* [PATCH 0/3] x86: assorted opcode processing adjustments
@ 2023-02-03 7:30 Jan Beulich
2023-02-03 7:32 ` [PATCH 1/3] x86: improve special casing of certain insns Jan Beulich
` (2 more replies)
0 siblings, 3 replies; 4+ messages in thread
From: Jan Beulich @ 2023-02-03 7:30 UTC (permalink / raw)
To: Binutils; +Cc: H.J. Lu
The first change follows what was already done for Intel syntax code.
The second is finally doing what I was meaning to do for quite some
time, noticing the inefficiencies once again while preparing patch 1.
The third change is perhaps just a first step towards re-arranging
how we represent template attributes; see there for possible further
steps.
1: improve special casing of certain insns
2: simplify a few expressions
3: move (and rename) opcodespace attribute
Jan
^ permalink raw reply [flat|nested] 4+ messages in thread
* [PATCH 1/3] x86: improve special casing of certain insns
2023-02-03 7:30 [PATCH 0/3] x86: assorted opcode processing adjustments Jan Beulich
@ 2023-02-03 7:32 ` Jan Beulich
2023-02-03 7:32 ` [PATCH 2/3] x86: simplify a few expressions Jan Beulich
2023-02-03 7:33 ` [PATCH 3/3] x86: move (and rename) opcodespace attribute Jan Beulich
2 siblings, 0 replies; 4+ messages in thread
From: Jan Beulich @ 2023-02-03 7:32 UTC (permalink / raw)
To: Binutils; +Cc: H.J. Lu
Now that we have identifiers for the mnemonic strings we can avoid
opcode based comparisons, for (in many cases) being more expensive and
(in a few cases) being a little fragile and not self-documenting.
Note that the MOV optimization can be engaged by the earlier LEA one,
and hence LEA also needs checking for there.
--- a/gas/config/tc-i386.c
+++ b/gas/config/tc-i386.c
@@ -3487,8 +3487,7 @@ want_disp32 (const insn_template *t)
{
return flag_code != CODE_64BIT
|| i.prefix[ADDR_PREFIX]
- || (t->base_opcode == 0x8d
- && t->opcode_modifier.opcodespace == SPACE_BASE
+ || (t->mnem_off == MN_lea
&& (!i.types[1].bitfield.qword
|| t->opcode_modifier.size == SIZE32));
}
@@ -4096,8 +4095,7 @@ optimize_encoding (void)
{
unsigned int j;
- if (i.tm.opcode_modifier.opcodespace == SPACE_BASE
- && i.tm.base_opcode == 0x8d)
+ if (i.tm.mnem_off == MN_lea)
{
/* Optimize: -O:
lea symbol, %rN -> mov $symbol, %rN
@@ -4245,15 +4243,12 @@ optimize_encoding (void)
}
if (optimize_for_space
- && i.tm.opcode_modifier.opcodespace == SPACE_BASE
+ && i.tm.mnem_off == MN_test
&& i.reg_operands == 1
&& i.imm_operands == 1
&& !i.types[1].bitfield.byte
&& i.op[0].imms->X_op == O_constant
- && fits_in_imm7 (i.op[0].imms->X_add_number)
- && (i.tm.base_opcode == 0xa8
- || (i.tm.base_opcode == 0xf6
- && i.tm.extension_opcode == 0x0)))
+ && fits_in_imm7 (i.op[0].imms->X_add_number))
{
/* Optimize: -Os:
test $imm7, %r64/%r32/%r16 -> test $imm7, %r8
@@ -4286,12 +4281,11 @@ optimize_encoding (void)
&& i.tm.extension_opcode == None
&& fits_in_unsigned_long (i.op[0].imms->X_add_number))
|| (fits_in_imm31 (i.op[0].imms->X_add_number)
- && ((i.tm.base_opcode == 0x24
- || i.tm.base_opcode == 0xa8)
+ && (i.tm.base_opcode == 0x24
|| (i.tm.base_opcode == 0x80
&& i.tm.extension_opcode == 0x4)
- || ((i.tm.base_opcode == 0xf6
- || (i.tm.base_opcode | 1) == 0xc7)
+ || i.tm.mnem_off == MN_test
+ || ((i.tm.base_opcode | 1) == 0xc7
&& i.tm.extension_opcode == 0x0)))
|| (fits_in_imm7 (i.op[0].imms->X_add_number)
&& i.tm.base_opcode == 0x83
@@ -4299,11 +4293,9 @@ optimize_encoding (void)
|| (i.types[0].bitfield.qword
&& ((i.reg_operands == 2
&& i.op[0].regs == i.op[1].regs
- && (i.tm.base_opcode == 0x30
- || i.tm.base_opcode == 0x28))
- || (i.reg_operands == 1
- && i.operands == 1
- && i.tm.base_opcode == 0x30)))))
+ && (i.tm.mnem_off == MN_xor
+ || i.tm.mnem_off == MN_sub))
+ || i.tm.mnem_off == MN_clr))))
