We can double import a dmabuf if we use it as a texture target and
a render target. Instead, let's unify render targets and texture dmabuf
imports to use wlr_gles2_buffer which manages the EGLImageKHR
Since imported textures will be based off of gles2_buffer we have
to destroy textures first or else they will have an invalid reference
to the buffers they are imported from.
If the external-only flag is set, then the EGLImage is only
supported for use with GL_TEXTURE_EXTERNAL_OES texture targets.
In particular, the EGLImage cannot be bound to a RBO.
It's possible that we don't have an EGLDevice if we created the
EGL context from a GBM device. Let's ensure all GPU-accelerated
renderers always have a DRM FD to return by falling back to GBM's
FD.
When a texel from the Vulkan format VK_FORMAT_B8G8R8A8_SRGB is read,
the sRGB to linear conversion is applied independently to the R, G,
and B channels; the A channel has no influence on this. However,
DRM_FORMAT_ARGB8888 buffers are, per Wayland protocol, not encoded
in this fashion; one must first unpremultiply the color channels
before doing sRGB to linear conversion. This commit switches to
handling ARGB8888 and ABGR8888 formats using the general fragment
shader conversion from electrical to optical values.
Not needs set GL_DEPTH_TEST, Because when rendering to a framebuffer
that has no depth buffer, depth testing always behaves as though
the test is disabled, The initial value for each capability with
the exception of GL_DITHER is GL_FALSE.
We could potentially leak a display here, but not really because the
display acts as a singleton that will be returned next time a renderer
of the same device is created.
If the compositor were to try to handle a GPU reset within the lost
signal (by recreating the renderer) we should avoid referencing renderer
resources after the lost signal. This prevents use after free for such
compositors.
We can just use a regular assignment instead. This is more
type-safe since there is no need to provide the struct size.
The remaining memcpy() calls perform array copies or copies from
void pointers (which may be unaligned).
This will become necessary when we switch away from scissoring. For the
time being, this cleans things up a bit and allows for a trivial
blending implementation for textures when that comes.
Avoids repeating the common bits between dma_tex_features and
dma_tex_ycbcr_features, and we will need just the YCbCr-related
flags for shm YCbCr support soon.
These features are required for shm only: the TRANSFER stuff is
for texture upload. We don't need these for DMA-BUFs. Make this
clearer by changing the name.
Also re-order the definitions to group all texture-related features
together.
Based on five calls:
wlr_render_timer_create - creates a timer which can be reused across
frames on the same renderer
wlr_renderer_begin_buffer_pass - now takes a timer so that backends can
record when the rendering starts and finishes
wlr_render_timer_get_time - should be called as late as possible so that
queries can make their way back from the GPU
wlr_render_timer_destroy - self-explanatory
The timer is exposed as an opaque `struct wlr_render_timer` so that
backends can store whatever they want in there.
The Vulkan spec states:
> For the purposes of range expansion and Y′CBCR model conversion,
> the R and B components contain color difference (chroma) values
> and the G component contains luma.
The equations below that sentence also help understand the mapping.
We don't actually need the REPEAT mode, and this makes things more
consistent with the YCbCr sampler (which requires CLAMP_TO_EDGE for
spec compliance).
Also drop borderColor which is unused for this mode.
Fixes the following validation error:
[ VUID-VkImageViewCreateInfo-pNext-01970 ] Object 0: handle = 0x62e00003c400, type = VK_OBJECT_TYPE_DEVICE; | MessageID = 0xf378e14b | vkCreateImageView(): If there is a VkSamplerYcbcrConversion, the imageView must be created with the identity swizzle.
Some formats like sub-sampled YCbCr use a block of bytes to
store the color values for more than one pixel. Update our format
table to be able to handle such formats.
Address Sanitizer reported two memory leaks because of
fmt not being freed in wlr_drm_format_set_copy.
This commit fixes the memory leak by removing the
allocation to the heap. The struct is allocated to the
stack anyway.
This is implemented by a two-subpass rendering scheme; the first
subpass draws (and blends) onto a linear R16G16B16A16_SFLOAT buffer,
while the second subpass performs linear->srgb conversion, writing
onto the actual output buffer.
This fixes the following validation errors when shutting down Sway:
00:00:01.263 [wlr] [render/vulkan/vulkan.c:65] Validation Error: [ VUID-vkDestroyFramebuffer-framebuffer-00892 ] Object 0: handle = 0x62e00003c400, type = VK_OBJECT_TYPE_DEVICE; | MessageID = 0xdb308312 | Cannot call vkDestroyFramebuffer on VkFramebuffer 0x2e2cd000000002b[] that is currently in use by a command buffer. The Vulkan spec states: All submitted commands that refer to framebuffer must have completed execution (https://www.khronos.org/registry/vulkan/specs/1.3-extensions/html/vkspec.html#VUID-vkDestroyFramebuffer-framebuffer-00892) (VUID-vkDestroyFramebuffer-framebuffer-00892)
00:00:01.264 [wlr] [render/vulkan/vulkan.c:65] Validation Error: [ VUID-vkDestroyImage-image-01000 ] Object 0: handle = 0x62e00003c400, type = VK_OBJECT_TYPE_DEVICE; | MessageID = 0xf2d29b5a | Cannot call vkDestroyImage on VkImage 0x3fbcd60000000028[] that is currently in use by a command buffer. The Vulkan spec states: All submitted commands that refer to image, either directly or via a VkImageView, must have completed execution (https://www.khronos.org/registry/vulkan/specs/1.3-extensions/html/vkspec.html#VUID-vkDestroyImage-image-01000) (VUID-vkDestroyImage-image-01000)
Remove debug logs when a texture is created, since this happens
pretty often. Use drmGetFormatName() and drmGetFormatModifierName()
to log DRM formats and modifiers.
This ensures that the pool sizes grow exponentially, making the number
of pools needed logarithmic in the number of descriptors, instead of
linear. Since the first pool's size is 256, this change only has an
effect when the compositor creates a large number of textures.
Goals:
- Extensibility: we need to be able to add new params to the calls
to render a texture/rect. For instance we'll need to add fences to
the render texture operation for explicit sync purposes.
- No implicit state: no more bind_buffer, begin, end.
- No matrices: these hurt Pixman and we don't need them.
- Clip regions for optimized damage repainting.
Closes: https://gitlab.freedesktop.org/wlroots/wlroots/-/issues/3188
Setting the GLESv2 parameter GL_PACK_ALIGNMENT to 1 ensures that the
stride of the glReadPixels output matches the value computed in
`pack_stride`. Since the default value of GL_PACK_ALIGNMENT is 4, this
does not make a difference under normal use; but without this patch
the stride can be incorrect; for example, with RGB565 buffers and
screenshots of regions with odd width.
We've had this struct for a while. It'd be useful for compositors
if they want to manage the swap chains themselves instead of being
forced to use wlr_output's. Some compositors might also want to use
a swapchain without an output.
Reverts commit c73e20628a.
This caused a regression in the GLES2 renderer because `egl->exts.EXT_device_drm` is set in
`egl_init_display()`, which is invoked after `get_egl_device_from_drm_fd()`. So the function will
always return `EGL_NO_DEVICE_EXT`.
Some formats like YUV are only supported by Mesa for sampling, not
for rendering. However we always unconditionally added the INVALID
modifier to the list of render formats.
Check whether all modifiers are external-only, in that case, don't
add INVALID to the list of render formats.
