/* tc-i386.c -- Assemble code for the Intel 80386
Copyright 1989, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011,
- 2012
+ 2012, 2013, 2014
Free Software Foundation, Inc.
This file is part of GAS, the GNU Assembler.
#define DATA_PREFIX 3
#define REP_PREFIX 4
#define HLE_PREFIX REP_PREFIX
+#define BND_PREFIX REP_PREFIX
#define LOCK_PREFIX 5
#define REX_PREFIX 6 /* must come last. */
#define MAX_PREFIXES 7 /* max prefixes per opcode */
#define QWORD_MNEM_SUFFIX 'q'
#define XMMWORD_MNEM_SUFFIX 'x'
#define YMMWORD_MNEM_SUFFIX 'y'
+#define ZMMWORD_MNEM_SUFFIX 'z'
/* Intel Syntax. Use a non-ascii letter since since it never appears
in instructions. */
#define LONG_DOUBLE_MNEM_SUFFIX '\1'
static const char *default_arch = DEFAULT_ARCH;
+/* This struct describes rounding control and SAE in the instruction. */
+struct RC_Operation
+{
+ enum rc_type
+ {
+ rne = 0,
+ rd,
+ ru,
+ rz,
+ saeonly
+ } type;
+ int operand;
+};
+
+static struct RC_Operation rc_op;
+
+/* The struct describes masking, applied to OPERAND in the instruction.
+ MASK is a pointer to the corresponding mask register. ZEROING tells
+ whether merging or zeroing mask is used. */
+struct Mask_Operation
+{
+ const reg_entry *mask;
+ unsigned int zeroing;
+ /* The operand where this operation is associated. */
+ int operand;
+};
+
+static struct Mask_Operation mask_op;
+
+/* The struct describes broadcasting, applied to OPERAND. FACTOR is
+ broadcast factor. */
+struct Broadcast_Operation
+{
+ /* Type of broadcast: no broadcast, {1to8}, or {1to16}. */
+ int type;
+
+ /* Index of broadcasted operand. */
+ int operand;
+};
+
+static struct Broadcast_Operation broadcast_op;
+
/* VEX prefix. */
typedef struct
{
- /* VEX prefix is either 2 byte or 3 byte. */
- unsigned char bytes[3];
+ /* VEX prefix is either 2 byte or 3 byte. EVEX is 4 byte. */
+ unsigned char bytes[4];
unsigned int length;
/* Destination or source register specifier. */
const reg_entry *register_specifier;
unsupported,
invalid_vsib_address,
invalid_vector_register_set,
- unsupported_vector_index_register
+ unsupported_vector_index_register,
+ unsupported_broadcast,
+ broadcast_not_on_src_operand,
+ broadcast_needed,
+ unsupported_masking,
+ mask_not_on_destination,
+ no_default_mask,
+ unsupported_rc_sae,
+ rc_sae_operand_not_last_imm,
+ invalid_register_operand,
+ try_vector_disp8
};
struct _i386_insn
addressing modes of this insn are encoded. */
modrm_byte rm;
rex_byte rex;
+ rex_byte vrex;
sib_byte sib;
vex_prefix vex;
+ /* Masking attributes. */
+ struct Mask_Operation *mask;
+
+ /* Rounding control and SAE attributes. */
+ struct RC_Operation *rounding;
+
+ /* Broadcasting attributes. */
+ struct Broadcast_Operation *broadcast;
+
+ /* Compressed disp8*N attribute. */
+ unsigned int memshift;
+
/* Swap operand in encoding. */
unsigned int swap_operand;
disp_encoding_32bit
} disp_encoding;
- /* Have HLE prefix. */
- unsigned int have_hle;
+ /* REP prefix. */
+ const char *rep_prefix;
+
+ /* HLE prefix. */
+ const char *hle_prefix;
+
+ /* Have BND prefix. */
+ const char *bnd_prefix;
+
+ /* Need VREX to support upper 16 registers. */
+ int need_vrex;
/* Error message. */
enum i386_error error;
typedef struct _i386_insn i386_insn;
+/* Link RC type with corresponding string, that'll be looked for in
+ asm. */
+struct RC_name
+{
+ enum rc_type type;
+ const char *name;
+ unsigned int len;
+};
+
+static const struct RC_name RC_NamesTable[] =
+{
+ { rne, STRING_COMMA_LEN ("rn-sae") },
+ { rd, STRING_COMMA_LEN ("rd-sae") },
+ { ru, STRING_COMMA_LEN ("ru-sae") },
+ { rz, STRING_COMMA_LEN ("rz-sae") },
+ { saeonly, STRING_COMMA_LEN ("sae") },
+};
+
/* List of chars besides those in app.c:symbol_chars that can start an
operand. Used to prevent the scrubber eating vital white-space. */
-const char extra_symbol_chars[] = "*%-(["
+const char extra_symbol_chars[] = "*%-([{"
#ifdef LEX_AT
"@"
#endif
static enum x86_elf_abi x86_elf_abi = I386_ABI;
#endif
-/* The names used to print error messages. */
-static const char *flag_code_names[] =
- {
- "32",
- "16",
- "64"
- };
-
/* 1 for intel syntax,
0 if att syntax. */
static int intel_syntax = 0;
/* 1 if register prefix % not required. */
static int allow_naked_reg = 0;
+/* 1 if the assembler should add BND prefix for all control-tranferring
+ instructions supporting it, even if this prefix wasn't specified
+ explicitly. */
+static int add_bnd_prefix = 0;
+
/* 1 if pseudo index register, eiz/riz, is allowed . */
static int allow_index_reg = 0;
vex256
} avxscalar;
+/* Encode scalar EVEX LIG instructions with specific vector length. */
+static enum
+ {
+ evexl128 = 0,
+ evexl256,
+ evexl512
+ } evexlig;
+
+/* Encode EVEX WIG instructions with specific evex.w. */
+static enum
+ {
+ evexw0 = 0,
+ evexw1
+ } evexwig;
+
/* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
static symbolS *GOT_symbol;
CPU_BDVER1_FLAGS, 0, 0 },
{ STRING_COMMA_LEN ("bdver2"), PROCESSOR_BD,
CPU_BDVER2_FLAGS, 0, 0 },
+ { STRING_COMMA_LEN ("bdver3"), PROCESSOR_BD,
+ CPU_BDVER3_FLAGS, 0, 0 },
+ { STRING_COMMA_LEN ("bdver4"), PROCESSOR_BD,
+ CPU_BDVER4_FLAGS, 0, 0 },
{ STRING_COMMA_LEN ("btver1"), PROCESSOR_BT,
CPU_BTVER1_FLAGS, 0, 0 },
{ STRING_COMMA_LEN ("btver2"), PROCESSOR_BT,
CPU_AVX_FLAGS, 0, 0 },
{ STRING_COMMA_LEN (".avx2"), PROCESSOR_UNKNOWN,
CPU_AVX2_FLAGS, 0, 0 },
+ { STRING_COMMA_LEN (".avx512f"), PROCESSOR_UNKNOWN,
+ CPU_AVX512F_FLAGS, 0, 0 },
+ { STRING_COMMA_LEN (".avx512cd"), PROCESSOR_UNKNOWN,
+ CPU_AVX512CD_FLAGS, 0, 0 },
+ { STRING_COMMA_LEN (".avx512er"), PROCESSOR_UNKNOWN,
+ CPU_AVX512ER_FLAGS, 0, 0 },
+ { STRING_COMMA_LEN (".avx512pf"), PROCESSOR_UNKNOWN,
+ CPU_AVX512PF_FLAGS, 0, 0 },
{ STRING_COMMA_LEN (".noavx"), PROCESSOR_UNKNOWN,
CPU_ANY_AVX_FLAGS, 0, 1 },
{ STRING_COMMA_LEN (".vmx"), PROCESSOR_UNKNOWN,
CPU_RDSEED_FLAGS, 0, 0 },
{ STRING_COMMA_LEN (".prfchw"), PROCESSOR_UNKNOWN,
CPU_PRFCHW_FLAGS, 0, 0 },
+ { STRING_COMMA_LEN (".smap"), PROCESSOR_UNKNOWN,
+ CPU_SMAP_FLAGS, 0, 0 },
+ { STRING_COMMA_LEN (".mpx"), PROCESSOR_UNKNOWN,
+ CPU_MPX_FLAGS, 0, 0 },
+ { STRING_COMMA_LEN (".sha"), PROCESSOR_UNKNOWN,
+ CPU_SHA_FLAGS, 0, 0 },
};
#ifdef I386COFF
= OPERAND_TYPE_ANYDISP;
static const i386_operand_type regxmm = OPERAND_TYPE_REGXMM;
static const i386_operand_type regymm = OPERAND_TYPE_REGYMM;
+static const i386_operand_type regzmm = OPERAND_TYPE_REGZMM;
+static const i386_operand_type regmask = OPERAND_TYPE_REGMASK;
static const i386_operand_type imm8 = OPERAND_TYPE_IMM8;
static const i386_operand_type imm8s = OPERAND_TYPE_IMM8S;
static const i386_operand_type imm16 = OPERAND_TYPE_IMM16;
|| (i.types[j].bitfield.xmmword
&& !t->operand_types[j].bitfield.xmmword)
|| (i.types[j].bitfield.ymmword
- && !t->operand_types[j].bitfield.ymmword)));
+ && !t->operand_types[j].bitfield.ymmword)
+ || (i.types[j].bitfield.zmmword
+ && !t->operand_types[j].bitfield.zmmword)));
}
/* Return 1 if there is no size conflict on any operands for
temp.bitfield.tbyte = 0;
temp.bitfield.xmmword = 0;
temp.bitfield.ymmword = 0;
+ temp.bitfield.zmmword = 0;
if (operand_type_all_zero (&temp))
goto mismatch;
static INLINE unsigned int
mode_from_disp_size (i386_operand_type t)
{
- if (t.bitfield.disp8)
+ if (t.bitfield.disp8 || t.bitfield.vec_disp8)
return 1;
else if (t.bitfield.disp16
|| t.bitfield.disp32
} /* fits_in_unsigned_long() */
static INLINE int
+fits_in_vec_disp8 (offsetT num)
+{
+ int shift = i.memshift;
+ unsigned int mask;
+
+ if (shift == -1)
+ abort ();
+
+ mask = (1 << shift) - 1;
+
+ /* Return 0 if NUM isn't properly aligned. */
+ if ((num & mask))
+ return 0;
+
+ /* Check if NUM will fit in 8bit after shift. */
+ return fits_in_signed_byte (num >> shift);
+}
+
+static INLINE int
fits_in_imm4 (offsetT num)
{
return (num & 0xf) == num;
(void *) core_optab);
if (hash_err)
{
- as_fatal (_("internal Error: Can't hash %s: %s"),
+ as_fatal (_("can't hash %s: %s"),
(optab - 1)->name,
hash_err);
}
{
hash_err = hash_insert (reg_hash, regtab->reg_name, (void *) regtab);
if (hash_err)
- as_fatal (_("internal Error: Can't hash %s: %s"),
+ as_fatal (_("can't hash %s: %s"),
regtab->reg_name,
hash_err);
}
register_chars[c] = mnemonic_chars[c];
operand_chars[c] = c;
}
+ else if (c == '{' || c == '}')
+ operand_chars[c] = c;
if (ISALPHA (c) || ISDIGIT (c))
identifier_chars[c] = c;
operand_chars[(unsigned char) *p] = *p;
}