{
/* Optimize: -O:
andq $imm31, %r64 -> andl $imm31, %r32
@@ -4328,7 +4320,7 @@ optimize_encoding (void)
}
i.types[1].bitfield.dword = 1;
i.types[1].bitfield.qword = 0;
- if (i.tm.base_opcode == 0xb8 || (i.tm.base_opcode | 1) == 0xc7)
+ if (i.tm.mnem_off == MN_mov || i.tm.mnem_off == MN_lea)
{
/* Handle
movq $imm31, %r64 -> movl $imm31, %r32
@@ -4351,11 +4343,9 @@ optimize_encoding (void)
}
else if (optimize > 1
&& !optimize_for_space
- && i.tm.opcode_modifier.opcodespace == SPACE_BASE
&& i.reg_operands == 2
&& i.op[0].regs == i.op[1].regs
- && ((i.tm.base_opcode & ~(Opcode_D | 1)) == 0x8
- || (i.tm.base_opcode & ~(Opcode_D | 1)) == 0x20)
+ && (i.tm.mnem_off == MN_and || i.tm.mnem_off == MN_or)
&& (flag_code != CODE_64BIT || !i.types[0].bitfield.dword))
{
/* Optimize: -O2:
@@ -4580,12 +4570,7 @@ load_insn_p (void)
if (any_vex_p)
{
- /* vldmxcsr. */
- if (i.tm.base_opcode == 0xae
- && i.tm.opcode_modifier.vex
- && i.tm.opcode_modifier.opcodespace == SPACE_0F
- && i.tm.opcode_modifier.opcodeprefix == PREFIX_NONE
- && i.tm.extension_opcode == 2)
+ if (i.tm.mnem_off == MN_vldmxcsr)
return 1;
}
else if (i.tm.opcode_modifier.opcodespace == SPACE_BASE)
@@ -4699,9 +4684,7 @@ load_insn_p (void)
|| (base_opcode | 2) == 0x87))
return 1;
- /* xadd. */
- if (i.tm.opcode_modifier.opcodespace == SPACE_0F
- && base_opcode == 0xc1)
+ if (i.tm.mnem_off == MN_xadd)
return 1;
/* Check for load instruction. */
@@ -5361,12 +5344,8 @@ md_assemble (char *line)
i.rex &= REX_OPCODE;
}
- /* 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. */
- if (i.tm.opcode_modifier.opcodespace == SPACE_BASE
- && i.tm.base_opcode == INT_OPCODE
- && !i.tm.opcode_modifier.modrm
+ /* Handle conversion of 'int $3' --> special int3 insn. */
+ if (i.tm.mnem_off == MN_int
&& i.op[0].imms->X_add_number == 3)
{
i.tm.base_opcode = INT3_OPCODE;
@@ -5476,10 +5455,7 @@ static INLINE bool q_suffix_allowed(cons
|| (t->opcode_modifier.opcodespace == SPACE_BASE
&& t->base_opcode == 0xdf
&& (t->extension_opcode & 1)) /* fild / fistp / fisttp */
- || (t->opcode_modifier.opcodespace == SPACE_0F
- && t->base_opcode == 0xc7
- && t->opcode_modifier.opcodeprefix == PREFIX_NONE
- && t->extension_opcode == 1) /* cmpxchg8b */;
+ || t->mnem_off == MN_cmpxchg8b;
}
static const char *
@@ -7201,7 +7177,7 @@ check_string (void)
static int
process_suffix (void)
{
- bool is_crc32 = false, is_movx = false;
+ bool is_movx = false;
/* If matched instruction specifies an explicit instruction mnemonic
suffix, use it. */
@@ -7224,11 +7200,6 @@ process_suffix (void)
&& i.tm.base_opcode == 0x63
&& i.tm.cpu_flags.bitfield.cpu64);
- /* CRC32 */
- is_crc32 = (i.tm.base_opcode == 0xf0
- && i.tm.opcode_modifier.opcodespace == SPACE_0F38
- && i.tm.opcode_modifier.opcodeprefix == PREFIX_0XF2);
-
/* movsx/movzx want only their source operand considered here, for the
ambiguity checking below. The suffix will be replaced afterwards
to represent the destination (register). */
@@ -7236,7 +7207,7 @@ process_suffix (void)
--i.operands;
/* crc32 needs REX.W set regardless of suffix / source operand size. */
- if (is_crc32 && i.tm.operand_types[1].bitfield.qword)
+ if (i.tm.mnem_off == MN_crc32 && i.tm.operand_types[1].bitfield.qword)
i.rex |= REX_W;
/* If there's no instruction mnemonic suffix we try to invent one
@@ -7247,7 +7218,7 @@ process_suffix (void)
Destination register type is more significant than source
register type. crc32 in SSE4.2 prefers source register
type. */
- unsigned int op = is_crc32 ? 1 : i.operands;
+ unsigned int op = i.tm.mnem_off == MN_crc32 ? 1 : i.operands;
while (op--)
if (i.tm.operand_types[op].bitfield.instance == InstanceNone
@@ -7571,8 +7542,7 @@ process_suffix (void)
