static bool
nds32_expand_setmem_loop_v3m (rtx dstmem, rtx size, rtx value)
{
rtx base_reg = copy_to_mode_reg (Pmode, XEXP (dstmem, 0));
rtx need_align_bytes = gen_reg_rtx (SImode);
rtx last_2_bit = gen_reg_rtx (SImode);
rtx byte_loop_base = gen_reg_rtx (SImode);
rtx byte_loop_size = gen_reg_rtx (SImode);
rtx remain_size = gen_reg_rtx (SImode);
rtx new_base_reg;
rtx value4byte, value4doubleword;
rtx byte_mode_size;
rtx last_byte_loop_label = gen_label_rtx ();
size = force_reg (SImode, size);
value4doubleword = nds32_gen_dup_8_byte_to_double_word_value (value);
value4byte = simplify_gen_subreg (QImode, value4doubleword, DImode,
subreg_lowpart_offset (QImode, DImode));
emit_move_insn (byte_loop_size, size);
emit_move_insn (byte_loop_base, base_reg);
/* Jump to last byte loop if size is less than 16. */
emit_cmp_and_jump_insns (size, gen_int_mode (16, SImode), LE, NULL,
SImode, 1, last_byte_loop_label);
/* Make sure align to 4 byte first since v3m can't unalign access. */
emit_insn (gen_andsi3 (last_2_bit,
base_reg,
gen_int_mode (0x3, SImode)));
emit_insn (gen_subsi3 (need_align_bytes,
gen_int_mode (4, SImode),
last_2_bit));
/* Align to 4 byte. */
new_base_reg = emit_setmem_byte_loop (base_reg,
need_align_bytes,
value4byte,
true);
/* Calculate remain size. */
emit_insn (gen_subsi3 (remain_size, size, need_align_bytes));
/* Set memory word by word. */
byte_mode_size = emit_setmem_doubleword_loop (new_base_reg,
remain_size,
value4doubleword);
emit_move_insn (byte_loop_base, new_base_reg);
emit_move_insn (byte_loop_size, byte_mode_size);
emit_label (last_byte_loop_label);
/* And set memory for remain bytes. */
emit_setmem_byte_loop (byte_loop_base, byte_loop_size, value4byte, false);
return true;
}
static rtx
nds32_gen_dup_4_byte_to_word_value_aux (rtx value, rtx value4word)
{
gcc_assert (GET_MODE (value) == QImode || CONST_INT_P (value));
if (CONST_INT_P (value))
{
unsigned HOST_WIDE_INT val = UINTVAL (value) & GET_MODE_MASK(QImode);
rtx new_val = gen_int_mode (val | (val << 8)
| (val << 16) | (val << 24), SImode);
/* Just calculate at here if it's constant value. */
emit_move_insn (value4word, new_val);
}
else
{
if (NDS32_EXT_DSP_P ())
{
/* ! prepare word
insb $tmp, $value, 1 ! $tmp <- 0x0000abab
pkbb16 $tmp6, $tmp2, $tmp2 ! $value4word <- 0xabababab */
rtx tmp = gen_reg_rtx (SImode);
convert_move (tmp, value, true);
emit_insn (
gen_insvsi_internal (tmp, gen_int_mode (0x8, SImode), tmp));
emit_insn (gen_pkbbsi_1 (value4word, tmp, tmp));
}
else
{
/* ! prepare word
andi $tmp1, $value, 0xff ! $tmp1 <- 0x000000ab
slli $tmp2, $tmp1, 8 ! $tmp2 <- 0x0000ab00
or $tmp3, $tmp1, $tmp2 ! $tmp3 <- 0x0000abab
slli $tmp4, $tmp3, 16 ! $tmp4 <- 0xabab0000
or $val4word, $tmp3, $tmp4 ! $value4word <- 0xabababab */
rtx tmp1, tmp2, tmp3, tmp4;
tmp1 = expand_binop (SImode, and_optab, value,
gen_int_mode (0xff, SImode),
NULL_RTX, 0, OPTAB_WIDEN);
tmp2 = expand_binop (SImode, ashl_optab, tmp1,
gen_int_mode (8, SImode),
NULL_RTX, 0, OPTAB_WIDEN);
tmp3 = expand_binop (SImode, ior_optab, tmp1, tmp2,
NULL_RTX, 0, OPTAB_WIDEN);
tmp4 = expand_binop (SImode, ashl_optab, tmp3,
gen_int_mode (16, SImode),
NULL_RTX, 0, OPTAB_WIDEN);
emit_insn (gen_iorsi3 (value4word, tmp3, tmp4));
}
}
return value4word;
}
/* Function to check if 'bitci' instruction can be used with IVAL. */
int
nds32_can_use_bitci_p (int ival)
{
/* If we are using V3 ISA, we have 'bitci' instruction.
