* - Add trace category capability and duration

* - Respect duration value if set.

* Fix redundant wait

* Parse rcs_state to determine if device is viable.

* Review pass.

And make it pass again.

Allows users to select what categories to trace for Fuchsia traces.

Implement VK_EXT_blend_operation_advanced

Implement VK_EXT_index_type_uint8

Show two big icons for the different trace modes and add two new
capture trace buttons, one for each.

fix access flags

* Modify protected feature and queue queries

Protected memory is not currently supported by capture/replay and it is
currently unknown if it can be supported. In the meantime, use the spy
layer to suppress support for protected memory advertised by the
underlying driver. This has to be done by overriding
vkGetPhysicalDeviceFeatures2 and
vkGetPhysicalDeviceQueueFamilyProperties since protected memory is a
core Vulkan 1.1 feature and not an extension.

* Swapchain layer surface capabilities 2

Add missing entry point vkGetPhysicalDeviceSurfaceCapabilities2KHR to
the virtual swapchain layer intercepts

Calculate size of texture to display in texture list
- Added size to the texture proto, calculated based on format
- Displays the total size from each depth/layer

When listing the commands for the command tree, support filtering by:

Host Commands
Begin/End Marker Commands
Non Renderpass Commands
Sync Commands

Implement Ext custom border color

Until the user presses the "Get Started" button in the first run dialog,
automatic update checks should be off.

If the user enables automatic update checks, don't wait for application
restart to do a check, but do so immediately.

Allow passing nil for the list of commands to process to
MutateWithSubcommands and use the capture's commands as a default if nil
is passed. This is the most common use case for this function and avoids
having to resolve the capture just to resolve it again.

Re-order the code a bit in GpuProfile and use better error constructor
functions when composing errors with a cause.

Maps are passed by reference by default.

- Fix VertexInputData and VertexAttributeDivisor
- Make VertexInputData optional(as spec states)
- Add VertexAttributeDivisor to stateRebuilder
- Fix the code around these two changes
- update the sample trace

Implement VK_KHR_Multiview

Partial implementation of KHR_buffer_device_address

* Add content view for replayer activity

* Fixed copyright msg and cleaned up dup files

* Fix copyright year

…e mechanisms (#971)

Added value representation to memory box to display char * in a more readable format

Sorry! Forgot to "git add"

* Implement VK_EXT_sampler_filter_minmax

* Fix format

Convert uint8 values to char for strings in memory view

If the framebuffer attachment drop down is closed before the thumbnail
finished loading, we end up calling an update function on a disposed
widget at 60Hz.

The function, as documented, is only intended to unpack dense maps. The
original naming is a bit misleading and the dense requirement may be
missed without careful reading of the doc comment.

The map of divisors is not a dense map. The existing map unpacking
functions require dense maps as documented. This change introduces a new
function that will properly unpack a non-dense map, ignoring the keys.

The lookup made assumptions about the tree layout that are no longer
correct when the tree is filtered.

The drop down is now filtered and its indices no longer correspond to the
list of devices.

If we're filtering the host commands, don't add the host coordination
groups.

Most traces are just one complete frame and so the group would just
group the entire trace, which is not helpful.

Also cleans up the grouping by frame so it doesn't unnecessarily loop
over the commands twice.

This is the server side implementation of a new flag for the command tree,
which allows users to filter the submit info nodes from the tree.

- Use a popup/dropdown to show the filtering options with a filter button
  that will show the menu.
- Save the selected filter items in the settings so they will be reloaded.
- Simplified the event handling code.
- Show the loading indicator if it takes time to load the new tree.
- Added new submit node filter option to the UI.

We only support one API, so grouping by API doesn't make sense, and there
is little benefit to group by thread as it's only host side.

We have no commands annotated with @draw_call or @clear, which means that
all code related to these annotations has been dead. This is all code that
was GLES releted.

Slices are passed by reference by default.

Has a side-effect that we no longer compute parent groups, but that is OK
as we will no longer need them once we combine the command tree with
PerfTab. For now, this flattens the tree in PerfTab, but it wasn't really
useful there anyways.

Attaches the profiling data to command ranges rather than individual
commands.

Changes some groups messages to id, that is references, so that we no
longer send the groups multiple times.