/* InOutPortReg */
|| i.tm.operand_types[0].bitfield.instance == RegD
|| i.tm.operand_types[1].bitfield.instance == RegD
- /* CRC32 */
- || is_crc32))))
+ || i.tm.mnem_off == MN_crc32))))
i.tm.base_opcode |= 1;
break;
}
@@ -7691,10 +7661,7 @@ check_byte_reg (void)
continue;
/* crc32 only wants its source operand checked here. */
- if (i.tm.base_opcode == 0xf0
- && i.tm.opcode_modifier.opcodespace == SPACE_0F38
- && i.tm.opcode_modifier.opcodeprefix == PREFIX_0XF2
- && op != 0)
+ if (i.tm.mnem_off == MN_crc32 && op != 0)
continue;
/* Any other register is bad. */
@@ -8160,9 +8127,7 @@ process_operands (void)
}
if ((i.seg[0] || i.prefix[SEG_PREFIX])
- && i.tm.base_opcode == 0x8d /* lea */
- && i.tm.opcode_modifier.opcodespace == SPACE_BASE
- && !is_any_vex_encoding(&i.tm))
+ && i.tm.mnem_off == MN_lea)
{
if (!quiet_warnings)
as_warn (_("segment override on `%s' is ineffectual"), insn_name (&i.tm));
@@ -9089,7 +9054,7 @@ output_jump (void)
break;
case 2:
- if (i.tm.base_opcode == 0xc7f8)
+ if (i.tm.mnem_off == MN_xbegin)
fixP->fx_signed = 1;
break;
@@ -9601,10 +9566,8 @@ output_insn (void)
x86_feature_2_used |= GNU_PROPERTY_X86_FEATURE_2_X87;
if ((i.xstate & xstate_mmx)
- || (i.tm.opcode_modifier.opcodespace == SPACE_0F
- && !is_any_vex_encoding (&i.tm)
- && (i.tm.base_opcode == 0x77 /* emms */
- || i.tm.base_opcode == 0x0e /* femms */)))
+ || i.tm.mnem_off == MN_emms
+ || i.tm.mnem_off == MN_femms)
x86_feature_2_used |= GNU_PROPERTY_X86_FEATURE_2_MMX;
if (i.index_reg)
@@ -9648,10 +9611,7 @@ output_insn (void)
if (x86_feature_2_used
|| i.tm.cpu_flags.bitfield.cpucmov
|| i.tm.cpu_flags.bitfield.cpusyscall
- || (i.tm.opcode_modifier.opcodespace == SPACE_0F
- && i.tm.base_opcode == 0xc7
- && i.tm.opcode_modifier.opcodeprefix == PREFIX_NONE
- && i.tm.extension_opcode == 1) /* cmpxchg8b */)
+ || i.tm.mnem_off == MN_cmpxchg8b)
x86_isa_1_used |= GNU_PROPERTY_X86_ISA_1_BASELINE;
if (i.tm.cpu_flags.bitfield.cpusse3
|| i.tm.cpu_flags.bitfield.cpussse3
@@ -11503,12 +11463,9 @@ i386_index_check (const char *operand_st
goto bad_address;
/* bndmk, bndldx, bndstx and mandatory non-vector SIB have special restrictions. */
- if ((t->opcode_modifier.opcodeprefix == PREFIX_0XF3
- && t->opcode_modifier.opcodespace == SPACE_0F
- && t->base_opcode == 0x1b)
- || (t->opcode_modifier.opcodeprefix == PREFIX_NONE
- && t->opcode_modifier.opcodespace == SPACE_0F
- && (t->base_opcode & ~1) == 0x1a)
+ if (t->mnem_off == MN_bndmk
+ || t->mnem_off == MN_bndldx
+ || t->mnem_off == MN_bndstx
|| t->opcode_modifier.sib == SIBMEM)
{
/* They cannot use RIP-relative addressing. */
@@ -11519,9 +11476,7 @@ i386_index_check (const char *operand_st
}
/* bndldx and bndstx ignore their scale factor. */
- if (t->opcode_modifier.opcodeprefix == PREFIX_NONE
- && t->opcode_modifier.opcodespace == SPACE_0F
- && (t->base_opcode & ~1) == 0x1a
+ if ((t->mnem_off == MN_bndldx || t->mnem_off == MN_bndstx)
&& i.log2_scale_factor)
as_warn (_("register scaling is being ignored here"));
}
^ permalink raw reply [flat|nested] 4+ messages in thread
* [PATCH 2/3] x86: simplify a few expressions
2023-02-03 7:30 [PATCH 0/3] x86: assorted opcode processing adjustments Jan Beulich
2023-02-03 7:32 ` [PATCH 1/3] x86: improve special casing of certain insns Jan Beulich
@ 2023-02-03 7:32 ` Jan Beulich
2023-02-03 7:33 ` [PATCH 3/3] x86: move (and rename) opcodespace attribute Jan Beulich
2 siblings, 0 replies; 4+ messages in thread
From: Jan Beulich @ 2023-02-03 7:32 UTC (permalink / raw)
To: Binutils; +Cc: H.J. Lu
Fold adjacent comparisons when, by ORing in a certain mask, the same
effect can be achieved by a single one. In load_insn_p() this extends
to further uses of an already available local variable.
--- a/gas/config/tc-i386.c
+++ b/gas/config/tc-i386.c
@@ -4576,8 +4576,7 @@ load_insn_p (void)
else if (i.tm.opcode_modifier.opcodespace == SPACE_BASE)
{
/* test, not, neg, mul, imul, div, idiv. */
- if ((i.tm.base_opcode == 0xf6 || i.tm.base_opcode == 0xf7)
- && i.tm.extension_opcode != 1)
+ if (base_opcode == 0xf7 && i.tm.extension_opcode != 1)
return 1;
/* inc, dec. */
@@ -4589,13 +4588,12 @@ load_insn_p (void)
return 1;
/* rol, ror, rcl, rcr, shl/sal, shr, sar. */
- if ((base_opcode == 0xc1
- || (i.tm.base_opcode >= 0xd0 && i.tm.base_opcode <= 0xd3))
+ if ((base_opcode == 0xc1 || (base_opcode | 2) == 0xd3)
&& i.tm.extension_opcode != 6)
return 1;
/* Check for x87 instructions. */
- if (base_opcode >= 0xd8 && base_opcode <= 0xdf)
+ if ((base_opcode | 6) == 0xdf)
{
/* Skip fst, fstp, fstenv, fstcw. */
if (i.tm.base_opcode == 0xd9
@@ -4638,7 +4636,7 @@ load_insn_p (void)
{
/* bt, bts, btr, btc. */
if (i.tm.base_opcode == 0xba
- && (i.tm.extension_opcode >= 4 && i.tm.extension_opcode <= 7))
+ && (i.tm.extension_opcode | 3) == 7)
return 1;
/* cmpxchg8b, cmpxchg16b, xrstors, vmptrld. */
@@ -4673,14 +4671,7 @@ load_insn_p (void)
/* add, or, adc, sbb, and, sub, xor, cmp, test, xchg. */
if (i.tm.opcode_modifier.opcodespace == SPACE_BASE
- && (base_opcode == 0x1
- || base_opcode == 0x9
- || base_opcode == 0x11
- || base_opcode == 0x19
- || base_opcode == 0x21
- || base_opcode == 0x29
- || base_opcode == 0x31
- || base_opcode == 0x39
+ && ((base_opcode | 0x38) == 0x39
|| (base_opcode | 2) == 0x87))
return 1;
@@ -4708,8 +4699,7 @@ insert_lfence_after (void)
chosen by the adversary using an LVI method,
then this data-dependent behavior may leak some aspect
of the secret. */
- if (((i.tm.base_opcode | 0x1) == 0xa7
- || (i.tm.base_opcode | 0x1) == 0xaf)
+ if (((i.tm.base_opcode | 0x9) == 0xaf)
&& i.prefix[REP_PREFIX])
{
as_warn (_("`%s` changes flags which would affect control flow behavior"),
@@ -4779,8 +4769,7 @@ insert_lfence_before (void)
/* Output or/not/shl and lfence before near ret. */
if (lfence_before_ret != lfence_before_ret_none
- && (i.tm.base_opcode == 0xc2
- || i.tm.base_opcode == 0xc3))
+ && (i.tm.base_opcode | 1) == 0xc3)
{
if (last_insn.kind != last_insn_other
&& last_insn.seg == now_seg)
^ permalink raw reply [flat|nested] 4+ messages in thread
* [PATCH 3/3] x86: move (and rename) opcodespace attribute
2023-02-03 7:30 [PATCH 0/3] x86: assorted opcode processing adjustments Jan Beulich
2023-02-03 7:32 ` [PATCH 1/3] x86: improve special casing of certain insns Jan Beulich
2023-02-03 7:32 ` [PATCH 2/3] x86: simplify a few expressions Jan Beulich
@ 2023-02-03 7:33 ` Jan Beulich
2 siblings, 0 replies; 4+ messages in thread
From: Jan Beulich @ 2023-02-03 7:33 UTC (permalink / raw)
To: Binutils; +Cc: H.J. Lu
This really isn't a "modifier" and rather ought to live next to the base
opcode anyway. Use the bits we presently have available to fit in the
field, renaming it to opcode_space. As an intended side effect this
helps readability at the use sites, by shortening the references quite a
bit.
In generated code arrange for human readable output, by using the
SPACE_* constants there rather than raw numbers. This may aid debugging
down the road.
---
I'd like to do the same to opcodeprefix, but that's going to be more
intrusive, as we don't have spare bits left in a useful position
(fitting the field right after mnem_off isn't a good idea imo). Hence
I'm considering to do away with i386_opcode_modifier as a structure
altogether, which would tidy use sites quite a bit (like is the case
here). Plus it would shrink total struct size (when, at the same time,
reducing mnem_off to a 16-bit bitfield) and hence overall table size.
Thoughts anyone?
--- a/gas/config/tc-i386.c
+++ b/gas/config/tc-i386.c
@@ -3203,8 +3203,8 @@ pte (insn_template *t)
fprintf (stdout, " %d operands ", t->operands);
if (opc_pfx[t->opcode_modifier.opcodeprefix])
fprintf (stdout, "pfx %x ", opc_pfx[t->opcode_modifier.opcodeprefix]);
- if (opc_spc[t->opcode_modifier.opcodespace])
- fprintf (stdout, "space %s ", opc_spc[t->opcode_modifier.opcodespace]);
+ if (opc_spc[t->opcode_space])
+ fprintf (stdout, "space %s ", opc_spc[t->opcode_space]);
fprintf (stdout, "opcode %x ", t->base_opcode);
