| /* Expands front end tree to back end RTL for GCC. |
| Copyright (C) 1987-2021 Free Software Foundation, Inc. |
| |
| This file is part of GCC. |
| |
| GCC is free software; you can redistribute it and/or modify it under |
| the terms of the GNU General Public License as published by the Free |
| Software Foundation; either version 3, or (at your option) any later |
| version. |
| |
| GCC is distributed in the hope that it will be useful, but WITHOUT ANY |
| WARRANTY; without even the implied warranty of MERCHANTABILITY or |
| FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
| for more details. |
| |
| You should have received a copy of the GNU General Public License |
| along with GCC; see the file COPYING3. If not see |
| <http://www.gnu.org/licenses/>. */ |
| |
| /* This file handles the generation of rtl code from tree structure |
| at the level of the function as a whole. |
| It creates the rtl expressions for parameters and auto variables |
| and has full responsibility for allocating stack slots. |
| |
| `expand_function_start' is called at the beginning of a function, |
| before the function body is parsed, and `expand_function_end' is |
| called after parsing the body. |
| |
| Call `assign_stack_local' to allocate a stack slot for a local variable. |
| This is usually done during the RTL generation for the function body, |
| but it can also be done in the reload pass when a pseudo-register does |
| not get a hard register. */ |
| |
| #include "config.h" |
| #include "system.h" |
| #include "coretypes.h" |
| #include "backend.h" |
| #include "target.h" |
| #include "rtl.h" |
| #include "tree.h" |
| #include "gimple-expr.h" |
| #include "cfghooks.h" |
| #include "df.h" |
| #include "memmodel.h" |
| #include "tm_p.h" |
| #include "stringpool.h" |
| #include "expmed.h" |
| #include "optabs.h" |
| #include "opts.h" |
| #include "regs.h" |
| #include "emit-rtl.h" |
| #include "recog.h" |
| #include "rtl-error.h" |
| #include "hard-reg-set.h" |
| #include "alias.h" |
| #include "fold-const.h" |
| #include "stor-layout.h" |
| #include "varasm.h" |
| #include "except.h" |
| #include "dojump.h" |
| #include "explow.h" |
| #include "calls.h" |
| #include "expr.h" |
| #include "optabs-tree.h" |
| #include "output.h" |
| #include "langhooks.h" |
| #include "common/common-target.h" |
| #include "gimplify.h" |
| #include "tree-pass.h" |
| #include "cfgrtl.h" |
| #include "cfganal.h" |
| #include "cfgbuild.h" |
| #include "cfgcleanup.h" |
| #include "cfgexpand.h" |
| #include "shrink-wrap.h" |
| #include "toplev.h" |
| #include "rtl-iter.h" |
| #include "tree-dfa.h" |
| #include "tree-ssa.h" |
| #include "stringpool.h" |
| #include "attribs.h" |
| #include "gimple.h" |
| #include "options.h" |
| #include "function-abi.h" |
| |
| /* So we can assign to cfun in this file. */ |
| #undef cfun |
| |
| #ifndef STACK_ALIGNMENT_NEEDED |
| #define STACK_ALIGNMENT_NEEDED 1 |
| #endif |
| |
| #define STACK_BYTES (STACK_BOUNDARY / BITS_PER_UNIT) |
| |
| /* Round a value to the lowest integer less than it that is a multiple of |
| the required alignment. Avoid using division in case the value is |
| negative. Assume the alignment is a power of two. */ |
| #define FLOOR_ROUND(VALUE,ALIGN) ((VALUE) & ~((ALIGN) - 1)) |
| |
| /* Similar, but round to the next highest integer that meets the |
| alignment. */ |
| #define CEIL_ROUND(VALUE,ALIGN) (((VALUE) + (ALIGN) - 1) & ~((ALIGN)- 1)) |
| |
| /* Nonzero once virtual register instantiation has been done. |
| assign_stack_local uses frame_pointer_rtx when this is nonzero. |
| calls.c:emit_library_call_value_1 uses it to set up |
| post-instantiation libcalls. */ |
| int virtuals_instantiated; |
| |
| /* Assign unique numbers to labels generated for profiling, debugging, etc. */ |
| static GTY(()) int funcdef_no; |
| |
| /* These variables hold pointers to functions to create and destroy |
| target specific, per-function data structures. */ |
| struct machine_function * (*init_machine_status) (void); |
| |
| /* The currently compiled function. */ |
| struct function *cfun = 0; |
| |
| /* These hashes record the prologue and epilogue insns. */ |
| |
| struct insn_cache_hasher : ggc_cache_ptr_hash<rtx_def> |
| { |
| static hashval_t hash (rtx x) { return htab_hash_pointer (x); } |
| static bool equal (rtx a, rtx b) { return a == b; } |
| }; |
| |
| static GTY((cache)) |
| hash_table<insn_cache_hasher> *prologue_insn_hash; |
| static GTY((cache)) |
| hash_table<insn_cache_hasher> *epilogue_insn_hash; |
| |
| |
| hash_table<used_type_hasher> *types_used_by_vars_hash = NULL; |
| vec<tree, va_gc> *types_used_by_cur_var_decl; |
| |
| /* Forward declarations. */ |
| |
| static class temp_slot *find_temp_slot_from_address (rtx); |
| static void pad_to_arg_alignment (struct args_size *, int, struct args_size *); |
| static void pad_below (struct args_size *, machine_mode, tree); |
| static void reorder_blocks_1 (rtx_insn *, tree, vec<tree> *); |
| static int all_blocks (tree, tree *); |
| static tree *get_block_vector (tree, int *); |
| extern tree debug_find_var_in_block_tree (tree, tree); |
| /* We always define `record_insns' even if it's not used so that we |
| can always export `prologue_epilogue_contains'. */ |
| static void record_insns (rtx_insn *, rtx, hash_table<insn_cache_hasher> **) |
| ATTRIBUTE_UNUSED; |
| static bool contains (const rtx_insn *, hash_table<insn_cache_hasher> *); |
| static void prepare_function_start (void); |
| static void do_clobber_return_reg (rtx, void *); |
| static void do_use_return_reg (rtx, void *); |
| |
| |
| /* Stack of nested functions. */ |
| /* Keep track of the cfun stack. */ |
| |
| static vec<function *> function_context_stack; |
| |
| /* Save the current context for compilation of a nested function. |
| This is called from language-specific code. */ |
| |
| void |
| push_function_context (void) |
| { |
| if (cfun == 0) |
| allocate_struct_function (NULL, false); |
| |
| function_context_stack.safe_push (cfun); |
| set_cfun (NULL); |
| } |
| |
| /* Restore the last saved context, at the end of a nested function. |
| This function is called from language-specific code. */ |
| |
| void |
| pop_function_context (void) |
| { |
| struct function *p = function_context_stack.pop (); |
| set_cfun (p); |
| current_function_decl = p->decl; |
| |
| /* Reset variables that have known state during rtx generation. */ |
| virtuals_instantiated = 0; |
| generating_concat_p = 1; |
| } |
| |
| /* Clear out all parts of the state in F that can safely be discarded |
| after the function has been parsed, but not compiled, to let |
| garbage collection reclaim the memory. */ |
| |
| void |
| free_after_parsing (struct function *f) |
| { |
| f->language = 0; |
| } |
| |
| /* Clear out all parts of the state in F that can safely be discarded |
| after the function has been compiled, to let garbage collection |
| reclaim the memory. */ |
| |
| void |
| free_after_compilation (struct function *f) |
| { |
| prologue_insn_hash = NULL; |
| epilogue_insn_hash = NULL; |
| |
| free (crtl->emit.regno_pointer_align); |
| |
| memset (crtl, 0, sizeof (struct rtl_data)); |
| f->eh = NULL; |
| f->machine = NULL; |
| f->cfg = NULL; |
| f->curr_properties &= ~PROP_cfg; |
| |
| regno_reg_rtx = NULL; |
| } |
| |
| /* Return size needed for stack frame based on slots so far allocated. |
| This size counts from zero. It is not rounded to PREFERRED_STACK_BOUNDARY; |
| the caller may have to do that. */ |
| |
| poly_int64 |
| get_frame_size (void) |
| { |
| if (FRAME_GROWS_DOWNWARD) |
| return -frame_offset; |
| else |
| return frame_offset; |
| } |
| |
| /* Issue an error message and return TRUE if frame OFFSET overflows in |
| the signed target pointer arithmetics for function FUNC. Otherwise |
| return FALSE. */ |
| |
| bool |
| frame_offset_overflow (poly_int64 offset, tree func) |
| { |
| poly_uint64 size = FRAME_GROWS_DOWNWARD ? -offset : offset; |
| unsigned HOST_WIDE_INT limit |
| = ((HOST_WIDE_INT_1U << (GET_MODE_BITSIZE (Pmode) - 1)) |
| /* Leave room for the fixed part of the frame. */ |
| - 64 * UNITS_PER_WORD); |
| |
| if (!coeffs_in_range_p (size, 0U, limit)) |
| { |
| unsigned HOST_WIDE_INT hwisize; |
| if (size.is_constant (&hwisize)) |
| error_at (DECL_SOURCE_LOCATION (func), |
| "total size of local objects %wu exceeds maximum %wu", |
| hwisize, limit); |
| else |
| error_at (DECL_SOURCE_LOCATION (func), |
| "total size of local objects exceeds maximum %wu", |
| limit); |
| return true; |
| } |
| |
| return false; |
| } |
| |
| /* Return the minimum spill slot alignment for a register of mode MODE. */ |
| |
| unsigned int |
| spill_slot_alignment (machine_mode mode ATTRIBUTE_UNUSED) |
| { |
| return STACK_SLOT_ALIGNMENT (NULL_TREE, mode, GET_MODE_ALIGNMENT (mode)); |
| } |
| |
| /* Return stack slot alignment in bits for TYPE and MODE. */ |
| |
| static unsigned int |
| get_stack_local_alignment (tree type, machine_mode mode) |
| { |
| unsigned int alignment; |
| |
| if (mode == BLKmode) |
| alignment = BIGGEST_ALIGNMENT; |
| else |
| alignment = GET_MODE_ALIGNMENT (mode); |
| |
| /* Allow the frond-end to (possibly) increase the alignment of this |
| stack slot. */ |
| if (! type) |
| type = lang_hooks.types.type_for_mode (mode, 0); |
| |
| return STACK_SLOT_ALIGNMENT (type, mode, alignment); |
| } |
| |
| /* Determine whether it is possible to fit a stack slot of size SIZE and |
| alignment ALIGNMENT into an area in the stack frame that starts at |
| frame offset START and has a length of LENGTH. If so, store the frame |
| offset to be used for the stack slot in *POFFSET and return true; |
| return false otherwise. This function will extend the frame size when |
| given a start/length pair that lies at the end of the frame. */ |
| |
| static bool |
| try_fit_stack_local (poly_int64 start, poly_int64 length, |
| poly_int64 size, unsigned int alignment, |
| poly_int64_pod *poffset) |
| { |
| poly_int64 this_frame_offset; |
| int frame_off, frame_alignment, frame_phase; |
| |
| /* Calculate how many bytes the start of local variables is off from |
| stack alignment. */ |
| frame_alignment = PREFERRED_STACK_BOUNDARY / BITS_PER_UNIT; |
| frame_off = targetm.starting_frame_offset () % frame_alignment; |
| frame_phase = frame_off ? frame_alignment - frame_off : 0; |
| |
| /* Round the frame offset to the specified alignment. */ |
| |
| if (FRAME_GROWS_DOWNWARD) |
| this_frame_offset |
| = (aligned_lower_bound (start + length - size - frame_phase, alignment) |
| + frame_phase); |
| else |
| this_frame_offset |
| = aligned_upper_bound (start - frame_phase, alignment) + frame_phase; |
| |
| /* See if it fits. If this space is at the edge of the frame, |
| consider extending the frame to make it fit. Our caller relies on |
| this when allocating a new slot. */ |
| if (maybe_lt (this_frame_offset, start)) |
| { |
| if (known_eq (frame_offset, start)) |
| frame_offset = this_frame_offset; |
| else |
| return false; |
| } |
| else if (maybe_gt (this_frame_offset + size, start + length)) |
| { |
| if (known_eq (frame_offset, start + length)) |
| frame_offset = this_frame_offset + size; |
| else |
| return false; |
| } |
| |
| *poffset = this_frame_offset; |
| return true; |
| } |
| |
| /* Create a new frame_space structure describing free space in the stack |
| frame beginning at START and ending at END, and chain it into the |
| function's frame_space_list. */ |
| |
| static void |
| add_frame_space (poly_int64 start, poly_int64 end) |
| { |
| class frame_space *space = ggc_alloc<frame_space> (); |
| space->next = crtl->frame_space_list; |
| crtl->frame_space_list = space; |
| space->start = start; |
| space->length = end - start; |
| } |
| |
| /* Allocate a stack slot of SIZE bytes and return a MEM rtx for it |
| with machine mode MODE. |
| |
| ALIGN controls the amount of alignment for the address of the slot: |
| 0 means according to MODE, |
| -1 means use BIGGEST_ALIGNMENT and round size to multiple of that, |
| -2 means use BITS_PER_UNIT, |
| positive specifies alignment boundary in bits. |
| |
| KIND has ASLK_REDUCE_ALIGN bit set if it is OK to reduce |
| alignment and ASLK_RECORD_PAD bit set if we should remember |
| extra space we allocated for alignment purposes. When we are |
| called from assign_stack_temp_for_type, it is not set so we don't |
| track the same stack slot in two independent lists. |
| |
| We do not round to stack_boundary here. */ |
| |
| rtx |
| assign_stack_local_1 (machine_mode mode, poly_int64 size, |
| int align, int kind) |
| { |
| rtx x, addr; |
| poly_int64 bigend_correction = 0; |
| poly_int64 slot_offset = 0, old_frame_offset; |
| unsigned int alignment, alignment_in_bits; |
| |
| if (align == 0) |
| { |
| alignment = get_stack_local_alignment (NULL, mode); |
| alignment /= BITS_PER_UNIT; |
| } |
| else if (align == -1) |
| { |
| alignment = BIGGEST_ALIGNMENT / BITS_PER_UNIT; |
| size = aligned_upper_bound (size, alignment); |
| } |
| else if (align == -2) |
| alignment = 1; /* BITS_PER_UNIT / BITS_PER_UNIT */ |
