/* Output routines for GCC for Renesas / SuperH SH.
Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001, 2002,
- 2003, 2004, 2005, 2006, 2007, 2008, 2009 Free Software Foundation, Inc.
+ 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010
+ Free Software Foundation, Inc.
Contributed by Steve Chamberlain (sac@cygnus.com).
Improved by Jim Wilson (wilson@cygnus.com).
#include "flags.h"
#include "expr.h"
#include "optabs.h"
+#include "reload.h"
#include "function.h"
#include "regs.h"
#include "hard-reg-set.h"
and returned from sh_reorder2. */
static short cached_can_issue_more;
+/* Unique number for UNSPEC_BBR pattern. */
+static unsigned int unspec_bbr_uid = 1;
+
/* Provides the class number of the smallest class containing
reg number. */
static void push_regs (HARD_REG_SET *, int);
static int calc_live_regs (HARD_REG_SET *);
static HOST_WIDE_INT rounded_frame_size (int);
+static bool sh_frame_pointer_required (void);
static rtx mark_constant_pool_use (rtx);
static tree sh_handle_interrupt_handler_attribute (tree *, tree, tree, int, bool *);
static tree sh_handle_resbank_handler_attribute (tree *, tree,
struct save_schedule_s *, int);
static rtx sh_struct_value_rtx (tree, int);
+static rtx sh_function_value (const_tree, const_tree, bool);
+static rtx sh_libcall_value (enum machine_mode, const_rtx);
static bool sh_return_in_memory (const_tree, const_tree);
static rtx sh_builtin_saveregs (void);
static void sh_setup_incoming_varargs (CUMULATIVE_ARGS *, enum machine_mode, tree, int *, int);
static tree sh_build_builtin_va_list (void);
static void sh_va_start (tree, rtx);
static tree sh_gimplify_va_arg_expr (tree, tree, gimple_seq *, gimple_seq *);
+static bool sh_promote_prototypes (const_tree);
static enum machine_mode sh_promote_function_mode (const_tree type,
enum machine_mode,
int *punsignedp,
#undef TARGET_PROMOTE_FUNCTION_MODE
#define TARGET_PROMOTE_FUNCTION_MODE sh_promote_function_mode
+#undef TARGET_FUNCTION_VALUE
+#define TARGET_FUNCTION_VALUE sh_function_value
+#undef TARGET_LIBCALL_VALUE
+#define TARGET_LIBCALL_VALUE sh_libcall_value
#undef TARGET_STRUCT_VALUE_RTX
#define TARGET_STRUCT_VALUE_RTX sh_struct_value_rtx
#undef TARGET_RETURN_IN_MEMORY
#undef TARGET_DWARF_CALLING_CONVENTION
#define TARGET_DWARF_CALLING_CONVENTION sh_dwarf_calling_convention
+#undef TARGET_FRAME_POINTER_REQUIRED
+#define TARGET_FRAME_POINTER_REQUIRED sh_frame_pointer_required
+
/* Return regmode weight for insn. */
#define INSN_REGMODE_WEIGHT(INSN, MODE) regmode_weight[((MODE) == SImode) ? 0 : 1][INSN_UID (INSN)]
{
if (level)
{
- flag_omit_frame_pointer = 2;
if (!size)
sh_div_str = "inv:minlat";
}
if (! VALID_REGISTER_P (ADDREGNAMES_REGNO (regno)))
sh_additional_register_names[regno][0] = '\0';
- if (flag_omit_frame_pointer == 2)
- {
- /* The debugging information is sufficient,
- but gdb doesn't implement this yet */
- if (0)
- flag_omit_frame_pointer
- = (PREFERRED_DEBUGGING_TYPE == DWARF2_DEBUG);
- else
- flag_omit_frame_pointer = 0;
- }
+ flag_omit_frame_pointer = (PREFERRED_DEBUGGING_TYPE == DWARF2_DEBUG);
if ((flag_pic && ! TARGET_PREFERGOT)
|| (TARGET_SHMEDIA && !TARGET_PT_FIXED))
flag_no_function_cse = 1;
- if (SMALL_REGISTER_CLASSES)
+ if (targetm.small_register_classes_for_mode_p (VOIDmode)) \
{
/* Never run scheduling before reload, since that can
break global alloc, and generates slower code anyway due
flag_schedule_insns = 0;
}
+ if ((target_flags_explicit & MASK_ACCUMULATE_OUTGOING_ARGS) == 0)