-#if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
- if (IS_ELF)
- {
- record_alignment (text_section, 2);
- record_alignment (data_section, 2);
- record_alignment (bss_section, 2);
- }
-#endif
-
if (flag_code == CODE_64BIT)
{
#if defined (OBJ_COFF) && defined (TE_PE)
|| x->types[j].bitfield.regmmx
|| x->types[j].bitfield.regxmm
|| x->types[j].bitfield.regymm
+ || x->types[j].bitfield.regzmm
|| x->types[j].bitfield.sreg2
|| x->types[j].bitfield.sreg3
|| x->types[j].bitfield.control
{ OPERAND_TYPE_DISP32, "d32" },
{ OPERAND_TYPE_DISP32S, "d32s" },
{ OPERAND_TYPE_DISP64, "d64" },
+ { OPERAND_TYPE_VEC_DISP8, "Vector d8" },
{ OPERAND_TYPE_INOUTPORTREG, "InOutPortReg" },
{ OPERAND_TYPE_SHIFTCOUNT, "ShiftCount" },
{ OPERAND_TYPE_CONTROL, "control reg" },
{ OPERAND_TYPE_REGMMX, "rMMX" },
{ OPERAND_TYPE_REGXMM, "rXMM" },
{ OPERAND_TYPE_REGYMM, "rYMM" },
+ { OPERAND_TYPE_REGZMM, "rZMM" },
+ { OPERAND_TYPE_REGMASK, "Mask reg" },
{ OPERAND_TYPE_ESSEG, "es" },
};
reloc (unsigned int size,
int pcrel,
int sign,
+ int bnd_prefix,
bfd_reloc_code_real_type other)
{
if (other != NO_RELOC)
break;
}
+#if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
+ if (other == BFD_RELOC_SIZE32)
+ {
+ if (size == 8)
+ return BFD_RELOC_SIZE64;
+ if (pcrel)
+ as_bad (_("there are no pc-relative size relocations"));
+ }
+#endif
+
/* Sign-checking 4-byte relocations in 16-/32-bit code is pointless. */
if (size == 4 && (flag_code != CODE_64BIT || disallow_64bit_reloc))
sign = -1;
{
case 1: return BFD_RELOC_8_PCREL;
case 2: return BFD_RELOC_16_PCREL;
- case 4: return BFD_RELOC_32_PCREL;
+ case 4: return (bnd_prefix && object_64bit
+ ? BFD_RELOC_X86_64_PC32_BND
+ : BFD_RELOC_32_PCREL);
case 8: return BFD_RELOC_64_PCREL;
}
as_bad (_("cannot do %u byte pc-relative relocation"), size);
&& fixP->fx_r_type == BFD_RELOC_32_PCREL)
return 0;
- /* adjust_reloc_syms doesn't know about the GOT. */
- if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
+ /* Adjust_reloc_syms doesn't know about the GOT. Need to keep symbol
+ for size relocations. */
+ if (fixP->fx_r_type == BFD_RELOC_SIZE32
+ || fixP->fx_r_type == BFD_RELOC_SIZE64
+ || fixP->fx_r_type == BFD_RELOC_386_GOTOFF
|| fixP->fx_r_type == BFD_RELOC_386_PLT32
|| fixP->fx_r_type == BFD_RELOC_386_GOT32
|| fixP->fx_r_type == BFD_RELOC_386_TLS_GD
/* Check register specifier. */
if (i.vex.register_specifier)
- register_specifier = ~register_number (i.vex.register_specifier) & 0xf;
+ {
+ register_specifier =
+ ~register_number (i.vex.register_specifier) & 0xf;
+ gas_assert ((i.vex.register_specifier->reg_flags & RegVRex) == 0);
+ }
else
register_specifier = 0xf;
}
}
+/* Build the EVEX prefix. */
+
+static void
+build_evex_prefix (void)
+{
+ unsigned int register_specifier;
+ unsigned int implied_prefix;
+ unsigned int m, w;
+ rex_byte vrex_used = 0;
+
+ /* Check register specifier. */
+ if (i.vex.register_specifier)
+ {
+ gas_assert ((i.vrex & REX_X) == 0);
+
+ register_specifier = i.vex.register_specifier->reg_num;
+ if ((i.vex.register_specifier->reg_flags & RegRex))
+ register_specifier += 8;
+ /* The upper 16 registers are encoded in the fourth byte of the
+ EVEX prefix. */
+ if (!(i.vex.register_specifier->reg_flags & RegVRex))
+ i.vex.bytes[3] = 0x8;
+ register_specifier = ~register_specifier & 0xf;
+ }
+ else
+ {
+ register_specifier = 0xf;
+
+ /* Encode upper 16 vector index register in the fourth byte of
+ the EVEX prefix. */
+ if (!(i.vrex & REX_X))
+ i.vex.bytes[3] = 0x8;
+ else
+ vrex_used |= REX_X;
+ }
+
+ switch ((i.tm.base_opcode >> 8) & 0xff)
+ {
+ case 0:
+ implied_prefix = 0;
+ break;
+ case DATA_PREFIX_OPCODE:
+ implied_prefix = 1;
+ break;
+ case REPE_PREFIX_OPCODE:
+ implied_prefix = 2;
+ break;
+ case REPNE_PREFIX_OPCODE:
+ implied_prefix = 3;
+ break;
+ default:
+ abort ();
+ }
+
+ /* 4 byte EVEX prefix. */
+ i.vex.length = 4;
+ i.vex.bytes[0] = 0x62;
+
+ /* mmmm bits. */
+ switch (i.tm.opcode_modifier.vexopcode)
+ {
+ case VEX0F:
+ m = 1;
+ break;
+ case VEX0F38:
+ m = 2;
+ break;
+ case VEX0F3A:
+ m = 3;
+ break;
+ default:
+ abort ();
+ break;
+ }
+
+ /* The high 3 bits of the second EVEX byte are 1's compliment of RXB
+ bits from REX. */
+ i.vex.bytes[1] = (~i.rex & 0x7) << 5 | m;
+
+ /* The fifth bit of the second EVEX byte is 1's compliment of the
+ REX_R bit in VREX. */
+ if (!(i.vrex & REX_R))
+ i.vex.bytes[1] |= 0x10;
+ else
+ vrex_used |= REX_R;
+
+ if ((i.reg_operands + i.imm_operands) == i.operands)
+ {
+ /* When all operands are registers, the REX_X bit in REX is not
+ used. We reuse it to encode the upper 16 registers, which is
+ indicated by the REX_B bit in VREX. The REX_X bit is encoded
+ as 1's compliment. */
+ if ((i.vrex & REX_B))
+ {
+ vrex_used |= REX_B;
+ i.vex.bytes[1] &= ~0x40;
+ }
+ }
+
+ /* EVEX instructions shouldn't need the REX prefix. */
+ i.vrex &= ~vrex_used;
+ gas_assert (i.vrex == 0);
+
+ /* Check the REX.W bit. */
+ w = (i.rex & REX_W) ? 1 : 0;
+ if (i.tm.opcode_modifier.vexw)
+ {
+ if (i.tm.opcode_modifier.vexw == VEXW1)
+ w = 1;
+ }
+ /* If w is not set it means we are dealing with WIG instruction. */
+ else if (!w)
+ {
+ if (evexwig == evexw1)
+ w = 1;
+ }
+
+ /* Encode the U bit. */
+ implied_prefix |= 0x4;
+
+ /* The third byte of the EVEX prefix. */
+ i.vex.bytes[2] = (w << 7 | register_specifier << 3 | implied_prefix);
+
+ /* The fourth byte of the EVEX prefix. */
+ /* The zeroing-masking bit. */
+ if (i.mask && i.mask->zeroing)
+ i.vex.bytes[3] |= 0x80;
+
+ /* Don't always set the broadcast bit if there is no RC. */
+ if (!i.rounding)
+ {
+ /* Encode the vector length. */
+ unsigned int vec_length;
+
+ switch (i.tm.opcode_modifier.evex)
+ {
+ case EVEXLIG: /* LL' is ignored */
+ vec_length = evexlig << 5;
+ break;
+ case EVEX128:
+ vec_length = 0 << 5;
+ break;
+ case EVEX256:
+ vec_length = 1 << 5;
+ break;
+ case EVEX512:
+ vec_length = 2 << 5;
+ break;
+ default:
+ abort ();
+ break;
+ }
+ i.vex.bytes[3] |= vec_length;
+ /* Encode the broadcast bit. */
+ if (i.broadcast)
+ i.vex.bytes[3] |= 0x10;
+ }
+ else
+ {
+ if (i.rounding->type != saeonly)
+ i.vex.bytes[3] |= 0x10 | (i.rounding->type << 5);
+ else
+ i.vex.bytes[3] |= 0x10;
+ }
+
+ if (i.mask && i.mask->mask)
+ i.vex.bytes[3] |= i.mask->mask->reg_num;
+}
+
static void
process_immext (void)
{
AVX instructions also use this encoding, for some of
3 argument instructions. */
- gas_assert (i.imm_operands == 0
+ gas_assert (i.imm_operands <= 1
&& (i.operands <= 2
- || (i.tm.opcode_modifier.vex
+ || ((i.tm.opcode_modifier.vex
+ || i.tm.opcode_modifier.evex)
&& i.operands <= 4)));
exp = &im_expressions[i.imm_operands++];
default:
abort ();
case HLEPrefixNone:
- if (i.prefix[HLE_PREFIX] == XACQUIRE_PREFIX_OPCODE)
- as_bad (_("invalid instruction `%s' after `xacquire'"),
- i.tm.name);
- else
- as_bad (_("invalid instruction `%s' after `xrelease'"),
- i.tm.name);
+ as_bad (_("invalid instruction `%s' after `%s'"),
+ i.tm.name, i.hle_prefix);
return 0;
case HLEPrefixLock:
if (i.prefix[LOCK_PREFIX])
return 1;
- if (i.prefix[HLE_PREFIX] == XACQUIRE_PREFIX_OPCODE)
- as_bad (_("missing `lock' with `xacquire'"));
- else
- as_bad (_("missing `lock' with `xrelease'"));
+ as_bad (_("missing `lock' with `%s'"), i.hle_prefix);
return 0;
case HLEPrefixAny:
return 1;
if (!add_prefix (FWAIT_OPCODE))
return;
+ /* Check if REP prefix is OK. */
+ if (i.rep_prefix && !i.tm.opcode_modifier.repprefixok)
+ {
+ as_bad (_("invalid instruction `%s' after `%s'"),
+ i.tm.name, i.rep_prefix);
+ return;
+ }
+
/* Check for lock without a lockable instruction. Destination operand
must be memory unless it is xchg (0x86). */
if (i.prefix[LOCK_PREFIX]
}
/* Check if HLE prefix is OK. */
- if (i.have_hle && !check_hle ())
+ if (i.hle_prefix && !check_hle ())
return;
+ /* Check BND prefix. */
+ if (i.bnd_prefix && !i.tm.opcode_modifier.bndprefixok)
+ as_bad (_("expecting valid branch instruction after `bnd'"));
+
+ if (i.tm.cpu_flags.bitfield.cpumpx
+ && flag_code == CODE_64BIT
+ && i.prefix[ADDR_PREFIX])
+ as_bad (_("32-bit address isn't allowed in 64-bit MPX instructions."));
+
+ /* Insert BND prefix. */
+ if (add_bnd_prefix
+ && i.tm.opcode_modifier.bndprefixok
+ && !i.prefix[BND_PREFIX])
+ add_prefix (BND_PREFIX_OPCODE);
+
/* Check string instruction segment overrides. */
if (i.tm.opcode_modifier.isstring && i.mem_operands != 0)
{
if (i.tm.opcode_modifier.vex)
build_vex_prefix (t);
+ if (i.tm.opcode_modifier.evex)
+ build_evex_prefix ();
+
/* Handle conversion of 'int $3' --> special int3 insn. XOP or FMA4
instructions may define INT_OPCODE as well, so avoid this corner
case for those instructions that use MODRM. */
const insn_template *t;
char *dot_p = NULL;
- /* Non-zero if we found a prefix only acceptable with string insns. */
- const char *expecting_string_instruction = NULL;
-
while (1)
{
mnem_p = mnemonic;
return NULL;
case PREFIX_REP:
if (current_templates->start->cpu_flags.bitfield.cpuhle)
- i.have_hle = 1;
+ i.hle_prefix = current_templates->start->name;
+ else if (current_templates->start->cpu_flags.bitfield.cpumpx)
+ i.bnd_prefix = current_templates->start->name;
else
- expecting_string_instruction = current_templates->start->name;
+ i.rep_prefix = current_templates->start->name;
break;
default:
break;
as_warn (_("use .code16 to ensure correct addressing mode"));
}
- /* Check for rep/repne without a string (or other allowed) instruction. */
- if (expecting_string_instruction)
- {
- static templates override;
-
- for (t = current_templates->start; t < current_templates->end; ++t)
- if (t->opcode_modifier.repprefixok)
- break;
- if (t >= current_templates->end)
- {
- as_bad (_("expecting string instruction after `%s'"),
- expecting_string_instruction);
- return NULL;
- }
- for (override.start = t; t < current_templates->end; ++t)
- if (!t->opcode_modifier.repprefixok)
- break;
- override.end = t;
- current_templates = &override;
- }
-
return l;
}
temp_reloc = i.reloc[xchg2];
i.reloc[xchg2] = i.reloc[xchg1];
i.reloc[xchg1] = temp_reloc;
+
+ if (i.mask)
+ {
+ if (i.mask->operand == xchg1)
+ i.mask->operand = xchg2;
+ else if (i.mask->operand == xchg2)
+ i.mask->operand = xchg1;
+ }
+ if (i.broadcast)
+ {
+ if (i.broadcast->operand == xchg1)
+ i.broadcast->operand = xchg2;
+ else if (i.broadcast->operand == xchg2)
+ i.broadcast->operand = xchg1;
+ }
+ if (i.rounding)
+ {
+ if (i.rounding->operand == xchg1)
+ i.rounding->operand = xchg2;
+ else if (i.rounding->operand == xchg2)
+ i.rounding->operand = xchg1;
+ }
}
static void
static int
check_VecOperands (const insn_template *t)
{
+ unsigned int op;
+
/* Without VSIB byte, we can't have a vector register for index. */
if (!t->opcode_modifier.vecsib
&& i.index_reg
&& (i.index_reg->reg_type.bitfield.regxmm
- || i.index_reg->reg_type.bitfield.regymm))
+ || i.index_reg->reg_type.bitfield.regymm
+ || i.index_reg->reg_type.bitfield.regzmm))
{
i.error = unsupported_vector_index_register;
return 1;
}
+ /* Check if default mask is allowed. */
+ if (t->opcode_modifier.nodefmask
+ && (!i.mask || i.mask->mask->reg_num == 0))
+ {
+ i.error = no_default_mask;
+ return 1;
+ }
+
/* For VSIB byte, we need a vector register for index, and all vector
registers must be distinct. */
if (t->opcode_modifier.vecsib)
|| !((t->opcode_modifier.vecsib == VecSIB128
&& i.index_reg->reg_type.bitfield.regxmm)
|| (t->opcode_modifier.vecsib == VecSIB256
- && i.index_reg->reg_type.bitfield.regymm)))
+ && i.index_reg->reg_type.bitfield.regymm)
+ || (t->opcode_modifier.vecsib == VecSIB512
+ && i.index_reg->reg_type.bitfield.regzmm)))
{
i.error = invalid_vsib_address;
return 1;
}
- gas_assert (i.reg_operands == 2);
- gas_assert (i.types[0].bitfield.regxmm
- || i.types[0].bitfield.regymm);
- gas_assert (i.types[2].bitfield.regxmm
- || i.types[2].bitfield.regymm);
+ gas_assert (i.reg_operands == 2 || i.mask);
+ if (i.reg_operands == 2 && !i.mask)
+ {
+ gas_assert (i.types[0].bitfield.regxmm
+ || i.types[0].bitfield.regymm
+ || i.types[0].bitfield.regzmm);
+ gas_assert (i.types[2].bitfield.regxmm
+ || i.types[2].bitfield.regymm
+ || i.types[2].bitfield.regzmm);
+ if (operand_check == check_none)
+ return 0;
+ if (register_number (i.op[0].regs)
+ != register_number (i.index_reg)
+ && register_number (i.op[2].regs)
+ != register_number (i.index_reg)
+ && register_number (i.op[0].regs)
+ != register_number (i.op[2].regs))
+ return 0;
+ if (operand_check == check_error)
+ {
+ i.error = invalid_vector_register_set;
+ return 1;
+ }
+ as_warn (_("mask, index, and destination registers should be distinct"));
+ }
+ }
+
+ /* Check if broadcast is supported by the instruction and is applied
+ to the memory operand. */
+ if (i.broadcast)
+ {
+ int broadcasted_opnd_size;
+
+ /* Check if specified broadcast is supported in this instruction,
+ and it's applied to memory operand of DWORD or QWORD type,
+ depending on VecESize. */
+ if (i.broadcast->type != t->opcode_modifier.broadcast
+ || !i.types[i.broadcast->operand].bitfield.mem
+ || (t->opcode_modifier.vecesize == 0
+ && !i.types[i.broadcast->operand].bitfield.dword
+ && !i.types[i.broadcast->operand].bitfield.unspecified)
+ || (t->opcode_modifier.vecesize == 1
+ && !i.types[i.broadcast->operand].bitfield.qword
+ && !i.types[i.broadcast->operand].bitfield.unspecified))
+ goto bad_broadcast;
+
+ broadcasted_opnd_size = t->opcode_modifier.vecesize ? 64 : 32;
+ if (i.broadcast->type == BROADCAST_1TO16)
+ broadcasted_opnd_size <<= 4; /* Broadcast 1to16. */
+ else if (i.broadcast->type == BROADCAST_1TO8)
+ broadcasted_opnd_size <<= 3; /* Broadcast 1to8. */
+ else
+ goto bad_broadcast;
- if (operand_check == check_none)
- return 0;
- if (register_number (i.op[0].regs) != register_number (i.index_reg)
- && register_number (i.op[2].regs) != register_number (i.index_reg)
- && register_number (i.op[0].regs) != register_number (i.op[2].regs))
- return 0;
- if (operand_check == check_error)
+ if ((broadcasted_opnd_size == 256
+ && !t->operand_types[i.broadcast->operand].bitfield.ymmword)
+ || (broadcasted_opnd_size == 512
+ && !t->operand_types[i.broadcast->operand].bitfield.zmmword))
{
- i.error = invalid_vector_register_set;
+ bad_broadcast:
+ i.error = unsupported_broadcast;
+ return 1;
+ }
+ }
+ /* If broadcast is supported in this instruction, we need to check if
+ operand of one-element size isn't specified without broadcast. */
+ else if (t->opcode_modifier.broadcast && i.mem_operands)
+ {
+ /* Find memory operand. */
+ for (op = 0; op < i.operands; op++)
+ if (operand_type_check (i.types[op], anymem))
+ break;
+ gas_assert (op < i.operands);
+ /* Check size of the memory operand. */
+ if ((t->opcode_modifier.vecesize == 0
+ && i.types[op].bitfield.dword)
+ || (t->opcode_modifier.vecesize == 1
+ && i.types[op].bitfield.qword))
+ {
+ i.error = broadcast_needed;
return 1;
}
- as_warn (_("mask, index, and destination registers should be distinct"));
}
+ /* Check if requested masking is supported. */
+ if (i.mask
+ && (!t->opcode_modifier.masking
+ || (i.mask->zeroing
+ && t->opcode_modifier.masking == MERGING_MASKING)))
+ {
+ i.error = unsupported_masking;
+ return 1;
+ }
+
+ /* Check if masking is applied to dest operand. */
+ if (i.mask && (i.mask->operand != (int) (i.operands - 1)))
+ {
+ i.error = mask_not_on_destination;
+ return 1;
+ }
+
+ /* Check RC/SAE. */
+ if (i.rounding)
+ {
+ if ((i.rounding->type != saeonly
+ && !t->opcode_modifier.staticrounding)
+ || (i.rounding->type == saeonly
+ && (t->opcode_modifier.staticrounding
+ || !t->opcode_modifier.sae)))
+ {
+ i.error = unsupported_rc_sae;
+ return 1;
+ }
+ /* If the instruction has several immediate operands and one of
+ them is rounding, the rounding operand should be the last
+ immediate operand. */
+ if (i.imm_operands > 1
+ && i.rounding->operand != (int) (i.imm_operands - 1))
+ {
+ i.error = rc_sae_operand_not_last_imm;
+ return 1;
+ }
+ }
+
+ /* Check vector Disp8 operand. */
+ if (t->opcode_modifier.disp8memshift)
+ {
+ if (i.broadcast)
+ i.memshift = t->opcode_modifier.vecesize ? 3 : 2;
+ else
+ i.memshift = t->opcode_modifier.disp8memshift;
+
+ for (op = 0; op < i.operands; op++)
+ if (operand_type_check (i.types[op], disp)
+ && i.op[op].disps->X_op == O_constant)
+ {
+ offsetT value = i.op[op].disps->X_add_number;
+ int vec_disp8_ok = fits_in_vec_disp8 (value);
+ if (t->operand_types [op].bitfield.vec_disp8)
+ {
+ if (vec_disp8_ok)
+ i.types[op].bitfield.vec_disp8 = 1;
+ else
+ {
+ /* Vector insn can only have Vec_Disp8/Disp32 in
+ 32/64bit modes, and Vec_Disp8/Disp16 in 16bit
+ mode. */
+ i.types[op].bitfield.disp8 = 0;
+ if (flag_code != CODE_16BIT)
+ i.types[op].bitfield.disp16 = 0;
+ }
+ }
+ else if (flag_code != CODE_16BIT)
+ {
+ /* One form of this instruction supports vector Disp8.
+ Try vector Disp8 if we need to use Disp32. */
+ if (vec_disp8_ok && !fits_in_signed_byte (value))
+ {
+ i.error = try_vector_disp8;
+ return 1;
+ }
+ }
+ }
+ }
+ else
+ i.memshift = -1;
+
return 0;
}
static int
VEX_check_operands (const insn_template *t)
{
+ /* VREX is only valid with EVEX prefix. */
+ if (i.need_vrex && !t->opcode_modifier.evex)
+ {
+ i.error = invalid_register_operand;
+ return 1;
+ }
+
if (!t->opcode_modifier.vex)
return 0;
: intel_float_operand (t->name) != 2)
&& ((!operand_types[0].bitfield.regmmx
&& !operand_types[0].bitfield.regxmm
- && !operand_types[0].bitfield.regymm)
+ && !operand_types[0].bitfield.regymm
+ && !operand_types[0].bitfield.regzmm)
|| (!operand_types[t->operands > 1].bitfield.regmmx
&& !!operand_types[t->operands > 1].bitfield.regxmm
- && !!operand_types[t->operands > 1].bitfield.regymm))
+ && !!operand_types[t->operands > 1].bitfield.regymm
+ && !!operand_types[t->operands > 1].bitfield.regzmm))
&& (t->base_opcode != 0x0fc7
|| t->extension_opcode != 1 /* cmpxchg8b */))
continue;
case unsupported_vector_index_register:
err_msg = _("unsupported vector index register");
break;
+ case unsupported_broadcast:
+ err_msg = _("unsupported broadcast");
+ break;
+ case broadcast_not_on_src_operand:
+ err_msg = _("broadcast not on source memory operand");
+ break;
+ case broadcast_needed:
+ err_msg = _("broadcast is needed for operand of such type");
+ break;
+ case unsupported_masking:
+ err_msg = _("unsupported masking");
+ break;
+ case mask_not_on_destination:
+ err_msg = _("mask not on destination operand");
+ break;
+ case no_default_mask:
+ err_msg = _("default mask isn't allowed");
+ break;
+ case unsupported_rc_sae:
+ err_msg = _("unsupported static rounding/sae");
+ break;
+ case rc_sae_operand_not_last_imm:
+ if (intel_syntax)
+ err_msg = _("RC/SAE operand must precede immediate operands");
+ else
+ err_msg = _("RC/SAE operand must follow immediate operands");
+ break;
+ case invalid_register_operand:
+ err_msg = _("invalid register operand");
+ break;
}
as_bad (_("%s for `%s'"), err_msg,
current_templates->start->name);
return 0;
}
else if (i.suffix == XMMWORD_MNEM_SUFFIX
- || i.suffix == YMMWORD_MNEM_SUFFIX)
+ || i.suffix == YMMWORD_MNEM_SUFFIX
+ || i.suffix == ZMMWORD_MNEM_SUFFIX)
{
- /* Skip if the instruction has x/y suffix. match_template
+ /* Skip if the instruction has x/y/z suffix. match_template
should check if it is a valid suffix. */
}
else if (intel_syntax && i.tm.opcode_modifier.ignoresize)
if (i.suffix
&& i.suffix != BYTE_MNEM_SUFFIX
&& i.suffix != XMMWORD_MNEM_SUFFIX
- && i.suffix != YMMWORD_MNEM_SUFFIX)
+ && i.suffix != YMMWORD_MNEM_SUFFIX
+ && i.suffix != ZMMWORD_MNEM_SUFFIX)
{
/* It's not a byte, select word/dword operation. */
if (i.tm.opcode_modifier.w)
|| i.types[op].bitfield.regmmx
|| i.types[op].bitfield.regxmm
|| i.types[op].bitfield.regymm
+ || i.types[op].bitfield.regzmm
|| i.types[op].bitfield.sreg2
|| i.types[op].bitfield.sreg3
|| i.types[op].bitfield.control
i.suffix);
return 0;
}
- /* Warn if the e prefix on a general reg is missing. */
+ /* Warn if the e prefix on a general reg is missing. */
else if ((!quiet_warnings || flag_code == CODE_64BIT)
&& i.types[op].bitfield.reg16
&& (i.tm.operand_types[op].bitfield.reg32
return 0;
}
#if REGISTER_WARNINGS
- else
- as_warn (_("using `%s%s' instead of `%s%s' due to `%c' suffix"),
- register_prefix,
- (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
- register_prefix,
- i.op[op].regs->reg_name,
- i.suffix);
+ as_warn (_("using `%s%s' instead of `%s%s' due to `%c' suffix"),
+ register_prefix,
+ (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
+ register_prefix, i.op[op].regs->reg_name, i.suffix);
#endif
}
- /* Warn if the r prefix on a general reg is missing. */
+ /* Warn if the r prefix on a general reg is present. */
else if (i.types[op].bitfield.reg64
&& (i.tm.operand_types[op].bitfield.reg32
|| i.tm.operand_types[op].bitfield.acc))
i.suffix);
return 0;
}
- /* Warn if the e prefix on a general reg is missing. */
+ /* Warn if the r prefix on a general reg is missing. */
else if ((i.types[op].bitfield.reg16
|| i.types[op].bitfield.reg32)
&& (i.tm.operand_types[op].bitfield.reg32
i.suffix);
return 0;
}
- /* Warn if the e prefix on a general reg is present. */
+ /* Warn if the e or r prefix on a general reg is present. */
else if ((!quiet_warnings || flag_code == CODE_64BIT)
- && i.types[op].bitfield.reg32
+ && (i.types[op].bitfield.reg32
+ || i.types[op].bitfield.reg64)
&& (i.tm.operand_types[op].bitfield.reg16
|| i.tm.operand_types[op].bitfield.acc))
{
i.suffix);
return 0;
}
- else
#if REGISTER_WARNINGS
- as_warn (_("using `%s%s' instead of `%s%s' due to `%c' suffix"),
- register_prefix,
- (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
- register_prefix,
- i.op[op].regs->reg_name,
- i.suffix);
+ as_warn (_("using `%s%s' instead of `%s%s' due to `%c' suffix"),
+ register_prefix,
+ (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
+ register_prefix, i.op[op].regs->reg_name, i.suffix);
#endif
}
return 1;
{
unsigned int j;
- /* The first operand is implicit and must be xmm0/ymm0. */
+ /* The first operand is implicit and must be xmm0/ymm0/zmm0. */
gas_assert (i.reg_operands
&& (operand_type_equal (&i.types[0], ®xmm)
- || operand_type_equal (&i.types[0], ®ymm)));
+ || operand_type_equal (&i.types[0], ®ymm)
+ || operand_type_equal (&i.types[0], ®zmm)));
if (register_number (i.op[0].regs) != 0)
return bad_implicit_operand (i.types[0].bitfield.regxmm);
/* There are 2 kinds of instructions:
1. 5 operands: 4 register operands or 3 register operands
plus 1 memory operand plus one Vec_Imm4 operand, VexXDS, and
- VexW0 or VexW1. The destination must be either XMM or YMM
- register.
+ VexW0 or VexW1. The destination must be either XMM, YMM or
+ ZMM register.
2. 4 operands: 4 register operands or 3 register operands
plus 1 memory operand, VexXDS, and VexImmExt */
gas_assert ((i.reg_operands == 4
&& (i.tm.opcode_modifier.vexw == VEXW0
|| i.tm.opcode_modifier.vexw == VEXW1)
&& (operand_type_equal (&i.tm.operand_types[dest], ®xmm)
- || operand_type_equal (&i.tm.operand_types[dest], ®ymm)))));
+ || operand_type_equal (&i.tm.operand_types[dest], ®ymm)
+ || operand_type_equal (&i.tm.operand_types[dest], ®zmm)))));
if (i.imm_operands == 0)
{
gas_assert (operand_type_equal (&i.tm.operand_types[reg_slot],
®xmm)
|| operand_type_equal (&i.tm.operand_types[reg_slot],
- ®ymm));
+ ®ymm)
+ || operand_type_equal (&i.tm.operand_types[reg_slot],
+ ®zmm));
exp->X_op = O_constant;
exp->X_add_number = register_number (i.op[reg_slot].regs) << 4;
- }
+ gas_assert ((i.op[reg_slot].regs->reg_flags & RegVRex) == 0);
+ }
else
{
unsigned int imm_slot;
gas_assert (operand_type_equal (&i.tm.operand_types[reg_slot],
®xmm)
|| operand_type_equal (&i.tm.operand_types[reg_slot],
- ®ymm));
+ ®ymm)
+ || operand_type_equal (&i.tm.operand_types[reg_slot],
+ ®zmm));
i.op[imm_slot].imms->X_add_number
|= register_number (i.op[reg_slot].regs) << 4;
+ gas_assert ((i.op[reg_slot].regs->reg_flags & RegVRex) == 0);
}
gas_assert (operand_type_equal (&i.tm.operand_types[nds], ®xmm)
|| operand_type_equal (&i.tm.operand_types[nds],
- ®ymm));
+ ®ymm)
+ || operand_type_equal (&i.tm.operand_types[nds],
+ ®zmm));
i.vex.register_specifier = i.op[nds].regs;
}
else
|| (i.tm.opcode_modifier.vexvvvv == VEXXDS
&& i.imm_operands == 1
&& (i.types[0].bitfield.imm8
- || i.types[i.operands - 1].bitfield.imm8)));
+ || i.types[i.operands - 1].bitfield.imm8
+ || i.rounding)));
if (i.imm_operands == 2)
source = 2;
else
}
break;
case 5:
+ if (i.tm.opcode_modifier.evex)
+ {
+ /* For EVEX instructions, when there are 5 operands, the
+ first one must be immediate operand. If the second one
+ is immediate operand, the source operand is the 3th
+ one. If the last one is immediate operand, the source
+ operand is the 2nd one. */
+ gas_assert (i.imm_operands == 2
+ && i.tm.opcode_modifier.sae
+ && operand_type_check (i.types[0], imm));
+ if (operand_type_check (i.types[1], imm))
+ source = 2;
+ else if (operand_type_check (i.types[4], imm))
+ source = 1;
+ else
+ abort ();
+ }
break;
default:
abort ();
{
dest = source + 1;
+ /* RC/SAE operand could be between DEST and SRC. That happens
+ when one operand is GPR and the other one is XMM/YMM/ZMM
+ register. */
+ if (i.rounding && i.rounding->operand == (int) dest)
+ dest++;
+
if (i.tm.opcode_modifier.vexvvvv == VEXXDS)
{
- /* For instructions with VexNDS, the register-only
- source operand must be 32/64bit integer, XMM or
- YMM register. It is encoded in VEX prefix. We
- need to clear RegMem bit before calling
- operand_type_equal. */
+ /* For instructions with VexNDS, the register-only source
+ operand must be 32/64bit integer, XMM, YMM or ZMM
+ register. It is encoded in VEX prefix. We need to
+ clear RegMem bit before calling operand_type_equal. */
i386_operand_type op;
unsigned int vvvv;
|| (op.bitfield.reg32 != 1
&& !op.bitfield.reg64 != 1
&& !operand_type_equal (&op, ®xmm)
- && !operand_type_equal (&op, ®ymm)))
+ && !operand_type_equal (&op, ®ymm)
+ && !operand_type_equal (&op, ®zmm)
+ && !operand_type_equal (&op, ®mask)))
abort ();
i.vex.register_specifier = i.op[vvvv].regs;
dest++;
i.rm.regmem = i.op[source].regs->reg_num;
if ((i.op[dest].regs->reg_flags & RegRex) != 0)
i.rex |= REX_R;
+ if ((i.op[dest].regs->reg_flags & RegVRex) != 0)
+ i.vrex |= REX_R;
if ((i.op[source].regs->reg_flags & RegRex) != 0)
i.rex |= REX_B;
+ if ((i.op[source].regs->reg_flags & RegVRex) != 0)
+ i.vrex |= REX_B;
}
else
{
i.rm.regmem = i.op[dest].regs->reg_num;
if ((i.op[dest].regs->reg_flags & RegRex) != 0)
i.rex |= REX_B;
+ if ((i.op[dest].regs->reg_flags & RegVRex) != 0)
+ i.vrex |= REX_B;
if ((i.op[source].regs->reg_flags & RegRex) != 0)
i.rex |= REX_R;
+ if ((i.op[source].regs->reg_flags & RegVRex) != 0)
+ i.vrex |= REX_R;
}
if (flag_code != CODE_64BIT && (i.rex & (REX_R | REX_B)))
{
{
i.sib.base = NO_BASE_REGISTER;
i.sib.scale = i.log2_scale_factor;
+ /* No Vec_Disp8 if there is no base. */
+ i.types[op].bitfield.vec_disp8 = 0;
i.types[op].bitfield.disp8 = 0;
i.types[op].bitfield.disp16 = 0;
i.types[op].bitfield.disp64 = 0;
i.sib.index = i.index_reg->reg_num;
if ((i.index_reg->reg_flags & RegRex) != 0)
i.rex |= REX_X;
+ if ((i.index_reg->reg_flags & RegVRex) != 0)
+ i.vrex |= REX_X;
}
default_seg = &ds;
i.sib.base = NO_BASE_REGISTER;
i.sib.scale = i.log2_scale_factor;
i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
+ /* No Vec_Disp8 if there is no base. */
+ i.types[op].bitfield.vec_disp8 = 0;
i.types[op].bitfield.disp8 = 0;
i.types[op].bitfield.disp16 = 0;
i.types[op].bitfield.disp64 = 0;
i.types[op].bitfield.disp32 = 0;
i.types[op].bitfield.disp32s = 1;
i.types[op].bitfield.disp64 = 0;
+ i.types[op].bitfield.vec_disp8 = 0;
i.flags[op] |= Operand_PCrel;
if (! i.disp_operands)
fake_zero_displacement = 1;
if (operand_type_check (i.types[op], disp) == 0)
{
/* fake (%bp) into 0(%bp) */
- i.types[op].bitfield.disp8 = 1;
+ if (i.tm.operand_types[op].bitfield.vec_disp8)
+ i.types[op].bitfield.vec_disp8 = 1;
+ else
+ i.types[op].bitfield.disp8 = 1;
fake_zero_displacement = 1;
}
}
i386_operand_type temp;
operand_type_set (&temp, 0);
temp.bitfield.disp8 = i.types[op].bitfield.disp8;
+ temp.bitfield.vec_disp8
+ = i.types[op].bitfield.vec_disp8;
i.types[op] = temp;
if (i.prefix[ADDR_PREFIX] == 0)
i.types[op].bitfield.disp32s = 1;
if (i.base_reg->reg_num == 5 && i.disp_operands == 0)
{
fake_zero_displacement = 1;
- i.types[op].bitfield.disp8 = 1;
+ if (i.tm.operand_types [op].bitfield.vec_disp8)
+ i.types[op].bitfield.vec_disp8 = 1;
+ else
+ i.types[op].bitfield.disp8 = 1;
}
i.sib.scale = i.log2_scale_factor;
if (i.index_reg == 0)
|| i.types[op].bitfield.regmmx
|| i.types[op].bitfield.regxmm
|| i.types[op].bitfield.regymm
+ || i.types[op].bitfield.regbnd
+ || i.types[op].bitfield.regzmm
+ || i.types[op].bitfield.regmask
|| i.types[op].bitfield.sreg2
|| i.types[op].bitfield.sreg3
|| i.types[op].bitfield.control
if (type->bitfield.reg32 != 1
&& type->bitfield.reg64 != 1
&& !operand_type_equal (type, ®xmm)
- && !operand_type_equal (type, ®ymm))
+ && !operand_type_equal (type, ®ymm)
+ && !operand_type_equal (type, ®zmm)
+ && !operand_type_equal (type, ®mask))
abort ();
i.vex.register_specifier = i.op[vex_reg].regs;
i.rm.regmem = i.op[op].regs->reg_num;
if ((i.op[op].regs->reg_flags & RegRex) != 0)
i.rex |= REX_B;
+ if ((i.op[op].regs->reg_flags & RegVRex) != 0)
+ i.vrex |= REX_B;
}
else
{
i.rm.reg = i.op[op].regs->reg_num;
if ((i.op[op].regs->reg_flags & RegRex) != 0)
i.rex |= REX_R;
+ if ((i.op[op].regs->reg_flags & RegVRex) != 0)
+ i.vrex |= REX_R;
}
}
i.prefixes--;
}
+ /* BND prefixed jump. */
+ if (i.prefix[BND_PREFIX] != 0)
+ {
+ FRAG_APPEND_1_CHAR (i.prefix[BND_PREFIX]);
+ i.prefixes -= 1;
+ }
+
if (i.prefixes != 0 && !intel_syntax)
as_warn (_("skipping prefixes on this instruction"));
/* 1 possible extra opcode + 4 byte displacement go in var part.
