When no descriptor mapping is specified, the backend will just
build the usual default mapping. Otherwise the explicit mapping
is used.
Once all backends support the explicit mapping, we'll be able to
handle generating the default mapping in the shader runner core
rather than having each backend implement its own algorithm.
So far only the d3d12 backend supports explicit descriptor
mapping.
Mostly to be able to associate a version number to each tag and
get rid of all the foo<1.2.3 tags. The new system also has fixed
tag slots, rather than dealing with strings, so we don't have to
manually adjust the size of the `tags' array.
With the new system each tag can be present or not, and if it is
present it can have an associated version number (of the form
major.minor.patch). If the version is not available, it is set to
0.0.0. Each tag can be queried for existence and for comparison
with the version number.
So we have a more recent version of SPIRV-Tools and also don't
have to recompile Mesa to test llvmpipe. This fixes a few failing
tests, but also breaks a couple.
Similar to how we have the "geometry-shader" cap. In principle shader
model 5+ implies support for tessellation shaders, but the Vulkan,
OpenGL, and Metal runners are able to support most of shader model 5+
without the underlying GPU (or API) necessarily supporting tessellation
shaders.
My main motivation to this is avoiding generating a lot of useless
log lines from other executors when I'm interested in just one of
them, but I can imagine this also somewhat improving efficiency.
Since this test depend on the specific code generated by the
native d3dcompiler we add the possibility to specify a "raw"
shader using a hex format. When the shader assembler is finally
available they should be replaced with assembly code.