| else |
| alignment = align / BITS_PER_UNIT; |
| |
| alignment_in_bits = alignment * BITS_PER_UNIT; |
| |
| /* Ignore alignment if it exceeds MAX_SUPPORTED_STACK_ALIGNMENT. */ |
| if (alignment_in_bits > MAX_SUPPORTED_STACK_ALIGNMENT) |
| { |
| alignment_in_bits = MAX_SUPPORTED_STACK_ALIGNMENT; |
| alignment = MAX_SUPPORTED_STACK_ALIGNMENT / BITS_PER_UNIT; |
| } |
| |
| if (SUPPORTS_STACK_ALIGNMENT) |
| { |
| if (crtl->stack_alignment_estimated < alignment_in_bits) |
| { |
| if (!crtl->stack_realign_processed) |
| crtl->stack_alignment_estimated = alignment_in_bits; |
| else |
| { |
| /* If stack is realigned and stack alignment value |
| hasn't been finalized, it is OK not to increase |
| stack_alignment_estimated. The bigger alignment |
| requirement is recorded in stack_alignment_needed |
| below. */ |
| gcc_assert (!crtl->stack_realign_finalized); |
| if (!crtl->stack_realign_needed) |
| { |
| /* It is OK to reduce the alignment as long as the |
| requested size is 0 or the estimated stack |
| alignment >= mode alignment. */ |
| gcc_assert ((kind & ASLK_REDUCE_ALIGN) |
| || known_eq (size, 0) |
| || (crtl->stack_alignment_estimated |
| >= GET_MODE_ALIGNMENT (mode))); |
| alignment_in_bits = crtl->stack_alignment_estimated; |
| alignment = alignment_in_bits / BITS_PER_UNIT; |
| } |
| } |
| } |
| } |
| |
| if (crtl->stack_alignment_needed < alignment_in_bits) |
| crtl->stack_alignment_needed = alignment_in_bits; |
| if (crtl->max_used_stack_slot_alignment < alignment_in_bits) |
| crtl->max_used_stack_slot_alignment = alignment_in_bits; |
| |
| if (mode != BLKmode || maybe_ne (size, 0)) |
| { |
| if (kind & ASLK_RECORD_PAD) |
| { |
| class frame_space **psp; |
| |
| for (psp = &crtl->frame_space_list; *psp; psp = &(*psp)->next) |
| { |
| class frame_space *space = *psp; |
| if (!try_fit_stack_local (space->start, space->length, size, |
| alignment, &slot_offset)) |
| continue; |
| *psp = space->next; |
| if (known_gt (slot_offset, space->start)) |
| add_frame_space (space->start, slot_offset); |
| if (known_lt (slot_offset + size, space->start + space->length)) |
| add_frame_space (slot_offset + size, |
| space->start + space->length); |
| goto found_space; |
| } |
| } |
| } |
| else if (!STACK_ALIGNMENT_NEEDED) |
| { |
| slot_offset = frame_offset; |
| goto found_space; |
| } |
| |
| old_frame_offset = frame_offset; |
| |
| if (FRAME_GROWS_DOWNWARD) |
| { |
| frame_offset -= size; |
| try_fit_stack_local (frame_offset, size, size, alignment, &slot_offset); |
| |
| if (kind & ASLK_RECORD_PAD) |
| { |
| if (known_gt (slot_offset, frame_offset)) |
| add_frame_space (frame_offset, slot_offset); |
| if (known_lt (slot_offset + size, old_frame_offset)) |
| add_frame_space (slot_offset + size, old_frame_offset); |
| } |
| } |
| else |
| { |
| frame_offset += size; |
| try_fit_stack_local (old_frame_offset, size, size, alignment, &slot_offset); |
| |
| if (kind & ASLK_RECORD_PAD) |
| { |
| if (known_gt (slot_offset, old_frame_offset)) |
| add_frame_space (old_frame_offset, slot_offset); |
| if (known_lt (slot_offset + size, frame_offset)) |
| add_frame_space (slot_offset + size, frame_offset); |
| } |
| } |
| |
| found_space: |
| /* On a big-endian machine, if we are allocating more space than we will use, |
| use the least significant bytes of those that are allocated. */ |
| if (mode != BLKmode) |
| { |
| /* The slot size can sometimes be smaller than the mode size; |
| e.g. the rs6000 port allocates slots with a vector mode |
| that have the size of only one element. However, the slot |
| size must always be ordered wrt to the mode size, in the |
| same way as for a subreg. */ |
| gcc_checking_assert (ordered_p (GET_MODE_SIZE (mode), size)); |
| if (BYTES_BIG_ENDIAN && maybe_lt (GET_MODE_SIZE (mode), size)) |
| bigend_correction = size - GET_MODE_SIZE (mode); |
| } |
| |
| /* If we have already instantiated virtual registers, return the actual |
| address relative to the frame pointer. */ |
| if (virtuals_instantiated) |
| addr = plus_constant (Pmode, frame_pointer_rtx, |
| trunc_int_for_mode |
| (slot_offset + bigend_correction |
| + targetm.starting_frame_offset (), Pmode)); |
| else |
| addr = plus_constant (Pmode, virtual_stack_vars_rtx, |
| trunc_int_for_mode |
| (slot_offset + bigend_correction, |
| Pmode)); |
| |
| x = gen_rtx_MEM (mode, addr); |
| set_mem_align (x, alignment_in_bits); |
| MEM_NOTRAP_P (x) = 1; |
| |
| vec_safe_push (stack_slot_list, x); |
| |
| if (frame_offset_overflow (frame_offset, current_function_decl)) |
| frame_offset = 0; |
| |
| return x; |
| } |
| |
| /* Wrap up assign_stack_local_1 with last parameter as false. */ |
| |
| rtx |
| assign_stack_local (machine_mode mode, poly_int64 size, int align) |
| { |
| return assign_stack_local_1 (mode, size, align, ASLK_RECORD_PAD); |
| } |
| |
| /* In order to evaluate some expressions, such as function calls returning |
| structures in memory, we need to temporarily allocate stack locations. |
| We record each allocated temporary in the following structure. |
| |
| Associated with each temporary slot is a nesting level. When we pop up |
| one level, all temporaries associated with the previous level are freed. |
| Normally, all temporaries are freed after the execution of the statement |
| in which they were created. However, if we are inside a ({...}) grouping, |
| the result may be in a temporary and hence must be preserved. If the |
| result could be in a temporary, we preserve it if we can determine which |
| one it is in. If we cannot determine which temporary may contain the |
| result, all temporaries are preserved. A temporary is preserved by |
| pretending it was allocated at the previous nesting level. */ |
| |
| class GTY(()) temp_slot { |
| public: |
| /* Points to next temporary slot. */ |
| class temp_slot *next; |
| /* Points to previous temporary slot. */ |
| class temp_slot *prev; |
| /* The rtx to used to reference the slot. */ |
| rtx slot; |
| /* The size, in units, of the slot. */ |
| poly_int64 size; |
| /* The type of the object in the slot, or zero if it doesn't correspond |
| to a type. We use this to determine whether a slot can be reused. |
| It can be reused if objects of the type of the new slot will always |
| conflict with objects of the type of the old slot. */ |
| tree type; |
| /* The alignment (in bits) of the slot. */ |
| unsigned int align; |
| /* Nonzero if this temporary is currently in use. */ |
| char in_use; |
| /* Nesting level at which this slot is being used. */ |
| int level; |
| /* The offset of the slot from the frame_pointer, including extra space |
| for alignment. This info is for combine_temp_slots. */ |
| poly_int64 base_offset; |
| /* The size of the slot, including extra space for alignment. This |
| info is for combine_temp_slots. */ |
| poly_int64 full_size; |
| }; |
| |
| /* Entry for the below hash table. */ |
| struct GTY((for_user)) temp_slot_address_entry { |
| hashval_t hash; |
| rtx address; |
| class temp_slot *temp_slot; |
| }; |
| |
| struct temp_address_hasher : ggc_ptr_hash<temp_slot_address_entry> |
| { |
| static hashval_t hash (temp_slot_address_entry *); |
| static bool equal (temp_slot_address_entry *, temp_slot_address_entry *); |
| }; |
| |
| /* A table of addresses that represent a stack slot. The table is a mapping |
| from address RTXen to a temp slot. */ |
| static GTY(()) hash_table<temp_address_hasher> *temp_slot_address_table; |
| static size_t n_temp_slots_in_use; |
| |
| /* Removes temporary slot TEMP from LIST. */ |
| |
| static void |
| cut_slot_from_list (class temp_slot *temp, class temp_slot **list) |
| { |
| if (temp->next) |
| temp->next->prev = temp->prev; |
| if (temp->prev) |
| temp->prev->next = temp->next; |
| else |
| *list = temp->next; |
| |
| temp->prev = temp->next = NULL; |
| } |
| |
| /* Inserts temporary slot TEMP to LIST. */ |
| |
| static void |
| insert_slot_to_list (class temp_slot *temp, class temp_slot **list) |
| { |
| temp->next = *list; |
| if (*list) |
| (*list)->prev = temp; |
| temp->prev = NULL; |
| *list = temp; |
| } |
| |
| /* Returns the list of used temp slots at LEVEL. */ |
| |
| static class temp_slot ** |
| temp_slots_at_level (int level) |
| { |
| if (level >= (int) vec_safe_length (used_temp_slots)) |
| vec_safe_grow_cleared (used_temp_slots, level + 1, true); |
| |
| return &(*used_temp_slots)[level]; |
| } |
| |
| /* Returns the maximal temporary slot level. */ |
| |
| static int |
| max_slot_level (void) |
| { |
| if (!used_temp_slots) |
| return -1; |
| |
| return used_temp_slots->length () - 1; |
| } |
| |
| /* Moves temporary slot TEMP to LEVEL. */ |
| |
| static void |
| move_slot_to_level (class temp_slot *temp, int level) |
| { |
| cut_slot_from_list (temp, temp_slots_at_level (temp->level)); |
| insert_slot_to_list (temp, temp_slots_at_level (level)); |
| temp->level = level; |
| } |
| |
| /* Make temporary slot TEMP available. */ |
| |
| static void |
| make_slot_available (class temp_slot *temp) |
| { |
| cut_slot_from_list (temp, temp_slots_at_level (temp->level)); |
| insert_slot_to_list (temp, &avail_temp_slots); |
| temp->in_use = 0; |
| temp->level = -1; |
| n_temp_slots_in_use--; |
| } |
| |
| /* Compute the hash value for an address -> temp slot mapping. |
| The value is cached on the mapping entry. */ |
| static hashval_t |
| temp_slot_address_compute_hash (struct temp_slot_address_entry *t) |
| { |
| int do_not_record = 0; |
| return hash_rtx (t->address, GET_MODE (t->address), |
| &do_not_record, NULL, false); |
| } |
| |
| /* Return the hash value for an address -> temp slot mapping. */ |
| hashval_t |
| temp_address_hasher::hash (temp_slot_address_entry *t) |
| { |
| return t->hash; |
| } |
| |
| /* Compare two address -> temp slot mapping entries. */ |
| bool |
| temp_address_hasher::equal (temp_slot_address_entry *t1, |
| temp_slot_address_entry *t2) |
| { |
| return exp_equiv_p (t1->address, t2->address, 0, true); |
| } |
| |
| /* Add ADDRESS as an alias of TEMP_SLOT to the addess -> temp slot mapping. */ |
| static void |
| insert_temp_slot_address (rtx address, class temp_slot *temp_slot) |
| { |
| struct temp_slot_address_entry *t = ggc_alloc<temp_slot_address_entry> (); |
| t->address = copy_rtx (address); |
| t->temp_slot = temp_slot; |
| t->hash = temp_slot_address_compute_hash (t); |
| *temp_slot_address_table->find_slot_with_hash (t, t->hash, INSERT) = t; |
| } |
| |
| /* Remove an address -> temp slot mapping entry if the temp slot is |
| not in use anymore. Callback for remove_unused_temp_slot_addresses. */ |
| int |
| remove_unused_temp_slot_addresses_1 (temp_slot_address_entry **slot, void *) |
| { |
| const struct temp_slot_address_entry *t = *slot; |
| if (! t->temp_slot->in_use) |
| temp_slot_address_table->clear_slot (slot); |
| return 1; |
| } |
| |
| /* Remove all mappings of addresses to unused temp slots. */ |
| static void |
| remove_unused_temp_slot_addresses (void) |
| { |
| /* Use quicker clearing if there aren't any active temp slots. */ |
| if (n_temp_slots_in_use) |
| temp_slot_address_table->traverse |
| <void *, remove_unused_temp_slot_addresses_1> (NULL); |
| else |
| temp_slot_address_table->empty (); |
| } |
| |
| /* Find the temp slot corresponding to the object at address X. */ |
| |
| static class temp_slot * |
| find_temp_slot_from_address (rtx x) |
| { |
| class temp_slot *p; |
| struct temp_slot_address_entry tmp, *t; |
| |
| /* First try the easy way: |
| See if X exists in the address -> temp slot mapping. */ |
| tmp.address = x; |
| tmp.temp_slot = NULL; |
| tmp.hash = temp_slot_address_compute_hash (&tmp); |
| t = temp_slot_address_table->find_with_hash (&tmp, tmp.hash); |
| if (t) |
| return t->temp_slot; |
| |
| /* If we have a sum involving a register, see if it points to a temp |
| slot. */ |
| if (GET_CODE (x) == PLUS && REG_P (XEXP (x, 0)) |
| && (p = find_temp_slot_from_address (XEXP (x, 0))) != 0) |
| return p; |
| else if (GET_CODE (x) == PLUS && REG_P (XEXP (x, 1)) |
| && (p = find_temp_slot_from_address (XEXP (x, 1))) != 0) |
| return p; |
| |
| /* Last resort: Address is a virtual stack var address. */ |
| poly_int64 offset; |
| if (strip_offset (x, &offset) == virtual_stack_vars_rtx) |
| { |
| int i; |
| for (i = max_slot_level (); i >= 0; i--) |
| for (p = *temp_slots_at_level (i); p; p = p->next) |
| if (known_in_range_p (offset, p->base_offset, p->full_size)) |
| return p; |
| } |
| |
| return NULL; |
| } |
| |
| /* Allocate a temporary stack slot and record it for possible later |
| reuse. |
| |
| MODE is the machine mode to be given to the returned rtx. |
| |
| SIZE is the size in units of the space required. We do no rounding here |
| since assign_stack_local will do any required rounding. |
| |
| TYPE is the type that will be used for the stack slot. */ |
| |
| rtx |
| assign_stack_temp_for_type (machine_mode mode, poly_int64 size, tree type) |
| { |
| unsigned int align; |
| class temp_slot *p, *best_p = 0, *selected = NULL, **pp; |
| rtx slot; |