+ target_flags |= MASK_ACCUMULATE_OUTGOING_ARGS;
+
+ /* Unwind info is not correct around the CFG unless either a frame
+ pointer is present or M_A_O_A is set. Fixing this requires rewriting
+ unwind info generation to be aware of the CFG and propagating states
+ around edges. */
+ if ((flag_unwind_tables || flag_asynchronous_unwind_tables
+ || flag_exceptions || flag_non_call_exceptions)
+ && flag_omit_frame_pointer
+ && !(target_flags & MASK_ACCUMULATE_OUTGOING_ARGS))
+ {
+ if (target_flags_explicit & MASK_ACCUMULATE_OUTGOING_ARGS)
+ warning (0, "unwind tables currently require either a frame pointer "
+ "or -maccumulate-outgoing-args for correctness");
+ target_flags |= MASK_ACCUMULATE_OUTGOING_ARGS;
+ }
+
/* Unwinding with -freorder-blocks-and-partition does not work on this
architecture, because it requires far jumps to label crossing between
hot/cold sections which are rejected on this architecture. */
int si_limit;
int hi_limit;
rtx orig = from;
+ rtx last_got = NULL_RTX;
+ rtx last_symoff = NULL_RTX;
/* For HImode: range is 510, add 4 because pc counts from address of
second instruction after this one, subtract 2 for the jump instruction
dst = SET_DEST (pat);
mode = GET_MODE (dst);
+ /* GOT pcrelat setting comes in pair of
+ mova .L8,r0
+ mov.l .L8,r12
+ instructions. (plus add r0,r12).
+ Remember if we see one without the other. */
+ if (GET_CODE (src) == UNSPEC && PIC_ADDR_P (XVECEXP (src, 0, 0)))
+ last_got = last_got ? NULL_RTX : from;
+ else if (PIC_ADDR_P (src))
+ last_got = last_got ? NULL_RTX : from;
+
/* We must explicitly check the mode, because sometimes the
front end will generate code to load unsigned constants into
HImode targets without properly sign extending them. */
{
switch (untangle_mova (&num_mova, &mova, from))
{
+ case 1:
+ if (flag_pic)
+ {
+ rtx src = SET_SRC (PATTERN (from));
+ if (GET_CODE (src) == CONST
+ && GET_CODE (XEXP (src, 0)) == UNSPEC
+ && XINT (XEXP (src, 0), 1) == UNSPEC_SYMOFF)
+ last_symoff = from;
+ }
+ break;
case 0: return find_barrier (0, 0, mova);
case 2:
{
&& ! TARGET_SMALLCODE)
new_align = 4;
+ /* There is a possibility that a bf is transformed into a bf/s by the
+ delay slot scheduler. */
+ if (JUMP_P (from) && !JUMP_TABLE_DATA_P (from)
+ && get_attr_type (from) == TYPE_CBRANCH
+ && GET_CODE (PATTERN (NEXT_INSN (PREV_INSN (from)))) != SEQUENCE)
+ inc += 2;
+
if (found_si)
{
count_si += inc;
so we'll make one. */
rtx label = gen_label_rtx ();
+ /* Don't emit a constant table in the middle of insns for
+ casesi_worker_2. This is a bit overkill but is enough
+ because casesi_worker_2 wouldn't appear so frequently. */
+ if (last_symoff)
+ from = last_symoff;
+
/* If we exceeded the range, then we must back up over the last
instruction we looked at. Otherwise, we just need to undo the
NEXT_INSN at the end of the loop. */
else
from = PREV_INSN (from);
+ /* Don't emit a constant table int the middle of global pointer setting,
+ since that that would move the addressing base GOT into another table.
+ We need the first mov instruction before the _GLOBAL_OFFSET_TABLE_
+ in the pool anyway, so just move up the whole constant pool. */
+ if (last_got)
+ from = PREV_INSN (last_got);
+
+ /* Don't insert the constant pool table at the position which
+ may be the landing pad. */
+ if (flag_exceptions
+ && CALL_P (from)
+ && find_reg_note (from, REG_EH_REGION, NULL_RTX))
+ from = PREV_INSN (from);
+
/* Walk back to be just before any jump or label.
