_queue, enum machine_mode outmode,
- int nargs, ...))
-{
-#ifndef ANSI_PROTOTYPES
- rtx orgfun;
- int no_queue;
- enum machine_mode outmode;
- int nargs;
-#endif
- va_list p;
- /* Total size in bytes of all the stack-parms scanned so far. */
- struct args_size args_size;
- /* Size of arguments before any adjustments (such as rounding). */
- struct args_size original_args_size;
- register int argnum;
- rtx fun;
- int inc;
- int count;
- 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 *argvec;
- int old_inhibit_defer_pop = inhibit_defer_pop;
- rtx call_fusage = 0;
- int reg_parm_stack_space = 0;
- int nothrow;
-#if defined(ACCUMULATE_OUTGOING_ARGS) && defined(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 */
-#endif
+ 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.
-#ifdef ACCUMULATE_OUTGOING_ARGS
- int initial_highest_arg_in_use = highest_outgoing_arg_in_use;
- char *initial_stack_usage_map = stack_usage_map;
- int needed;
+ Another approach might be to try to reorder the argument
+ evaluations to avoid this conflicting stack usage. */
+
+#ifndef OUTGOING_REG_PARM_STACK_SPACE
+ /* Since we will be writing into the entire argument area,
+ the map must be allocated for its entire size, not just
+ the part that is the responsibility of the caller. */
+ needed += reg_parm_stack_space;
#endif
-#ifdef REG_PARM_STACK_SPACE
- /* Size of the stack reserved for parameter registers. */
-#ifdef MAYBE_REG_PARM_STACK_SPACE
- reg_parm_stack_space = MAYBE_REG_PARM_STACK_SPACE;
+#ifdef ARGS_GROW_DOWNWARD
+ highest_outgoing_arg_in_use = MAX (initial_highest_arg_in_use,
+ needed + 1);
#else
- reg_parm_stack_space = REG_PARM_STACK_SPACE ((tree) 0);
-#endif
+ highest_outgoing_arg_in_use = MAX (initial_highest_arg_in_use,
+ needed);
#endif
+ stack_usage_map
+ = (char *) alloca (highest_outgoing_arg_in_use);
- VA_START (p, nargs);
+ if (initial_highest_arg_in_use)
+ memcpy (stack_usage_map, initial_stack_usage_map,
+ initial_highest_arg_in_use);
-#ifndef ANSI_PROTOTYPES
- orgfun = va_arg (p, rtx);
- no_queue = va_arg (p, int);
- outmode = va_arg (p, enum machine_mode);
- nargs = va_arg (p, int);
-#endif
+ if (initial_highest_arg_in_use != highest_outgoing_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;
- fun = orgfun;
+ /* 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. */
- nothrow = libfunc_nothrow (fun);
+ 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
+ (unadjusted_args_size,
+ &adjusted_args_size,
+ preferred_unit_stack_boundary));
+
+ /* combine_pending_stack_adjustment_and_call computes
+ an adjustment before the arguments are allocated.
+ Account for them and see whether or not the stack
+ needs to go up or down. */
+ needed = unadjusted_args_size - needed;
+
+ if (needed < 0)
+ {
+ /* We're releasing stack space. */
+ /* ??? We can avoid any adjustment at all if we're
+ already aligned. FIXME. */
+ pending_stack_adjust = -needed;
+ do_pending_stack_adjust ();
+ needed = 0;
+ }
+ else
+ /* We need to allocate space. We'll do that in
+ push_block below. */
+ pending_stack_adjust = 0;
+ }
- /* Copy all the libcall-arguments out of the varargs data
- and into a vector ARGVEC.
+ /* Special case this because overhead of `push_block' in
+ this case is non-trivial. */
+ if (needed == 0)
+ argblock = virtual_outgoing_args_rtx;
+ else
+ argblock = push_block (GEN_INT (needed), 0, 0);
+
+ /* We only really need to call `copy_to_reg' in the case
+ where push insns are going to be used to pass ARGBLOCK
+ to a function call in ARGS. In that case, the stack
+ pointer changes value from the allocation point to the
+ call point, and hence the value of
+ VIRTUAL_OUTGOING_ARGS_RTX changes as well. But might
+ as well always do it. */
+ argblock = copy_to_reg (argblock);
+
+ /* The save/restore code in store_one_arg handles all
+ cases except one: a constructor call (including a C
+ function returning a BLKmode struct) to initialize
+ an argument. */
+ if (stack_arg_under_construction)
+ {
+#ifndef OUTGOING_REG_PARM_STACK_SPACE
+ rtx push_size = GEN_INT (reg_parm_stack_space
+ + adjusted_args_size.constant);
+#else
+ rtx push_size = GEN_INT (adjusted_args_size.constant);
+#endif
+ if (old_stack_level == 0)
+ {
+ emit_stack_save (SAVE_BLOCK, &old_stack_level,
+ NULL_RTX);
+ old_pending_adj = pending_stack_adjust;
+ pending_stack_adjust = 0;
+ /* stack_arg_under_construction says whether a stack
+ arg is being constructed at the old stack level.
+ Pushing the stack gets a clean outgoing argument
+ block. */
+ old_stack_arg_under_construction
+ = stack_arg_under_construction;
+ stack_arg_under_construction = 0;
+ /* Make a new map for the new argument list. */
+ stack_usage_map = (char *)
+ alloca (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,
+ BITS_PER_UNIT);
+ }
+ /* If argument evaluation might modify the stack pointer,
+ copy the address of the argument list to a register. */
+ for (i = 0; i < num_actuals; i++)
+ if (args[i].pass_on_stack)
+ {
+ argblock = copy_addr_to_reg (argblock);
+ break;
+ }
+ }
+ }
+ }
- Compute how to pass each argument. We only support a very small subset
- of the full argument passing conventions to limit complexity here since
- library functions shouldn't have many args. */
+ compute_argument_addresses (args, argblock, num_actuals);
- argvec = (struct arg *) alloca (nargs * sizeof (struct arg));
- bzero ((char *) argvec, nargs * sizeof (struct arg));
+#ifdef PREFERRED_STACK_BOUNDARY
+ /* If we push args individually in reverse order, perform stack alignment
+ before the first push (the last arg). */
+ if (PUSH_ARGS_REVERSED && argblock == 0
+ && 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
+ && ! (flags & (ECF_CONST | ECF_PURE))
+ && ! inhibit_defer_pop)
+ {
+ pending_stack_adjust
+ = (combine_pending_stack_adjustment_and_call
+ (unadjusted_args_size,
+ &adjusted_args_size,
+ preferred_unit_stack_boundary));
+ do_pending_stack_adjust ();
+ }
+ else if (argblock == 0)
+ anti_adjust_stack (GEN_INT (adjusted_args_size.constant
+ - unadjusted_args_size));
+ }
+ /* Now that the stack is properly aligned, pops can't safely
+ be deferred during the evaluation of the arguments. */
+ NO_DEFER_POP;
+#endif
+ /* Don't try to defer pops if preallocating, not even from the first arg,
+ since ARGBLOCK probably refers to the SP. */
+ if (argblock)
+ NO_DEFER_POP;
- INIT_CUMULATIVE_ARGS (args_so_far, NULL_TREE, fun, 0);
+ funexp = rtx_for_function_call (fndecl, exp);
- args_size.constant = 0;
- args_size.var = 0;
+ /* Figure out the register where the value, if any, will come back. */
+ valreg = 0;
+ if (TYPE_MODE (TREE_TYPE (exp)) != VOIDmode
+ && ! structure_value_addr)
+ {
+ if (pcc_struct_value)
+ valreg = hard_function_value (build_pointer_type (TREE_TYPE (exp)),
+ fndecl, (pass == 0));
+ else
+ valreg = hard_function_value (TREE_TYPE (exp), fndecl, (pass == 0));
+ }
- push_temp_slots ();
+ /* Precompute all register parameters. It isn't safe to compute anything
+ once we have started filling any specific hard regs. */
+ precompute_register_parameters (num_actuals, args, ®_parm_seen);
-#ifdef PREFERRED_STACK_BOUNDARY
- /* Ensure current function's preferred stack boundary is at least
- what we need. */
- if (cfun->preferred_stack_boundary < PREFERRED_STACK_BOUNDARY)
- cfun->preferred_stack_boundary = PREFERRED_STACK_BOUNDARY;
+#ifdef REG_PARM_STACK_SPACE
+ /* Save the fixed argument area if it's part of the caller's frame and
+ is clobbered by argument setup for this call. */
+ if (ACCUMULATE_OUTGOING_ARGS && pass)
+ save_area = save_fixed_argument_area (reg_parm_stack_space, argblock,
+ &low_to_save, &high_to_save);
#endif
- for (count = 0; count < nargs; count++)
- {
- rtx val = va_arg (p, rtx);
- enum machine_mode mode = va_arg (p, enum machine_mode);
-
- /* We cannot convert the arg value to the mode the library wants here;
- must do it earlier where we know the signedness of the arg. */
- if (mode == BLKmode
- || (GET_MODE (val) != mode && GET_MODE (val) != VOIDmode))
- abort ();
-
- /* On some machines, there's no way to pass a float to a library fcn.