This time mapping more closely to the command tree.

This makes some assumptions about command/sub command IDs:
1. we only get groups for command buffers, renderpasses and draw calls.
2. no overlaps.
3. the sub command ids are [cmdId, submission, cmdbuff, cmd].

All these assumptions currently hold and are also made in other parts of
the code in some way. The assumptions will need to be codified as part of
the command/sub-command refactor that is already planned.

* Add new types added with VK_KHR_create_renderpass2

* Implement VkCreateRenderpass2

This change also includes:
     - Refactoring RenderPassObject and VkCreateRenderPass in terms of:
       - API files
       - State Reconstruction
       - And usage of it throughout the codebase

* Implement VkBeginRenderPass2(KHR)

This includes:
     - Command Recording
     - Command Implementation
     - Command Rebuilding
     - Refactoring VkBeginRenderPass and it's usage

* Implement VkEndRenderPass2(KHR)

This includes:
     - Command Recording
     - Command Implementation
     - Command Rebuilding
     - Refactoring VkEndRenderPass and it's usage

* Implement VkCmdNextSubpass2(KHR)

    This includes:
         - Command Recording
         - Command Implementation
         - Command Rebuilding
         - Refactoring VkCmdNextSubpass and it's usage

Leaving it selected after destruction was causing a UI exception on
macOS.

Bug: b/191637172
Test: Manual

When showing the system profiler view in the frame profiler, hide some of
the UI elements that don't make sense in this view.

This reverts commit ae5da6c.

This, somehow broke replay profiling.

Building with the MacOSX12.1 SDK will pull in TargetConditionals.h which
unconditionally defines conflicting macros `TARGET_OS_LINUX` and
`TARGET_OS_WINDOWS`. This change fixes the name conflict by prefixing
the gapid macros with `GAPID_`.

* Change the location of the dmgbuild to update the build machine to Big-sur
and use python3 explicitly for pip

* Set Java version to prevent ANDROID_SDK crash

* update XCode version

Implements the VK_KHR_create_renderpass2

Adds the following counters:
  - Vertex Count
  - Primitive Count
  - Byte Count
  - Vertex Size
  - Primitive Size

For the following draw calls:
  - vkCmdDraw
  - vkCmdDrawIndexed
  - vkCmdDrawIndirect
  - vkCmdDrawIndexedIndirect
  - vkCmdDrawIndirectCountKHR/AMD
  - vkCmdDrawIndexedIndirectCountKHR/AMD

Tested locally with modified cube sample that exercised all of the
implemented draw calls.

They are already referenced by the counters and we will need the tree to
merge in the draw call stats.

can use them outside of the vulkan package.

combining with the Perfetto based profiling data.

This is currently done in a hacky way via a channel.

The slices->counters code is written on top of slices structs from the
proto and will be replaced to use this slice data, so this makes this
transition easier.

This allows sharing some more code between GPU specific implementations
and prepares for merging of the GPU and static analysis data.

- Only try to expand/collapse items if they are not currently in the
  desired state, when handling the custom key/mouse events.
- Fire SWT.Expand/SWT.Collapse events

Overrides users of old AGI's default stored layout with a new default that
is more suitable for the new command tab.

 * Add driver version to GPU protobuf
 * Introduce a hardware-agnostic mapping between `id` and Mali counter
 * Use the new driverVersion field in the GPU protobuf to determine
   which set of counters to use for validation
 * Note: this is a breaking change for users who have a text-based
   perfetto config, since counter IDs have changed

This is a work-around for b/196832502.

- RG formats will be rendered into RGBA
- R formats will be rendered into luminance

For this purpose, this PR adds:

- R16 to Luminance Float conversion support
- Some fixing with the format naming to prevent empty format names in logs

Fix for: b/216478949

GAPIR does not do profile replays correctly when in the background. Kill it on pause.
The next replay after a kill on pause will spawn a new GAPIR.

* Add validation info to gapis
- Always download perfetto trace from validation
- Include path to trace in result proto

Pre-computing shader static analysis for every shader as part of
ResourceData was taking too long. This change splits static analysis
data into a new ResourceExtras request that is made as needed for each
individual shader.

* Add missing defines for VK_EXT_sampler_filter_minmax
* Fixed missing newline