[PATCH] Add vpermil2pd/vpermil2ps for AMD Orochi processor

[Quentin Neill]

On Wed, Feb 10, 2010 at 10:34 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
> On Wed, Feb 10, 2010 at 8:20 AM, Quentin Neill
> <quentin.neill.gnu@gmail.com> wrote:
>> On Wed, Feb 10, 2010 at 9:55 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
>>> On Wed, Feb 10, 2010 at 7:38 AM, Quentin Neill
>>> <quentin.neill.gnu@gmail.com> wrote:
>>>> This patch restores vpermil2pd/vpermil2ps insns with 5 operand support.
>>>>
>>>> I started with the original Intel code removed back in
>>>>    http://sourceware.org/ml/binutils-cvs/2009-01/msg00025.html
>>>> and then adjusted for the AMD implementation.
>>>>
>>>> Tested with make -k check.
>>>>
>>>> Okay to commit?
>>>>
>>>> gas:
>>>>        * config/tc-i386.c (vex_imm4) New operand type.
>>>>        (fits_in_imm4): New.
>>>>        (VEX_check_operands): New.
>>>>        (check_reverse): Call VEX_check_operands.
>>>>        (build_modrm_byte): Reintroduce code for 5
>>>>        operand insns.  Fix whitespace.
>>>>
>>>> gas/testsuite:
>>>>        * gas/i386/x86-64-xop.d: Add vpermil2p[sd] tests.
>>>>        * testsuite/gas/i386/x86-64-xop.s: Likewise.
>>>>        * testsuite/gas/i386/xop.d: Likewise.
>>>>        * testsuite/gas/i386/xop.s: Likewise.
>>>>
>>>> opcodes:
>>>>        * i386-dis.c (OP_EX_VexImmW): Reintroduced
>>>>        function to handle 5th imm8 operand.
>>>>        (PREFIX_VEX_3A48): Added.
>>>>        (PREFIX_VEX_3A49): Added.
>>>>        (VEX_W_3A48_P_2): Added.
>>>>        (VEX_W_3A49_P_2): Added.
>>>>        (prefix table): Added entries for PREFIX_VEX_3A48
>>>>        and PREFIX_VEX_3A49.
>>>>        (vex table): Added entries for VEX_W_3A48_P_2 and
>>>>        and VEX_W_3A49_P_2.
>>>>        * i386-gen.c (operand_type_init): Added OPERAND_TYPE_VEX_IMM4
>>>>        for VEX_Imm4 operands.
>>>>        * i386-opc.h (enum): Added Vex_Imm4.
>>>>        (i386_operand_type): Added vex_imm4.
>>>>        * i386-opc.tbl: Add entries for vpermilp[ds].
>>>>        * i386-init.h: Regenerated.
>>>>        * i386-tbl.h: Regenerated.
>>>>
>>>
>>> Please change VEX_Imm4 to Vec_Imm4.
>>>
>>> What is // in
>>>
>>> +      gas_assert ((i.reg_operands == 4
>>> +                   || (i.reg_operands == 3 && i.mem_operands == 1))
>>> +                  && i.tm.opcode_modifier.vexvvvv == VEXXDS
>>> +                  && i.tm.opcode_modifier.veximmext
>>> +                  && (operand_type_equal (&i.tm.operand_types[dest], &regxmm)
>>> +                      || operand_type_equal
>>> (&i.tm.operand_types[dest], &regymm))
>>> +                  && ((dest == 4
>>> +                       && i.imm_operands == 1
>>> +                       //&& i.types[0].bitfield.veximmext
>>>                       ^^^^^^^^^^^^^^^^^^^^^^^^^
>>> +                       && (i.tm.opcode_modifier.vexw == VEXW0
>>> +                           || i.tm.opcode_modifier.vexw == VEXW1))
>>> +                      || (dest == 3
>>> +                          && (i.imm_operands == 0
>>> +                              || (i.imm_operands == 1
>>> +                                  && i.tm.opcode_modifier.immext))
>>> +                          && i.tm.opcode_modifier.veximmext)));
>>>
>>> for?
>>> --
>>> H.J.
>>
>> Hi H.J.
>>
>> Thanks for the quick reply.  You wrote:
>>
>>> Please change VEX_Imm4 to Vec_Imm4
>>
>> Did you mean "Vex_Imm4" (spelled with x, not c)?
>
> I mean Vec_Imm4 with c not x. I will add other Vec_ImmX in the future.
>
>>
>>> +                       //&& i.types[0].bitfield.veximmext
>>>                       ^^^^^^^^^^^^^^^^^^^^^^^^^
>>
>> As for the commented conditional, it was left unintentionally in the
>> patch, but I am revisiting the logic in this gas_assert.
>>
>
> Please submit a new patch.
>
> Thanks.
> --
> H.J.

Changed Vex_Imm4 to Vec_Imm4.

Also updated the gas_assert() in build_modrm_byte.

Tested with make -k check.

Okay to commit?

gas:
	* config/tc-i386.c (vec_imm4) New operand type.
	(fits_in_imm4): New.
	(VEX_check_operands): New.
	(check_reverse): Call VEX_check_operands.
	(build_modrm_byte): Reintroduce code for 5
	operand insns.  Fix whitespace.

gas/testsuite:
	* gas/i386/x86-64-xop.d: Add vpermil2p[sd] tests.
	* gas/i386/x86-64-xop.s: Likewise.
	* gas/i386/xop.d: Likewise.
	* gas/i386/xop.s: Likewise.

opcodes:
	* i386-dis.c (OP_EX_VexImmW): Reintroduced
	function to handle 5th imm8 operand.
	(PREFIX_VEX_3A48): Added.
	(PREFIX_VEX_3A49): Added.
	(VEX_W_3A48_P_2): Added.
	(VEX_W_3A49_P_2): Added.
	(prefix table): Added entries for PREFIX_VEX_3A48
	and PREFIX_VEX_3A49.
	(vex table): Added entries for VEX_W_3A48_P_2 and
	and VEX_W_3A49_P_2.
	* i386-gen.c (operand_type_init): Added OPERAND_TYPE_VEC_IMM4
	for Vec_Imm4 operands.
	* i386-opc.h (enum): Added Vec_Imm4.
	(i386_operand_type): Added vec_imm4.
	* i386-opc.tbl: Add entries for vpermilp[ds].
	* i386-init.h: Regenerated.
	* i386-tbl.h: Regenerated.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 7009_vpermil2px.diff
Type: application/octet-stream
Size: 40904 bytes
Desc: not available
URL: <https://sourceware.org/pipermail/binutils/attachments/20100210/adbee74a/attachment.obj>