Try to see if we can present 'andi' semantic with
such 'bit-clear-immediate' operation.
For example, 'andi $r0,$r0,0xfffffffc' can be
presented with 'bitci $r0,$r0,3'. */
return (TARGET_ISA_V3
&& (ival < 0)
&& satisfies_constraint_Iu15 (gen_int_mode (~ival, SImode)));
}
static rtx
emit_setmem_doubleword_loop (rtx itr, rtx size, rtx value)
{
rtx word_mode_label = gen_label_rtx ();
rtx word_mode_end_label = gen_label_rtx ();
rtx byte_mode_size = gen_reg_rtx (SImode);
rtx byte_mode_size_tmp = gen_reg_rtx (SImode);
rtx word_mode_end = gen_reg_rtx (SImode);
rtx size_for_word = gen_reg_rtx (SImode);
/* and $size_for_word, $size, #~0x7 */
size_for_word = expand_binop (SImode, and_optab, size,
gen_int_mode (~0x7, SImode),
NULL_RTX, 0, OPTAB_WIDEN);
emit_move_insn (byte_mode_size, size);
/* beqz $size_for_word, .Lbyte_mode_entry */
emit_cmp_and_jump_insns (size_for_word, const0_rtx, EQ, NULL,
SImode, 1, word_mode_end_label);
/* add $word_mode_end, $dst, $size_for_word */
word_mode_end = expand_binop (Pmode, add_optab, itr, size_for_word,
NULL_RTX, 0, OPTAB_WIDEN);
/* andi $byte_mode_size, $size, 0x7 */
byte_mode_size_tmp = expand_binop (SImode, and_optab, size, GEN_INT (0x7),
NULL_RTX, 0, OPTAB_WIDEN);
emit_move_insn (byte_mode_size, byte_mode_size_tmp);
/* .Lword_mode: */
emit_label (word_mode_label);
/* ! word-mode set loop
smw.bim $value4word, [$dst_itr], $value4word, 0
bne $word_mode_end, $dst_itr, .Lword_mode */
emit_insn (gen_unaligned_store_update_base_dw (itr,
itr,
value));
emit_cmp_and_jump_insns (word_mode_end, itr, NE, NULL,
Pmode, 1, word_mode_label);
emit_label (word_mode_end_label);
return byte_mode_size;
}
static bool
update_reg_equal_equiv_notes (rtx insn, enum machine_mode new_mode,
enum machine_mode old_mode, enum rtx_code code)
{
rtx *loc = ®_NOTES (insn);
while (*loc)
{
enum reg_note kind = REG_NOTE_KIND (*loc);
if (kind == REG_EQUAL || kind == REG_EQUIV)
{
rtx orig_src = XEXP (*loc, 0);
/* Update equivalency constants. Recall that RTL constants are
sign-extended. */
if (GET_CODE (orig_src) == CONST_INT
&& HOST_BITS_PER_WIDE_INT >= GET_MODE_BITSIZE (new_mode))
{
if (INTVAL (orig_src) >= 0 || code == SIGN_EXTEND)
/* Nothing needed. */;
else
{
/* Zero-extend the negative constant by masking out the
bits outside the source mode. */
rtx new_const_int
= gen_int_mode (INTVAL (orig_src)
& GET_MODE_MASK (old_mode),
new_mode);
if (!validate_change (insn, &XEXP (*loc, 0),
new_const_int, true))
return false;
}
loc = &XEXP (*loc, 1);
}
/* Drop all other notes, they assume a wrong mode. */
else if (!validate_change (insn, loc, XEXP (*loc, 1), true))
return false;
}
else
loc = &XEXP (*loc, 1);
}
return true;
}
static void
canonicalize_address (rtx x)
{
for (;;)
switch (GET_CODE (x))
{
case ASHIFT:
if (CONST_INT_P (XEXP (x, 1))
&& INTVAL (XEXP (x, 1)) < GET_MODE_BITSIZE (GET_MODE (x))
&& INTVAL (XEXP (x, 1)) >= 0)
{
HOST_WIDE_INT shift = INTVAL (XEXP (x, 1));
PUT_CODE (x, MULT);
XEXP (x, 1) = gen_int_mode (HOST_WIDE_INT_1 << shift,
GET_MODE (x));
}
x = XEXP (x, 0);
break;
case PLUS:
if (GET_CODE (XEXP (x, 0)) == PLUS
|| GET_CODE (XEXP (x, 0)) == ASHIFT
|| GET_CODE (XEXP (x, 0)) == CONST)
canonicalize_address (XEXP (x, 0));
x = XEXP (x, 1);
break;
case CONST:
x = XEXP (x, 0);
break;
default:
return;
}
}
static bool
combine_set_extension (ext_cand *cand, rtx_insn *curr_insn, rtx *orig_set)
{
rtx orig_src = SET_SRC (*orig_set);
machine_mode orig_mode = GET_MODE (SET_DEST (*orig_set));
rtx new_set;
rtx cand_pat = PATTERN (cand->insn);
/* If the extension's source/destination registers are not the same
then we need to change the original load to reference the destination
of the extension. Then we need to emit a copy from that destination
to the original destination of the load. */
rtx new_reg;
bool copy_needed
= (REGNO (SET_DEST (cand_pat)) != REGNO (XEXP (SET_SRC (cand_pat), 0)));
if (copy_needed)
new_reg = gen_rtx_REG (cand->mode, REGNO (SET_DEST (cand_pat)));
else
new_reg = gen_rtx_REG (cand->mode, REGNO (SET_DEST (*orig_set)));
#if 0
/* Rethinking test. Temporarily disabled. */
/* We're going to be widening the result of DEF_INSN, ensure that doing so
doesn't change the number of hard registers needed for the result. */
if (HARD_REGNO_NREGS (REGNO (new_reg), cand->mode)
!= HARD_REGNO_NREGS (REGNO (SET_DEST (*orig_set)),
GET_MODE (SET_DEST (*orig_set))))
return false;
#endif
/* Merge constants by directly moving the constant into the register under
some conditions. Recall that RTL constants are sign-extended. */
if (GET_CODE (orig_src) == CONST_INT
&& HOST_BITS_PER_WIDE_INT >= GET_MODE_BITSIZE (cand->mode))
{
if (INTVAL (orig_src) >= 0 || cand->code == SIGN_EXTEND)
new_set = gen_rtx_SET (new_reg, orig_src);
else
{
/* Zero-extend the negative constant by masking out the bits outside
the source mode. */
rtx new_const_int
= gen_int_mode (INTVAL (orig_src) & GET_MODE_MASK (orig_mode),
GET_MODE (new_reg));
new_set = gen_rtx_SET (new_reg, new_const_int);
}
}
else if (GET_MODE (orig_src) == VOIDmode)
{
/* This is mostly due to a call insn that should not be optimized. */
return false;
}
else if (GET_CODE (orig_src) == cand->code)
{
/* Here is a sequence of two extensions. Try to merge them. */
rtx temp_extension
= gen_rtx_fmt_e (cand->code, cand->mode, XEXP (orig_src, 0));
rtx simplified_temp_extension = simplify_rtx (temp_extension);
if (simplified_temp_extension)
temp_extension = simplified_temp_extension;
new_set = gen_rtx_SET (new_reg, temp_extension);
}
else if (GET_CODE (orig_src) == IF_THEN_ELSE)
{
/* Only IF_THEN_ELSE of phi-type copies are combined. Otherwise,
in general, IF_THEN_ELSE should not be combined. */
return false;
}
else
{
/* This is the normal case. */
rtx temp_extension
= gen_rtx_fmt_e (cand->code, cand->mode, orig_src);
