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

[Ser, Simon]

On Fri, 2019-05-24 at 09:33 +0100, 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.
> 
> > > +     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()?
> 
> > 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.

I don't understand the motivation here. Can you elaborate?