Pass reloc in fr_var. */
- frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
+ frag_var (rs_machine_dependent, 5,
+ ((!object_64bit
+ || i.reloc[0] != NO_RELOC
+ || (i.bnd_prefix == NULL && !add_bnd_prefix))
+ ? i.reloc[0]
+ : BFD_RELOC_X86_64_PC32_BND),
+ subtype, sym, off, p);
}
static void
i.prefixes -= 1;
}
+ /* BND prefixed jump. */
+ if (i.prefix[BND_PREFIX] != 0)
+ {
+ FRAG_APPEND_1_CHAR (i.prefix[BND_PREFIX]);
+ i.prefixes -= 1;
+ }
+
if (i.prefixes != 0 && !intel_syntax)
as_warn (_("skipping prefixes on this instruction"));
}
fixP = fix_new_exp (frag_now, p - frag_now->fr_literal, size,
- i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
+ i.op[0].disps, 1, reloc (size, 1, 1,
+ (i.bnd_prefix != NULL
+ || add_bnd_prefix),
+ i.reloc[0]));
/* All jumps handled here are signed, but don't use a signed limit
check for 32 and 16 bit jumps as we want to allow wrap around at
}
else
fix_new_exp (frag_now, p - frag_now->fr_literal, size,
- i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
+ i.op[1].imms, 0, reloc (size, 0, 0, 0, i.reloc[1]));
if (i.op[0].imms->X_op != O_constant)
as_bad (_("can't handle non absolute segment in `%s'"),
i.tm.name);
unsigned int j;
unsigned int prefix;
- /* Since the VEX prefix contains the implicit prefix, we don't
- need the explicit prefix. */
- if (!i.tm.opcode_modifier.vex)
+ /* Since the VEX/EVEX prefix contains the implicit prefix, we
+ don't need the explicit prefix. */
+ if (!i.tm.opcode_modifier.vex && !i.tm.opcode_modifier.evex)
{
switch (i.tm.opcode_length)
{
abort ();
}
+ /* For EVEX instructions i.vrex should become 0 after
+ build_evex_prefix. For VEX instructions upper 16 registers
+ aren't available, so VREX should be 0. */
+ if (i.vrex)
+ abort ();
/* Now the VEX prefix. */
p = frag_more (i.vex.length);
for (j = 0; j < i.vex.length; j++)
{
switch (i.tm.opcode_length)
{
+ case 4:
+ p = frag_more (4);
+ *p++ = (i.tm.base_opcode >> 24) & 0xff;
+ *p++ = (i.tm.base_opcode >> 16) & 0xff;
+ break;
case 3:
p = frag_more (3);
*p++ = (i.tm.base_opcode >> 16) & 0xff;
disp_size (unsigned int n)
{
int size = 4;
- if (i.types[n].bitfield.disp64)
+
+ /* Vec_Disp8 has to be 8bit. */
+ if (i.types[n].bitfield.vec_disp8)
+ size = 1;
+ else if (i.types[n].bitfield.disp64)
size = 8;
else if (i.types[n].bitfield.disp8)
size = 1;
for (n = 0; n < i.operands; n++)
{
- if (operand_type_check (i.types[n], disp))
+ if (i.types[n].bitfield.vec_disp8
+ || operand_type_check (i.types[n], disp))
{
if (i.op[n].disps->X_op == O_constant)
{
int size = disp_size (n);
- offsetT val;
+ offsetT val = i.op[n].disps->X_add_number;
- val = offset_in_range (i.op[n].disps->X_add_number,
- size);
+ if (i.types[n].bitfield.vec_disp8)
+ val >>= i.memshift;
+ val = offset_in_range (val, size);
p = frag_more (size);
md_number_to_chars (p, val, size);
}
}
p = frag_more (size);
- reloc_type = reloc (size, pcrel, sign, i.reloc[n]);
+ reloc_type = reloc (size, pcrel, sign,
+ (i.bnd_prefix != NULL
+ || add_bnd_prefix),
+ i.reloc[n]);
if (GOT_symbol
&& GOT_symbol == i.op[n].disps->X_add_symbol
&& (((reloc_type == BFD_RELOC_32
for (n = 0; n < i.operands; n++)
{
+ /* Skip SAE/RC Imm operand in EVEX. They are already handled. */
+ if (i.rounding && (int) n == i.rounding->operand)
+ continue;
+
if (operand_type_check (i.types[n], imm))
{
if (i.op[n].imms->X_op == O_constant)
sign = 0;
p = frag_more (size);
- reloc_type = reloc (size, 0, sign, i.reloc[n]);
+ reloc_type = reloc (size, 0, sign, 0, i.reloc[n]);
/* This is tough to explain. We end up with this one if we
* have operands that look like
x86_cons_fix_new (fragS *frag, unsigned int off, unsigned int len,
expressionS *exp)
{
- enum bfd_reloc_code_real r = reloc (len, 0, cons_sign, got_reloc);
+ enum bfd_reloc_code_real r = reloc (len, 0, cons_sign, 0, got_reloc);
got_reloc = NO_RELOC;
#if !(defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined (OBJ_MACH_O)) \
|| defined (LEX_AT)
-# define lex_got(reloc, adjust, types) NULL
+# define lex_got(reloc, adjust, types, bnd_prefix) NULL
#else
/* Parse operands of the form
<symbol>@GOTOFF+<nnn>
static char *
lex_got (enum bfd_reloc_code_real *rel,
int *adjust,
- i386_operand_type *types)
+ i386_operand_type *types,
+ int bnd_prefix)
{
/* Some of the relocations depend on the size of what field is to
be relocated. But in our callers i386_immediate and i386_displacement
const enum bfd_reloc_code_real rel[2];
const i386_operand_type types64;
} gotrel[] = {
+#if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
+ { STRING_COMMA_LEN ("SIZE"), { BFD_RELOC_SIZE32,
+ BFD_RELOC_SIZE32 },
+ OPERAND_TYPE_IMM32_64 },
+#endif
{ STRING_COMMA_LEN ("PLTOFF"), { _dummy_first_bfd_reloc_code_real,
BFD_RELOC_X86_64_PLTOFF64 },
OPERAND_TYPE_IMM64 },
char *tmpbuf, *past_reloc;
*rel = gotrel[j].rel[object_64bit];
- if (adjust)
- *adjust = len;
if (types)
{
*types = gotrel[j].types64;
}
- if (GOT_symbol == NULL)
+ if (j != 0 && GOT_symbol == NULL)
GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
/* The length of the first part of our input line. */
/* Replace the relocation token with ' ', so that
errors like foo@GOTOFF1 will be detected. */
tmpbuf[first++] = ' ';
+ else
+ /* Increment length by 1 if the relocation token is
+ removed. */
+ len++;
+ if (adjust)
+ *adjust = len;
memcpy (tmpbuf + first, past_reloc, second);
tmpbuf[first + second] = '\0';
+ if (bnd_prefix && *rel == BFD_RELOC_X86_64_PLT32)
+ *rel = BFD_RELOC_X86_64_PLT32_BND;
return tmpbuf;
}
input string, minus the `@SECREL32' into a malloc'd buffer for
parsing by the calling routine. Return this buffer, and if ADJUST
is non-null set it to the length of the string we removed from the
- input line. Otherwise return NULL.
-
+ input line. Otherwise return NULL.
+
This function is copied from the ELF version above adjusted for PE targets. */
static char *
lex_got (enum bfd_reloc_code_real *rel ATTRIBUTE_UNUSED,
int *adjust ATTRIBUTE_UNUSED,
- i386_operand_type *types ATTRIBUTE_UNUSED)
+ i386_operand_type *types,
+ int bnd_prefix ATTRIBUTE_UNUSED)
{
static const struct
{
int adjust = 0;
save = input_line_pointer;
- gotfree_input_line = lex_got (&got_reloc, &adjust, NULL);
+ gotfree_input_line = lex_got (&got_reloc, &adjust, NULL, 0);
if (gotfree_input_line)
input_line_pointer = gotfree_input_line;
}
#endif
+/* Handle Vector operations. */
+
+static char *
+check_VecOperations (char *op_string, char *op_end)
+{
+ const reg_entry *mask;
+ const char *saved;
+ char *end_op;
+
+ while (*op_string
+ && (op_end == NULL || op_string < op_end))
+ {
+ saved = op_string;
+ if (*op_string == '{')
+ {
+ op_string++;
+
+ /* Check broadcasts. */
+ if (strncmp (op_string, "1to", 3) == 0)
+ {
+ int bcst_type;
+
+ if (i.broadcast)
+ goto duplicated_vec_op;
+
+ op_string += 3;
+ if (*op_string == '8')
+ bcst_type = BROADCAST_1TO8;
+ else if (*op_string == '1'
+ && *(op_string+1) == '6')
+ {
+ bcst_type = BROADCAST_1TO16;
+ op_string++;
+ }
+ else
+ {
+ as_bad (_("Unsupported broadcast: `%s'"), saved);
+ return NULL;
+ }
+ op_string++;
+
+ broadcast_op.type = bcst_type;
+ broadcast_op.operand = this_operand;
+ i.broadcast = &broadcast_op;
+ }
+ /* Check masking operation. */
+ else if ((mask = parse_register (op_string, &end_op)) != NULL)
+ {
+ /* k0 can't be used for write mask. */
+ if (mask->reg_num == 0)
+ {
+ as_bad (_("`%s' can't be used for write mask"),
+ op_string);
+ return NULL;
+ }
+
+ if (!i.mask)
+ {
+ mask_op.mask = mask;
+ mask_op.zeroing = 0;
+ mask_op.operand = this_operand;
+ i.mask = &mask_op;
+ }
+ else
+ {
+ if (i.mask->mask)
+ goto duplicated_vec_op;
+
+ i.mask->mask = mask;
+
+ /* Only "{z}" is allowed here. No need to check
+ zeroing mask explicitly. */
+ if (i.mask->operand != this_operand)
+ {
+ as_bad (_("invalid write mask `%s'"), saved);
+ return NULL;
+ }
+ }
+
+ op_string = end_op;
+ }
+ /* Check zeroing-flag for masking operation. */
+ else if (*op_string == 'z')
+ {
+ if (!i.mask)
+ {
+ mask_op.mask = NULL;
+ mask_op.zeroing = 1;
+ mask_op.operand = this_operand;
+ i.mask = &mask_op;
+ }
+ else
+ {
+ if (i.mask->zeroing)
+ {
+ duplicated_vec_op:
+ as_bad (_("duplicated `%s'"), saved);
+ return NULL;
+ }
+
+ i.mask->zeroing = 1;
+
+ /* Only "{%k}" is allowed here. No need to check mask
+ register explicitly. */
+ if (i.mask->operand != this_operand)
+ {
+ as_bad (_("invalid zeroing-masking `%s'"),
+ saved);
+ return NULL;
+ }
+ }
+
+ op_string++;
+ }
+ else
+ goto unknown_vec_op;
+
+ if (*op_string != '}')
+ {
+ as_bad (_("missing `}' in `%s'"), saved);
+ return NULL;
+ }
+ op_string++;
+ continue;
+ }
+ unknown_vec_op:
+ /* We don't know this one. */
+ as_bad (_("unknown vector operation: `%s'"), saved);
+ return NULL;
+ }
+
+ return op_string;
+}
+
static int
i386_immediate (char *imm_start)
{
save_input_line_pointer = input_line_pointer;
input_line_pointer = imm_start;
- gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
+ gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types,
+ (i.bnd_prefix != NULL
+ || add_bnd_prefix));
if (gotfree_input_line)
input_line_pointer = gotfree_input_line;
exp_seg = expression (exp);
SKIP_WHITESPACE ();
+
+ /* Handle vector operations. */
+ if (*input_line_pointer == '{')
+ {
+ input_line_pointer = check_VecOperations (input_line_pointer,
+ NULL);
+ if (input_line_pointer == NULL)
+ return 0;
+ }
+
if (*input_line_pointer)
as_bad (_("junk `%s' after expression"), input_line_pointer);
*displacement_string_end = '0';
}
#endif
- gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
+ gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types,
+ (i.bnd_prefix != NULL
+ || add_bnd_prefix));
if (gotfree_input_line)
input_line_pointer = gotfree_input_line;
static int
i386_index_check (const char *operand_string)
{
- int ok;
const char *kind = "base/index";
+ enum flag_code addr_mode;
+
+ if (i.prefix[ADDR_PREFIX])
+ addr_mode = flag_code == CODE_32BIT ? CODE_16BIT : CODE_32BIT;
+ else
+ {
+ addr_mode = flag_code;
+
#if INFER_ADDR_PREFIX
- int fudged = 0;
+ if (i.mem_operands == 0)
+ {
+ /* Infer address prefix from the first memory operand. */
+ const reg_entry *addr_reg = i.base_reg;
- tryprefix:
+ if (addr_reg == NULL)
+ addr_reg = i.index_reg;
+
+ if (addr_reg)
+ {
+ if (addr_reg->reg_num == RegEip
+ || addr_reg->reg_num == RegEiz
+ || addr_reg->reg_type.bitfield.reg32)
+ addr_mode = CODE_32BIT;
+ else if (flag_code != CODE_64BIT
+ && addr_reg->reg_type.bitfield.reg16)
+ addr_mode = CODE_16BIT;
+
+ if (addr_mode != flag_code)
+ {
+ i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
+ i.prefixes += 1;
+ /* Change the size of any displacement too. At most one
+ of Disp16 or Disp32 is set.
