/* Analyze RTL for C-Compiler
Copyright (C) 1987, 1988, 1992, 1993, 1994, 1995, 1996, 1997, 1998,
- 1999, 2000, 2001, 2002, 2003, 2004 Free Software Foundation, Inc.
+ 1999, 2000, 2001, 2002, 2003, 2004, 2005 Free Software Foundation, Inc.
This file is part of GCC.
You should have received a copy of the GNU General Public License
along with GCC; see the file COPYING. If not, write to the Free
-Software Foundation, 59 Temple Place - Suite 330, Boston, MA
-02111-1307, USA. */
+Software Foundation, 51 Franklin Street, Fifth Floor, Boston, MA
+02110-1301, USA. */
#include "config.h"
static unsigned int num_sign_bit_copies1 (rtx, enum machine_mode, rtx,
enum machine_mode, unsigned int);
+/* Offset of the first 'e', 'E' or 'V' operand for each rtx code, or
+ -1 if a code has no such operand. */
+static int non_rtx_starting_operands[NUM_RTX_CODE];
+
/* Bit flags that specify the machine subtype we are compiling for.
Bits are tested using macros TARGET_... defined in the tm.h file
and set by `-m...' switches. Must be defined in rtlanal.c. */
return 0;
}
-/* Return 0 if the use of X as an address in a MEM can cause a trap. */
+/* Return nonzero if the use of X as an address in a MEM can cause a trap.
+ MODE is the mode of the MEM (not that of X) and UNALIGNED_MEMS controls
+ whether nonzero is returned for unaligned memory accesses on strict
+ alignment machines. */
-int
-rtx_addr_can_trap_p (rtx x)
+static int
+rtx_addr_can_trap_p_1 (rtx x, enum machine_mode mode, bool unaligned_mems)
{
enum rtx_code code = GET_CODE (x);
return 1;
case CONST:
- return rtx_addr_can_trap_p (XEXP (x, 0));
+ return rtx_addr_can_trap_p_1 (XEXP (x, 0), mode, unaligned_mems);
case PLUS:
- /* An address is assumed not to trap if it is an address that can't
- trap plus a constant integer or it is the pic register plus a
- constant. */
- return ! ((! rtx_addr_can_trap_p (XEXP (x, 0))
- && GET_CODE (XEXP (x, 1)) == CONST_INT)
- || (XEXP (x, 0) == pic_offset_table_rtx
- && CONSTANT_P (XEXP (x, 1))));
+ /* An address is assumed not to trap if:
+ - it is an address that can't trap plus a constant integer,
+ with the proper remainder modulo the mode size if we are
+ considering unaligned memory references. */
+ if (!rtx_addr_can_trap_p_1 (XEXP (x, 0), mode, unaligned_mems)
+ && GET_CODE (XEXP (x, 1)) == CONST_INT)
+ {
+ HOST_WIDE_INT offset;
+
+ if (!STRICT_ALIGNMENT
+ || !unaligned_mems
+ || GET_MODE_SIZE (mode) == 0)
+ return 0;
+
+ offset = INTVAL (XEXP (x, 1));
+
+#ifdef SPARC_STACK_BOUNDARY_HACK
+ /* ??? The SPARC port may claim a STACK_BOUNDARY higher than
+ the real alignment of %sp. However, when it does this, the
+ alignment of %sp+STACK_POINTER_OFFSET is STACK_BOUNDARY. */
+ if (SPARC_STACK_BOUNDARY_HACK
+ && (XEXP (x, 0) == stack_pointer_rtx
+ || XEXP (x, 0) == hard_frame_pointer_rtx))
+ offset -= STACK_POINTER_OFFSET;
+#endif
+
+ return offset % GET_MODE_SIZE (mode) != 0;
+ }
+
+ /* - or it is the pic register plus a constant. */