Putting it before a label reduces the number of times the branch
around the constant pool table will be hit. Putting it before
branch; simplejump_p fails for indirect jumps even if they have
a JUMP_LABEL. */
rtx insn = emit_insn_before (gen_indirect_jump_scratch
- (reg, GEN_INT (INSN_UID (JUMP_LABEL (jump))))
- , jump);
+ (reg, GEN_INT (unspec_bbr_uid++)),
+ jump);
/* ??? We would like this to have the scope of the jump, but that
scope will change when a delay slot insn of an inner scope is added.
Hence, after delay slot scheduling, we'll have to expect
/* We can't use JUMP_LABEL here because it might be undefined
when not optimizing. */
return emit_insn_before (gen_block_branch_redirect
- (GEN_INT (INSN_UID (XEXP (SET_SRC (PATTERN (jump)), 0))))
- , jump);
+ (GEN_INT (unspec_bbr_uid++)),
+ jump);
return prev;
}
if (bp->far_label)
(emit_insn_after
(gen_stuff_delay_slot
- (GEN_INT (INSN_UID (XEXP (SET_SRC (PATTERN (jump)), 0))),
+ (GEN_INT (unspec_bbr_uid++),
GEN_INT (recog_memoized (insn) == CODE_FOR_branch_false)),
insn));
/* Prevent reorg from undoing our splits. */
HOST_WIDE_INT size = get_frame_size ();
HOST_WIDE_INT align = STACK_BOUNDARY / BITS_PER_UNIT;
+ if (ACCUMULATE_OUTGOING_ARGS)
+ size += crtl->outgoing_args_size;
+
return ((size + pushed + align - 1) & -align) - pushed;
}
pop (PR_REG);
}
- /* Banked registers are poped first to avoid being scheduled in the
+ /* Banked registers are popped first to avoid being scheduled in the
delay slot. RTE switches banks before the ds instruction. */
if (current_function_interrupt)
{
- for (i = FIRST_BANKED_REG; i <= LAST_BANKED_REG; i++)
- if (TEST_HARD_REG_BIT (live_regs_mask, i))
- pop (LAST_BANKED_REG - i);
+ for (i = LAST_BANKED_REG; i >= FIRST_BANKED_REG; i--)
+ if (TEST_HARD_REG_BIT (live_regs_mask, i))
+ pop (i);
last_reg = FIRST_PSEUDO_REGISTER - LAST_BANKED_REG - 1;
}
pr_offset = rounded_frame_size (d);
emit_insn (GEN_MOV (tmp, GEN_INT (pr_offset)));
- emit_insn (GEN_ADD3 (tmp, tmp, hard_frame_pointer_rtx));
+
+ if (frame_pointer_needed)
+ emit_insn (GEN_ADD3 (tmp, tmp, hard_frame_pointer_rtx));
+ else
+ emit_insn (GEN_ADD3 (tmp, tmp, stack_pointer_rtx));
tmp = gen_frame_mem (Pmode, tmp);
emit_insn (GEN_MOV (tmp, ra));
return mode;
}
-bool
+static bool
sh_promote_prototypes (const_tree type)
{
if (TARGET_HITACHI)
return gen_rtx_REG (Pmode, 2);
}
+/* Worker function for TARGET_FUNCTION_VALUE.
+
+ For the SH, this is like LIBCALL_VALUE, except that we must change the
+ mode like PROMOTE_MODE does.
+ ??? PROMOTE_MODE is ignored for non-scalar types. The set of types
+ tested here has to be kept in sync with the one in explow.c:promote_mode.