| |
| gcc_assert (known_size_p (size)); |
| |
| align = get_stack_local_alignment (type, mode); |
| |
| /* Try to find an available, already-allocated temporary of the proper |
| mode which meets the size and alignment requirements. Choose the |
| smallest one with the closest alignment. |
| |
| If assign_stack_temp is called outside of the tree->rtl expansion, |
| we cannot reuse the stack slots (that may still refer to |
| VIRTUAL_STACK_VARS_REGNUM). */ |
| if (!virtuals_instantiated) |
| { |
| for (p = avail_temp_slots; p; p = p->next) |
| { |
| if (p->align >= align |
| && known_ge (p->size, size) |
| && GET_MODE (p->slot) == mode |
| && objects_must_conflict_p (p->type, type) |
| && (best_p == 0 |
| || (known_eq (best_p->size, p->size) |
| ? best_p->align > p->align |
| : known_ge (best_p->size, p->size)))) |
| { |
| if (p->align == align && known_eq (p->size, size)) |
| { |
| selected = p; |
| cut_slot_from_list (selected, &avail_temp_slots); |
| best_p = 0; |
| break; |
| } |
| best_p = p; |
| } |
| } |
| } |
| |
| /* Make our best, if any, the one to use. */ |
| if (best_p) |
| { |
| selected = best_p; |
| cut_slot_from_list (selected, &avail_temp_slots); |
| |
| /* If there are enough aligned bytes left over, make them into a new |
| temp_slot so that the extra bytes don't get wasted. Do this only |
| for BLKmode slots, so that we can be sure of the alignment. */ |
| if (GET_MODE (best_p->slot) == BLKmode) |
| { |
| int alignment = best_p->align / BITS_PER_UNIT; |
| poly_int64 rounded_size = aligned_upper_bound (size, alignment); |
| |
| if (known_ge (best_p->size - rounded_size, alignment)) |
| { |
| p = ggc_alloc<temp_slot> (); |
| p->in_use = 0; |
| p->size = best_p->size - rounded_size; |
| p->base_offset = best_p->base_offset + rounded_size; |
| p->full_size = best_p->full_size - rounded_size; |
| p->slot = adjust_address_nv (best_p->slot, BLKmode, rounded_size); |
| p->align = best_p->align; |
| p->type = best_p->type; |
| insert_slot_to_list (p, &avail_temp_slots); |
| |
| vec_safe_push (stack_slot_list, p->slot); |
| |
| best_p->size = rounded_size; |
| best_p->full_size = rounded_size; |
| } |
| } |
| } |
| |
| /* If we still didn't find one, make a new temporary. */ |
| if (selected == 0) |
| { |
| poly_int64 frame_offset_old = frame_offset; |
| |
| p = ggc_alloc<temp_slot> (); |
| |
| /* We are passing an explicit alignment request to assign_stack_local. |
| One side effect of that is assign_stack_local will not round SIZE |
| to ensure the frame offset remains suitably aligned. |
| |
| So for requests which depended on the rounding of SIZE, we go ahead |
| and round it now. We also make sure ALIGNMENT is at least |
| BIGGEST_ALIGNMENT. */ |
| gcc_assert (mode != BLKmode || align == BIGGEST_ALIGNMENT); |
| p->slot = assign_stack_local_1 (mode, |
| (mode == BLKmode |
| ? aligned_upper_bound (size, |
| (int) align |
| / BITS_PER_UNIT) |
| : size), |
| align, 0); |
| |
| p->align = align; |
| |
| /* The following slot size computation is necessary because we don't |
| know the actual size of the temporary slot until assign_stack_local |
| has performed all the frame alignment and size rounding for the |
| requested temporary. Note that extra space added for alignment |
| can be either above or below this stack slot depending on which |
| way the frame grows. We include the extra space if and only if it |
| is above this slot. */ |
| if (FRAME_GROWS_DOWNWARD) |
| p->size = frame_offset_old - frame_offset; |
| else |
| p->size = size; |
| |
| /* Now define the fields used by combine_temp_slots. */ |
| if (FRAME_GROWS_DOWNWARD) |
| { |
| p->base_offset = frame_offset; |
| p->full_size = frame_offset_old - frame_offset; |
| } |
| else |
| { |
| p->base_offset = frame_offset_old; |
| p->full_size = frame_offset - frame_offset_old; |
| } |
| |
| selected = p; |
| } |
| |
| p = selected; |
| p->in_use = 1; |
| p->type = type; |
| p->level = temp_slot_level; |
| n_temp_slots_in_use++; |
| |
| pp = temp_slots_at_level (p->level); |
| insert_slot_to_list (p, pp); |
| insert_temp_slot_address (XEXP (p->slot, 0), p); |
| |
| /* Create a new MEM rtx to avoid clobbering MEM flags of old slots. */ |
| slot = gen_rtx_MEM (mode, XEXP (p->slot, 0)); |
| vec_safe_push (stack_slot_list, slot); |
| |
| /* If we know the alias set for the memory that will be used, use |
| it. If there's no TYPE, then we don't know anything about the |
| alias set for the memory. */ |
| set_mem_alias_set (slot, type ? get_alias_set (type) : 0); |
| set_mem_align (slot, align); |
| |
| /* If a type is specified, set the relevant flags. */ |
| if (type != 0) |
| MEM_VOLATILE_P (slot) = TYPE_VOLATILE (type); |
| MEM_NOTRAP_P (slot) = 1; |
| |
| return slot; |
| } |
| |
| /* Allocate a temporary stack slot and record it for possible later |
| reuse. First two arguments are same as in preceding function. */ |
| |
| rtx |
| assign_stack_temp (machine_mode mode, poly_int64 size) |
| { |
| return assign_stack_temp_for_type (mode, size, NULL_TREE); |
| } |
| |
| /* Assign a temporary. |
| If TYPE_OR_DECL is a decl, then we are doing it on behalf of the decl |
| and so that should be used in error messages. In either case, we |
| allocate of the given type. |
| MEMORY_REQUIRED is 1 if the result must be addressable stack memory; |
| it is 0 if a register is OK. |
| DONT_PROMOTE is 1 if we should not promote values in register |
| to wider modes. */ |
| |
| rtx |
| assign_temp (tree type_or_decl, int memory_required, |
| int dont_promote ATTRIBUTE_UNUSED) |
| { |
| tree type, decl; |
| machine_mode mode; |
| #ifdef PROMOTE_MODE |
| int unsignedp; |
| #endif |
| |
| if (DECL_P (type_or_decl)) |
| decl = type_or_decl, type = TREE_TYPE (decl); |
| else |
| decl = NULL, type = type_or_decl; |
| |
| mode = TYPE_MODE (type); |
| #ifdef PROMOTE_MODE |
| unsignedp = TYPE_UNSIGNED (type); |
| #endif |
| |
| /* Allocating temporaries of TREE_ADDRESSABLE type must be done in the front |
| end. See also create_tmp_var for the gimplification-time check. */ |
| gcc_assert (!TREE_ADDRESSABLE (type) && COMPLETE_TYPE_P (type)); |
| |
| if (mode == BLKmode || memory_required) |
| { |
| poly_int64 size; |
| rtx tmp; |
| |
| /* Unfortunately, we don't yet know how to allocate variable-sized |
| temporaries. However, sometimes we can find a fixed upper limit on |
| the size, so try that instead. */ |
| if (!poly_int_tree_p (TYPE_SIZE_UNIT (type), &size)) |
| size = max_int_size_in_bytes (type); |
| |
| /* Zero sized arrays are a GNU C extension. Set size to 1 to avoid |
| problems with allocating the stack space. */ |
| if (known_eq (size, 0)) |
| size = 1; |
| |
| /* The size of the temporary may be too large to fit into an integer. */ |
| /* ??? Not sure this should happen except for user silliness, so limit |
| this to things that aren't compiler-generated temporaries. The |
| rest of the time we'll die in assign_stack_temp_for_type. */ |
| if (decl |
| && !known_size_p (size) |
| && TREE_CODE (TYPE_SIZE_UNIT (type)) == INTEGER_CST) |
| { |
| error ("size of variable %q+D is too large", decl); |
| size = 1; |
| } |
| |
| tmp = assign_stack_temp_for_type (mode, size, type); |
| return tmp; |
| } |
| |
| #ifdef PROMOTE_MODE |
| if (! dont_promote) |
| mode = promote_mode (type, mode, &unsignedp); |
| #endif |
| |
| return gen_reg_rtx (mode); |
| } |
| |
| /* Combine temporary stack slots which are adjacent on the stack. |
| |
| This allows for better use of already allocated stack space. This is only |
| done for BLKmode slots because we can be sure that we won't have alignment |
| problems in this case. */ |
| |
| static void |
| combine_temp_slots (void) |
| { |
| class temp_slot *p, *q, *next, *next_q; |
| int num_slots; |
| |
| /* We can't combine slots, because the information about which slot |
| is in which alias set will be lost. */ |
| if (flag_strict_aliasing) |
| return; |
| |
| /* If there are a lot of temp slots, don't do anything unless |
| high levels of optimization. */ |
| if (! flag_expensive_optimizations) |
| for (p = avail_temp_slots, num_slots = 0; p; p = p->next, num_slots++) |
| if (num_slots > 100 || (num_slots > 10 && optimize == 0)) |
| return; |
| |
| for (p = avail_temp_slots; p; p = next) |
| { |
| int delete_p = 0; |
| |
| next = p->next; |
| |
| if (GET_MODE (p->slot) != BLKmode) |
| continue; |
| |
| for (q = p->next; q; q = next_q) |
| { |
| int delete_q = 0; |
| |
| next_q = q->next; |
| |
| if (GET_MODE (q->slot) != BLKmode) |
| continue; |
| |
| if (known_eq (p->base_offset + p->full_size, q->base_offset)) |
| { |
| /* Q comes after P; combine Q into P. */ |
| p->size += q->size; |
| p->full_size += q->full_size; |
| delete_q = 1; |
| } |
| else if (known_eq (q->base_offset + q->full_size, p->base_offset)) |
| { |
| /* P comes after Q; combine P into Q. */ |
| q->size += p->size; |
| q->full_size += p->full_size; |
| delete_p = 1; |
| break; |
| } |
| if (delete_q) |
| cut_slot_from_list (q, &avail_temp_slots); |
| } |
| |
| /* Either delete P or advance past it. */ |
| if (delete_p) |
| cut_slot_from_list (p, &avail_temp_slots); |
| } |
| } |
| |
| /* Indicate that NEW_RTX is an alternate way of referring to the temp |
| slot that previously was known by OLD_RTX. */ |
| |
| void |
| update_temp_slot_address (rtx old_rtx, rtx new_rtx) |
| { |
| class temp_slot *p; |
| |
| if (rtx_equal_p (old_rtx, new_rtx)) |
| return; |
| |
| p = find_temp_slot_from_address (old_rtx); |
| |
| /* If we didn't find one, see if both OLD_RTX is a PLUS. If so, and |
| NEW_RTX is a register, see if one operand of the PLUS is a |
| temporary location. If so, NEW_RTX points into it. Otherwise, |
| if both OLD_RTX and NEW_RTX are a PLUS and if there is a register |
| in common between them. If so, try a recursive call on those |
| values. */ |
| if (p == 0) |
| { |
| if (GET_CODE (old_rtx) != PLUS) |
| return; |
| |
| if (REG_P (new_rtx)) |
| { |
| update_temp_slot_address (XEXP (old_rtx, 0), new_rtx); |
| update_temp_slot_address (XEXP (old_rtx, 1), new_rtx); |
| return; |
| } |
| else if (GET_CODE (new_rtx) != PLUS) |
| return; |
| |
| if (rtx_equal_p (XEXP (old_rtx, 0), XEXP (new_rtx, 0))) |
| update_temp_slot_address (XEXP (old_rtx, 1), XEXP (new_rtx, 1)); |
| else if (rtx_equal_p (XEXP (old_rtx, 1), XEXP (new_rtx, 0))) |
| update_temp_slot_address (XEXP (old_rtx, 0), XEXP (new_rtx, 1)); |
| else if (rtx_equal_p (XEXP (old_rtx, 0), XEXP (new_rtx, 1))) |
| update_temp_slot_address (XEXP (old_rtx, 1), XEXP (new_rtx, 0)); |
| else if (rtx_equal_p (XEXP (old_rtx, 1), XEXP (new_rtx, 1))) |
| update_temp_slot_address (XEXP (old_rtx, 0), XEXP (new_rtx, 0)); |
| |
| return; |
| } |
| |
| /* Otherwise add an alias for the temp's address. */ |
| insert_temp_slot_address (new_rtx, p); |
| } |
| |
| /* If X could be a reference to a temporary slot, mark that slot as |
| belonging to the to one level higher than the current level. If X |
| matched one of our slots, just mark that one. Otherwise, we can't |
| easily predict which it is, so upgrade all of them. |
| |
| This is called when an ({...}) construct occurs and a statement |
| returns a value in memory. */ |
| |
| void |
| preserve_temp_slots (rtx x) |
| { |
| class temp_slot *p = 0, *next; |
| |
| if (x == 0) |
| return; |
| |
| /* If X is a register that is being used as a pointer, see if we have |
| a temporary slot we know it points to. */ |
| if (REG_P (x) && REG_POINTER (x)) |
| p = find_temp_slot_from_address (x); |
| |
| /* If X is not in memory or is at a constant address, it cannot be in |
| a temporary slot. */ |
| if (p == 0 && (!MEM_P (x) || CONSTANT_P (XEXP (x, 0)))) |
| return; |
| |
| /* First see if we can find a match. */ |
| if (p == 0) |
| p = find_temp_slot_from_address (XEXP (x, 0)); |
| |
| if (p != 0) |
| { |
| if (p->level == temp_slot_level) |
| move_slot_to_level (p, temp_slot_level - 1); |
| return; |
| } |
| |
| /* Otherwise, preserve all non-kept slots at this level. */ |
| for (p = *temp_slots_at_level (temp_slot_level); p; p = next) |
| { |
| next = p->next; |
| move_slot_to_level (p, temp_slot_level - 1); |
| } |
| } |
| |
| /* Free all temporaries used so far. This is normally called at the |
| end of generating code for a statement. */ |
| |
| void |
| free_temp_slots (void) |
| { |
| class temp_slot *p, *next; |
| bool some_available = false; |
| |
| for (p = *temp_slots_at_level (temp_slot_level); p; p = next) |
| { |
| next = p->next; |
| make_slot_available (p); |
| some_available = true; |
| } |
| |
| if (some_available) |
| { |
| remove_unused_temp_slot_addresses (); |
| combine_temp_slots (); |
| } |
| } |
| |
| /* Push deeper into the nesting level for stack temporaries. */ |
| |
| void |
| push_temp_slots (void) |
| { |
| temp_slot_level++; |
| } |
| |
| /* Pop a temporary nesting level. All slots in use in the current level |
| are freed. */ |
| |
| void |
| pop_temp_slots (void) |
| { |
| free_temp_slots (); |
| temp_slot_level--; |
| } |
| |
| /* Initialize temporary slots. */ |
| |
| void |