+ if (XEXP (x, 0) == pic_offset_table_rtx && CONSTANT_P (XEXP (x, 1)))
+ return 0;
+
+ return 1;
case LO_SUM:
case PRE_MODIFY:
- return rtx_addr_can_trap_p (XEXP (x, 1));
+ return rtx_addr_can_trap_p_1 (XEXP (x, 1), mode, unaligned_mems);
case PRE_DEC:
case PRE_INC:
case POST_DEC:
case POST_INC:
case POST_MODIFY:
- return rtx_addr_can_trap_p (XEXP (x, 0));
+ return rtx_addr_can_trap_p_1 (XEXP (x, 0), mode, unaligned_mems);
default:
break;
return 1;
}
+/* Return nonzero if the use of X as an address in a MEM can cause a trap. */
+
+int
+rtx_addr_can_trap_p (rtx x)
+{
+ return rtx_addr_can_trap_p_1 (x, VOIDmode, false);
+}
+
/* Return true if X is an address that is known to not be zero. */
bool
for (insn = NEXT_INSN (from_insn); insn != to_insn; insn = NEXT_INSN (insn))
if (INSN_P (insn)
&& (reg_overlap_mentioned_p (reg, PATTERN (insn))
- || (CALL_P (insn)
- && (find_reg_fusage (insn, USE, reg)
- || find_reg_fusage (insn, CLOBBER, reg)))))
+ || (CALL_P (insn) && find_reg_fusage (insn, USE, reg))))
return 1;
return 0;
}
}
/* Similar to reg_set_between_p, but check all registers in X. Return 0
- only if none of them are modified between START and END. Do not
- consider non-registers one way or the other. */
-
-int
-regs_set_between_p (rtx x, rtx start, rtx end)
-{
- enum rtx_code code = GET_CODE (x);
- const char *fmt;
- int i, j;
-
- switch (code)
- {
- case CONST_INT:
- case CONST_DOUBLE:
- case CONST_VECTOR:
- case CONST:
- case SYMBOL_REF:
- case LABEL_REF:
- case PC:
- case CC0:
- return 0;
-
- case REG:
- return reg_set_between_p (x, start, end);
-
- default:
- break;
- }
-
- fmt = GET_RTX_FORMAT (code);
- for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
- {
- if (fmt[i] == 'e' && regs_set_between_p (XEXP (x, i), start, end))
- return 1;
-
- else if (fmt[i] == 'E')
- for (j = XVECLEN (x, i) - 1; j >= 0; j--)
- if (regs_set_between_p (XVECEXP (x, i, j), start, end))
- return 1;
- }
-
- return 0;
-}
-
-/* Similar to reg_set_between_p, but check all registers in X. Return 0
only if none of them are modified between START and END. Return 1 if
X contains a MEM; this routine does usememory aliasing. */
return 1;
case MEM:
- if (MEM_READONLY_P (x))
- return 0;
if (modified_between_p (XEXP (x, 0), start, end))
return 1;
+ if (MEM_READONLY_P (x))
+ return 0;
for (insn = NEXT_INSN (start); insn != end; insn = NEXT_INSN (insn))
if (memory_modified_in_insn_p (x, insn))
return 1;
return 1;
case MEM:
- if (MEM_READONLY_P (x))
- return 0;
if (modified_in_p (XEXP (x, 0), insn))
return 1;
+ if (MEM_READONLY_P (x))
+ return 0;
if (memory_modified_in_insn_p (x, insn))
return 1;
return 0;
fmt = GET_RTX_FORMAT (GET_CODE (in));
for (i = GET_RTX_LENGTH (GET_CODE (in)) - 1; i >= 0; i--)
- if (fmt[i] == 'e' && reg_overlap_mentioned_p (x, XEXP (in, i)))
- return 1;
+ if (fmt[i] == 'e')
+ {
+ if (reg_overlap_mentioned_p (x, XEXP (in, i)))
+ return 1;
+ }
+ else if (fmt[i] == 'E')
+ {
+ int j;