- Pass it as a double instead. */
-#ifdef LIBGCC_NEEDS_DOUBLE
- if (LIBGCC_NEEDS_DOUBLE && mode == SFmode)
- val = convert_modes (DFmode, SFmode, val, 0), mode = DFmode;
-#endif
+ /* Now store (and compute if necessary) all non-register parms.
+ These come before register parms, since they can require block-moves,
+ which could clobber the registers used for register parms.
+ Parms which have partial registers are not stored here,
+ but we do preallocate space here if they want that. */
- /* There's no need to call protect_from_queue, because
- either emit_move_insn or emit_push_insn will do that. */
+ for (i = 0; i < num_actuals; i++)
+ if (args[i].reg == 0 || args[i].pass_on_stack)
+ {
+ 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;
+ }
- /* Make sure it is a reasonable operand for a move or push insn. */
- if (GET_CODE (val) != REG && GET_CODE (val) != MEM
- && ! (CONSTANT_P (val) && LEGITIMATE_CONSTANT_P (val)))
- val = force_operand (val, NULL_RTX);
+ /* 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,
+ make a group of pseudos that correspond to each register that we
+ will later fill. */
+ if (STRICT_ALIGNMENT)
+ store_unaligned_arguments_into_pseudos (args, num_actuals);
+
+ /* Now store any partially-in-registers parm.
+ This is the last place a block-move can happen. */
+ if (reg_parm_seen)
+ for (i = 0; i < num_actuals; i++)
+ if (args[i].partial != 0 && ! args[i].pass_on_stack)
+ {
+ 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 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);
- mode = Pmode;
- }
+#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 (adjusted_args_size.constant
+ - unadjusted_args_size));
#endif
- argvec[count].value = val;
- argvec[count].mode = mode;
-
- argvec[count].reg = FUNCTION_ARG (args_so_far, mode, NULL_TREE, 1);
-
-#ifdef FUNCTION_ARG_PARTIAL_NREGS
- argvec[count].partial
- = FUNCTION_ARG_PARTIAL_NREGS (args_so_far, mode, NULL_TREE, 1);
-#else
- argvec[count].partial = 0;
+ /* If register arguments require space on the stack and stack space
+ was not preallocated, allocate stack space here for arguments
+ passed in registers. */
+#ifdef OUTGOING_REG_PARM_STACK_SPACE
+ if (!ACCUMULATE_OUTGOING_ARGS
+ && must_preallocate == 0 && reg_parm_stack_space > 0)
+ anti_adjust_stack (GEN_INT (reg_parm_stack_space));
#endif
- locate_and_pad_parm (mode, NULL_TREE,
- argvec[count].reg && argvec[count].partial == 0,
- NULL_TREE, &args_size, &argvec[count].offset,
- &argvec[count].size, &alignment_pad);
+ /* Pass the function the address in which to return a
+ structure value. */
+ if (pass != 0 && structure_value_addr && ! structure_value_addr_parm)
+ {
+ emit_move_insn (struct_value_rtx,
+ force_reg (Pmode,
+ force_operand (structure_value_addr,
+ NULL_RTX)));
+
+ /* Mark the memory for the aggregate as write-only. */
+ if (current_function_check_memory_usage)
+ emit_library_call (chkr_set_right_libfunc, LCT_CONST_MAKE_BLOCK,
+ VOIDmode, 3,
+ structure_value_addr, ptr_mode,
+ GEN_INT (struct_value_size),
+ TYPE_MODE (sizetype),
+ GEN_INT (MEMORY_USE_WO),
+ TYPE_MODE (integer_type_node));
+
+ if (GET_CODE (struct_value_rtx) == REG)
+ use_reg (&call_fusage, struct_value_rtx);
+ }
- if (argvec[count].size.var)
- abort ();
+ funexp = prepare_call_address (funexp, fndecl, &call_fusage,
+ reg_parm_seen, pass == 0);
- if (reg_parm_stack_space == 0 && argvec[count].partial)
- argvec[count].size.constant -= argvec[count].partial * UNITS_PER_WORD;
+ load_register_parameters (args, num_actuals, &call_fusage, flags);
- if (argvec[count].reg == 0 || argvec[count].partial != 0
- || reg_parm_stack_space > 0)
- args_size.constant += argvec[count].size.constant;
+ /* Perform postincrements before actually calling the function. */
+ emit_queue ();
- FUNCTION_ARG_ADVANCE (args_so_far, mode, (tree) 0, 1);
- }
- va_end (p);
+ /* Save a pointer to the last insn before the call, so that we can
+ later safely search backwards to find the CALL_INSN. */
+ before_call = get_last_insn ();
-#ifdef FINAL_REG_PARM_STACK_SPACE
- reg_parm_stack_space = FINAL_REG_PARM_STACK_SPACE (args_size.constant,
- args_size.var);
+ /* Set up next argument register. For sibling calls on machines
+ with register windows this should be the incoming register. */
+#ifdef FUNCTION_INCOMING_ARG
+ if (pass == 0)
+ next_arg_reg = FUNCTION_INCOMING_ARG (args_so_far, VOIDmode,
+ void_type_node, 1);
+ else
#endif
-
- /* If this machine requires an external definition for library
- functions, write one out. */
- assemble_external_libcall (fun);
+ next_arg_reg = FUNCTION_ARG (args_so_far, VOIDmode,
+ void_type_node, 1);
+
+ /* All arguments and registers used for the call must be set up by
+ now! */
- original_args_size = args_size;
#ifdef PREFERRED_STACK_BOUNDARY
- args_size.constant = (((args_size.constant + (STACK_BYTES - 1))
- / STACK_BYTES) * STACK_BYTES);
+ /* Stack must be properly aligned now. */
+ if (pass && stack_pointer_delta % preferred_unit_stack_boundary)
+ abort ();
#endif
- args_size.constant = MAX (args_size.constant,
- reg_parm_stack_space);
+ /* Generate the actual call instruction. */
+ emit_call_1 (funexp, fndecl, funtype, unadjusted_args_size,
+ adjusted_args_size.constant, struct_value_size,
+ next_arg_reg, valreg, old_inhibit_defer_pop, call_fusage,
+ flags);
-#ifndef OUTGOING_REG_PARM_STACK_SPACE
- args_size.constant -= reg_parm_stack_space;
-#endif
+ /* Verify that we've deallocated all the stack we used. */
+ if (pass
+ && old_stack_allocated != stack_pointer_delta - pending_stack_adjust)
+ abort ();
- if (args_size.constant > current_function_outgoing_args_size)
- current_function_outgoing_args_size = args_size.constant;
+ /* If call is cse'able, make appropriate pair of reg-notes around it.
+ Test valreg so we don't crash; may safely ignore `const'
+ if return type is void. Disable for PARALLEL return values, because
+ we have no way to move such values into a pseudo register. */
+ if (pass
+ && (flags & (ECF_CONST | ECF_PURE))
+ && valreg != 0 && GET_CODE (valreg) != PARALLEL)
+ {
+ rtx note = 0;
+ rtx temp = gen_reg_rtx (GET_MODE (valreg));
+ rtx insns;
-#ifdef ACCUMULATE_OUTGOING_ARGS
- /* Since the stack pointer will never be pushed, it is possible for
- the evaluation of a parm to clobber something we have already
- written to the stack. Since most function calls on RISC machines
- do not use the stack, this is uncommon, but must work correctly.
+ /* Mark the return value as a pointer if needed. */
+ if (TREE_CODE (TREE_TYPE (exp)) == POINTER_TYPE)
+ mark_reg_pointer (temp, TYPE_ALIGN (TREE_TYPE (TREE_TYPE (exp))));
- 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.
