[Mesa-dev] [PATCH 02/18] anv/image: Add anv_layout_to_aux_usage()
Jason Ekstrand
jason at jlekstrand.net
Tue Feb 28 19:00:14 UTC 2017
On Tue, Feb 28, 2017 at 10:53 AM, Nanley Chery <nanleychery at gmail.com>
wrote:
> On Tue, Feb 28, 2017 at 10:38:12AM -0800, Jason Ekstrand wrote:
> > On Tue, Feb 28, 2017 at 10:32 AM, Nanley Chery <nanleychery at gmail.com>
> > wrote:
> >
> > > On Tue, Feb 28, 2017 at 08:26:56AM -0800, Jason Ekstrand wrote:
> > > > On Mon, Feb 27, 2017 at 5:20 PM, Nanley Chery <nanleychery at gmail.com
> >
> > > wrote:
> > > >
> > > > > This function supersedes layout_to_hiz_usage().
> > > > >
> > > > > Signed-off-by: Nanley Chery <nanley.g.chery at intel.com>
> > > > > ---
> > > > > src/intel/vulkan/anv_image.c | 149
> ++++++++++++++++++++++++++++++
> > > > > +++++++++++
> > > > > src/intel/vulkan/anv_private.h | 4 ++
> > > > > 2 files changed, 153 insertions(+)
> > > > >
> > > > > diff --git a/src/intel/vulkan/anv_image.c
> > > b/src/intel/vulkan/anv_image.c
> > > > > index cd142938e7..716cdf3a38 100644
> > > > > --- a/src/intel/vulkan/anv_image.c
> > > > > +++ b/src/intel/vulkan/anv_image.c
> > > > > @@ -432,6 +432,155 @@ void anv_GetImageSubresourceLayout(
> > > > > }
> > > > > }
> > > > >
> > > > > +/**
> > > > > + * @brief This function determines the optimal buffer to use for
> > > device
> > > > > + * accesses given a VkImageLayout and other pieces of information
> > > needed
> > > > > to
> > > > > + * make that determination. Device accesses may include normal
> > > sampling
> > > > > and
> > > > > + * rendering operations or resolves.
> > > > > + *
> > > > > + * @param gen The generation of the Intel GPU.
> > > > > + * @param image The image that may contain a collection of
> buffers.
> > > > > + * @param aspects The aspect(s) of the image to be accessed.
> > > > > + * @param layout The current layout of the image aspect(s).
> > > > > + * @param future_layout The next layout of the image aspect(s), if
> > > known.
> > > > > + * Otherwise this should be equal to the
> current
> > > > > layout.
> > > > > + *
> > > > > + * @return The primary buffer that should be used for the given
> > > layout.
> > > > > + */
> > > > > +enum isl_aux_usage
> > > > > +anv_layout_to_aux_usage(const uint8_t gen, const struct anv_image
> *
> > > const
> > > > > image,
> > > > >
> > > >
> > > > Might be better to take a gen_device_info rather than just the
> integer.
> > > > Since we're doing run-time checks anyway, the cost should be small.
> > > >
> > >
> > > What does this buy us?
> > >
> >
> > Nothing at the moment. However, it is the more standard thing to pass in
> > and, if we ever need more information in the future, device_info is what
> > we'll want. Or you could just pass in the anv_device to guarantee that
> we
> > have everything we would ever need.
> >
>
> I'm in favor of adding it when we need it, but I can make it take the
> device_info if you insist. My rationale is that if the function takes a
> device_info or anv_device, the function prototype would be unnecessarily
> ambiguous with respect to what device-related information is necessary
> to perform the mapping.
>
Certainly anv_device is probably a bit much since it doesn't (yet) depend
on any run-time information. I could see that changing in the future
(environment variables, per-app parameters, etc.) but we can cross that
bridge when we come to it. I don't really see an "information" difference
between an integer "gen" and a device info. The one completely describes
the hardware and the other sort-of describes the hardware. Also, I
wouldn't be terribly surprised if, in the future, whether or not we can
sample from HiZ starts to depend on half-gen or whether or not it's an atom
part.
> -Nanley
>
> >
> > > -Nanley
> > >
> > > >
> > > > > + const VkImageAspectFlags aspects,
> > > VkImageLayout
> > > > > layout,
> > > > > + const VkImageLayout future_layout)
> > > > > +{
> > > > > + /* Validate the inputs. */
> > > > > +
> > > > > + /* Intel GPUs prior to Gen7 are not supported in anv. */
> > > > > + assert(gen >= 7);
> > > > > +
> > > > > + /* The layout of a NULL image is not properly defined. */
> > > > > + assert(image != NULL);
> > > > > +
> > > > > + /* The aspects must be a subset of the image aspects. */
> > > > > + assert(aspects & image->aspects && aspects <= image->aspects);
> > > > > +
> > > > > + /* According to the Vulkan Spec, the following layouts are
> valid
> > > only
> > > > > as
> > > > > + * initial layouts in a layout transition and don't support
> device
> > > > > access.
> > > > > + * Therefore, the caller should not be setting the future
> layout to
> > > > > either.
> > > > > + */
> > > > > + assert(future_layout != VK_IMAGE_LAYOUT_UNDEFINED &&
> > > > > + future_layout != VK_IMAGE_LAYOUT_PREINITIALIZED);
> > > > > +
> > > > > + /* Determine the optimal buffer. */
> > > > > +
> > > > > + /* If there is no auxiliary surface allocated, we must use the
> one
> > > and
> > > > > only
> > > > > + * main buffer.
> > > > > + */
> > > > > + if (image->aux_surface.isl.size == 0)
> > > > > + return ISL_AUX_USAGE_NONE;
> > > > > +
> > > > > + /* All images that use an auxiliary surface are required to be
> > > tiled.
> > > > > */
> > > > > + assert(image->tiling == VK_IMAGE_TILING_OPTIMAL);
> > > > > +
> > > > > + /* On BDW+, when clearing the stencil aspect of a depth stencil
> > > image,
> > > > > + * the HiZ buffer allows us to record the clear with a
> relatively
> > > small
> > > > > + * number of packets. Prior to BDW, the HiZ buffer provides no
> > > known
> > > > > benefit
> > > > > + * to the stencil aspect.
> > > > > + */
> > > > > + if (gen < 8 && aspects == VK_IMAGE_ASPECT_STENCIL_BIT)
> > > > > + return ISL_AUX_USAGE_NONE;
> > > > > +
> > > > > + /* The undefined layout indicates that the user doesn't care
> about
> > > the
> > > > > data
> > > > > + * that's currently in the buffer. Therefore, the optimal
> buffer to
> > > > > use is
> > > > > + * the same buffer that would be used in the next layout. This
> > > avoids
> > > > > the
> > > > > + * possibility of having to resolve in order to maintain
> coherency.
> > > > > + *
> > > > > + * The pre-initialized layout is undefined for optimally-tiled
> > > images.
> > > > > As
> > > > > + * guaranteed by the assertion above, all images that have
> reached
> > > this
> > > > > + * point are tiled.
> > > > > + */
> > > > > + if (layout == VK_IMAGE_LAYOUT_UNDEFINED ||
> > > > > + layout == VK_IMAGE_LAYOUT_PREINITIALIZED)
> > > > > + layout = future_layout;
> > > > > +
> > > > > + const bool has_depth = aspects & VK_IMAGE_ASPECT_DEPTH_BIT;
> > > > > + const bool color_aspect = aspects == VK_IMAGE_ASPECT_COLOR_BIT;
> > > > > +
> > > > > + /* The following switch currently only handles depth stencil
> > > aspects.
> > > > > + * TODO: Handle the color aspect.
> > > > > + */
> > > > > + if (color_aspect)
> > > > > + return image->aux_usage;
> > > > > +
> > > > > + switch (layout) {
> > > > > +
> > > > > + /* Invalid Layouts */
> > > > > + case VK_IMAGE_LAYOUT_UNDEFINED:
> > > > > + case VK_IMAGE_LAYOUT_PREINITIALIZED:
> > > > > + case VK_IMAGE_LAYOUT_RANGE_SIZE:
> > > > > + case VK_IMAGE_LAYOUT_MAX_ENUM:
> > > > > + unreachable("Invalid image layout.");
> > > > > +
> > > > > +
> > > > > + /* Transfer Layouts
> > > > > + *
> > > > > + * This buffer could be a depth buffer used in a transfer
> > > operation.
> > > > > BLORP
> > > > > + * currently doesn't use HiZ for transfer operations so we
> must use
> > > > > the main
> > > > > + * buffer for this layout. TODO: Enable HiZ in BLORP.
> > > > > + */
> > > > > + case VK_IMAGE_LAYOUT_GENERAL:
> > > > > + case VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL:
> > > > > + case VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL:
> > > > > + return ISL_AUX_USAGE_NONE;
> > > > > +
> > > > > +
> > > > > + /* Sampling Layouts */
> > > > > + case VK_IMAGE_LAYOUT_DEPTH_STENCIL_READ_ONLY_OPTIMAL:
> > > > > + assert(!color_aspect);
> > > > > + /* Fall-through */
> > > > > + case VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL:
> > > > > + if (has_depth && anv_can_sample_with_hiz(gen,
> image->samples))
> > > > > + return ISL_AUX_USAGE_HIZ;
> > > > > + else
> > > > > + return ISL_AUX_USAGE_NONE;
> > > > > +
> > > > > + case VK_IMAGE_LAYOUT_PRESENT_SRC_KHR:
> > > > > + assert(color_aspect);
> > > > > +
> > > > > + /* On SKL+, the render buffer can be decompressed by the
> > > > > presentation
> > > > > + * engine. Support for this feature has not yet landed in
> the
> > > wider
> > > > > + * ecosystem. TODO: Update this code when support lands.
> > > > > + *
> > > > > + * From the BDW PRM, Vol 7, Render Target Resolve:
> > > > > + *
> > > > > + * If the MCS is enabled on a non-multisampled render
> > > target, the
> > > > > + * render target must be resolved before being used for
> other
> > > > > + * purposes (display, texture, CPU lock) The clear value
> from
> > > > > + * SURFACE_STATE is written into pixels in the render
> target
> > > > > + * indicated as clear in the MCS.
> > > > > + *
> > > > > + * Pre-SKL, the render buffer must be resolved before being
> > > used for
> > > > > + * presentation. We can infer that the auxiliary buffer is
> not
> > > used.
> > > > > + */
> > > > > + return ISL_AUX_USAGE_NONE;
> > > > > +
> > > > > +
> > > > > + /* Rendering Layouts */
> > > > > + case VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL:
> > > > > + assert(color_aspect);
> > > > > + unreachable("Color images are not yet supported.");
> > > > > +
> > > > > + case VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL:
> > > > > + assert(!color_aspect);
> > > > > + return ISL_AUX_USAGE_HIZ;
> > > > > + }
> > > > > +
> > > > > + /* If the layout isn't recognized in the exhaustive switch
> above,
> > > the
> > > > > + * VkImageLayout value is not defined in vulkan.h.
> > > > > + */
> > > > > + unreachable("layout is not a VkImageLayout enumeration
> member.");
> > > > > +}
> > > > > +
> > > > > +
> > > > > static struct anv_state
> > > > > alloc_surface_state(struct anv_device *device)
> > > > > {
> > > > > diff --git a/src/intel/vulkan/anv_private.h
> b/src/intel/vulkan/anv_
> > > > > private.h
> > > > > index 2527c2cc5a..5ce9027f88 100644
> > > > > --- a/src/intel/vulkan/anv_private.h
> > > > > +++ b/src/intel/vulkan/anv_private.h
> > > > > @@ -1671,6 +1671,10 @@ anv_gen8_hiz_op_resolve(struct
> anv_cmd_buffer
> > > > > *cmd_buffer,
> > > > > const struct anv_image *image,
> > > > > enum blorp_hiz_op op);
> > > > >
> > > > > +enum isl_aux_usage
> > > > > +anv_layout_to_aux_usage(const uint8_t gen, const struct anv_image
> > > *image,
> > > > > + const VkImageAspectFlags aspects,
> > > VkImageLayout
> > > > > layout,
> > > > > + const VkImageLayout future_layout);
> > > > > static inline uint32_t
> > > > > anv_get_layerCount(const struct anv_image *image,
> > > > > const VkImageSubresourceRange *range)
> > > > > --
> > > > > 2.11.1
> > > > >
> > > > > _______________________________________________
> > > > > mesa-dev mailing list
> > > > > mesa-dev at lists.freedesktop.org
> > > > > https://lists.freedesktop.org/mailman/listinfo/mesa-dev
> > > > >
> > >
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20170228/958af1a3/attachment-0001.html>
More information about the mesa-dev
mailing list