+ for (j = XVECLEN (in, i) - 1; j >= 0; --j)
+ if (reg_overlap_mentioned_p (x, XVECEXP (in, i, j)))
+ return 1;
+ }
return 0;
}
return 0;
}
\f
-/* Return nonzero if evaluating rtx X might cause a trap. */
+/* Return nonzero if evaluating rtx X might cause a trap. UNALIGNED_MEMS
+ controls whether nonzero is returned for unaligned memory accesses on
+ strict alignment machines. */
-int
-may_trap_p (rtx x)
+static int
+may_trap_p_1 (rtx x, bool unaligned_mems)
{
int i;
enum rtx_code code;
/* Memory ref can trap unless it's a static var or a stack slot. */
case MEM:
- if (MEM_NOTRAP_P (x))
+ if (MEM_NOTRAP_P (x)
+ && (!STRICT_ALIGNMENT || !unaligned_mems))
return 0;
- return rtx_addr_can_trap_p (XEXP (x, 0));
+ return
+ rtx_addr_can_trap_p_1 (XEXP (x, 0), GET_MODE (x), unaligned_mems);
/* Division by a non-constant might trap. */
case DIV:
case UMOD:
if (HONOR_SNANS (GET_MODE (x)))
return 1;
- if (! CONSTANT_P (XEXP (x, 1))
- || (GET_MODE_CLASS (GET_MODE (x)) == MODE_FLOAT
- && flag_trapping_math))
- return 1;
- if (XEXP (x, 1) == const0_rtx)
+ if (GET_MODE_CLASS (GET_MODE (x)) == MODE_FLOAT)
+ return flag_trapping_math;
+ if (!CONSTANT_P (XEXP (x, 1)) || (XEXP (x, 1) == const0_rtx))
return 1;
break;
case NEG:
case ABS:
+ case SUBREG:
/* These operations don't trap even with floating point. */
break;
{
if (fmt[i] == 'e')
{
- if (may_trap_p (XEXP (x, i)))
+ if (may_trap_p_1 (XEXP (x, i), unaligned_mems))
return 1;
}
else if (fmt[i] == 'E')
{
int j;
for (j = 0; j < XVECLEN (x, i); j++)
- if (may_trap_p (XVECEXP (x, i, j)))
+ if (may_trap_p_1 (XVECEXP (x, i, j), unaligned_mems))
return 1;
}
}
return 0;
}
+
+/* Return nonzero if evaluating rtx X might cause a trap. */
+
+int
+may_trap_p (rtx x)
+{
+ return may_trap_p_1 (x, false);
+}
+
+/* Same as above, but additionally return non-zero if evaluating rtx X might
+ cause a fault. We define a fault for the purpose of this function as a
+ erroneous execution condition that cannot be encountered during the normal
+ execution of a valid program; the typical example is an unaligned memory
+ access on a strict alignment machine. The compiler guarantees that it
+ doesn't generate code that will fault from a valid program, but this
+ guarantee doesn't mean anything for individual instructions. Consider
+ the following example:
+
+ struct S { int d; union { char *cp; int *ip; }; };
+
+ int foo(struct S *s)
+ {
+ if (s->d == 1)
+ return *s->ip;
+ else
+ return *s->cp;
+ }
+
+ on a strict alignment machine. In a valid program, foo will never be
+ invoked on a structure for which d is equal to 1 and the underlying
+ unique field of the union not aligned on a 4-byte boundary, but the
+ expression *s->ip might cause a fault if considered individually.
+
+ At the RTL level, potentially problematic expressions will almost always
+ verify may_trap_p; for example, the above dereference can be emitted as
+ (mem:SI (reg:P)) and this expression is may_trap_p for a generic register.