+ /* Construct an "equal form" for the value which mentions all the
+ arguments in order as well as the function name. */
+ for (i = 0; i < num_actuals; i++)
+ note = gen_rtx_EXPR_LIST (VOIDmode, args[i].initial_value, note);
+ note = gen_rtx_EXPR_LIST (VOIDmode, funexp, note);
- Another approach might be to try to reorder the argument
- evaluations to avoid this conflicting stack usage. */
+ insns = get_insns ();
+ end_sequence ();
- needed = args_size.constant;
+ if (flags & ECF_PURE)
+ note = gen_rtx_EXPR_LIST (VOIDmode,
+ gen_rtx_USE (VOIDmode,
+ gen_rtx_MEM (BLKmode,
+ gen_rtx_SCRATCH (VOIDmode))), note);
-#ifndef OUTGOING_REG_PARM_STACK_SPACE
- /* Since we will be writing into the entire argument area, the
- map must be allocated for its entire size, not just the part that
- is the responsibility of the caller. */
- needed += reg_parm_stack_space;
-#endif
+ emit_libcall_block (insns, temp, valreg, note);
-#ifdef ARGS_GROW_DOWNWARD
- highest_outgoing_arg_in_use = MAX (initial_highest_arg_in_use,
- needed + 1);
-#else
- highest_outgoing_arg_in_use = MAX (initial_highest_arg_in_use,
- needed);
-#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);
-
- if (initial_highest_arg_in_use != highest_outgoing_arg_in_use)
- bzero (&stack_usage_map[initial_highest_arg_in_use],
- 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.
- */
-
- argblock = virtual_outgoing_args_rtx;
-#else /* not ACCUMULATE_OUTGOING_ARGS */
-#ifndef PUSH_ROUNDING
- argblock = push_block (GEN_INT (args_size.constant), 0, 0);
-#endif
-#endif
+ valreg = temp;
+ }
+ else if (flags & (ECF_CONST | ECF_PURE))
+ {
+ /* Otherwise, just write out the sequence without a note. */
+ rtx insns = get_insns ();
-#ifdef PUSH_ARGS_REVERSED
-#ifdef PREFERRED_STACK_BOUNDARY
- /* If we push args individually in reverse order, perform stack alignment
- before the first push (the last arg). */
- if (argblock == 0)
- anti_adjust_stack (GEN_INT (args_size.constant
- - original_args_size.constant));
-#endif
-#endif
+ end_sequence ();
+ emit_insns (insns);
+ }
+ else if (flags & ECF_MALLOC)
+ {
+ rtx temp = gen_reg_rtx (GET_MODE (valreg));
+ rtx last, insns;
+
+ /* 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);
+
+ emit_move_insn (temp, valreg);
+
+ /* The return value from a malloc-like function can not alias
+ anything else. */
+ last = get_last_insn ();
+ REG_NOTES (last) =
+ gen_rtx_EXPR_LIST (REG_NOALIAS, temp, REG_NOTES (last));
+
+ /* Write out the sequence. */
+ insns = get_insns ();
+ end_sequence ();
+ emit_insns (insns);
+ valreg = temp;
+ }
-#ifdef PUSH_ARGS_REVERSED
- inc = -1;
- argnum = nargs - 1;
-#else
- inc = 1;
- argnum = 0;
-#endif
+ /* 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 defined(ACCUMULATE_OUTGOING_ARGS) && defined(REG_PARM_STACK_SPACE)
- /* 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.
+ if ((flags & (ECF_RETURNS_TWICE | ECF_NORETURN | ECF_LONGJMP)) || pass == 0)
+ {
+ /* 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. */
- Here we compute the boundary of the that needs to be saved, if any. */
+ 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 ();
+ }
-#ifdef ARGS_GROW_DOWNWARD
- for (count = 0; count < reg_parm_stack_space + 1; count++)
-#else
- for (count = 0; count < reg_parm_stack_space; count++)
-#endif
- {
- if (count >= highest_outgoing_arg_in_use
- || stack_usage_map[count] == 0)
- continue;
+ if (flags & ECF_RETURNS_TWICE)
+ {
+ emit_note_after (NOTE_INSN_SETJMP, last);
+ current_function_calls_setjmp = 1;
+ }
+ else
+ emit_barrier_after (last);
+ }
- if (low_to_save == -1)
- low_to_save = count;
+ if (flags & ECF_LONGJMP)
+ current_function_calls_longjmp = 1;
- high_to_save = count;
- }
+ /* If this function is returning into a memory location marked as
+ readonly, it means it is initializing that location. But we normally
+ treat functions as not clobbering such locations, so we need to
+ specify that this one does. */
+ if (target != 0 && GET_CODE (target) == MEM
+ && structure_value_addr != 0 && RTX_UNCHANGING_P (target))
+ emit_insn (gen_rtx_CLOBBER (VOIDmode, target));
- if (low_to_save >= 0)
- {
- int num_to_save = high_to_save - low_to_save + 1;
- enum machine_mode save_mode
- = mode_for_size (num_to_save * BITS_PER_UNIT, MODE_INT, 1);
- rtx stack_area;
+ /* If value type not void, return an rtx for the value. */
- /* 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)))
- save_mode = BLKmode;
+ /* If there are cleanups to be called, don't use a hard reg as target.
+ We need to double check this and see if it matters anymore. */
+ if (any_pending_cleanups (1))
+ {
+ if (target && REG_P (target)
+ && REGNO (target) < FIRST_PSEUDO_REGISTER)
+ target = 0;
+ sibcall_failure = 1;
+ }
-#ifdef ARGS_GROW_DOWNWARD
- 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,
- plus_constant (argblock,
- low_to_save)));
-#endif
- if (save_mode == BLKmode)
+ if (TYPE_MODE (TREE_TYPE (exp)) == VOIDmode
+ || ignore)
+ {
+ target = const0_rtx;
+ }
+ else if (structure_value_addr)
+ {
+ 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));
+ set_mem_attributes (target, exp, 1);
+ }
+ }
+ else if (pcc_struct_value)
{
- save_area = assign_stack_temp (BLKmode, num_to_save, 0);
- emit_block_move (validize_mem (save_area), stack_area,
- GEN_INT (num_to_save),
- PARM_BOUNDARY / BITS_PER_UNIT);
+ /* This is the special C++ case where we need to
+ know what the true target was. We take care to
+ never use this value more than once in one expression. */
+ target = gen_rtx_MEM (TYPE_MODE (TREE_TYPE (exp)),
+ copy_to_reg (valreg));
+ set_mem_attributes (target, exp, 1);
}
- else
+ /* Handle calls that return values in multiple non-contiguous locations.
+ The Irix 6 ABI has examples of this. */
+ else if (GET_CODE (valreg) == PARALLEL)
{
- save_area = gen_reg_rtx (save_mode);
- emit_move_insn (save_area, stack_area);
+ if (target == 0)
+ {
+ /* 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,
+ int_size_in_bytes (TREE_TYPE (exp)),
+ TYPE_ALIGN (TREE_TYPE (exp)));
+
+ /* We can not support sibling calls for this case. */
+ sibcall_failure = 1;
}
- }
-#endif
-
- /* Push the args that need to be pushed. */
+ else if (target
+ && GET_MODE (target) == TYPE_MODE (TREE_TYPE (exp))
+ && GET_MODE (target) == GET_MODE (valreg))
+ {
+ /* TARGET and VALREG cannot be equal at this point because the
+ latter would not have REG_FUNCTION_VALUE_P true, while the
+ former would if it were referring to the same register.
