| Age | Commit message (Collapse) | Author |
|---|
|
Freeing object memory before destroying object resulted in a segfault.
This is illegal but shouldn't crash the layer.
|
|
Now that debug report functionality is in place can use
that mechanism to handle logging enabled by vk_layer_settings.txt
|
|
Performance analysis showed a lot of time being spent
formating unused log messages (particularly INFO). By
moving the formatting into the logging function we can
only do the formatting if the message will be used.
|
|
|
|
Hopefully makes code easier to read when using the
dispatch table access functions.
|
|
Needed to make sure we had a dispatchable object to
perform logging. Can't trust that mem or object are
correct, so have caller pass in user's dispatchable object.
|
|
|
|
Need to intialize the record put on the list.
Removed the unused, newed record.
|
|
|
|
This patch eliminates global variables and accesses
everything through pre-instance or per-device storage.
Have basic template supporting get_my_data_ptr function working.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Change all layers and loader interface to init dispatch tables on
GPA("GetXXXProcAddr"). After that initialization rest of dispatch
tables are inited via unwrapped object using the GPA in the dispatch
table. This also allows App generated GPA calls that the loader can't resolve
to function correctly.
|
|
Conflicts:
layers/mem_tracker.cpp
|
|
This patch starts restructuring the various components
(loader, driver, layers, etc.) to support global and
device extensions. Require GetProcAddr to access the
extension functions and related support.
|
|
Also cleanup layer/loader GPA's to only return device level entrypoints.
|
|
|
|
Also add to instance dispatch table CreateInstance and GetGlobalExtensionInfo
|
|
Make the layer init separate from either device or instance dispatch
table init, since these are done at different times.
|
|
Loader will use this for constructing instance layer instance chains.
Layer instance dispatch table is not yet added.
|
|
Non-sparse memory bindings are immutable. Remove unbinding checks,
unbinding before free in demos/tests, and add validation check for
rebinding objects, appropriately renamed memtracker support routines.
|
|
Remove one unnecessary level of indirection.
Conflicts:
layers/mem_tracker.cpp
|
|
Reverse sense of fenceMap using VkFence instead of fenceID as key.
This eliminates iterating over all of fenceMap.
Stop creating internal fences and let vkQueueSubmit pass NULL fence.
Track lastSubmittedFence and lastSubmittedQueue in command buffer info.
Use unordered_map instead of map for better performance.
Reuse iterator returned by map find whenever available.
It is faster than repeating a find operation with "[]" notation.
Remove map entries at time of VkDestroy instead of other calls such as Wait.
|
|
updateFenceTracking was using iterator for deleted item.
|
|
Remove unneeded getFenceFromId() that crashed outside of mutex.
|
|
|
|
Remove multiple memory allocation requirements from API, and supporting
changes in driver, demos, layers, and tests.
|
|
|
|
Added layer validation tests for several ObjectTracker and
MemTracker validation cases. Also improved error handling
in object tracker layer to handle validation failures that
could cause unrecoverable driver errors.
|
|
New names and behavior for QueueBind* APIs.
|
|
|
|
header: 0.93.0
svn: 30980
|
|
svn ID: 30918
Change the VkCmdClearColroImage color parameter
be a pointer.
|
|
This includes suppressing compiler warning C4065, where VS complains if a
switch statement has a "default", but not "case" statements. This was true in
many functions of the auto-generated "helper" header files.
This also includes removing unreferenced local variables, and dealing with some
type differences (sometimes with casting, sometimes with changing the type).
|
|
|
|
|
|
The loop was blindly incrementing the iterator and
it should not. In the delete case the loop should
resume with the iterator returned
by the delete as it's the next element in the list.
|
|
Cleaned up alignment, made function definitions consistent,
shortened some line lengths.
|
|
|
|
The upstream vulkan.h header removed maxLayerCount parameter
from vkEnumerateLayers. This patch implements that change
for the sample driver and layers.
|
|
|
|
Switch from VK_WSI_X11 to VK_WSI_LunarG
v2:
- split out glave changes
- redo mem_tracker changes
- rebase
Conflicts:
demos/cube.c
demos/tri.c
v3 (from Ian):
- Put "#if 0" around non-upstream members of VkDisplayPropertiesWSI.
|
|
The blit tests were skipping this step resulting in tons of memtracker
validation errors.
|
|
Rename some vkCmdBindDescriptorSets parameters and add a dynamic offset count.
v2: Add a check against dynamicOffsetCount before copying dset data.
|
|
|
|
This is part of r30446 on vulkan.h, discussed in bug 13632.
Meaty chunks of this commit done by Olv, basic hook up and testing by Cody.
|
