| Age | Commit message (Collapse) | Author |
|
|
|
|
|
Finish moving the rest of layer generation to using generate_intercept.
Make APIDump variants for File and NoAddr subclasses of APIDump and APIDumpCpp.
|
|
Move layer_name from parameters to class attribute.
Move no_addr setting from parameters to class attribute.
Access generate_intercept method through class instead of parameter.
|
|
Change ObjectTracker generation to use _generate_dispatch_entrypoints_with_func
|
|
Changed _generate_dispatch_entrypoints in xgl-layer-generate.py
to remove specific layer handling.
Instead pass it a "generate_intercept" function to decide which
API calls to intercept and output layer specific intercept code.
Output layer_intercept_proc() function per layer instead of using
xgl_generic_intercept_proc_helper.h so intercept lists can vary
for different generated layers.
|
|
Make the code to generate DbgRegisterMsgCallback and
DbgUnregisterMsgCallback in _generate_dispatch_entrypoints
more uniform across layers.
This also fixes the ObjectTracker layer to pass on these calls
to the next layer or icd.
|
|
ParamChecker is no longer generated by xgl-layer-generate.py
|
|
|
|
|
|
|
|
|
|
|
|
A recent "vulkan.h" header change means we should use the
VK_DEFINE_DISP_SUBCLASS_HANDLE() macor instead of the VK_DEFINE_HANDLE() macro.
Also, we needed to extend the VkObjectType enum/typedef.
|
|
|
|
v2: enable drawState too (olv)
|
|
|
|
|
|
|
|
This update makes it possible to bind multiple descriptor sets to the command
buffer. We introduced intel_cmd_dset_data to hold the descriptor set offsets
in the global descriptor region as well as their dynamic offsets. We also
changed the descriptor reading routines from
intel_desc_set_read_surface()
intel_desc_set_read_sampler()
to
intel_desc_region_read_surface()
intel_desc_region_read_sampler()
v2: fix a potential crash when allocation of dynamic_desc_indices fails
|
|
v2: fix a typo in the type of shared_immutable_sampler (olv)
|
|
Replace a chain of XGL_DESCRIPTOR_SET_LAYOUT by
XGL_DESCRIPTOR_SET_LAYOUT_CHAIN. The change also drops
stageFlags/pSetBindPoints, and affects xglUpdateDescriptors().
xglCmdBindDescriptorSet() is not updated yet by this commit, which makes it
impossible to bind multiple sets.
The change also encourages the use of arrays of resources, which we do not
support.
v2: fix a typo in intel_desc_set_update_samplers() that prevents it from
working (olv)
|
|
|
|
Rename it to xglResetDescriptorPool().
|
|
XGL_DESCRIPTOR_SET_LAYOUT_CREATE_INFO now has an array of bindings. Rename
intel_desc_layout_range to intel_desc_layout_binding.
|
|
This should be trivial, except that the intel driver has intel_desc_pool
internally. Exchange the names of intel_desc_pool and intel_desc_region.
|
|
|
|
|
|
Move almost all linked-lists to unordered_maps and vectors.
Clean up malloc/free with new/delete.
Skip write-strings compile warnings for graphviz_helper.
Still need to migrate the DS update tree to c++.
|
|
Take advantage of containers, memory allocation, and strings.
|
|
Bug #13402
header version: 0.64.0
|
|
|
|
Distinguish between freeObj calls that are being done for
layer internal reasons (i.e. tracking presentable_image objects)
and user actions to eliminate unnecessary warning from layer.
|
|
|
|
|
|
bug #13473
header: r29790 (0.63.0)
|
|
bug #13425
header: r29769 (0.62.0)
v2: cast uint64_t to int64_t with range check (olv)
|
|
bug #13230
header revision: r29764 (0.62.0)
Part 1: Change BUFFER_USAGE, BUFFER_VIEW and BUFFER_VIEW_CREATE_INFO.
|
|
header: r29714 (0.61.0)
|
|
commit 57a757fda4f6a62cb1187b015f28119ce9c5a2b2
Author: Courtney Goeltzenleuchter <courtney@LunarG.com>
Date: Wed Mar 25 15:37:49 2015 -0600
xgl: Remove PIPELINE_DELTA.
bug: 13065
header: 0.60.0
Remove PIPELINE_DELTA.