| init_temp_slots (void) |
| { |
| /* We have not allocated any temporaries yet. */ |
| avail_temp_slots = 0; |
| vec_alloc (used_temp_slots, 0); |
| temp_slot_level = 0; |
| n_temp_slots_in_use = 0; |
| |
| /* Set up the table to map addresses to temp slots. */ |
| if (! temp_slot_address_table) |
| temp_slot_address_table = hash_table<temp_address_hasher>::create_ggc (32); |
| else |
| temp_slot_address_table->empty (); |
| } |
| |
| /* Functions and data structures to keep track of the values hard regs |
| had at the start of the function. */ |
| |
| /* Private type used by get_hard_reg_initial_reg, get_hard_reg_initial_val, |
| and has_hard_reg_initial_val.. */ |
| struct GTY(()) initial_value_pair { |
| rtx hard_reg; |
| rtx pseudo; |
| }; |
| /* ??? This could be a VEC but there is currently no way to define an |
| opaque VEC type. This could be worked around by defining struct |
| initial_value_pair in function.h. */ |
| struct GTY(()) initial_value_struct { |
| int num_entries; |
| int max_entries; |
| initial_value_pair * GTY ((length ("%h.num_entries"))) entries; |
| }; |
| |
| /* If a pseudo represents an initial hard reg (or expression), return |
| it, else return NULL_RTX. */ |
| |
| rtx |
| get_hard_reg_initial_reg (rtx reg) |
| { |
| struct initial_value_struct *ivs = crtl->hard_reg_initial_vals; |
| int i; |
| |
| if (ivs == 0) |
| return NULL_RTX; |
| |
| for (i = 0; i < ivs->num_entries; i++) |
| if (rtx_equal_p (ivs->entries[i].pseudo, reg)) |
| return ivs->entries[i].hard_reg; |
| |
| return NULL_RTX; |
| } |
| |
| /* Make sure that there's a pseudo register of mode MODE that stores the |
| initial value of hard register REGNO. Return an rtx for such a pseudo. */ |
| |
| rtx |
| get_hard_reg_initial_val (machine_mode mode, unsigned int regno) |
| { |
| struct initial_value_struct *ivs; |
| rtx rv; |
| |
| rv = has_hard_reg_initial_val (mode, regno); |
| if (rv) |
| return rv; |
| |
| ivs = crtl->hard_reg_initial_vals; |
| if (ivs == 0) |
| { |
| ivs = ggc_alloc<initial_value_struct> (); |
| ivs->num_entries = 0; |
| ivs->max_entries = 5; |
| ivs->entries = ggc_vec_alloc<initial_value_pair> (5); |
| crtl->hard_reg_initial_vals = ivs; |
| } |
| |
| if (ivs->num_entries >= ivs->max_entries) |
| { |
| ivs->max_entries += 5; |
| ivs->entries = GGC_RESIZEVEC (initial_value_pair, ivs->entries, |
| ivs->max_entries); |
| } |
| |
| ivs->entries[ivs->num_entries].hard_reg = gen_rtx_REG (mode, regno); |
| ivs->entries[ivs->num_entries].pseudo = gen_reg_rtx (mode); |
| |
| return ivs->entries[ivs->num_entries++].pseudo; |
| } |
| |
| /* See if get_hard_reg_initial_val has been used to create a pseudo |
| for the initial value of hard register REGNO in mode MODE. Return |
| the associated pseudo if so, otherwise return NULL. */ |
| |
| rtx |
| has_hard_reg_initial_val (machine_mode mode, unsigned int regno) |
| { |
| struct initial_value_struct *ivs; |
| int i; |
| |
| ivs = crtl->hard_reg_initial_vals; |
| if (ivs != 0) |
| for (i = 0; i < ivs->num_entries; i++) |
| if (GET_MODE (ivs->entries[i].hard_reg) == mode |
| && REGNO (ivs->entries[i].hard_reg) == regno) |
| return ivs->entries[i].pseudo; |
| |
| return NULL_RTX; |
| } |
| |
| unsigned int |
| emit_initial_value_sets (void) |
| { |
| struct initial_value_struct *ivs = crtl->hard_reg_initial_vals; |
| int i; |
| rtx_insn *seq; |
| |
| if (ivs == 0) |
| return 0; |
| |
| start_sequence (); |
| for (i = 0; i < ivs->num_entries; i++) |
| emit_move_insn (ivs->entries[i].pseudo, ivs->entries[i].hard_reg); |
| seq = get_insns (); |
| end_sequence (); |
| |
| emit_insn_at_entry (seq); |
| return 0; |
| } |
| |
| /* Return the hardreg-pseudoreg initial values pair entry I and |
| TRUE if I is a valid entry, or FALSE if I is not a valid entry. */ |
| bool |
| initial_value_entry (int i, rtx *hreg, rtx *preg) |
| { |
| struct initial_value_struct *ivs = crtl->hard_reg_initial_vals; |
| if (!ivs || i >= ivs->num_entries) |
| return false; |
| |
| *hreg = ivs->entries[i].hard_reg; |
| *preg = ivs->entries[i].pseudo; |
| return true; |
| } |
| |
| /* These routines are responsible for converting virtual register references |
| to the actual hard register references once RTL generation is complete. |
| |
| The following four variables are used for communication between the |
| routines. They contain the offsets of the virtual registers from their |
| respective hard registers. */ |
| |
| static poly_int64 in_arg_offset; |
| static poly_int64 var_offset; |
| static poly_int64 dynamic_offset; |
| static poly_int64 out_arg_offset; |
| static poly_int64 cfa_offset; |
| |
| /* In most machines, the stack pointer register is equivalent to the bottom |
| of the stack. */ |
| |
| #ifndef STACK_POINTER_OFFSET |
| #define STACK_POINTER_OFFSET 0 |
| #endif |
| |
| #if defined (REG_PARM_STACK_SPACE) && !defined (INCOMING_REG_PARM_STACK_SPACE) |
| #define INCOMING_REG_PARM_STACK_SPACE REG_PARM_STACK_SPACE |
| #endif |
| |
| /* If not defined, pick an appropriate default for the offset of dynamically |
| allocated memory depending on the value of ACCUMULATE_OUTGOING_ARGS, |
| INCOMING_REG_PARM_STACK_SPACE, and OUTGOING_REG_PARM_STACK_SPACE. */ |
| |
| #ifndef STACK_DYNAMIC_OFFSET |
| |
| /* The bottom of the stack points to the actual arguments. If |
| REG_PARM_STACK_SPACE is defined, this includes the space for the register |
| parameters. However, if OUTGOING_REG_PARM_STACK space is not defined, |
| stack space for register parameters is not pushed by the caller, but |
| rather part of the fixed stack areas and hence not included in |
| `crtl->outgoing_args_size'. Nevertheless, we must allow |
| for it when allocating stack dynamic objects. */ |
| |
| #ifdef INCOMING_REG_PARM_STACK_SPACE |
| #define STACK_DYNAMIC_OFFSET(FNDECL) \ |
| ((ACCUMULATE_OUTGOING_ARGS \ |
| ? (crtl->outgoing_args_size \ |
| + (OUTGOING_REG_PARM_STACK_SPACE ((!(FNDECL) ? NULL_TREE : TREE_TYPE (FNDECL))) ? 0 \ |
| : INCOMING_REG_PARM_STACK_SPACE (FNDECL))) \ |
| : 0) + (STACK_POINTER_OFFSET)) |
| #else |
| #define STACK_DYNAMIC_OFFSET(FNDECL) \ |
| ((ACCUMULATE_OUTGOING_ARGS ? crtl->outgoing_args_size : poly_int64 (0)) \ |
| + (STACK_POINTER_OFFSET)) |
| #endif |
| #endif |
| |
| |
| /* Given a piece of RTX and a pointer to a HOST_WIDE_INT, if the RTX |
| is a virtual register, return the equivalent hard register and set the |
| offset indirectly through the pointer. Otherwise, return 0. */ |
| |
| static rtx |
| instantiate_new_reg (rtx x, poly_int64_pod *poffset) |
| { |
| rtx new_rtx; |
| poly_int64 offset; |
| |
| if (x == virtual_incoming_args_rtx) |
| { |
| if (stack_realign_drap) |
| { |
| /* Replace virtual_incoming_args_rtx with internal arg |
| pointer if DRAP is used to realign stack. */ |
| new_rtx = crtl->args.internal_arg_pointer; |
| offset = 0; |
| } |
| else |
| new_rtx = arg_pointer_rtx, offset = in_arg_offset; |
| } |
| else if (x == virtual_stack_vars_rtx) |
| new_rtx = frame_pointer_rtx, offset = var_offset; |
| else if (x == virtual_stack_dynamic_rtx) |
| new_rtx = stack_pointer_rtx, offset = dynamic_offset; |
| else if (x == virtual_outgoing_args_rtx) |
| new_rtx = stack_pointer_rtx, offset = out_arg_offset; |
| else if (x == virtual_cfa_rtx) |
| { |
| #ifdef FRAME_POINTER_CFA_OFFSET |
| new_rtx = frame_pointer_rtx; |
| #else |
| new_rtx = arg_pointer_rtx; |
| #endif |
| offset = cfa_offset; |
| } |
| else if (x == virtual_preferred_stack_boundary_rtx) |
| { |
| new_rtx = GEN_INT (crtl->preferred_stack_boundary / BITS_PER_UNIT); |
| offset = 0; |
| } |
| else |
| return NULL_RTX; |
| |
| *poffset = offset; |
| return new_rtx; |
| } |
| |
| /* A subroutine of instantiate_virtual_regs. Instantiate any virtual |
| registers present inside of *LOC. The expression is simplified, |
| as much as possible, but is not to be considered "valid" in any sense |
| implied by the target. Return true if any change is made. */ |
| |
| static bool |
| instantiate_virtual_regs_in_rtx (rtx *loc) |
| { |
| if (!*loc) |
| return false; |
| bool changed = false; |
| subrtx_ptr_iterator::array_type array; |
| FOR_EACH_SUBRTX_PTR (iter, array, loc, NONCONST) |
| { |
| rtx *loc = *iter; |
| if (rtx x = *loc) |
| { |
| rtx new_rtx; |
| poly_int64 offset; |
| switch (GET_CODE (x)) |
| { |
| case REG: |
| new_rtx = instantiate_new_reg (x, &offset); |
| if (new_rtx) |
| { |
| *loc = plus_constant (GET_MODE (x), new_rtx, offset); |
| changed = true; |
| } |
| iter.skip_subrtxes (); |
| break; |
| |
| case PLUS: |
| new_rtx = instantiate_new_reg (XEXP (x, 0), &offset); |
| if (new_rtx) |
| { |
| XEXP (x, 0) = new_rtx; |
| *loc = plus_constant (GET_MODE (x), x, offset, true); |
| changed = true; |
| iter.skip_subrtxes (); |
| break; |
| } |
| |
| /* FIXME -- from old code */ |
| /* If we have (plus (subreg (virtual-reg)) (const_int)), we know |
| we can commute the PLUS and SUBREG because pointers into the |
| frame are well-behaved. */ |
| break; |
| |
| default: |
| break; |
| } |
| } |
| } |
| return changed; |
| } |
| |
| /* A subroutine of instantiate_virtual_regs_in_insn. Return true if X |
| matches the predicate for insn CODE operand OPERAND. */ |
| |
| static int |
| safe_insn_predicate (int code, int operand, rtx x) |
| { |
| return code < 0 || insn_operand_matches ((enum insn_code) code, operand, x); |
| } |
| |
| /* A subroutine of instantiate_virtual_regs. Instantiate any virtual |
| registers present inside of insn. The result will be a valid insn. */ |
| |
| static void |
| instantiate_virtual_regs_in_insn (rtx_insn *insn) |
| { |
| poly_int64 offset; |
| int insn_code, i; |
| bool any_change = false; |
| rtx set, new_rtx, x; |
| rtx_insn *seq; |
| |
| /* There are some special cases to be handled first. */ |
| set = single_set (insn); |
| if (set) |
| { |
| /* We're allowed to assign to a virtual register. This is interpreted |
| to mean that the underlying register gets assigned the inverse |
| transformation. This is used, for example, in the handling of |
| non-local gotos. */ |
| new_rtx = instantiate_new_reg (SET_DEST (set), &offset); |
| if (new_rtx) |
| { |
| start_sequence (); |
| |
| instantiate_virtual_regs_in_rtx (&SET_SRC (set)); |
| x = simplify_gen_binary (PLUS, GET_MODE (new_rtx), SET_SRC (set), |
| gen_int_mode (-offset, GET_MODE (new_rtx))); |
| x = force_operand (x, new_rtx); |
| if (x != new_rtx) |
| emit_move_insn (new_rtx, x); |
| |
| seq = get_insns (); |
| end_sequence (); |
| |
| emit_insn_before (seq, insn); |
| delete_insn (insn); |
| return; |
| } |
| |
| /* Handle a straight copy from a virtual register by generating a |
| new add insn. The difference between this and falling through |
| to the generic case is avoiding a new pseudo and eliminating a |
| move insn in the initial rtl stream. */ |
| new_rtx = instantiate_new_reg (SET_SRC (set), &offset); |
| if (new_rtx |
| && maybe_ne (offset, 0) |
| && REG_P (SET_DEST (set)) |
| && REGNO (SET_DEST (set)) > LAST_VIRTUAL_REGISTER) |
| { |
| start_sequence (); |
| |
| x = expand_simple_binop (GET_MODE (SET_DEST (set)), PLUS, new_rtx, |
| gen_int_mode (offset, |
| GET_MODE (SET_DEST (set))), |
| SET_DEST (set), 1, OPTAB_LIB_WIDEN); |
| if (x != SET_DEST (set)) |
| emit_move_insn (SET_DEST (set), x); |
| |
| seq = get_insns (); |
| end_sequence (); |
| |
| emit_insn_before (seq, insn); |
| delete_insn (insn); |
| return; |
| } |
| |
| extract_insn (insn); |
| insn_code = INSN_CODE (insn); |
| |
| /* Handle a plus involving a virtual register by determining if the |
| operands remain valid if they're modified in place. */ |
| poly_int64 delta; |
| if (GET_CODE (SET_SRC (set)) == PLUS |
| && recog_data.n_operands >= 3 |
| && recog_data.operand_loc[1] == &XEXP (SET_SRC (set), 0) |
| && recog_data.operand_loc[2] == &XEXP (SET_SRC (set), 1) |
| && poly_int_rtx_p (recog_data.operand[2], &delta) |
| && (new_rtx = instantiate_new_reg (recog_data.operand[1], &offset))) |
| { |
| offset += delta; |
| |
| /* If the sum is zero, then replace with a plain move. */ |
| if (known_eq (offset, 0) |
| && REG_P (SET_DEST (set)) |
| && REGNO (SET_DEST (set)) > LAST_VIRTUAL_REGISTER) |
| { |
| start_sequence (); |
| emit_move_insn (SET_DEST (set), new_rtx); |
| seq = get_insns (); |
| end_sequence (); |
| |
| emit_insn_before (seq, insn); |
| delete_insn (insn); |
| return; |
| } |
| |
| x = gen_int_mode (offset, recog_data.operand_mode[2]); |
| |
| /* Using validate_change and apply_change_group here leaves |
| recog_data in an invalid state. Since we know exactly what |
| we want to check, do those two by hand. */ |
| if (safe_insn_predicate (insn_code, 1, new_rtx) |
| && safe_insn_predicate (insn_code, 2, x)) |
| { |
| *recog_data.operand_loc[1] = recog_data.operand[1] = new_rtx; |
| *recog_data.operand_loc[2] = recog_data.operand[2] = x; |
| any_change = true; |
| |
| /* Fall through into the regular operand fixup loop in |
| order to take care of operands other than 1 and 2. */ |
| } |
| } |
| } |
| else |
| { |
| extract_insn (insn); |
| insn_code = INSN_CODE (insn); |
| } |
| |
