[Intel-gfx] [PATCH i-g-t] benchmarks/gem_wsim: Heap allocate VLA structs

[Tvrtko Ursulin]

On 24/05/2019 09:33, Chris Wilson wrote:
> Quoting Tvrtko Ursulin (2019-05-24 09:20:47)
>>
>> On 24/05/2019 08:25, Chris Wilson wrote:
>>> Apparently VLA structs (e.g. struct { int array[count] }) is a gcc
>>> extension that clang refuses to support as handling memory layout is too
>>> difficult for it.
>>>
>>> Move the on-stack VLA to the heap.
>>>
>>> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
>>> Cc: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
>>> ---
>>>    benchmarks/gem_wsim.c | 146 +++++++++++++++++++++++++++---------------
>>>    1 file changed, 95 insertions(+), 51 deletions(-)
>>>
>>> diff --git a/benchmarks/gem_wsim.c b/benchmarks/gem_wsim.c
>>> index e2ffb93a9..0a0032bff 100644
>>> --- a/benchmarks/gem_wsim.c
>>> +++ b/benchmarks/gem_wsim.c
>>> @@ -1441,6 +1441,48 @@ set_ctx_sseu(struct ctx *ctx, uint64_t slice_mask)
>>>        return slice_mask;
>>>    }
>>>    
>>> +static size_t sizeof_load_balance(int count)
>>> +{
>>> +     struct i915_context_engines_load_balance *ptr;
>>> +
>>> +     assert(sizeof(ptr->engines[count]) == count * sizeof(ptr->engines[0]));
>>
>> This seems wrong - is bound to trigger.
> 
> Why does it seem wrong? That's the calculation used previously, and the
> ptr->engines[] was meant to be packed in order for
> sizeof(ptr->engines[count]) == count * sizeof(ptr->engines[0]). Anyway,
> I threw it in there to check if the calculation was sane.

Because sizeof(ptr->engines[0]) == sizeof(ptr->engines[N]), since the 
code is not declaring N big array, just referencing the element N. So 
for more than one engine I expect it explodes. Unless I am way wrong.. I 
guess someone needs to run it.. :)

>>> +     return sizeof(*ptr) + sizeof(ptr->engines[count]);
>>
>> So size of of engine needs to be multiplied by count.
> 
> (Just note this is the what the current VLA evaluates to :)
> 
>>> +}
>>> +
>>> +static struct i915_context_engines_load_balance *
>>> +alloc_load_balance(int count)
>>> +{
>>> +     return calloc(1, sizeof_load_balance(count));
>>
>> How about alloca so cleanup is simpler? Or is alloca also on the
>> unpopular list?
> 
> I don't mind. Would shave a few lines indeed, but we need the memsets
> back. #define alloca0()?

And a helper macro to generically deal with struct header + engines 
array so it doesn't need to be repeated three times. Yadayada too much 
work.. :) ...

>> Or possibly what Simon suggested, just a large temporary stack arrays
>> would be enough and easiest diff. Just with an assert that it fits.
> 
> I don't think that is as clean for the long term.

... this should be just fine for now so I'd vote for it.

Regards,

Tvrtko