if (t->extension_opcode != None)
fprintf (stdout, "ext %x ", t->extension_opcode);
@@ -3587,7 +3587,7 @@ build_vex_prefix (const insn_template *t
&& i.dir_encoding == dir_encoding_default
&& i.operands == i.reg_operands
&& operand_type_equal (&i.types[0], &i.types[i.operands - 1])
- && i.tm.opcode_modifier.opcodespace == SPACE_0F
+ && i.tm.opcode_space == SPACE_0F
&& (i.tm.opcode_modifier.load || i.tm.opcode_modifier.d)
&& i.rex == REX_B)
{
@@ -3633,7 +3633,7 @@ build_vex_prefix (const insn_template *t
union i386_op temp_op;
i386_operand_type temp_type;
- gas_assert (i.tm.opcode_modifier.opcodespace == SPACE_0F);
+ gas_assert (i.tm.opcode_space == SPACE_0F);
gas_assert (!i.tm.opcode_modifier.sae);
gas_assert (operand_type_equal (&i.types[i.operands - 2],
&i.types[i.operands - 3]));
@@ -3686,7 +3686,7 @@ build_vex_prefix (const insn_template *t
/* Use 2-byte VEX prefix if possible. */
if (w == 0
&& i.vec_encoding != vex_encoding_vex3
- && i.tm.opcode_modifier.opcodespace == SPACE_0F
+ && i.tm.opcode_space == SPACE_0F
&& (i.rex & (REX_W | REX_X | REX_B)) == 0)
{
/* 2-byte VEX prefix. */
@@ -3707,7 +3707,7 @@ build_vex_prefix (const insn_template *t
/* 3-byte VEX prefix. */
i.vex.length = 3;
- switch (i.tm.opcode_modifier.opcodespace)
+ switch (i.tm.opcode_space)
{
case SPACE_0F:
case SPACE_0F38:
@@ -3725,7 +3725,7 @@ build_vex_prefix (const insn_template *t
/* 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 | i.tm.opcode_modifier.opcodespace;
+ i.vex.bytes[1] = (~i.rex & 0x7) << 5 | i.tm.opcode_space;
i.vex.bytes[2] = (w << 7
| register_specifier << 3
@@ -3860,9 +3860,9 @@ build_evex_prefix (void)
/* The high 3 bits of the second EVEX byte are 1's compliment of RXB
bits from REX. */
- gas_assert (i.tm.opcode_modifier.opcodespace >= SPACE_0F);
- gas_assert (i.tm.opcode_modifier.opcodespace <= SPACE_EVEXMAP6);
- i.vex.bytes[1] = (~i.rex & 0x7) << 5 | i.tm.opcode_modifier.opcodespace;
+ gas_assert (i.tm.opcode_space >= SPACE_0F);
+ gas_assert (i.tm.opcode_space <= SPACE_EVEXMAP6);
+ i.vex.bytes[1] = (~i.rex & 0x7) << 5 | i.tm.opcode_space;
/* The fifth bit of the second EVEX byte is 1's compliment of the
REX_R bit in VREX. */
@@ -4072,14 +4072,14 @@ encode_with_unaligned_vector_move (void)
case 0x28: /* Load instructions. */
case 0x29: /* Store instructions. */
/* movaps/movapd/vmovaps/vmovapd. */
- if (i.tm.opcode_modifier.opcodespace == SPACE_0F
+ if (i.tm.opcode_space == SPACE_0F
&& i.tm.opcode_modifier.opcodeprefix <= PREFIX_0X66)
i.tm.base_opcode = 0x10 | (i.tm.base_opcode & 1);
break;
case 0x6f: /* Load instructions. */
case 0x7f: /* Store instructions. */
/* movdqa/vmovdqa/vmovdqa64/vmovdqa32. */
- if (i.tm.opcode_modifier.opcodespace == SPACE_0F
+ if (i.tm.opcode_space == SPACE_0F
&& i.tm.opcode_modifier.opcodeprefix == PREFIX_0X66)
i.tm.opcode_modifier.opcodeprefix = PREFIX_0XF3;
break;
@@ -4221,7 +4221,7 @@ optimize_encoding (void)
{
if (flag_code != CODE_32BIT)
return;
- i.tm.opcode_modifier.opcodespace = SPACE_0F;
+ i.tm.opcode_space = SPACE_0F;
i.tm.base_opcode = 0xb7;
}
else
@@ -4272,7 +4272,7 @@ optimize_encoding (void)
}
}
else if (flag_code == CODE_64BIT
- && i.tm.opcode_modifier.opcodespace == SPACE_BASE
+ && i.tm.opcode_space == SPACE_BASE
&& ((i.types[1].bitfield.qword
&& i.reg_operands == 1
&& i.imm_operands == 1
@@ -4374,7 +4374,7 @@ optimize_encoding (void)
|| i.tm.cpu_flags.bitfield.cpuavx512vl
|| (i.tm.operand_types[2].bitfield.zmmword
&& i.types[2].bitfield.ymmword))))
- && i.tm.opcode_modifier.opcodespace == SPACE_0F
+ && i.tm.opcode_space == SPACE_0F
&& ((i.tm.base_opcode | 2) == 0x57
|| i.tm.base_opcode == 0xdf
|| i.tm.base_opcode == 0xef
@@ -4545,7 +4545,7 @@ load_insn_p (void)
return 1;
}
- if (i.tm.opcode_modifier.opcodespace == SPACE_BASE)
+ if (i.tm.opcode_space == SPACE_BASE)
{
/* popf, popa. */
if (i.tm.base_opcode == 0x9d
@@ -4573,7 +4573,7 @@ load_insn_p (void)
if (i.tm.mnem_off == MN_vldmxcsr)
return 1;
}
- else if (i.tm.opcode_modifier.opcodespace == SPACE_BASE)
+ else if (i.tm.opcode_space == SPACE_BASE)
{
/* test, not, neg, mul, imul, div, idiv. */
if (base_opcode == 0xf7 && i.tm.extension_opcode != 1)
@@ -4632,7 +4632,7 @@ load_insn_p (void)
return 1;
}
}
