[Mesa-dev] [PATCH 14/16] anv: Advertise both 32-bit and 48-bit heaps when we have enough memory

[Nanley Chery]

On Fri, May 19, 2017 at 10:53:12AM -0700, Jason Ekstrand wrote:
> On Thu, May 18, 2017 at 2:01 PM, Jason Ekstrand <jason at jlekstrand.net>
> wrote:
> 
> > ---
> >  src/intel/vulkan/anv_device.c | 42 ++++++++++++++++++++++++++++++
> > ++++++------
> >  1 file changed, 36 insertions(+), 6 deletions(-)
> >
> > diff --git a/src/intel/vulkan/anv_device.c b/src/intel/vulkan/anv_device.c
> > index 6ea8dfe..8eed7f3 100644
> > --- a/src/intel/vulkan/anv_device.c
> > +++ b/src/intel/vulkan/anv_device.c
> > @@ -112,12 +112,42 @@ anv_physical_device_init_heaps(struct
> > anv_physical_device *device, int fd)
> >     if (result != VK_SUCCESS)
> >        return result;
> >
> > -   device->memory.heap_count = 1;
> > -   device->memory.heaps[0] = (struct anv_memory_heap) {
> > -      .size = heap_size,
> > -      .flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
> > -      .supports_48bit_addresses = device->supports_48bit_addresses,
> > -   };
> > +   if (heap_size <= 3ull * (1ull << 30)) {
> > +      /* In this case, everything fits nicely into the 32-bit address
> > space,
> > +       * so there's no need for supporting 48bit addresses on
> > clinet-allocated
> > +       * memory objects.
> > +       */
> > +      device->memory.heap_count = 1;
> > +      device->memory.heaps[0] = (struct anv_memory_heap) {
> > +         .size = heap_size,
> > +         .flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
> > +         .supports_48bit_addresses = false,
> > +      };
> > +   } else {
> > +      /* Not everything will fit nicely into a 32-bit address space.  In
> > this
> > +       * case we need a 64-bit heap.  Advertise a small 32-bit heap and a
> > +       * larger 48-bit heap.  If we're in this case, then we have a total
> > heap
> > +       * size larger than 3GiB which most likely means they have 8 GiB of
> > +       * video memory and so carving off 2 GiB for the 32-bit heap should
> > be
> > +       * reasonable.
> > +       */
> > +      const uint32_t heap_size_32bit = 2ull * (1ull << 30);
> >
> 
> FYI: I keep waflling on whether the 32-bit carve-out should be 2 GiB or 1
> GiB.  No one should need more than 1 GiB for vertex buffers and it means
> that, on my laptop, the space is split roughly in half (2 GiB and ~3 GiB)
> which is a bit awkward.  I'm kind-of inclined to decrease it.  Opinions?
> 
> 

I'm not sure what a good size is either. It may be helpful if we ran
some games to see how they behaved with different values.

> > +      const uint32_t heap_size_48bit = heap_size - heap_size_32bit;
> > +
> > +      assert(device->supports_48bit_addresses);
> > +
> > +      device->memory.heap_count = 2;
> > +      device->memory.heaps[0] = (struct anv_memory_heap) {
> > +         .size = heap_size_48bit,
> > +         .flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
> > +         .supports_48bit_addresses = true,
> > +      };
> > +      device->memory.heaps[1] = (struct anv_memory_heap) {
> > +         .size = heap_size_32bit,
> > +         .flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
> > +         .supports_48bit_addresses = false,
> > +      };
> > +   }
> >
> >     uint32_t type_count = 0;
> >     for (uint32_t heap = 0; heap < device->memory.heap_count; heap++) {
> > --
> > 2.5.0.400.gff86faf
> >
> >

> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev