/* 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
+ 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009
Free Software Foundation, Inc.
This file is part of GAS, the GNU Assembler.
#define LONG_MNEM_SUFFIX 'l'
#define QWORD_MNEM_SUFFIX 'q'
#define XMMWORD_MNEM_SUFFIX 'x'
+#define YMMWORD_MNEM_SUFFIX 'y'
/* Intel Syntax. Use a non-ascii letter since since it never appears
in instructions. */
#define LONG_DOUBLE_MNEM_SUFFIX '\1'
}
sib_byte;
-enum processor_type
-{
- PROCESSOR_UNKNOWN,
- PROCESSOR_I386,
- PROCESSOR_I486,
- PROCESSOR_PENTIUM,
- PROCESSOR_PENTIUMPRO,
- PROCESSOR_PENTIUM4,
- PROCESSOR_NOCONA,
- PROCESSOR_CORE,
- PROCESSOR_CORE2,
- PROCESSOR_K6,
- PROCESSOR_ATHLON,
- PROCESSOR_K8,
- PROCESSOR_GENERIC32,
- PROCESSOR_GENERIC64,
- PROCESSOR_AMDFAM10
-};
-
/* x86 arch names, types and features */
typedef struct
{
static void set_intel_syntax (int);
static void set_intel_mnemonic (int);
static void set_allow_index_reg (int);
+static void set_sse_check (int);
static void set_cpu_arch (int);
#ifdef TE_PE
static void pe_directive_secrel (int);
static void swap_2_operands (int, int);
static void optimize_imm (void);
static void optimize_disp (void);
-static int match_template (void);
+static const template *match_template (void);
static int check_string (void);
static int process_suffix (void);
static int check_byte_reg (void);
static const char *default_arch = DEFAULT_ARCH;
+/* VEX prefix. */
+typedef struct
+{
+ /* VEX prefix is either 2 byte or 3 byte. */
+ unsigned char bytes[3];
+ unsigned int length;
+ /* Destination or source register specifier. */
+ const reg_entry *register_specifier;
+} vex_prefix;
+
/* 'md_assemble ()' gathers together information and puts it into a
i386_insn. */
rex_byte rex;
sib_byte sib;
drex_byte drex;
+ vex_prefix vex;
+
+ /* Swap operand in encoding. */
+ unsigned int swap_operand : 1;
};
typedef struct _i386_insn i386_insn;
/* 1 if support old (<= 2.8.1) versions of gcc. */
static int old_gcc = OLDGCC_COMPAT;
+/* 1 if pseudo registers are permitted. */
+static int allow_pseudo_reg = 0;
+
/* 1 if register prefix % not required. */
static int allow_naked_reg = 0;
/* 1 if pseudo index register, eiz/riz, is allowed . */
static int allow_index_reg = 0;
+static enum
+ {
+ sse_check_none = 0,
+ sse_check_warning,
+ sse_check_error
+ }
+sse_check;
+
/* Register prefix used for error message. */
static const char *register_prefix = "%";
static int cpu_arch_tune_set = 0;
/* Cpu we are generating instructions for. */
-static enum processor_type cpu_arch_tune = PROCESSOR_UNKNOWN;
+enum processor_type cpu_arch_tune = PROCESSOR_UNKNOWN;
/* CPU feature flags of cpu we are generating instructions for. */
static i386_cpu_flags cpu_arch_tune_flags;
/* CPU instruction set architecture used. */
-static enum processor_type cpu_arch_isa = PROCESSOR_UNKNOWN;
+enum processor_type cpu_arch_isa = PROCESSOR_UNKNOWN;
/* CPU feature flags of instruction set architecture used. */
-static i386_cpu_flags cpu_arch_isa_flags;
+i386_cpu_flags cpu_arch_isa_flags;
/* If set, conditional jumps are not automatically promoted to handle
larger than a byte offset. */
static unsigned int no_cond_jump_promotion = 0;
+/* Encode SSE instructions with VEX prefix. */
+static unsigned int sse2avx;
+
/* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
static symbolS *GOT_symbol;
CPU_CORE2_FLAGS },
{ "core2", PROCESSOR_CORE2,
CPU_CORE2_FLAGS },
+ { "corei7", PROCESSOR_COREI7,
+ CPU_COREI7_FLAGS },
{ "k6", PROCESSOR_K6,
CPU_K6_FLAGS },
{ "k6_2", PROCESSOR_K6,
CPU_SSE4_2_FLAGS },
{ ".sse4", PROCESSOR_UNKNOWN,
CPU_SSE4_2_FLAGS },
+ { ".avx", PROCESSOR_UNKNOWN,
+ CPU_AVX_FLAGS },
{ ".vmx", PROCESSOR_UNKNOWN,
CPU_VMX_FLAGS },
{ ".smx", PROCESSOR_UNKNOWN,
CPU_SMX_FLAGS },
{ ".xsave", PROCESSOR_UNKNOWN,
CPU_XSAVE_FLAGS },
+ { ".aes", PROCESSOR_UNKNOWN,
+ CPU_AES_FLAGS },
+ { ".pclmul", PROCESSOR_UNKNOWN,
+ CPU_PCLMUL_FLAGS },
+ { ".clmul", PROCESSOR_UNKNOWN,
+ CPU_PCLMUL_FLAGS },
+ { ".fma", PROCESSOR_UNKNOWN,
+ CPU_FMA_FLAGS },
+ { ".movbe", PROCESSOR_UNKNOWN,
+ CPU_MOVBE_FLAGS },
+ { ".ept", PROCESSOR_UNKNOWN,
+ CPU_EPT_FLAGS },
+ { ".clflush", PROCESSOR_UNKNOWN,
+ CPU_CLFLUSH_FLAGS },
+ { ".syscall", PROCESSOR_UNKNOWN,
+ CPU_SYSCALL_FLAGS },
+ { ".rdtscp", PROCESSOR_UNKNOWN,
+ CPU_RDTSCP_FLAGS },
{ ".3dnow", PROCESSOR_UNKNOWN,
CPU_3DNOW_FLAGS },
{ ".3dnowa", PROCESSOR_UNKNOWN,
CPU_SSE5_FLAGS },
};
+#ifdef I386COFF
+/* Like s_lcomm_internal in gas/read.c but the alignment string
+ is allowed to be optional. */
+
+static symbolS *
+pe_lcomm_internal (int needs_align, symbolS *symbolP, addressT size)
+{
+ addressT align = 0;
+
+ SKIP_WHITESPACE ();
+
+ if (needs_align
+ && *input_line_pointer == ',')
+ {
+ align = parse_align (needs_align - 1);
+
+ if (align == (addressT) -1)
+ return NULL;
+ }
+ else
+ {
+ if (size >= 8)
+ align = 3;
+ else if (size >= 4)
+ align = 2;
+ else if (size >= 2)
+ align = 1;
+ else
+ align = 0;
+ }
+
+ bss_alloc (symbolP, size, align);
+ return symbolP;
+}
+
+static void
+pe_lcomm (int needs_align)
+{
+ s_comm_internal (needs_align * 2, pe_lcomm_internal);
+}
+#endif
+
const pseudo_typeS md_pseudo_table[] =
{
#if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
{"arch", set_cpu_arch, 0},
#ifndef I386COFF
{"bss", s_bss, 0},
+#else
+ {"lcomm", pe_lcomm, 1},
#endif
{"ffloat", float_cons, 'f'},
{"dfloat", float_cons, 'd'},
{"att_mnemonic", set_intel_mnemonic, 0},
{"allow_index_reg", set_allow_index_reg, 1},
{"disallow_index_reg", set_allow_index_reg, 0},
+ {"sse_check", set_sse_check, 0},
#if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
{"largecomm", handle_large_common, 0},
#else
1. For PROCESSOR_I386, PROCESSOR_I486, PROCESSOR_PENTIUM and
PROCESSOR_GENERIC32, f32_patt will be used.
2. For PROCESSOR_PENTIUMPRO, PROCESSOR_PENTIUM4, PROCESSOR_NOCONA,
- PROCESSOR_CORE, PROCESSOR_CORE2, and PROCESSOR_GENERIC64,
- alt_long_patt will be used.
+ PROCESSOR_CORE, PROCESSOR_CORE2, PROCESSOR_COREI7, and
+ PROCESSOR_GENERIC64, alt_long_patt will be used.
3. For PROCESSOR_ATHLON, PROCESSOR_K6, PROCESSOR_K8 and
PROCESSOR_AMDFAM10, alt_short_patt will be used.