+ However, suppose that foo is inlined in a caller that causes s->cp to
+ point to a local character variable and guarantees that s->d is not set
+ to 1; foo may have been effectively translated into pseudo-RTL as:
+
+ if ((reg:SI) == 1)
+ (set (reg:SI) (mem:SI (%fp - 7)))
+ else
+ (set (reg:QI) (mem:QI (%fp - 7)))
+
+ Now (mem:SI (%fp - 7)) is considered as not may_trap_p since it is a
+ memory reference to a stack slot, but it will certainly cause a fault
+ on a strict alignment machine. */
+
+int
+may_trap_or_fault_p (rtx x)
+{
+ return may_trap_p_1 (x, true);
+}
\f
/* Return nonzero if X contains a comparison that is not either EQ or NE,
i.e., an inequality. */
return 0;
}
+/* Optimized loop of for_each_rtx, trying to avoid useless recursive
+ calls. Processes the subexpressions of EXP and passes them to F. */
+static int
+for_each_rtx_1 (rtx exp, int n, rtx_function f, void *data)
+{
+ int result, i, j;
+ const char *format = GET_RTX_FORMAT (GET_CODE (exp));
+ rtx *x;
+
+ for (; format[n] != '\0'; n++)
+ {
+ switch (format[n])
+ {
+ case 'e':
+ /* Call F on X. */
+ x = &XEXP (exp, n);
+ result = (*f) (x, data);
+ if (result == -1)
+ /* Do not traverse sub-expressions. */
+ continue;
+ else if (result != 0)
+ /* Stop the traversal. */
+ return result;
+
+ if (*x == NULL_RTX)
+ /* There are no sub-expressions. */
+ continue;
+
+ i = non_rtx_starting_operands[GET_CODE (*x)];
+ if (i >= 0)
+ {
+ result = for_each_rtx_1 (*x, i, f, data);
+ if (result != 0)
+ return result;
+ }
+ break;
+
+ case 'V':
+ case 'E':
+ if (XVEC (exp, n) == 0)
+ continue;
+ for (j = 0; j < XVECLEN (exp, n); ++j)
+ {
+ /* Call F on X. */
+ x = &XVECEXP (exp, n, j);
+ result = (*f) (x, data);
+ if (result == -1)
+ /* Do not traverse sub-expressions. */
+ continue;
+ else if (result != 0)
+ /* Stop the traversal. */
+ return result;
+
+ if (*x == NULL_RTX)
+ /* There are no sub-expressions. */
+ continue;
+
+ i = non_rtx_starting_operands[GET_CODE (*x)];
+ if (i >= 0)
+ {
+ result = for_each_rtx_1 (*x, i, f, data);
+ if (result != 0)
+ return result;
+ }
+ }
+ break;
+
+ default:
+ /* Nothing to do. */
+ break;
+ }
+ }
+
+ return 0;
+}
+
/* Traverse X via depth-first search, calling F for each
sub-expression (including X itself). F is also passed the DATA.
If F returns -1, do not traverse sub-expressions, but continue
for_each_rtx (rtx *x, rtx_function f, void *data)
{
int result;
- int length;
- const char *format;
int i;
/* Call F on X. */
/* There are no sub-expressions. */
return 0;
- length = GET_RTX_LENGTH (GET_CODE (*x));
- format = GET_RTX_FORMAT (GET_CODE (*x));
-
- for (i = 0; i < length; ++i)
- {
- switch (format[i])
- {
- case 'e':
- result = for_each_rtx (&XEXP (*x, i), f, data);
- if (result != 0)
- return result;
- break;
-
- case 'V':
- case 'E':
- if (XVEC (*x, i) != 0)
- {
- int j;
- for (j = 0; j < XVECLEN (*x, i); ++j)
- {
- result = for_each_rtx (&XVECEXP (*x, i, j), f, data);
- if (result != 0)
- return result;
- }
- }
- break;
-
- default:
- /* Nothing to do. */
- break;
- }
-
- }
+ i = non_rtx_starting_operands[GET_CODE (*x)];
+ if (i < 0)
+ return 0;
- return 0;
+ return for_each_rtx_1 (*x, i, f, data);
}
+
/* Searches X for any reference to REGNO, returning the rtx of the
reference found if any. Otherwise, returns NULL_RTX. */
subreg_regno_offset (unsigned int xregno, enum machine_mode xmode,
unsigned int offset, enum machine_mode ymode)
{
- int nregs_xmode, nregs_ymode;
+ int nregs_xmode, nregs_ymode, nregs_xmode_unit_int;
int mode_multiple, nregs_multiple;
int y_offset;
+ enum machine_mode xmode_unit, xmode_unit_int;
gcc_assert (xregno < FIRST_PSEUDO_REGISTER);
- nregs_xmode = hard_regno_nregs[xregno][xmode];
+ if (GET_MODE_INNER (xmode) == VOIDmode)
+ xmode_unit = xmode;
+ else
+ xmode_unit = GET_MODE_INNER (xmode);
+
+ if (FLOAT_MODE_P (xmode_unit))
+ {
+ xmode_unit_int = int_mode_for_mode (xmode_unit);
+ if (xmode_unit_int == BLKmode)
+ /* It's probably bad to be here; a port should have an integer mode
+ that's the same size as anything of which it takes a SUBREG. */
+ xmode_unit_int = xmode_unit;
+ }
+ else
+ xmode_unit_int = xmode_unit;
+
+ nregs_xmode_unit_int = hard_regno_nregs[xregno][xmode_unit_int];
+
+ /* Adjust nregs_xmode to allow for 'holes'. */
+ if (nregs_xmode_unit_int != hard_regno_nregs[xregno][xmode_unit])
+ nregs_xmode = nregs_xmode_unit_int * GET_MODE_NUNITS (xmode);
+ else
+ nregs_xmode = hard_regno_nregs[xregno][xmode];
+
nregs_ymode = hard_regno_nregs[xregno][ymode];
/* If this is a big endian paradoxical subreg, which uses more actual
if (offset == 0 || nregs_xmode == nregs_ymode)
return 0;
- /* size of ymode must not be greater than the size of xmode. */
+ /* Size of ymode must not be greater than the size of xmode. */
mode_multiple = GET_MODE_SIZE (xmode) / GET_MODE_SIZE (ymode);
gcc_assert (mode_multiple != 0);
xmode - The mode of xregno.