- /* ARGNUM indexes the ARGVEC array in the order in which the arguments
- are to be pushed. */
- for (count = 0; count < nargs; count++, argnum += inc)
- {
- register enum machine_mode mode = argvec[argnum].mode;
- register rtx val = argvec[argnum].value;
- rtx reg = argvec[argnum].reg;
- int partial = argvec[argnum].partial;
-#ifdef ACCUMULATE_OUTGOING_ARGS
- int lower_bound, upper_bound, i;
-#endif
+ If they refer to the same register, this move will be a no-op,
+ except when function inlining is being done. */
+ emit_move_insn (target, valreg);
+ }
+ else if (TYPE_MODE (TREE_TYPE (exp)) == BLKmode)
+ {
+ target = copy_blkmode_from_reg (target, valreg, TREE_TYPE (exp));
- if (! (reg != 0 && partial == 0))
+ /* We can not support sibling calls for this case. */
+ sibcall_failure = 1;
+ }
+ else
+ target = copy_to_reg (valreg);
+
+#ifdef PROMOTE_FUNCTION_RETURN
+ /* If we promoted this return value, make the proper SUBREG. TARGET
+ might be const0_rtx here, so be careful. */
+ if (GET_CODE (target) == REG
+ && TYPE_MODE (TREE_TYPE (exp)) != BLKmode
+ && GET_MODE (target) != TYPE_MODE (TREE_TYPE (exp)))
{
-#ifdef ACCUMULATE_OUTGOING_ARGS
- /* If this is being stored into a pre-allocated, fixed-size, stack
- area, save any previous data at that location. */
+ tree type = TREE_TYPE (exp);
+ int unsignedp = TREE_UNSIGNED (type);
+ int offset = 0;
-#ifdef ARGS_GROW_DOWNWARD
- /* stack_slot is negative, but we want to index stack_usage_map
- with positive values. */
- upper_bound = -argvec[argnum].offset.constant + 1;
- lower_bound = upper_bound - argvec[argnum].size.constant;
-#else
- lower_bound = argvec[argnum].offset.constant;
- upper_bound = lower_bound + argvec[argnum].size.constant;
+ /* If we don't promote as expected, something is wrong. */
+ if (GET_MODE (target)
+ != promote_mode (type, TYPE_MODE (type), &unsignedp, 1))
+ abort ();
+
+ 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;
+ }
#endif
- for (i = lower_bound; i < upper_bound; i++)
- if (stack_usage_map[i]
- /* Don't store things in the fixed argument area at this point;
- it has already been saved. */
- && i > reg_parm_stack_space)
- break;
+ /* If size of args is variable or this was a constructor call for a stack
+ argument, restore saved stack-pointer value. */
- if (i != upper_bound)
+ 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_arg_under_construction = old_stack_arg_under_construction;
+ highest_outgoing_arg_in_use = initial_highest_arg_in_use;
+ stack_usage_map = initial_stack_usage_map;
+ sibcall_failure = 1;
+ }
+ else if (ACCUMULATE_OUTGOING_ARGS && pass)
+ {
+#ifdef REG_PARM_STACK_SPACE
+ if (save_area)
{
- /* We need to make a save area. See what mode we can make it. */
- enum machine_mode save_mode
- = mode_for_size (argvec[argnum].size.constant * BITS_PER_UNIT,
- MODE_INT, 1);
- rtx stack_area
- = gen_rtx_MEM
- (save_mode,
- memory_address
- (save_mode,
- plus_constant (argblock,
- argvec[argnum].offset.constant)));
-
- argvec[argnum].save_area = gen_reg_rtx (save_mode);
- emit_move_insn (argvec[argnum].save_area, stack_area);
+ restore_fixed_argument_area (save_area, argblock,
+ high_to_save, low_to_save);
}
#endif
- emit_push_insn (val, mode, NULL_TREE, NULL_RTX, 0, partial, reg, 0,
- argblock, GEN_INT (argvec[argnum].offset.constant),
- reg_parm_stack_space, ARGS_SIZE_RTX (alignment_pad));
-#ifdef ACCUMULATE_OUTGOING_ARGS
- /* Now mark the segment we just used. */
- for (i = lower_bound; i < upper_bound; i++)
- stack_usage_map[i] = 1;
-#endif
+ /* If we saved any argument areas, restore them. */
+ for (i = 0; i < num_actuals; i++)
+ if (args[i].save_area)
+ {
+ enum machine_mode save_mode = GET_MODE (args[i].save_area);
+ rtx stack_area
+ = gen_rtx_MEM (save_mode,
+ memory_address (save_mode,
+ XEXP (args[i].stack_slot, 0)));
+
+ if (save_mode != BLKmode)
+ emit_move_insn (stack_area, args[i].save_area);
+ else
+ emit_block_move (stack_area,
+ validize_mem (args[i].save_area),
+ GEN_INT (args[i].size.constant),
+ PARM_BOUNDARY);
+ }
- NO_DEFER_POP;
+ highest_outgoing_arg_in_use = initial_highest_arg_in_use;
+ stack_usage_map = initial_stack_usage_map;
}
- }
-
-#ifndef PUSH_ARGS_REVERSED
-#ifdef PREFERRED_STACK_BOUNDARY
- /* If we pushed args in forward order, perform stack alignment
- after pushing the last arg. */
- if (argblock == 0)
- anti_adjust_stack (GEN_INT (args_size.constant
- - original_args_size.constant));
-#endif
-#endif
-#ifdef PUSH_ARGS_REVERSED
- argnum = nargs - 1;
-#else
- argnum = 0;
-#endif
+ /* 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. */
- fun = prepare_call_address (fun, NULL_TREE, &call_fusage, 0);
+ if ((flags & ECF_MAY_BE_ALLOCA) && nonlocal_goto_handler_slots != 0)
+ emit_stack_save (SAVE_NONLOCAL, &nonlocal_goto_stack_level, NULL_RTX);
- /* Now load any reg parms into their regs. */
+ pop_temp_slots ();
- /* ARGNUM indexes the ARGVEC array in the order in which the arguments
- are to be pushed. */
- for (count = 0; count < nargs; count++, argnum += inc)
- {
- register rtx val = argvec[argnum].value;
- rtx reg = argvec[argnum].reg;
- int partial = argvec[argnum].partial;
+ /* Free up storage we no longer need. */
+ for (i = 0; i < num_actuals; ++i)
+ if (args[i].aligned_regs)
+ free (args[i].aligned_regs);
- /* Handle calls that pass values in multiple non-contiguous
- locations. The PA64 has examples of this for library calls. */
- if (reg != 0 && GET_CODE (reg) == PARALLEL)
- emit_group_load (reg, val,
- GET_MODE_SIZE (GET_MODE (val)),
- GET_MODE_ALIGNMENT (GET_MODE (val)));
- else if (reg != 0 && partial == 0)
- emit_move_insn (reg, val);
+ if (pass == 0)
+ {
+ /* Undo the fake expand_start_target_temps we did earlier. If
+ there had been any cleanups created, we've already set
+ sibcall_failure. */
+ expand_end_target_temps ();
+ }
- NO_DEFER_POP;
- }
+ insns = get_insns ();
+ end_sequence ();
- /* For version 1.37, try deleting this entirely. */
- if (! no_queue)
- emit_queue ();
+ if (pass == 0)
+ {
+ tail_call_insns = insns;
- /* Any regs containing parms remain in use through the call. */
- for (count = 0; count < nargs; count++)
- {
- rtx reg = argvec[count].reg;
- if (reg != 0 && GET_CODE (argvec[count].reg) == PARALLEL)
- use_group_regs (&call_fusage, reg);
- else if (reg != 0)
- use_reg (&call_fusage, reg);
- }
+ /* Restore the pending stack adjustment now that we have
+ finished generating the sibling call sequence. */
- /* Don't allow popping to be deferred, since then
- cse'ing of library calls could delete a call and leave the pop. */
- NO_DEFER_POP;
+ pending_stack_adjust = save_pending_stack_adjust;
+ stack_pointer_delta = save_stack_pointer_delta;
- /* We pass the old value of inhibit_defer_pop + 1 to emit_call_1, which
- will set inhibit_defer_pop to that value. */
+ /* 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;
+ }
- /* The return type is needed to decide how many bytes the function pops.
- Signedness plays no role in that, so for simplicity, we pretend it's
- always signed. We also assume that the list of arguments passed has
- no impact, so we pretend it is unknown. */
+ sbitmap_free (stored_args_map);
+ }
+ else
+ normal_call_insns = insns;
- 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, 0,
- FUNCTION_ARG (args_so_far, VOIDmode, void_type_node, 1),
- outmode != VOIDmode ? hard_libcall_value (outmode) : NULL_RTX,
- old_inhibit_defer_pop + 1, call_fusage, no_queue, nothrow);
+ /* If something prevents making this a sibling call,
+ zero out the sequence. */
+ if (sibcall_failure)
+ tail_call_insns = NULL_RTX;
+ }
- pop_temp_slots ();
+ /* The function optimize_sibling_and_tail_recursive_calls doesn't
+ handle CALL_PLACEHOLDERs inside other CALL_PLACEHOLDERs. This
+ can happen if the arguments to this function call an inline
+ 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. */
+
+ for (insn = normal_call_insns; insn; insn = NEXT_INSN (insn))
+ if (GET_CODE (insn) == CALL_INSN
+ && GET_CODE (PATTERN (insn)) == CALL_PLACEHOLDER)
+ replace_call_placeholder (insn, sibcall_use_normal);
+
+ for (insn = tail_call_insns; insn; insn = NEXT_INSN (insn))
+ if (GET_CODE (insn) == CALL_INSN
+ && GET_CODE (PATTERN (insn)) == CALL_PLACEHOLDER)
+ replace_call_placeholder (insn, sibcall_use_normal);
+
+ for (insn = tail_recursion_insns; insn; insn = NEXT_INSN (insn))
+ if (GET_CODE (insn) == CALL_INSN
+ && GET_CODE (PATTERN (insn)) == CALL_PLACEHOLDER)
+ replace_call_placeholder (insn, sibcall_use_normal);
+
+ /* If this was a potential tail recursion site, then emit a
+ CALL_PLACEHOLDER with the normal and the tail recursion streams.