{
const char *const *patt = NULL;
- if (cpu_arch_isa == PROCESSOR_UNKNOWN)
+ if (fragP->tc_frag_data.isa == PROCESSOR_UNKNOWN)
{
/* PROCESSOR_UNKNOWN means that all ISAs may be used. */
switch (cpu_arch_tune)
case PROCESSOR_UNKNOWN:
/* We use cpu_arch_isa_flags to check if we SHOULD
optimize for Cpu686. */
- if (cpu_arch_isa_flags.bitfield.cpui686)
+ if (fragP->tc_frag_data.isa_flags.bitfield.cpui686)
patt = alt_long_patt;
else
patt = f32_patt;
case PROCESSOR_NOCONA:
case PROCESSOR_CORE:
case PROCESSOR_CORE2:
+ case PROCESSOR_COREI7:
case PROCESSOR_GENERIC64:
patt = alt_long_patt;
break;
}
else
{
- switch (cpu_arch_tune)
+ switch (fragP->tc_frag_data.tune)
{
case PROCESSOR_UNKNOWN:
- /* When cpu_arch_isa is net, cpu_arch_tune shouldn't be
+ /* When cpu_arch_isa is set, cpu_arch_tune shouldn't be
PROCESSOR_UNKNOWN. */
abort ();
break;
case PROCESSOR_GENERIC32:
/* We use cpu_arch_isa_flags to check if we CAN optimize
for Cpu686. */
- if (cpu_arch_isa_flags.bitfield.cpui686)
+ if (fragP->tc_frag_data.isa_flags.bitfield.cpui686)
patt = alt_short_patt;
else
patt = f32_patt;
case PROCESSOR_NOCONA:
case PROCESSOR_CORE:
case PROCESSOR_CORE2:
- if (cpu_arch_isa_flags.bitfield.cpui686)
+ case PROCESSOR_COREI7:
+ if (fragP->tc_frag_data.isa_flags.bitfield.cpui686)
patt = alt_long_patt;
else
patt = f32_patt;
}
static INLINE int
-uints_all_zero (const unsigned int *x, unsigned int size)
+operand_type_all_zero (const union i386_operand_type *x)
{
- switch (size)
+ switch (ARRAY_SIZE(x->array))
{
case 3:
- if (x[2])
+ if (x->array[2])
return 0;
case 2:
- if (x[1])
+ if (x->array[1])
return 0;
case 1:
- return !x[0];
+ return !x->array[0];
default:
abort ();
}
}
static INLINE void
-uints_set (unsigned int *x, unsigned int v, unsigned int size)
+operand_type_set (union i386_operand_type *x, unsigned int v)
{
- switch (size)
+ switch (ARRAY_SIZE(x->array))
{
case 3:
- x[2] = v;
+ x->array[2] = v;
case 2:
- x[1] = v;
+ x->array[1] = v;
case 1:
- x[0] = v;
+ x->array[0] = v;
break;
default:
abort ();
}
static INLINE int
-uints_equal (const unsigned int *x, const unsigned int *y,
- unsigned int size)
+operand_type_equal (const union i386_operand_type *x,
+ const union i386_operand_type *y)
{
- switch (size)
+ switch (ARRAY_SIZE(x->array))
{
case 3:
- if (x[2] != y [2])
+ if (x->array[2] != y->array[2])
return 0;
case 2:
- if (x[1] != y [1])
+ if (x->array[1] != y->array[1])
return 0;
case 1:
- return x[0] == y [0];
+ return x->array[0] == y->array[0];
break;
default:
abort ();
}
}
-#define UINTS_ALL_ZERO(x) \
- uints_all_zero ((x).array, ARRAY_SIZE ((x).array))
-#define UINTS_SET(x, v) \
- uints_set ((x).array, v, ARRAY_SIZE ((x).array))
-#define UINTS_CLEAR(x) \
- uints_set ((x).array, 0, ARRAY_SIZE ((x).array))
-#define UINTS_EQUAL(x, y) \
- uints_equal ((x).array, (y).array, ARRAY_SIZE ((x).array))
+static INLINE int
+cpu_flags_all_zero (const union i386_cpu_flags *x)
+{
+ switch (ARRAY_SIZE(x->array))
+ {
+ case 3:
+ if (x->array[2])
+ return 0;
+ case 2:
+ if (x->array[1])
+ return 0;
+ case 1:
+ return !x->array[0];
+ default:
+ abort ();
+ }
+}
+
+static INLINE void
+cpu_flags_set (union i386_cpu_flags *x, unsigned int v)
+{
+ switch (ARRAY_SIZE(x->array))
+ {
+ case 3:
+ x->array[2] = v;
+ case 2:
+ x->array[1] = v;
+ case 1:
+ x->array[0] = v;
+ break;
+ default:
+ abort ();
+ }
+}
+
+static INLINE int
+cpu_flags_equal (const union i386_cpu_flags *x,
+ const union i386_cpu_flags *y)
+{
+ switch (ARRAY_SIZE(x->array))
+ {
+ case 3:
+ if (x->array[2] != y->array[2])
+ return 0;
+ case 2:
+ if (x->array[1] != y->array[1])
+ return 0;
+ case 1:
+ return x->array[0] == y->array[0];
+ break;
+ default:
+ abort ();
+ }
+}
static INLINE int
cpu_flags_check_cpu64 (i386_cpu_flags f)
return x;
}
-/* Return 3 if there is a perfect match, 2 if compatible with 64bit,
- 1 if compatible with arch, 0 if there is no match. */
+#define CPU_FLAGS_ARCH_MATCH 0x1
+#define CPU_FLAGS_64BIT_MATCH 0x2
+#define CPU_FLAGS_AES_MATCH 0x4
+#define CPU_FLAGS_PCLMUL_MATCH 0x8
+#define CPU_FLAGS_AVX_MATCH 0x10
+
+#define CPU_FLAGS_32BIT_MATCH \
+ (CPU_FLAGS_ARCH_MATCH | CPU_FLAGS_AES_MATCH \
+ | CPU_FLAGS_PCLMUL_MATCH | CPU_FLAGS_AVX_MATCH)
+#define CPU_FLAGS_PERFECT_MATCH \
+ (CPU_FLAGS_32BIT_MATCH | CPU_FLAGS_64BIT_MATCH)
+
+/* Return CPU flags match bits. */
static int
-cpu_flags_match (i386_cpu_flags x)
+cpu_flags_match (const template *t)
{
- int overlap = cpu_flags_check_cpu64 (x) ? 2 : 0;
+ i386_cpu_flags x = t->cpu_flags;
+ int match = cpu_flags_check_cpu64 (x) ? CPU_FLAGS_64BIT_MATCH : 0;
x.bitfield.cpu64 = 0;
x.bitfield.cpuno64 = 0;
- if (UINTS_ALL_ZERO (x))
- overlap |= 1;
+ if (cpu_flags_all_zero (&x))
+ {
+ /* This instruction is available on all archs. */
+ match |= CPU_FLAGS_32BIT_MATCH;
+ }
else
{
+ /* This instruction is available only on some archs. */
i386_cpu_flags cpu = cpu_arch_flags;
cpu.bitfield.cpu64 = 0;
cpu.bitfield.cpuno64 = 0;
cpu = cpu_flags_and (x, cpu);
- overlap |= UINTS_ALL_ZERO (cpu) ? 0 : 1;
+ if (!cpu_flags_all_zero (&cpu))
+ {
+ if (x.bitfield.cpuavx)
+ {
+ /* We only need to check AES/PCLMUL/SSE2AVX with AVX. */
+ if (cpu.bitfield.cpuavx)
+ {
+ /* Check SSE2AVX. */
+ if (!t->opcode_modifier.sse2avx|| sse2avx)
+ {
+ match |= (CPU_FLAGS_ARCH_MATCH
+ | CPU_FLAGS_AVX_MATCH);
+ /* Check AES. */
+ if (!x.bitfield.cpuaes || cpu.bitfield.cpuaes)
+ match |= CPU_FLAGS_AES_MATCH;
+ /* Check PCLMUL. */
+ if (!x.bitfield.cpupclmul
+ || cpu.bitfield.cpupclmul)
+ match |= CPU_FLAGS_PCLMUL_MATCH;
+ }
+ }
+ else
+ match |= CPU_FLAGS_ARCH_MATCH;
+ }
+ else
+ match |= CPU_FLAGS_32BIT_MATCH;
+ }
}
- return overlap;
+ return match;
}
static INLINE i386_operand_type
static const i386_operand_type acc32 = OPERAND_TYPE_ACC32;
static const i386_operand_type acc64 = OPERAND_TYPE_ACC64;
static const i386_operand_type control = OPERAND_TYPE_CONTROL;
+static const i386_operand_type inoutportreg
+ = OPERAND_TYPE_INOUTPORTREG;
static const i386_operand_type reg16_inoutportreg
= OPERAND_TYPE_REG16_INOUTPORTREG;
static const i386_operand_type disp16 = OPERAND_TYPE_DISP16;
static const i386_operand_type anydisp
= 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 imm8 = OPERAND_TYPE_IMM8;
static const i386_operand_type imm8s = OPERAND_TYPE_IMM8S;
static const i386_operand_type imm16 = OPERAND_TYPE_IMM16;
default:
abort ();
}
+
+ return 0;
}
/* Return 1 if there is no conflict in 8bit/16bit/32bit/64bit on
|| (i.types[j].bitfield.tbyte
&& !t->operand_types[j].bitfield.tbyte)
|| (i.types[j].bitfield.xmmword
- && !t->operand_types[j].bitfield.xmmword)));
+ && !t->operand_types[j].bitfield.xmmword)
+ || (i.types[j].bitfield.ymmword
+ && !t->operand_types[j].bitfield.ymmword)));
}
/* Return 1 if there is no size conflict on any operands for
temp.bitfield.qword = 0;
temp.bitfield.tbyte = 0;
temp.bitfield.xmmword = 0;
- if (UINTS_ALL_ZERO (temp))
+ temp.bitfield.ymmword = 0;
+ if (operand_type_all_zero (&temp))
return 0;
return (given.bitfield.baseindex == overlap.bitfield.baseindex
{
i386_operand_type t;
- UINTS_CLEAR (t);
+ operand_type_set (&t, 0);
t.bitfield.imm64 = 1;
if (cpu_arch_tune != PROCESSOR_I486 && num == 1)
allow_index_reg = flag;
}
+static void
+set_sse_check (int dummy ATTRIBUTE_UNUSED)
+{
+ SKIP_WHITESPACE ();
+
+ if (!is_end_of_line[(unsigned char) *input_line_pointer])
+ {
+ char *string = input_line_pointer;
+ int e = get_symbol_end ();
+
+ if (strcmp (string, "none") == 0)
+ sse_check = sse_check_none;
+ else if (strcmp (string, "warning") == 0)
+ sse_check = sse_check_warning;
+ else if (strcmp (string, "error") == 0)
+ sse_check = sse_check_error;
+ else
+ as_bad (_("bad argument to sse_check directive."));
+ *input_line_pointer = e;
+ }
+ else
+ as_bad (_("missing argument for sse_check directive"));
+
+ demand_empty_rest_of_line ();
+}
+
static void
set_cpu_arch (int dummy ATTRIBUTE_UNUSED)
{
flags = cpu_flags_or (cpu_arch_flags,
cpu_arch[i].flags);
- if (!UINTS_EQUAL (flags, cpu_arch_flags))
+ if (!cpu_flags_equal (&flags, &cpu_arch_flags))
{
if (cpu_sub_arch_name)
{
core_optab->end = optab;
hash_err = hash_insert (op_hash,
(optab - 1)->name,
- (PTR) core_optab);
+ (void *) core_optab);
if (hash_err)
{
as_fatal (_("Internal Error: Can't hash %s: %s"),
for (regtab = i386_regtab; regtab_size--; regtab++)
{
- hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
+ hash_err = hash_insert (reg_hash, regtab->reg_name, (void *) regtab);
if (hash_err)
as_fatal (_("Internal Error: Can't hash %s: %s"),
regtab->reg_name,
operand_chars['?'] = '?';
#endif
digit_chars['-'] = '-';
+ mnemonic_chars['_'] = '_';
mnemonic_chars['-'] = '-';
mnemonic_chars['.'] = '.';
identifier_chars['_'] = '_';
|| x->types[i].bitfield.reg64
|| x->types[i].bitfield.regmmx
|| x->types[i].bitfield.regxmm
+ || x->types[i].bitfield.regymm
|| x->types[i].bitfield.sreg2
|| x->types[i].bitfield.sreg3
|| x->types[i].bitfield.control
{ OPERAND_TYPE_JUMPABSOLUTE, "Jump Absolute" },
{ OPERAND_TYPE_REGMMX, "rMMX" },
{ OPERAND_TYPE_REGXMM, "rXMM" },
+ { OPERAND_TYPE_REGYMM, "rYMM" },
{ OPERAND_TYPE_ESSEG, "es" },
};
for (j = 0; j < ARRAY_SIZE (type_names); j++)
{
a = operand_type_and (t, type_names[j].mask);
- if (!UINTS_ALL_ZERO (a))
+ if (!operand_type_all_zero (&a))
fprintf (stdout, "%s, ", type_names[j].name);
}
fflush (stdout);
sign > 0 ? "signed" : "unsigned", size);