rtx simplified_temp_extension = simplify_rtx (temp_extension);
if (simplified_temp_extension)
temp_extension = simplified_temp_extension;
new_set = gen_rtx_SET (new_reg, temp_extension);
}
/* This change is a part of a group of changes. Hence,
validate_change will not try to commit the change. */
if (validate_change (curr_insn, orig_set, new_set, true)
&& update_reg_equal_equiv_notes (curr_insn, cand->mode, orig_mode,
cand->code))
{
if (dump_file)
{
fprintf (dump_file,
"Tentatively merged extension with definition %s:\n",
(copy_needed) ? "(copy needed)" : "");
print_rtl_single (dump_file, curr_insn);
}
return true;
}
return false;
}
static bool
combine_set_extension (ext_cand *cand, rtx curr_insn, rtx *orig_set)
{
rtx orig_src = SET_SRC (*orig_set);
rtx new_reg = gen_rtx_REG (cand->mode, REGNO (SET_DEST (*orig_set)));
rtx new_set;
/* Merge constants by directly moving the constant into the register under
some conditions. Recall that RTL constants are sign-extended. */
if (GET_CODE (orig_src) == CONST_INT
&& HOST_BITS_PER_WIDE_INT >= GET_MODE_BITSIZE (cand->mode))
{
if (INTVAL (orig_src) >= 0 || cand->code == SIGN_EXTEND)
new_set = gen_rtx_SET (VOIDmode, new_reg, orig_src);
else
{
/* Zero-extend the negative constant by masking out the bits outside
the source mode. */
enum machine_mode src_mode = GET_MODE (SET_DEST (*orig_set));
rtx new_const_int
= gen_int_mode (INTVAL (orig_src) & GET_MODE_MASK (src_mode),
GET_MODE (new_reg));
new_set = gen_rtx_SET (VOIDmode, new_reg, new_const_int);
}
}
else if (GET_MODE (orig_src) == VOIDmode)
{
/* This is mostly due to a call insn that should not be optimized. */
return false;
}
else if (GET_CODE (orig_src) == cand->code)
{
/* Here is a sequence of two extensions. Try to merge them. */
rtx temp_extension
= gen_rtx_fmt_e (cand->code, cand->mode, XEXP (orig_src, 0));
rtx simplified_temp_extension = simplify_rtx (temp_extension);
if (simplified_temp_extension)
temp_extension = simplified_temp_extension;
new_set = gen_rtx_SET (VOIDmode, new_reg, temp_extension);
}
else if (GET_CODE (orig_src) == IF_THEN_ELSE)
{
/* Only IF_THEN_ELSE of phi-type copies are combined. Otherwise,
in general, IF_THEN_ELSE should not be combined. */
return false;
}
else
{
/* This is the normal case. */
rtx temp_extension
= gen_rtx_fmt_e (cand->code, cand->mode, orig_src);
rtx simplified_temp_extension = simplify_rtx (temp_extension);
if (simplified_temp_extension)
temp_extension = simplified_temp_extension;
new_set = gen_rtx_SET (VOIDmode, new_reg, temp_extension);
}
/* This change is a part of a group of changes. Hence,
validate_change will not try to commit the change. */
if (validate_change (curr_insn, orig_set, new_set, true))
{
if (dump_file)
{
fprintf (dump_file,
"Tentatively merged extension with definition:\n");
print_rtl_single (dump_file, curr_insn);
}
return true;
}
return false;
}