| /* In the general case, we expect virtual registers to appear only in |
| operands, and then only as either bare registers or inside memories. */ |
| for (i = 0; i < recog_data.n_operands; ++i) |
| { |
| x = recog_data.operand[i]; |
| switch (GET_CODE (x)) |
| { |
| case MEM: |
| { |
| rtx addr = XEXP (x, 0); |
| |
| if (!instantiate_virtual_regs_in_rtx (&addr)) |
| continue; |
| |
| start_sequence (); |
| x = replace_equiv_address (x, addr, true); |
| /* It may happen that the address with the virtual reg |
| was valid (e.g. based on the virtual stack reg, which might |
| be acceptable to the predicates with all offsets), whereas |
| the address now isn't anymore, for instance when the address |
| is still offsetted, but the base reg isn't virtual-stack-reg |
| anymore. Below we would do a force_reg on the whole operand, |
| but this insn might actually only accept memory. Hence, |
| before doing that last resort, try to reload the address into |
| a register, so this operand stays a MEM. */ |
| if (!safe_insn_predicate (insn_code, i, x)) |
| { |
| addr = force_reg (GET_MODE (addr), addr); |
| x = replace_equiv_address (x, addr, true); |
| } |
| seq = get_insns (); |
| end_sequence (); |
| if (seq) |
| emit_insn_before (seq, insn); |
| } |
| break; |
| |
| case REG: |
| new_rtx = instantiate_new_reg (x, &offset); |
| if (new_rtx == NULL) |
| continue; |
| if (known_eq (offset, 0)) |
| x = new_rtx; |
| else |
| { |
| start_sequence (); |
| |
| /* Careful, special mode predicates may have stuff in |
| insn_data[insn_code].operand[i].mode that isn't useful |
| to us for computing a new value. */ |
| /* ??? Recognize address_operand and/or "p" constraints |
| to see if (plus new offset) is a valid before we put |
| this through expand_simple_binop. */ |
| x = expand_simple_binop (GET_MODE (x), PLUS, new_rtx, |
| gen_int_mode (offset, GET_MODE (x)), |
| NULL_RTX, 1, OPTAB_LIB_WIDEN); |
| seq = get_insns (); |
| end_sequence (); |
| emit_insn_before (seq, insn); |
| } |
| break; |
| |
| case SUBREG: |
| new_rtx = instantiate_new_reg (SUBREG_REG (x), &offset); |
| if (new_rtx == NULL) |
| continue; |
| if (maybe_ne (offset, 0)) |
| { |
| start_sequence (); |
| new_rtx = expand_simple_binop |
| (GET_MODE (new_rtx), PLUS, new_rtx, |
| gen_int_mode (offset, GET_MODE (new_rtx)), |
| NULL_RTX, 1, OPTAB_LIB_WIDEN); |
| seq = get_insns (); |
| end_sequence (); |
| emit_insn_before (seq, insn); |
| } |
| x = simplify_gen_subreg (recog_data.operand_mode[i], new_rtx, |
| GET_MODE (new_rtx), SUBREG_BYTE (x)); |
| gcc_assert (x); |
| break; |
| |
| default: |
| continue; |
| } |
| |
| /* At this point, X contains the new value for the operand. |
| Validate the new value vs the insn predicate. Note that |
| asm insns will have insn_code -1 here. */ |
| if (!safe_insn_predicate (insn_code, i, x)) |
| { |
| start_sequence (); |
| if (REG_P (x)) |
| { |
| gcc_assert (REGNO (x) <= LAST_VIRTUAL_REGISTER); |
| x = copy_to_reg (x); |
| } |
| else |
| x = force_reg (insn_data[insn_code].operand[i].mode, x); |
| seq = get_insns (); |
| end_sequence (); |
| if (seq) |
| emit_insn_before (seq, insn); |
| } |
| |
| *recog_data.operand_loc[i] = recog_data.operand[i] = x; |
| any_change = true; |
| } |
| |
| if (any_change) |
| { |
| /* Propagate operand changes into the duplicates. */ |
| for (i = 0; i < recog_data.n_dups; ++i) |
| *recog_data.dup_loc[i] |
| = copy_rtx (recog_data.operand[(unsigned)recog_data.dup_num[i]]); |
| |
| /* Force re-recognition of the instruction for validation. */ |
| INSN_CODE (insn) = -1; |
| } |
| |
| if (asm_noperands (PATTERN (insn)) >= 0) |
| { |
| if (!check_asm_operands (PATTERN (insn))) |
| { |
| error_for_asm (insn, "impossible constraint in %<asm%>"); |
| /* For asm goto, instead of fixing up all the edges |
| just clear the template and clear input and output operands |
| and strip away clobbers. */ |
| if (JUMP_P (insn)) |
| { |
| rtx asm_op = extract_asm_operands (PATTERN (insn)); |
| PATTERN (insn) = asm_op; |
| PUT_MODE (asm_op, VOIDmode); |
| ASM_OPERANDS_TEMPLATE (asm_op) = ggc_strdup (""); |
| ASM_OPERANDS_OUTPUT_CONSTRAINT (asm_op) = ""; |
| ASM_OPERANDS_OUTPUT_IDX (asm_op) = 0; |
| ASM_OPERANDS_INPUT_VEC (asm_op) = rtvec_alloc (0); |
| ASM_OPERANDS_INPUT_CONSTRAINT_VEC (asm_op) = rtvec_alloc (0); |
| } |
| else |
| delete_insn (insn); |
| } |
| } |
| else |
| { |
| if (recog_memoized (insn) < 0) |
| fatal_insn_not_found (insn); |
| } |
| } |
| |
| /* Subroutine of instantiate_decls. Given RTL representing a decl, |
| do any instantiation required. */ |
| |
| void |
| instantiate_decl_rtl (rtx x) |
| { |
| rtx addr; |
| |
| if (x == 0) |
| return; |
| |
| /* If this is a CONCAT, recurse for the pieces. */ |
| if (GET_CODE (x) == CONCAT) |
| { |
| instantiate_decl_rtl (XEXP (x, 0)); |
| instantiate_decl_rtl (XEXP (x, 1)); |
| return; |
| } |
| |
| /* If this is not a MEM, no need to do anything. Similarly if the |
| address is a constant or a register that is not a virtual register. */ |
| if (!MEM_P (x)) |
| return; |
| |
| addr = XEXP (x, 0); |
| if (CONSTANT_P (addr) |
| || (REG_P (addr) |
| && (REGNO (addr) < FIRST_VIRTUAL_REGISTER |
| || REGNO (addr) > LAST_VIRTUAL_REGISTER))) |
| return; |
| |
| instantiate_virtual_regs_in_rtx (&XEXP (x, 0)); |
| } |
| |
| /* Helper for instantiate_decls called via walk_tree: Process all decls |
| in the given DECL_VALUE_EXPR. */ |
| |
| static tree |
| instantiate_expr (tree *tp, int *walk_subtrees, void *data ATTRIBUTE_UNUSED) |
| { |
| tree t = *tp; |
| if (! EXPR_P (t)) |
| { |
| *walk_subtrees = 0; |
| if (DECL_P (t)) |
| { |
| if (DECL_RTL_SET_P (t)) |
| instantiate_decl_rtl (DECL_RTL (t)); |
| if (TREE_CODE (t) == PARM_DECL && DECL_NAMELESS (t) |
| && DECL_INCOMING_RTL (t)) |
| instantiate_decl_rtl (DECL_INCOMING_RTL (t)); |
| if ((VAR_P (t) || TREE_CODE (t) == RESULT_DECL) |
| && DECL_HAS_VALUE_EXPR_P (t)) |
| { |
| tree v = DECL_VALUE_EXPR (t); |
| walk_tree (&v, instantiate_expr, NULL, NULL); |
| } |
| } |
| } |
| return NULL; |
| } |
| |
| /* Subroutine of instantiate_decls: Process all decls in the given |
| BLOCK node and all its subblocks. */ |
| |
| static void |
| instantiate_decls_1 (tree let) |
| { |
| tree t; |
| |
| for (t = BLOCK_VARS (let); t; t = DECL_CHAIN (t)) |
| { |
| if (DECL_RTL_SET_P (t)) |
| instantiate_decl_rtl (DECL_RTL (t)); |
| if (VAR_P (t) && DECL_HAS_VALUE_EXPR_P (t)) |
| { |
| tree v = DECL_VALUE_EXPR (t); |
| walk_tree (&v, instantiate_expr, NULL, NULL); |
| } |
| } |
| |
| /* Process all subblocks. */ |
| for (t = BLOCK_SUBBLOCKS (let); t; t = BLOCK_CHAIN (t)) |
| instantiate_decls_1 (t); |
| } |
| |
| /* Scan all decls in FNDECL (both variables and parameters) and instantiate |
| all virtual registers in their DECL_RTL's. */ |
| |
| static void |
| instantiate_decls (tree fndecl) |
| { |
| tree decl; |
| unsigned ix; |
| |
| /* Process all parameters of the function. */ |
| for (decl = DECL_ARGUMENTS (fndecl); decl; decl = DECL_CHAIN (decl)) |
| { |
| instantiate_decl_rtl (DECL_RTL (decl)); |
| instantiate_decl_rtl (DECL_INCOMING_RTL (decl)); |
| if (DECL_HAS_VALUE_EXPR_P (decl)) |
| { |
| tree v = DECL_VALUE_EXPR (decl); |
| walk_tree (&v, instantiate_expr, NULL, NULL); |
| } |
| } |
| |
| if ((decl = DECL_RESULT (fndecl)) |
| && TREE_CODE (decl) == RESULT_DECL) |
| { |
| if (DECL_RTL_SET_P (decl)) |
| instantiate_decl_rtl (DECL_RTL (decl)); |
| if (DECL_HAS_VALUE_EXPR_P (decl)) |
| { |
| tree v = DECL_VALUE_EXPR (decl); |
| walk_tree (&v, instantiate_expr, NULL, NULL); |
| } |
| } |
| |
| /* Process the saved static chain if it exists. */ |
| decl = DECL_STRUCT_FUNCTION (fndecl)->static_chain_decl; |
| if (decl && DECL_HAS_VALUE_EXPR_P (decl)) |
| instantiate_decl_rtl (DECL_RTL (DECL_VALUE_EXPR (decl))); |
| |
| /* Now process all variables defined in the function or its subblocks. */ |
| if (DECL_INITIAL (fndecl)) |
| instantiate_decls_1 (DECL_INITIAL (fndecl)); |
| |
| FOR_EACH_LOCAL_DECL (cfun, ix, decl) |
| if (DECL_RTL_SET_P (decl)) |
| instantiate_decl_rtl (DECL_RTL (decl)); |
| vec_free (cfun->local_decls); |
| } |
| |
| /* Pass through the INSNS of function FNDECL and convert virtual register |
| references to hard register references. */ |
| |
| static unsigned int |
| instantiate_virtual_regs (void) |
| { |
| rtx_insn *insn; |
| |
| /* Compute the offsets to use for this function. */ |
| in_arg_offset = FIRST_PARM_OFFSET (current_function_decl); |
| var_offset = targetm.starting_frame_offset (); |
| dynamic_offset = STACK_DYNAMIC_OFFSET (current_function_decl); |
| out_arg_offset = STACK_POINTER_OFFSET; |
| #ifdef FRAME_POINTER_CFA_OFFSET |
| cfa_offset = FRAME_POINTER_CFA_OFFSET (current_function_decl); |
| #else |
| cfa_offset = ARG_POINTER_CFA_OFFSET (current_function_decl); |
| #endif |
| |
| /* Initialize recognition, indicating that volatile is OK. */ |
| init_recog (); |
| |
| /* Scan through all the insns, instantiating every virtual register still |
| present. */ |
| for (insn = get_insns (); insn; insn = NEXT_INSN (insn)) |
| if (INSN_P (insn)) |
| { |
| /* These patterns in the instruction stream can never be recognized. |
| Fortunately, they shouldn't contain virtual registers either. */ |
| if (GET_CODE (PATTERN (insn)) == USE |
| || GET_CODE (PATTERN (insn)) == CLOBBER |
| || GET_CODE (PATTERN (insn)) == ASM_INPUT |
| || DEBUG_MARKER_INSN_P (insn)) |
| continue; |
| else if (DEBUG_BIND_INSN_P (insn)) |
| instantiate_virtual_regs_in_rtx (INSN_VAR_LOCATION_PTR (insn)); |
| else |
| instantiate_virtual_regs_in_insn (insn); |
| |
| if (insn->deleted ()) |
| continue; |
| |
| instantiate_virtual_regs_in_rtx (®_NOTES (insn)); |
| |
| /* Instantiate any virtual registers in CALL_INSN_FUNCTION_USAGE. */ |
| if (CALL_P (insn)) |
| instantiate_virtual_regs_in_rtx (&CALL_INSN_FUNCTION_USAGE (insn)); |
| } |
| |
| /* Instantiate the virtual registers in the DECLs for debugging purposes. */ |
| instantiate_decls (current_function_decl); |
| |
| targetm.instantiate_decls (); |
| |
| /* Indicate that, from now on, assign_stack_local should use |
| frame_pointer_rtx. */ |
| virtuals_instantiated = 1; |
| |
| return 0; |
| } |
| |
| namespace { |
| |
| const pass_data pass_data_instantiate_virtual_regs = |
| { |
| RTL_PASS, /* type */ |
| "vregs", /* name */ |
| OPTGROUP_NONE, /* optinfo_flags */ |
| TV_NONE, /* tv_id */ |
| 0, /* properties_required */ |
| 0, /* properties_provided */ |
| 0, /* properties_destroyed */ |
| 0, /* todo_flags_start */ |
| 0, /* todo_flags_finish */ |
| }; |
| |
| class pass_instantiate_virtual_regs : public rtl_opt_pass |
| { |
| public: |
| pass_instantiate_virtual_regs (gcc::context *ctxt) |
| : rtl_opt_pass (pass_data_instantiate_virtual_regs, ctxt) |
| {} |
| |
| /* opt_pass methods: */ |
| virtual unsigned int execute (function *) |
| { |
| return instantiate_virtual_regs (); |
| } |
| |
| }; // class pass_instantiate_virtual_regs |
| |
| } // anon namespace |
| |
| rtl_opt_pass * |
| make_pass_instantiate_virtual_regs (gcc::context *ctxt) |
| { |
| return new pass_instantiate_virtual_regs (ctxt); |
| } |
| |
| |
| /* Return 1 if EXP is an aggregate type (or a value with aggregate type). |
| This means a type for which function calls must pass an address to the |
| function or get an address back from the function. |
| EXP may be a type node or an expression (whose type is tested). */ |
| |
| int |
| aggregate_value_p (const_tree exp, const_tree fntype) |
| { |
| const_tree type = (TYPE_P (exp)) ? exp : TREE_TYPE (exp); |
| int i, regno, nregs; |
| rtx reg; |
| |
| if (fntype) |
| switch (TREE_CODE (fntype)) |
| { |
| case CALL_EXPR: |
| { |
| tree fndecl = get_callee_fndecl (fntype); |
| if (fndecl) |
| fntype = TREE_TYPE (fndecl); |
| else if (CALL_EXPR_FN (fntype)) |
| fntype = TREE_TYPE (TREE_TYPE (CALL_EXPR_FN (fntype))); |
| else |
| /* For internal functions, assume nothing needs to be |
| returned in memory. */ |
| return 0; |
| } |
| break; |
| case FUNCTION_DECL: |
| fntype = TREE_TYPE (fntype); |
| break; |
| case FUNCTION_TYPE: |
| case METHOD_TYPE: |
| break; |
| case IDENTIFIER_NODE: |
| fntype = NULL_TREE; |
| break; |
| default: |
| /* We don't expect other tree types here. */ |
| gcc_unreachable (); |
| } |
| |
| if (VOID_TYPE_P (type)) |
| return 0; |
| |
| /* If a record should be passed the same as its first (and only) member |
| don't pass it as an aggregate. */ |
| if (TREE_CODE (type) == RECORD_TYPE && TYPE_TRANSPARENT_AGGR (type)) |
| return aggregate_value_p (first_field (type), fntype); |
| |
| /* If the front end has decided that this needs to be passed by |
| reference, do so. */ |
| if ((TREE_CODE (exp) == PARM_DECL || TREE_CODE (exp) == RESULT_DECL) |
| && DECL_BY_REFERENCE (exp)) |
| return 1; |
| |
| /* Function types that are TREE_ADDRESSABLE force return in memory. */ |
| if (fntype && TREE_ADDRESSABLE (fntype)) |
| return 1; |
| |
| /* Types that are TREE_ADDRESSABLE must be constructed in memory, |
| and thus can't be returned in registers. */ |
| if (TREE_ADDRESSABLE (type)) |