+ FIXME. There doesn't seem to be any real need for
+ separate Disp16 and Disp32 flags. The same goes for
+ Imm16 and Imm32. Removing them would probably clean
+ up the code quite a lot. */
+ if (flag_code != CODE_64BIT
+ && (i.types[this_operand].bitfield.disp16
+ || i.types[this_operand].bitfield.disp32))
+ i.types[this_operand]
+ = operand_type_xor (i.types[this_operand], disp16_32);
+ }
+ }
+ }
#endif
- ok = 1;
+ }
+
if (current_templates->start->opcode_modifier.isstring
&& !current_templates->start->opcode_modifier.immext
&& (current_templates->end[-1].opcode_modifier.isstring
{
/* Memory operands of string insns are special in that they only allow
a single register (rDI, rSI, or rBX) as their memory address. */
- unsigned int expected;
+ const reg_entry *expected_reg;
+ static const char *di_si[][2] =
+ {
+ { "esi", "edi" },
+ { "si", "di" },
+ { "rsi", "rdi" }
+ };
+ static const char *bx[] = { "ebx", "bx", "rbx" };
kind = "string address";
&& current_templates->end[-1].operand_types[1]
.bitfield.baseindex))
type = current_templates->end[-1].operand_types[1];
- expected = type.bitfield.esseg ? 7 /* rDI */ : 6 /* rSI */;
+ expected_reg = hash_find (reg_hash,
+ di_si[addr_mode][type.bitfield.esseg]);
+
}
else
- expected = 3 /* rBX */;
+ expected_reg = hash_find (reg_hash, bx[addr_mode]);
- if (!i.base_reg || i.index_reg
+ if (i.base_reg != expected_reg
+ || i.index_reg
|| operand_type_check (i.types[this_operand], disp))
- ok = -1;
- else if (!(flag_code == CODE_64BIT
- ? i.prefix[ADDR_PREFIX]
- ? i.base_reg->reg_type.bitfield.reg32
- : i.base_reg->reg_type.bitfield.reg64
- : (flag_code == CODE_16BIT) ^ !i.prefix[ADDR_PREFIX]
- ? i.base_reg->reg_type.bitfield.reg32
- : i.base_reg->reg_type.bitfield.reg16))
- ok = 0;
- else if (register_number (i.base_reg) != expected)
- ok = -1;
-
- if (ok < 0)
- {
- unsigned int j;
-
- for (j = 0; j < i386_regtab_size; ++j)
- if ((flag_code == CODE_64BIT
- ? i.prefix[ADDR_PREFIX]
- ? i386_regtab[j].reg_type.bitfield.reg32
- : i386_regtab[j].reg_type.bitfield.reg64
- : (flag_code == CODE_16BIT) ^ !i.prefix[ADDR_PREFIX]
- ? i386_regtab[j].reg_type.bitfield.reg32
- : i386_regtab[j].reg_type.bitfield.reg16)
- && register_number(i386_regtab + j) == expected)
- break;
- gas_assert (j < i386_regtab_size);
+ {
+ /* The second memory operand must have the same size as
+ the first one. */
+ if (i.mem_operands
+ && i.base_reg
+ && !((addr_mode == CODE_64BIT
+ && i.base_reg->reg_type.bitfield.reg64)
+ || (addr_mode == CODE_32BIT
+ ? i.base_reg->reg_type.bitfield.reg32
+ : i.base_reg->reg_type.bitfield.reg16)))
+ goto bad_address;
+
as_warn (_("`%s' is not valid here (expected `%c%s%s%c')"),
operand_string,
intel_syntax ? '[' : '(',
register_prefix,
- i386_regtab[j].reg_name,
+ expected_reg->reg_name,
intel_syntax ? ']' : ')');
- ok = 1;
- }
- }
- else if (flag_code == CODE_64BIT)
- {
- if ((i.base_reg
- && ((i.prefix[ADDR_PREFIX] == 0
- && !i.base_reg->reg_type.bitfield.reg64)
- || (i.prefix[ADDR_PREFIX]
- && !i.base_reg->reg_type.bitfield.reg32))
- && (i.index_reg
- || i.base_reg->reg_num !=
- (i.prefix[ADDR_PREFIX] == 0 ? RegRip : RegEip)))
- || (i.index_reg
- && !(i.index_reg->reg_type.bitfield.regxmm
- || i.index_reg->reg_type.bitfield.regymm)
- && (!i.index_reg->reg_type.bitfield.baseindex
- || (i.prefix[ADDR_PREFIX] == 0
- && i.index_reg->reg_num != RegRiz
- && !i.index_reg->reg_type.bitfield.reg64
- )
- || (i.prefix[ADDR_PREFIX]
- && i.index_reg->reg_num != RegEiz
- && !i.index_reg->reg_type.bitfield.reg32))))
- ok = 0;
+ return 1;
+ }
+ else
+ return 1;
+
+bad_address:
+ as_bad (_("`%s' is not a valid %s expression"),
+ operand_string, kind);
+ return 0;
}
else
{
- if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
+ if (addr_mode != CODE_16BIT)
+ {
+ /* 32-bit/64-bit checks. */
+ if ((i.base_reg
+ && (addr_mode == CODE_64BIT
+ ? !i.base_reg->reg_type.bitfield.reg64
+ : !i.base_reg->reg_type.bitfield.reg32)
+ && (i.index_reg
+ || (i.base_reg->reg_num
+ != (addr_mode == CODE_64BIT ? RegRip : RegEip))))
+ || (i.index_reg
+ && !i.index_reg->reg_type.bitfield.regxmm
+ && !i.index_reg->reg_type.bitfield.regymm
+ && !i.index_reg->reg_type.bitfield.regzmm
+ && ((addr_mode == CODE_64BIT
+ ? !(i.index_reg->reg_type.bitfield.reg64
+ || i.index_reg->reg_num == RegRiz)
+ : !(i.index_reg->reg_type.bitfield.reg32
+ || i.index_reg->reg_num == RegEiz))
+ || !i.index_reg->reg_type.bitfield.baseindex)))
+ goto bad_address;
+ }
+ else
{
- /* 16bit checks. */
+ /* 16-bit checks. */
if ((i.base_reg
&& (!i.base_reg->reg_type.bitfield.reg16
|| !i.base_reg->reg_type.bitfield.baseindex))
&& i.base_reg->reg_num < 6
&& i.index_reg->reg_num >= 6
&& i.log2_scale_factor == 0))))
- ok = 0;
+ goto bad_address;
}
- else
+ }
+ return 1;
+}
+
+/* Handle vector immediates. */
+
+static int
+RC_SAE_immediate (const char *imm_start)
+{
+ unsigned int match_found, j;
+ const char *pstr = imm_start;
+ expressionS *exp;
+
+ if (*pstr != '{')
+ return 0;
+
+ pstr++;
+ match_found = 0;
+ for (j = 0; j < ARRAY_SIZE (RC_NamesTable); j++)
+ {
+ if (!strncmp (pstr, RC_NamesTable[j].name, RC_NamesTable[j].len))
{
- /* 32bit checks. */
- if ((i.base_reg
- && !i.base_reg->reg_type.bitfield.reg32)
- || (i.index_reg
- && !i.index_reg->reg_type.bitfield.regxmm
- && !i.index_reg->reg_type.bitfield.regymm
- && ((!i.index_reg->reg_type.bitfield.reg32
- && i.index_reg->reg_num != RegEiz)
- || !i.index_reg->reg_type.bitfield.baseindex)))
- ok = 0;
+ if (!i.rounding)
+ {
+ rc_op.type = RC_NamesTable[j].type;
+ rc_op.operand = this_operand;
+ i.rounding = &rc_op;
+ }
+ else
+ {
+ as_bad (_("duplicated `%s'"), imm_start);
+ return 0;
+ }
+ pstr += RC_NamesTable[j].len;
+ match_found = 1;
+ break;
}
}
- if (!ok)
+ if (!match_found)
+ return 0;
+
+ if (*pstr++ != '}')
{
-#if INFER_ADDR_PREFIX
- if (!i.mem_operands && !i.prefix[ADDR_PREFIX])
- {
- i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
- i.prefixes += 1;
- /* Change the size of any displacement too. At most one of
- Disp16 or Disp32 is set.
