Previously the uniform array flattening feature would trigger on loose
uniform arrays of any basic type (e.g, floats). This PR restricts it
to sampler and texture arrays. Other arrays would end up in their own
uniform block (anonymous or otherwise). (Atomic counter arrays might be an
exception, but those are not currently flattened).
This checkin adds a --flatten-uniform-arrays option which can break
uniform arrays of samplers, textures, or UBOs up into individual
scalars named (e.g) myarray[0], myarray[1], etc. These appear as
individual linkage objects.
Code notes:
- shouldFlatten internally calls shouldFlattenIO, and shouldFlattenUniform,
but is the only flattening query directly called.
- flattenVariable will handle structs or arrays (but not yet arrayed structs;
this is tested an an error is generated).
- There's some error checking around unhandled situations. E.g, flattening
uniform arrays with initializer lists is not implemented.
- This piggybacks on as much of the existing mechanism for struct flattening
as it can. E.g, it uses the same flattenMap, and the same
flattenAccess() method.
- handleAssign() has been generalized to cope with either structs or arrays.
- Extended test infrastructure to test flattening ability.
This PR adds the ability to offset sampler, texture, and UBO bindings
from provided base bindings, and to auto-number bindings that are not
provided with explicit register numbers. The mechanism works as
follows:
- Offsets may be given on the command line for all stages, or
individually for one or more single stages, in which case the
offset will be auto-selected according to the stage being
compiled. There is also an API to set them. The new command line
options are --shift-sampler-binding, --shift-texture-binding, and
--shift-UBO-binding.
- Uniforms which are not given explicit bindings in the source code
are auto-numbered if and only if they are in live code as
determined by the algorithm used to build the reflection
database, and the --auto-map-bindings option is given. This auto-numbering
avoids using any binding slots which were explicitly provided in
the code, whether or not that explicit use was live. E.g, "uniform
Texture1D foo : register(t3);" with --shift-texture-binding 10 will
reserve binding 13, whether or not foo is used in live code.
- Shorter synonyms for the command line options are available. See
the --help output.
The testing infrastructure is slightly extended to allow use of the
binding offset API, and two new tests spv.register.(no)autoassign.frag are
added for comparing the resulting SPIR-V.
Added -C option to request cascading errors. By default, will exit early,
to avoid all error-recovery-based crashes.
This works by simulating end-of-file in input on first error, so no
need for exception handling, or stack unwinding, or any complex error
checking/handling to get out of the stack.
Previously GlslangToSpv() reported missing/TBD functionalities
by directly writing to stdout using printf. That could cause
problems to callers of GlslangToSpv(). This patch cleans up
the error reporting logic in GlslangToSpv(), TGlslangToSpvTraverser,
and spv::Builder a little bit to use ostringstream.
Also fixed the usage of GlslangToSpv() in GTest fixtures to
capture warnings/errors reported when translating AST to SPIR-V.
The existing test harness is a homemade shell script. All the tests
and the expected results are written in plain text files. The harness
just reads in a test, invoke the glslangValidator binary on it, and
compare the result with the golden file. All tests are kinda
integration tests.
This patch add Google Test as an external project, which provides a
new harness for reading shader source files, compile to SPIR-V, and
then compare with the expected output.
This plumbs both the current file path and the include depth
back up to the includer. This allows the includer to properly
support relative paths.
This also replaces the string copy that was done during include
with a zero-copy method of accomplishing the same thing. This
prevents extra copies of entire files.
Separating file I/O from compile/link lets the compile/link be done
repeatedly in a loop for testing and measuring of performance and
memory footprint, including seeing memory growth over time for
functional-level memory-leak testing.
While the older compile-only mode already had this functionality,
and typically showed no memory leaks, SPIR-V uses the link path,
has pending "TODO" for memory freeing, and this shows several
kilobytes of leaking per compile-link. Most likely, pending
merge request 131 will address much of this.
Using platform-neutral osinclude.h makes it easier to substitute
implementation when necessary and eliminates some variability between
build configurations.
* Linux folder has been renamed to Unix, to match defines and so that it
compiles on OS X.
* This removes the need for a per-platform include search path for the
right OS folder
The new make-revision script regenerates glslang/Include/revision.h,
used as it always has been, but made with a git-tag version and the
the number of commits on master.
I have a pre-commit hook that will automatically do this on master,
likely often enough to work in practice, without needing pull requests
to include it.
When an include directive is recognized by the preprocessor, it
executes a callback on the filepath argument to obtain the file
contents. That way the compilation client can deal with the file
system, include paths, etc.
Currently only accepts quoted filepaths -- no angle brackets yet.