+ One of them will be selected later. */
+ if (tail_recursion_insns || tail_call_insns)
+ {
+ /* The tail recursion label must be kept around. We could expose
+ its use in the CALL_PLACEHOLDER, but that creates unwanted edges
+ and makes determining true tail recursion sites difficult.
+
+ So we set LABEL_PRESERVE_P here, then clear it when we select
+ one of the call sequences after rtl generation is complete. */
+ if (tail_recursion_insns)
+ LABEL_PRESERVE_P (tail_recursion_label) = 1;
+ emit_call_insn (gen_rtx_CALL_PLACEHOLDER (VOIDmode, normal_call_insns,
+ tail_call_insns,
+ tail_recursion_insns,
+ tail_recursion_label));
+ }
+ else
+ emit_insns (normal_call_insns);
- /* Now restore inhibit_defer_pop to its actual original value. */
- OK_DEFER_POP;
+ currently_expanding_call--;
-#ifdef ACCUMULATE_OUTGOING_ARGS
-#ifdef REG_PARM_STACK_SPACE
- if (save_area)
+ /* 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)
{
- enum machine_mode save_mode = GET_MODE (save_area);
-#ifdef ARGS_GROW_DOWNWARD
- rtx stack_area
- = gen_rtx_MEM (save_mode,
- memory_address (save_mode,
- plus_constant (argblock,
- - high_to_save)));
-#else
- rtx stack_area
- = gen_rtx_MEM (save_mode,
- memory_address (save_mode,
- plus_constant (argblock, low_to_save)));
-#endif
-
- if (save_mode != BLKmode)
- emit_move_insn (stack_area, save_area);
- else
- emit_block_move (stack_area, validize_mem (save_area),
- GEN_INT (high_to_save - low_to_save + 1),
- PARM_BOUNDARY / BITS_PER_UNIT);
+ 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 ();
}
-#endif
-
- /* If we saved any argument areas, restore them. */
- for (count = 0; count < nargs; count++)
- if (argvec[count].save_area)
- {
- enum machine_mode save_mode = GET_MODE (argvec[count].save_area);
- rtx stack_area
- = gen_rtx_MEM (save_mode,
- memory_address
- (save_mode,
- plus_constant (argblock,
- argvec[count].offset.constant)));
-
- emit_move_insn (stack_area, argvec[count].save_area);
- }
- highest_outgoing_arg_in_use = initial_highest_arg_in_use;
- stack_usage_map = initial_stack_usage_map;
-#endif
+ return target;
}
\f
-/* Like emit_library_call except that an extra argument, VALUE,
- comes second and says where to store the result.
- (If VALUE is zero, this function chooses a convenient way
- to return the value.
-
- This function returns an rtx for where the value is to be found.
- If VALUE is nonzero, VALUE is returned. */
-
-rtx
-emit_library_call_value VPARAMS((rtx orgfun, rtx value, int no_queue,
- enum machine_mode outmode, int nargs, ...))
+/* 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 below. */
+static rtx
+emit_library_call_value_1 (retval, orgfun, value, fn_type, outmode, nargs, p)
+ int retval;
+ rtx orgfun;
+ rtx value;
+ enum libcall_type fn_type;
+ enum machine_mode outmode;
+ int nargs;
+ va_list p;
{
-#ifndef ANSI_PROTOTYPES
- rtx orgfun;
- rtx value;
- int no_queue;
- enum machine_mode outmode;
- int nargs;
-#endif
- va_list p;
/* Total size in bytes of all the stack-parms scanned so far. */
struct args_size args_size;
/* Size of arguments before any adjustments (such as rounding). */
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 mem_value = 0;
+ rtx valreg;
int pcc_struct_value = 0;
int struct_value_size = 0;
- int is_const;
+ int flags;
int reg_parm_stack_space = 0;
- int nothrow;
-#ifdef ACCUMULATE_OUTGOING_ARGS
int needed;
-#endif
+ rtx before_call;
-#if defined(ACCUMULATE_OUTGOING_ARGS) && defined(REG_PARM_STACK_SPACE)
+#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
-#ifdef ACCUMULATE_OUTGOING_ARGS
/* Size of the stack reserved for parameter registers. */
int initial_highest_arg_in_use = highest_outgoing_arg_in_use;
char *initial_stack_usage_map = stack_usage_map;
-#endif
#ifdef REG_PARM_STACK_SPACE
#ifdef MAYBE_REG_PARM_STACK_SPACE
#endif
#endif
- VA_START (p, nargs);
-
-#ifndef ANSI_PROTOTYPES
- orgfun = va_arg (p, rtx);
- value = va_arg (p, rtx);
- no_queue = va_arg (p, int);
- outmode = va_arg (p, enum machine_mode);
- nargs = va_arg (p, int);
-#endif
+ /* By default, library functions can not throw. */
+ flags = ECF_NOTHROW;
- is_const = no_queue;
+ 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;
- nothrow = libfunc_nothrow (fun);
-
#ifdef PREFERRED_STACK_BOUNDARY
/* Ensure current function's preferred stack boundary is at least
what we need. */
/* If this kind of value comes back in memory,
decide where in memory it should come back. */
- if (aggregate_value_p (type_for_mode (outmode, 0)))
+ if (outmode != VOIDmode && aggregate_value_p (type_for_mode (outmode, 0)))
{
#ifdef PCC_STATIC_STRUCT_RETURN
rtx pointer_reg
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. */
- is_const = 0;
+ flags &= ~(ECF_CONST | ECF_PURE);
}
/* ??? Unfinished: must pass the memory address as an argument. */
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;
count = 0;
+ /* Now we are about to start emitting insns that can be deleted
+ if a libcall is deleted. */
+ if (flags & (ECF_CONST | ECF_PURE))
+ start_sequence ();
+
push_temp_slots ();
/* If there's a structure value address to be passed,
#endif
locate_and_pad_parm (Pmode, NULL_TREE,
- argvec[count].reg && argvec[count].partial == 0,
+#ifdef STACK_PARMS_IN_REG_PARM_AREA
+ 1,
+#else
+ argvec[count].reg != 0,
+#endif
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 = XEXP (slot, 0);
+ 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
#endif
locate_and_pad_parm (mode, NULL_TREE,
- argvec[count].reg && argvec[count].partial == 0,
+#ifdef STACK_PARMS_IN_REG_PARM_AREA
+ 1,
+#else
+ argvec[count].reg != 0,
+#endif
NULL_TREE, &args_size, &argvec[count].offset,
&argvec[count].size, &alignment_pad);
FUNCTION_ARG_ADVANCE (args_so_far, mode, (tree) 0, 1);
}
- va_end (p);
#ifdef FINAL_REG_PARM_STACK_SPACE
reg_parm_stack_space = FINAL_REG_PARM_STACK_SPACE (args_size.constant,
original_args_size = args_size;
#ifdef PREFERRED_STACK_BOUNDARY
- args_size.constant = (((args_size.constant + (STACK_BYTES - 1))
- / STACK_BYTES) * STACK_BYTES);
+ args_size.constant = (((args_size.constant
+ + stack_pointer_delta
+ + STACK_BYTES - 1)
+ / STACK_BYTES
+ * STACK_BYTES)
+ - stack_pointer_delta);
#endif
args_size.constant = MAX (args_size.constant,
if (args_size.constant > current_function_outgoing_args_size)
current_function_outgoing_args_size = args_size.constant;
-#ifdef ACCUMULATE_OUTGOING_ARGS
- /* Since the stack pointer will never be pushed, it is possible for
- the evaluation of a parm to clobber something we have already
- written to the stack. Since most function calls on RISC machines
- do not use the stack, this is uncommon, but must work correctly.
+ if (ACCUMULATE_OUTGOING_ARGS)
+ {
+ /* Since the stack pointer will never be pushed, it is possible for
+ the evaluation of a parm to clobber something we have already
+ written to the stack. Since most function calls on RISC machines
+ do not use the stack, this is uncommon, but must work correctly.
- 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.
+ 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.