- else if (i.tm.opcode_modifier.opcodespace == SPACE_0F)
+ else if (i.tm.opcode_space == SPACE_0F)
{
/* bt, bts, btr, btc. */
if (i.tm.base_opcode == 0xba
@@ -4670,7 +4670,7 @@ load_insn_p (void)
dest--;
/* add, or, adc, sbb, and, sub, xor, cmp, test, xchg. */
- if (i.tm.opcode_modifier.opcodespace == SPACE_BASE
+ if (i.tm.opcode_space == SPACE_BASE
&& ((base_opcode | 0x38) == 0x39
|| (base_opcode | 2) == 0x87))
return 1;
@@ -4719,7 +4719,7 @@ insert_lfence_before (void)
{
char *p;
- if (i.tm.opcode_modifier.opcodespace != SPACE_BASE)
+ if (i.tm.opcode_space != SPACE_BASE)
return;
if (i.tm.base_opcode == 0xff
@@ -4841,9 +4841,9 @@ static INLINE bool may_need_pass2 (const
return t->opcode_modifier.sse2avx
/* Note that all SSE2AVX templates have at least one operand. */
? t->operand_types[t->operands - 1].bitfield.class == RegSIMD
- : (t->opcode_modifier.opcodespace == SPACE_0F
+ : (t->opcode_space == SPACE_0F
&& (t->base_opcode | 1) == 0xbf)
- || (t->opcode_modifier.opcodespace == SPACE_BASE
+ || (t->opcode_space == SPACE_BASE
&& t->base_opcode == 0x63);
}
@@ -5116,8 +5116,8 @@ md_assemble (char *line)
if (sse_check != check_none
/* The opcode space check isn't strictly needed; it's there only to
bypass the logic below when easily possible. */
- && t->opcode_modifier.opcodespace >= SPACE_0F
- && t->opcode_modifier.opcodespace <= SPACE_0F3A
+ && t->opcode_space >= SPACE_0F
+ && t->opcode_space <= SPACE_0F3A
&& !i.tm.cpu_flags.bitfield.cpusse4a
&& !is_any_vex_encoding (t))
{
@@ -5240,7 +5240,7 @@ md_assemble (char *line)
instructions (base opcodes: 0x6c, 0x6e, 0xec, 0xee). */
if (i.input_output_operand
&& ((i.tm.base_opcode | 0x82) != 0xee
- || i.tm.opcode_modifier.opcodespace != SPACE_BASE))
+ || i.tm.opcode_space != SPACE_BASE))
{
as_bad (_("input/output port address isn't allowed with `%s'"),
insn_name (&i.tm));
@@ -5441,7 +5441,7 @@ md_assemble (char *line)
static INLINE bool q_suffix_allowed(const insn_template *t)
{
return flag_code == CODE_64BIT
- || (t->opcode_modifier.opcodespace == SPACE_BASE
+ || (t->opcode_space == SPACE_BASE
&& t->base_opcode == 0xdf
&& (t->extension_opcode & 1)) /* fild / fistp / fisttp */
|| t->mnem_off == MN_cmpxchg8b;
@@ -5678,7 +5678,7 @@ parse_insn (const char *line, char *mnem
if (current_templates != NULL
/* MOVS or CMPS */
&& (current_templates->start->base_opcode | 2) == 0xa6
- && current_templates->start->opcode_modifier.opcodespace
+ && current_templates->start->opcode_space
== SPACE_BASE
&& mnem_p[-2] == 's')
{
@@ -6853,14 +6853,14 @@ match_template (char mnem_suffix)
zero-extend %eax to %rax. */
if (flag_code == CODE_64BIT
&& t->base_opcode == 0x90
- && t->opcode_modifier.opcodespace == SPACE_BASE
+ && t->opcode_space == SPACE_BASE
&& i.types[0].bitfield.instance == Accum
&& i.types[0].bitfield.dword
&& i.types[1].bitfield.instance == Accum)
continue;
if (t->base_opcode == MOV_AX_DISP32
- && t->opcode_modifier.opcodespace == SPACE_BASE
+ && t->opcode_space == SPACE_BASE
&& t->mnem_off != MN_movabs)
{
/* Force 0x8b encoding for "mov foo@GOT, %eax". */
@@ -6972,8 +6972,8 @@ match_template (char mnem_suffix)
found_reverse_match = Opcode_VexW;
goto check_operands_345;
}
- else if (t->opcode_modifier.opcodespace != SPACE_BASE
- && (t->opcode_modifier.opcodespace != SPACE_0F
+ else if (t->opcode_space != SPACE_BASE
+ && (t->opcode_space != SPACE_0F
/* MOV to/from CR/DR/TR, as an exception, follow
the base opcode space encoding model. */
|| (t->base_opcode | 7) != 0x27))
@@ -7183,9 +7183,9 @@ process_suffix (void)
unsigned int numop = i.operands;
/* MOVSX/MOVZX */
- is_movx = (i.tm.opcode_modifier.opcodespace == SPACE_0F
+ is_movx = (i.tm.opcode_space == SPACE_0F
&& (i.tm.base_opcode | 8) == 0xbe)
- || (i.tm.opcode_modifier.opcodespace == SPACE_BASE
+ || (i.tm.opcode_space == SPACE_BASE
&& i.tm.base_opcode == 0x63
&& i.tm.cpu_flags.bitfield.cpu64);
@@ -7289,7 +7289,7 @@ process_suffix (void)
&& (i.tm.opcode_modifier.jump == JUMP_ABSOLUTE
|| i.tm.opcode_modifier.jump == JUMP_BYTE
|| i.tm.opcode_modifier.jump == JUMP_INTERSEGMENT
- || (i.tm.opcode_modifier.opcodespace == SPACE_0F
+ || (i.tm.opcode_space == SPACE_0F
&& i.tm.base_opcode == 0x01 /* [ls][gi]dt */
&& i.tm.extension_opcode <= 3)))
{
@@ -7511,7 +7511,7 @@ process_suffix (void)
