+2002-02-21 Eric Christopher <echristo@redhat.com>
+
+ * expmed.c (store_bit_field): Try to simplify the subreg
+ before generating a new one when when the mode size of
+ value is less than maxmode.
+
2002-02-21 Richard Henderson <rth@redhat.com>
* emit-rtl.c (offset_address): Use simplify_gen_binary rather
bits of SImode const_int.
(includes_rshift_p): Likewise.
(print_operand): Call mask_operand and mask64_operand with correct
- mode.
+ mode.
(rs6000_output_function_epilogue): Pad traceback table to word.
* config/rs6000/rs6000.h (MASK_64BIT): Correct comment.
(EXTRA_CONSTRAINT, 'S' and 'T'): Call mask_operand and
(ix86_expand_unop1_builtin): Likewise.
2002-02-20 Philip Blundell <philb@gnu.org>
-
+
PR 5705
* config/arm/arm.h (HARD_REGNO_RENAME_OK): New macro.
* cppinit.c (init_dependency_output): Take deps output file
from -o if none given with -MF. Suppress normal output.
- * gcc.c (cpp_unique_options): Have -M and -MM imply -E.
+ * gcc.c (cpp_unique_options): Have -M and -MM imply -E.
* doc/cpp.texi, doc/invoke.texi: Update.
2002-02-19 Zack Weinberg <zack@codesourcery.com>
really_call_used[VRSAVE_REGNO] if not Altivec.
2002-02-19 Alan Modra <amodra@bigpond.net.au>
-
+
* config/rs6000/rs6000.c (u_short_cint_operand): Mask op with
MODE_MASK.
(constant_pool_expr_1): Fix formatting.
(arm_assemble_integer): Likewise.
* config/arm/arm.h (ARM_ENCODE_CALL_TYPE): Allow any decl to be
marked local.
-
+
2002-02-19 matthew green <mrg@eterna.com.au>
* config.gcc (sparc-*-netbsdelf*): Enable target.
(decode_rtx_const): Add case for vector.
* config/rs6000/rs6000-protos.h: Add zero_constant.
-
+
* config/rs6000/rs6000.c (rs6000_emit_move): Handle vector
constants. Force easy vector constants into memory.
(easy_vector_constant): New.
2002-02-18 David Billinghurst <David.Billinghurst@riotinto.com>
* objc/objc-act.c (handle_impent): Remove leading '*'
- from objc_class_name.
+ from objc_class_name.
2002-02-17 Richard Henderson <rth@redhat.com>
* config/arc/arc.c config/fr30/fr30.c config/m32r/m32r.c
config/m88k/m88k.c : Adapt all calls to output_operand_lossage.
Update copyright date where necessary.
-
+
* config/i386/i386.c (print_operand): Likewise. Remove use of
sprintf.
2002-02-16 Ulrich Weigand <uweigand@de.ibm.com>
- * config/s390/s390.c (pool_stop_uid, other_chunk, far_away,
+ * config/s390/s390.c (pool_stop_uid, other_chunk, far_away,
check_and_change_labels, s390_final_chunkify): Delete.
(s390_split_branches, s390_chunkify_pool): New functions.
(s390_function_prologue): Call them.
-
+
* config/s390/s390.h (S390_REL_MAX): Delete.
(S390_CHUNK_MAX, S390_CHUNK_OV): Adjust values.
- * config/s390/s390.md (cjump, icjump, jump): Fix length
+ * config/s390/s390.md (cjump, icjump, jump): Fix length
attribute calculation.
enum machine_mode, rtx));
/* Non-zero means divides or modulus operations are relatively cheap for
- powers of two, so don't use branches; emit the operation instead.
+ powers of two, so don't use branches; emit the operation instead.
Usually, this will mean that the MD file will emit non-branch
sequences. */
If the target is memory, storing any naturally aligned field can be
done with a simple store. For targets that support fast unaligned
memory, any naturally sized, unit aligned field can be done directly. */
-
+
byte_offset = (bitnum % BITS_PER_WORD) / BITS_PER_UNIT
+ (offset * UNITS_PER_WORD);
&& GET_MODE_BITSIZE (bestmode) > MEM_ALIGN (op0)))
goto insv_loses;
- /* Adjust address to point to the containing unit of that mode.
+ /* Adjust address to point to the containing unit of that mode.