}
- abort ();
- return BFD_RELOC_NONE;
+ return NO_RELOC;
}
/* Here we decide which fixups can be adjusted to make them relative to
return 1;
}
+/* Build the VEX prefix. */
+
+static void
+build_vex_prefix (const template *t)
+{
+ unsigned int register_specifier;
+ unsigned int implied_prefix;
+ unsigned int vector_length;
+
+ /* Check register specifier. */
+ if (i.vex.register_specifier)
+ {
+ register_specifier = i.vex.register_specifier->reg_num;
+ if ((i.vex.register_specifier->reg_flags & RegRex))
+ register_specifier += 8;
+ register_specifier = ~register_specifier & 0xf;
+ }
+ else
+ register_specifier = 0xf;
+
+ /* Use 2-byte VEX prefix by swappping destination and source
+ operand. */
+ if (!i.swap_operand
+ && i.operands == i.reg_operands
+ && i.tm.opcode_modifier.vex0f
+ && i.tm.opcode_modifier.s
+ && i.rex == REX_B)
+ {
+ unsigned int xchg = i.operands - 1;
+ union i386_op temp_op;
+ i386_operand_type temp_type;
+
+ temp_type = i.types[xchg];
+ i.types[xchg] = i.types[0];
+ i.types[0] = temp_type;
+ temp_op = i.op[xchg];
+ i.op[xchg] = i.op[0];
+ i.op[0] = temp_op;
+
+ assert (i.rm.mode == 3);
+
+ i.rex = REX_R;
+ xchg = i.rm.regmem;
+ i.rm.regmem = i.rm.reg;
+ i.rm.reg = xchg;
+
+ /* Use the next insn. */
+ i.tm = t[1];
+ }
+
+ vector_length = i.tm.opcode_modifier.vex256 ? 1 : 0;
+
+ 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 ();
+ }
+
+ /* Use 2-byte VEX prefix if possible. */
+ if (i.tm.opcode_modifier.vex0f
+ && (i.rex & (REX_W | REX_X | REX_B)) == 0)
+ {
+ /* 2-byte VEX prefix. */
+ unsigned int r;
+
+ i.vex.length = 2;
+ i.vex.bytes[0] = 0xc5;
+
+ /* Check the REX.R bit. */
+ r = (i.rex & REX_R) ? 0 : 1;
+ i.vex.bytes[1] = (r << 7
+ | register_specifier << 3
+ | vector_length << 2
+ | implied_prefix);
+ }
+ else
+ {
+ /* 3-byte VEX prefix. */
+ unsigned int m, w;
+
+ if (i.tm.opcode_modifier.vex0f)
+ m = 0x1;
+ else if (i.tm.opcode_modifier.vex0f38)
+ m = 0x2;
+ else if (i.tm.opcode_modifier.vex0f3a)
+ m = 0x3;
+ else
+ abort ();
+
+ i.vex.length = 3;
+ i.vex.bytes[0] = 0xc4;
+
+ /* The high 3 bits of the second VEX byte are 1's compliment
+ of RXB bits from REX. */
+ i.vex.bytes[1] = (~i.rex & 0x7) << 5 | m;
+
+ /* Check the REX.W bit. */
+ w = (i.rex & REX_W) ? 1 : 0;
+ if (i.tm.opcode_modifier.vexw0 || i.tm.opcode_modifier.vexw1)
+ {
+ if (w)
+ abort ();
+
+ if (i.tm.opcode_modifier.vexw1)
+ w = 1;
+ }
+
+ i.vex.bytes[2] = (w << 7
+ | register_specifier << 3
+ | vector_length << 2
+ | implied_prefix);
+ }
+}
+
+static void
+process_immext (void)
+{
+ expressionS *exp;
+
+ if (i.tm.cpu_flags.bitfield.cpusse3 && i.operands > 0)
+ {
+ /* SSE3 Instructions have the fixed operands with an opcode
+ suffix which is coded in the same place as an 8-bit immediate
+ field would be. Here we check those operands and remove them
+ afterwards. */
+ unsigned int x;
+
+ for (x = 0; x < i.operands; x++)
+ if (i.op[x].regs->reg_num != x)
+ as_bad (_("can't use register '%s%s' as operand %d in '%s'."),
+ register_prefix, i.op[x].regs->reg_name, x + 1,
+ i.tm.name);
+
+ i.operands = 0;
+ }
+
+ /* These AMD 3DNow! and SSE2 instructions have an opcode suffix
+ which is coded in the same place as an 8-bit immediate field
+ would be. Here we fake an 8-bit immediate operand from the
+ opcode suffix stored in tm.extension_opcode.
+
+ SSE5 and AVX instructions also use this encoding, for some of
+ 3 argument instructions. */
+
+ assert (i.imm_operands == 0
+ && (i.operands <= 2
+ || (i.tm.cpu_flags.bitfield.cpusse5
+ && i.operands <= 3)
+ || (i.tm.opcode_modifier.vex
+ && i.operands <= 4)));
+
+ exp = &im_expressions[i.imm_operands++];
+ i.op[i.operands].imms = exp;
+ i.types[i.operands] = imm8;
+ i.operands++;
+ exp->X_op = O_constant;
+ exp->X_add_number = i.tm.extension_opcode;
+ i.tm.extension_opcode = None;
+}
+
/* This is the guts of the machine-dependent assembler. LINE points to a
machine dependent instruction. This function is supposed to emit
the frags/bytes it assembles to. */
void
-md_assemble (line)
- char *line;
+md_assemble (char *line)
{
unsigned int j;
char mnemonic[MAX_MNEM_SIZE];
+ const template *t;
/* Initialize globals. */
memset (&i, '\0', sizeof (i));
making sure the overlap of the given operands types is consistent
with the template operand types. */
- if (!match_template ())
+ if (!(t = match_template ()))
return;
+ if (sse_check != sse_check_none
+ && !i.tm.opcode_modifier.noavx
+ && (i.tm.cpu_flags.bitfield.cpusse
+ || i.tm.cpu_flags.bitfield.cpusse2
+ || i.tm.cpu_flags.bitfield.cpusse3
+ || i.tm.cpu_flags.bitfield.cpussse3
+ || i.tm.cpu_flags.bitfield.cpusse4_1
+ || i.tm.cpu_flags.bitfield.cpusse4_2))
+ {
+ (sse_check == sse_check_warning
+ ? as_warn
+ : as_bad) (_("SSE instruction `%s' is used"), i.tm.name);
+ }
+
/* Zap movzx and movsx suffix. The suffix has been set from
"word ptr" or "byte ptr" on the source operand in Intel syntax
or extracted from mnemonic in AT&T syntax. But we'll use
{
if (!check_string ())
return;
+ i.disp_operands = 0;
}
if (!process_suffix ())
|| i.types[j].bitfield.floatacc)
i.reg_operands--;
- if (i.tm.opcode_modifier.immext)
- {
- expressionS *exp;
-
- if (i.tm.cpu_flags.bitfield.cpusse3 && i.operands > 0)
- {
- /* Streaming SIMD extensions 3 Instructions have the fixed
- operands with an opcode suffix which is coded in the same
- place as an 8-bit immediate field would be. Here we check
- those operands and remove them afterwards. */
- unsigned int x;
-
- for (x = 0; x < i.operands; x++)
- if (i.op[x].regs->reg_num != x)
- as_bad (_("can't use register '%s%s' as operand %d in '%s'."),
- register_prefix,
- i.op[x].regs->reg_name,
- x + 1,
- i.tm.name);
- i.operands = 0;
- }
-
- /* These AMD 3DNow! and Intel Katmai New Instructions have an
- opcode suffix which is coded in the same place as an 8-bit
- immediate field would be. Here we fake an 8-bit immediate
- operand from the opcode suffix stored in tm.extension_opcode.
- SSE5 also uses this encoding, for some of its 3 argument
- instructions. */
-
- assert (i.imm_operands == 0
- && (i.operands <= 2
- || (i.tm.cpu_flags.bitfield.cpusse5
- && i.operands <= 3)));
-
- exp = &im_expressions[i.imm_operands++];
- i.op[i.operands].imms = exp;
- UINTS_CLEAR (i.types[i.operands]);
- i.types[i.operands].bitfield.imm8 = 1;
- i.operands++;
- exp->X_op = O_constant;
- exp->X_add_number = i.tm.extension_opcode;
- i.tm.extension_opcode = None;
- }
+ /* ImmExt should be processed after SSE2AVX. */
+ if (!i.tm.opcode_modifier.sse2avx
+ && i.tm.opcode_modifier.immext)
+ process_immext ();
/* For insns with operands there are more diddles to do to the opcode. */
if (i.operands)
as_warn (_("translating to `%sp'"), i.tm.name);
}
+ if (i.tm.opcode_modifier.vex)
+ build_vex_prefix (t);
+
/* Handle conversion of 'int $3' --> special int3 insn. */
if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
{
char *mnem_p;
int supported;
const template *t;
+ char *dot_p = NULL;
/* Non-zero if we found a prefix only acceptable with string insns. */
const char *expecting_string_instruction = NULL;
mnem_p = mnemonic;
while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
{
+ if (*mnem_p == '.')
+ dot_p = mnem_p;
mnem_p++;
if (mnem_p >= mnemonic + MAX_MNEM_SIZE)
{
if (!current_templates)
{
+ /* Check if we should swap operand in encoding. */
+ if (mnem_p - 2 == dot_p && dot_p[1] == 's')
+ i.swap_operand = 1;
+ else
+ goto check_suffix;
+ mnem_p = dot_p;
+ *dot_p = '\0';
+ current_templates = hash_find (op_hash, mnemonic);
+ }
+
+ if (!current_templates)
+ {
+check_suffix:
/* See if we can get a match by trimming off a suffix. */
switch (mnem_p[-1])
{
supported = 0;
for (t = current_templates->start; t < current_templates->end; ++t)
{
- supported |= cpu_flags_match (t->cpu_flags);
- if (supported == 3)
+ supported |= cpu_flags_match (t);
+ if (supported == CPU_FLAGS_PERFECT_MATCH)
goto skip;
}
- if (!(supported & 2))
+ if (!(supported & CPU_FLAGS_64BIT_MATCH))
{
as_bad (flag_code == CODE_64BIT
? _("`%s' is not supported in 64-bit mode")
current_templates->start->name);
return NULL;
}
- if (!(supported & 1))
+ if (supported != CPU_FLAGS_PERFECT_MATCH)
{
as_bad (_("`%s' is not supported on `%s%s'"),
- current_templates->start->name, cpu_arch_name,
+ current_templates->start->name,
+ cpu_arch_name ? cpu_arch_name : default_arch,
cpu_sub_arch_name ? cpu_sub_arch_name : "");
return NULL;
}
{
switch (i.operands)
{
+ case 5:
case 4:
swap_2_operands (1, i.operands - 2);
case 3:
i386_operand_type mask, allowed;
const template *t;
- UINTS_CLEAR (mask);
- UINTS_CLEAR (allowed);
+ operand_type_set (&mask, 0);
+ operand_type_set (&allowed, 0);
for (t = current_templates->start;
t < current_templates->end;
break;
}
allowed = operand_type_and (mask, allowed);
- if (!UINTS_ALL_ZERO (allowed))
+ if (!operand_type_all_zero (&allowed))
i.types[op] = operand_type_and (i.types[op], mask);
}
break;
}
}
-static int
+static const template *
match_template (void)
{
/* Points to template once we've found it. */
const template *t;
i386_operand_type overlap0, overlap1, overlap2, overlap3;
+ i386_operand_type overlap4;
unsigned int found_reverse_match;
i386_opcode_modifier suffix_check;
i386_operand_type operand_types [MAX_OPERANDS];
unsigned int found_cpu_match;
unsigned int check_register;
-#if MAX_OPERANDS != 4
-# error "MAX_OPERANDS must be 4."