| return 1; |
| |
| if (TYPE_EMPTY_P (type)) |
| return 0; |
| |
| if (flag_pcc_struct_return && AGGREGATE_TYPE_P (type)) |
| return 1; |
| |
| if (targetm.calls.return_in_memory (type, fntype)) |
| return 1; |
| |
| /* Make sure we have suitable call-clobbered regs to return |
| the value in; if not, we must return it in memory. */ |
| reg = hard_function_value (type, 0, fntype, 0); |
| |
| /* If we have something other than a REG (e.g. a PARALLEL), then assume |
| it is OK. */ |
| if (!REG_P (reg)) |
| return 0; |
| |
| /* Use the default ABI if the type of the function isn't known. |
| The scheme for handling interoperability between different ABIs |
| requires us to be able to tell when we're calling a function with |
| a nondefault ABI. */ |
| const predefined_function_abi &abi = (fntype |
| ? fntype_abi (fntype) |
| : default_function_abi); |
| regno = REGNO (reg); |
| nregs = hard_regno_nregs (regno, TYPE_MODE (type)); |
| for (i = 0; i < nregs; i++) |
| if (!fixed_regs[regno + i] && !abi.clobbers_full_reg_p (regno + i)) |
| return 1; |
| |
| return 0; |
| } |
| |
| /* Return true if we should assign DECL a pseudo register; false if it |
| should live on the local stack. */ |
| |
| bool |
| use_register_for_decl (const_tree decl) |
| { |
| if (TREE_CODE (decl) == SSA_NAME) |
| { |
| /* We often try to use the SSA_NAME, instead of its underlying |
| decl, to get type information and guide decisions, to avoid |
| differences of behavior between anonymous and named |
| variables, but in this one case we have to go for the actual |
| variable if there is one. The main reason is that, at least |
| at -O0, we want to place user variables on the stack, but we |
| don't mind using pseudos for anonymous or ignored temps. |
| Should we take the SSA_NAME, we'd conclude all SSA_NAMEs |
| should go in pseudos, whereas their corresponding variables |
| might have to go on the stack. So, disregarding the decl |
| here would negatively impact debug info at -O0, enable |
| coalescing between SSA_NAMEs that ought to get different |
| stack/pseudo assignments, and get the incoming argument |
| processing thoroughly confused by PARM_DECLs expected to live |
| in stack slots but assigned to pseudos. */ |
| if (!SSA_NAME_VAR (decl)) |
| return TYPE_MODE (TREE_TYPE (decl)) != BLKmode |
| && !(flag_float_store && FLOAT_TYPE_P (TREE_TYPE (decl))); |
| |
| decl = SSA_NAME_VAR (decl); |
| } |
| |
| /* Honor volatile. */ |
| if (TREE_SIDE_EFFECTS (decl)) |
| return false; |
| |
| /* Honor addressability. */ |
| if (TREE_ADDRESSABLE (decl)) |
| return false; |
| |
| /* RESULT_DECLs are a bit special in that they're assigned without |
| regard to use_register_for_decl, but we generally only store in |
| them. If we coalesce their SSA NAMEs, we'd better return a |
| result that matches the assignment in expand_function_start. */ |
| if (TREE_CODE (decl) == RESULT_DECL) |
| { |
| /* If it's not an aggregate, we're going to use a REG or a |
| PARALLEL containing a REG. */ |
| if (!aggregate_value_p (decl, current_function_decl)) |
| return true; |
| |
| /* If expand_function_start determines the return value, we'll |
| use MEM if it's not by reference. */ |
| if (cfun->returns_pcc_struct |
| || (targetm.calls.struct_value_rtx |
| (TREE_TYPE (current_function_decl), 1))) |
| return DECL_BY_REFERENCE (decl); |
| |
| /* Otherwise, we're taking an extra all.function_result_decl |
| argument. It's set up in assign_parms_augmented_arg_list, |
| under the (negated) conditions above, and then it's used to |
| set up the RESULT_DECL rtl in assign_params, after looping |
| over all parameters. Now, if the RESULT_DECL is not by |
| reference, we'll use a MEM either way. */ |
| if (!DECL_BY_REFERENCE (decl)) |
| return false; |
| |
| /* Otherwise, if RESULT_DECL is DECL_BY_REFERENCE, it will take |
| the function_result_decl's assignment. Since it's a pointer, |
| we can short-circuit a number of the tests below, and we must |
| duplicate them because we don't have the function_result_decl |
| to test. */ |
| if (!targetm.calls.allocate_stack_slots_for_args ()) |
| return true; |
| /* We don't set DECL_IGNORED_P for the function_result_decl. */ |
| if (optimize) |
| return true; |
| if (cfun->tail_call_marked) |
| return true; |
| /* We don't set DECL_REGISTER for the function_result_decl. */ |
| return false; |
| } |
| |
| /* Only register-like things go in registers. */ |
| if (DECL_MODE (decl) == BLKmode) |
| return false; |
| |
| /* If -ffloat-store specified, don't put explicit float variables |
| into registers. */ |
| /* ??? This should be checked after DECL_ARTIFICIAL, but tree-ssa |
| propagates values across these stores, and it probably shouldn't. */ |
| if (flag_float_store && FLOAT_TYPE_P (TREE_TYPE (decl))) |
| return false; |
| |
| if (!targetm.calls.allocate_stack_slots_for_args ()) |
| return true; |
| |
| /* If we're not interested in tracking debugging information for |
| this decl, then we can certainly put it in a register. */ |
| if (DECL_IGNORED_P (decl)) |
| return true; |
| |
| if (optimize) |
| return true; |
| |
| /* Thunks force a tail call even at -O0 so we need to avoid creating a |
| dangling reference in case the parameter is passed by reference. */ |
| if (TREE_CODE (decl) == PARM_DECL && cfun->tail_call_marked) |
| return true; |
| |
| if (!DECL_REGISTER (decl)) |
| return false; |
| |
| /* When not optimizing, disregard register keyword for types that |
| could have methods, otherwise the methods won't be callable from |
| the debugger. */ |
| if (RECORD_OR_UNION_TYPE_P (TREE_TYPE (decl))) |
| return false; |
| |
| return true; |
| } |
| |
| /* Structures to communicate between the subroutines of assign_parms. |
| The first holds data persistent across all parameters, the second |
| is cleared out for each parameter. */ |
| |
| struct assign_parm_data_all |
| { |
| /* When INIT_CUMULATIVE_ARGS gets revamped, allocating CUMULATIVE_ARGS |
| should become a job of the target or otherwise encapsulated. */ |
| CUMULATIVE_ARGS args_so_far_v; |
| cumulative_args_t args_so_far; |
| struct args_size stack_args_size; |
| tree function_result_decl; |
| tree orig_fnargs; |
| rtx_insn *first_conversion_insn; |
| rtx_insn *last_conversion_insn; |
| HOST_WIDE_INT pretend_args_size; |
| HOST_WIDE_INT extra_pretend_bytes; |
| int reg_parm_stack_space; |
| }; |
| |
| struct assign_parm_data_one |
| { |
| tree nominal_type; |
| function_arg_info arg; |
| rtx entry_parm; |
| rtx stack_parm; |
| machine_mode nominal_mode; |
| machine_mode passed_mode; |
| struct locate_and_pad_arg_data locate; |
| int partial; |
| }; |
| |
| /* A subroutine of assign_parms. Initialize ALL. */ |
| |
| static void |
| assign_parms_initialize_all (struct assign_parm_data_all *all) |
| { |
| tree fntype ATTRIBUTE_UNUSED; |
| |
| memset (all, 0, sizeof (*all)); |
| |
| fntype = TREE_TYPE (current_function_decl); |
| |
| #ifdef INIT_CUMULATIVE_INCOMING_ARGS |
| INIT_CUMULATIVE_INCOMING_ARGS (all->args_so_far_v, fntype, NULL_RTX); |
| #else |
| INIT_CUMULATIVE_ARGS (all->args_so_far_v, fntype, NULL_RTX, |
| current_function_decl, -1); |
| #endif |
| all->args_so_far = pack_cumulative_args (&all->args_so_far_v); |
| |
| #ifdef INCOMING_REG_PARM_STACK_SPACE |
| all->reg_parm_stack_space |
| = INCOMING_REG_PARM_STACK_SPACE (current_function_decl); |
| #endif |
| } |
| |
| /* If ARGS contains entries with complex types, split the entry into two |
| entries of the component type. Return a new list of substitutions are |
| needed, else the old list. */ |
| |
| static void |
| split_complex_args (vec<tree> *args) |
| { |
| unsigned i; |
| tree p; |
| |
| FOR_EACH_VEC_ELT (*args, i, p) |
| { |
| tree type = TREE_TYPE (p); |
| if (TREE_CODE (type) == COMPLEX_TYPE |
| && targetm.calls.split_complex_arg (type)) |
| { |
| tree decl; |
| tree subtype = TREE_TYPE (type); |
| bool addressable = TREE_ADDRESSABLE (p); |
| |
| /* Rewrite the PARM_DECL's type with its component. */ |
| p = copy_node (p); |
| TREE_TYPE (p) = subtype; |
| DECL_ARG_TYPE (p) = TREE_TYPE (DECL_ARG_TYPE (p)); |
| SET_DECL_MODE (p, VOIDmode); |
| DECL_SIZE (p) = NULL; |
| DECL_SIZE_UNIT (p) = NULL; |
| /* If this arg must go in memory, put it in a pseudo here. |
| We can't allow it to go in memory as per normal parms, |
| because the usual place might not have the imag part |
| adjacent to the real part. */ |
| DECL_ARTIFICIAL (p) = addressable; |
| DECL_IGNORED_P (p) = addressable; |
| TREE_ADDRESSABLE (p) = 0; |
| layout_decl (p, 0); |
| (*args)[i] = p; |
| |
| /* Build a second synthetic decl. */ |
| decl = build_decl (EXPR_LOCATION (p), |
| PARM_DECL, NULL_TREE, subtype); |
| DECL_ARG_TYPE (decl) = DECL_ARG_TYPE (p); |
| DECL_ARTIFICIAL (decl) = addressable; |
| DECL_IGNORED_P (decl) = addressable; |
| layout_decl (decl, 0); |
| args->safe_insert (++i, decl); |
| } |
| } |
| } |
| |
| /* A subroutine of assign_parms. Adjust the parameter list to incorporate |
| the hidden struct return argument, and (abi willing) complex args. |
| Return the new parameter list. */ |
| |
| static vec<tree> |
| assign_parms_augmented_arg_list (struct assign_parm_data_all *all) |
| { |
| tree fndecl = current_function_decl; |
| tree fntype = TREE_TYPE (fndecl); |
| vec<tree> fnargs = vNULL; |
| tree arg; |
| |
| for (arg = DECL_ARGUMENTS (fndecl); arg; arg = DECL_CHAIN (arg)) |
| fnargs.safe_push (arg); |
| |
| all->orig_fnargs = DECL_ARGUMENTS (fndecl); |
| |
| /* If struct value address is treated as the first argument, make it so. */ |
| if (aggregate_value_p (DECL_RESULT (fndecl), fndecl) |
| && ! cfun->returns_pcc_struct |
| && targetm.calls.struct_value_rtx (TREE_TYPE (fndecl), 1) == 0) |
| { |
| tree type = build_pointer_type (TREE_TYPE (fntype)); |
| tree decl; |
| |
| decl = build_decl (DECL_SOURCE_LOCATION (fndecl), |
| PARM_DECL, get_identifier (".result_ptr"), type); |
| DECL_ARG_TYPE (decl) = type; |
| DECL_ARTIFICIAL (decl) = 1; |
| DECL_NAMELESS (decl) = 1; |
| TREE_CONSTANT (decl) = 1; |
| /* We don't set DECL_IGNORED_P or DECL_REGISTER here. If this |
| changes, the end of the RESULT_DECL handling block in |
| use_register_for_decl must be adjusted to match. */ |
| |
| DECL_CHAIN (decl) = all->orig_fnargs; |
| all->orig_fnargs = decl; |
| fnargs.safe_insert (0, decl); |
| |
| all->function_result_decl = decl; |
| } |
| |
| /* If the target wants to split complex arguments into scalars, do so. */ |
| if (targetm.calls.split_complex_arg) |
| split_complex_args (&fnargs); |
| |
| return fnargs; |
| } |
| |
| /* A subroutine of assign_parms. Examine PARM and pull out type and mode |
| data for the parameter. Incorporate ABI specifics such as pass-by- |
| reference and type promotion. */ |
| |
| static void |
| assign_parm_find_data_types (struct assign_parm_data_all *all, tree parm, |
| struct assign_parm_data_one *data) |
| { |
| int unsignedp; |
| |
| #ifndef BROKEN_VALUE_INITIALIZATION |
| *data = assign_parm_data_one (); |
| #else |
| /* Old versions of GCC used to miscompile the above by only initializing |
| the members with explicit constructors and copying garbage |
| to the other members. */ |
| assign_parm_data_one zero_data = {}; |
| *data = zero_data; |
| #endif |
| |
| /* NAMED_ARG is a misnomer. We really mean 'non-variadic'. */ |
| if (!cfun->stdarg) |
| data->arg.named = 1; /* No variadic parms. */ |
| else if (DECL_CHAIN (parm)) |
| data->arg.named = 1; /* Not the last non-variadic parm. */ |
| else if (targetm.calls.strict_argument_naming (all->args_so_far)) |
| data->arg.named = 1; /* Only variadic ones are unnamed. */ |
| else |
| data->arg.named = 0; /* Treat as variadic. */ |
| |
| data->nominal_type = TREE_TYPE (parm); |
| data->arg.type = DECL_ARG_TYPE (parm); |
| |
| /* Look out for errors propagating this far. Also, if the parameter's |
| type is void then its value doesn't matter. */ |
| if (TREE_TYPE (parm) == error_mark_node |
| /* This can happen after weird syntax errors |
| or if an enum type is defined among the parms. */ |
| || TREE_CODE (parm) != PARM_DECL |
| || data->arg.type == NULL |
| || VOID_TYPE_P (data->nominal_type)) |
| { |
| data->nominal_type = data->arg.type = void_type_node; |
| data->nominal_mode = data->passed_mode = data->arg.mode = VOIDmode; |
| return; |
| } |
| |
| /* Find mode of arg as it is passed, and mode of arg as it should be |
| during execution of this function. */ |
| data->passed_mode = data->arg.mode = TYPE_MODE (data->arg.type); |
| data->nominal_mode = TYPE_MODE (data->nominal_type); |
| |
| /* If the parm is to be passed as a transparent union or record, use the |
| type of the first field for the tests below. We have already verified |
| that the modes are the same. */ |
| if (RECORD_OR_UNION_TYPE_P (data->arg.type) |
| && TYPE_TRANSPARENT_AGGR (data->arg.type)) |
| data->arg.type = TREE_TYPE (first_field (data->arg.type)); |
| |
| /* See if this arg was passed by invisible reference. */ |
| if (apply_pass_by_reference_rules (&all->args_so_far_v, data->arg)) |
| { |