Add functions for creating and loading pipeline derivatives.
|
|
bug: 13065
header: 0.60.0
Remove PIPELINE_DELTA.
Add functions for creating and loading pipeline derivatives.
v2: fix warnings (olv)
|
|
commit 6f5bca9057be71e818982c4cece0cb5ed1743b0f
Author: Courtney Goeltzenleuchter <courtney@LunarG.com>
Date: Wed Mar 25 11:25:10 2015 -0600
xgl: Add IMAGE_LAYOUT to Cmd*Image* calls
This is part 2 of bug #12904.
(2) Blit operations need to know what layout images are in. How do they figure that out?
- Added new parameter(s) specifying the image layout(s) to the following functions:
xglCmdCopyImage
xglCmdCopyBufferToImage
xglCmdCopyImageToBuffer
xglCmdClearColorImage
xglCmdClearColorImageRaw
xglCmdClearDepthStencilImage
xglCmdResolveImage
|
|
This is part 2 of bug #12904.
(2) Blit operations need to know what layout images are in. How do they figure that out?
- Added new parameter(s) specifying the image layout(s) to the following functions:
xglCmdCopyImage
xglCmdCopyBufferToImage
xglCmdCopyImageToBuffer
xglCmdClearColorImage
xglCmdClearColorImageRaw
xglCmdClearDepthStencilImage
xglCmdResolveImage
|
|
This is part 2 of bug #12904.
(2) Blit operations need to know what layout images are in. How do they figure that out?
- Added new parameter(s) specifying the image layout(s) to the following functions:
xglCmdCopyImage
xglCmdCopyBufferToImage
xglCmdCopyImageToBuffer
xglCmdClearColorImage
xglCmdClearColorImageRaw
xglCmdClearDepthStencilImage
xglCmdResolveImage
|
|
Rebase updates
Part one of adding support for abitrary scaled blits and format conversions.
Just interface here. Functionality to follow.
bug #13071
alpha rev: r29665
|
|
part 1 of bug #12904:
(1) What layout are images in when they are created?
- Added new XGL_IMAGE_LAYOUT_UNINITIALIZED constant. This is the layout images are in right after creation. The application has to transition images to some useable layout before using the image. It can only be used as the source layout of a resource transition.
An image is considered to have UNDEFINED layout in the following cases:
* right after creation
* if the contents of the memory backing the image has been changed by performing an operation on another resource which (partially or fully) shares the same memory backing
The content of an image that is in UNDEFINED layout is considered undefined and unless the image is transitioned to another layout any operation on the image results in the content of the image to stay undefined.
The application can transition any image from UNDEFINED layout to another layout anytime. Right after that the content of the image still remains undefined but subsequent operations performed on the image (e.g. clear) will have well defined results.
|
|
part 3 of bug #12904
alpha header: r29684 (0.59.0)
(3) Where does xglCmdResetEvent update the event's status? Unlike xglCmdSetEvent this function doesn't take a parameter specifying this.
- Added pipeEvent parameter also to xglCmdResetEvent. Also renamed XGL_SET_EVENT to XGL_PIPE_EVENT as now it's no longer used only for set operations.
|
|
bug #13066
alpha header: r29666
Currently not using this bit in sample driver.
|
|
Part one of adding support for abitrary scaled blits and format conversions.
Just interface here. Functionality to follow.
bug #13071
alpha rev: r29665
|
|
bug # 13323
alpha header: r29635
Don't actually remove XGL_FRAMEBUFFER but do change how
RenderPass and Framebuffer are connected. Some comments
from the bug:
- Created a new structure XGL_RENDER_PASS_BEGIN that contains
both the XGL_RENDER_PASS and XGL_FRAMEBUFFER.
- XGL_CMD_BUFFER_GRAPHICS_BEGIN_INFO and xglCmdBeginRenderPass
both use XGL_RENDER_PASS_BEGIN to ensure they stay consistent.
- Renamed the member in XGL_CMD_BUFFER_GRAPHICS_BEGIN_INFO to
renderPassContinue with a comment to clarify that this is
only needed when a render pass is split across two command buffer.
The last has the biggest impact on this patch. The tests now
directly call CmdBeginRenderPass and CmdEndRenderPass in the
command buffer rather than set the BEGIN_INFO to a render pass and
have the driver implicitly do BeginRenderPass and EndRenderPass.
It would probably still work, but didn't seem to match the intent
with this change in the header file.
|