Compute offset as multiple of this unit, counting in bytes. */
unit = GET_MODE_BITSIZE (bestmode);
offset = (bitnum / unit) * GET_MODE_SIZE (bestmode);
if we must narrow it, be sure we do it correctly. */
if (GET_MODE_SIZE (GET_MODE (value)) < GET_MODE_SIZE (maxmode))
- value1 = simplify_gen_subreg (maxmode, value1,
- GET_MODE (value1), 0);
+ {
+ rtx tmp;
+
+ tmp = simplify_subreg (maxmode, value1, GET_MODE (value), 0);
+ if (! tmp)
+ tmp = simplify_gen_subreg (maxmode,
+ force_reg (GET_MODE (value),
+ value1),
+ GET_MODE (value), 0);
+ value1 = tmp;
+ }
else
value1 = gen_lowpart (maxmode, value1);
}
{
/* Get the proper mode to use for this field. We want a mode that
includes the entire field. If such a mode would be larger than
- a word, we won't be doing the extraction the normal way.
+ a word, we won't be doing the extraction the normal way.
We don't want a mode bigger than the destination. */
mode = GET_MODE (op0);
/* ??? We currently assume TARGET is at least as big as BITSIZE.
If that's wrong, the solution is to test for it and set TARGET to 0
if needed. */
-
+
/* If OP0 is a register, BITPOS must count within a word.
But as we have it, it counts within whatever size OP0 now has.
On a bigendian machine, these are not the same, so convert. */
}
/* Handle fields bigger than a word. */
-
+
if (bitsize > BITS_PER_WORD)
{
/* Here we transfer the words of the field
build_int_2 (GET_MODE_BITSIZE (mode) - bitsize, 0),
NULL_RTX, 0);
}
-
+
/* From here on we know the desired field is smaller than a word. */
/* Check if there is a correspondingly-sized integer field, so we can
safely extract it as one size of integer, if necessary; then
truncate or extend to the size that is wanted; then use SUBREGs or
convert_to_mode to get one of the modes we really wanted. */
-
+
int_mode = int_mode_for_mode (tmode);
if (int_mode == BLKmode)
int_mode = int_mode_for_mode (mode);
}
else
extzv_loses:
- target = extract_fixed_bit_field (int_mode, op0, offset, bitsize,
+ target = extract_fixed_bit_field (int_mode, op0, offset, bitsize,
bitpos, target, 1);
}
else
target = extract_fixed_bit_field (int_mode, op0, offset, bitsize,
bitpos, target, 0);
}
- }
+ }
else
extv_loses:
- target = extract_fixed_bit_field (int_mode, op0, offset, bitsize,
+ target = extract_fixed_bit_field (int_mode, op0, offset, bitsize,
bitpos, target, 0);
}
if (target == spec_target)
}
return expand_shift (RSHIFT_EXPR, mode, op0,
- build_int_2 (GET_MODE_BITSIZE (mode) - bitsize, 0),
+ build_int_2 (GET_MODE_BITSIZE (mode) - bitsize, 0),
target, 0);
}
\f
if (bitpos + bitsize < HOST_BITS_PER_WIDE_INT)
masklow &= ((unsigned HOST_WIDE_INT) -1
>> (HOST_BITS_PER_WIDE_INT - bitpos - bitsize));
-
+
if (bitpos <= HOST_BITS_PER_WIDE_INT)
maskhigh = -1;
else
&& (unsigned int) INTVAL (op1) < GET_MODE_BITSIZE (mode))
temp = expand_binop (mode,
left ? rotr_optab : rotl_optab,
- shifted,
+ shifted,
GEN_INT (GET_MODE_BITSIZE (mode)
- INTVAL (op1)),
target, unsignedp, methods);
}
/* We used to try extzv here for logical right shifts, but that was
- only useful for one machine, the VAX, and caused poor code
+ only useful for one machine, the VAX, and caused poor code
generation there for lshrdi3, so the code was deleted and a
define_expand for lshrsi3 was added to vax.md. */
}
for (w = 1; (w & t) != 0; w <<= 1)
;
/* If T was -1, then W will be zero after the loop. This is another
- case where T ends with ...111. Handling this with (T + 1) and
+ case where T ends with ...111. Handling this with (T + 1) and
subtract 1 produces slightly better code and results in algorithm
selection much faster than treating it like the ...0111 case
below. */
&& ! preserve)
? target : 0;
rtx accum_target = preserve ? 0 : accum;
-
+
switch (alg.op[opno])
{
case alg_shift:
}
insn = get_last_insn ();
- set_unique_reg_note (insn,
+ set_unique_reg_note (insn,
REG_EQUAL,
gen_rtx_MULT (nmode, tem,
GEN_INT (val_so_far)));
/* This used to use umul_optab if unsigned, but for non-widening multiply
there is no difference between signed and unsigned. */
- op0 = expand_binop (mode,
+ op0 = expand_binop (mode,
! unsignedp
