X-Git-Url: http://git.sourceforge.jp/view?a=blobdiff_plain;f=gcc%2Fcalls.c;h=0d0c0f262029a3a36fe47bf5d4b16f0107962ec8;hb=1ebefdcd36f307f94ec94376af932b1472089322;hp=2c952ab1b68ab3dd130ded443a36700b015f5786;hpb=85d654ddd4b94ecbd2d2f291a6b6404fcde249ec;p=pf3gnuchains%2Fgcc-fork.git diff --git a/gcc/calls.c b/gcc/calls.c index 2c952ab1b68..0d0c0f26202 100644 --- a/gcc/calls.c +++ b/gcc/calls.c @@ -1,6 +1,6 @@ /* Convert function calls to rtl insns, for GNU C compiler. Copyright (C) 1989, 1992, 1993, 1994, 1995, 1996, 1997, 1998, - 1999, 2000, 2001, 2002, 2003 Free Software Foundation, Inc. + 1999, 2000, 2001, 2002, 2003, 2004 Free Software Foundation, Inc. This file is part of GCC. @@ -41,10 +41,6 @@ Software Foundation, 59 Temple Place - Suite 330, Boston, MA #include "cgraph.h" #include "except.h" -#ifndef STACK_POINTER_OFFSET -#define STACK_POINTER_OFFSET 0 -#endif - /* Like PREFERRED_STACK_BOUNDARY but in units of bytes, not bits. */ #define STACK_BYTES (PREFERRED_STACK_BOUNDARY / BITS_PER_UNIT) @@ -68,6 +64,9 @@ struct arg_data This is not the same register as for normal calls on machines with register windows. */ rtx tail_call_reg; + /* If REG is a PARALLEL, this is a copy of VALUE pulled into the correct + form for emit_group_move. */ + rtx parallel_value; /* If REG was promoted from the actual mode of the argument expression, indicates whether the promotion is sign- or zero-extended. */ int unsignedp; @@ -121,10 +120,7 @@ static sbitmap stored_args_map; argument list for the constructor call. */ int stack_arg_under_construction; -static int calls_function (tree, int); -static int calls_function_1 (tree, int); - -static void emit_call_1 (rtx, tree, tree, HOST_WIDE_INT, HOST_WIDE_INT, +static void emit_call_1 (rtx, tree, tree, tree, HOST_WIDE_INT, HOST_WIDE_INT, HOST_WIDE_INT, rtx, rtx, int, rtx, int, CUMULATIVE_ARGS *); static void precompute_register_parameters (int, struct arg_data *, int *); @@ -137,7 +133,8 @@ static int compute_argument_block_size (int, struct args_size *, int); static void initialize_argument_information (int, struct arg_data *, struct args_size *, int, tree, tree, CUMULATIVE_ARGS *, int, - rtx *, int *, int *, int *); + rtx *, int *, int *, int *, + bool *, bool); static void compute_argument_addresses (struct arg_data *, rtx, int); static rtx rtx_for_function_call (tree, tree); static void load_register_parameters (struct arg_data *, int, rtx *, int, @@ -145,135 +142,19 @@ static void load_register_parameters (struct arg_data *, int, rtx *, int, static rtx emit_library_call_value_1 (int, rtx, rtx, enum libcall_type, enum machine_mode, int, va_list); static int special_function_p (tree, int); -static rtx try_to_integrate (tree, tree, rtx, int, tree, rtx); static int check_sibcall_argument_overlap_1 (rtx); static int check_sibcall_argument_overlap (rtx, struct arg_data *, int); static int combine_pending_stack_adjustment_and_call (int, struct args_size *, - int); -static tree fix_unsafe_tree (tree); + unsigned int); +static tree split_complex_values (tree); +static tree split_complex_types (tree); #ifdef REG_PARM_STACK_SPACE static rtx save_fixed_argument_area (int, rtx, int *, int *); static void restore_fixed_argument_area (rtx, rtx, int, int); #endif -/* If WHICH is 1, return 1 if EXP contains a call to the built-in function - `alloca'. - - If WHICH is 0, return 1 if EXP contains a call to any function. - Actually, we only need return 1 if evaluating EXP would require pushing - arguments on the stack, but that is too difficult to compute, so we just - assume any function call might require the stack. */ - -static tree calls_function_save_exprs; - -static int -calls_function (tree exp, 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 (tree exp, int which) -{ - int i; - enum tree_code code = TREE_CODE (exp); - 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; - - 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) - && (special_function_p (TREE_OPERAND (TREE_OPERAND (exp, 0), 0), - 0) - & ECF_MAY_BE_ALLOCA)) - return 1; - - break; - - case CONSTRUCTOR: - { - tree tem; - - for (tem = CONSTRUCTOR_ELTS (exp); tem != 0; tem = TREE_CHAIN (tem)) - if (calls_function_1 (TREE_VALUE (tem), which)) - return 1; - } - - return 0; - - case SAVE_EXPR: - if (SAVE_EXPR_RTL (exp) != 0) - return 0; - if (value_member (exp, calls_function_save_exprs)) - return 0; - calls_function_save_exprs = tree_cons (NULL_TREE, exp, - calls_function_save_exprs); - return (TREE_OPERAND (exp, 0) != 0 - && calls_function_1 (TREE_OPERAND (exp, 0), which)); - - case BLOCK: - { - tree local; - 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; - - for (subblock = BLOCK_SUBBLOCKS (exp); - subblock; - subblock = TREE_CHAIN (subblock)) - if (calls_function_1 (subblock, which)) - return 1; - } - return 0; - - case TREE_LIST: - for (; exp != 0; exp = TREE_CHAIN (exp)) - if (calls_function_1 (TREE_VALUE (exp), which)) - return 1; - return 0; - - default: - break; - } - - /* Only expressions and blocks can contain calls. */ - if (! IS_EXPR_CODE_CLASS (class) && class != 'b') - return 0; - - for (i = 0; i < length; i++) - if (TREE_OPERAND (exp, i) != 0 - && calls_function_1 (TREE_OPERAND (exp, i), which)) - return 1; - - return 0; -} - /* Force FUNEXP into a form suitable for the address of a CALL, and return that as an rtx. Also load the static chain register if FNDECL is a nested function. @@ -282,18 +163,10 @@ calls_function_1 (tree exp, int which) CALL_INSN_FUNCTION_USAGE information. */ rtx -prepare_call_address (rtx funexp, tree fndecl, rtx *call_fusage, - int reg_parm_seen, int sibcallp) +prepare_call_address (rtx funexp, rtx static_chain_value, + rtx *call_fusage, int reg_parm_seen, int sibcallp) { - rtx static_chain_value = 0; - - funexp = protect_from_queue (funexp, 0); - - if (fndecl != 0) - /* Get possible static chain value for nested function in C. */ - static_chain_value = lookup_static_chain (fndecl); - - /* Make a valid memory address and copy constants thru pseudo-regs, + /* Make a valid memory address and copy constants through pseudo-regs, but not for a constant address if -fno-function-cse. */ if (GET_CODE (funexp) != SYMBOL_REF) /* If we are using registers for parameters, force the @@ -305,18 +178,16 @@ prepare_call_address (rtx funexp, tree fndecl, rtx *call_fusage, { #ifndef NO_FUNCTION_CSE if (optimize && ! flag_no_function_cse) -#ifdef NO_RECURSIVE_FUNCTION_CSE - if (fndecl != current_function_decl) -#endif - funexp = force_reg (Pmode, funexp); + funexp = force_reg (Pmode, funexp); #endif } if (static_chain_value != 0) { + static_chain_value = convert_memory_address (Pmode, static_chain_value); emit_move_insn (static_chain_rtx, static_chain_value); - if (GET_CODE (static_chain_rtx) == REG) + if (REG_P (static_chain_rtx)) use_reg (call_fusage, static_chain_rtx); } @@ -364,7 +235,8 @@ prepare_call_address (rtx funexp, tree fndecl, rtx *call_fusage, denote registers used by the called function. */ static void -emit_call_1 (rtx funexp, tree fndecl ATTRIBUTE_UNUSED, tree funtype ATTRIBUTE_UNUSED, +emit_call_1 (rtx funexp, tree fntree, tree fndecl ATTRIBUTE_UNUSED, + tree funtype ATTRIBUTE_UNUSED, HOST_WIDE_INT stack_size ATTRIBUTE_UNUSED, HOST_WIDE_INT rounded_stack_size, HOST_WIDE_INT struct_value_size ATTRIBUTE_UNUSED, @@ -481,7 +353,7 @@ emit_call_1 (rtx funexp, tree fndecl ATTRIBUTE_UNUSED, tree funtype ATTRIBUTE_UN } else #endif - abort (); + gcc_unreachable (); /* Find the call we just emitted. */ call_insn = last_call_insn (); @@ -508,7 +380,16 @@ emit_call_1 (rtx funexp, tree fndecl ATTRIBUTE_UNUSED, tree funtype ATTRIBUTE_UN REG_NOTES (call_insn) = gen_rtx_EXPR_LIST (REG_EH_REGION, const0_rtx, REG_NOTES (call_insn)); else - note_eh_region_may_contain_throw (); + { + int rn = lookup_stmt_eh_region (fntree); + + /* If rn < 0, then either (1) tree-ssa not used or (2) doesn't + throw, which we already took care of. */ + if (rn > 0) + REG_NOTES (call_insn) = gen_rtx_EXPR_LIST (REG_EH_REGION, GEN_INT (rn), + REG_NOTES (call_insn)); + note_current_region_may_contain_throw (); + } if (ecf_flags & ECF_NORETURN) REG_NOTES (call_insn) = gen_rtx_EXPR_LIST (REG_NORETURN, const0_rtx, @@ -530,10 +411,6 @@ emit_call_1 (rtx funexp, tree fndecl ATTRIBUTE_UNUSED, tree funtype ATTRIBUTE_UN if the context of the call as a whole permits. */ inhibit_defer_pop = old_inhibit_defer_pop; - /* Don't bother cleaning up after a noreturn function. */ - if (ecf_flags & (ECF_NORETURN | ECF_LONGJMP)) - return; - if (n_popped > 0) { if (!already_popped) @@ -557,7 +434,7 @@ emit_call_1 (rtx funexp, tree fndecl ATTRIBUTE_UNUSED, tree funtype ATTRIBUTE_UN if (rounded_stack_size != 0) { - if (ecf_flags & ECF_SP_DEPRESSED) + if (ecf_flags & (ECF_SP_DEPRESSED | ECF_NORETURN)) /* Just pretend we did the pop. */ stack_pointer_delta -= rounded_stack_size; else if (flag_defer_pop && inhibit_defer_pop == 0 @@ -596,8 +473,7 @@ emit_call_1 (rtx funexp, tree fndecl ATTRIBUTE_UNUSED, tree