| data->nominal_type = data->arg.type; |
| data->passed_mode = data->nominal_mode = data->arg.mode; |
| } |
| |
| /* Find mode as it is passed by the ABI. */ |
| unsignedp = TYPE_UNSIGNED (data->arg.type); |
| data->arg.mode |
| = promote_function_mode (data->arg.type, data->arg.mode, &unsignedp, |
| TREE_TYPE (current_function_decl), 0); |
| } |
| |
| /* A subroutine of assign_parms. Invoke setup_incoming_varargs. */ |
| |
| static void |
| assign_parms_setup_varargs (struct assign_parm_data_all *all, |
| struct assign_parm_data_one *data, bool no_rtl) |
| { |
| int varargs_pretend_bytes = 0; |
| |
| function_arg_info last_named_arg = data->arg; |
| last_named_arg.named = true; |
| targetm.calls.setup_incoming_varargs (all->args_so_far, last_named_arg, |
| &varargs_pretend_bytes, no_rtl); |
| |
| /* If the back-end has requested extra stack space, record how much is |
| needed. Do not change pretend_args_size otherwise since it may be |
| nonzero from an earlier partial argument. */ |
| if (varargs_pretend_bytes > 0) |
| all->pretend_args_size = varargs_pretend_bytes; |
| } |
| |
| /* A subroutine of assign_parms. Set DATA->ENTRY_PARM corresponding to |
| the incoming location of the current parameter. */ |
| |
| static void |
| assign_parm_find_entry_rtl (struct assign_parm_data_all *all, |
| struct assign_parm_data_one *data) |
| { |
| HOST_WIDE_INT pretend_bytes = 0; |
| rtx entry_parm; |
| bool in_regs; |
| |
| if (data->arg.mode == VOIDmode) |
| { |
| data->entry_parm = data->stack_parm = const0_rtx; |
| return; |
| } |
| |
| targetm.calls.warn_parameter_passing_abi (all->args_so_far, |
| data->arg.type); |
| |
| entry_parm = targetm.calls.function_incoming_arg (all->args_so_far, |
| data->arg); |
| if (entry_parm == 0) |
| data->arg.mode = data->passed_mode; |
| |
| /* Determine parm's home in the stack, in case it arrives in the stack |
| or we should pretend it did. Compute the stack position and rtx where |
| the argument arrives and its size. |
| |
| There is one complexity here: If this was a parameter that would |
| have been passed in registers, but wasn't only because it is |
| __builtin_va_alist, we want locate_and_pad_parm to treat it as if |
| it came in a register so that REG_PARM_STACK_SPACE isn't skipped. |
| In this case, we call FUNCTION_ARG with NAMED set to 1 instead of 0 |
| as it was the previous time. */ |
| in_regs = (entry_parm != 0); |
| #ifdef STACK_PARMS_IN_REG_PARM_AREA |
| in_regs = true; |
| #endif |
| if (!in_regs && !data->arg.named) |
| { |
| if (targetm.calls.pretend_outgoing_varargs_named (all->args_so_far)) |
| { |
| rtx tem; |
| function_arg_info named_arg = data->arg; |
| named_arg.named = true; |
| tem = targetm.calls.function_incoming_arg (all->args_so_far, |
| named_arg); |
| in_regs = tem != NULL; |
| } |
| } |
| |
| /* If this parameter was passed both in registers and in the stack, use |
| the copy on the stack. */ |
| if (targetm.calls.must_pass_in_stack (data->arg)) |
| entry_parm = 0; |
| |
| if (entry_parm) |
| { |
| int partial; |
| |
| partial = targetm.calls.arg_partial_bytes (all->args_so_far, data->arg); |
| data->partial = partial; |
| |
| /* The caller might already have allocated stack space for the |
| register parameters. */ |
| if (partial != 0 && all->reg_parm_stack_space == 0) |
| { |
| /* Part of this argument is passed in registers and part |
| is passed on the stack. Ask the prologue code to extend |
| the stack part so that we can recreate the full value. |
| |
| PRETEND_BYTES is the size of the registers we need to store. |
| CURRENT_FUNCTION_PRETEND_ARGS_SIZE is the amount of extra |
| stack space that the prologue should allocate. |
| |
| Internally, gcc assumes that the argument pointer is aligned |
| to STACK_BOUNDARY bits. This is used both for alignment |
| optimizations (see init_emit) and to locate arguments that are |
| aligned to more than PARM_BOUNDARY bits. We must preserve this |
| invariant by rounding CURRENT_FUNCTION_PRETEND_ARGS_SIZE up to |
| a stack boundary. */ |
| |
| /* We assume at most one partial arg, and it must be the first |
| argument on the stack. */ |
| gcc_assert (!all->extra_pretend_bytes && !all->pretend_args_size); |
| |
| pretend_bytes = partial; |
| all->pretend_args_size = CEIL_ROUND (pretend_bytes, STACK_BYTES); |
| |
| /* We want to align relative to the actual stack pointer, so |
| don't include this in the stack size until later. */ |
| all->extra_pretend_bytes = all->pretend_args_size; |
| } |
| } |
| |
| locate_and_pad_parm (data->arg.mode, data->arg.type, in_regs, |
| all->reg_parm_stack_space, |
| entry_parm ? data->partial : 0, current_function_decl, |
| &all->stack_args_size, &data->locate); |
| |
| /* Update parm_stack_boundary if this parameter is passed in the |
| stack. */ |
| if (!in_regs && crtl->parm_stack_boundary < data->locate.boundary) |
| crtl->parm_stack_boundary = data->locate.boundary; |
| |
| /* Adjust offsets to include the pretend args. */ |
| pretend_bytes = all->extra_pretend_bytes - pretend_bytes; |
| data->locate.slot_offset.constant += pretend_bytes; |
| data->locate.offset.constant += pretend_bytes; |
| |
| data->entry_parm = entry_parm; |
| } |
| |
| /* A subroutine of assign_parms. If there is actually space on the stack |
| for this parm, count it in stack_args_size and return true. */ |
| |
| static bool |
| assign_parm_is_stack_parm (struct assign_parm_data_all *all, |
| struct assign_parm_data_one *data) |
| { |
| /* Trivially true if we've no incoming register. */ |
| if (data->entry_parm == NULL) |
| ; |
| /* Also true if we're partially in registers and partially not, |
| since we've arranged to drop the entire argument on the stack. */ |
| else if (data->partial != 0) |
| ; |
| /* Also true if the target says that it's passed in both registers |
| and on the stack. */ |
| else if (GET_CODE (data->entry_parm) == PARALLEL |
| && XEXP (XVECEXP (data->entry_parm, 0, 0), 0) == NULL_RTX) |
| ; |
| /* Also true if the target says that there's stack allocated for |
| all register parameters. */ |
| else if (all->reg_parm_stack_space > 0) |
| ; |
| /* Otherwise, no, this parameter has no ABI defined stack slot. */ |
| else |
| return false; |
| |
| all->stack_args_size.constant += data->locate.size.constant; |
| if (data->locate.size.var) |
| ADD_PARM_SIZE (all->stack_args_size, data->locate.size.var); |
| |
| return true; |
| } |
| |
| /* A subroutine of assign_parms. Given that this parameter is allocated |
| stack space by the ABI, find it. */ |
| |
| static void |
| assign_parm_find_stack_rtl (tree parm, struct assign_parm_data_one *data) |
| { |
| rtx offset_rtx, stack_parm; |
| unsigned int align, boundary; |
| |
| /* If we're passing this arg using a reg, make its stack home the |
| aligned stack slot. */ |
| if (data->entry_parm) |
| offset_rtx = ARGS_SIZE_RTX (data->locate.slot_offset); |
| else |
| offset_rtx = ARGS_SIZE_RTX (data->locate.offset); |
| |
| stack_parm = crtl->args.internal_arg_pointer; |
| if (offset_rtx != const0_rtx) |
| stack_parm = gen_rtx_PLUS (Pmode, stack_parm, offset_rtx); |
| stack_parm = gen_rtx_MEM (data->arg.mode, stack_parm); |
| |
| if (!data->arg.pass_by_reference) |
| { |
| set_mem_attributes (stack_parm, parm, 1); |
| /* set_mem_attributes could set MEM_SIZE to the passed mode's size, |
| while promoted mode's size is needed. */ |
| if (data->arg.mode != BLKmode |
| && data->arg.mode != DECL_MODE (parm)) |
| { |
| set_mem_size (stack_parm, GET_MODE_SIZE (data->arg.mode)); |
| if (MEM_EXPR (stack_parm) && MEM_OFFSET_KNOWN_P (stack_parm)) |
| { |
| poly_int64 offset = subreg_lowpart_offset (DECL_MODE (parm), |
| data->arg.mode); |
| if (maybe_ne (offset, 0)) |
| set_mem_offset (stack_parm, MEM_OFFSET (stack_parm) - offset); |
| } |
| } |
| } |
| |
| boundary = data->locate.boundary; |
| align = BITS_PER_UNIT; |
| |
| /* If we're padding upward, we know that the alignment of the slot |
| is TARGET_FUNCTION_ARG_BOUNDARY. If we're using slot_offset, we're |
| intentionally forcing upward padding. Otherwise we have to come |
| up with a guess at the alignment based on OFFSET_RTX. */ |
| poly_int64 offset; |
| if (data->locate.where_pad == PAD_NONE || data->entry_parm) |
| align = boundary; |
| else if (data->locate.where_pad == PAD_UPWARD) |
| { |
| align = boundary; |
| /* If the argument offset is actually more aligned than the nominal |
| stack slot boundary, take advantage of that excess alignment. |
| Don't make any assumptions if STACK_POINTER_OFFSET is in use. */ |
| if (poly_int_rtx_p (offset_rtx, &offset) |
| && known_eq (STACK_POINTER_OFFSET, 0)) |
| { |
| unsigned int offset_align = known_alignment (offset) * BITS_PER_UNIT; |
| if (offset_align == 0 || offset_align > STACK_BOUNDARY) |
| offset_align = STACK_BOUNDARY; |
| align = MAX (align, offset_align); |
| } |
| } |
| else if (poly_int_rtx_p (offset_rtx, &offset)) |
| { |
| align = least_bit_hwi (boundary); |
| unsigned int offset_align = known_alignment (offset) * BITS_PER_UNIT; |
| if (offset_align != 0) |
| align = MIN (align, offset_align); |
| } |
| set_mem_align (stack_parm, align); |
| |
| if (data->entry_parm) |
| set_reg_attrs_for_parm (data->entry_parm, stack_parm); |
| |
| data->stack_parm = stack_parm; |
| } |
| |
| /* A subroutine of assign_parms. Adjust DATA->ENTRY_RTL such that it's |
| always valid and contiguous. */ |
| |
| static void |
| assign_parm_adjust_entry_rtl (struct assign_parm_data_one *data) |
| { |
| rtx entry_parm = data->entry_parm; |
| rtx stack_parm = data->stack_parm; |
| |
| /* If this parm was passed part in regs and part in memory, pretend it |
| arrived entirely in memory by pushing the register-part onto the stack. |
| In the special case of a DImode or DFmode that is split, we could put |
| it together in a pseudoreg directly, but for now that's not worth |
| bothering with. */ |
| if (data->partial != 0) |
| { |
| /* Handle calls that pass values in multiple non-contiguous |
| locations. The Irix 6 ABI has examples of this. */ |
| if (GET_CODE (entry_parm) == PARALLEL) |
| emit_group_store (validize_mem (copy_rtx (stack_parm)), entry_parm, |
| data->arg.type, int_size_in_bytes (data->arg.type)); |
| else |
| { |
| gcc_assert (data->partial % UNITS_PER_WORD == 0); |
| move_block_from_reg (REGNO (entry_parm), |
| validize_mem (copy_rtx (stack_parm)), |
| data->partial / UNITS_PER_WORD); |
| } |
| |
| entry_parm = stack_parm; |
| } |
| |
| /* If we didn't decide this parm came in a register, by default it came |
| on the stack. */ |
| else if (entry_parm == NULL) |
| entry_parm = stack_parm; |
| |
| /* When an argument is passed in multiple locations, we can't make use |
| of this information, but we can save some copying if the whole argument |
| is passed in a single register. */ |
| else if (GET_CODE (entry_parm) == PARALLEL |
| && data->nominal_mode != BLKmode |
| && data->passed_mode != BLKmode) |
| { |
| size_t i, len = XVECLEN (entry_parm, 0); |
| |
| for (i = 0; i < len; i++) |
| if (XEXP (XVECEXP (entry_parm, 0, i), 0) != NULL_RTX |
| && REG_P (XEXP (XVECEXP (entry_parm, 0, i), 0)) |
| && (GET_MODE (XEXP (XVECEXP (entry_parm, 0, i), 0)) |
| == data->passed_mode) |
| && INTVAL (XEXP (XVECEXP (entry_parm, 0, i), 1)) == 0) |
| { |
| entry_parm = XEXP (XVECEXP (entry_parm, 0, i), 0); |
| break; |
| } |
| } |
| |
| data->entry_parm = entry_parm; |
| } |
| |
| /* A subroutine of assign_parms. Reconstitute any values which were |
| passed in multiple registers and would fit in a single register. */ |
| |
| static void |
| assign_parm_remove_parallels (struct assign_parm_data_one *data) |
| { |
| rtx entry_parm = data->entry_parm; |
| |
| /* Convert the PARALLEL to a REG of the same mode as the parallel. |
| This can be done with register operations rather than on the |
| stack, even if we will store the reconstituted parameter on the |
| stack later. */ |
| if (GET_CODE (entry_parm) == PARALLEL && GET_MODE (entry_parm) != BLKmode) |
| { |
| rtx parmreg = gen_reg_rtx (GET_MODE (entry_parm)); |
| emit_group_store (parmreg, entry_parm, data->arg.type, |
| GET_MODE_SIZE (GET_MODE (entry_parm))); |
| entry_parm = parmreg; |
| } |
| |
| data->entry_parm = entry_parm; |
| } |
| |
| /* A subroutine of assign_parms. Adjust DATA->STACK_RTL such that it's |
| always valid and properly aligned. */ |
| |
| static void |
| assign_parm_adjust_stack_rtl (struct assign_parm_data_one *data) |
| { |
| rtx stack_parm = data->stack_parm; |
| |
| /* If we can't trust the parm stack slot to be aligned enough for its |
| ultimate type, don't use that slot after entry. We'll make another |
| stack slot, if we need one. */ |
| if (stack_parm |
| && ((GET_MODE_ALIGNMENT (data->nominal_mode) > MEM_ALIGN (stack_parm) |
| && ((optab_handler (movmisalign_optab, data->nominal_mode) |
| != CODE_FOR_nothing) |
| || targetm.slow_unaligned_access (data->nominal_mode, |
| MEM_ALIGN (stack_parm)))) |
| || (data->nominal_type |
| && TYPE_ALIGN (data->nominal_type) > MEM_ALIGN (stack_parm) |
