Hi Kwok,

On 30/11/2022 19:50, Kwok Cheung Yeung wrote:
> Hello PA,
> 
>> --- libgomp/config/gcn/selector.c
>> +++ libgomp/config/gcn/selector.c
>> @@ -36,7 +36,7 @@ GOMP_evaluate_current_device (const char *kind, 
>> const char *arch,
>>    if (kind && strcmp (kind, "gpu") != 0)
>>      return false;
>>
>> -  if (arch && strcmp (arch, "gcn") != 0)
>> +  if (arch && (strcmp (arch, "gcn") != 0 || strcmp (arch, "amdgcn") 
>> != 0))
>>      return false;
> 
> The logic here looks wrong to me - surely it should return false if arch 
> is not 'gcn' AND it is not 'amdgcn'?

Sure. Fixed in revised patch.

>> @@ -48,8 +48,17 @@ GOMP_evaluate_current_device (const char *kind, 
>> const char *arch,
>>  #endif
>>
>>  #ifdef __GCN5__
>> -  if (strcmp (isa, "gfx900") == 0 || strcmp (isa, "gfx906") != 0
>> -      || strcmp (isa, "gfx908") == 0)
>> +  if (strcmp (isa, "gfx900") == 0 || strcmp (isa, "gfx906") != 0)
>> +    return true;
>> +#endif
>> +
>> +#ifdef __CDNA1__
>> +  if (strcmp (isa, "gfx908") == 0)
>> +    return true;
>> +#endif
>> +
>> +#ifdef __CDNA2__
>> +  if (strcmp (isa, "gfx90a") == 0)
>>      return true;
>>  #endif
> 
> Okay for gfx908 and gfx90a, but is there any way of distinguishing 
> between 'gfx900' and 'gfx906' ISAs? I don't think these are mutually 
> compatible.
>

Since I did not find any existing builtin to check the exact ISA, I 
added all of them for consistency. Let me know if that looks good to you.

Thanks,
-- 
PA