| Age | Commit message (Collapse) | Author |
|---|
|
|
|
|
|
|
|
|
|
https://cvs.khronos.org/bugzilla/show_bug.cgi?id=14352
|
|
Occasional saw issues with parallel make because dependency
was not specified for the shader SPIRV components used by tri
and cube.
|
|
https://cvs.khronos.org/bugzilla/show_bug.cgi?id=14330
Accepted in gl_common 7/27
|
|
Added to the Image validation layer.
|
|
Had missed initializing some WSI function pointers for
the device dispatch table.
|
|
This patch fixes LunarXchange issue #84.
Also adds corresponding stub entry point to the Intel
sample driver.
|
|
|
|
A copy and paste error from the spec made it seem that all ICDs will support
VK_PRESENT_MODE_IMMEDIATE_WSI (a.k.a. "immediate"). That's not true. They all
need to support VK_PRESENT_MODE_FIFO_WSI. Changed the comment and code so that
the preference of these demos is:
1) VK_PRESENT_MODE_MAILBOX_WSI
2) VK_PRESENT_MODE_IMMEDIATE_WSI
3) VK_PRESENT_MODE_FIFO_WSI
|
|
Previously we passed `count` down to the validator and then dropped it
on the floor, ignoring anything other than pipeline 0.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Courtney Goeltzenleuchter <courtney@lunarg.com>
|
|
A cut+paste error resulted in using the same value
for the explicit and implicit layer config paths.
This change remedies that.
Signed-off-by: James Jones <jajones@nvidia.com>
|
|
Reviewed-by: James Jones <jajones@nvidia.com>
|
|
The previous libdir/confdir patches mistakenly
used /usr/etc rather than /etc as the default
location for configuration files. This patch
corrects that mistake.
Signed-off-by: James Jones <jajones@nvidia.com>
Tested-by: Liam Middlebrook <lmiddlebrook@nvidia.com>
|
|
The previous attempt at using GNU standard paths
limited searching to the prefix where the loader
was installed. What we really want though is to
always search in /usr, also search in whatever
local prefix the user installed at if it wasn't
/usr, and use the standard GNU libdir, sysconfdir,
and datadir under those prefixes. This change
implements that behavior.
Signed-off-by: James Jones <jajones@nvidia.com>
Tested-by: Liam Middlebrook <lmiddlebrook@nvidia.com>
|
|
These should get inherited from the build environment, which should
handle the variation between /usr/lib/$triple on debian vs /usr/lib{,64}
on redhat.
Signed-off-by: Adam Jackson <ajax@redhat.com>
|
|
This reverts commit 3d2dc08ed28b159f1d86476add4719b3ec975b1b.
Just adding attributiones here does not match our SDK
and loader redistributable plans. Will add attributions to
specific areas rather than a global attribution file.
|
|
|
|
The loaderMagic grew to be 64-bits to look better
but need to mask it to 32bits for the validation check.
|
|
|
|
This effects Windows layers.
|
|
Also fix several other missing entrypoints from dispatch table related functions.
Also reorder some of these functions to better match header list order so
can more easily spot missing entry points.
|
|
|
|
|
|
|
|
|
|
|
|
Twice we've have feedback on this structure not being
correct - folks missed the union statment - so changing
this to be the same size as a pointer to make it more
clear that loaderMagic is the same size as *loaderData.
|
|
When cube sees --break on the command line it will
install the DebugBreak helper callback for layer validation
errors and warnings.
|
|
Implement BreakCallback that an application can use
to set a debugger breakpoint when running on Visual Studio
or gdb for requested debug message flags.
|
|
The DEBUG_REPORT callback helpers do not need to be
included in the dispatch table as they are not API entrypoints.
They are functions specifically and only for DEBUG_REPORT
DbgCreateMsgCallback.
|
|
|
|
The convention is to scan /etc for host-specific settings before looking
in /usr/share, which may be a shared network filesystem.
Signed-off-by: Adam Jackson <ajax@redhat.com>
|
|
|
|
|
|
In vk_debug_report_lunarg.h
|
|
This eliminates a ton of bad destroy messages, also removes check for
leftover instances when device is destroyed.
|
|
report_data is only defined if DEBUG_REPORT is enabled,
so asserting that it must be non-null prevents things from
running for no reason.
|
|
|
|
gcc 5.1 will complain (and refuse to compile) if you use 'auto' in the
C++11 sense but haven't asked for C++11.
|
|
|
|
Changes to "tri" are similar to the fixes for "cube":
1) WSI function were being called with an uninitialized pPlatformWindow.
Fixed by splitting demo_init_vk into two functions.
2) VkAttachmentViewCreateInfo.format wasn't being set.
3) VkShaderCreateInfo.pName wasn't being set.
4) demo->cmdPool was being leaked.
|
|
demo_init_vk(), which sets the surface_description.pPlatformWindow was
getting called before the window was created so surface_description.pPlatformWindow
was always NULL. I added a new demo_init_vk_wsi() which splits demo_init_vk()
in two so that the window is created before surface_description.pPlatformWindow is assigned.
|
|
|
|
Need to know the number of queues in multiple places so just
save it as part of the demo info data structure.
|
|
Remove the direct queries for the layer extensions and properties.
|
|
|
|
|