- Another approach might be to try to reorder the argument
- evaluations to avoid this conflicting stack usage. */
+ Another approach might be to try to reorder the argument
+ evaluations to avoid this conflicting stack usage. */
- needed = args_size.constant;
+ needed = args_size.constant;
#ifndef OUTGOING_REG_PARM_STACK_SPACE
- /* Since we will be writing into the entire argument area, the
- map must be allocated for its entire size, not just the part that
- is the responsibility of the caller. */
- needed += reg_parm_stack_space;
+ /* Since we will be writing into the entire argument area, the
+ map must be allocated for its entire size, not just the part that
+ is the responsibility of the caller. */
+ needed += reg_parm_stack_space;
#endif
#ifdef ARGS_GROW_DOWNWARD
- highest_outgoing_arg_in_use = MAX (initial_highest_arg_in_use,
- needed + 1);
+ highest_outgoing_arg_in_use = MAX (initial_highest_arg_in_use,
+ needed + 1);
#else
- highest_outgoing_arg_in_use = MAX (initial_highest_arg_in_use,
- needed);
-#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);
-
- if (initial_highest_arg_in_use != highest_outgoing_arg_in_use)
- bzero (&stack_usage_map[initial_highest_arg_in_use],
- 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.
- */
-
- argblock = virtual_outgoing_args_rtx;
-#else /* not ACCUMULATE_OUTGOING_ARGS */
-#ifndef PUSH_ROUNDING
- argblock = push_block (GEN_INT (args_size.constant), 0, 0);
-#endif
+ highest_outgoing_arg_in_use = MAX (initial_highest_arg_in_use,
+ needed);
#endif
+ stack_usage_map = (char *) alloca (highest_outgoing_arg_in_use);
+
+ if (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)
+ memset (&stack_usage_map[initial_highest_arg_in_use], 0,
+ highest_outgoing_arg_in_use - initial_highest_arg_in_use);
+ needed = 0;
+
+ /* 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. */
+
+ if (virtuals_instantiated)
+ argblock = plus_constant (stack_pointer_rtx, STACK_POINTER_OFFSET);
+ else
+ argblock = virtual_outgoing_args_rtx;
+ }
+ else
+ {
+ if (!PUSH_ARGS)
+ argblock = push_block (GEN_INT (args_size.constant), 0, 0);
+ }
-#ifdef PUSH_ARGS_REVERSED
#ifdef PREFERRED_STACK_BOUNDARY
/* If we push args individually in reverse order, perform stack alignment
before the first push (the last arg). */
- if (argblock == 0)
+ if (argblock == 0 && PUSH_ARGS_REVERSED)
anti_adjust_stack (GEN_INT (args_size.constant
- original_args_size.constant));
#endif
-#endif
-
-#ifdef PUSH_ARGS_REVERSED
- inc = -1;
- argnum = nargs - 1;
-#else
- inc = 1;
- argnum = 0;
-#endif
-#if defined(ACCUMULATE_OUTGOING_ARGS) && defined(REG_PARM_STACK_SPACE)
- /* 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.
-
- Here we compute the boundary of the that needs to be saved, if any. */
-
-#ifdef ARGS_GROW_DOWNWARD
- for (count = 0; count < reg_parm_stack_space + 1; count++)
-#else
- for (count = 0; count < reg_parm_stack_space; count++)
-#endif
+ if (PUSH_ARGS_REVERSED)
{
- if (count >= highest_outgoing_arg_in_use
- || stack_usage_map[count] == 0)
- continue;
-
- if (low_to_save == -1)
- low_to_save = count;
-
- high_to_save = count;
+ inc = -1;
+ argnum = nargs - 1;
+ }
+ else
+ {
+ inc = 1;
+ argnum = 0;
}
- if (low_to_save >= 0)
+#ifdef REG_PARM_STACK_SPACE
+ if (ACCUMULATE_OUTGOING_ARGS)
{
- int num_to_save = high_to_save - low_to_save + 1;
- enum machine_mode save_mode
- = mode_for_size (num_to_save * BITS_PER_UNIT, MODE_INT, 1);
- rtx stack_area;
+ /* 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.
- /* 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)))
- save_mode = BLKmode;
+ Here we compute the boundary of the that needs to be saved, if any. */
#ifdef ARGS_GROW_DOWNWARD
- stack_area = gen_rtx_MEM (save_mode,
- memory_address (save_mode,
- plus_constant (argblock,
- - high_to_save)));
+ for (count = 0; count < reg_parm_stack_space + 1; count++)
#else
- stack_area = gen_rtx_MEM (save_mode,
- memory_address (save_mode,
- plus_constant (argblock,
- low_to_save)));
+ for (count = 0; count < reg_parm_stack_space; count++)
#endif
- if (save_mode == BLKmode)
{
- save_area = assign_stack_temp (BLKmode, num_to_save, 0);
- emit_block_move (validize_mem (save_area), stack_area,
- GEN_INT (num_to_save),
- PARM_BOUNDARY / BITS_PER_UNIT);
+ if (count >= highest_outgoing_arg_in_use
+ || stack_usage_map[count] == 0)
+ continue;
+
+ if (low_to_save == -1)
+ low_to_save = count;
+
+ high_to_save = count;
}
- else
+
+ if (low_to_save >= 0)
{
- save_area = gen_reg_rtx (save_mode);
- emit_move_insn (save_area, stack_area);
+ int num_to_save = high_to_save - low_to_save + 1;
+ enum machine_mode save_mode
+ = mode_for_size (num_to_save * BITS_PER_UNIT, MODE_INT, 1);
+ rtx stack_area;
+
+ /* 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)))
+ save_mode = BLKmode;
+
+#ifdef ARGS_GROW_DOWNWARD
+ 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,
+ plus_constant (argblock,
+ low_to_save)));
+#endif
+ if (save_mode == BLKmode)
+ {
+ save_area = assign_stack_temp (BLKmode, num_to_save, 0);
+ emit_block_move (validize_mem (save_area), stack_area,
+ GEN_INT (num_to_save), PARM_BOUNDARY);
+ }
+ else
+ {
+ save_area = gen_reg_rtx (save_mode);
+ emit_move_insn (save_area, stack_area);
+ }
}
}
#endif
-
+
/* Push the args that need to be pushed. */
/* ARGNUM indexes the ARGVEC array in the order in which the arguments
register rtx val = argvec[argnum].value;
rtx reg = argvec[argnum].reg;
int partial = argvec[argnum].partial;
-#ifdef ACCUMULATE_OUTGOING_ARGS
- int lower_bound, upper_bound, i;
-#endif
+ int lower_bound = 0, upper_bound = 0, i;
if (! (reg != 0 && partial == 0))
{
-#ifdef ACCUMULATE_OUTGOING_ARGS
- /* If this is being stored into a pre-allocated, fixed-size, stack
- area, save any previous data at that location. */
+ if (ACCUMULATE_OUTGOING_ARGS)
+ {
+ /* If this is being stored into a pre-allocated, fixed-size,
+ stack area, save any previous data at that location. */
#ifdef ARGS_GROW_DOWNWARD
- /* stack_slot is negative, but we want to index stack_usage_map
- with positive values. */
- upper_bound = -argvec[argnum].offset.constant + 1;
- lower_bound = upper_bound - argvec[argnum].size.constant;
+ /* stack_slot is negative, but we want to index stack_usage_map
+ with positive values. */
+ upper_bound = -argvec[argnum].offset.constant + 1;
+ lower_bound = upper_bound - argvec[argnum].size.constant;
#else
- lower_bound = argvec[argnum].offset.constant;
- upper_bound = lower_bound + argvec[argnum].size.constant;
+ lower_bound = argvec[argnum].offset.constant;
+ upper_bound = lower_bound + argvec[argnum].size.constant;
#endif
- for (i = lower_bound; i < upper_bound; i++)
- if (stack_usage_map[i]
- /* Don't store things in the fixed argument area at this point;
- it has already been saved. */
- && i > reg_parm_stack_space)
- break;
+ for (i = lower_bound; i < upper_bound; i++)
+ if (stack_usage_map[i]
+ /* Don't store things in the fixed argument area at this
+ point; it has already been saved. */
+ && i > reg_parm_stack_space)
+ break;
- if (i != upper_bound)
- {
- /* We need to make a save area. See what mode we can make it. */
- enum machine_mode save_mode
- = mode_for_size (argvec[argnum].size.constant * BITS_PER_UNIT,
- MODE_INT, 1);
- rtx stack_area
- = gen_rtx_MEM
- (save_mode,
- memory_address
- (save_mode,
- plus_constant (argblock,
- argvec[argnum].offset.constant)));
- argvec[argnum].save_area = gen_reg_rtx (save_mode);
-
- emit_move_insn (argvec[argnum].save_area, stack_area);
+ if (i != upper_bound)
+ {