funtype ATTRIBUTE_UN static int special_function_p (tree fndecl, int flags) { - if (! (flags & ECF_MALLOC) - && fndecl && DECL_NAME (fndecl) + if (fndecl && DECL_NAME (fndecl) && IDENTIFIER_LENGTH (DECL_NAME (fndecl)) <= 17 /* Exclude functions not at the file scope, or not `extern', since they are not the magic functions we would otherwise @@ -606,7 +482,7 @@ special_function_p (tree fndecl, int flags) hacky imitation of DECL_ASSEMBLER_NAME. It's (also) wrong because you can declare fork() inside a function if you wish. */ - && (DECL_CONTEXT (fndecl) == NULL_TREE + && (DECL_CONTEXT (fndecl) == NULL_TREE || TREE_CODE (DECL_CONTEXT (fndecl)) == TRANSLATION_UNIT_DECL) && TREE_PUBLIC (fndecl)) { @@ -648,7 +524,7 @@ special_function_p (tree fndecl, int flags) if (tname[1] == 'i' && ! strcmp (tname, "siglongjmp")) - flags |= ECF_LONGJMP; + flags |= ECF_NORETURN; } else if ((tname[0] == 'q' && tname[1] == 's' && ! strcmp (tname, "qsetjmp")) @@ -658,22 +534,9 @@ special_function_p (tree fndecl, int flags) else if (tname[0] == 'l' && tname[1] == 'o' && ! strcmp (tname, "longjmp")) - flags |= ECF_LONGJMP; - - else if ((tname[0] == 'f' && tname[1] == 'o' - && ! strcmp (tname, "fork")) - /* Linux specific: __clone. check NAME to insist on the - leading underscores, to avoid polluting the ISO / POSIX - namespace. */ - || (name[0] == '_' && name[1] == '_' - && ! strcmp (tname, "clone")) - || (tname[0] == 'e' && tname[1] == 'x' && tname[2] == 'e' - && tname[3] == 'c' && (tname[4] == 'l' || tname[4] == 'v') - && (tname[5] == '\0' - || ((tname[5] == 'p' || tname[5] == 'e') - && tname[6] == '\0')))) - flags |= ECF_FORK_OR_EXEC; + flags |= ECF_NORETURN; } + return flags; } @@ -732,10 +595,11 @@ flags_from_decl_or_type (tree exp) flags |= ECF_NOTHROW; if (TREE_READONLY (exp) && ! TREE_THIS_VOLATILE (exp)) - flags |= ECF_LIBCALL_BLOCK; - } + flags |= ECF_LIBCALL_BLOCK | ECF_CONST; - if (TREE_READONLY (exp) && ! TREE_THIS_VOLATILE (exp)) + flags = special_function_p (exp, flags); + } + else if (TYPE_P (exp) && TYPE_READONLY (exp) && ! TREE_THIS_VOLATILE (exp)) flags |= ECF_CONST; if (TREE_THIS_VOLATILE (exp)) @@ -752,6 +616,28 @@ flags_from_decl_or_type (tree exp) return flags; } +/* Detect flags from a CALL_EXPR. */ + +int +call_expr_flags (tree t) +{ + int flags; + tree decl = get_callee_fndecl (t); + + if (decl) + flags = flags_from_decl_or_type (decl); + else + { + t = TREE_TYPE (TREE_OPERAND (t, 0)); + if (t && TREE_CODE (t) == POINTER_TYPE) + flags = flags_from_decl_or_type (TREE_TYPE (t)); + else + flags = 0; + } + + return flags; +} + /* Precompute all register parameters as described by ARGS, storing values into fields within the ARGS array. @@ -760,7 +646,8 @@ flags_from_decl_or_type (tree exp) Set REG_PARM_SEEN if we encounter a register parameter. */ static void -precompute_register_parameters (int num_actuals, struct arg_data *args, int *reg_parm_seen) +precompute_register_parameters (int num_actuals, struct arg_data *args, + int *reg_parm_seen) { int i; @@ -778,10 +665,6 @@ precompute_register_parameters (int num_actuals, struct arg_data *args, int *reg VOIDmode, 0); preserve_temp_slots (args[i].value); pop_temp_slots (); - - /* ANSI doesn't require a sequence point here, - but PCC has one, so this will avoid some problems. */ - emit_queue (); } /* If the value is a non-legitimate constant, force it into a @@ -799,6 +682,17 @@ precompute_register_parameters (int num_actuals, struct arg_data *args, int *reg TYPE_MODE (TREE_TYPE (args[i].tree_value)), args[i].value, args[i].unsignedp); + /* If we're going to have to load the value by parts, pull the + parts into pseudos. The part extraction process can involve + non-trivial computation. */ + if (GET_CODE (args[i].reg) == PARALLEL) + { + tree type = TREE_TYPE (args[i].tree_value); + args[i].parallel_value + = emit_group_load_into_temps (args[i].reg, args[i].value, + type, int_size_in_bytes (type)); + } + /* 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. @@ -807,13 +701,13 @@ precompute_register_parameters (int num_actuals, struct arg_data *args, int *reg register parameters. This is to avoid reload conflicts while loading the parameters registers. */ - if ((! (GET_CODE (args[i].value) == 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) > COSTS_N_INSNS (1) - && ((SMALL_REGISTER_CLASSES && *reg_parm_seen) - || preserve_subexpressions_p ())) + else if ((! (REG_P (args[i].value) + || (GET_CODE (args[i].value) == SUBREG + && REG_P (SUBREG_REG (args[i].value))))) + && args[i].mode != BLKmode + && rtx_cost (args[i].value, SET) > COSTS_N_INSNS (1) + && ((SMALL_REGISTER_CLASSES && *reg_parm_seen) + || optimize)) args[i].value = copy_to_mode_reg (args[i].mode, args[i].value); } } @@ -938,10 +832,19 @@ store_unaligned_arguments_into_pseudos (struct arg_data *args, int num_actuals) < (unsigned int) MIN (BIGGEST_ALIGNMENT, BITS_PER_WORD))) { int bytes = int_size_in_bytes (TREE_TYPE (args[i].tree_value)); - int nregs = (bytes + UNITS_PER_WORD - 1) / UNITS_PER_WORD; int endian_correction = 0; - args[i].n_aligned_regs = args[i].partial ? args[i].partial : nregs; + if (args[i].partial) + { + gcc_assert (args[i].partial % UNITS_PER_WORD == 0); + args[i].n_aligned_regs = args[i].partial / UNITS_PER_WORD; + } + else + { + args[i].n_aligned_regs + = (bytes + UNITS_PER_WORD - 1) / UNITS_PER_WORD; + } + args[i].aligned_regs = xmalloc (sizeof (rtx) * args[i].n_aligned_regs); /* Structures smaller than a word are normally aligned to the @@ -967,7 +870,7 @@ store_unaligned_arguments_into_pseudos (struct arg_data *args, int num_actuals) args[i].aligned_regs[j] = reg; word = extract_bit_field (word, bitsize, 0, 1, NULL_RTX, - word_mode, word_mode, BITS_PER_WORD); + word_mode, word_mode); /* There is no need to restrict this code to loading items in TYPE_ALIGN sized hunks. The bitfield instructions can @@ -984,7 +887,7 @@ store_unaligned_arguments_into_pseudos (struct arg_data *args, int num_actuals) bytes -= bitsize / BITS_PER_UNIT; store_bit_field (reg, bitsize, endian_correction, word_mode, - word, BITS_PER_WORD); + word); } } } @@ -1008,7 +911,13 @@ store_unaligned_arguments_into_pseudos (struct arg_data *args, int num_actuals) and may be modified by this routine. OLD_PENDING_ADJ, MUST_PREALLOCATE and FLAGS are pointers to integer - flags which may may be modified by this routine. */ + flags which may may be modified by this routine. + + MAY_TAILCALL is cleared if we encounter an invisible pass-by-reference + that requires allocation of stack space. + + CALL_FROM_THUNK_P is true if this call is the jump from a thunk to + the thunked-to function. */ static void initialize_argument_information (int num_actuals ATTRIBUTE_UNUSED, @@ -1019,7 +928,8 @@ initialize_argument_information (int num_actuals ATTRIBUTE_UNUSED, CUMULATIVE_ARGS *args_so_far, int reg_parm_stack_space, rtx *old_stack_level, int *old_pending_adj, - int *must_preallocate, int *ecf_flags) + int *must_preallocate, int *ecf_flags, + bool *may_tailcall, bool call_from_thunk_p) { /* 1 if scanning parms front to back, -1 if scanning back to front. */ int inc; @@ -1072,7 +982,7 @@ initialize_argument_information (int num_actuals ATTRIBUTE_UNUSED, args[i].reg is nonzero if all or part is passed in registers. args[i].partial is nonzero if part but not all is passed in registers, - and the exact value says how many words are passed in registers. + and the exact value says how many bytes are passed in registers. args[i].pass_on_stack is nonzero if the argument must at least be computed on the stack. It may then be loaded back into registers @@ -1082,56 +992,33 @@ initialize_argument_information (int num_actuals ATTRIBUTE_UNUSED, with those made by function.c. */ /* See if this argument should be passed by invisible reference. */ - if (CONTAINS_PLACEHOLDER_P (TYPE_SIZE (type)) - || TREE_ADDRESSABLE (type) -#ifdef FUNCTION_ARG_PASS_BY_REFERENCE - || FUNCTION_ARG_PASS_BY_REFERENCE (*args_so_far, TYPE_MODE (type), - type, argpos < n_named_args) -#endif - ) + if (pass_by_reference (args_so_far, TYPE_MODE (type), + type, argpos < n_named_args)) { - /* If we're compiling a thunk, pass through invisible - references instead of making a copy. */ - if (current_function_is_thunk -#ifdef FUNCTION_ARG_CALLEE_COPIES - || (FUNCTION_ARG_CALLEE_COPIES (*args_so_far, TYPE_MODE (type), - type, argpos < n_named_args) - /* If it's in a register, we must make a copy of it too. */ - /* ??? Is this a sufficient test? Is there a better one? */ - && !