Re: [PATCH 09/10] Support Intel AMX-FP16

[Jan Beulich <jbeulich@suse.com>]

On 19.10.2022 16:41, Cui, Lili wrote:
>>>>> --- a/opcodes/i386-gen.c
>>>>> +++ b/opcodes/i386-gen.c
>>>>> @@ -319,6 +319,8 @@ static initializer cpu_flag_init[] =
>>>>>      "CPU_AMX_TILE_FLAGS|CpuAMX_INT8" },
>>>>>    { "CPU_AMX_BF16_FLAGS",
>>>>>      "CPU_AMX_TILE_FLAGS|CpuAMX_BF16" },
>>>>> +  { "CPU_AMX_FP16_FLAGS",
>>>>> +    "CPU_AMX_TILE_FLAGS|CpuAMX_FP16" },
>>>>>    { "CPU_AMX_TILE_FLAGS",
>>>>>      "CpuAMX_TILE" },
>>>>>    { "CPU_MOVDIRI_FLAGS",
>>>>> @@ -425,8 +427,10 @@ static initializer cpu_flag_init[] =
>>>>>      "CpuAMX_INT8" },
>>>>>    { "CPU_ANY_AMX_BF16_FLAGS",
>>>>>      "CpuAMX_BF16" },
>>>>> +  { "CPU_ANY_AMX_FP16_FLAGS",
>>>>> +    "CpuAMX_FP16" },
>>>>
>>>> ... you request it to be constructed here.
>>>
>>> HAHA, you mean if there is no instruction that depends on CpuAMX_FP16,
>> we don’t need to add *ANY* for it in arch_entry cpu_arch[]?  Maybe when
>> the dependency happens, we need to change it back.
>>
>> Right, that's why I did ask whether there are dependent features known to
>> appear. If you go look you'll find that we don't have ANY_* for a fair part (if
>> not a majority) of the features).
> 
> I removed ANY_* for AMX_FP16,

You did remove its use, but not ...

> --- a/opcodes/i386-gen.c
> +++ b/opcodes/i386-gen.c
> @@ -319,6 +319,8 @@ static initializer cpu_flag_init[] =
>      "CPU_AMX_TILE_FLAGS|CpuAMX_INT8" },
>    { "CPU_AMX_BF16_FLAGS",
>      "CPU_AMX_TILE_FLAGS|CpuAMX_BF16" },
> +  { "CPU_AMX_FP16_FLAGS",
> +    "CPU_AMX_TILE_FLAGS|CpuAMX_FP16" },
>    { "CPU_AMX_TILE_FLAGS",
>      "CpuAMX_TILE" },
>    { "CPU_MOVDIRI_FLAGS",
> @@ -425,8 +427,10 @@ static initializer cpu_flag_init[] =
>      "CpuAMX_INT8" },
>    { "CPU_ANY_AMX_BF16_FLAGS",
>      "CpuAMX_BF16" },
> +  { "CPU_ANY_AMX_FP16_FLAGS",
> +    "CpuAMX_FP16" },

... it being created. This is the only remaining issue I see.

Jan