Re: [PATCH v3] Support AVX10_V2_AUX instructions

Jan Beulich <[email protected]>
Newsgroups gmane.comp.gnu.binutils
Message-ID <[email protected]>
On 27.08.2026 09:13, Jiang, Haochen wrote:
>> From: Rohan Shenoy <[email protected]>
>> Sent: Thursday, August 27, 2026 2:44 PM
>>
>> diff --git a/opcodes/i386-dis-evex-prefix.h b/opcodes/i386-dis-evex-prefix.h
>> index 1aeac210c5d..0135703912a 100644
>> --- a/opcodes/i386-dis-evex-prefix.h
>> +++ b/opcodes/i386-dis-evex-prefix.h
>> +  /* PREFIX_EVEX_MAP5_3C_W_0 */
>> +  {
>> +    { Bad_Opcode },
>> +    { "vcvthf82hf6s",	{ XM, Ux }, 0 },
>> +  },
>> +  /* PREFIX_EVEX_MAP5_3D */
>> +  {
>> +    { Bad_Opcode },
>> +    { "vcvt%HB82bf4s",	{ EXxmmq, XM }, 0 },
>> +  },
>> +  /* PREFIX_EVEX_MAP5_3E_W_1 */
>> +  {
>> +    { Bad_Opcode },
>> +    { "vcvtbf82bf6s",	{ XM, Ux }, 0 },
>> +  },
> 
> 
> We need to hold the patch (or part of the patch) for these
> insns.
> 
> This line
>> +    { "vcvt%HB82bf4s",	{ EXxmmq, XM }, 0 },
> is not matching the documentation where documentation
> uses modrm.reg for operand1 and modrm.r/m for operand2.
> It is a documentation bug since we have memory for operand1.
> 
> While for those FP8 to FP6 insns, I am not sure if the current
> documentation is the correct version. I saw a version using
> modrm.r/m for operand1 and modrm.reg for operand2, where
> current SDE is also using this version.

In fact doc version 1.15 section 9.4.2 is inconsistent in itself here:

Form
xmm1/m64, xmm2
xmm1/m128, ymm2
ymm1/m256, zmm2

vs

Instruction	Operand 1	Operand 2	Operand 3
VCVTBF82BF4S
VCVTHF82BF4S	MODRM.REG(w)	MODRM.R/M(r)	N/A

Jan
lmpx.com only provides a reader for public news (NNTP) servers. It is not affiliated with the servers or forums shown here and is not responsible for the content of articles, which is written by their respective authors.