Re: i386 intel syntax integer load/store fp instructions

["Jan Beulich" <JBeulich at novell dot com>]

An update to the previously submitted x86 Intel syntax assembly
adjustments patch, now retaining fildq/fistpq/fisttpq in AT&T mode.

Built and tested on i686-pc-linux-gnu.

Thanks, Jan

gas/
2004-10-18 Jan Beulich <jbeulich@novell.com>

	* config/tc-i386.c (set_intel_syntax): Allow % in symbol names
when
	intel syntax and no register prefix, allow $ in symbol names
when
	intel syntax.
	(set_16bit_gcc_code_flag): Replace literal 'l' by
LONG_MNEM_SUFFIX.
	(intel_float_operand): Add fourth return value indicating math
control
	operations. Make classification more precise.
	(md_assemble): Complain if memory operand of mov[sz]x has no
size
	specified.
	(parse_insn): Translate word operands to floating point
instructions
	operating on integers as well as control instructions to short
ones
	as expected by AT&T syntax. Translate 'd' suffix to short one
only for
	floating point instructions operating on non-integer operands.
	(match_template): Remove fldcw special case. Adjust q-suffix
handling
	to permit it on fild/fistp/fisttp in AT&T mode.
	(process_suffix): Don't guess DefaultSize insns' suffix from
	stackop_size for certain floating point control instructions.
Guess
	suffix for branch and [ls][gi]dt based on flag_code. Split
error
	messages for Intel and AT&T syntax, and make the condition more
strict
	for the former. Adjust suppressing of generation of operand
size
	overrides.
	(intel parser): Allow the full set of MASM operators. Add FWORD,
TBYTE,
	OWORD, and XMMWORD operand size specifiers (TBYTE replaces
XWORD). Add
	more error checking.
	* config/tc-i386.h (BYTE_PTR WORD_PTR DWORD_PTR QWORD_PTR
XWORD_PTR
	SHORT OFFSET_FLAT FLAT NONE_FOUND): Remove unused defines.

gas/testsuite/
2004-10-18 Jan Beulich <jbeulich@novell.com>
	* gas/i386/i386.exp: Execute new tests intelbad and intelok.
	* gas/i386/intelbad.[sl]: New test to check for various things
not
	permitted in Intel mode.
	* gas/i386/intel.d, gas/i386/opcode.d,
gas/i386/x86-64-opcode.d:
	Adjust for change to segment register store.
	* gas/i386/intelok.[sd]: New test to check various Intel mode
specific
	things get handled correctly.
	* gas/i386/x86_64.[sd]: Remove unsupported constructs referring
to
	'high' and 'low' parts of an operand, which the parser
previously
	accepted while neither telling that it's not supported nor that
it
	ignored the remainder of the line following these supposed
keywords.

include/opcode/
2004-10-18 Jan Beulich <jbeulich@novell.com>

	* i386.h (sldx_Suf): Remove.
	(FP, l_FP, sl_FP, x_FP): Don't imply IgnoreSize.
	(q_FP): Define, implying no REX64.
	(x_FP, sl_FP): Imply FloatMF.
	(i386_optab): Split reg and mem forms of moving from segment
registers
	so that the memory forms can ignore the 16-/32-bit operand size
	distinction. Adjust a few others for Intel mode. Remove *FP uses
from
	all non-floating-point instructions. Unite 32- and 64-bit forms
of
	movsx, movzx, and movd. Adjust floating point operations for the
above
	changes to the *FP macros. Add DefaultSize to floating point
controls
	insns operating on larger memory ranges. Remove left over
comments
	hinting at certain being Intel-syntax ones where the ones
actually
	meant are already gone.

opcodes/
2004-10-18 Jan Beulich <jbeulich@novell.com>

	* i386-dis.c (Eq, Edqw, indirEp, Gdq, I1): Define.
	(indirEb): Remove.
	(Mp): Use f_mode rather than none at all.
	(t_mode, dq_mode, dqw_mode, f_mode, const_1_mode): Define.
t_mode
	replaces what previously was x_mode; x_mode now means 128-bit
SSE
	operands.
	(dis386): Make far jumps and calls have an 'l' prefix only in
AT&T
	mode. movmskpX's, pextrw's, and pmovmskb's first operands are
Gdq.
	pinsrw's second operand is Edqw.
	(grps): 1-bit shifts' and rotates' second operands are I1.
cmpxchg8b's
	operand is Eq. movntq's and movntdq's first operands are EM.
s[gi]dt,
	fldenv, frstor, fsave, fstenv all should also have suffixes in
Intel
	mode when an operand size override is present or always
suffixing.
	More instructions will need to be added to this group.
	(putop): Handle new macro chars 'C' (short/long suffix
selector),
	'I' (Intel mode override for following macro char), and 'J'
(for
	adding the 'l' prefix to far branches in AT&T mode). When an
	alternative was specified in the template, honor macro character
when
	specified for Intel mode.
	(OP_E): Handle new *_mode values. Correct pointer specifications
for
	memory operands. Consolidate output of index register.
	(OP_G): Handle new *_mode values.
	(OP_I): Handle const_1_mode.
	(OP_ESreg, OP_DSreg): Generate pointer specifications. Indicate
	respective opcode prefix bits have been consumed.
	(OP_EM, OP_EX): Provide some default handling for generating
pointer
	specifications.


>>> Alan Modra <amodra@bigpond.net.au> 15.10.04 07:03:56 >>>
On Wed, Oct 13, 2004 at 10:38:21AM +0200, Jan Beulich wrote:
> Since the patch to remove the (superfluous) Intel syntax fildd,
fistpd,
> and fisttpd instructions (24Jun2004) left intact the comments saying
the
> next instruction would be Intel syntax, these comments now appear to
> apply to fildq, fistpq, and fisttpq respectively. Whether this is
> correct I cannot judge since I haven't been able to locate a formal
AT&T
> syntax specification (in order to see whether these instructions
> actually exist there).

I believe the correct instructions are ones with an 'll' suffix rather
than 'q'.  However, a quick google search shows that people use the
first two instructions in att mode.  fisttpq doesn't get many hits.

I think they should remain valid in att mode.

-- 
Alan Modra
IBM OzLabs - Linux Technology Centre

Attachment: binutils-mainline-x86-intel-syntax.patch
Description: Binary data