| && MEM_ALIGN (stack_parm) < PREFERRED_STACK_BOUNDARY))) |
| stack_parm = NULL; |
| |
| /* If parm was passed in memory, and we need to convert it on entry, |
| don't store it back in that same slot. */ |
| else if (data->entry_parm == stack_parm |
| && data->nominal_mode != BLKmode |
| && data->nominal_mode != data->passed_mode) |
| stack_parm = NULL; |
| |
| /* If stack protection is in effect for this function, don't leave any |
| pointers in their passed stack slots. */ |
| else if (crtl->stack_protect_guard |
| && (flag_stack_protect == SPCT_FLAG_ALL |
| || data->arg.pass_by_reference |
| || POINTER_TYPE_P (data->nominal_type))) |
| stack_parm = NULL; |
| |
| data->stack_parm = stack_parm; |
| } |
| |
| /* A subroutine of assign_parms. Return true if the current parameter |
| should be stored as a BLKmode in the current frame. */ |
| |
| static bool |
| assign_parm_setup_block_p (struct assign_parm_data_one *data) |
| { |
| if (data->nominal_mode == BLKmode) |
| return true; |
| if (GET_MODE (data->entry_parm) == BLKmode) |
| return true; |
| |
| #ifdef BLOCK_REG_PADDING |
| /* Only assign_parm_setup_block knows how to deal with register arguments |
| that are padded at the least significant end. */ |
| if (REG_P (data->entry_parm) |
| && known_lt (GET_MODE_SIZE (data->arg.mode), UNITS_PER_WORD) |
| && (BLOCK_REG_PADDING (data->passed_mode, data->arg.type, 1) |
| == (BYTES_BIG_ENDIAN ? PAD_UPWARD : PAD_DOWNWARD))) |
| return true; |
| #endif |
| |
| return false; |
| } |
| |
| /* A subroutine of assign_parms. Arrange for the parameter to be |
| present and valid in DATA->STACK_RTL. */ |
| |
| static void |
| assign_parm_setup_block (struct assign_parm_data_all *all, |
| tree parm, struct assign_parm_data_one *data) |
| { |
| rtx entry_parm = data->entry_parm; |
| rtx stack_parm = data->stack_parm; |
| rtx target_reg = NULL_RTX; |
| bool in_conversion_seq = false; |
| HOST_WIDE_INT size; |
| HOST_WIDE_INT size_stored; |
| |
| if (GET_CODE (entry_parm) == PARALLEL) |
| entry_parm = emit_group_move_into_temps (entry_parm); |
| |
| /* If we want the parameter in a pseudo, don't use a stack slot. */ |
| if (is_gimple_reg (parm) && use_register_for_decl (parm)) |
| { |
| tree def = ssa_default_def (cfun, parm); |
| gcc_assert (def); |
| machine_mode mode = promote_ssa_mode (def, NULL); |
| rtx reg = gen_reg_rtx (mode); |
| if (GET_CODE (reg) != CONCAT) |
| stack_parm = reg; |
| else |
| { |
| target_reg = reg; |
| /* Avoid allocating a stack slot, if there isn't one |
| preallocated by the ABI. It might seem like we should |
| always prefer a pseudo, but converting between |
| floating-point and integer modes goes through the stack |
| on various machines, so it's better to use the reserved |
| stack slot than to risk wasting it and allocating more |
| for the conversion. */ |
| if (stack_parm == NULL_RTX) |
| { |
| int save = generating_concat_p; |
| generating_concat_p = 0; |
| stack_parm = gen_reg_rtx (mode); |
| generating_concat_p = save; |
| } |
| } |
| data->stack_parm = NULL; |
| } |
| |
| size = int_size_in_bytes (data->arg.type); |
| size_stored = CEIL_ROUND (size, UNITS_PER_WORD); |
| if (stack_parm == 0) |
| { |
| HOST_WIDE_INT parm_align |
| = (STRICT_ALIGNMENT |
| ? MAX (DECL_ALIGN (parm), BITS_PER_WORD) : DECL_ALIGN (parm)); |
| |
| SET_DECL_ALIGN (parm, parm_align); |
| if (DECL_ALIGN (parm) > MAX_SUPPORTED_STACK_ALIGNMENT) |
| { |
| rtx allocsize = gen_int_mode (size_stored, Pmode); |
| get_dynamic_stack_size (&allocsize, 0, DECL_ALIGN (parm), NULL); |
| stack_parm = assign_stack_local (BLKmode, UINTVAL (allocsize), |
| MAX_SUPPORTED_STACK_ALIGNMENT); |
| rtx addr = align_dynamic_address (XEXP (stack_parm, 0), |
| DECL_ALIGN (parm)); |
| mark_reg_pointer (addr, DECL_ALIGN (parm)); |
| stack_parm = gen_rtx_MEM (GET_MODE (stack_parm), addr); |
| MEM_NOTRAP_P (stack_parm) = 1; |
| } |
| else |
| stack_parm = assign_stack_local (BLKmode, size_stored, |
| DECL_ALIGN (parm)); |
| if (known_eq (GET_MODE_SIZE (GET_MODE (entry_parm)), size)) |
| PUT_MODE (stack_parm, GET_MODE (entry_parm)); |
| set_mem_attributes (stack_parm, parm, 1); |
| } |
| |
| /* If a BLKmode arrives in registers, copy it to a stack slot. Handle |
| calls that pass values in multiple non-contiguous locations. */ |
| if (REG_P (entry_parm) || GET_CODE (entry_parm) == PARALLEL) |
| { |
| rtx mem; |
| |
| /* Note that we will be storing an integral number of words. |
| So we have to be careful to ensure that we allocate an |
| integral number of words. We do this above when we call |
| assign_stack_local if space was not allocated in the argument |
| list. If it was, this will not work if PARM_BOUNDARY is not |
| a multiple of BITS_PER_WORD. It isn't clear how to fix this |
| if it becomes a problem. Exception is when BLKmode arrives |
| with arguments not conforming to word_mode. */ |
| |
| if (data->stack_parm == 0) |
| ; |
| else if (GET_CODE (entry_parm) == PARALLEL) |
| ; |
| else |
| gcc_assert (!size || !(PARM_BOUNDARY % BITS_PER_WORD)); |
| |
| mem = validize_mem (copy_rtx (stack_parm)); |
| |
| /* Handle values in multiple non-contiguous locations. */ |
| if (GET_CODE (entry_parm) == PARALLEL && !MEM_P (mem)) |
| emit_group_store (mem, entry_parm, data->arg.type, size); |
| else if (GET_CODE (entry_parm) == PARALLEL) |
| { |
| push_to_sequence2 (all->first_conversion_insn, |
| all->last_conversion_insn); |
| emit_group_store (mem, entry_parm, data->arg.type, size); |
| all->first_conversion_insn = get_insns (); |
| all->last_conversion_insn = get_last_insn (); |
| end_sequence (); |
| in_conversion_seq = true; |
| } |
| |
| else if (size == 0) |
| ; |
| |
| /* If SIZE is that of a mode no bigger than a word, just use |
| that mode's store operation. */ |
| else if (size <= UNITS_PER_WORD) |
| { |
| unsigned int bits = size * BITS_PER_UNIT; |
| machine_mode mode = int_mode_for_size (bits, 0).else_blk (); |
| |
| if (mode != BLKmode |
| #ifdef BLOCK_REG_PADDING |
| && (size == UNITS_PER_WORD |
| || (BLOCK_REG_PADDING (mode, data->arg.type, 1) |
| != (BYTES_BIG_ENDIAN ? PAD_UPWARD : PAD_DOWNWARD))) |
| #endif |
| ) |
| { |
| rtx reg; |
| |
| /* We are really truncating a word_mode value containing |
| SIZE bytes into a value of mode MODE. If such an |
| operation requires no actual instructions, we can refer |
| to the value directly in mode MODE, otherwise we must |
| start with the register in word_mode and explicitly |
| convert it. */ |
| if (mode == word_mode |
| || TRULY_NOOP_TRUNCATION_MODES_P (mode, word_mode)) |
| reg = gen_rtx_REG (mode, REGNO (entry_parm)); |
| else |
| { |
| reg = gen_rtx_REG (word_mode, REGNO (entry_parm)); |
| reg = convert_to_mode (mode, copy_to_reg (reg), 1); |
| } |
| emit_move_insn (change_address (mem, mode, 0), reg); |
| } |
| |
| #ifdef BLOCK_REG_PADDING |
| /* Storing the register in memory as a full word, as |
| move_block_from_reg below would do, and then using the |
| MEM in a smaller mode, has the effect of shifting right |
| if BYTES_BIG_ENDIAN. If we're bypassing memory, the |
| shifting must be explicit. */ |
| else if (!MEM_P (mem)) |
| { |
| rtx x; |
| |
| /* If the assert below fails, we should have taken the |
| mode != BLKmode path above, unless we have downward |
| padding of smaller-than-word arguments on a machine |
| with little-endian bytes, which would likely require |
| additional changes to work correctly. */ |
| gcc_checking_assert (BYTES_BIG_ENDIAN |
| && (BLOCK_REG_PADDING (mode, |
| data->arg.type, 1) |
| == PAD_UPWARD)); |
| |
| int by = (UNITS_PER_WORD - size) * BITS_PER_UNIT; |
| |
| x = gen_rtx_REG (word_mode, REGNO (entry_parm)); |
| x = expand_shift (RSHIFT_EXPR, word_mode, x, by, |
| NULL_RTX, 1); |
| x = force_reg (word_mode, x); |
| x = gen_lowpart_SUBREG (GET_MODE (mem), x); |
| |
| emit_move_insn (mem, x); |
| } |
| #endif |
| |
| /* Blocks smaller than a word on a BYTES_BIG_ENDIAN |
| machine must be aligned to the left before storing |
| to memory. Note that the previous test doesn't |
| handle all cases (e.g. SIZE == 3). */ |
| else if (size != UNITS_PER_WORD |
| #ifdef BLOCK_REG_PADDING |
| && (BLOCK_REG_PADDING (mode, data->arg.type, 1) |
| == PAD_DOWNWARD) |
| #else |
| && BYTES_BIG_ENDIAN |
| #endif |
| ) |
| { |
| rtx tem, x; |
| int by = (UNITS_PER_WORD - size) * BITS_PER_UNIT; |
| rtx reg = gen_rtx_REG (word_mode, REGNO (entry_parm)); |
| |
| x = expand_shift (LSHIFT_EXPR, word_mode, reg, by, NULL_RTX, 1); |
| tem = change_address (mem, word_mode, 0); |
| emit_move_insn (tem, x); |
| } |
| else |
| move_block_from_reg (REGNO (entry_parm), mem, |
| size_stored / UNITS_PER_WORD); |
| } |
| else if (!MEM_P (mem)) |
| { |
| gcc_checking_assert (size > UNITS_PER_WORD); |
| #ifdef BLOCK_REG_PADDING |
| gcc_checking_assert (BLOCK_REG_PADDING (GET_MODE (mem), |
| data->arg.type, 0) |
| == PAD_UPWARD); |
| #endif |
| emit_move_insn (mem, entry_parm); |
| } |
| else |
| move_block_from_reg (REGNO (entry_parm), mem, |
| size_stored / UNITS_PER_WORD); |
| } |
| else if (data->stack_parm == 0 && !TYPE_EMPTY_P (data->arg.type)) |
| { |
| push_to_sequence2 (all->first_conversion_insn, all->last_conversion_insn); |
| emit_block_move (stack_parm, data->entry_parm, GEN_INT (size), |
| BLOCK_OP_NORMAL); |
| all->first_conversion_insn = get_insns (); |
| all->last_conversion_insn = get_last_insn (); |
| end_sequence (); |
| in_conversion_seq = true; |
| } |
| |
| if (target_reg) |
| { |
| if (!in_conversion_seq) |
| emit_move_insn (target_reg, stack_parm); |
| else |
| { |
| push_to_sequence2 (all->first_conversion_insn, |
| all->last_conversion_insn); |
| emit_move_insn (target_reg, stack_parm); |
| all->first_conversion_insn = get_insns (); |
| all->last_conversion_insn = get_last_insn (); |
| end_sequence (); |
| } |
| stack_parm = target_reg; |
| } |
| |
| data->stack_parm = stack_parm; |
| set_parm_rtl (parm, stack_parm); |
| } |
| |
| /* A subroutine of assign_parms. Allocate a pseudo to hold the current |
| parameter. Get it there. Perform all ABI specified conversions. */ |
| |
| static void |
| assign_parm_setup_reg (struct assign_parm_data_all *all, tree parm, |
| struct assign_parm_data_one *data) |
| { |
| rtx parmreg, validated_mem; |
| rtx equiv_stack_parm; |
| machine_mode promoted_nominal_mode; |
| int unsignedp = TYPE_UNSIGNED (TREE_TYPE (parm)); |
| bool did_conversion = false; |
| bool need_conversion, moved; |
| enum insn_code icode; |
| rtx rtl; |
| |
| /* Store the parm in a pseudoregister during the function, but we may |
| need to do it in a wider mode. Using 2 here makes the result |
| consistent with promote_decl_mode and thus expand_expr_real_1. */ |
| promoted_nominal_mode |
| = promote_function_mode (data->nominal_type, data->nominal_mode, &unsignedp, |
| TREE_TYPE (current_function_decl), 2); |
| |
| parmreg = gen_reg_rtx (promoted_nominal_mode); |
| if (!DECL_ARTIFICIAL (parm)) |
| mark_user_reg (parmreg); |
| |
| /* If this was an item that we received a pointer to, |
| set rtl appropriately. */ |
| if (data->arg.pass_by_reference) |
| { |
| rtl = gen_rtx_MEM (TYPE_MODE (TREE_TYPE (data->arg.type)), parmreg); |
| set_mem_attributes (rtl, parm, 1); |
| } |
| else |
| rtl = parmreg; |
| |
| assign_parm_remove_parallels (data); |
| |
| /* Copy the value into the register, thus bridging between |
| assign_parm_find_data_types and expand_expr_real_1. */ |
| |
| equiv_stack_parm = data->stack_parm; |
| validated_mem = validize_mem (copy_rtx (data->entry_parm)); |
| |
| need_conversion = (data->nominal_mode != data->passed_mode |
| || promoted_nominal_mode != data->arg.mode); |
| moved = false; |
| |
| if (need_conversion |
| && GET_MODE_CLASS (data->nominal_mode) == MODE_INT |
| && data->nominal_mode == data->passed_mode |
| && data->nominal_mode == GET_MODE (data->entry_parm)) |
| { |
| /* ENTRY_PARM has been converted to PROMOTED_MODE, its |
| mode, by the caller. We now have to convert it to |
| NOMINAL_MODE, if different. However, PARMREG may be in |
| a different mode than NOMINAL_MODE if it is being stored |
| promoted. |
| |
| If ENTRY_PARM is a hard register, it might be in a register |
| not valid for operating in its mode (e.g., an odd-numbered |
| register for a DFmode). In that case, moves are the only |
| thing valid, so we can't do a convert from there. This |
| occurs when the calling sequence allow such misaligned |
| usages. |
| |
| In addition, the conversion may involve a call, which could |
| clobber parameters which haven't been copied to pseudo |
| registers yet. |
| |
| First, we try to emit an insn which performs the necessary |
| conversion. We verify that this insn does not clobber any |
| hard registers. */ |
| |
| rtx op0, op1; |
| |
| icode = can_extend_p (promoted_nominal_mode, data->passed_mode, |
| unsignedp); |
| |
| op0 = parmreg; |
| op1 = validated_mem; |
| if (icode != CODE_FOR_nothing |
| && insn_operand_matches (icode, 0, op0) |
| && insn_operand_matches (icode, 1, op1)) |
| { |
| enum rtx_code code = unsignedp ? ZERO_EXTEND : SIGN_EXTEND; |
| rtx_insn *insn, *insns |