+ /* We need to make a save area. See what mode we can make
+ it. */
+ enum machine_mode save_mode
+ = mode_for_size (argvec[argnum].size.constant
+ * BITS_PER_UNIT,
+ MODE_INT, 1);
+ rtx stack_area
+ = gen_rtx_MEM
+ (save_mode,
+ memory_address
+ (save_mode,
+ plus_constant (argblock,
+ argvec[argnum].offset.constant)));
+ argvec[argnum].save_area = gen_reg_rtx (save_mode);
+
+ emit_move_insn (argvec[argnum].save_area, stack_area);
+ }
}
-#endif
+
emit_push_insn (val, mode, NULL_TREE, NULL_RTX, 0, partial, reg, 0,
argblock, GEN_INT (argvec[argnum].offset.constant),
reg_parm_stack_space, ARGS_SIZE_RTX (alignment_pad));
-#ifdef ACCUMULATE_OUTGOING_ARGS
/* Now mark the segment we just used. */
- for (i = lower_bound; i < upper_bound; i++)
- stack_usage_map[i] = 1;
-#endif
+ if (ACCUMULATE_OUTGOING_ARGS)
+ for (i = lower_bound; i < upper_bound; i++)
+ stack_usage_map[i] = 1;
NO_DEFER_POP;
}
}
-#ifndef PUSH_ARGS_REVERSED
#ifdef PREFERRED_STACK_BOUNDARY
/* If we pushed args in forward order, perform stack alignment
after pushing the last arg. */
- if (argblock == 0)
+ if (argblock == 0 && !PUSH_ARGS_REVERSED)
anti_adjust_stack (GEN_INT (args_size.constant
- original_args_size.constant));
#endif
-#endif
-#ifdef PUSH_ARGS_REVERSED
- argnum = nargs - 1;
-#else
- argnum = 0;
-#endif
+ if (PUSH_ARGS_REVERSED)
+ argnum = nargs - 1;
+ 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. */
NO_DEFER_POP;
}
-#if 0
- /* For version 1.37, try deleting this entirely. */
- if (! no_queue)
- emit_queue ();
-#endif
-
/* Any regs containing parms remain in use through the call. */
for (count = 0; count < nargs; count++)
{
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
cse'ing of library calls could delete a call and leave the pop. */
NO_DEFER_POP;
+ valreg = (mem_value == 0 && outmode != VOIDmode
+ ? hard_libcall_value (outmode) : NULL_RTX);
+
+#ifdef PREFERRED_STACK_BOUNDARY
+ /* Stack must be properly aligned now. */
+ if (stack_pointer_delta & (PREFERRED_STACK_BOUNDARY / BITS_PER_UNIT - 1))
+ 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. */
- /* See the comment in emit_library_call about the function type we build
- and pass here. */
+ /* The return type is needed to decide how many bytes the function pops.
+ Signedness plays no role in that, so for simplicity, we pretend it's
+ 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)),
- build_function_type (type_for_mode (outmode, 0), NULL_TREE),
- original_args_size.constant, args_size.constant,
+ 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,
struct_value_size,
FUNCTION_ARG (args_so_far, VOIDmode, void_type_node, 1),
- mem_value == 0 ? hard_libcall_value (outmode) : NULL_RTX,
- old_inhibit_defer_pop + 1, call_fusage, is_const, nothrow);
+ 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;
+ /* If call is cse'able, make appropriate pair of reg-notes around it.
+ Test valreg so we don't crash; may safely ignore `const'
+ if return type is void. Disable for PARALLEL return values, because
+ we have no way to move such values into a pseudo register. */
+ if ((flags & (ECF_CONST | ECF_PURE))
+ && valreg != 0 && GET_CODE (valreg) != PARALLEL)
+ {
+ rtx note = 0;
+ rtx temp = gen_reg_rtx (GET_MODE (valreg));
+ rtx insns;
+ int i;
+
+ /* Construct an "equal form" for the value which mentions all the
+ arguments in order as well as the function name. */
+ for (i = 0; i < nargs; i++)
+ note = gen_rtx_EXPR_LIST (VOIDmode, argvec[i].value, note);
+ note = gen_rtx_EXPR_LIST (VOIDmode, fun, note);
+
+ insns = get_insns ();
+ end_sequence ();
+
+ if (flags & ECF_PURE)
+ note = gen_rtx_EXPR_LIST (VOIDmode,
+ gen_rtx_USE (VOIDmode,
+ gen_rtx_MEM (BLKmode,
+ gen_rtx_SCRATCH (VOIDmode))), note);
+
+ emit_libcall_block (insns, temp, valreg, note);
+
+ valreg = temp;
+ }
+ else if (flags & (ECF_CONST | ECF_PURE))
+ {
+ /* Otherwise, just write out the sequence without a note. */
+ rtx insns = get_insns ();
+
+ end_sequence ();
+ emit_insns (insns);
+ }
pop_temp_slots ();
/* Copy the value to the right place. */
- if (outmode != VOIDmode)
+ if (outmode != VOIDmode && retval)
{
if (mem_value)
{
value = hard_libcall_value (outmode);
}
-#ifdef ACCUMULATE_OUTGOING_ARGS
-#ifdef REG_PARM_STACK_SPACE
- if (save_area)
+ if (ACCUMULATE_OUTGOING_ARGS)
{
- enum machine_mode save_mode = GET_MODE (save_area);
+#ifdef REG_PARM_STACK_SPACE
+ if (save_area)
+ {
+ enum machine_mode save_mode = GET_MODE (save_area);
#ifdef ARGS_GROW_DOWNWARD
- rtx stack_area
- = gen_rtx_MEM (save_mode,
- memory_address (save_mode,
- plus_constant (argblock,
- - high_to_save)));
+ rtx stack_area
+ = gen_rtx_MEM (save_mode,
+ memory_address (save_mode,
+ plus_constant (argblock,
+ - high_to_save)));
#else
- rtx stack_area
- = gen_rtx_MEM (save_mode,
- memory_address (save_mode,
- plus_constant (argblock, low_to_save)));
+ rtx stack_area
+ = gen_rtx_MEM (save_mode,
+ memory_address (save_mode,
+ plus_constant (argblock, low_to_save)));
#endif
- if (save_mode != BLKmode)
- emit_move_insn (stack_area, save_area);
- else
- emit_block_move (stack_area, validize_mem (save_area),
- GEN_INT (high_to_save - low_to_save + 1),
- PARM_BOUNDARY / BITS_PER_UNIT);
+ if (save_mode != BLKmode)
+ emit_move_insn (stack_area, save_area);
+ else
+ emit_block_move (stack_area, validize_mem (save_area),
+ GEN_INT (high_to_save - low_to_save + 1),
+ PARM_BOUNDARY);
+ }
+#endif
+
+ /* If we saved any argument areas, restore them. */
+ for (count = 0; count < nargs; count++)
+ if (argvec[count].save_area)
+ {
+ enum machine_mode save_mode = GET_MODE (argvec[count].save_area);
+ rtx stack_area
+ = gen_rtx_MEM (save_mode,
+ memory_address
+ (save_mode,
+ plus_constant (argblock,
+ argvec[count].offset.constant)));
+
+ emit_move_insn (stack_area, argvec[count].save_area);
+ }
+
+ highest_outgoing_arg_in_use = initial_highest_arg_in_use;
+ stack_usage_map = initial_stack_usage_map;
}
+
+ return value;
+
+}
+\f
+/* Output a library call to function FUN (a SYMBOL_REF rtx)
+ (emitting the queue unless NO_QUEUE is nonzero),
+ for a value of mode OUTMODE,
+ with NARGS different arguments, passed as alternating rtx values
+ and machine_modes to convert them to.
+ The rtx values should have been passed through protect_from_queue already.
+
+ FN_TYPE will is zero for `normal' calls, one for `const' calls, wich
+ which will be enclosed in REG_LIBCALL/REG_RETVAL notes and two for `pure'
+ calls, that are handled like `const' calls with extra
+ (use (memory (scratch)). */
+
+void
+emit_library_call VPARAMS((rtx orgfun, enum libcall_type fn_type,
+ enum machine_mode outmode, int nargs, ...))
+{
+#ifndef ANSI_PROTOTYPES
+ rtx orgfun;
+ int fn_type;
+ enum machine_mode outmode;
+ int nargs;
#endif
-
- /* If we saved any argument areas, restore them. */
- for (count = 0; count < nargs; count++)
- if (argvec[count].save_area)
- {
- enum machine_mode save_mode = GET_MODE (argvec[count].save_area);
- rtx stack_area
- = gen_rtx_MEM (save_mode,
- memory_address
- (save_mode,
- plus_constant (argblock,
- argvec[count].offset.constant)));
-
- emit_move_insn (stack_area, argvec[count].save_area);
- }
+ va_list p;
+
+ VA_START (p, nargs);
+
+#ifndef ANSI_PROTOTYPES
+ orgfun = va_arg (p, rtx);
+ fn_type = va_arg (p, int);
+ outmode = va_arg (p, enum machine_mode);
+ nargs = va_arg (p, int);
+#endif
+
+ emit_library_call_value_1 (0, orgfun, NULL_RTX, fn_type, outmode, nargs, p);
+
+ va_end (p);
+}
+\f
+/* Like emit_library_call except that an extra argument, VALUE,
+ comes second and says where to store the result.