+#if MAX_OPERANDS != 5
+# error "MAX_OPERANDS must be 5."
#endif
found_reverse_match = 0;
continue;
/* Check processor support. */
- found_cpu_match = cpu_flags_match (t->cpu_flags) == 3;
+ found_cpu_match = (cpu_flags_match (t)
+ == CPU_FLAGS_PERFECT_MATCH);
if (!found_cpu_match)
continue;
&& !intel_float_operand (t->name))
: intel_float_operand (t->name) != 2)
&& ((!operand_types[0].bitfield.regmmx
- && !operand_types[0].bitfield.regxmm)
+ && !operand_types[0].bitfield.regxmm
+ && !operand_types[0].bitfield.regymm)
|| (!operand_types[t->operands > 1].bitfield.regmmx
- && !!operand_types[t->operands > 1].bitfield.regxmm))
+ && !!operand_types[t->operands > 1].bitfield.regxmm
+ && !!operand_types[t->operands > 1].bitfield.regymm))
&& (t->base_opcode != 0x0fc7
|| t->extension_opcode != 1 /* cmpxchg8b */))
continue;
+ /* In general, don't allow 32-bit operands on pre-386. */
+ else if (i.suffix == LONG_MNEM_SUFFIX
+ && !cpu_arch_flags.bitfield.cpui386
+ && (intel_syntax
+ ? (!t->opcode_modifier.ignoresize
+ && !intel_float_operand (t->name))
+ : intel_float_operand (t->name) != 2)
+ && ((!operand_types[0].bitfield.regmmx
+ && !operand_types[0].bitfield.regxmm)
+ || (!operand_types[t->operands > 1].bitfield.regmmx
+ && !!operand_types[t->operands > 1].bitfield.regxmm)))
+ continue;
+
/* Do not verify operands when there are none. */
else
{
zero-extend %eax to %rax. */
if (flag_code == CODE_64BIT
&& t->base_opcode == 0x90
- && UINTS_EQUAL (i.types [0], acc32)
- && UINTS_EQUAL (i.types [1], acc32))
+ && operand_type_equal (&i.types [0], &acc32)
+ && operand_type_equal (&i.types [1], &acc32))
continue;
+ if (i.swap_operand)
+ {
+ /* If we swap operand in encoding, we either match
+ the next one or reverse direction of operands. */
+ if (t->opcode_modifier.s)
+ continue;
+ else if (t->opcode_modifier.d)
+ goto check_reverse;
+ }
+
case 3:
+ /* If we swap operand in encoding, we match the next one. */
+ if (i.swap_operand && t->opcode_modifier.s)
+ continue;
case 4:
+ case 5:
overlap1 = operand_type_and (i.types[1], operand_types[1]);
if (!operand_type_match (overlap0, i.types[0])
|| !operand_type_match (overlap1, i.types[1])
if (!t->opcode_modifier.d && !t->opcode_modifier.floatd)
continue;
+check_reverse:
/* Try reversing direction of operands. */
overlap0 = operand_type_and (i.types[0], operand_types[1]);
overlap1 = operand_type_and (i.types[1], operand_types[0]);
/* Found a forward 2 operand match here. */
switch (t->operands)
{
+ case 5:
+ overlap4 = operand_type_and (i.types[4],
+ operand_types[4]);
case 4:
overlap3 = operand_type_and (i.types[3],
operand_types[3]);
switch (t->operands)
{
+ case 5:
+ if (!operand_type_match (overlap4, i.types[4])
+ || !operand_type_register_match (overlap3,
+ i.types[3],
+ operand_types[3],
+ overlap4,
+ i.types[4],
+ operand_types[4]))
+ continue;
case 4:
if (!operand_type_match (overlap3, i.types[3])
|| (check_register
found_reverse_match = 0;
continue;
}
+
/* We've found a match; break out of loop. */
break;
}
if (t == current_templates->end)
{
/* We found no match. */
- as_bad (_("suffix or operands invalid for `%s'"),
- current_templates->start->name);
- return 0;
+ if (intel_syntax)
+ as_bad (_("ambiguous operand size or operands invalid for `%s'"),
+ current_templates->start->name);
+ else
+ as_bad (_("suffix or operands invalid for `%s'"),
+ current_templates->start->name);
+ return NULL;
}
if (!quiet_warnings)
i.tm.operand_types[1] = operand_types[0];
}
- return 1;
+ return t;
}
static int
{
if (i.seg[0] != NULL && i.seg[0] != &es)
{
- as_bad (_("`%s' operand %d must use `%%es' segment"),
+ as_bad (_("`%s' operand %d must use `%ses' segment"),
i.tm.name,
- mem_op + 1);
+ mem_op + 1,
+ register_prefix);
return 0;
}
/* There's only ever one segment override allowed per instruction.
{
if (i.seg[1] != NULL && i.seg[1] != &es)
{
- as_bad (_("`%s' operand %d must use `%%es' segment"),
+ as_bad (_("`%s' operand %d must use `%ses' segment"),
i.tm.name,
- mem_op + 2);
+ mem_op + 2,
+ register_prefix);
return 0;
}
}
if (!check_word_reg ())
return 0;
}
- else if (i.suffix == XMMWORD_MNEM_SUFFIX)
+ else if (i.suffix == XMMWORD_MNEM_SUFFIX
+ || i.suffix == YMMWORD_MNEM_SUFFIX)
{
- /* Skip if the instruction has x suffix. match_template
+ /* Skip if the instruction has x/y 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 != XMMWORD_MNEM_SUFFIX
+ && i.suffix != YMMWORD_MNEM_SUFFIX)
{
/* It's not a byte, select word/dword operation. */
if (i.tm.opcode_modifier.w)
if (! (i.operands == 2
&& i.tm.base_opcode == 0x90
&& i.tm.extension_opcode == None
- && UINTS_EQUAL (i.types [0], acc64)
- && UINTS_EQUAL (i.types [1], acc64))
+ && operand_type_equal (&i.types [0], &acc64)
+ && operand_type_equal (&i.types [1], &acc64))
&& ! (i.operands == 1
&& i.tm.base_opcode == 0xfc7
&& i.tm.extension_opcode == 1
|| i.types[op].bitfield.reg64
|| i.types[op].bitfield.regmmx
|| i.types[op].bitfield.regxmm
+ || i.types[op].bitfield.regymm
|| i.types[op].bitfield.sreg2
|| i.types[op].bitfield.sreg3
|| i.types[op].bitfield.control
|| overlap.bitfield.imm32
|| overlap.bitfield.imm32s
|| overlap.bitfield.imm64)
- && !UINTS_EQUAL (overlap, imm8)
- && !UINTS_EQUAL (overlap, imm8s)
- && !UINTS_EQUAL (overlap, imm16)
- && !UINTS_EQUAL (overlap, imm32)
- && !UINTS_EQUAL (overlap, imm32s)
- && !UINTS_EQUAL (overlap, imm64))
+ && !operand_type_equal (&overlap, &imm8)
+ && !operand_type_equal (&overlap, &imm8s)
+ && !operand_type_equal (&overlap, &imm16)
+ && !operand_type_equal (&overlap, &imm32)
+ && !operand_type_equal (&overlap, &imm32s)
+ && !operand_type_equal (&overlap, &imm64))
{
if (i.suffix)
{
i386_operand_type temp;
- UINTS_CLEAR (temp);
+ operand_type_set (&temp, 0);
if (i.suffix == BYTE_MNEM_SUFFIX)
{
temp.bitfield.imm8 = overlap.bitfield.imm8;
temp.bitfield.imm32 = overlap.bitfield.imm32;
overlap = temp;
}
- else if (UINTS_EQUAL (overlap, imm16_32_32s)
- || UINTS_EQUAL (overlap, imm16_32)
- || UINTS_EQUAL (overlap, imm16_32s))
+ else if (operand_type_equal (&overlap, &imm16_32_32s)
+ || operand_type_equal (&overlap, &imm16_32)
+ || operand_type_equal (&overlap, &imm16_32s))
{
- UINTS_CLEAR (overlap);
if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
- overlap.bitfield.imm16 = 1;
+ overlap = imm16;
else
- overlap.bitfield.imm32s = 1;
+ overlap = imm32s;
}
- if (!UINTS_EQUAL (overlap, imm8)
- && !UINTS_EQUAL (overlap, imm8s)
- && !UINTS_EQUAL (overlap, imm16)
- && !UINTS_EQUAL (overlap, imm32)
- && !UINTS_EQUAL (overlap, imm32s)
- && !UINTS_EQUAL (overlap, imm64))
+ if (!operand_type_equal (&overlap, &imm8)
+ && !operand_type_equal (&overlap, &imm8s)
+ && !operand_type_equal (&overlap, &imm16)
+ && !operand_type_equal (&overlap, &imm32)
+ && !operand_type_equal (&overlap, &imm32s)
+ && !operand_type_equal (&overlap, &imm64))
{
as_bad (_("no instruction mnemonic suffix given; "
"can't determine immediate size"));
&& i.op[0].regs->reg_flags == i.op[3].regs->reg_flags)
{
/* Clear the arguments that are stored in drex. */
- UINTS_CLEAR (i.types[0]);
- UINTS_CLEAR (i.types[3]);
+ operand_type_set (&i.types[0], 0);
+ operand_type_set (&i.types[3], 0);
i.reg_operands -= 2;
/* There are two different ways to encode a 4 operand
&& i.op[0].regs->reg_flags == i.op[3].regs->reg_flags)
{
/* clear the arguments that are stored in drex */
- UINTS_CLEAR (i.types[0]);
- UINTS_CLEAR (i.types[3]);
+ operand_type_set (&i.types[0], 0);
+ operand_type_set (&i.types[3], 0);
i.reg_operands -= 2;
/* Specify the modrm encoding for memory addressing. Include
&& i.op[0].regs->reg_flags == i.op[3].regs->reg_flags)
{
/* Clear the arguments that are stored in drex. */
- UINTS_CLEAR (i.types[0]);
- UINTS_CLEAR (i.types[3]);
+ operand_type_set (&i.types[0], 0);
+ operand_type_set (&i.types[3], 0);
i.reg_operands -= 2;
/* Specify the modrm encoding for memory addressing. Include
&& i.op[2].regs->reg_flags == i.op[3].regs->reg_flags)
{
/* clear the arguments that are stored in drex */
- UINTS_CLEAR (i.types[2]);
- UINTS_CLEAR (i.types[3]);
+ operand_type_set (&i.types[2], 0);
+ operand_type_set (&i.types[3], 0);
i.reg_operands -= 2;
/* There are two different ways to encode a 4 operand
&& i.op[2].regs->reg_flags == i.op[3].regs->reg_flags)
{
/* Clear the arguments that are stored in drex. */
- UINTS_CLEAR (i.types[2]);
- UINTS_CLEAR (i.types[3]);
+ operand_type_set (&i.types[2], 0);
+ operand_type_set (&i.types[3], 0);
i.reg_operands -= 2;
/* Specify the modrm encoding and remember the register