offset - The byte offset.
ymode - The mode of a top level SUBREG (or what may become one).
- RETURN - The regno offset which would be used. */
+ RETURN - Whether the offset is representable. */
bool
subreg_offset_representable_p (unsigned int xregno, enum machine_mode xmode,
unsigned int offset, enum machine_mode ymode)
{
- int nregs_xmode, nregs_ymode;
+ int nregs_xmode, nregs_ymode, nregs_xmode_unit, nregs_xmode_unit_int;
int mode_multiple, nregs_multiple;
int y_offset;
+ enum machine_mode xmode_unit, xmode_unit_int;
gcc_assert (xregno < FIRST_PSEUDO_REGISTER);
- nregs_xmode = hard_regno_nregs[xregno][xmode];
+ if (GET_MODE_INNER (xmode) == VOIDmode)
+ xmode_unit = xmode;
+ else
+ xmode_unit = GET_MODE_INNER (xmode);
+
+ if (FLOAT_MODE_P (xmode_unit))
+ {
+ xmode_unit_int = int_mode_for_mode (xmode_unit);
+ if (xmode_unit_int == BLKmode)
+ /* It's probably bad to be here; a port should have an integer mode
+ that's the same size as anything of which it takes a SUBREG. */
+ xmode_unit_int = xmode_unit;
+ }
+ else
+ xmode_unit_int = xmode_unit;
+
+ nregs_xmode_unit = hard_regno_nregs[xregno][xmode_unit];
+ nregs_xmode_unit_int = hard_regno_nregs[xregno][xmode_unit_int];
+
+ /* If there are holes in a non-scalar mode in registers, we expect
+ that it is made up of its units concatenated together. */
+ if (nregs_xmode_unit != nregs_xmode_unit_int)
+ {
+ gcc_assert (nregs_xmode_unit * GET_MODE_NUNITS (xmode)
+ == hard_regno_nregs[xregno][xmode]);
+
+ /* You can only ask for a SUBREG of a value with holes in the middle
+ if you don't cross the holes. (Such a SUBREG should be done by
+ picking a different register class, or doing it in memory if
+ necessary.) An example of a value with holes is XCmode on 32-bit
+ x86 with -m128bit-long-double; it's represented in 6 32-bit registers,
+ 3 for each part, but in memory it's two 128-bit parts.
