Yes, this is only ever provided by an ANGLE extension, and as you might
have guessed, The Great Google Overlords didn't bother caring to have
this supported outside of their browser.
Since the main point of the low-level API is to make use of the
EXT_multi_draw_arrays / ANGLE_multi_draw / WEBGL_multi_draw extensions
for a reduced driver overhead *and* the builtin gl_DrawID variable, it
just doesn't make sense to provide a fallback to draw() in a loop. When
the extensions are not available, the user has to do extra work in order
to supply proper UBO draw offset for each draw call, and thus the
fallback path has basically no practical use.
On the other hand, draw() taking the MeshView instances is kept as-is --
this one is rather old and so breaking compatibility would be an
unnecessary pain point. Plus, since it has to allocate the contiguous
arrays internally, it's not desirable for any fast-path rendering
anyway.
It doesn't allow anything except full RGBA and 32-bit types to be read
from the framebuffer. Chromium browsers weren't complaining but Firefox
did. I remember a similar restriction from GLES2 but all implementations
I tested with including ANGLE and SwiftShader allow that already, so
it's practically just a WebGL limitation.
This doesn't actually fix the WebGL-specific error (there it
additionally disallows mismatched signedness in a vertex attribute) but
is nevertheless good to have correct.
Because a MeshView might not be the best thing to have when you are
submitting a batch of thousand draws. It takes a strided array views to
allow for more flexibility, but can also detect if the input is already
contiguous and use it as-is.
UNFORTUNATELY the GL 1.0 legacy still continues to stink and so there
has to be a 64-bit-specific overload which is the *actual* variant that
doesn't allocate because glMultiDrawElements takes a `void**` for INDEX
OFFSETS and it's IN BYTES! Which foolish soul designed such a thing back
in the 1860s, I wonder. There's no reason to not have an index offset
in elements because all indices have to have the same type ANYWAY. And
yes, I wasted about three hours debugging driver crashes because I
THOUGHT this parameter takes offset in elements, not bytes.
Also note: on 32-bit platforms this depends on latest Corrade with the
CORRADE_TARGET_32BIT definition. Spent an embarrassing amount of time
wondering why all local builds but Emscripten work.
Or, the horror when I comment-out half of it and it still kinda passes.
The new code tests it more thoroughly, with actually checking that
different draw produce different results instead of one draw overwriting
the other.
Plus also verifying interaction with gl_VertexID and gl_DrawID to ensure
it really behaves correctly:
- gl_VertexID should get the same value regardless of whether we
perform a multi-draw or a one-by-one fallback, so this catches issues
where the mesh would get reconfigured in some nasty way, losing this
information
- gl_DrawID should get properly populated if and only if we use the
multi-draw code path, so this catches issues where the fallback would
get used instead of the actual multi-draw code path
Also test it properly and add a check for extension availability. Sigh,
seems like nobody ran this test on an Apple device since this thing got
added in 75d238f50b ...
Its only use was for specifying N-dimensional SamplerWrapping because,
compared to a Math::Vector, it had an implicit constructor from a single
value (whereas the Vector has it explicit). I solved that by simply
adding a few single-value overloads where it mattered. There, done. No
need for this weird thing and confusion with Containers::Array anymore.
All places that used it now use Math::VectorN<SamplerWrapping>, but the
class is still included for backwards compatibility purposes, together
with providing implicit conversion from and to a Vector.
Ah the good old times where my brain was not overflowing with
unnecessarily deep knowledge about how GPUs (and OpenGL drivers in
particular) work. Nope, switching a blend state and calling glFinish()
won't generate any GPU work, that will be all deferred until there's
something to be actually drawn, cleared or copied.
Disabling engine startup log or modifying enabled extensions /
workarounds from the application side was one of the common pain
points and this should *finally* solve the problem. This Configuration
is now inherited by the usual Platform::*Application::GLConfiguration /
Platform::Windowless*Application::Configuration classes people are used
to, so for the end user it's just as if these classes got a bunch new
options.
Having this, I also extended the ContextGLTest to verify that the
Configuration and command-line options do what's expected because that
hadn't automated tests until now. The test is mostly a copy of what I
did for Vulkan already, nothing special. Additionally all
Platform*ApplicationTest executables gained a new --quiet option to
verify that the GL::Context::Configuration subset gets correctly passed
from the Application code, because that's something we can't really
verify in an automated way.
We're going to eventually include this class in all Application classes
(need that in order to inherit a to-be-created Configuration class) and
the <string> and <vector> would be just too much. This change caused
magnum-gl-info.wasm (WebGL 2 build) to go down from 247 to 245 kB. Not
much, but that's I guess because there's still a lot other vectors of
strings elsewhere.
There's a lot more places to clean up, will do those in separate
commits. This change is the most atomic I could do, and it introduces a
breaking change to all APIs that returned a std::vector or a
std::string. Fortunately (or as I hope) those weren't used that much, so
it shouldn't cause build breakages for that many people.
Quite a lot of the optimization ideas is borrowed from the new Vk
library -- such as "interning" the driver workaround strings to avoid
allocating their copies.
Vladimír Vondruš