&& i.op[2].regs->reg_flags == i.op[3].regs->reg_flags)
{
/* clear the arguments that are stored in drex */
- UINTS_CLEAR (i.types[2]);
- UINTS_CLEAR (i.types[3]);
+ operand_type_set (&i.types[2], 0);
+ operand_type_set (&i.types[3], 0);
i.reg_operands -= 2;
/* Specify the modrm encoding and remember the register
&& i.op[0].regs->reg_flags == i.op[3].regs->reg_flags)
{
/* clear the arguments that are stored in drex */
- UINTS_CLEAR (i.types[0]);
- UINTS_CLEAR (i.types[3]);
+ operand_type_set (&i.types[0], 0);
+ operand_type_set (&i.types[3], 0);
i.reg_operands -= 2;
/* Specify the modrm encoding and remember the register
&& i.types[2].bitfield.regxmm != 0)
{
/* Clear the arguments that are stored in drex. */
- UINTS_CLEAR (i.types[2]);
+ operand_type_set (&i.types[2], 0);
i.reg_operands--;
/* Specify the modrm encoding and remember the register
&& i.types[2].bitfield.regxmm != 0)
{
/* Clear the arguments that are stored in drex. */
- UINTS_CLEAR (i.types[2]);
+ operand_type_set (&i.types[2], 0);
i.reg_operands--;
/* Specify the modrm encoding and remember the register
&& i.types[2].bitfield.regxmm != 0)
{
/* Clear the arguments that are stored in drex. */
- UINTS_CLEAR (i.types[2]);
+ operand_type_set (&i.types[2], 0);
i.reg_operands--;
/* Specify the modrm encoding and remember the register
&& i.types[3].bitfield.regxmm != 0)
{
/* clear the arguments that are stored in drex */
- UINTS_CLEAR (i.types[3]);
+ operand_type_set (&i.types[3], 0);
i.reg_operands--;
/* Specify the modrm encoding and remember the register
&& operand_type_check (i.types[3], imm) != 0)
{
/* clear the arguments that are stored in drex */
- UINTS_CLEAR (i.types[2]);
+ operand_type_set (&i.types[2], 0);
i.reg_operands--;
/* Specify the modrm encoding and remember the register
as_bad (_("Internal error for the '%s' instruction"), i.tm.name);
}
+static int
+bad_implicit_operand (int xmm)
+{
+ const char *reg = xmm ? "xmm0" : "ymm0";
+ if (intel_syntax)
+ as_bad (_("the last operand of `%s' must be `%s%s'"),
+ i.tm.name, register_prefix, reg);
+ else
+ as_bad (_("the first operand of `%s' must be `%s%s'"),
+ i.tm.name, register_prefix, reg);
+ return 0;
+}
+
static int
process_operands (void)
{
|| i.tm.opcode_modifier.drexc)
process_drex ();
- if (i.tm.opcode_modifier.firstxmm0)
+ if (i.tm.opcode_modifier.sse2avx
+ && (i.tm.opcode_modifier.vexnds
+ || i.tm.opcode_modifier.vexndd))
{
+ unsigned int dup = i.operands;
+ unsigned int dest = dup - 1;
unsigned int j;
- /* The first operand is implicit and must be xmm0. */
- assert (i.reg_operands && UINTS_EQUAL (i.types[0], regxmm));
- if (i.op[0].regs->reg_num != 0)
+ /* The destination must be an xmm register. */
+ assert (i.reg_operands
+ && MAX_OPERANDS > dup
+ && operand_type_equal (&i.types[dest], ®xmm));
+
+ if (i.tm.opcode_modifier.firstxmm0)
{
- if (intel_syntax)
- as_bad (_("the last operand of `%s' must be `%sxmm0'"),
- i.tm.name, register_prefix);
+ /* The first operand is implicit and must be xmm0. */
+ assert (operand_type_equal (&i.types[0], ®xmm));
+ if (i.op[0].regs->reg_num != 0)
+ return bad_implicit_operand (1);
+
+ if (i.tm.opcode_modifier.vex3sources)
+ {
+ /* Keep xmm0 for instructions with VEX prefix and 3
+ sources. */
+ goto duplicate;
+ }
else
- as_bad (_("the first operand of `%s' must be `%sxmm0'"),
- i.tm.name, register_prefix);
- return 0;
+ {
+ /* We remove the first xmm0 and keep the number of
+ operands unchanged, which in fact duplicates the
+ destination. */
+ for (j = 1; j < i.operands; j++)
+ {
+ i.op[j - 1] = i.op[j];
+ i.types[j - 1] = i.types[j];
+ i.tm.operand_types[j - 1] = i.tm.operand_types[j];
+ }
+ }
+ }
+ else if (i.tm.opcode_modifier.implicit1stxmm0)
+ {
+ assert ((MAX_OPERANDS - 1) > dup
+ && i.tm.opcode_modifier.vex3sources);
+
+ /* Add the implicit xmm0 for instructions with VEX prefix
+ and 3 sources. */
+ for (j = i.operands; j > 0; j--)
+ {
+ i.op[j] = i.op[j - 1];
+ i.types[j] = i.types[j - 1];
+ i.tm.operand_types[j] = i.tm.operand_types[j - 1];
+ }
+ i.op[0].regs
+ = (const reg_entry *) hash_find (reg_hash, "xmm0");
+ i.types[0] = regxmm;
+ i.tm.operand_types[0] = regxmm;
+
+ i.operands += 2;
+ i.reg_operands += 2;
+ i.tm.operands += 2;
+
+ dup++;
+ dest++;
+ i.op[dup] = i.op[dest];
+ i.types[dup] = i.types[dest];
+ i.tm.operand_types[dup] = i.tm.operand_types[dest];
}
+ else
+ {
+duplicate:
+ i.operands++;
+ i.reg_operands++;
+ i.tm.operands++;
+
+ i.op[dup] = i.op[dest];
+ i.types[dup] = i.types[dest];
+ i.tm.operand_types[dup] = i.tm.operand_types[dest];
+ }
+
+ if (i.tm.opcode_modifier.immext)
+ process_immext ();
+ }
+ else if (i.tm.opcode_modifier.firstxmm0)
+ {
+ unsigned int j;
+
+ /* The first operand is implicit and must be xmm0/ymm0. */
+ assert (i.reg_operands
+ && (operand_type_equal (&i.types[0], ®xmm)
+ || operand_type_equal (&i.types[0], ®ymm)));
+ if (i.op[0].regs->reg_num != 0)
+ return bad_implicit_operand (i.types[0].bitfield.regxmm);
for (j = 1; j < i.operands; j++)
{
if (i.tm.base_opcode == POP_SEG_SHORT
&& i.op[0].regs->reg_num == 1)
{
- as_bad (_("you can't `pop %%cs'"));
+ as_bad (_("you can't `pop %scs'"), register_prefix);
return 0;
}
i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
{
/* Reversed arguments on faddp, fsubp, etc. */
as_warn (_("translating to `%s %s%s,%s%s'"), i.tm.name,
- register_prefix, i.op[1].regs->reg_name,
- register_prefix, i.op[0].regs->reg_name);
+ register_prefix, i.op[!intel_syntax].regs->reg_name,
+ register_prefix, i.op[intel_syntax].regs->reg_name);
}
else
{
build_modrm_byte (void)
{
const seg_entry *default_seg = 0;
+ unsigned int source, dest;
+ int vex_3_sources;
+
+ /* The first operand of instructions with VEX prefix and 3 sources
+ must be VEX_Imm4. */
+ vex_3_sources = i.tm.opcode_modifier.vex3sources;
+ if (vex_3_sources)
+ {
+ unsigned int nds, reg;
+
+ dest = i.operands - 1;
+ nds = dest - 1;
+ source = 1;
+ reg = 0;
+
+ /* This instruction must have 4 operands: 4 register operands
+ or 3 register operands plus 1 memory operand. It must have
+ VexNDS and VexImmExt. */
+ assert (i.operands == 4
+ && (i.reg_operands == 4
+ || (i.reg_operands == 3 && i.mem_operands == 1))
+ && i.tm.opcode_modifier.vexnds
+ && i.tm.opcode_modifier.veximmext
+ && (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[nds],
+ ®xmm)
+ || operand_type_equal (&i.tm.operand_types[nds],
+ ®ymm))
+ && (operand_type_equal (&i.tm.operand_types[reg],
+ ®xmm)
+ || operand_type_equal (&i.tm.operand_types[reg],
+ ®ymm)));
+
+ /* Generate an 8bit immediate operand to encode the register
+ operand. */
+ expressionS *exp = &im_expressions[i.imm_operands++];
+ i.op[i.operands].imms = exp;
+ i.types[i.operands] = imm8;
+ i.operands++;
+ exp->X_op = O_constant;
+ exp->X_add_number
+ = ((i.op[0].regs->reg_num
+ + ((i.op[0].regs->reg_flags & RegRex) ? 8 : 0)) << 4);
+
+ i.vex.register_specifier = i.op[nds].regs;
+ }
+ else
+ source = dest = 0;
/* SSE5 4 operand instructions are encoded in such a way that one of
the inputs must match the destination register. Process_drex hides
}
/* i.reg_operands MUST be the number of real register operands;
- implicit registers do not count. */
- else if (i.reg_operands == 2)
+ implicit registers do not count. If there are 3 register
+ operands, it must be a instruction with VexNDS. For a
+ instruction with VexNDD, the destination register is encoded
+ in VEX prefix. If there are 4 register operands, it must be
+ a instruction with VEX prefix and 3 sources. */
+ else if (i.mem_operands == 0
+ && ((i.reg_operands == 2
+ && !i.tm.opcode_modifier.vexndd)
+ || (i.reg_operands == 3
+ && i.tm.opcode_modifier.vexnds)
+ || (i.reg_operands == 4 && vex_3_sources)))
{
- unsigned int source, dest;
-
switch (i.operands)
{
case 2:
case 3:
/* When there are 3 operands, one of them may be immediate,
which may be the first or the last operand. Otherwise,
- the first operand must be shift count register (cl). */
+ the first operand must be shift count register (cl) or it
+ is an instruction with VexNDS. */
assert (i.imm_operands == 1
|| (i.imm_operands == 0
- && i.types[0].bitfield.shiftcount));
+ && (i.tm.opcode_modifier.vexnds
+ || i.types[0].bitfield.shiftcount)));
if (operand_type_check (i.types[0], imm)
|| i.types[0].bitfield.shiftcount)
source = 1;
case 4:
/* When there are 4 operands, the first two must be 8bit
immediate operands. The source operand will be the 3rd
- one. */
- assert (i.imm_operands == 2
- && i.types[0].bitfield.imm8
- && i.types[1].bitfield.imm8);
- source = 2;
+ one.