+ Padding is assumed to be at the end (not necessarily the 'high part')
+ of each unit. */
+ if (nregs_xmode_unit != nregs_xmode_unit_int
+ && (offset / GET_MODE_SIZE (xmode_unit_int) + 1
+ < GET_MODE_NUNITS (xmode))
+ && (offset / GET_MODE_SIZE (xmode_unit_int)
+ != ((offset + GET_MODE_SIZE (ymode) - 1)
+ / GET_MODE_SIZE (xmode_unit_int))))
+ return false;
+
+ nregs_xmode = nregs_xmode_unit_int * GET_MODE_NUNITS (xmode);
+ }
+ else
+ nregs_xmode = hard_regno_nregs[xregno][xmode];
+
nregs_ymode = hard_regno_nregs[xregno][ymode];
- /* Paradoxical subregs are always valid. */
+ /* Paradoxical subregs are otherwise valid. */
if (offset == 0
&& nregs_ymode > nregs_xmode
&& (GET_MODE_SIZE (ymode) > UNITS_PER_WORD
? WORDS_BIG_ENDIAN : BYTES_BIG_ENDIAN))
return true;
- /* Lowpart subregs are always valid. */
+ /* Lowpart subregs are otherwise valid. */
if (offset == subreg_lowpart_offset (ymode, xmode))
return true;
- /* This should always pass, otherwise we don't know how to verify the
- constraint. These conditions may be relaxed but subreg_offset would
- need to be redesigned. */
+ /* This should always pass, otherwise we don't know how to verify
+ the constraint. These conditions may be relaxed but
+ subreg_regno_offset would need to be redesigned. */
gcc_assert ((GET_MODE_SIZE (xmode) % GET_MODE_SIZE (ymode)) == 0);
- gcc_assert ((GET_MODE_SIZE (ymode) % nregs_ymode) == 0);
gcc_assert ((nregs_xmode % nregs_ymode) == 0);
/* The XMODE value can be seen as a vector of NREGS_XMODE
/ nregs_xmode,
MODE_INT, 0));
- /* size of ymode must not be greater than the size of xmode. */
+ /* Size of ymode must not be greater than the size of xmode. */
mode_multiple = GET_MODE_SIZE (xmode) / GET_MODE_SIZE (ymode);
gcc_assert (mode_multiple != 0);
}
/* Look backward for first parameter to be loaded.
+ Note that loads of all parameters will not necessarily be
+ found if CSE has eliminated some of them (e.g., an argument
+ to the outer function is passed down as a parameter).
Do not skip BOUNDARY. */
rtx
find_first_parameter_load (rtx call_insn, rtx boundary)
{
struct parms_set_data parm;
- rtx p, before;
+ rtx p, before, first_set;
/* Since different machines initialize their parameter registers
in different orders, assume nothing. Collect the set of all
parm.nregs++;
}
before = call_insn;
+ first_set = call_insn;
/* Search backward for the first set of a register in this set. */
while (parm.nregs && before != boundary)
}
if (INSN_P (before))
- note_stores (PATTERN (before), parms_set, &parm);
+ {
+ int nregs_old = parm.nregs;
+ note_stores (PATTERN (before), parms_set, &parm);
+ /* If we found something that did not set a parameter reg,
+ we're done. Do not keep going, as that might result
+ in hoisting an insn before the setting of a pseudo
+ that is used by the hoisted insn. */
+ if (nregs_old != parm.nregs)
+ first_set = before;
+ else
+ break;
+ }
}
- return before;
+ return first_set;
}
/* Return true if we should avoid inserting code between INSN and preceding
return 0;
case SUBREG:
+ total = 0;
/* If we can't tie these modes, make this expensive. The larger
the mode, the more expensive it is. */
if (! MODES_TIEABLE_P (GET_MODE (x), GET_MODE (SUBREG_REG (x))))
case GE: case GEU: case UNGE:
case LE: case LEU: case UNLE:
case UNORDERED: case ORDERED:
-
/* If this produces an integer result, we know which bits are set.
Code here used to clear bits outside the mode of X, but that is
now done above. */
-
- if (GET_MODE_CLASS (mode) == MODE_INT
+ /* Mind that MODE is the mode the caller wants to look at this
+ operation in, and not the actual operation mode. We can wind
+ up with (subreg:DI (gt:V4HI x y)), and we don't have anything
+ that describes the results of a vector compare. */
+ if (GET_MODE_CLASS (GET_MODE (x)) == MODE_INT
&& mode_width <= HOST_BITS_PER_WIDE_INT)
nonzero = STORE_FLAG_VALUE;
break;
allow_cc_mode, valid_at_insn_p);
}
+\f
+/* Initialize non_rtx_starting_operands, which is used to speed up
+ for_each_rtx. */
+void
+init_rtlanal (void)
+{
+ int i;
+ for (i = 0; i < NUM_RTX_CODE; i++)
+ {
+ const char *format = GET_RTX_FORMAT (i);
+ const char *first = strpbrk (format, "eEV");
+ non_rtx_starting_operands[i] = first ? first - format : -1;
+ }
+}