need rex64. */
&& ! (i.operands == 2
&& i.tm.base_opcode == 0x90
- && i.tm.opcode_modifier.opcodespace == SPACE_BASE
+ && i.tm.opcode_space == SPACE_BASE
&& i.types[0].bitfield.instance == Accum
&& i.types[0].bitfield.qword
&& i.types[1].bitfield.instance == Accum))
@@ -8083,14 +8083,14 @@ process_operands (void)
return 0;
}
if (i.op[0].regs->reg_num > 3
- && i.tm.opcode_modifier.opcodespace == SPACE_BASE )
+ && i.tm.opcode_space == SPACE_BASE )
{
i.tm.base_opcode ^= (POP_SEG_SHORT ^ POP_SEG386_SHORT) & 0xff;
- i.tm.opcode_modifier.opcodespace = SPACE_0F;
+ i.tm.opcode_space = SPACE_0F;
}
i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
}
- else if (i.tm.opcode_modifier.opcodespace == SPACE_BASE
+ else if (i.tm.opcode_space == SPACE_BASE
&& (i.tm.base_opcode & ~3) == MOV_AX_DISP32)
{
default_seg = reg_ds;
@@ -9325,7 +9325,7 @@ maybe_fused_with_jcc_p (enum mf_cmp_kind
return 0;
/* No opcodes outside of base encoding space. */
- if (i.tm.opcode_modifier.opcodespace != SPACE_BASE)
+ if (i.tm.opcode_space != SPACE_BASE)
return 0;
/* add, sub without add/sub m, imm. */
@@ -9449,7 +9449,7 @@ add_branch_padding_frag_p (enum align_br
if (!align_branch_power
|| now_seg == absolute_section
|| !cpu_arch_flags.bitfield.cpui386
- || i.tm.opcode_modifier.opcodespace != SPACE_BASE)
+ || i.tm.opcode_space != SPACE_BASE)
return 0;
add_padding = 0;
@@ -9611,7 +9611,7 @@ output_insn (void)
/* LAHF-SAHF insns in 64-bit mode. */
|| (flag_code == CODE_64BIT
&& (i.tm.base_opcode | 1) == 0x9f
- && i.tm.opcode_modifier.opcodespace == SPACE_BASE))
+ && i.tm.opcode_space == SPACE_BASE))
x86_isa_1_used |= GNU_PROPERTY_X86_ISA_1_V2;
if (i.tm.cpu_flags.bitfield.cpuavx
|| i.tm.cpu_flags.bitfield.cpuavx2
@@ -9870,7 +9870,7 @@ output_insn (void)
/* Now the opcode; be careful about word order here! */
j = i.opcode_length;
if (!i.vex.length)
- switch (i.tm.opcode_modifier.opcodespace)
+ switch (i.tm.opcode_space)
{
case SPACE_BASE:
break;
@@ -9895,11 +9895,11 @@ output_insn (void)
{
p = frag_more (j);
if (!i.vex.length
- && i.tm.opcode_modifier.opcodespace != SPACE_BASE)
+ && i.tm.opcode_space != SPACE_BASE)
{
*p++ = 0x0f;
- if (i.tm.opcode_modifier.opcodespace != SPACE_0F)
- *p++ = i.tm.opcode_modifier.opcodespace == SPACE_0F38
+ if (i.tm.opcode_space != SPACE_0F)
+ *p++ = i.tm.opcode_space == SPACE_0F38
? 0x38 : 0x3a;
}
@@ -9971,7 +9971,7 @@ output_insn (void)
/* Count prefixes for extended opcode maps. */
if (!i.vex.length)
- switch (i.tm.opcode_modifier.opcodespace)
+ switch (i.tm.opcode_space)
{
case SPACE_BASE:
break;
@@ -10209,7 +10209,7 @@ output_disp (fragS *insn_start_frag, off
&& i.rm.regmem == 5))
&& (i.rm.mode == 2
|| (i.rm.mode == 0 && i.rm.regmem == 5))
- && i.tm.opcode_modifier.opcodespace == SPACE_BASE
+ && i.tm.opcode_space == SPACE_BASE
&& ((i.operands == 1
&& i.tm.base_opcode == 0xff
&& (i.rm.reg == 2 || i.rm.reg == 4))
--- a/opcodes/i386-gen.c
+++ b/opcodes/i386-gen.c
@@ -430,7 +430,6 @@ static bitfield opcode_modifiers[] =
BITFIELD (Vex),
BITFIELD (VexVVVV),
BITFIELD (VexW),
- BITFIELD (OpcodeSpace),
BITFIELD (OpcodePrefix),
BITFIELD (VexSources),
BITFIELD (SIB),
@@ -957,10 +956,24 @@ get_element_size (char **opnd, int linen
static void
process_i386_opcode_modifier (FILE *table, char *mod, unsigned int space,
- unsigned int prefix, char **opnd, int lineno)
+ unsigned int prefix, const char *extension_opcode,
+ char **opnd, int lineno)
{
char *str, *next, *last;
bitfield modifiers [ARRAY_SIZE (opcode_modifiers)];
+ static const char *const spaces[] = {
+#define SPACE(n) [SPACE_##n] = #n
+ SPACE(BASE),
+ SPACE(0F),
+ SPACE(0F38),
+ SPACE(0F3A),
+ SPACE(EVEXMAP5),
+ SPACE(EVEXMAP6),
+ SPACE(XOP08),
+ SPACE(XOP09),
+ SPACE(XOP0A),
+#undef SPACE
+ };
active_isstring = 0;
@@ -978,6 +991,34 @@ process_i386_opcode_modifier (FILE *tabl
if (str)
{
int val = 1;
+
+ if (strncmp(str, "OpcodeSpace", 11) == 0)
+ {
+ char *end;
+
+ if (str[11] != '=')
+ fail ("%s:%d: Missing value for `OpcodeSpace'\n",
+ filename, lineno);
+
+ val = strtol (str + 12, &end, 0);
+ if (*end)
+ fail ("%s:%d: Bogus value `%s' for `OpcodeSpace'\n",
+ filename, lineno, end);
+
+ if (space)
+ {
+ if (val != space)
+ fail ("%s:%d: Conflicting opcode space specifications\n",