+
+ For instructions with VexNDS, if the first operand
+ an imm8, the source operand is the 2nd one. If the last
+ operand is imm8, the source operand is the first one. */
+ assert ((i.imm_operands == 2
+ && i.types[0].bitfield.imm8
+ && i.types[1].bitfield.imm8)
+ || (i.tm.opcode_modifier.vexnds
+ && i.imm_operands == 1
+ && (i.types[0].bitfield.imm8
+ || i.types[i.operands - 1].bitfield.imm8)));
+ if (i.tm.opcode_modifier.vexnds)
+ {
+ if (i.types[0].bitfield.imm8)
+ source = 1;
+ else
+ source = 0;
+ }
+ else
+ source = 2;
+ break;
+ case 5:
break;
default:
abort ();
}
- dest = source + 1;
+ if (!vex_3_sources)
+ {
+ dest = source + 1;
+
+ if (i.tm.opcode_modifier.vexnds)
+ {
+ /* For instructions with VexNDS, the register-only
+ source operand must be XMM or YMM register. It is
+ encoded in VEX prefix. We need to clear RegMem bit
+ before calling operand_type_equal. */
+ i386_operand_type op = i.tm.operand_types[dest];
+ op.bitfield.regmem = 0;
+ if ((dest + 1) >= i.operands
+ || (!operand_type_equal (&op, ®xmm)
+ && !operand_type_equal (&op, ®ymm)))
+ abort ();
+ i.vex.register_specifier = i.op[dest].regs;
+ dest++;
+ }
+ }
i.rm.mode = 3;
/* One of the register operands will be encoded in the i.tm.reg
}
else
{ /* If it's not 2 reg operands... */
+ unsigned int mem;
+
if (i.mem_operands)
{
unsigned int fake_zero_displacement = 0;
&& operand_type_check (i.types[op], disp))
{
i386_operand_type temp;
- UINTS_CLEAR (temp);
+ operand_type_set (&temp, 0);
temp.bitfield.disp8 = i.types[op].bitfield.disp8;
i.types[op] = temp;
if (i.prefix[ADDR_PREFIX] == 0)
exp->X_add_symbol = (symbolS *) 0;
exp->X_op_symbol = (symbolS *) 0;
}
+
+ mem = op;
}
+ else
+ mem = ~0;
/* Fill in i.rm.reg or i.rm.regmem field with register operand
(if any) based on i.tm.extension_opcode. Again, we must be
}
else
{
+ unsigned int vex_reg = ~0;
+
for (op = 0; op < i.operands; op++)
if (i.types[op].bitfield.reg8
|| i.types[op].bitfield.reg16
|| i.types[op].bitfield.reg64
|| i.types[op].bitfield.regmmx
|| i.types[op].bitfield.regxmm
+ || i.types[op].bitfield.regymm
|| i.types[op].bitfield.sreg2
|| i.types[op].bitfield.sreg3
|| i.types[op].bitfield.control
|| i.types[op].bitfield.test)
break;
- assert (op < i.operands);
+ if (vex_3_sources)
+ op = dest;
+ else if (i.tm.opcode_modifier.vexnds)
+ {
+ /* For instructions with VexNDS, the register-only
+ source operand is encoded in VEX prefix. */
+ assert (mem != (unsigned int) ~0);
+
+ if (op > mem)
+ {
+ vex_reg = op++;
+ assert (op < i.operands);
+ }
+ else
+ {
+ vex_reg = op + 1;
+ assert (vex_reg < i.operands);
+ }
+ }
+ else if (i.tm.opcode_modifier.vexndd)
+ {
+ /* For instructions with VexNDD, there should be
+ no memory operand and the register destination
+ is encoded in VEX prefix. */
+ assert (i.mem_operands == 0
+ && (op + 2) == i.operands);
+ vex_reg = op + 1;
+ }
+ else
+ assert (op < i.operands);
+
+ if (vex_reg != (unsigned int) ~0)
+ {
+ assert (i.reg_operands == 2);
+
+ if (!operand_type_equal (&i.tm.operand_types[vex_reg],
+ & regxmm)
+ && !operand_type_equal (&i.tm.operand_types[vex_reg],
+ ®ymm))
+ abort ();
+ i.vex.register_specifier = i.op[vex_reg].regs;
+ }
/* If there is an extension opcode to put here, the
register number must be put into the regmem field. */
unsigned int j;
unsigned int prefix;
- switch (i.tm.opcode_length)
+ /* Since the VEX prefix contains the implicit prefix, we don't
+ need the explicit prefix. */
+ if (!i.tm.opcode_modifier.vex)
{
- case 3:
- if (i.tm.base_opcode & 0xff000000)
- {
- prefix = (i.tm.base_opcode >> 24) & 0xff;
- goto check_prefix;
- }
- break;
- case 2:
- if ((i.tm.base_opcode & 0xff0000) != 0)
+ switch (i.tm.opcode_length)
{
- prefix = (i.tm.base_opcode >> 16) & 0xff;
- if (i.tm.cpu_flags.bitfield.cpupadlock)
+ case 3:
+ if (i.tm.base_opcode & 0xff000000)
+ {
+ prefix = (i.tm.base_opcode >> 24) & 0xff;
+ goto check_prefix;
+ }
+ break;
+ case 2:
+ if ((i.tm.base_opcode & 0xff0000) != 0)
{
+ prefix = (i.tm.base_opcode >> 16) & 0xff;
+ if (i.tm.cpu_flags.bitfield.cpupadlock)
+ {
check_prefix:
- if (prefix != REPE_PREFIX_OPCODE
- || i.prefix[LOCKREP_PREFIX] != REPE_PREFIX_OPCODE)
+ if (prefix != REPE_PREFIX_OPCODE
+ || (i.prefix[LOCKREP_PREFIX]
+ != REPE_PREFIX_OPCODE))
+ add_prefix (prefix);
+ }
+ else
add_prefix (prefix);
}
- else
- add_prefix (prefix);
+ break;
+ case 1:
+ break;
+ default:
+ abort ();
}
- break;
- case 1:
- break;
- default:
- abort ();
+
+ /* The prefix bytes. */
+ for (j = ARRAY_SIZE (i.prefix), q = i.prefix; j > 0; j--, q++)
+ if (*q)
+ FRAG_APPEND_1_CHAR (*q);
}
- /* The prefix bytes. */
- for (j = ARRAY_SIZE (i.prefix), q = i.prefix; j > 0; j--, q++)
- if (*q)
- FRAG_APPEND_1_CHAR (*q);
+ if (i.tm.opcode_modifier.vex)
+ {
+ for (j = 0, q = i.prefix; j < ARRAY_SIZE (i.prefix); j++, q++)
+ if (*q)
+ switch (j)
+ {
+ case REX_PREFIX:
+ /* REX byte is encoded in VEX prefix. */
+ break;
+ case SEG_PREFIX:
+ case ADDR_PREFIX:
+ FRAG_APPEND_1_CHAR (*q);
+ break;
+ default:
+ /* There should be no other prefixes for instructions
+ with VEX prefix. */
+ abort ();
+ }
+
+ /* Now the VEX prefix. */
+ p = frag_more (i.vex.length);
+ for (j = 0; j < i.vex.length; j++)
+ p[j] = i.vex.bytes[j];
+ }
/* Now the opcode; be careful about word order here! */
if (i.tm.opcode_length == 1)
expressionS *exp;
i386_operand_type types;
- UINTS_SET (types, ~0);
+ operand_type_set (&types, ~0);
if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
{
return 0;
}
- UINTS_CLEAR (bigdisp);
+ operand_type_set (&bigdisp, 0);
if ((i.types[this_operand].bitfield.jumpabsolute)
|| (!current_templates->start->opcode_modifier.jump
&& !current_templates->start->opcode_modifier.jumpdword))
bigdisp.bitfield.disp32 = 0;
bigdisp.bitfield.disp32s = 0;
bigdisp.bitfield.disp64 = 0;
- if (UINTS_ALL_ZERO (bigdisp))
+ if (operand_type_all_zero (&bigdisp))
i.types[this_operand] = operand_type_and (i.types[this_operand],
types);
i386_index_check (const char *operand_string)
{
int ok;
+ const char *kind = "base/index";
#if INFER_ADDR_PREFIX
int fudged = 0;
tryprefix:
#endif
ok = 1;
- if (flag_code == CODE_64BIT)
+ if (current_templates->start->opcode_modifier.isstring
+ && !current_templates->start->opcode_modifier.immext
+ && (current_templates->end[-1].opcode_modifier.isstring
+ || i.mem_operands))
+ {
+ /* 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;
+
+ kind = "string address";
+
+ if (current_templates->start->opcode_modifier.w)
+ {
+ i386_operand_type type = current_templates->end[-1].operand_types[0];
+
+ if (!type.bitfield.baseindex
+ || ((!i.mem_operands != !intel_syntax)
+ && 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 */;
+ }
+ else
+ expected = 3 /* rBX */;
+
+ if (!i.base_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 (i.base_reg->reg_num != 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)
+ && i386_regtab[j].reg_num == expected)
+ break;
+ assert (j < i386_regtab_size);
+ as_warn (_("`%s' is not valid here (expected `%c%s%s%c')"),
+ operand_string,
+ intel_syntax ? '[' : '(',
+ register_prefix,
+ i386_regtab[j].reg_name,
+ intel_syntax ? ']' : ')');
+ ok = 1;
+ }
+ }
+ else if (flag_code == CODE_64BIT)
{
if ((i.base_reg
&& ((i.prefix[ADDR_PREFIX] == 0
if (!ok)
{
#if INFER_ADDR_PREFIX
- if (i.prefix[ADDR_PREFIX] == 0)
+ if (!i.mem_operands && !i.prefix[ADDR_PREFIX])
{
i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
i.prefixes += 1;
goto tryprefix;
}
if (fudged)
- as_bad (_("`%s' is not a valid base/index expression"),
- operand_string);
+ as_bad (_("`%s' is not a valid %s expression"),
+ operand_string,
+ kind);
else
#endif
- as_bad (_("`%s' is not a valid %s bit base/index expression"),
+ as_bad (_("`%s' is not a valid %s-bit %s expression"),
operand_string,
- flag_code_names[flag_code]);
+ flag_code_names[i.prefix[ADDR_PREFIX]
+ ? flag_code == CODE_32BIT
+ ? CODE_16BIT
+ : CODE_32BIT
+ : flag_code],
+ kind);
}
return ok;
}
-/* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
+/* Parse OPERAND_STRING into the i386_insn structure I. Returns zero
on error. */
static int
/* Special case for (%dx) while doing input/output op. */
if (i.base_reg
- && UINTS_EQUAL (i.base_reg->reg_type, reg16_inoutportreg)
+ && operand_type_equal (&i.base_reg->reg_type,
+ ®16_inoutportreg)
&& i.index_reg == 0
&& i.log2_scale_factor == 0
&& i.seg[i.mem_operands] == 0
&& !operand_type_check (i.types[this_operand], disp))
{
- UINTS_CLEAR (i.types[this_operand]);
- i.types[this_operand].bitfield.inoutportreg = 1;
+ i.types[this_operand] = inoutportreg;
return 1;
}
}
}
+ if (r == NULL || allow_pseudo_reg)
+ return r;
+
+ if (operand_type_all_zero (&r->reg_type))
+ return (const reg_entry *) NULL;
+
+ if ((r->reg_type.bitfield.reg32
+ || r->reg_type.bitfield.sreg3
+ || r->reg_type.bitfield.control
+ || r->reg_type.bitfield.debug
+ || r->reg_type.bitfield.test)
+ && !cpu_arch_flags.bitfield.cpui386)
+ return (const reg_entry *) NULL;
+
+ if (r->reg_type.bitfield.regmmx && !cpu_arch_flags.bitfield.cpummx)
+ return (const reg_entry *) NULL;
+
+ if (r->reg_type.bitfield.regxmm && !cpu_arch_flags.bitfield.cpusse)
+ return (const reg_entry *) NULL;
+
+ if (r->reg_type.bitfield.regymm && !cpu_arch_flags.bitfield.cpuavx)
+ return (const reg_entry *) NULL;
+
/* Don't allow fake index register unless allow_index_reg isn't 0. */
- if (r != NULL
- && !allow_index_reg
+ if (!allow_index_reg
&& (r->reg_num == RegEiz || r->reg_num == RegRiz))
return (const reg_entry *) NULL;
- if (r != NULL
- && ((r->reg_flags & (RegRex64 | RegRex))
- || r->reg_type.bitfield.reg64)
+ if (((r->reg_flags & (RegRex64 | RegRex))
+ || r->reg_type.bitfield.reg64)
&& (!cpu_arch_flags.bitfield.cpulm
- || !UINTS_EQUAL (r->reg_type, control))
+ || !operand_type_equal (&r->reg_type, &control))
&& flag_code != CODE_64BIT)
return (const reg_entry *) NULL;
+ if (r->reg_type.bitfield.sreg3 && r->reg_num == RegFlat && !intel_syntax)
+ return (const reg_entry *) NULL;
+
return r;
}
#define OPTION_MINDEX_REG (OPTION_MD_BASE + 7)
#define OPTION_MNAKED_REG (OPTION_MD_BASE + 8)
#define OPTION_MOLD_GCC (OPTION_MD_BASE + 9)
+#define OPTION_MSSE2AVX (OPTION_MD_BASE + 10)
+#define OPTION_MSSE_CHECK (OPTION_MD_BASE + 11)
struct option md_longopts[] =
{
{"mindex-reg", no_argument, NULL, OPTION_MINDEX_REG},
{"mnaked-reg", no_argument, NULL, OPTION_MNAKED_REG},
{"mold-gcc", no_argument, NULL, OPTION_MOLD_GCC},
+ {"msse2avx", no_argument, NULL, OPTION_MSSE2AVX},
+ {"msse-check", required_argument, NULL, OPTION_MSSE_CHECK},
{NULL, no_argument, NULL, 0}
};
size_t md_longopts_size = sizeof (md_longopts);
i386_cpu_flags flags;
flags = cpu_flags_or (cpu_arch_flags,
cpu_arch[i].flags);
- if (!UINTS_EQUAL (flags, cpu_arch_flags))
+ if (!cpu_flags_equal (&flags, &cpu_arch_flags))
{
if (cpu_sub_arch_name)
{
old_gcc = 1;
break;
+ case OPTION_MSSE2AVX:
+ sse2avx = 1;
+ break;
+
+ case OPTION_MSSE_CHECK:
+ if (strcasecmp (arg, "error") == 0)
+ sse_check = sse_check_error;
+ else if (strcasecmp (arg, "warning") == 0)
+ sse_check = sse_check_warning;
+ else if (strcasecmp (arg, "none") == 0)
+ sse_check = sse_check_none;
+ else
+ as_fatal (_("Invalid -msse-check= option: `%s'"), arg);
+ break;
+
default:
return 0;
}
generate code for CPU and EXTENSION, CPU is one of:\n\
i8086, i186, i286, i386, i486, pentium, pentiumpro,\n\
pentiumii, pentiumiii, pentium4, prescott, nocona,\n\
- core, core2, k6, k6_2, athlon, k8, amdfam10,\n\
+ core, core2, corei7, k6, k6_2, athlon, k8, amdfam10,\n\
generic32, generic64\n\
EXTENSION is combination of:\n\
mmx, sse, sse2, sse3, ssse3, sse4.1, sse4.2, sse4,\n\
- vmx, smx, xsave, 3dnow, 3dnowa, sse4a, sse5, svme,\n\
- abm, padlock\n"));
+ avx, vmx, smx, xsave, movbe, ept, aes, pclmul, fma,\n\
+ clflush, syscall, rdtscp, 3dnow, 3dnowa, sse4a,\n\
+ sse5, svme, abm, padlock\n"));
fprintf (stream, _("\
-mtune=CPU optimize for CPU, CPU is one of:\n\
i8086, i186, i286, i386, i486, pentium, pentiumpro,\n\
pentiumii, pentiumiii, pentium4, prescott, nocona,\n\
- core, core2, k6, k6_2, athlon, k8, amdfam10,\n\
+ core, core2, corei7, k6, k6_2, athlon, k8, amdfam10,\n\
generic32, generic64\n"));
fprintf (stream, _("\
+ -msse2avx encode SSE instructions with VEX prefix\n"));
+ fprintf (stream, _("\
+ -msse-check=[none|error|warning]\n\
+ check SSE instructions\n"));
+ fprintf (stream, _("\
-mmnemonic=[att|intel] use AT&T/Intel mnemonic\n"));
fprintf (stream, _("\
-msyntax=[att|intel] use AT&T/Intel syntax\n"));
if (!strcmp (default_arch, "x86_64"))
{
set_code_flag (CODE_64BIT);
- if (UINTS_ALL_ZERO (cpu_arch_isa_flags))
+ if (cpu_flags_all_zero (&cpu_arch_isa_flags))
{
cpu_arch_isa_flags.bitfield.cpui186 = 1;
cpu_arch_isa_flags.bitfield.cpui286 = 1;
cpu_arch_isa_flags.bitfield.cpui486 = 1;
cpu_arch_isa_flags.bitfield.cpui586 = 1;
cpu_arch_isa_flags.bitfield.cpui686 = 1;
- cpu_arch_isa_flags.bitfield.cpup4 = 1;
+ cpu_arch_isa_flags.bitfield.cpuclflush = 1;
cpu_arch_isa_flags.bitfield.cpummx= 1;
cpu_arch_isa_flags.bitfield.cpusse = 1;
cpu_arch_isa_flags.bitfield.cpusse2 = 1;
}
- if (UINTS_ALL_ZERO (cpu_arch_tune_flags))
+ if (cpu_flags_all_zero (&cpu_arch_tune_flags))
{
cpu_arch_tune_flags.bitfield.cpui186 = 1;
cpu_arch_tune_flags.bitfield.cpui286 = 1;
cpu_arch_tune_flags.bitfield.cpui486 = 1;
cpu_arch_tune_flags.bitfield.cpui586 = 1;
cpu_arch_tune_flags.bitfield.cpui686 = 1;
- cpu_arch_tune_flags.bitfield.cpup4 = 1;
+ cpu_arch_tune_flags.bitfield.cpuclflush = 1;
cpu_arch_tune_flags.bitfield.cpummx= 1;
cpu_arch_tune_flags.bitfield.cpusse = 1;
cpu_arch_tune_flags.bitfield.cpusse2 = 1;
else if (!strcmp (default_arch, "i386"))
{
set_code_flag (CODE_32BIT);
- if (UINTS_ALL_ZERO (cpu_arch_isa_flags))
+ if (cpu_flags_all_zero (&cpu_arch_isa_flags))
{
cpu_arch_isa_flags.bitfield.cpui186 = 1;
cpu_arch_isa_flags.bitfield.cpui286 = 1;
cpu_arch_isa_flags.bitfield.cpui386 = 1;
}
- if (UINTS_ALL_ZERO (cpu_arch_tune_flags))
+ if (cpu_flags_all_zero (&cpu_arch_tune_flags))
{
cpu_arch_tune_flags.bitfield.cpui186 = 1;
cpu_arch_tune_flags.bitfield.cpui286 = 1;
{
#ifdef TE_PEP
case bfd_target_coff_flavour:
- return flag_code == CODE_64BIT ? COFF_TARGET_FORMAT : "coff-i386";
+ return flag_code == CODE_64BIT ? COFF_TARGET_FORMAT : "pe-i386";
break;
#endif
#ifdef OBJ_MAYBE_AOUT
constant digits [[ radixOverride ]]
- dataType BYTE | WORD | DWORD | FWORD | QWORD | TBYTE | OWORD | XMMWORD
+ dataType BYTE | WORD | DWORD | FWORD | QWORD | TBYTE | OWORD | XMMWORD | YMMWORD
digits decdigit
| digits decdigit
| TBYTE
| OWORD
| XMMWORD
+ | YMMWORD
| .