- FIXME. There doesn't seem to be any real need for separate
- Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
- Removing them would probably clean up the code quite a lot. */
- if (flag_code != CODE_64BIT
- && (i.types[this_operand].bitfield.disp16
- || i.types[this_operand].bitfield.disp32))
- i.types[this_operand]
- = operand_type_xor (i.types[this_operand], disp16_32);
- fudged = 1;
- goto tryprefix;
- }
- if (fudged)
- as_bad (_("`%s' is not a valid %s expression"),
- operand_string,
- kind);
- else
-#endif
- as_bad (_("`%s' is not a valid %s-bit %s expression"),
- operand_string,
- flag_code_names[i.prefix[ADDR_PREFIX]
- ? flag_code == CODE_32BIT
- ? CODE_16BIT
- : CODE_32BIT
- : flag_code],
- kind);
+ as_bad (_("Missing '}': '%s'"), imm_start);
+ return 0;
}
- return ok;
+ /* RC/SAE immediate string should contain nothing more. */;
+ if (*pstr != 0)
+ {
+ as_bad (_("Junk after '}': '%s'"), imm_start);
+ return 0;
+ }
+
+ exp = &im_expressions[i.imm_operands++];
+ i.op[this_operand].imms = exp;
+
+ exp->X_op = O_constant;
+ exp->X_add_number = 0;
+ exp->X_add_symbol = (symbolS *) 0;
+ exp->X_op_symbol = (symbolS *) 0;
+
+ i.types[this_operand].bitfield.imm8 = 1;
+ return 1;
}
/* Parse OPERAND_STRING into the i386_insn structure I. Returns zero
}
goto do_memory_reference;
}
+
+ /* Handle vector operations. */
+ if (*op_string == '{')
+ {
+ op_string = check_VecOperations (op_string, NULL);
+ if (op_string == NULL)
+ return 0;
+ }
+
if (*op_string)
{
as_bad (_("junk `%s' after register"), op_string);
if (!i386_immediate (op_string))
return 0;
}
+ else if (RC_SAE_immediate (operand_string))
+ {
+ /* If it is a RC or SAE immediate, do nothing. */
+ ;
+ }
else if (is_digit_char (*op_string)
|| is_identifier_char (*op_string)
|| *op_string == '(')
/* Start and end of displacement string expression (if found). */
char *displacement_string_start;
char *displacement_string_end;
+ char *vop_start;
do_memory_reference:
if ((i.mem_operands == 1
after the '('. */
base_string = op_string + strlen (op_string);
+ /* Handle vector operations. */
+ vop_start = strchr (op_string, '{');
+ if (vop_start && vop_start < base_string)
+ {
+ if (check_VecOperations (vop_start, base_string) == NULL)
+ return 0;
+ base_string = vop_start;
+ }
+
--base_string;
if (is_space_char (*base_string))
--base_string;
&& (fixP->fx_r_type == BFD_RELOC_32_PCREL
|| fixP->fx_r_type == BFD_RELOC_64_PCREL
|| fixP->fx_r_type == BFD_RELOC_16_PCREL
- || fixP->fx_r_type == BFD_RELOC_8_PCREL)
+ || fixP->fx_r_type == BFD_RELOC_8_PCREL
+ || fixP->fx_r_type == BFD_RELOC_X86_64_PC32_BND)
&& !use_rela_relocations)
{
/* This is a hack. There should be a better way to handle this.
{
case BFD_RELOC_386_PLT32:
case BFD_RELOC_X86_64_PLT32:
+ case BFD_RELOC_X86_64_PLT32_BND:
/* Make the jump instruction point to the address of the operand. At
runtime we merely add the offset to the actual PLT entry. */
value = -4;
if (r->reg_type.bitfield.regymm && !cpu_arch_flags.bitfield.cpuavx)
return (const reg_entry *) NULL;
+ if ((r->reg_type.bitfield.regzmm || r->reg_type.bitfield.regmask)
+ && !cpu_arch_flags.bitfield.cpuavx512f)
+ return (const reg_entry *) NULL;
+
/* Don't allow fake index register unless allow_index_reg isn't 0. */
if (!allow_index_reg
&& (r->reg_num == RegEiz || r->reg_num == RegRiz))
return (const reg_entry *) NULL;
+ /* Upper 16 vector register is only available with VREX in 64bit
+ mode. */
+ if ((r->reg_flags & RegVRex))
+ {
+ if (!cpu_arch_flags.bitfield.cpuvrex
+ || flag_code != CODE_64BIT)
+ return (const reg_entry *) NULL;
+
+ i.need_vrex = 1;
+ }
+
if (((r->reg_flags & (RegRex64 | RegRex))
|| r->reg_type.bitfield.reg64)
&& (!cpu_arch_flags.bitfield.cpulm
#define OPTION_MOPERAND_CHECK (OPTION_MD_BASE + 12)
#define OPTION_MAVXSCALAR (OPTION_MD_BASE + 13)
#define OPTION_X32 (OPTION_MD_BASE + 14)
+#define OPTION_MADD_BND_PREFIX (OPTION_MD_BASE + 15)
+#define OPTION_MEVEXLIG (OPTION_MD_BASE + 16)
+#define OPTION_MEVEXWIG (OPTION_MD_BASE + 17)
struct option md_longopts[] =
{
{"msse-check", required_argument, NULL, OPTION_MSSE_CHECK},
{"moperand-check", required_argument, NULL, OPTION_MOPERAND_CHECK},
{"mavxscalar", required_argument, NULL, OPTION_MAVXSCALAR},
+ {"madd-bnd-prefix", no_argument, NULL, OPTION_MADD_BND_PREFIX},
+ {"mevexlig", required_argument, NULL, OPTION_MEVEXLIG},
+ {"mevexwig", required_argument, NULL, OPTION_MEVEXWIG},
{NULL, no_argument, NULL, 0}
};
size_t md_longopts_size = sizeof (md_longopts);
as_fatal (_("invalid -mavxscalar= option: `%s'"), arg);
break;
+ case OPTION_MADD_BND_PREFIX:
+ add_bnd_prefix = 1;
+ break;
+
+ case OPTION_MEVEXLIG:
+ if (strcmp (arg, "128") == 0)
+ evexlig = evexl128;
+ else if (strcmp (arg, "256") == 0)
+ evexlig = evexl256;
+ else if (strcmp (arg, "512") == 0)
+ evexlig = evexl512;
+ else
+ as_fatal (_("invalid -mevexlig= option: `%s'"), arg);
+ break;
+
+ case OPTION_MEVEXWIG:
+ if (strcmp (arg, "0") == 0)
+ evexwig = evexw0;
+ else if (strcmp (arg, "1") == 0)
+ evexwig = evexw1;
+ else
+ as_fatal (_("invalid -mevexwig= option: `%s'"), arg);
+ break;
+
default:
return 0;
}
-mavxscalar=[128|256] encode scalar AVX instructions with specific vector\n\
length\n"));
fprintf (stream, _("\
+ -mevexlig=[128|256|512] encode scalar EVEX instructions with specific vector\n\
+ length\n"));
+ fprintf (stream, _("\
+ -mevexwig=[0|1] encode EVEX instructions with specific EVEX.W value\n\
+ for EVEX.W bit ignored instructions\n"));
+ fprintf (stream, _("\
-mmnemonic=[att|intel] use AT&T/Intel mnemonic\n"));
fprintf (stream, _("\
-msyntax=[att|intel] use AT&T/Intel syntax\n"));
-mnaked-reg don't require `%%' prefix for registers\n"));
fprintf (stream, _("\
-mold-gcc support old (<= 2.8.1) versions of gcc\n"));
+ fprintf (stream, _("\
+ -madd-bnd-prefix add BND prefix for all valid branches\n"));
}
#if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
switch (fixp->fx_r_type)
{
+#if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
+ case BFD_RELOC_SIZE32:
+ case BFD_RELOC_SIZE64:
+ if (S_IS_DEFINED (fixp->fx_addsy)
+ && !S_IS_EXTERNAL (fixp->fx_addsy))
+ {
+ /* Resolve size relocation against local symbol to size of
+ the symbol plus addend. */
+ valueT value = S_GET_SIZE (fixp->fx_addsy) + fixp->fx_offset;
+ if (fixp->fx_r_type == BFD_RELOC_SIZE32
+ && !fits_in_unsigned_long (value))
+ as_bad_where (fixp->fx_file, fixp->fx_line,
+ _("symbol size computation overflow"));
+ fixp->fx_addsy = NULL;
+ fixp->fx_subsy = NULL;
+ md_apply_fix (fixp, (valueT *) &value, NULL);
+ return NULL;
+ }
+#endif
+
case BFD_RELOC_X86_64_PLT32:
+ case BFD_RELOC_X86_64_PLT32_BND:
case BFD_RELOC_X86_64_GOT32:
case BFD_RELOC_X86_64_GOTPCREL:
case BFD_RELOC_386_PLT32:
break;
case 1: code = BFD_RELOC_8_PCREL; break;
case 2: code = BFD_RELOC_16_PCREL; break;
- case 4: code = BFD_RELOC_32_PCREL; break;
+ case 4:
+ code = (fixp->fx_r_type == BFD_RELOC_X86_64_PC32_BND
+ ? fixp-> fx_r_type : BFD_RELOC_32_PCREL);
+ break;
#ifdef BFD64
case 8: code = BFD_RELOC_64_PCREL; break;
#endif
switch (code)
{
case BFD_RELOC_X86_64_PLT32:
+ case BFD_RELOC_X86_64_PLT32_BND:
case BFD_RELOC_X86_64_GOT32:
case BFD_RELOC_X86_64_GOTPCREL:
case BFD_RELOC_X86_64_TLSGD: