We have an issue with occlusion queries (PIPE_CONTROL depth writes)
after using the pipeline with the VS disabled. We work around it by
using a depth cache flush PIPE_CONTROL before doing a depth write.
Fixes dEQP-VK.query_pool.*
This exposes a case where we want to anv_CmdCopyBufferToImage() on an
image that wasn't created with VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT and
end up using uninitialized color_rt_surface_state from the meta image
view.
If VK_COMMAND_BUFFER_USAGE_RENDER_PASS_CONTINUE_BIT is not set in
pBeginInfo->flags, we don't have a render pass or framebuffer. Change
the condition that guard looking up render pass and framebuffer to test
for VK_COMMAND_BUFFER_USAGE_RENDER_PASS_CONTINUE_BIT instead of
VK_COMMAND_BUFFER_LEVEL_SECONDARY.
Fixes all remaining crashes in dEQP-VK.api.command_buffers.*.
SKL needs this to make sure we flush the push constants. It gets a
little tricky, since we also need to emit binding tables before push
constants, since that may affect the push constants (dynamic buffer
offsets and storage image parameters). This patch splits emitting
binding tables from emitting the pointers so that we can emit push
constants after binding tables but before emitting binding table
pointers.
Previously, anv_image_view had a anv_format pointer that we used for
everything. This commit replaces that pointer with a VkFormat enum copied
from the API and an isl_format. In order to implement RGB formats, we have
to use a different isl_format for the actual surface state than the obvious
one from the VkFormat. Separating the two helps us keep things streight.
This is what image_view does. Also, we really need to do this so that we
can properly handle the combined offsets from the buffer and from
pCreateInfo.
This fixes some of the nonzero offset buffer view CTS tests.
Similar to anv_cmd_buffer_push_constants, but handles the compute
pipeline, which requires different setup from the other stages.
This also handles initializing the compute shader local IDs.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
This made for an unfortunately large amount of work since we were using it
fairly heavily internally. However, gl_shader_stage does basically the
same things, so it's not too bad.
This file contains code that can be shared across gens modulo recompiling.
In particular, we can share STATE_BASE_ADDRESS setup and handling of the
vkPipelineBarrier call. Not sharing STATE_BASE_ADDRESS setup has already
been a source of bugs and the gen7 and gen8 implementations of
PipelineBarrier were line-for-line identical.
Incidentally, this should fix MOCS settings for dynamic and surface state
on Haswell.
The new mechanism should be able to handle SSBOs as well as properly handle
emitting surface state on gen7 where we need different strides depending on
shader stage.
We never *actually* supported them, we just used them for binding UBOs.
Now that we have BufferInfo and we aren't supporting texture buffers yet,
we should get rid of them until we can do them properly.
The anv_state is supposed to be a flyweight so we're not really saving
anything by using a pointer. Also, we were creating one, setting a pointer
to it, and then having it go out-of-scope which is bad.
When emitting the binding table for the fragment shader stage, we no
longer "walk all of the attachments, [inserting only] the color
attachments into the binding table". Instead, we iterate only over the
subpass's color attachments, which is the minimal possible iteration.
While killing the comment, also rename the variable 'attachments' to
'color_count', as it's no longer a count of all framebuffer attachments
but only the subpass's color attachment count.
Prefix all anv_cmd_state dirty bit tokens with ANV_CMD_DIRTY. For
example:
old -> new
ANV_DYNAMIC_VIEWPORT_DIRTY -> ANV_CMD_DIRTY_DYNAMIC_VIEWPORT
ANV_CMD_BUFFER_PIPELINE_DIRTY -> ANV_CMD_DIRTY_PIPELINE
Change type of anv_cmd_state::dirty and ::compute_dirty from uint32_t to
the self-documenting type anv_cmd_dirty_mask_t.
This patch reworks a bunch of stuff in the way we do descriptor set
layouts. Our previous approach had a couple of problems. First, it was
based on a misunderstanding of arrays in descriptor sets. Second, it
didn't properly handle descriptor sets where some bindings were missing
stages. The new apporach should be correct and also makes some operations,
particularly those on the hot-path, a bit easier.
We use the descriptor set layout for four things:
1) To determine the map from bindings to the actual flattened descriptor
set in vkUpdateDescriptorSets().
2) To determine the descriptor <-> binding table entry mapping to use in
anv_cmd_buffer_flush_descriptor_sets().
3) To determine the mappings of dynamic indices.
4) To determine the (set, binding, array index) -> binding table entry
mapping inside of shaders.
The new approach is directly taylored towards these operations.