+*/
+
+static rtx
+sh_function_value (const_tree valtype,
+ const_tree fn_decl_or_type,
+ bool outgoing ATTRIBUTE_UNUSED)
+{
+ if (fn_decl_or_type
+ && !DECL_P (fn_decl_or_type))
+ fn_decl_or_type = NULL;
+
+ return gen_rtx_REG (
+ ((GET_MODE_CLASS (TYPE_MODE (valtype)) == MODE_INT
+ && GET_MODE_SIZE (TYPE_MODE (valtype)) < 4
+ && (TREE_CODE (valtype) == INTEGER_TYPE
+ || TREE_CODE (valtype) == ENUMERAL_TYPE
+ || TREE_CODE (valtype) == BOOLEAN_TYPE
+ || TREE_CODE (valtype) == REAL_TYPE
+ || TREE_CODE (valtype) == OFFSET_TYPE))
+ && sh_promote_prototypes (fn_decl_or_type)
+ ? (TARGET_SHMEDIA64 ? DImode : SImode) : TYPE_MODE (valtype)),
+ BASE_RETURN_VALUE_REG (TYPE_MODE (valtype)));
+}
+
+/* Worker function for TARGET_LIBCALL_VALUE. */
+
+static rtx
+sh_libcall_value (enum machine_mode mode, const_rtx fun ATTRIBUTE_UNUSED)
+{
+ return gen_rtx_REG (mode, BASE_RETURN_VALUE_REG (mode));
+}
+
+/* Worker function for FUNCTION_VALUE_REGNO_P. */
+
+bool
+sh_function_value_regno_p (const unsigned int regno)
+{
+ return ((regno) == FIRST_RET_REG
+ || (TARGET_SH2E && (regno) == FIRST_FP_RET_REG)
+ || (TARGET_SHMEDIA_FPU && (regno) == FIRST_FP_RET_REG));
+}
+
/* Worker function for TARGET_RETURN_IN_MEMORY. */
static bool
if (! TEST_HARD_REG_BIT (regs_live, 1))
return gen_rtx_REG (Pmode, 1);
- /* Hard reg 1 is live; since this is a SMALL_REGISTER_CLASSES target,
+ /* Hard reg 1 is live; since this is a small register classes target,
there shouldn't be anything but a jump before the function end. */
gcc_assert (!TEST_HARD_REG_BIT (regs_live, 7));
return gen_rtx_REG (Pmode, 7);
&& GET_CODE (PATTERN (insn)) != USE
&& GET_CODE (PATTERN (insn)) != CLOBBER)
|| CALL_P (insn)
- || (JUMP_P (insn)
- && GET_CODE (PATTERN (insn)) != ADDR_DIFF_VEC
- && GET_CODE (PATTERN (insn)) != ADDR_VEC))
+ || (JUMP_P (insn) && !JUMP_TABLE_DATA_P (insn)))
&& GET_CODE (PATTERN (NEXT_INSN (PREV_INSN (insn)))) != SEQUENCE
&& get_attr_needs_delay_slot (insn) == NEEDS_DELAY_SLOT_YES)
return 2;
/* SH2e has a bug that prevents the use of annulled branches, so if
the delay slot is not filled, we'll have to put a NOP in it. */
if (sh_cpu_attr == CPU_SH2E
- && JUMP_P (insn)
- && GET_CODE (PATTERN (insn)) != ADDR_DIFF_VEC
- && GET_CODE (PATTERN (insn)) != ADDR_VEC
+ && JUMP_P (insn) && !JUMP_TABLE_DATA_P (insn)
&& get_attr_type (insn) == TYPE_CBRANCH
&& GET_CODE (PATTERN (NEXT_INSN (PREV_INSN (insn)))) != SEQUENCE)
return 2;
|| XINT (x, 1) == UNSPEC_GOTPLT
|| XINT (x, 1) == UNSPEC_GOTTPOFF
|| XINT (x, 1) == UNSPEC_DTPOFF
+ || XINT (x, 1) == UNSPEC_TPOFF
|| XINT (x, 1) == UNSPEC_PLT
|| XINT (x, 1) == UNSPEC_SYMOFF
|| XINT (x, 1) == UNSPEC_PCREL_SYMOFF))
return x;
}
+/* Attempt to replace *P, which is an address that needs reloading, with
+ a valid memory address for an operand of mode MODE.
+ Like for sh_legitimize_address, for the SH we try to get a normal form
+ of the address. That will allow inheritance of the address reloads. */
+
+bool
+sh_legitimize_reload_address (rtx *p, enum machine_mode mode, int opnum,
+ int itype)
+{
+ enum reload_type type = (enum reload_type) itype;
+
+ if (GET_CODE (*p) == PLUS
+ && (GET_MODE_SIZE (mode) == 4 || GET_MODE_SIZE (mode) == 8)
+ && CONST_INT_P (XEXP (*p, 1))
+ && MAYBE_BASE_REGISTER_RTX_P (XEXP (*p, 0), true)
+ && ! TARGET_SHMEDIA
+ && ! (TARGET_SH4 && mode == DFmode)
+ && ! (mode == PSImode && type == RELOAD_FOR_INPUT_ADDRESS)
+ && (ALLOW_INDEXED_ADDRESS
+ || XEXP (*p, 0) == stack_pointer_rtx
+ || XEXP (*p, 0) == hard_frame_pointer_rtx))
+ {
+ rtx index_rtx = XEXP (*p, 1);
+ HOST_WIDE_INT offset = INTVAL (index_rtx), offset_base;
+ rtx sum;
+
+ if (TARGET_SH2A && mode == DFmode && (offset & 0x7))
+ {
+ push_reload (*p, NULL_RTX, p, NULL,
+ BASE_REG_CLASS, Pmode, VOIDmode, 0, 0, opnum, type);
+ goto win;
+ }
+ if (TARGET_SH2E && mode == SFmode)
+ {
+ *p = copy_rtx (*p);
+ push_reload (*p, NULL_RTX, p, NULL,
+ BASE_REG_CLASS, Pmode, VOIDmode, 0, 0, opnum, type);
+ goto win;
+ }
+ /* Instead of offset_base 128..131 use 124..127, so that
+ simple add suffices. */
+ if (offset > 127)
+ offset_base = ((offset + 4) & ~60) - 4;
+ else
+ offset_base = offset & ~60;
+ /* Sometimes the normal form does not suit DImode. We could avoid
+ that by using smaller ranges, but that would give less optimized
+ code when SImode is prevalent. */
+ if (GET_MODE_SIZE (mode) + offset - offset_base <= 64)
+ {
+ sum = gen_rtx_PLUS (Pmode, XEXP (*p, 0), GEN_INT (offset_base));
+ *p = gen_rtx_PLUS (Pmode, sum, GEN_INT (offset - offset_base));
+ push_reload (sum, NULL_RTX, &XEXP (*p, 0), NULL,
+ BASE_REG_CLASS, Pmode, VOIDmode, 0, 0, opnum, type);
+ goto win;
+ }
+ }
+ /* We must re-recognize what we created before. */
+ else if (GET_CODE (*p) == PLUS
+ && (GET_MODE_SIZE (mode) == 4 || GET_MODE_SIZE (mode) == 8)
+ && GET_CODE (XEXP (*p, 0)) == PLUS
+ && CONST_INT_P (XEXP (XEXP (*p, 0), 1))
+ && MAYBE_BASE_REGISTER_RTX_P (XEXP (XEXP (*p, 0), 0), true)
+ && CONST_INT_P (XEXP (*p, 1))
+ && ! TARGET_SHMEDIA
+ && ! (TARGET_SH2E && mode == SFmode))
+ {
+ /* Because this address is so complex, we know it must have
+ been created by LEGITIMIZE_RELOAD_ADDRESS before; thus,
+ it is already unshared, and needs no further unsharing. */
+ push_reload (XEXP (*p, 0), NULL_RTX, &XEXP (*p, 0), NULL,
+ BASE_REG_CLASS, Pmode, VOIDmode, 0, 0, opnum, type);
+ goto win;
+ }
+
+ return false;
+
+ win:
+ return true;
+}
+
/* Mark the use of a constant in the literal table. If the constant
has multiple labels, make it unique. */
static rtx
return false;
}
+bool
+sh_frame_pointer_required (void)
+{
+/* If needed override this in other tm.h files to cope with various OS
+ lossage requiring a frame pointer. */
+ if (SUBTARGET_FRAME_POINTER_REQUIRED)
+ return true;
+
+ if (crtl->profile)
+ return true;
+
+ return false;
+}
+
/* Implements target hook dwarf_calling_convention. Return an enum
of dwarf_calling_convention. */
int
return 0;
}
+/* Return true if registers in machine mode MODE will likely be
+ allocated to registers in small register classes. */
+
+static bool
+sh_small_register_classes_for_mode_p (enum machine_mode mode ATTRIBUTE_UNUSED)
+{
+ return (! TARGET_SHMEDIA);
+}
/* If ADDRESS refers to a CODE_LABEL, add NUSES to the number of times
that label is used. */