+ (If VALUE is zero, this function chooses a convenient way
+ to return the value.
+
+ This function returns an rtx for where the value is to be found.
+ If VALUE is nonzero, VALUE is returned. */
+
+rtx
+emit_library_call_value VPARAMS((rtx orgfun, rtx value,
+ enum libcall_type fn_type,
+ enum machine_mode outmode, int nargs, ...))
+{
+#ifndef ANSI_PROTOTYPES
+ rtx orgfun;
+ rtx value;
+ int fn_type;
+ enum machine_mode outmode;
+ int nargs;
+#endif
+ va_list p;
- highest_outgoing_arg_in_use = initial_highest_arg_in_use;
- stack_usage_map = initial_stack_usage_map;
+ VA_START (p, nargs);
+
+#ifndef ANSI_PROTOTYPES
+ orgfun = va_arg (p, rtx);
+ value = va_arg (p, rtx);
+ fn_type = va_arg (p, int);
+ outmode = va_arg (p, enum machine_mode);
+ nargs = va_arg (p, int);
#endif
+ value = emit_library_call_value_1 (1, orgfun, value, fn_type, outmode, nargs, p);
+
+ va_end (p);
+
return value;
}
\f
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, may_be_alloca, 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 may_be_alloca;
+ int flags;
int variable_size ATTRIBUTE_UNUSED;
int reg_parm_stack_space;
{
rtx reg = 0;
int partial = 0;
int used = 0;
-#ifdef ACCUMULATE_OUTGOING_ARGS
int i, lower_bound = 0, upper_bound = 0;
-#endif
+ 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. */
push_temp_slots ();
-#ifdef ACCUMULATE_OUTGOING_ARGS
- /* If this is being stored into a pre-allocated, fixed-size, stack area,
- save any previous data at that location. */
- if (argblock && ! variable_size && arg->stack)
+ if (ACCUMULATE_OUTGOING_ARGS && !(flags & ECF_SIBCALL))
{
+ /* If this is being stored into a pre-allocated, fixed-size, stack area,
+ save any previous data at that location. */
+ if (argblock && ! variable_size && arg->stack)
+ {
#ifdef ARGS_GROW_DOWNWARD
- /* stack_slot is negative, but we want to index stack_usage_map
- with positive values. */
- if (GET_CODE (XEXP (arg->stack_slot, 0)) == PLUS)
- upper_bound = -INTVAL (XEXP (XEXP (arg->stack_slot, 0), 1)) + 1;
- else
- upper_bound = 0;
+ /* stack_slot is negative, but we want to index stack_usage_map
+ with positive values. */
+ if (GET_CODE (XEXP (arg->stack_slot, 0)) == PLUS)
+ upper_bound = -INTVAL (XEXP (XEXP (arg->stack_slot, 0), 1)) + 1;
+ else
+ upper_bound = 0;
- lower_bound = upper_bound - arg->size.constant;
+ lower_bound = upper_bound - arg->size.constant;
#else
- if (GET_CODE (XEXP (arg->stack_slot, 0)) == PLUS)
- lower_bound = INTVAL (XEXP (XEXP (arg->stack_slot, 0), 1));
- else
- lower_bound = 0;
+ if (GET_CODE (XEXP (arg->stack_slot, 0)) == PLUS)
+ lower_bound = INTVAL (XEXP (XEXP (arg->stack_slot, 0), 1));
+ else
+ lower_bound = 0;
- upper_bound = lower_bound + arg->size.constant;
+ upper_bound = lower_bound + arg->size.constant;
#endif
- for (i = lower_bound; i < upper_bound; i++)
- if (stack_usage_map[i]
- /* Don't store things in the fixed argument area at this point;
- it has already been saved. */
- && i > reg_parm_stack_space)
- break;
-
- if (i != upper_bound)
- {
- /* We need to make a save area. See what mode we can make it. */
- enum machine_mode save_mode
- = mode_for_size (arg->size.constant * BITS_PER_UNIT, MODE_INT, 1);
- rtx stack_area
- = gen_rtx_MEM (save_mode,
- memory_address (save_mode,
- XEXP (arg->stack_slot, 0)));
+ for (i = lower_bound; i < upper_bound; i++)
+ if (stack_usage_map[i]
+ /* Don't store things in the fixed argument area at this point;
+ it has already been saved. */
+ && i > reg_parm_stack_space)
+ break;
- 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)));
- preserve_temp_slots (arg->save_area);
- emit_block_move (validize_mem (arg->save_area), stack_area,
- GEN_INT (arg->size.constant),
- PARM_BOUNDARY / BITS_PER_UNIT);
- }
- else
+ if (i != upper_bound)
{
- arg->save_area = gen_reg_rtx (save_mode);
- emit_move_insn (arg->save_area, stack_area);
+ /* We need to make a save area. See what mode we can make it. */
+ enum machine_mode save_mode
+ = mode_for_size (arg->size.constant * BITS_PER_UNIT, MODE_INT, 1);
+ rtx stack_area
+ = gen_rtx_MEM (save_mode,
+ memory_address (save_mode,
+ XEXP (arg->stack_slot, 0)));
+
+ if (save_mode == BLKmode)
+ {
+ 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,
+ expr_size (arg->tree_value),
+ MIN (PARM_BOUNDARY, TYPE_ALIGN (nt)));
+ }
+ else
+ {
+ arg->save_area = gen_reg_rtx (save_mode);
+ emit_move_insn (arg->save_area, stack_area);
+ }
}
}
+ /* Now that we have saved any slots that will be overwritten by this
+ store, mark all slots this store will use. We must do this before
+ we actually expand the argument since the expansion itself may
+ trigger library calls which might need to use the same stack slot. */
+ if (argblock && ! variable_size && arg->stack)
+ for (i = lower_bound; i < upper_bound; i++)
+ stack_usage_map[i] = 1;
}
- /* Now that we have saved any slots that will be overwritten by this
- store, mark all slots this store will use. We must do this before
- we actually expand the argument since the expansion itself may
- trigger library calls which might need to use the same stack slot. */
- if (argblock && ! variable_size && arg->stack)
- for (i = lower_bound; i < upper_bound; i++)
- stack_usage_map[i] = 1;
-#endif
-
/* If this isn't going to be placed on both the stack and in registers,
set up the register and number of words. */
if (! arg->pass_on_stack)
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)
{
-#ifdef ACCUMULATE_OUTGOING_ARGS
/* stack_arg_under_construction is nonzero if a function argument is
being evaluated directly into the outgoing argument list and
expand_call must take special action to preserve the argument list
if (arg->pass_on_stack)
stack_arg_under_construction++;
-#endif
+
arg->value = expand_expr (pval,
(partial
|| TYPE_MODE (TREE_TYPE (pval)) != arg->mode)
arg->value = convert_modes (arg->mode, TYPE_MODE (TREE_TYPE (pval)),
arg->value, arg->unsignedp);
-#ifdef ACCUMULATE_OUTGOING_ARGS
if (arg->pass_on_stack)
stack_arg_under_construction--;
-#endif
}
/* Don't allow anything left on stack from computation
of argument to alloca. */
- if (may_be_alloca)
+ if (flags & ECF_MAY_BE_ALLOCA)
do_pending_stack_adjust ();
if (arg->value == arg->stack)
/* 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.
partial, reg, used - size, argblock,
ARGS_SIZE_RTX (arg->offset), reg_parm_stack_space,
ARGS_SIZE_RTX (arg->alignment_pad));
-
- arg_space_so_far += used;
}
else
{
excess = (arg->size.constant - int_size_in_bytes (TREE_TYPE (pval))
+ partial * UNITS_PER_WORD);
size_rtx = expr_size (pval);
- arg_space_so_far += excess + INTVAL (size_rtx);
}
+ 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)) / BITS_PER_UNIT, partial,
- reg, excess, argblock, ARGS_SIZE_RTX (arg->offset),
+ TYPE_ALIGN (TREE_TYPE (pval)), partial, reg, excess,
+ argblock, ARGS_SIZE_RTX (arg->offset),
reg_parm_stack_space,
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