| $
| register
#define T_ID 13
#define T_SHL 14
#define T_SHR 15
+#define T_YMMWORD 16
/* Prototypes for intel parser functions. */
static int intel_match_token (int);
{
int ret;
char *p;
+ const reg_entry *final_base = i.base_reg;
+ const reg_entry *final_index = i.index_reg;
p = intel_parser.op_string = xstrdup (operand_string);
intel_parser.disp = (char *) xmalloc (strlen (operand_string) + 1);
intel_parser.disp[0] = '\0';
intel_parser.next_operand = NULL;
+ i.base_reg = NULL;
+ i.index_reg = NULL;
+
/* Read the first token and start the parser. */
intel_get_token ();
ret = intel_expr ();
else
{
char *s = intel_parser.disp;
- i.types[this_operand].bitfield.mem = 1;
- i.mem_operands++;
if (!quiet_warnings && intel_parser.is_mem < 0)
/* See the comments in intel_bracket_expr. */
}
ret = i386_index_check (operand_string);
}
+ if (ret)
+ {
+ i.types[this_operand].bitfield.mem = 1;
+ i.mem_operands++;
+ }
}
}
/* Constant and OFFSET expressions are handled by i386_immediate. */
else if ((intel_parser.op_modifier & (1 << T_OFFSET))
|| intel_parser.reg == NULL)
- ret = i386_immediate (intel_parser.disp);
+ {
+ if (i.mem_operands < 2 && i.seg[i.mem_operands])
+ {
+ if (!(intel_parser.op_modifier & (1 << T_OFFSET)))
+ as_warn (_("Segment override ignored"));
+ i.seg[i.mem_operands] = NULL;
+ }
+ ret = i386_immediate (intel_parser.disp);
+ }
+
+ if (!final_base && !final_index)
+ {
+ final_base = i.base_reg;
+ final_index = i.index_reg;
+ }
if (intel_parser.next_operand && this_operand >= MAX_OPERANDS - 1)
ret = 0;
free (p);
free (intel_parser.disp);
+ if (final_base || final_index)
+ {
+ i.base_reg = final_base;
+ i.index_reg = final_index;
+ }
+
return ret;
}
else if (prev_token.code == T_WORD)
{
- if (current_templates->start->name[0] == 'l'
- && current_templates->start->name[2] == 's'
- && current_templates->start->name[3] == 0)
+ if ((current_templates->start->name[0] == 'l'
+ && current_templates->start->name[2] == 's'
+ && current_templates->start->name[3] == 0)
+ || current_templates->start->base_opcode == 0x62 /* bound */)
suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
else if (intel_parser.got_a_float == 2) /* "fi..." */
suffix = SHORT_MNEM_SUFFIX;
else if (prev_token.code == T_DWORD)
{
- if (current_templates->start->name[0] == 'l'
- && current_templates->start->name[2] == 's'
- && current_templates->start->name[3] == 0)
+ if ((current_templates->start->name[0] == 'l'
+ && current_templates->start->name[2] == 's'
+ && current_templates->start->name[3] == 0)
+ || current_templates->start->base_opcode == 0x62 /* bound */)
suffix = WORD_MNEM_SUFFIX;
else if (flag_code == CODE_16BIT
&& (current_templates->start->opcode_modifier.jump
else if (prev_token.code == T_QWORD)
{
- if (intel_parser.got_a_float == 1) /* "f..." */
+ if (current_templates->start->base_opcode == 0x62 /* bound */
+ || intel_parser.got_a_float == 1) /* "f..." */
suffix = LONG_MNEM_SUFFIX;
else
- suffix = QWORD_MNEM_SUFFIX;
+ suffix = QWORD_MNEM_SUFFIX;
i.types[this_operand].bitfield.qword = 1;
}
i.types[this_operand].bitfield.xmmword = 1;
}
+ else if (prev_token.code == T_YMMWORD)
+ {
+ suffix = YMMWORD_MNEM_SUFFIX;
+ i.types[this_operand].bitfield.ymmword = 1;
+ }
+
else
{
as_bad (_("Unknown operand modifier `%s'"), prev_token.str);
| TBYTE
| OWORD
| XMMWORD
+ | YMMWORD
| $
| .
| register
reg->reg_name);
return 0;
}
+ else if (i.mem_operands >= 2)
+ as_warn (_("Segment override ignored"));
else if (i.seg[i.mem_operands])
as_warn (_("Extra segment override ignored"));
else
}
}
+ else if (reg->reg_type.bitfield.sreg3 && reg->reg_num == RegFlat)
+ {
+ as_bad (_("cannot use `FLAT' here"));
+ return 0;
+ }
+
/* Not a segment register. Check for register scaling. */
else if (cur_token.code == '*')
{
| QWORD
| TBYTE
| OWORD
- | XMMWORD */
+ | XMMWORD
+ | YMMWORD */
case T_BYTE:
case T_WORD:
case T_DWORD:
case T_QWORD:
case T_TBYTE:
case T_XMMWORD:
+ case T_YMMWORD:
intel_match_token (cur_token.code);
if (cur_token.code == T_PTR)
|| strcasecmp (new_token.str, "OWORD") == 0)
new_token.code = T_XMMWORD;
+ else if (strcasecmp (new_token.str, "YMMWORD") == 0)
+ new_token.code = T_YMMWORD;
+
else if (strcasecmp (new_token.str, "PTR") == 0)
new_token.code = T_PTR;
strcat (new_token.str, " FLAT:");
}
- /* ??? This is not mentioned in the MASM grammar. */
- else if (strcasecmp (new_token.str, "FLAT") == 0)
- {
- new_token.code = T_OFFSET;
- if (*q == ':')
- strcat (new_token.str, ":");
- else
- as_bad (_("`:' expected"));
- }
-
else
new_token.code = T_ID;
}
prev_token.str = NULL;
}
-int
-tc_x86_regname_to_dw2regnum (char *regname)
-{
- unsigned int regnum;
- unsigned int regnames_count;
- static const char *const regnames_32[] =
- {
- "eax", "ecx", "edx", "ebx",
- "esp", "ebp", "esi", "edi",
- "eip", "eflags", NULL,
- "st0", "st1", "st2", "st3",
- "st4", "st5", "st6", "st7",
- NULL, NULL,
- "xmm0", "xmm1", "xmm2", "xmm3",
- "xmm4", "xmm5", "xmm6", "xmm7",
- "mm0", "mm1", "mm2", "mm3",
- "mm4", "mm5", "mm6", "mm7",
- "fcw", "fsw", "mxcsr",
- "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
- "tr", "ldtr"
- };
- static const char *const regnames_64[] =
- {
- "rax", "rdx", "rcx", "rbx",
- "rsi", "rdi", "rbp", "rsp",
- "r8", "r9", "r10", "r11",
- "r12", "r13", "r14", "r15",
- "rip",
- "xmm0", "xmm1", "xmm2", "xmm3",
- "xmm4", "xmm5", "xmm6", "xmm7",
- "xmm8", "xmm9", "xmm10", "xmm11",
- "xmm12", "xmm13", "xmm14", "xmm15",
- "st0", "st1", "st2", "st3",
- "st4", "st5", "st6", "st7",
- "mm0", "mm1", "mm2", "mm3",
- "mm4", "mm5", "mm6", "mm7",
- "rflags",
- "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
- "fs.base", "gs.base", NULL, NULL,
- "tr", "ldtr",
- "mxcsr", "fcw", "fsw"
- };
- const char *const *regnames;
+void
+tc_x86_parse_to_dw2regnum (expressionS *exp)
+{
+ int saved_naked_reg;
+ char saved_register_dot;
- if (flag_code == CODE_64BIT)
- {
- regnames = regnames_64;
- regnames_count = ARRAY_SIZE (regnames_64);
- }
- else
+ saved_naked_reg = allow_naked_reg;
+ allow_naked_reg = 1;
+ saved_register_dot = register_chars['.'];
+ register_chars['.'] = '.';
+ allow_pseudo_reg = 1;
+ expression_and_evaluate (exp);
+ allow_pseudo_reg = 0;
+ register_chars['.'] = saved_register_dot;
+ allow_naked_reg = saved_naked_reg;
+
+ if (exp->X_op == O_register && exp->X_add_number >= 0)
{
- regnames = regnames_32;
- regnames_count = ARRAY_SIZE (regnames_32);
+ if ((addressT) exp->X_add_number < i386_regtab_size)
+ {
+ exp->X_op = O_constant;
+ exp->X_add_number = i386_regtab[exp->X_add_number]
+ .dw2_regnum[flag_code >> 1];
+ }
+ else
+ exp->X_op = O_illegal;
}
-
- for (regnum = 0; regnum < regnames_count; regnum++)
- if (regnames[regnum] != NULL
- && strcmp (regname, regnames[regnum]) == 0)
- return regnum;
-
- return -1;
}
void
tc_x86_frame_initial_instructions (void)
{
- static unsigned int sp_regno;
+ static unsigned int sp_regno[2];
+
+ if (!sp_regno[flag_code >> 1])
+ {
+ char *saved_input = input_line_pointer;
+ char sp[][4] = {"esp", "rsp"};
+ expressionS exp;
- if (!sp_regno)
- sp_regno = tc_x86_regname_to_dw2regnum (flag_code == CODE_64BIT
- ? "rsp" : "esp");
+ input_line_pointer = sp[flag_code >> 1];
+ tc_x86_parse_to_dw2regnum (&exp);
+ assert (exp.X_op == O_constant);
+ sp_regno[flag_code >> 1] = exp.X_add_number;
+ input_line_pointer = saved_input;
+ }
- cfi_add_CFA_def_cfa (sp_regno, -x86_cie_data_alignment);
+ cfi_add_CFA_def_cfa (sp_regno[flag_code >> 1], -x86_cie_data_alignment);
cfi_add_CFA_offset (x86_dwarf2_return_column, x86_cie_data_alignment);
}
return -1;
}
+#ifdef TE_SOLARIS
+void
+i386_solaris_fix_up_eh_frame (segT sec)
+{
+ if (flag_code == CODE_64BIT)
+ elf_section_type (sec) = SHT_X86_64_UNWIND;
+}
+#endif
+
#ifdef TE_PE
void
tc_pe_dwarf2_emit_offset (symbolS *symbol, unsigned int size)
#if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
/* For ELF on x86-64, add support for SHF_X86_64_LARGE. */
-int
+bfd_vma
x86_64_section_letter (int letter, char **ptr_msg)
{
if (flag_code == CODE_64BIT)
return -1;
}
-int
+bfd_vma
x86_64_section_word (char *str, size_t len)
{
if (len == 5 && flag_code == CODE_64BIT && CONST_STRNEQ (str, "large"))