(TREE_CODE (args[i].tree_value) == VAR_DECL - && REG_P (DECL_RTL (args[i].tree_value))) - && ! TREE_ADDRESSABLE (type)) -#endif - ) + bool callee_copies; + tree base; + + callee_copies + = reference_callee_copied (args_so_far, TYPE_MODE (type), + type, argpos < n_named_args); + + /* If we're compiling a thunk, pass through invisible references + instead of making a copy. */ + if (call_from_thunk_p + || (callee_copies + && !TREE_ADDRESSABLE (type) + && (base = get_base_address (args[i].tree_value)) + && (!DECL_P (base) || MEM_P (DECL_RTL (base))))) { - /* C++ uses a TARGET_EXPR to indicate that we want to make a - new object from the argument. If we are passing by - invisible reference, the callee will do that for us, so we - can strip off the TARGET_EXPR. This is not always safe, - 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. */ - - if (TREE_CODE (args[i].tree_value) == TARGET_EXPR - && (DECL_P (TREE_OPERAND (args[i].tree_value, 1))) - && ! REG_P (DECL_RTL (TREE_OPERAND (args[i].tree_value, 1)))) - args[i].tree_value = TREE_OPERAND (args[i].tree_value, 1); - - args[i].tree_value = build1 (ADDR_EXPR, - build_pointer_type (type), - args[i].tree_value); - type = build_pointer_type (type); - } - else if (TREE_CODE (args[i].tree_value) == TARGET_EXPR) - { - /* In the V3 C++ ABI, parameters are destroyed in the caller. - We implement this by passing the address of the temporary - rather than expanding it into another allocated slot. */ - args[i].tree_value = build1 (ADDR_EXPR, - build_pointer_type (type), - args[i].tree_value); - type = build_pointer_type (type); + /* We can't use sibcalls if a callee-copied argument is + stored in the current function's frame. */ + if (!call_from_thunk_p && DECL_P (base) && !TREE_STATIC (base)) + *may_tailcall = false; + + args[i].tree_value = build_fold_addr_expr (args[i].tree_value); + type = TREE_TYPE (args[i].tree_value); + + *ecf_flags &= ~(ECF_CONST | ECF_LIBCALL_BLOCK); } else { @@ -1165,17 +1052,21 @@ initialize_argument_information (int num_actuals ATTRIBUTE_UNUSED, copy = assign_temp (type, 0, 1, 0); store_expr (args[i].tree_value, copy, 0); - *ecf_flags &= ~(ECF_CONST | ECF_PURE | ECF_LIBCALL_BLOCK); - args[i].tree_value = build1 (ADDR_EXPR, - build_pointer_type (type), - make_tree (type, copy)); - type = build_pointer_type (type); + if (callee_copies) + *ecf_flags &= ~(ECF_CONST | ECF_LIBCALL_BLOCK); + else + *ecf_flags &= ~(ECF_CONST | ECF_PURE | ECF_LIBCALL_BLOCK); + + args[i].tree_value + = build_fold_addr_expr (make_tree (type, copy)); + type = TREE_TYPE (args[i].tree_value); + *may_tailcall = false; } } mode = TYPE_MODE (type); - unsignedp = TREE_UNSIGNED (type); + unsignedp = TYPE_UNSIGNED (type); if (targetm.calls.promote_function_args (fndecl ? TREE_TYPE (fndecl) : 0)) mode = promote_mode (type, mode, &unsignedp, 1); @@ -1195,14 +1086,12 @@ initialize_argument_information (int num_actuals ATTRIBUTE_UNUSED, args[i].tail_call_reg = args[i].reg; #endif -#ifdef FUNCTION_ARG_PARTIAL_NREGS if (args[i].reg) args[i].partial - = FUNCTION_ARG_PARTIAL_NREGS (*args_so_far, mode, type, - argpos < n_named_args); -#endif + = targetm.calls.arg_partial_bytes (args_so_far, mode, type, + argpos < n_named_args); - args[i].pass_on_stack = MUST_PASS_IN_STACK (mode, type); + args[i].pass_on_stack = targetm.calls.must_pass_in_stack (mode, type); /* If FUNCTION_ARG returned a (parallel [(expr_list (nil) ...) ...]), it means that we are to pass this arg in the register(s) designated @@ -1294,8 +1183,7 @@ compute_argument_block_size (int reg_parm_stack_space, /* We don't handle this case yet. To handle it correctly we have to add the delta, round and subtract the delta. Currently no machine description requires this support. */ - if (stack_pointer_delta & (preferred_stack_boundary - 1)) - abort (); + gcc_assert (!(stack_pointer_delta & (preferred_stack_boundary - 1))); args_size->var = round_up (args_size->var, preferred_stack_boundary); } @@ -1329,11 +1217,6 @@ compute_argument_block_size (int reg_parm_stack_space, args_size->constant = MAX (args_size->constant, reg_parm_stack_space); -#ifdef MAYBE_REG_PARM_STACK_SPACE - if (reg_parm_stack_space == 0) - args_size->constant = 0; -#endif - #ifndef OUTGOING_REG_PARM_STACK_SPACE args_size->constant -= reg_parm_stack_space; #endif @@ -1356,63 +1239,43 @@ precompute_arguments (int flags, int num_actuals, struct arg_data *args) { int i; - /* If this function call is cse'able, precompute all the parameters. - Note that if the parameter is constructed into a temporary, this will - cause an additional copy because the parameter will be constructed - into a temporary location and then copied into the outgoing arguments. - If a parameter contains a call to alloca and this function uses the - stack, precompute the parameter. */ - - /* If we preallocated the stack space, and some arguments must be passed - 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. (we have code to avoid - such case by saving the outgoing stack arguments, but it results in - worse code) */ + /* If this is a libcall, then precompute all arguments so that we do not + get extraneous instructions emitted as part of the libcall sequence. */ + if ((flags & ECF_LIBCALL_BLOCK) == 0) + return; for (i = 0; i < num_actuals; i++) - if ((flags & ECF_LIBCALL_BLOCK) - || 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].value - = expand_expr (args[i].tree_value, NULL_RTX, VOIDmode, 0); + { + enum machine_mode mode; - /* ANSI doesn't require a sequence point here, - but PCC has one, so this will avoid some problems. */ - emit_queue (); + /* If this is an addressable type, we cannot pre-evaluate it. */ + gcc_assert (!TREE_ADDRESSABLE (TREE_TYPE (args[i].tree_value))); - args[i].initial_value = args[i].value - = protect_from_queue (args[i].value, 0); + args[i].initial_value = args[i].value + = expand_expr (args[i].tree_value, NULL_RTX, VOIDmode, 0); - mode = TYPE_MODE (TREE_TYPE (args[i].tree_value)); - if (mode != args[i].mode) - { - args[i].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 - pseudo, so convert it down to the declared mode using - a SUBREG. */ - if (GET_CODE (args[i].value) == REG - && GET_MODE_CLASS (args[i].mode) == MODE_INT) - { - args[i].initial_value - = gen_lowpart_SUBREG (mode, args[i].value); - SUBREG_PROMOTED_VAR_P (args[i].initial_value) = 1; - SUBREG_PROMOTED_UNSIGNED_SET (args[i].initial_value, - args[i].unsignedp); - } + mode = TYPE_MODE (TREE_TYPE (args[i].tree_value)); + if (mode != args[i].mode) + { + args[i].value + = convert_modes (args[i].mode, mode, + args[i].value, args[i].unsignedp); +#if defined(PROMOTE_FUNCTION_MODE) && !defined(PROMOTE_MODE) + /* CSE will replace this only if it contains args[i].value + pseudo, so convert it down to the declared mode using + a SUBREG. */ + if (REG_P (args[i].value) + && GET_MODE_CLASS (args[i].mode) == MODE_INT) + { + args[i].initial_value + = gen_lowpart_SUBREG (mode, args[i].value); + SUBREG_PROMOTED_VAR_P (args[i].initial_value) = 1; + SUBREG_PROMOTED_UNSIGNED_SET (args[i].initial_value, + args[i].unsignedp); + } #endif - } - } + } + } } /* Given the current state of MUST_PREALLOCATE and information about @@ -1564,7 +1427,6 @@ rtx_for_function_call (tree fndecl, tree addr) push_temp_slots (); funexp = expand_expr (addr, NULL_RTX, VOIDmode, 0); pop_temp_slots (); /* FUNEXP can't be BLKmode. */ - emit_queue (); } return funexp; } @@ -1586,11 +1448,7 @@ load_register_parameters (struct arg_data *args, int num_actuals, { int i, j; -#ifdef LOAD_ARGS_REVERSED - for (i = num_actuals - 1; i >= 0; i--) -#else for (i = 0; i < num_actuals; i++) -#endif { rtx reg = ((flags & ECF_SIBCALL) ? args[i].tail_call_reg : args[i].reg); @@ -1605,8 +1463,13 @@ load_register_parameters (struct arg_data *args, int num_actuals, we just use a normal move insn. This value can be zero if the argument is a zero size structure with no fields. */ nregs = -1; - if (partial) - nregs = partial; + if (GET_CODE (reg) == PARALLEL) + ; + else if (partial) + { + gcc_assert (partial % UNITS_PER_WORD == 0); + nregs = partial / UNITS_PER_WORD; + } else if (TYPE_MODE (TREE_TYPE (args[i].tree_value)) == BLKmode) { size = int_size_in_bytes (TREE_TYPE (args[i].tree_value)); @@ -1619,11 +1482,7 @@ load_register_parameters (struct arg_data *args, int num_actuals, locations. The Irix 6 ABI has examples of this. */ if (GET_CODE (reg) == PARALLEL) - { - tree type = TREE_TYPE (args[i].tree_value); - emit_group_load (reg, args[i].value, type, - int_size_in_bytes (type)); - } + emit_group_move (reg, args[i].parallel_value); /* If simple case, just do move. If normal partial, store_one_arg has already loaded the register for us. In all other cases, @@ -1648,8 +1507,9 @@ load_register_parameters (struct arg_data *args, int num_actuals, call only uses SIZE bytes at the msb end, but it doesn't seem worth generating rtl to say that. */ reg = gen_rtx_REG (word_mode, REGNO (reg)); - x = expand_binop (word_mode, ashl_optab, reg, - GEN_INT (shift), reg, 1, OPTAB_WIDEN); + x = expand_shift (LSHIFT_EXPR, word_mode, reg, + build_int_cst (NULL_TREE, shift), + reg, 1); if (x != reg) emit_move_insn (reg, x); } @@ -1668,25 +1528,30 @@ load_register_parameters (struct arg_data *args, int num_actuals, { rtx mem = validize_mem (args[i].value); -#ifdef BLOCK_REG_PADDING /* Handle a BLKmode that needs shifting. */ if (nregs == 1 && size < UNITS_PER_WORD - && args[i].locate.where_pad == downward) +#ifdef BLOCK_REG_PADDING + && args[i].locate.where_pad == downward +#else + && BYTES_BIG_ENDIAN +#endif + ) { rtx tem = operand_subword_force (mem, 0, args[i].mode); rtx ri = gen_rtx_REG (word_mode, REGNO (reg)); rtx x = gen_reg_rtx (word_mode); int shift = (UNITS_PER_WORD - size) * BITS_PER_UNIT; - optab dir = BYTES_BIG_ENDIAN ? lshr_optab : ashl_optab; + enum tree_code dir = BYTES_BIG_ENDIAN ? RSHIFT_EXPR + : LSHIFT_EXPR; emit_move_insn (x, tem); - x = expand_binop (word_mode, dir, x, GEN_INT (shift), - ri, 1, OPTAB_WIDEN); + x = expand_shift (dir, word_mode, x, + build_int_cst (NULL_TREE, shift), + ri, 1); if (x != ri) emit_move_insn (ri, x); } else -#endif move_block_to_reg (REGNO (reg), mem, nregs, args[i].mode); } @@ -1703,128 +1568,10 @@ load_register_parameters (struct arg_data *args, int num_actuals, use_group_regs (call_fusage, reg); else if (nregs == -1) use_reg (call_fusage, reg); - else - use_regs (call_fusage, REGNO (reg), nregs == 0 ? 1 : nregs); - } - } -} - -/* Try to integrate function. See expand_inline_function for documentation - about the parameters. */ - -static rtx -try_to_integrate (tree fndecl, tree actparms, rtx target, int ignore, - tree type, rtx structure_value_addr) -{ - rtx temp; - rtx before_call; - int i; - rtx old_stack_level = 0; - int reg_parm_stack_space = 0; - -#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 - - before_call = get_last_insn (); - - timevar_push (TV_INTEGRATION); - - temp = expand_inline_function (fndecl, actparms, target, - ignore, type, - structure_value_addr); - - timevar_pop (TV_INTEGRATION); - - /* If inlining succeeded, return. */ - if (temp != (rtx) (size_t) - 1) - { - if (ACCUMULATE_OUTGOING_ARGS) - { - /* If the outgoing argument list must be preserved, push - the stack before executing the inlined function if it - makes any calls. */ - - i = reg_parm_stack_space; - if (i > highest_outgoing_arg_in_use) - i = highest_outgoing_arg_in_use; - while (--i >= 0 && stack_usage_map[i] == 0) - ; - - if (stack_arg_under_construction || i >= 0) - { - rtx first_insn - = before_call ? NEXT_INSN (before_call) : get_insns (); - 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 == 0) - for (insn = first_insn; insn; insn = NEXT_INSN (insn)) - if (GET_CODE (insn) == CALL_INSN) - break; - - if (insn) - { - /* Reserve enough stack space so that the largest - argument list of any function call in the inline - function does not overlap the argument list being - evaluated. This is usually an overestimate because - allocate_dynamic_stack_space reserves space for an - 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. - - Add the stack space reserved for register arguments, if - any, in the inline function. What is really needed is the - largest value of reg_parm_stack_space in the inline - function, but that is not available. Using the current - value of reg_parm_stack_space is wrong, but gives - correct results on all supported machines. */ - - int adjust = (DECL_SAVED_INSNS (fndecl)->outgoing_args_size - + reg_parm_stack_space); - - start_sequence (); - emit_stack_save (SAVE_BLOCK, &old_stack_level, NULL_RTX); - allocate_dynamic_stack_space (GEN_INT (adjust), - NULL_RTX, BITS_PER_UNIT); - seq = get_insns (); - end_sequence (); - emit_insn_before (seq, first_insn); - emit_stack_restore (SAVE_BLOCK, old_stack_level, NULL_RTX); - } - } + else if (nregs > 0) + use_regs (call_fusage, REGNO (reg), nregs); } - - /* 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 ("%Hinlining failed in call to '%F'", - &DECL_SOURCE_LOCATION (fndecl), fndecl); - warning ("called from here"); } - (*lang_hooks.mark_addressable) (fndecl); - return (rtx) (size_t) - 1; } /* We need to pop PENDING_STACK_ADJUST bytes. But, if the arguments @@ -1839,14 +1586,14 @@ try_to_integrate (tree fndecl, tree actparms, rtx target, int ignore, static int combine_pending_stack_adjustment_and_call (int unadjusted_args_size, struct args_size *args_size, - int preferred_unit_stack_boundary) + unsigned 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; + unsigned HOST_WIDE_INT unadjusted_alignment; unadjusted_alignment = ((stack_pointer_delta + unadjusted_args_size) @@ -1912,7 +1659,7 @@ check_sibcall_argument_overlap_1 (rtx x) && GET_CODE (XEXP (XEXP (x, 0), 1)) == CONST_INT) i = INTVAL (XEXP (XEXP (x, 0), 1)); else - return 0; + return 1; #ifdef ARGS_GROW_DOWNWARD i = -i - GET_MODE_SIZE (GET_MODE (x)); @@ -1981,32 +1728,51 @@ check_sibcall_argument_overlap (rtx insn, struct arg_data *arg, int mark_stored_ return insn != NULL_RTX; } -static tree -fix_unsafe_tree (tree t) +/* Given that a function returns a value of mode MODE at the most + significant end of hard register VALUE, shift VALUE left or right + as specified by LEFT_P. Return true if some action was needed. */ + +bool +shift_return_value (enum machine_mode mode, bool left_p, rtx value) { - switch (unsafe_for_reeval (t)) - { - case 0: /* Safe. */ - break; + HOST_WIDE_INT shift; - case 1: /* Mildly unsafe. */ - t = unsave_expr (t); - break; + gcc_assert (REG_P (value) && HARD_REGISTER_P (value)); + shift = GET_MODE_BITSIZE (GET_MODE (value)) - GET_MODE_BITSIZE (mode); + if (shift == 0) + return false; - case 2: /* Wildly unsafe. */ - { - tree var = build_decl (VAR_DECL, NULL_TREE, - TREE_TYPE (t)); - SET_DECL_RTL (var, - expand_expr (t, NULL_RTX, VOIDmode, EXPAND_NORMAL)); - t = var; - } - break; + /* Use ashr rather than lshr for right shifts. This is for the benefit + of the MIPS port, which requires SImode values to be sign-extended + when stored in 64-bit registers. */ + if (!force_expand_binop (GET_MODE (value), left_p ? ashl_optab : ashr_optab, + value, GEN_INT (shift), value, 1, OPTAB_WIDEN)) + gcc_unreachable (); + return true; +} + +/* Remove all REG_EQUIV notes found in the insn chain. */ + +static void +purge_reg_equiv_notes (void) +{ + rtx insn; - default: - abort (); + for (insn = get_insns (); insn; insn = NEXT_INSN (insn)) + { + while (1) + { + rtx note = find_reg_note (insn, REG_EQUIV, 0); + if (note) + { + /* Remove the note and keep looking at the notes for + this insn. */ + remove_note (insn, note); + continue; + } + break; + } } - return t; } /* Generate all the code for a function call @@ -2025,11 +1791,9 @@ expand_call (tree exp, rtx target, int ignore) 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". */ + /* Sequence of insns to perform a tail "call". */ rtx tail_call_insns = NULL_RTX; /* Data type of the function. */ tree funtype; @@ -2037,9 +1801,9 @@ expand_call (tree exp, rtx target, int ignore) /* 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; + /* The type of the function being called. */ + tree fntype; + bool try_tail_call = CALL_EXPR_TAILCALL (exp); int pass; /* Register in which non-BLKmode value will be returned, @@ -2100,8 +1864,6 @@ expand_call (tree exp, rtx target, int ignore) /* 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 saved, if any. */ @@ -2130,65 +1892,30 @@ expand_call (tree exp, rtx target, int ignore) tree addr = TREE_OPERAND (exp, 0); int i; /* The alignment of the stack, in bits. */ - HOST_WIDE_INT preferred_stack_boundary; + unsigned HOST_WIDE_INT preferred_stack_boundary; /* The alignment of the stack, in bytes. */ - HOST_WIDE_INT preferred_unit_stack_boundary; - + unsigned HOST_WIDE_INT preferred_unit_stack_boundary; + /* The static chain value to use for this call. */ + rtx static_chain_value; /* 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. */ - + /* See if we can find a DECL-node for the actual function, and get the + function attributes (flags) from the function decl or type node. */ fndecl = get_callee_fndecl (exp); if (fndecl) { - 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 ("%Hcan't inline call to '%F'", - &DECL_SOURCE_LOCATION (fndecl), fndecl); - warning ("called from here"); - } - (*lang_hooks.mark_addressable) (fndecl); - } - - if (ignore - && lookup_attribute ("warn_unused_result", - TYPE_ATTRIBUTES (TREE_TYPE (fndecl)))) - warning ("ignoring return value of `%D', " - "declared with attribute warn_unused_result", fndecl); - + fntype = TREE_TYPE (fndecl); 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 { - if (ignore - && lookup_attribute ("warn_unused_result", - TYPE_ATTRIBUTES (TREE_TYPE (TREE_TYPE (p))))) - warning ("ignoring return value of function " - "declared with attribute warn_unused_result"); - flags |= flags_from_decl_or_type (TREE_TYPE (TREE_TYPE (p))); + fntype = TREE_TYPE (TREE_TYPE (p)); + flags |= flags_from_decl_or_type (fntype); } - struct_value = targetm.calls.struct_value_rtx (fndecl ? TREE_TYPE (fndecl) : 0, 0); + struct_value = targetm.calls.struct_value_rtx (fntype, 0); /* Warn if this value is an aggregate type, regardless of which calling convention we are using for it. */ @@ -2222,12 +1949,8 @@ expand_call (tree exp, rtx target, int ignore) } #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 #ifndef OUTGOING_REG_PARM_STACK_SPACE if (reg_parm_stack_space > 0 && PUSH_ARGS) @@ -2245,15 +1968,6 @@ expand_call (tree exp, rtx target, int ignore) #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)) - (*lang_hooks.mark_addressable) (fndecl); - is_integrable = 0; - } } #else /* not PCC_STATIC_STRUCT_RETURN */ { @@ -2269,7 +1983,7 @@ expand_call (tree exp, rtx target, int ignore) structure_value_addr = expand_expr (return_arg, NULL_RTX, VOIDmode, EXPAND_NORMAL); } - else if (target && GET_CODE (target) == MEM) + else if (target && MEM_P (target)) structure_value_addr = XEXP (target, 0); else { @@ -2286,17 +2000,6 @@ expand_call (tree exp, rtx target, int ignore) #endif /* not PCC_STATIC_STRUCT_RETURN */ } - /* If called function is inline, try to integrate it. */ - - if (is_integrable) - { - rtx temp = try_to_integrate (fndecl, actparms, target, - ignore, TREE_TYPE (exp), - structure_value_addr); - if (temp != (rtx) (size_t) - 1) - return temp; - } - /* Figure out the amount to which the stack should be aligned. */ preferred_stack_boundary = PREFERRED_STACK_BOUNDARY; if (fndecl) @@ -2308,13 +2011,12 @@ expand_call (tree exp, rtx target, int ignore) /* Operand 0 is a pointer-to-function; get the type of the function. */ funtype = TREE_TYPE (addr); - if (! POINTER_TYPE_P (funtype)) - abort (); + gcc_assert (POINTER_TYPE_P (funtype)); funtype = TREE_TYPE (funtype); /* Munge the tree to split complex arguments into their imaginary and real parts. */ - if (SPLIT_COMPLEX_ARGS) + if (targetm.calls.split_complex_arg) { type_arg_types = split_complex_types (TYPE_ARG_TYPES (funtype)); actparms = split_complex_values (actparms); @@ -2322,10 +2024,6 @@ expand_call (tree exp, rtx target, int ignore) else type_arg_types = TYPE_ARG_TYPES (funtype); - /* See if this is a call to a function that can return more than once - or a call to longjmp or malloc. */ - flags |= special_function_p (fndecl, flags); - if (flags & ECF_MAY_BE_ALLOCA) current_function_calls_alloca = 1; @@ -2338,11 +2036,12 @@ expand_call (tree exp, rtx target, int ignore) 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 + rtx temp = (!REG_P (structure_value_addr) || (ACCUMULATE_OUTGOING_ARGS && stack_arg_under_construction && structure_value_addr == virtual_outgoing_args_rtx) - ? copy_addr_to_reg (structure_value_addr) + ? copy_addr_to_reg (convert_memory_address + (Pmode, structure_value_addr)) : structure_value_addr); actparms @@ -2357,41 +2056,53 @@ expand_call (tree exp, rtx target, int ignore) for (p = actparms, num_actuals = 0; p; p = TREE_CHAIN (p)) num_actuals++; + /* Compute number of named args. + First, do a raw count of the args for INIT_CUMULATIVE_ARGS. */ + + if (type_arg_types != 0) + n_named_args + = (list_length (type_arg_types) + /* Count the struct value address, if it is passed as a parm. */ + + structure_value_addr_parm); + else + /* If we know nothing, treat all args as named. */ + n_named_args = num_actuals; + /* Start updating where the next arg would go. - On some machines (such as the PA) indirect calls have a difuferent - calling convention than normal calls. The last argument in + On some machines (such as the PA) indirect calls have a different + calling convention than normal calls. The fourth 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); + INIT_CUMULATIVE_ARGS (args_so_far, funtype, NULL_RTX, fndecl, n_named_args); - /* Compute number of named args. + /* Now possibly adjust the 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. + We do include the last named arg if + targetm.calls.strict_argument_naming() returns nonzero. (If no anonymous args follow, the result of list_length is actually one too large. This is harmless.) - If PRETEND_OUTGOING_VARARGS_NAMED is set and STRICT_ARGUMENT_NAMING is - zero, this machine will be able to place unnamed args that were - passed in registers into the stack. So treat all args as named. - This allows the insns emitting for a specific argument list to be + If targetm.calls.pretend_outgoing_varargs_named() returns + nonzero, and targetm.calls.strict_argument_naming() returns zero, + this machine will be able to place unnamed args that were passed + in registers into the stack. So treat all args as named. This + allows the insns emitting for a specific argument list to be independent of the function declaration. - If PRETEND_OUTGOING_VARARGS_NAMED is not set, we do not have any - reliable way to pass unnamed args in registers, so we must force - them into memory. */ + If targetm.calls.pretend_outgoing_varargs_named() returns zero, + we do not have any reliable way to pass unnamed args in + registers, so we must force them into memory. */ - if ((targetm.calls.strict_argument_naming (&args_so_far) - || ! targetm.calls.pretend_outgoing_varargs_named (&args_so_far)) - && type_arg_types != 0) - n_named_args - = (list_length (type_arg_types) - /* Don't include the last named arg. */ - - (targetm.calls.strict_argument_naming (&args_so_far) ? 0 : 1) - /* Count the struct value address, if it is passed as a parm. */ - + structure_value_addr_parm); + if (type_arg_types != 0 + && targetm.calls.strict_argument_naming (&args_so_far)) + ; + else if (type_arg_types != 0 + && ! targetm.calls.pretend_outgoing_varargs_named (&args_so_far)) + /* Don't include the last named arg. */ + --n_named_args; else - /* If we know nothing, treat all args as named. */ + /* Treat all args as named. */ n_named_args = num_actuals; /* Make a vector to hold all the information about each arg. */ @@ -2404,7 +2115,8 @@ expand_call (tree exp, rtx target, int ignore) n_named_args, actparms, fndecl, &args_so_far, reg_parm_stack_space, &old_stack_level, &old_pending_adj, - &must_preallocate, &flags); + &must_preallocate, &flags, + &try_tail_call, CALL_FROM_THUNK_P (exp)); if (args_size.var) { @@ -2434,29 +2146,16 @@ expand_call (tree exp, rtx target, int ignore) || (!ACCUMULATE_OUTGOING_ARGS && args_size.constant))) structure_value_addr = copy_to_reg (structure_value_addr); - /* Tail calls can make things harder to debug, and we're traditionally + /* Tail calls can make things harder to debug, and we've 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. */ + there's cleanups, as we know there's code to follow the call. */ if (currently_expanding_call++ != 0 || !flag_optimize_sibling_calls - || !rtx_equal_function_value_matters - || any_pending_cleanups () - || 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 (addr) != ADDR_EXPR - || TREE_OPERAND (addr, 0) != current_function_decl) - try_tail_recursion = 0; + || args_size.var + || lookup_stmt_eh_region (exp) >= 0) + try_tail_call = 0; /* Rest of purposes for tail call optimizations to fail. */ if ( @@ -2473,10 +2172,10 @@ expand_call (tree exp, rtx target, int ignore) || structure_value_addr != NULL_RTX /* Check whether the target is able to optimize the call into a sibcall. */ - || !(*targetm.function_ok_for_sibcall) (fndecl, exp) + || !targetm.function_ok_for_sibcall (fndecl, exp) /* Functions that do not return exactly once may not be sibcall optimized. */ - || (flags & (ECF_RETURNS_TWICE | ECF_LONGJMP | ECF_NORETURN)) + || (flags & (ECF_RETURNS_TWICE | ECF_NORETURN)) || TYPE_VOLATILE (TREE_TYPE (TREE_TYPE (addr))) /* If the called function is nested in the current one, it might access some of the caller's arguments, but could clobber them beforehand if @@ -2491,113 +2190,9 @@ expand_call (tree exp, rtx target, int ignore) != RETURN_POPS_ARGS (current_function_decl, TREE_TYPE (current_function_decl), current_function_args_size)) - || !(*lang_hooks.decls.ok_for_sibcall) (fndecl)) + || !lang_hooks.decls.ok_for_sibcall (fndecl)) try_tail_call = 0; - if (try_tail_call || try_tail_recursion) - { - int end, inc; - actparms = NULL_TREE; - /* Ok, we're going to give the tail call the old college try. - This means we're going to evaluate the function arguments - up to three times. There are two degrees of badness we can - 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 order. */ - - if (PUSH_ARGS_REVERSED) - { - inc = 1; - i = 0; - end = num_actuals; - } - else - { - inc = -1; - i = num_actuals - 1; - end = -1; - } - - for (; i != end; i += inc) - { - args[i].tree_value = fix_unsafe_tree (args[i].tree_value); - /* 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); - } - /* Do the same for the function address if it is an expression. */ - if (!fndecl) - addr = fix_unsafe_tree (addr); - /* Expanding one of those dangerous arguments could have added - cleanups, but otherwise give it a whirl. */ - if (any_pending_cleanups ()) - try_tail_call = try_tail_recursion = 0; - } - - /* Generate a tail recursion sequence when calling ourselves. */ - - if (try_tail_recursion) - { - /* We want to emit any pending stack adjustments before the tail - recursion "call". That way we know any adjustment after the tail - 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 (); - - /* 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 ())) - { - if (any_pending_cleanups ()) - try_tail_call = try_tail_recursion = 0; - else - tail_recursion_insns = get_insns (); - } - expand_end_target_temps (); - end_sequence (); - - /* Restore the original pending stack adjustment for the sibling and - normal call cases below. */ - pending_stack_adjust = save_pending_stack_adjust; - stack_pointer_delta = save_stack_pointer_delta; - } - - if (profile_arc_flag && (flags & ECF_FORK_OR_EXEC)) - { - /* A fork duplicates the profile information, and an exec discards - it. We can't rely on fork/exec to be paired. So write out the - profile information we have gathered so far, and clear it. */ - /* ??? When Linux's __clone is called with CLONE_VM set, profiling - is subject to race conditions, just as with multithreaded - programs. */ - - emit_library_call (gcov_flush_libfunc, LCT_ALWAYS_RETURN, VOIDmode, 0); - } - /* Ensure current function's preferred stack boundary is at least what we need. We don't have to increase alignment for recursive functions. */ @@ -2609,8 +2204,6 @@ expand_call (tree exp, rtx target, int ignore) 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. */ @@ -2619,7 +2212,7 @@ expand_call (tree exp, rtx target, int ignore) int sibcall_failure = 0; /* 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 + recursion call can be ignored if we indeed use the tail call expansion. */ int save_pending_stack_adjust = 0; int save_stack_pointer_delta = 0; @@ -2628,10 +2221,6 @@ expand_call (tree exp, rtx target, int ignore) if (pass == 0) { - /* 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; @@ -2653,15 +2242,6 @@ expand_call (tree exp, rtx target, int ignore) sibcall_failure instead of continuing the loop. */ start_sequence (); - if (pass == 0) - { - /* We know at this point that there are not currently any - pending cleanups. If, however, in the process of evaluating - the arguments we were to create some, we'll need to be - able to get rid of them. */ - expand_start_target_temps (); - } - /* Don't let pending stack adjusts add up to too much. Also, do all pending adjustments now if there is any chance this might be a call to alloca or if we are expanding a sibling @@ -2678,10 +2258,6 @@ expand_call (tree exp, rtx target, int ignore) if (pass && (flags & ECF_LIBCALL_BLOCK)) NO_DEFER_POP; -#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); @@ -2955,6 +2531,12 @@ expand_call (tree exp, rtx target, int ignore) once we have started filling any specific hard regs. */ precompute_register_parameters (num_actuals, args, ®_parm_seen); + if (TREE_OPERAND (exp, 2)) + static_chain_value = expand_expr (TREE_OPERAND (exp, 2), + NULL_RTX, VOIDmode, 0); + else + static_chain_value = 0; + #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. */ @@ -2981,6 +2563,14 @@ expand_call (tree exp, rtx target, int ignore) && check_sibcall_argument_overlap (before_arg, &args[i], 1))) sibcall_failure = 1; + + if (flags & ECF_CONST + && args[i].stack + && args[i].value == args[i].stack) + call_fusage = gen_rtx_EXPR_LIST (VOIDmode, + gen_rtx_USE (VOIDmode, + args[i].value), + call_fusage); } /* If we have a parm that is passed in registers but not in memory @@ -3026,26 +2616,23 @@ expand_call (tree exp, rtx target, int ignore) structure value. */ if (pass != 0 && structure_value_addr && ! structure_value_addr_parm) { - structure_value_addr + structure_value_addr = convert_memory_address (Pmode, structure_value_addr); emit_move_insn (struct_value, force_reg (Pmode, force_operand (structure_value_addr, NULL_RTX))); - if (GET_CODE (struct_value) == REG) + if (REG_P (struct_value)) use_reg (&call_fusage, struct_value); } - funexp = prepare_call_address (funexp, fndecl, &call_fusage, - reg_parm_seen, pass == 0); + funexp = prepare_call_address (funexp, static_chain_value, + &call_fusage, reg_parm_seen, pass == 0); load_register_parameters (args, num_actuals, &call_fusage, flags, pass == 0, &sibcall_failure); - /* Perform postincrements before actually calling the function. */ - emit_queue (); - /* 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 (); @@ -3065,15 +2652,29 @@ expand_call (tree exp, rtx target, int ignore) now! */ /* Stack must be properly aligned now. */ - if (pass && stack_pointer_delta % preferred_unit_stack_boundary) - abort (); + gcc_assert (!pass + || !(stack_pointer_delta % preferred_unit_stack_boundary)); /* Generate the actual call instruction. */ - emit_call_1 (funexp, fndecl, funtype, unadjusted_args_size, + emit_call_1 (funexp, exp, fndecl, funtype, unadjusted_args_size, adjusted_args_size.constant, struct_value_size, next_arg_reg, valreg, old_inhibit_defer_pop, call_fusage, flags, & args_so_far); + /* If a non-BLKmode value is returned at the most significant end + of a register, shift the register right by the appropriate amount + and update VALREG accordingly. BLKmode values are handled by the + group load/store machinery below. */ + if (!structure_value_addr + && !pcc_struct_value + && TYPE_MODE (TREE_TYPE (exp)) != BLKmode + && targetm.calls.return_in_msb (TREE_TYPE (exp))) + { + if (shift_return_value (TYPE_MODE (TREE_TYPE (exp)), false, valreg)) + sibcall_failure = 1; + valreg = gen_rtx_REG (TYPE_MODE (TREE_TYPE (exp)), REGNO (valreg)); + } + /* 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 @@ -3089,7 +2690,7 @@ expand_call (tree exp, rtx target, int ignore) /* Expansion of block moves possibly introduced a loop that may not appear inside libcall block. */ for (insn = insns; insn; insn = NEXT_INSN (insn)) - if (GET_CODE (insn) == JUMP_INSN) + if (JUMP_P (insn)) failed = true; if (failed) @@ -3107,22 +2708,33 @@ expand_call (tree exp, rtx target, int ignore) mark_reg_pointer (temp, TYPE_ALIGN (TREE_TYPE (TREE_TYPE (exp)))); - /* 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); - end_sequence (); - - if (flags & ECF_PURE) - note = gen_rtx_EXPR_LIST (VOIDmode, + if (flag_unsafe_math_optimizations + && fndecl + && DECL_BUILT_IN (fndecl) + && (DECL_FUNCTION_CODE (fndecl) == BUILT_IN_SQRT + || DECL_FUNCTION_CODE (fndecl) == BUILT_IN_SQRTF + || DECL_FUNCTION_CODE (fndecl) == BUILT_IN_SQRTL)) + note = gen_rtx_fmt_e (SQRT, + GET_MODE (temp), + args[0].initial_value); + else + { + /* 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); + + 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; @@ -3156,49 +2768,40 @@ expand_call (tree exp, rtx target, int ignore) if nonvolatile values are live. For functions that cannot return, inform flow that control does not fall through. */ - if ((flags & (ECF_NORETURN | ECF_LONGJMP)) || pass == 0) + if ((flags & ECF_NORETURN) || pass == 0) { /* The barrier 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) + while (!CALL_P (last)) { last = PREV_INSN (last); /* There was no CALL_INSN? */ - if (last == before_call) - abort (); + gcc_assert (last != before_call); } emit_barrier_after (last); - /* Stack adjustments after a noreturn call are dead code. */ - stack_pointer_delta = old_stack_allocated; - pending_stack_adjust = 0; + /* Stack adjustments after a noreturn call are dead code. + However when NO_DEFER_POP is in effect, we must preserve + stack_pointer_delta. */ + if (inhibit_defer_pop == 0) + { + stack_pointer_delta = old_stack_allocated; + pending_stack_adjust = 0; + } } - if (flags & ECF_LONGJMP) - current_function_calls_longjmp = 1; - /* If value type not void, return an rtx for the value. */ - /* 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 ()) - { - if (target && REG_P (target) - && REGNO (target) < FIRST_PSEUDO_REGISTER) - target = 0; - sibcall_failure = 1; - } - if (TYPE_MODE (TREE_TYPE (exp)) == VOIDmode || ignore) target = const0_rtx; else if (structure_value_addr) { - if (target == 0 || GET_CODE (target) != MEM) + if (target == 0 || !MEM_P (target)) { target = gen_rtx_MEM (TYPE_MODE (TREE_TYPE (exp)), @@ -3228,7 +2831,6 @@ expand_call (tree exp, rtx target, int ignore) | TYPE_QUAL_CONST)); target = assign_temp (nt, 0, 1, 1); - preserve_temp_slots (target); } if (! rtx_equal_p (target, valreg)) @@ -3253,7 +2855,7 @@ expand_call (tree exp, rtx target, int ignore) /* If we are setting a MEM, this code must be executed. Since it is emitted after the call insn, sibcall optimization cannot be performed in that case. */ - if (GET_CODE (target) == MEM) + if (MEM_P (target)) sibcall_failure = 1; } else if (TYPE_MODE (TREE_TYPE (exp)) == BLKmode) @@ -3268,36 +2870,36 @@ expand_call (tree exp, rtx target, int ignore) if (targetm.calls.promote_function_return(funtype)) { - /* 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))) - { - tree type = TREE_TYPE (exp); - int unsignedp = TREE_UNSIGNED (type); - int offset = 0; - - /* If we don't promote as expected, something is wrong. */ - if (GET_MODE (target) - != promote_mode (type, TYPE_MODE (type), &unsignedp, 1)) - abort (); - - 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_SET (target, unsignedp); - } + /* If we promoted this return value, make the proper SUBREG. + TARGET might be const0_rtx here, so be careful. */ + if (REG_P (target) + && TYPE_MODE (TREE_TYPE (exp)) != BLKmode + && GET_MODE (target) != TYPE_MODE (TREE_TYPE (exp))) + { + tree type = TREE_TYPE (exp); + int unsignedp = TYPE_UNSIGNED (type); + int offset = 0; + enum machine_mode pmode; + + pmode = promote_mode (type, TYPE_MODE (type), &unsignedp, 1); + /* If we don't promote as expected, something is wrong. */ + gcc_assert (GET_MODE (target) == pmode); + + 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_SET (target, unsignedp); + } } /* If size of args is variable or this was a constructor call for a stack @@ -3308,6 +2910,7 @@ expand_call (tree exp, rtx target, int ignore) emit_stack_restore (SAVE_BLOCK, old_stack_level, NULL_RTX); stack_pointer_delta = old_stack_pointer_delta; pending_stack_adjust = old_pending_adj; + old_stack_allocated = stack_pointer_delta - pending_stack_adjust; 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; @@ -3347,38 +2950,14 @@ expand_call (tree exp, rtx target, int ignore) Check for the handler slots since we might not have a save area for non-local gotos. */ - if ((flags & ECF_MAY_BE_ALLOCA) && nonlocal_goto_handler_slots != 0) - emit_stack_save (SAVE_NONLOCAL, &nonlocal_goto_stack_level, NULL_RTX); + if ((flags & ECF_MAY_BE_ALLOCA) && cfun->nonlocal_goto_save_area != 0) + update_nonlocal_goto_save_area (); /* 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) - { - /* 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 (); - } - - /* If this function is returning into a memory location marked as - readonly, it means it is initializing that location. We normally treat - functions as not clobbering such locations, so we need to specify that - this one does. We do this by adding the appropriate CLOBBER to the - CALL_INSN function usage list. This cannot be done by emitting a - standalone CLOBBER after the call because the latter would be ignored - by at least the delay slot scheduling pass. We do this now instead of - adding to call_fusage before the call to emit_call_1 because TARGET - may be modified in the meantime. */ - if (structure_value_addr != 0 && target != 0 - && GET_CODE (target) == MEM && RTX_UNCHANGING_P (target)) - add_function_usage_to - (last_call_insn (), - gen_rtx_EXPR_LIST (VOIDmode, gen_rtx_CLOBBER (VOIDmode, target), - NULL_RTX)); - insns = get_insns (); end_sequence (); @@ -3407,58 +2986,25 @@ expand_call (tree exp, rtx target, int ignore) normal_call_insns = insns; /* Verify that we've deallocated all the stack we used. */ - if (! (flags & (ECF_NORETURN | ECF_LONGJMP)) - && old_stack_allocated != stack_pointer_delta - - pending_stack_adjust) - abort (); + gcc_assert ((flags & ECF_NORETURN) + || (old_stack_allocated + == stack_pointer_delta - pending_stack_adjust)); } /* If something prevents making this a sibling call, zero out the sequence. */ if (sibcall_failure) tail_call_insns = NULL_RTX; + else + break; } - /* 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) + /* If tail call production succeeded, we need to remove REG_EQUIV notes on + arguments too, as argument area is now clobbered by the call. */ + if (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)); + emit_insn (tail_call_insns); + cfun->tail_call_emit = true; } else emit_insn (normal_call_insns); @@ -3471,21 +3017,47 @@ expand_call (tree exp, rtx target, int ignore) if (flags & ECF_SP_DEPRESSED) { clear_pending_stack_adjust (); - emit_insn (gen_rtx (CLOBBER, VOIDmode, stack_pointer_rtx)); + emit_insn (gen_rtx_CLOBBER (VOIDmode, stack_pointer_rtx)); emit_move_insn (virtual_stack_dynamic_rtx, stack_pointer_rtx); - save_stack_pointer (); } return target; } +/* A sibling call sequence invalidates any REG_EQUIV notes made for + this function's incoming arguments. + + At the start of RTL generation we know the only REG_EQUIV notes + in the rtl chain are those for incoming arguments, so we can safely + flush any REG_EQUIV note. + + This is (slight) overkill. We could keep track of the highest + argument we clobber and be more selective in removing notes, but it + does not seem to be worth the effort. */ +void +fixup_tail_calls (void) +{ + purge_reg_equiv_notes (); +} + /* Traverse an argument list in VALUES and expand all complex arguments into their components. */ -tree +static tree split_complex_values (tree values) { tree p; + /* Before allocating memory, check for the common case of no complex. */ + for (p = values; p; p = TREE_CHAIN (p)) + { + tree type = TREE_TYPE (TREE_VALUE (p)); + if (type && TREE_CODE (type) == COMPLEX_TYPE + && targetm.calls.split_complex_arg (type)) + goto found; + } + return values; + + found: values = copy_list (values); for (p = values; p; p = TREE_CHAIN (p)) @@ -3497,7 +3069,8 @@ split_complex_values (tree values) if (!complex_type) continue; - if (TREE_CODE (complex_type) == COMPLEX_TYPE) + if (TREE_CODE (complex_type) == COMPLEX_TYPE + && targetm.calls.split_complex_arg (complex_type)) { tree subtype; tree real, imag, next; @@ -3523,18 +3096,30 @@ split_complex_values (tree values) /* Traverse a list of TYPES and expand all complex types into their components. */ -tree +static tree split_complex_types (tree types) { tree p; + /* Before allocating memory, check for the common case of no complex. */ + for (p = types; p; p = TREE_CHAIN (p)) + { + tree type = TREE_VALUE (p); + if (TREE_CODE (type) == COMPLEX_TYPE + && targetm.calls.split_complex_arg (type)) + goto found; + } + return types; + + found: types = copy_list (types); for (p = types; p; p = TREE_CHAIN (p)) { tree complex_type = TREE_VALUE (p); - if (TREE_CODE (complex_type) == COMPLEX_TYPE) + if (TREE_CODE (complex_type) == COMPLEX_TYPE + && targetm.calls.split_complex_arg (complex_type)) { tree next, imag; @@ -3610,12 +3195,8 @@ emit_library_call_value_1 (int retval, rtx orgfun, rtx value, rtx struct_value = targetm.calls.struct_value_rtx (0, 0); #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 ((tree) 0); #endif -#endif /* By default, library functions can not throw. */ flags = ECF_NOTHROW; @@ -3660,7 +3241,7 @@ emit_library_call_value_1 (int retval, rtx orgfun, rtx value, decide where in memory it should come back. */ if (outmode != VOIDmode) { - tfom = (*lang_hooks.types.type_for_mode) (outmode, 0); + tfom = lang_hooks.types.type_for_mode (outmode, 0); if (aggregate_value_p (tfom, 0)) { #ifdef PCC_STATIC_STRUCT_RETURN @@ -3672,7 +3253,7 @@ emit_library_call_value_1 (int retval, rtx orgfun, rtx value, value = gen_reg_rtx (outmode); #else /* not PCC_STATIC_STRUCT_RETURN */ struct_value_size = GET_MODE_SIZE (outmode); - if (value != 0 && GET_CODE (value) == MEM) + if (value != 0 && MEM_P (value)) mem_value = value; else mem_value = assign_temp (tfom, 0, 1, 1); @@ -3699,7 +3280,7 @@ emit_library_call_value_1 (int retval, rtx orgfun, rtx value, #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); + INIT_CUMULATIVE_ARGS (args_so_far, NULL_TREE, fun, 0, nargs); #endif args_size.constant = 0; @@ -3719,10 +3300,11 @@ emit_library_call_value_1 (int retval, rtx orgfun, rtx value, if (mem_value && struct_value == 0 && ! pcc_struct_value) { rtx addr = XEXP (mem_value, 0); + nargs++; /* Make sure it is a reasonable operand for a move or push insn. */ - if (GET_CODE (addr) != REG && GET_CODE (addr) != MEM + if (!REG_P (addr) && !MEM_P (addr) && ! (CONSTANT_P (addr) && LEGITIMATE_CONSTANT_P (addr))) addr = force_operand (addr, NULL_RTX); @@ -3731,10 +3313,8 @@ emit_library_call_value_1 (int retval, rtx orgfun, rtx value, argvec[count].partial = 0; argvec[count].reg = FUNCTION_ARG (args_so_far, Pmode, NULL_TREE, 1); -#ifdef FUNCTION_ARG_PARTIAL_NREGS - if (FUNCTION_ARG_PARTIAL_NREGS (args_so_far, Pmode, NULL_TREE, 1)) - abort (); -#endif + gcc_assert (targetm.calls.arg_partial_bytes (&args_so_far, Pmode, + NULL_TREE, 1) == 0); locate_and_pad_parm (Pmode, NULL_TREE, #ifdef STACK_PARMS_IN_REG_PARM_AREA @@ -3760,28 +3340,19 @@ emit_library_call_value_1 (int retval, rtx orgfun, rtx value, /* 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 (); - - /* There's no need to call protect_from_queue, because - either emit_move_insn or emit_push_insn will do that. */ + gcc_assert (mode != BLKmode + && (GET_MODE (val) == mode || GET_MODE (val) == VOIDmode)); /* Make sure it is a reasonable operand for a move or push insn. */ - if (GET_CODE (val) != REG && GET_CODE (val) != MEM + if (!REG_P (val) && !MEM_P (val) && ! (CONSTANT_P (val) && LEGITIMATE_CONSTANT_P (val))) val = force_operand (val, NULL_RTX); -#ifdef FUNCTION_ARG_PASS_BY_REFERENCE - if (FUNCTION_ARG_PASS_BY_REFERENCE (args_so_far, mode, NULL_TREE, 1)) + if (pass_by_reference (&args_so_far, mode, NULL_TREE, 1)) { rtx slot; - int must_copy = 1 -#ifdef FUNCTION_ARG_CALLEE_COPIES - && ! FUNCTION_ARG_CALLEE_COPIES (args_so_far, mode, - NULL_TREE, 1) -#endif - ; + int must_copy + = !reference_callee_copied (&args_so_far, mode, NULL_TREE, 1); /* loop.c won't look at CALL_INSN_FUNCTION_USAGE of const/pure functions, so we have to pretend this isn't such a function. */ @@ -3801,25 +3372,14 @@ emit_library_call_value_1 (int retval, rtx orgfun, rtx value, flags |= ECF_PURE; } - if (GET_MODE (val) == MEM && ! must_copy) + if (GET_MODE (val) == MEM && !must_copy) slot = val; - else if (must_copy) + else { - slot = assign_temp ((*lang_hooks.types.type_for_mode) (mode, 0), + slot = assign_temp (lang_hooks.types.type_for_mode (mode, 0), 0, 1, 1); emit_move_insn (slot, val); } - else - { - tree type = (*lang_hooks.types.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), @@ -3833,19 +3393,14 @@ emit_library_call_value_1 (int retval, rtx orgfun, rtx value, mode = Pmode; val = force_operand (XEXP (slot, 0), NULL_RTX); } -#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; -#endif + = targetm.calls.arg_partial_bytes (&args_so_far, mode, NULL_TREE, 1); locate_and_pad_parm (mode, NULL_TREE, #ifdef STACK_PARMS_IN_REG_PARM_AREA @@ -3856,8 +3411,7 @@ emit_library_call_value_1 (int retval, rtx orgfun, rtx value, argvec[count].partial, NULL_TREE, &args_size, &argvec[count].locate); - if (argvec[count].locate.size.var) - abort (); + gcc_assert (!argvec[count].locate.size.var); if (argvec[count].reg == 0 || argvec[count].partial != 0 || reg_parm_stack_space > 0) @@ -3866,10 +3420,6 @@ emit_library_call_value_1 (int retval, rtx orgfun, rtx value, FUNCTION_ARG_ADVANCE (args_so_far, mode, (tree) 0, 1); } -#ifdef FINAL_REG_PARM_STACK_SPACE - reg_parm_stack_space = FINAL_REG_PARM_STACK_SPACE (args_size.constant, - args_size.var); -#endif /* If this machine requires an external definition for library functions, write one out. */ assemble_external_libcall (fun); @@ -4074,7 +3624,7 @@ emit_library_call_value_1 (int retval, rtx orgfun, rtx value, else argnum = 0; - fun = prepare_call_address (fun, NULL_TREE, &call_fusage, 0, 0); + fun = prepare_call_address (fun, NULL, &call_fusage, 0, 0); /* Now load any reg parms into their regs. */ @@ -4082,6 +3632,7 @@ emit_library_call_value_1 (int retval, rtx orgfun, rtx value, are to be pushed. */ for (count = 0; count < nargs; count++, argnum += inc) { + enum machine_mode mode = argvec[argnum].mode; rtx val = argvec[argnum].value; rtx reg = argvec[argnum].reg; int partial = argvec[argnum].partial; @@ -4089,7 +3640,7 @@ emit_library_call_value_1 (int retval, rtx orgfun, rtx value, /* 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, NULL_TREE, GET_MODE_SIZE (GET_MODE (val))); + emit_group_load (reg, val, NULL_TREE, GET_MODE_SIZE (mode)); else if (reg != 0 && partial == 0) emit_move_insn (reg, val); @@ -4113,7 +3664,7 @@ emit_library_call_value_1 (int retval, rtx orgfun, rtx value, force_reg (Pmode, force_operand (XEXP (mem_value, 0), NULL_RTX))); - if (GET_CODE (struct_value) == REG) + if (REG_P (struct_value)) use_reg (&call_fusage, struct_value); } @@ -4124,8 +3675,8 @@ emit_library_call_value_1 (int retval, rtx orgfun, rtx value, ? hard_libcall_value (outmode) : NULL_RTX); /* Stack must be properly aligned now. */ - if (stack_pointer_delta & (PREFERRED_STACK_BOUNDARY / BITS_PER_UNIT - 1)) - abort (); + gcc_assert (!(stack_pointer_delta + & (PREFERRED_STACK_BOUNDARY / BITS_PER_UNIT - 1))); before_call = get_last_insn (); @@ -4136,7 +3687,7 @@ emit_library_call_value_1 (int retval, rtx orgfun, rtx value, always signed. We also assume that the list of arguments passed has no impact, so we pretend it is unknown. */ - emit_call_1 (fun, + emit_call_1 (fun, NULL, get_identifier (XSTR (orgfun, 0)), build_function_type (tfom, NULL_TREE), original_args_size.constant, args_size.constant, @@ -4149,19 +3700,18 @@ emit_library_call_value_1 (int retval, rtx orgfun, rtx value, if nonvolatile values are live. For functions that cannot return, inform flow that control does not fall through. */ - if (flags & (ECF_NORETURN | ECF_LONGJMP)) + if (flags & ECF_NORETURN) { /* The barrier 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) + while (!CALL_P (last)) { last = PREV_INSN (last); /* There was no CALL_INSN? */ - if (last == before_call) - abort (); + gcc_assert (last != before_call); } emit_barrier_after (last); @@ -4193,7 +3743,7 @@ emit_library_call_value_1 (int retval, rtx orgfun, rtx value, if (GET_CODE (valreg) == PARALLEL) { temp = gen_reg_rtx (outmode); - emit_group_store (temp, valreg, NULL_TREE, + emit_group_store (temp, valreg, NULL_TREE, GET_MODE_SIZE (outmode)); valreg = temp; } @@ -4285,7 +3835,6 @@ emit_library_call_value_1 (int retval, rtx orgfun, rtx value, 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 should be LCT_NORMAL for `normal' calls, LCT_CONST for `const' calls, LCT_PURE for `pure' calls, LCT_CONST_MAKE_BLOCK for `const' calls @@ -4438,11 +3987,10 @@ store_one_arg (struct arg_data *arg, rtx argblock, int flags, partial = arg->partial; } - if (reg != 0 && partial == 0) - /* Being passed entirely in a register. We shouldn't be called in - this case. */ - abort (); - + /* Being passed entirely in a register. We shouldn't be called in + this case. */ + gcc_assert (reg == 0 || partial != 0); + /* If this arg needs special alignment, don't load the registers here. */ if (arg->n_aligned_regs != 0) @@ -4562,11 +4110,11 @@ store_one_arg (struct arg_data *arg, rtx argblock, int flags, } else { - /* PUSH_ROUNDING has no effect on us, because - emit_push_insn for BLKmode is careful to avoid it. */ + /* PUSH_ROUNDING has no effect on us, because emit_push_insn + for BLKmode is careful to avoid it. */ excess = (arg->locate.size.constant - int_size_in_bytes (TREE_TYPE (pval)) - + partial * UNITS_PER_WORD); + + partial); size_rtx = expand_expr (size_in_bytes (TREE_TYPE (pval)), NULL_RTX, TYPE_MODE (sizetype), 0); } @@ -4588,7 +4136,7 @@ store_one_arg (struct arg_data *arg, rtx argblock, int flags, } } - if ((flags & ECF_SIBCALL) && GET_CODE (arg->value) == MEM) + if ((flags & ECF_SIBCALL) && MEM_P (arg->value)) { /* emit_push_insn might not work properly if arg->value and argblock + arg->locate.offset areas overlap. */ @@ -4604,9 +4152,9 @@ store_one_arg (struct arg_data *arg, rtx argblock, int flags, if (XEXP (x, 0) != current_function_internal_arg_pointer) i = INTVAL (XEXP (XEXP (x, 0), 1)); - /* expand_call should ensure this */ - if (arg->locate.offset.var || GET_CODE (size_rtx) != CONST_INT) - abort (); + /* expand_call should ensure this. */ + gcc_assert (!arg->locate.offset.var + && GET_CODE (size_rtx) == CONST_INT); if (arg->locate.offset.constant > i) { @@ -4638,6 +4186,14 @@ store_one_arg (struct arg_data *arg, rtx argblock, int flags, arg->value = arg->stack_slot; } + if (arg->reg && GET_CODE (arg->reg) == PARALLEL) + { + tree type = TREE_TYPE (arg->tree_value); + arg->parallel_value + = emit_group_load_into_temps (arg->reg, arg->value, type, + int_size_in_bytes (type)); + } + /* Mark all slots this store used. */ if (ACCUMULATE_OUTGOING_ARGS && !(flags & ECF_SIBCALL) && argblock && ! variable_size && arg->stack) @@ -4648,10 +4204,6 @@ store_one_arg (struct arg_data *arg, rtx argblock, int flags, be deferred during the rest of the arguments. */ NO_DEFER_POP; - /* ANSI doesn't require a sequence point here, - but PCC has one, so this will avoid some problems. */ - emit_queue (); - /* Free any temporary slots made in processing this argument. Show that we might have taken the address of something and pushed that as an operand. */ @@ -4662,24 +4214,33 @@ store_one_arg (struct arg_data *arg, rtx argblock, int flags, return sibcall_failure; } -/* Nonzero if we do not know how to pass TYPE solely in registers. - We cannot do so in the following cases: +/* Nonzero if we do not know how to pass TYPE solely in registers. */ + +bool +must_pass_in_stack_var_size (enum machine_mode mode ATTRIBUTE_UNUSED, + tree type) +{ + if (!type) + return false; + + /* If the type has variable size... */ + if (TREE_CODE (TYPE_SIZE (type)) != INTEGER_CST) + return true; - - if the type has variable size - - if the type is marked as addressable (it is required to be constructed - into the stack) - - if the padding and mode of the type is such that a copy into a register - would put it into the wrong part of the register. + /* If the type is marked as addressable (it is required + to be constructed into the stack)... */ + if (TREE_ADDRESSABLE (type)) + return true; - Which padding can't be supported depends on the byte endianness. + return false; +} - A value in a register is implicitly padded at the most significant end. - On a big-endian machine, that is the lower end in memory. - So a value padded in memory at the upper end can't go in a register. - For a little-endian machine, the reverse is true. */ +/* Another version of the TARGET_MUST_PASS_IN_STACK hook. This one + takes trailing padding of a structure into account. */ +/* ??? Should be able to merge these two by examining BLOCK_REG_PADDING. */ bool -default_must_pass_in_stack (enum machine_mode mode, tree type) +must_pass_in_stack_var_size_or_pad (enum machine_mode mode, tree type) { if (!type) return false;