+ filename, lineno);
+ fprintf (stderr,
+ "%s:%d: Warning: redundant opcode space specification\n",
+ filename, lineno);
+ }
+
+ space = val;
+ continue;
+ }
+
if (strcasecmp(str, "Broadcast") == 0)
val = get_element_size (opnd, lineno) + BYTE_BROADCAST;
else if (strcasecmp(str, "Disp8MemShift") == 0)
@@ -1002,19 +1043,6 @@ process_i386_opcode_modifier (FILE *tabl
}
}
- if (space)
- {
- if (!modifiers[OpcodeSpace].value)
- modifiers[OpcodeSpace].value = space;
- else if (modifiers[OpcodeSpace].value != space)
- fail ("%s:%d: Conflicting opcode space specifications\n",
- filename, lineno);
- else
- fprintf (stderr,
- "%s:%d: Warning: redundant opcode space specification\n",
- filename, lineno);
- }
-
if (prefix)
{
if (!modifiers[OpcodePrefix].value)
@@ -1038,6 +1066,13 @@ process_i386_opcode_modifier (FILE *tabl
"%s: %d: W modifier without Word/Dword/Qword operand(s)\n",
filename, lineno);
}
+
+ if (space >= ARRAY_SIZE (spaces) || !spaces[space])
+ fail ("%s:%d: Unknown opcode space %u\n", filename, lineno, space);
+
+ fprintf (table, " SPACE_%s, %s,\n",
+ spaces[space], extension_opcode ? extension_opcode : "None");
+
output_opcode_modifier (table, modifiers, ARRAY_SIZE (modifiers));
}
@@ -1257,13 +1292,12 @@ output_i386_opcode (FILE *table, const c
filename, lineno, name, 2 * length, opcode);
ident = mkident (name);
- fprintf (table, " { MN_%s, 0x%0*llx%s, %lu, %s,\n",
- ident, 2 * (int)length, opcode, end, i,
- extension_opcode ? extension_opcode : "None");
+ fprintf (table, " { MN_%s, 0x%0*llx%s, %u,",
+ ident, 2 * (int)length, opcode, end, i);
free (ident);
process_i386_opcode_modifier (table, opcode_modifier, space, prefix,
- operand_types, lineno);
+ extension_opcode, operand_types, lineno);
process_i386_cpu_flag (table, cpu_flags, NULL, ",", " ", lineno, CpuMax);
--- a/opcodes/i386-opc.h
+++ b/opcodes/i386-opc.h
@@ -605,28 +605,6 @@ enum
#define VEXW1 2
#define VEXWIG 3
VexW,
- /* Opcode encoding space (values chosen to be usable directly in
- VEX/XOP mmmmm and EVEX mm fields):
- 0: Base opcode space.
- 1: 0F opcode prefix / space.
- 2: 0F38 opcode prefix / space.
- 3: 0F3A opcode prefix / space.
- 5: EVEXMAP5 opcode prefix / space.
- 6: EVEXMAP6 opcode prefix / space.
- 8: XOP 08 opcode space.
- 9: XOP 09 opcode space.
- A: XOP 0A opcode space.
- */
-#define SPACE_BASE 0
-#define SPACE_0F 1
-#define SPACE_0F38 2
-#define SPACE_0F3A 3
-#define SPACE_EVEXMAP5 5
-#define SPACE_EVEXMAP6 6
-#define SPACE_XOP08 8
-#define SPACE_XOP09 9
-#define SPACE_XOP0A 0xA
- OpcodeSpace,
/* Opcode prefix (values chosen to be usable directly in
VEX/XOP/EVEX pp fields):
0: None
@@ -759,7 +737,6 @@ typedef struct i386_opcode_modifier
unsigned int vex:2;
unsigned int vexvvvv:2;
unsigned int vexw:2;
- unsigned int opcodespace:4;
unsigned int opcodeprefix:2;
unsigned int vexsources:2;
unsigned int sib:3;
@@ -942,8 +919,29 @@ typedef struct insn_template
/* how many operands */
unsigned int operands:3;
- /* spare bits */
- unsigned int :4;
+ /* opcode space */
+ unsigned int opcode_space:4;
+ /* Opcode encoding space (values chosen to be usable directly in
+ VEX/XOP mmmmm and EVEX mm fields):
+ 0: Base opcode space.
+ 1: 0F opcode prefix / space.
+ 2: 0F38 opcode prefix / space.
+ 3: 0F3A opcode prefix / space.
+ 5: EVEXMAP5 opcode prefix / space.
+ 6: EVEXMAP6 opcode prefix / space.
+ 8: XOP 08 opcode space.
+ 9: XOP 09 opcode space.
+ A: XOP 0A opcode space.
+ */
+#define SPACE_BASE 0
+#define SPACE_0F 1
+#define SPACE_0F38 2
+#define SPACE_0F3A 3
+#define SPACE_EVEXMAP5 5
+#define SPACE_EVEXMAP6 6
+#define SPACE_XOP08 8
+#define SPACE_XOP09 9
+#define SPACE_XOP0A 0xA
/* (Fake) base opcode value for pseudo prefixes. */
#define PSEUDO_PREFIX 0
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2023-02-03 7:33 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-02-03 7:30 [PATCH 0/3] x86: assorted opcode processing adjustments Jan Beulich
2023-02-03 7:32 ` [PATCH 1/3] x86: improve special casing of certain insns Jan Beulich
2023-02-03 7:32 ` [PATCH 2/3] x86: simplify a few expressions Jan Beulich
2023-02-03 7:33 ` [PATCH 3/3] x86: move (and rename) opcodespace attribute Jan Beulich
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).