&& flag_trapv && (GET_MODE_CLASS(mode) == MODE_INT)
? smulv_optab : smul_optab,
{
op1 = force_reg (mode, op1);
goto try;
- }
+ }
/* Try widening the mode and perform a non-widening multiplication. */
moptab = smul_optab;
if (insn != last
&& (set = single_set (insn)) != 0
&& SET_DEST (set) == quotient)
- set_unique_reg_note (insn,
+ set_unique_reg_note (insn,
REG_EQUAL,
gen_rtx_UDIV (compute_mode, op0, op1));
}
&& SET_DEST (set) == quotient
&& abs_d < ((unsigned HOST_WIDE_INT) 1
<< (HOST_BITS_PER_WIDE_INT - 1)))
- set_unique_reg_note (insn,
+ set_unique_reg_note (insn,
REG_EQUAL,
gen_rtx_DIV (compute_mode,
op0,
if (insn != last
&& (set = single_set (insn)) != 0
&& SET_DEST (set) == quotient)
- set_unique_reg_note (insn,
+ set_unique_reg_note (insn,
REG_EQUAL,
gen_rtx_DIV (compute_mode, op0, op1));
}
emit_label (label);
}
return gen_lowpart (mode, rem_flag ? remainder : quotient);
-
+
default:
abort ();
}
elt = CONST_VECTOR_ELT (x, i);
t = tree_cons (NULL_TREE, make_tree (type, elt), t);
}
-
+
return build_vector (type, t);
}
case PLUS:
return fold (build (PLUS_EXPR, type, make_tree (type, XEXP (x, 0)),
make_tree (type, XEXP (x, 1))));
-
+
case MINUS:
return fold (build (MINUS_EXPR, type, make_tree (type, XEXP (x, 0)),
make_tree (type, XEXP (x, 1))));
-
+
case NEG:
return fold (build1 (NEGATE_EXPR, type, make_tree (type, XEXP (x, 0))));
case MULT:
return fold (build (MULT_EXPR, type, make_tree (type, XEXP (x, 0)),
make_tree (type, XEXP (x, 1))));
-
+
case ASHIFT:
return fold (build (LSHIFT_EXPR, type, make_tree (type, XEXP (x, 0)),
make_tree (type, XEXP (x, 1))));
-
+
case LSHIFTRT:
return fold (convert (type,
build (RSHIFT_EXPR, unsigned_type (type),
make_tree (unsigned_type (type),
XEXP (x, 0)),
make_tree (type, XEXP (x, 1)))));
-
+
case ASHIFTRT:
return fold (convert (type,
build (RSHIFT_EXPR, signed_type (type),
make_tree (signed_type (type), XEXP (x, 0)),
make_tree (type, XEXP (x, 1)))));
-
+
case DIV:
if (TREE_CODE (type) != REAL_TYPE)
t = signed_type (type);
Return 0 if that cannot be done.
MODE is the mode to use for OP0 and OP1 should they be CONST_INTs. If
- it is VOIDmode, they cannot both be CONST_INT.
+ it is VOIDmode, they cannot both be CONST_INT.
UNSIGNEDP is for the case where we have to widen the operands
to perform the operation. It says to use zero-extension.
if (mode == VOIDmode)
mode = GET_MODE (op0);
- /* For some comparisons with 1 and -1, we can convert this to
+ /* For some comparisons with 1 and -1, we can convert this to
comparisons with zero. This will often produce more opportunities for
store-flag insns. */
else
abort ();
- /* If we were converting to a smaller mode, do the
+ /* If we were converting to a smaller mode, do the
conversion now. */
if (target_mode != compare_mode)
{
return tem;
}
- /* Some other cases we can do are EQ, NE, LE, and GT comparisons with
+ /* Some other cases we can do are EQ, NE, LE, and GT comparisons with
the constant zero. Reject all other comparisons at this point. Only
do LE and GT if branches are expensive since they are expensive on
2-operand machines. */
tem = expand_binop (mode, sub_optab, tem, op0, subtarget, 0,
OPTAB_WIDEN);
}
-
+
if (code == EQ || code == NE)
{
/* For EQ or NE, one way to do the comparison is to apply an operation
that converts the operand into a positive number if it is non-zero
or zero if it was originally zero. Then, for EQ, we subtract 1 and
for NE we negate. This puts the result in the sign bit. Then we
- normalize with a shift, if needed.
+ normalize with a shift, if needed.
Two operations that can do the above actions are ABS and FFS, so try
them. If that doesn't work, and MODE is smaller than a full word,
we can use zero-extension to the wider mode (an unsigned conversion)
as the operation. */
- /* Note that ABS doesn't yield a positive number for INT_MIN, but
- that is compensated by the subsequent overflow when subtracting
+ /* Note that ABS doesn't yield a positive number for INT_MIN, but
+ that is compensated by the subsequent overflow when subtracting
one / negating. */
if (abs_optab->handlers[(int) mode].insn_code != CODE_FOR_nothing)
OSDN Git Service
